Merge pull request #329 from telemt/bump

Update Cargo.toml

.github/codeql/codeql-config.yml

@@ -0,0 +1,19 @@
+name: "Rust without tests"
+
+disable-default-queries: false
+
+queries:
+  - uses: security-extended
+  - uses: security-and-quality
+  - uses: ./.github/codeql/queries
+
+query-filters:
+  - exclude:
+      id:
+        - rust/unwrap-on-option
+        - rust/unwrap-on-result
+        - rust/expect-used
+
+analysis:
+  dataflow:
+    default-precision: high

.github/codeql/queries/common/ProductionOnly.qll

@@ -0,0 +1,20 @@
+import rust
+
+predicate isTestOnly(Item i) {
+  exists(ConditionalCompilation cc |
+    cc.getItem() = i and
+    cc.getCfg().toString() = "test"
+  )
+}
+
+predicate hasTestAttribute(Item i) {
+  exists(Attribute a |
+    a.getItem() = i and
+    a.getName() = "test"
+  )
+}
+
+predicate isProductionCode(Item i) {
+  not isTestOnly(i) and
+  not hasTestAttribute(i)
+}

.github/codeql/queries/qlpack.yml

@@ -0,0 +1,4 @@
+name: rust-production-only
+version: 0.0.1
+dependencies:
+  codeql/rust-all: "*"

.github/workflows/codeql.yml

@@ -0,0 +1,45 @@
+name: "CodeQL Advanced"
+
+on:
+  push:
+    branches: [ "*" ]
+  pull_request:
+    branches: [ "*" ]
+  schedule:
+    - cron: '0 0 * * 0'
+
+jobs:
+  analyze:
+    name: Analyze (${{ matrix.language }})
+    runs-on: ${{ (matrix.language == 'swift' && 'macos-latest') || 'ubuntu-latest' }}
+
+    permissions:
+      security-events: write
+      packages: read
+      actions: read
+      contents: read
+
+    strategy:
+      fail-fast: false
+      matrix:
+        include:
+          - language: actions
+            build-mode: none
+          - language: rust
+            build-mode: none
+
+    steps:
+      - name: Checkout repository
+        uses: actions/checkout@v4
+
+      - name: Initialize CodeQL
+        uses: github/codeql-action/init@v4
+        with:
+          languages: ${{ matrix.language }}
+          build-mode: ${{ matrix.build-mode }}
+          config-file: .github/codeql/codeql-config.yml
+
+      - name: Perform CodeQL Analysis
+        uses: github/codeql-action/analyze@v4
+        with:
+          category: "/language:${{ matrix.language }}"

.github/workflows/release.yml

@@ -0,0 +1,139 @@
+name: Release
+
+on:
+  push:
+    tags:
+      - '[0-9]+.[0-9]+.[0-9]+'
+  workflow_dispatch:
+
+permissions:
+  contents: read
+  packages: write
+
+env:
+  CARGO_TERM_COLOR: always
+
+jobs:
+  build:
+    name: Build ${{ matrix.target }}
+    runs-on: ubuntu-latest
+    permissions:
+      contents: read
+
+    strategy:
+      fail-fast: false
+      matrix:
+        include:
+          - target: x86_64-unknown-linux-gnu
+            artifact_name: telemt
+            asset_name: telemt-x86_64-linux-gnu
+          - target: aarch64-unknown-linux-gnu
+            artifact_name: telemt
+            asset_name: telemt-aarch64-linux-gnu
+          - target: x86_64-unknown-linux-musl
+            artifact_name: telemt
+            asset_name: telemt-x86_64-linux-musl
+          - target: aarch64-unknown-linux-musl
+            artifact_name: telemt
+            asset_name: telemt-aarch64-linux-musl
+
+    steps:
+      - uses: actions/checkout@v4
+
+      - uses: dtolnay/rust-toolchain@v1
+        with:
+          toolchain: stable
+          targets: ${{ matrix.target }}
+
+      - name: Install cross-compilation tools
+        run: |
+          sudo apt-get update
+          sudo apt-get install -y gcc-aarch64-linux-gnu
+
+      - uses: actions/cache@v4
+        with:
+          path: |
+            ~/.cargo/registry
+            ~/.cargo/git
+            target
+          key: ${{ runner.os }}-${{ matrix.target }}-cargo-${{ hashFiles('**/Cargo.lock') }}
+          restore-keys: |
+            ${{ runner.os }}-${{ matrix.target }}-cargo-
+
+      - name: Install cross
+        run: cargo install cross --git https://github.com/cross-rs/cross
+
+      - name: Build Release
+        env:
+          RUSTFLAGS: ${{ contains(matrix.target, 'musl') && '-C target-feature=+crt-static' || '' }}
+        run: cross build --release --target ${{ matrix.target }}
+
+      - name: Package binary
+        run: |
+          cd target/${{ matrix.target }}/release
+          tar -czvf ${{ matrix.asset_name }}.tar.gz ${{ matrix.artifact_name }}
+          sha256sum ${{ matrix.asset_name }}.tar.gz > ${{ matrix.asset_name }}.sha256
+
+      - uses: actions/upload-artifact@v4
+        with:
+          name: ${{ matrix.asset_name }}
+          path: |
+            target/${{ matrix.target }}/release/${{ matrix.asset_name }}.tar.gz
+            target/${{ matrix.target }}/release/${{ matrix.asset_name }}.sha256
+
+  build-docker-image:
+    needs: build
+    runs-on: ubuntu-latest
+    permissions:
+      contents: read
+      packages: write
+
+    steps:
+      - uses: actions/checkout@v4
+
+      - uses: docker/setup-qemu-action@v3
+      - uses: docker/setup-buildx-action@v3
+
+      - name: Login to GHCR
+        uses: docker/login-action@v3
+        with:
+          registry: ghcr.io
+          username: ${{ github.actor }}
+          password: ${{ secrets.GITHUB_TOKEN }}
+
+      - name: Extract version
+        id: vars
+        run: echo "VERSION=${GITHUB_REF#refs/tags/}" >> $GITHUB_OUTPUT
+
+      - name: Build and push
+        uses: docker/build-push-action@v6
+        with:
+          context: .
+          push: true
+          tags: |
+            ghcr.io/${{ github.repository }}:${{ steps.vars.outputs.VERSION }}
+            ghcr.io/${{ github.repository }}:latest
+
+  release:
+    name: Create Release
+    needs: build
+    runs-on: ubuntu-latest
+    permissions:
+      contents: write
+
+    steps:
+      - uses: actions/checkout@v4
+        with:
+          fetch-depth: 0
+
+      - uses: actions/download-artifact@v4
+        with:
+          path: artifacts
+
+      - name: Create Release
+        uses: softprops/action-gh-release@v2
+        with:
+          files: artifacts/**/*
+          generate_release_notes: true
+          draft: false
+          prerelease: ${{ contains(github.ref, '-rc') || contains(github.ref, '-beta') || contains(github.ref, '-alpha') }}

.github/workflows/rust.yml

@@ -2,18 +2,23 @@ name: Rust
 
 on:
   push:
-    branches: [ main ]
+    branches: [ "*" ]
   pull_request:
-    branches: [ main ]
+    branches: [ "*" ]
 
 env:
   CARGO_TERM_COLOR: always
 
 jobs:
-  build-and-test:
-    name: Build & Test
+  build:
+    name: Build
     runs-on: ubuntu-latest
 
+    permissions:
+      contents: read
+      actions: write
+      checks: write
+
     steps:
     - name: Checkout repository
       uses: actions/checkout@v4
@@ -37,5 +42,13 @@ jobs:
     - name: Build Release
       run: cargo build --release --verbose
 
+    - name: Run tests
+      run: cargo test --verbose
+
+# clippy dont fail on warnings because of active development of telemt 
+# and many warnings
+    - name: Run clippy
+      run: cargo clippy -- --cap-lints warn
+
     - name: Check for unused dependencies
       run: cargo udeps || true

.gitignore

@@ -19,3 +19,5 @@ target
 #  and can be added to the global gitignore or merged into this file.  For a more nuclear
 #  option (not recommended) you can uncomment the following to ignore the entire idea folder.
 #.idea/
+
+proxy-secret

.kilocode/rules-architect/AGENTS.md

@@ -0,0 +1,58 @@
+# Architect Mode Rules for Telemt
+
+## Architecture Overview
+
+```mermaid
+graph TB
+    subgraph Entry
+        Client[Clients] --> Listener[TCP/Unix Listener]
+    end
+    
+    subgraph Proxy Layer
+        Listener --> ClientHandler[ClientHandler]
+        ClientHandler --> Handshake[Handshake Validator]
+        Handshake --> |Valid| Relay[Relay Layer]
+        Handshake --> |Invalid| Masking[Masking/TLS Fronting]
+    end
+    
+    subgraph Transport
+        Relay --> MiddleProxy[Middle-End Proxy Pool]
+        Relay --> DirectRelay[Direct DC Relay]
+        MiddleProxy --> TelegramDC[Telegram DCs]
+        DirectRelay --> TelegramDC
+    end
+```
+
+## Module Dependencies
+- [`src/main.rs`](src/main.rs) - Entry point, spawns all async tasks
+- [`src/config/`](src/config/) - Configuration loading with auto-migration
+- [`src/error.rs`](src/error.rs) - Error types, must be used by all modules
+- [`src/crypto/`](src/crypto/) - AES, SHA, random number generation
+- [`src/protocol/`](src/protocol/) - MTProto constants, frame encoding, obfuscation
+- [`src/stream/`](src/stream/) - Stream wrappers, buffer pool, frame codecs
+- [`src/proxy/`](src/proxy/) - Client handling, handshake, relay logic
+- [`src/transport/`](src/transport/) - Upstream management, middle-proxy, SOCKS support
+- [`src/stats/`](src/stats/) - Statistics and replay protection
+- [`src/ip_tracker.rs`](src/ip_tracker.rs) - Per-user IP tracking
+
+## Key Architectural Constraints
+
+### Middle-End Proxy Mode
+- Requires public IP on interface OR 1:1 NAT with STUN probing
+- Uses separate `proxy-secret` from Telegram (NOT user secrets)
+- Falls back to direct mode automatically on STUN mismatch
+
+### TLS Fronting
+- Invalid handshakes are transparently proxied to `mask_host`
+- This is critical for DPI evasion - do not change this behavior
+- `mask_unix_sock` and `mask_host` are mutually exclusive
+
+### Stream Architecture
+- Buffer pool is shared globally via Arc - prevents allocation storms
+- Frame codecs implement tokio-util Encoder/Decoder traits
+- State machine in [`src/stream/state.rs`](src/stream/state.rs) manages stream transitions
+
+### Configuration Migration
+- [`ProxyConfig::load()`](src/config/mod.rs:641) mutates config in-place
+- New fields must have sensible defaults
+- DC203 override is auto-injected for CDN/media support

.kilocode/rules-code/AGENTS.md

@@ -0,0 +1,23 @@
+# Code Mode Rules for Telemt
+
+## Error Handling
+- Always use [`ProxyError`](src/error.rs:168) from [`src/error.rs`](src/error.rs) for proxy operations
+- [`HandshakeResult<T,R,W>`](src/error.rs:292) returns streams on bad client - these MUST be returned for masking, never dropped
+- Use [`Recoverable`](src/error.rs:110) trait to check if errors are retryable
+
+## Configuration Changes
+- [`ProxyConfig::load()`](src/config/mod.rs:641) auto-mutates config - new fields should have defaults
+- DC203 override is auto-injected if missing - do not remove this behavior
+- When adding config fields, add migration logic in [`ProxyConfig::load()`](src/config/mod.rs:641)
+
+## Crypto Code
+- [`SecureRandom`](src/crypto/random.rs) from [`src/crypto/random.rs`](src/crypto/random.rs) must be used for all crypto operations
+- Never use `rand::thread_rng()` directly - use the shared `Arc<SecureRandom>`
+
+## Stream Handling
+- Buffer pool [`BufferPool`](src/stream/buffer_pool.rs) is shared via Arc - always use it instead of allocating
+- Frame codecs in [`src/stream/frame_codec.rs`](src/stream/frame_codec.rs) implement tokio-util's Encoder/Decoder traits
+
+## Testing
+- Tests are inline in modules using `#[cfg(test)]`
+- Use `cargo test --lib <module_name>` to run tests for specific modules

.kilocode/rules-debug/AGENTS.md

@@ -0,0 +1,27 @@
+# Debug Mode Rules for Telemt
+
+## Logging
+- `RUST_LOG` environment variable takes absolute priority over all config log levels
+- Log levels: `trace`, `debug`, `info`, `warn`, `error`
+- Use `RUST_LOG=debug cargo run` for detailed operational logs
+- Use `RUST_LOG=trace cargo run` for full protocol-level debugging
+
+## Middle-End Proxy Debugging
+- Set `ME_DIAG=1` environment variable for high-precision cryptography diagnostics
+- STUN probe results are logged at startup - check for mismatch between local and reflected IP
+- If Middle-End fails, check `proxy_secret_path` points to valid file from https://core.telegram.org/getProxySecret
+
+## Connection Issues
+- DC connectivity is logged at startup with RTT measurements
+- If DC ping fails, check `dc_overrides` for custom addresses
+- Use `prefer_ipv6=false` in config if IPv6 is unreliable
+
+## TLS Fronting Issues
+- Invalid handshakes are proxied to `mask_host` - check this host is reachable
+- `mask_unix_sock` and `mask_host` are mutually exclusive - only one can be set
+- If `mask_unix_sock` is set, socket must exist before connections arrive
+
+## Common Errors
+- `ReplayAttack` - client replayed a handshake nonce, potential attack
+- `TimeSkew` - client clock is off, can disable with `ignore_time_skew=true`
+- `TgHandshakeTimeout` - upstream DC connection failed, check network

AGENTS.md

@@ -0,0 +1,410 @@
+## System Prompt — Production Rust Codebase: Modification and Architecture Guidelines
+
+You are a senior Rust Engineer and pricipal Rust Architect acting as a strict code reviewer and implementation partner. 
+Your responses are precise, minimal, and architecturally sound. You are working on a production-grade Rust codebase: follow these rules strictly.
+
+---
+
+### 0. Priority Resolution — Scope Control
+
+This section resolves conflicts between code quality enforcement and scope limitation.
+
+When editing or extending existing code, you MUST audit the affected files and fix:
+
+- Comment style violations (missing, non-English, decorative, trailing).
+- Missing or incorrect documentation on public items.
+- Comment placement issues (trailing comments → move above the code).
+
+These are **coordinated changes** — they are always in scope.
+
+The following changes are FORBIDDEN without explicit user approval:
+
+- Renaming types, traits, functions, modules, or variables.
+- Altering business logic, control flow, or data transformations.
+- Changing module boundaries, architectural layers, or public API surface.
+- Adding or removing functions, structs, enums, or trait implementations.
+- Fixing compiler warnings or removing unused code.
+
+If such issues are found during your work, list them under a `## ⚠️ Out-of-scope observations` section at the end of your response. Include file path, context, and a brief description. Do not apply these changes.
+
+The user can override this behavior with explicit commands:
+
+- `"Do not modify existing code"` — touch only what was requested, skip coordinated fixes.
+- `"Make minimal changes"` — no coordinated fixes, narrowest possible diff.
+- `"Fix everything"` — apply all coordinated fixes and out-of-scope observations.
+
+### Core Rule
+
+The codebase must never enter an invalid intermediate state.
+No response may leave the repository in a condition that requires follow-up fixes.
+
+---
+
+### 1. Comments and Documentation
+
+- All comments MUST be written in English.
+- Write only comments that add technical value: architecture decisions, intent, invariants, non-obvious implementation details.
+- Place all comments on separate lines above the relevant code.
+- Use `///` doc-comments for public items. Use `//` for internal clarifications.
+
+Correct example:
+
+```rust
+// Handles MTProto client authentication and establishes encrypted session state.
+fn handle_authenticated_client(...) { ... }
+```
+
+Incorrect examples:
+
+```rust
+let x = 5; // set x to 5
+```
+
+```rust
+// This function does stuff
+fn do_stuff() { ... }
+```
+
+---
+
+### 2. File Size and Module Structure
+
+- Files MUST NOT exceed 350–550 lines.
+- If a file exceeds this limit, split it into submodules organized by responsibility (e.g., protocol, transport, state, handlers).
+- Parent modules MUST declare and describe their submodules.
+- Maintain clear architectural boundaries between modules.
+
+Correct example:
+
+```rust
+// Client connection handling logic.
+// Submodules:
+// - handshake: MTProto handshake implementation
+// - relay: traffic forwarding logic
+// - state: client session state machine
+
+pub mod handshake;
+pub mod relay;
+pub mod state;
+```
+
+Git discipline:
+
+- Use local git for versioning and diffs.
+- Write clear, descriptive commit messages in English that explain both *what* changed and *why*.
+
+---
+
+### 3. Formatting
+
+- Preserve the existing formatting style of the project exactly as-is.
+- Reformat code only when explicitly instructed to do so.
+- Do not run `cargo fmt` unless explicitly instructed.
+
+---
+
+### 4. Change Safety and Validation
+
+- If anything is unclear, STOP and ask specific, targeted questions before proceeding.
+- List exactly what is ambiguous and offer possible interpretations for the user to choose from.
+- Prefer clarification over assumptions. Do not guess intent, behavior, or missing requirements.
+- Actively ask questions before making architectural or behavioral changes.
+
+---
+
+### 5. Warnings and Unused Code
+
+- Leave all warnings, unused variables, functions, imports, and dead code untouched unless explicitly instructed to modify them.
+- These may be intentional or part of work-in-progress code.
+- `todo!()` and `unimplemented!()` are permitted and should not be removed or replaced unless explicitly instructed.
+
+---
+
+### 6. Architectural Integrity
+
+- Preserve existing architecture unless explicitly instructed to refactor.
+- Do not introduce hidden behavioral changes.
+- Do not introduce implicit refactors.
+- Keep changes minimal, isolated, and intentional.
+
+---
+
+### 7. When Modifying Code
+
+You MUST:
+
+- Maintain architectural consistency with the existing codebase.
+- Document non-obvious logic with comments that describe *why*, not *what*.
+- Limit changes strictly to the requested scope (plus coordinated fixes per Section 0).
+- Keep all existing symbol names unless renaming is explicitly requested.
+- Preserve global formatting as-is
+- Result every modification in a self-contained, compilable, runnable state of the codebase
+
+You MUST NOT:
+
+- Use placeholders: no `// ... rest of code`, no `// implement here`, no `/* TODO */` stubs that replace existing working code. Write full, working implementation. If the implementation is unclear, ask first
+- Refactor code outside the requested scope
+- Make speculative improvements
+- Spawn multiple agents for EDITING
+- Produce partial changes
+- Introduce references to entities that are not yet implemented
+- Leave TODO placeholders in production paths
+
+Note: `todo!()` and `unimplemented!()` are allowed as idiomatic Rust markers for genuinely unfinished code paths.
+
+Every change must:
+   - compile,
+   - pass type checks,
+   - have no broken imports,
+   - preserve invariants,
+   - not rely on future patches.
+
+If the task requires multiple phases:
+   - either implement all required phases,
+   - or explicitly refuse and explain missing dependencies.
+
+---
+
+### 8. Decision Process for Complex Changes
+
+When facing a non-trivial modification, follow this sequence:
+
+1. **Clarify**: Restate the task in one sentence to confirm understanding.
+2. **Assess impact**: Identify which modules, types, and invariants are affected.
+3. **Propose**: Describe the intended change before implementing it.
+4. **Implement**: Make the minimal, isolated change.
+5. **Verify**: Explain why the change preserves existing behavior and architectural integrity.
+
+---
+
+### 9. Context Awareness
+
+- When provided with partial code, assume the rest of the codebase exists and functions correctly unless stated otherwise.
+- Reference existing types, functions, and module structures by their actual names as shown in the provided code.
+- When the provided context is insufficient to make a safe change, request the missing context explicitly.
+- Spawn multiple agents for SEARCHING information, code, functions
+
+---
+
+### 10. Response Format
+
+#### Language Policy
+
+- Code, comments, commit messages, documentation ONLY ON **English**!
+- Reasoning and explanations in response text on language from promt
+
+#### Response Structure
+
+Your response MUST consist of two sections:
+
+**Section 1: `## Reasoning`**
+
+- What needs to be done and why.
+- Which files and modules are affected.
+- Architectural decisions and their rationale.
+- Potential risks or side effects.
+
+**Section 2: `## Changes`**
+
+- For each modified or created file: the filename on a separate line in backticks, followed by the code block.
+- For files **under 200 lines**: return the full file with all changes applied.
+- For files **over 200 lines**: return only the changed functions/blocks with at least 3 lines of surrounding context above and below. If the user requests the full file, provide it.
+- New files: full file content.
+- End with a suggested git commit message in English.
+
+#### Reporting Out-of-Scope Issues
+
+If during modification you discover issues outside the requested scope (potential bugs, unsafe code, architectural concerns, missing error handling, unused imports, dead code):
+
+- Do not fix them silently.
+- List them under `## ⚠️ Out-of-scope observations` at the end of your response.
+- Include: file path, line/function context, brief description of the issue, and severity estimate.
+
+#### Splitting Protocol
+
+If the response exceeds the output limit:
+
+1. End the current part with: **SPLIT: PART N — CONTINUE? (remaining: file_list)**
+2. List the files that will be provided in subsequent parts.
+3. Wait for user confirmation before continuing.
+4. No single file may be split across parts.
+
+## 11. Anti-LLM Degeneration Safeguards (Principal-Paranoid, Visionary)
+
+This section exists to prevent common LLM failure modes: scope creep, semantic drift, cargo-cult refactors, performance regressions, contract breakage, and hidden behavior changes.
+
+### 11.1 Non-Negotiable Invariants
+
+- **No semantic drift:** Do not reinterpret requirements, rename concepts, or change meaning of existing terms.
+- **No “helpful refactors”:** Any refactor not explicitly requested is forbidden.
+- **No architectural drift:** Do not introduce new layers, patterns, abstractions, or “clean architecture” migrations unless requested.
+- **No dependency drift:** Do not add crates, features, or versions unless explicitly requested.
+- **No behavior drift:** If a change could alter runtime behavior, you MUST call it out explicitly in `## Reasoning` and justify it.
+
+### 11.2 Minimal Surface Area Rule
+
+- Touch the smallest number of files possible.
+- Prefer local changes over cross-cutting edits.
+- Do not “align style” across a file/module—only adjust the modified region.
+- Do not reorder items, imports, or code unless required for correctness.
+
+### 11.3 No Implicit Contract Changes
+
+Contracts include:
+- public APIs, trait bounds, visibility, error types, timeouts/retries, logging semantics, metrics semantics,
+- protocol formats, framing, padding, keepalive cadence, state machine transitions,
+- concurrency guarantees, cancellation behavior, backpressure behavior.
+
+Rule:
+- If you change a contract, you MUST update all dependents in the same patch AND document the contract delta explicitly.
+
+### 11.4 Hot-Path Preservation (Performance Paranoia)
+
+- Do not introduce extra allocations, cloning, or formatting in hot paths.
+- Do not add logging/metrics on hot paths unless requested.
+- Do not add new locks or broaden lock scope.
+- Prefer `&str` / slices / borrowed data where the codebase already does so.
+- Avoid `String` building for errors/logs if it changes current patterns.
+
+If you cannot prove performance neutrality, label it as risk in `## Reasoning`.
+
+### 11.5 Async / Concurrency Safety (Cancellation & Backpressure)
+
+- No blocking calls inside async contexts.
+- Preserve cancellation safety: do not introduce `await` between lock acquisition and critical invariants unless already present.
+- Preserve backpressure: do not replace bounded channels with unbounded, do not remove flow control.
+- Do not change task lifecycle semantics (spawn patterns, join handles, shutdown order) unless requested.
+- Do not introduce `tokio::spawn` / background tasks unless explicitly requested.
+
+### 11.6 Error Semantics Integrity
+
+- Do not replace structured errors with generic strings.
+- Do not widen/narrow error types or change error categories without explicit approval.
+- Avoid introducing panics in production paths (`unwrap`, `expect`) unless the codebase already treats that path as impossible and documented.
+
+### 11.7 “No New Abstractions” Default
+
+Default stance:
+- No new traits, generics, macros, builder patterns, type-level cleverness, or “frameworking”.
+- If abstraction is necessary, prefer the smallest possible local helper (private function) and justify it.
+
+### 11.8 Negative-Diff Protection
+
+Avoid “diff inflation” patterns:
+- mass edits,
+- moving code between files,
+- rewrapping long lines,
+- rearranging module order,
+- renaming for aesthetics.
+
+If a diff becomes large, STOP and ask before proceeding.
+
+### 11.9 Consistency with Existing Style (But Not Style Refactors)
+
+- Follow existing conventions of the touched module (naming, error style, return patterns).
+- Do not enforce global “best practices” that the codebase does not already use.
+
+### 11.10 Two-Phase Safety Gate (Plan → Patch)
+
+For non-trivial changes:
+1) Provide a micro-plan (1–5 bullets): what files, what functions, what invariants, what risks.
+2) Implement exactly that plan—no extra improvements.
+
+### 11.11 Pre-Response Checklist (Hard Gate)
+
+Before final output, verify internally:
+
+- No unresolved symbols / broken imports.
+- No partially updated call sites.
+- No new public surface changes unless requested.
+- No transitional states / TODO placeholders replacing working code.
+- Changes are atomic: the repository remains buildable and runnable.
+- Any behavior change is explicitly stated.
+
+If any check fails: fix it before responding.
+
+### 11.12 Truthfulness Policy (No Hallucinated Claims)
+
+- Do not claim “this compiles” or “tests pass” unless you actually verified with the available tooling/context.
+- If verification is not possible, state: “Not executed; reasoning-based consistency check only.”
+
+### 11.13 Visionary Guardrail: Preserve Optionality
+
+When multiple valid designs exist, prefer the one that:
+- minimally constrains future evolution,
+- preserves existing extension points,
+- avoids locking the project into a new paradigm,
+- keeps interfaces stable and implementation local.
+
+Default to reversible changes.
+
+### 11.14 Stop Conditions
+
+STOP and ask targeted questions if:
+- required context is missing,
+- a change would cross module boundaries,
+- a contract might change,
+- concurrency/protocol invariants are unclear,
+- the diff is growing beyond a minimal patch.
+
+No guessing.
+
+### 12. Invariant Preservation
+
+You MUST explicitly preserve:
+- Thread-safety guarantees (`Send` / `Sync` expectations).
+- Memory safety assumptions (no hidden `unsafe` expansions).
+- Lock ordering and deadlock invariants.
+- State machine correctness (no new invalid transitions).
+- Backward compatibility of serialized formats (if applicable).
+
+If a change touches concurrency, networking, protocol logic, or state machines,
+you MUST explain why existing invariants remain valid.
+
+### 13. Error Handling Policy
+
+- Do not replace structured errors with generic strings.
+- Preserve existing error propagation semantics.
+- Do not widen or narrow error types without approval.
+- Avoid introducing panics in production paths.
+- Prefer explicit error mapping over implicit conversions.
+
+### 14. Test Safety
+
+- Do not modify existing tests unless the task explicitly requires it.
+- Do not weaken assertions.
+- Preserve determinism in testable components.
+
+### 15. Security Constraints
+
+- Do not weaken cryptographic assumptions.
+- Do not modify key derivation logic without explicit request.
+- Do not change constant-time behavior.
+- Do not introduce logging of secrets.
+- Preserve TLS/MTProto protocol correctness.
+
+### 16. Logging Policy
+
+- Do not introduce excessive logging in hot paths.
+- Do not log sensitive data.
+- Preserve existing log levels and style.
+
+### 17. Pre-Response Verification Checklist
+
+Before producing the final answer, verify internally:
+
+- The change compiles conceptually.
+- No unresolved symbols exist.
+- All modified call sites are updated.
+- No accidental behavioral changes were introduced.
+- Architectural boundaries remain intact.
+
+### 18. Atomic Change Principle
+Every patch must be **atomic and production-safe**.
+* **Self-contained** — no dependency on future patches or unimplemented components.
+* **Build-safe** — the project must compile successfully after the change.
+* **Contract-consistent** — no partial interface or behavioral changes; all dependent code must be updated within the same patch.
+* **No transitional states** — no placeholders, incomplete refactors, or temporary inconsistencies.
+
+**Invariant:** After any single patch, the repository remains fully functional and buildable.
+

CONTRIBUTING.md

@@ -0,0 +1,19 @@
+# Issues - Rules
+## What it is not
+- NOT Question and Answer
+- NOT Helpdesk
+
+# Pull Requests - Rules
+## General
+- ONLY signed and verified commits
+- ONLY from your name
+- DO NOT commit with `codex` or `claude` as author/commiter
+- PREFER `flow` branch for development, not `main`
+
+## AI
+We are not against modern tools, like AI, where you act as a principal or architect, but we consider it important:
+
+- you really understand what you're doing
+- you understand the relationships and dependencies of the components being modified
+- you understand the architecture of Telegram MTProto, MTProxy, Middle-End KDF at least generically
+- you DO NOT commit for the sake of commits, but to help the community, core-developers and ordinary users

Cargo.toml

@@ -1,16 +1,15 @@
 [package]
 name = "telemt"
-version = "1.0.0"
-edition = "2021"
-rust-version = "1.75"
+version = "3.3.2"
+edition = "2024"
 
 [dependencies]
 # C
 libc = "0.2"
 
 # Async runtime
-tokio = { version = "1.35", features = ["full", "tracing"] }
-tokio-util = { version = "0.7", features = ["codec"] }
+tokio = { version = "1.42", features = ["full", "tracing"] }
+tokio-util = { version = "0.7", features = ["full"] }
 
 # Crypto
 aes = "0.8"
@@ -20,42 +19,57 @@ sha2 = "0.10"
 sha1 = "0.10"
 md-5 = "0.10"
 hmac = "0.12"
-crc32fast = "1.3"
+crc32fast = "1.4"
+crc32c = "0.6"
+zeroize = { version = "1.8", features = ["derive"] }
 
 # Network
 socket2 = { version = "0.5", features = ["all"] }
-rustls = "0.22"
+nix = { version = "0.28", default-features = false, features = ["net"] }
 
-# Serial
+# Serialization
 serde = { version = "1.0", features = ["derive"] }
 serde_json = "1.0"
 toml = "0.8"
+x509-parser = "0.15"
 
 # Utils
-bytes = "1.5"
-thiserror = "1.0"
+bytes = "1.9"
+thiserror = "2.0"
 tracing = "0.1"
 tracing-subscriber = { version = "0.3", features = ["env-filter"] }
 parking_lot = "0.12"
 dashmap = "5.5"
-lru = "0.12"
-rand = "0.8"
+lru = "0.16"
+rand = "0.9"
 chrono = { version = "0.4", features = ["serde"] }
 hex = "0.4"
-base64 = "0.21"
+base64 = "0.22"
 url = "2.5"
-regex = "1.10"
-once_cell = "1.19"
+regex = "1.11"
 crossbeam-queue = "0.3"
+num-bigint = "0.4"
+num-traits = "0.2"
+anyhow = "1.0"
 
 # HTTP
-reqwest = { version = "0.11", features = ["rustls-tls"], default-features = false }
+reqwest = { version = "0.12", features = ["rustls-tls"], default-features = false }
+notify = { version = "6", features = ["macos_fsevent"] }
+ipnetwork = "0.20"
+hyper = { version = "1", features = ["server", "http1"] }
+hyper-util = { version = "0.1", features = ["tokio", "server-auto"] }
+http-body-util = "0.1"
+httpdate = "1.0"
+tokio-rustls = { version = "0.26", default-features = false, features = ["tls12"] }
+rustls = { version = "0.23", default-features = false, features = ["std", "tls12", "ring"] }
+webpki-roots = "0.26"
 
 [dev-dependencies]
 tokio-test = "0.4"
 criterion = "0.5"
 proptest = "1.4"
+futures = "0.3"
 
 [[bench]]
 name = "crypto_bench"
-harness = false
+harness = false

Dockerfile

@@ -0,0 +1,43 @@
+# ==========================
+# Stage 1: Build
+# ==========================
+FROM rust:1.88-slim-bookworm AS builder
+
+RUN apt-get update && apt-get install -y --no-install-recommends \
+    pkg-config \
+    && rm -rf /var/lib/apt/lists/*
+
+WORKDIR /build
+
+COPY Cargo.toml Cargo.lock* ./
+RUN mkdir src && echo 'fn main() {}' > src/main.rs && \
+    cargo build --release 2>/dev/null || true && \
+    rm -rf src
+
+COPY . .
+RUN cargo build --release && strip target/release/telemt
+
+# ==========================
+# Stage 2: Runtime
+# ==========================
+FROM debian:bookworm-slim
+
+RUN apt-get update && apt-get install -y --no-install-recommends \
+    ca-certificates \
+    && rm -rf /var/lib/apt/lists/*
+
+RUN useradd -r -s /usr/sbin/nologin telemt
+
+WORKDIR /app
+
+COPY --from=builder /build/target/release/telemt /app/telemt
+COPY config.toml /app/config.toml
+
+RUN chown -R telemt:telemt /app
+USER telemt
+
+EXPOSE 443
+EXPOSE 9090
+
+ENTRYPOINT ["/app/telemt"]
+CMD ["config.toml"]

LICENSING.md

@@ -0,0 +1,17 @@
+# LICENSING
+## Licenses for Versions
+| Version | License       |
+|---------|---------------|
+| 1.0     | NO LICNESE    |
+| 1.1     | NO LICENSE    |
+| 1.2     | NO LICENSE    |
+| 2.0     | NO LICENSE    |
+| 3.0     | TELEMT UL 1   |
+
+### License Types
+- **NO LICENSE** = ***ALL RIGHT RESERVED***
+- **TELEMT UL1** - work in progress license for source code of `telemt`, which encourages:
+  - fair use, 
+  - contributions, 
+  - distribution, 
+  - but prohibits NOT mentioning the authors

README.md

@@ -1,6 +1,85 @@
 # Telemt - MTProxy on Rust + Tokio
 
-**Telemt** is a fast, secure, and feature-rich server written in Rust: it fully implements the official Telegram proxy algo and adds many production-ready improvements such as connection pooling, replay protection, detailed statistics, masking from "prying" eyes
+***Löst Probleme, bevor andere überhaupt wissen, dass sie existieren*** / ***It solves problems before others even realize they exist***
+
+**Telemt** is a fast, secure, and feature-rich server written in Rust: it fully implements the official Telegram proxy algo and adds many production-ready improvements such as:
+- ME Pool + Reader/Writer + Registry + Refill + Adaptive Floor + Trio-State + Generation Lifecycle
+- [Full-covered API w/ management](https://github.com/telemt/telemt/blob/main/docs/API.md)
+- Anti-Replay on Sliding Window
+- Prometheus-format Metrics
+- TLS-Fronting and TCP-Splicing for masking from "prying" eyes
+
+[**Telemt Chat in Telegram**](https://t.me/telemtrs)
+
+## NEWS and EMERGENCY
+### ✈️ Telemt 3 is released!
+<table>
+<tr>
+<td width="50%" valign="top">
+
+### 🇷🇺 RU
+
+#### Релиз 3.0.15 — 25 февраля
+
+25 февраля мы выпустили версию **3.0.15**
+
+Мы предполагаем, что она станет завершающей версией поколения 3.0 и уже сейчас мы рассматриваем её как **LTS-кандидата** для версии **3.1.0**!
+
+После нескольких дней детального анализа особенностей работы Middle-End мы спроектировали и реализовали продуманный режим **ротации ME Writer**. Данный режим позволяет поддерживать стабильно высокую производительность в long-run сценариях без возникновения ошибок, связанных с некорректной конфигурацией прокси
+
+Будем рады вашему фидбеку и предложениям по улучшению — особенно в части **статистики** и **UX**
+
+Релиз:  
+[3.0.15](https://github.com/telemt/telemt/releases/tag/3.0.15)
+
+---
+
+Если у вас есть компетенции в:
+
+- Асинхронных сетевых приложениях  
+- Анализе трафика  
+- Реверс-инжиниринге  
+- Сетевых расследованиях  
+
+Мы открыты к архитектурным предложениям, идеям и pull requests
+</td>
+<td width="50%" valign="top">
+
+### 🇬🇧 EN
+
+#### Release 3.0.15 — February 25
+
+On February 25, we released version **3.0.15**
+
+We expect this to become the final release of the 3.0 generation and at this point, we already see it as a strong **LTS candidate** for the upcoming **3.1.0** release!
+
+After several days of deep analysis of Middle-End behavior, we designed and implemented a well-engineered **ME Writer rotation mode**. This mode enables sustained high throughput in long-run scenarios while preventing proxy misconfiguration errors
+
+We are looking forward to your feedback and improvement proposals — especially regarding **statistics** and **UX**
+
+Release:  
+[3.0.15](https://github.com/telemt/telemt/releases/tag/3.0.15)
+
+---
+
+If you have expertise in:
+
+- Asynchronous network applications  
+- Traffic analysis  
+- Reverse engineering  
+- Network forensics  
+
+We welcome ideas, architectural feedback, and pull requests.
+</td>
+</tr>
+</table>
+
+# Features
+💥 The configuration structure has changed since version 1.1.0.0. change it in your environment!
+
+⚓ Our implementation of **TLS-fronting** is one of the most deeply debugged, focused, advanced and *almost* **"behaviorally consistent to real"**:  we are confident we have it right - [see evidence on our validation and traces](#recognizability-for-dpi-and-crawler)
+
+⚓ Our ***Middle-End Pool*** is fastest by design in standard scenarios, compared to other implementations of connecting to the Middle-End Proxy: non dramatically, but usual
 
 # GOTO
 - [Features](#features)
@@ -16,10 +95,13 @@
       - [IP](#bind-on-ip)
       - [SOCKS](#socks45-as-upstream)
 - [FAQ](#faq)
+  - [Recognizability for DPI + crawler](#recognizability-for-dpi-and-crawler)
   - [Telegram Calls](#telegram-calls-via-mtproxy)
   - [DPI](#how-does-dpi-see-mtproxy-tls)
   - [Whitelist on Network Level](#whitelist-on-ip)
+  - [Too many open files](#too-many-open-files)
 - [Build](#build)
+- [Docker](#docker)
 - [Why Rust?](#why-rust)
 
 ## Features
@@ -35,127 +117,28 @@
 - Extensive logging via `trace` and `debug` with `RUST_LOG` method
 
 ## Quick Start Guide
-**This software is designed for Debian-based OS: in addition to Debian, these are Ubuntu, Mint, Kali, MX and many other Linux**
-1. Download release
-```bash
-wget https://github.com/telemt/telemt/releases/latest/download/telemt
-```
-2. Move to Bin Folder
-```bash
-mv telemt /bin
-```
-4. Make Executable
-```bash
-chmod +x /bin/telemt
-```
-5. Go to [How to use?](#how-to-use) section for for further steps
 
-## How to use?
-### Telemt via Systemd
-**This instruction "assume" that you:**
-- logged in as root or executed `su -` / `sudo su`
-- you already have an assembled and executable `telemt` in /bin folder as a result of the [Quick Start Guide](#quick-start-guide) or [Build](#build)
+### [Quick Start Guide RU](docs/QUICK_START_GUIDE.ru.md)  
+### [Quick Start Guide EN](docs/QUICK_START_GUIDE.en.md)
 
-**0. Check port and generate secrets**
 
-The port you have selected for use should be MISSING from the list, when:
-```bash
-netstat -lnp
-```
-
-Generate 16 bytes/32 characters HEX with OpenSSL or another way:
-```bash
-openssl rand -hex 16
-```
-OR
-```bash
-xxd -l 16 -p /dev/urandom
-```
-OR
-```bash
-python3 -c 'import os; print(os.urandom(16).hex())'
-```
-
-**1. Place your config to /etc/telemt.toml**
-
-Open nano
-```bash
-nano /etc/telemt.toml
-```
-paste your config from [Configuration](#configuration) section
-
-then Ctrl+X -> Y -> Enter to save
-
-**2. Create service on /etc/systemd/system/telemt.service**
-
-Open nano
-```bash
-nano /etc/systemd/system/telemt.service
-```
-paste this Systemd Module
-```bash
-[Unit]
-Description=Telemt
-After=network.target
-
-[Service]
-Type=simple
-WorkingDirectory=/bin
-ExecStart=/bin/telemt /etc/telemt.toml
-Restart=on-failure
-
-[Install]
-WantedBy=multi-user.target
-```
-then Ctrl+X -> Y -> Enter to save
-
-**3.**  In Shell type `systemctl start telemt` - it must start with zero exit-code
-
-**4.** In Shell type `systemctl status telemt` - there you can reach info about current MTProxy status
-
-**5.** In Shell type `systemctl enable telemt` - then telemt will start with system startup, after the network is up
-
-## Configuration
-### Minimal Configuration for First Start
-```toml
-port = 443                              # Listening port
-show_links = ["tele", "hello"]          # Specify users, for whom will be displayed the links
-
-[users]
-tele = "00000000000000000000000000000000" # Replace the secret with one generated before
-hello = "00000000000000000000000000000000" # Replace the secret with one generated before
-
-[modes]
-classic = false                         # Plain obfuscated mode
-secure = false                          # dd-prefix mode
-tls = true                              # Fake TLS - ee-prefix
-
-tls_domain = "petrovich.ru"             # Domain for ee-secret and masking
-mask = true                             # Enable masking of bad traffic
-mask_host = "petrovich.ru"              # Optional override for mask destination
-mask_port = 443                         # Port for masking
-
-prefer_ipv6 = false                     # Try IPv6 DCs first if true
-fast_mode = true                        # Use "fast" obfuscation variant
-
-client_keepalive = 600                  # Seconds
-client_ack_timeout = 300                # Seconds
-```
 ### Advanced
-#### Adtag
-To use channel advertising and usage statistics from Telegram, get Adtag from [@mtproxybot](https://t.me/mtproxybot), add this parameter to the end of config.toml and specify it
+#### Adtag (per-user)
+To use channel advertising and usage statistics from Telegram, get an Adtag from [@mtproxybot](https://t.me/mtproxybot). Set it per user in `[access.user_ad_tags]` (32 hex chars):
 ```toml
-ad_tag = "00000000000000000000000000000000" # Replace zeros to your adtag from @mtproxybot
+[access.user_ad_tags]
+username1 = "11111111111111111111111111111111"  # Replace with your tag from @mtproxybot
+username2 = "22222222222222222222222222222222"
 ```
 #### Listening and Announce IPs
-To specify listening address and/or address in links, add to the end of config.toml:
+To specify listening address and/or address in links, add to section `[[server.listeners]]` of config.toml:
 ```toml
-[[listeners]]
+[[server.listeners]]
 ip = "0.0.0.0"          # 0.0.0.0 = all IPs; your IP = specific listening
 announce_ip = "1.2.3.4" # IP in links; comment with # if not used
 ```
 #### Upstream Manager
-To specify upstream, add to the end of config.toml:
+To specify upstream, add to section `[[upstreams]]` of config.toml:
 ##### Bind on IP
 ```toml
 [[upstreams]]
@@ -186,6 +169,95 @@ enabled = true
 ```
 
 ## FAQ
+### Recognizability for DPI and crawler
+Since version 1.1.0.0, we have debugged masking perfectly: for all clients without "presenting" a key, 
+we transparently direct traffic to the target host!
+
+- We consider this a breakthrough aspect, which has no stable analogues today
+- Based on this: if `telemt` configured correctly, **TLS mode is completely identical to real-life handshake + communication** with a specified host
+- Here is our evidence:
+    - 212.220.88.77 - "dummy" host, running `telemt`
+    - `petrovich.ru` - `tls` + `masking` host, in HEX: `706574726f766963682e7275`
+    - **No MITM + No Fake Certificates/Crypto** = pure transparent *TCP Splice* to "best" upstream: MTProxy or tls/mask-host:
+      - DPI see legitimate HTTPS to `tls_host`, including *valid chain-of-trust* and entropy
+      - Crawlers completely satisfied receiving responses from `mask_host`
+  #### Client WITH secret-key accesses the MTProxy resource:
+  
+  <img width="360" height="439" alt="telemt" src="https://github.com/user-attachments/assets/39352afb-4a11-4ecc-9d91-9e8cfb20607d" />
+  
+  #### Client WITHOUT secret-key gets transparent access to the specified resource:
+    - with trusted certificate
+    - with original handshake
+    - with full request-response way
+    - with low-latency overhead
+```bash
+root@debian:~/telemt# curl -v -I --resolve petrovich.ru:443:212.220.88.77 https://petrovich.ru/
+* Added petrovich.ru:443:212.220.88.77 to DNS cache
+* Hostname petrovich.ru was found in DNS cache
+*   Trying 212.220.88.77:443...
+* Connected to petrovich.ru (212.220.88.77) port 443 (#0)
+* ALPN: offers h2,http/1.1
+* TLSv1.3 (OUT), TLS handshake, Client hello (1):
+*  CAfile: /etc/ssl/certs/ca-certificates.crt
+*  CApath: /etc/ssl/certs
+* TLSv1.3 (IN), TLS handshake, Server hello (2):
+* TLSv1.3 (IN), TLS handshake, Encrypted Extensions (8):
+* TLSv1.3 (IN), TLS handshake, Certificate (11):
+* TLSv1.3 (IN), TLS handshake, CERT verify (15):
+* TLSv1.3 (IN), TLS handshake, Finished (20):
+* TLSv1.3 (OUT), TLS change cipher, Change cipher spec (1):
+* TLSv1.3 (OUT), TLS handshake, Finished (20):
+* SSL connection using TLSv1.3 / TLS_AES_256_GCM_SHA384
+* ALPN: server did not agree on a protocol. Uses default.
+* Server certificate:
+*  subject: C=RU; ST=Saint Petersburg; L=Saint Petersburg; O=STD Petrovich; CN=*.petrovich.ru
+*  start date: Jan 28 11:21:01 2025 GMT
+*  expire date: Mar  1 11:21:00 2026 GMT
+*  subjectAltName: host "petrovich.ru" matched cert's "petrovich.ru"
+*  issuer: C=BE; O=GlobalSign nv-sa; CN=GlobalSign RSA OV SSL CA 2018
+*  SSL certificate verify ok.
+* using HTTP/1.x
+> HEAD / HTTP/1.1
+> Host: petrovich.ru
+> User-Agent: curl/7.88.1
+> Accept: */*
+> 
+* TLSv1.3 (IN), TLS handshake, Newsession Ticket (4):
+* TLSv1.3 (IN), TLS handshake, Newsession Ticket (4):
+* old SSL session ID is stale, removing
+< HTTP/1.1 200 OK
+HTTP/1.1 200 OK
+< Server: Variti/0.9.3a
+Server: Variti/0.9.3a
+< Date: Thu, 01 Jan 2026 00:0000 GMT
+Date: Thu, 01 Jan 2026 00:0000 GMT
+< Access-Control-Allow-Origin: *
+Access-Control-Allow-Origin: *
+< Content-Type: text/html
+Content-Type: text/html
+< Cache-Control: no-store
+Cache-Control: no-store
+< Expires: Thu, 01 Jan 2026 00:0000 GMT
+Expires: Thu, 01 Jan 2026 00:0000 GMT
+< Pragma: no-cache
+Pragma: no-cache
+< Set-Cookie: ipp_uid=XXXXX/XXXXX/XXXXX==; Expires=Tue, 31 Dec 2040 23:59:59 GMT; Domain=.petrovich.ru; Path=/
+Set-Cookie: ipp_uid=XXXXX/XXXXX/XXXXX==; Expires=Tue, 31 Dec 2040 23:59:59 GMT; Domain=.petrovich.ru; Path=/
+< Content-Type: text/html
+Content-Type: text/html
+< Content-Length: 31253
+Content-Length: 31253
+< Connection: keep-alive
+Connection: keep-alive
+< Keep-Alive: timeout=60
+Keep-Alive: timeout=60
+
+< 
+* Connection #0 to host petrovich.ru left intact
+
+```
+- We challenged ourselves, we kept trying and we didn't only *beat the air*: now, we have something to show you
+  - Do not just take our word for it? - This is great and we respect that: you can build your own `telemt` or download a build and check it right now
 ### Telegram Calls via MTProxy
 - Telegram architecture **does NOT allow calls via MTProxy**, but only via SOCKS5, which cannot be obfuscated
 ### How does DPI see MTProxy TLS?
@@ -206,6 +278,23 @@ enabled = true
   - in China behind the Great Firewall
   - in Russia on mobile networks, less in wired networks
   - in Iran during "activity"
+### Too many open files
+- On a fresh Linux install the default open file limit is low; under load `telemt` may fail with `Accept error: Too many open files`
+- **Systemd**: add `LimitNOFILE=65536` to the `[Service]` section (already included in the example above)
+- **Docker**: add `--ulimit nofile=65536:65536` to your `docker run` command, or in `docker-compose.yml`:
+```yaml
+ulimits:
+  nofile:
+    soft: 65536
+    hard: 65536
+```
+- **System-wide** (optional): add to `/etc/security/limits.conf`:
+```
+*       soft    nofile  1048576
+*       hard    nofile  1048576
+root    soft    nofile  1048576
+root    hard    nofile  1048576
+```
 
 
 ## Build
@@ -224,13 +313,52 @@ chmod +x /bin/telemt
 telemt config.toml
 ```
 
+## Docker
+**Quick start (Docker Compose)**
+
+1. Edit `config.toml` in repo root (at least: port, users secrets, tls_domain)
+2. Start container:
+```bash
+docker compose up -d --build
+```
+3. Check logs:
+```bash
+docker compose logs -f telemt
+```
+4. Stop:
+```bash
+docker compose down
+```
+
+**Notes**
+- `docker-compose.yml` maps `./config.toml` to `/app/config.toml` (read-only)
+- By default it publishes `443:443` and runs with dropped capabilities (only `NET_BIND_SERVICE` is added)
+- If you really need host networking (usually only for some IPv6 setups) uncomment `network_mode: host`
+
+**Run without Compose**
+```bash
+docker build -t telemt:local .
+docker run --name telemt --restart unless-stopped \
+  -p 443:443 \
+  -e RUST_LOG=info \
+  -v "$PWD/config.toml:/app/config.toml:ro" \
+  --read-only \
+  --cap-drop ALL --cap-add NET_BIND_SERVICE \
+  --ulimit nofile=65536:65536 \
+  telemt:local
+```
+
 ## Why Rust?
 - Long-running reliability and idempotent behavior
-- Rust’s deterministic resource management - RAII 
+- Rust's deterministic resource management - RAII 
 - No garbage collector
 - Memory safety and reduced attack surface
 - Tokio's asynchronous architecture
 
+## Issues
+- ✅ [SOCKS5 as Upstream](https://github.com/telemt/telemt/issues/1) -> added Upstream Management
+- ✅ [iOS - Media Upload Hanging-in-Loop](https://github.com/telemt/telemt/issues/2)
+
 ## Roadmap
 - Public IP in links
 - Config Reload-on-fly

ROADMAP.md

@@ -0,0 +1,34 @@
+### 3.0.0 Anschluss
+- **Middle Proxy now is stable**, confirmed on canary-deploy over ~20 users
+- Ad-tag now is working
+- DC=203/CDN now is working over ME
+- `getProxyConfig` and `ProxySecret` are automated
+- Version order is now in format `3.0.0` - without Windows-style "microfixes"
+
+### 3.0.1 Kabelsammler
+- Handshake timeouts fixed
+- Connectivity logging refactored
+- Docker: tmpfs for ProxyConfig and ProxySecret
+- Public Host and Port in config
+- ME Relays Head-of-Line Blocking fixed
+- ME Ping
+
+### 3.0.2 Microtrencher
+- New [network] section
+- ME Fixes
+- Small bugs coverage
+
+### 3.0.3 Ausrutscher
+- ME as stateful, no conn-id migration
+- No `flush()` on datapath after RpcWriter
+- Hightech parser for IPv6 without regexp
+- `nat_probe = true` by default
+- Timeout for `recv()` in STUN-client
+- ConnRegistry review
+- Dualstack emergency reconnect
+
+### 3.0.4 Schneeflecken
+- Only WARN and Links in Normal log
+- Consistent IP-family detection
+- Includes for config
+- `nonce_frame_hex` in log only with `DEBUG`

config.full.toml

@@ -0,0 +1,697 @@
+# ==============================================================================
+#
+#   TELEMT — Advanced Rust-based Telegram MTProto Proxy
+#   Full Configuration Reference
+#
+#   This file is both a working config and a complete documentation.
+#   Every parameter is explained. Read it top to bottom before deploying.
+#
+#   Quick Start:
+#     1. Set [server].port to your desired port (443 recommended)
+#     2. Generate a secret: openssl rand -hex 16
+#     3. Put it in [access.users] under a name you choose
+#     4. Set [censorship].tls_domain to a popular unblocked HTTPS site
+#     5. Set your public IP in [general].middle_proxy_nat_ip
+#        and [general.links].public_host
+#     6. Set announce IP in [[server.listeners]]
+#     7. Run Telemt. It prints a tg:// link. Send it to your users.
+#
+#   Modes of Operation:
+#     Direct Mode  (use_middle_proxy = false)
+#       Connects straight to Telegram DCs via TCP. Simple, fast, low overhead.
+#       No ad_tag support. No CDN DC support (203, etc).
+#
+#     Middle-Proxy Mode  (use_middle_proxy = true)
+#       Connects to Telegram Middle-End servers via RPC protocol.
+#       Required for ad_tag monetization and CDN support.
+#       Requires proxy_secret_path and a valid public IP.
+#
+# ==============================================================================
+
+
+# ==============================================================================
+# LEGACY TOP-LEVEL FIELDS
+# ==============================================================================
+
+# Deprecated. Use [general.links].show instead.
+# Accepts "*" for all users, or an array like ["alice", "bob"].
+show_link = ["0"]
+
+# Fallback Datacenter index (1-5) when a client requests an unknown DC ID.
+# DC 2 is Amsterdam (Europe), closest for most CIS users.
+# default_dc = 2
+
+
+# ==============================================================================
+# GENERAL SETTINGS
+# ==============================================================================
+
+[general]
+
+# ------------------------------------------------------------------------------
+# Core Protocol
+# ------------------------------------------------------------------------------
+
+# Coalesce the MTProto handshake and first data payload into a single TCP packet.
+# Significantly reduces connection latency. No reason to disable.
+fast_mode = true
+
+# How the proxy connects to Telegram servers.
+# false = Direct TCP to Telegram DCs (simple, low overhead)
+# true  = Middle-End RPC protocol (required for ad_tag and CDN DCs)
+use_middle_proxy = true
+
+# 32-char hex Ad-Tag from @MTProxybot for sponsored channel injection.
+# Only works when use_middle_proxy = true.
+# Obtain yours: message @MTProxybot on Telegram, register your proxy.
+# ad_tag = "00000000000000000000000000000000"
+
+# ------------------------------------------------------------------------------
+# Middle-End Authentication
+# ------------------------------------------------------------------------------
+
+# Path to the Telegram infrastructure AES key file.
+# Auto-downloaded from https://core.telegram.org/getProxySecret on first run.
+# This key authenticates your proxy with Middle-End servers.
+proxy_secret_path = "proxy-secret"
+
+# ------------------------------------------------------------------------------
+# Public IP Configuration (Critical for Middle-Proxy Mode)
+# ------------------------------------------------------------------------------
+
+# Your server's PUBLIC IPv4 address.
+# Middle-End servers need this for the cryptographic Key Derivation Function.
+# If your server has a direct public IP, set it here.
+# If behind NAT (AWS, Docker, etc.), this MUST be your external IP.
+# If omitted, Telemt uses STUN to auto-detect (see middle_proxy_nat_probe).
+# middle_proxy_nat_ip = "203.0.113.10"
+
+# Auto-detect public IP via STUN servers defined in [network].
+# Set to false if you hardcoded middle_proxy_nat_ip above.
+# Set to true if you want automatic detection.
+middle_proxy_nat_probe = true
+
+# ------------------------------------------------------------------------------
+# Middle-End Connection Pool
+# ------------------------------------------------------------------------------
+
+# Number of persistent multiplexed RPC connections to ME servers.
+# All client traffic is routed through these "fat pipes".
+# 8 handles thousands of concurrent users comfortably.
+middle_proxy_pool_size = 8
+
+# Legacy field. Connections kept initialized but idle as warm standby.
+middle_proxy_warm_standby = 16
+
+# ------------------------------------------------------------------------------
+# Middle-End Keepalive
+# Telegram ME servers aggressively kill idle TCP connections.
+# These settings send periodic RPC_PING frames to keep pipes alive.
+# ------------------------------------------------------------------------------
+
+me_keepalive_enabled = true
+
+# Base interval between pings in seconds.
+me_keepalive_interval_secs = 25
+
+# Random jitter added to interval to prevent all connections pinging simultaneously.
+me_keepalive_jitter_secs = 5
+
+# Randomize ping payload bytes to prevent DPI from fingerprinting ping patterns.
+me_keepalive_payload_random = true
+
+# ------------------------------------------------------------------------------
+# Client-Side Limits
+# ------------------------------------------------------------------------------
+
+# Max buffered ciphertext per client (bytes) when upstream is slow.
+# Acts as backpressure to prevent memory exhaustion. 256KB is safe.
+crypto_pending_buffer = 262144
+
+# Maximum single MTProto frame size from client. 16MB is protocol standard.
+max_client_frame = 16777216
+
+# ------------------------------------------------------------------------------
+# Crypto Desynchronization Logging
+# Desync errors usually mean DPI/GFW is tampering with connections.
+# ------------------------------------------------------------------------------
+
+# true  = full forensics (trace ID, IP hash, hex dumps) for EVERY desync event
+# false = deduplicated logging, one entry per time window (prevents log spam)
+# Set true if you are actively debugging DPI interference.
+desync_all_full = true
+
+# ------------------------------------------------------------------------------
+# Beobachten — Built-in Honeypot / Active Probe Tracker
+# Tracks IPs that fail handshakes or behave like TLS scanners.
+# Output file can be fed into fail2ban or iptables for auto-blocking.
+# ------------------------------------------------------------------------------
+
+beobachten = true
+
+# How long (minutes) to remember a suspicious IP before expiring it.
+beobachten_minutes = 30
+
+# How often (seconds) to flush tracker state to disk.
+beobachten_flush_secs = 15
+
+# File path for the tracker output.
+beobachten_file = "cache/beobachten.txt"
+
+# ------------------------------------------------------------------------------
+# Hardswap — Zero-Downtime ME Pool Rotation
+# When Telegram updates ME server IPs, Hardswap creates a completely new pool,
+# waits until it is fully ready, migrates traffic, then kills the old pool.
+# Users experience zero interruption.
+# ------------------------------------------------------------------------------
+
+hardswap = true
+
+# ------------------------------------------------------------------------------
+# ME Pool Warmup Staggering
+# When creating a new pool, connections are opened one by one with delays
+# to avoid a burst of SYN packets that could trigger ISP flood protection.
+# ------------------------------------------------------------------------------
+
+me_warmup_stagger_enabled = true
+
+# Delay between each connection creation (milliseconds).
+me_warmup_step_delay_ms = 500
+
+# Random jitter added to the delay (milliseconds).
+me_warmup_step_jitter_ms = 300
+
+# ------------------------------------------------------------------------------
+# ME Reconnect Backoff
+# If an ME server drops the connection, Telemt retries with this strategy.
+# ------------------------------------------------------------------------------
+
+# Max simultaneous reconnect attempts per DC.
+me_reconnect_max_concurrent_per_dc = 8
+
+# Exponential backoff base (milliseconds).
+me_reconnect_backoff_base_ms = 500
+
+# Backoff ceiling (milliseconds). Will never wait longer than this.
+me_reconnect_backoff_cap_ms = 30000
+
+# Number of instant retries before switching to exponential backoff.
+me_reconnect_fast_retry_count = 12
+
+# ------------------------------------------------------------------------------
+# NAT Mismatch Behavior
+# If STUN-detected IP differs from local interface IP (you are behind NAT).
+# false = abort ME mode (safe default)
+# true  = force ME mode anyway (use if you know your NAT setup is correct)
+# ------------------------------------------------------------------------------
+
+stun_iface_mismatch_ignore = false
+
+# ------------------------------------------------------------------------------
+# Logging
+# ------------------------------------------------------------------------------
+
+# File to log unknown DC requests (DC IDs outside standard 1-5).
+unknown_dc_log_path = "unknown-dc.txt"
+
+# Verbosity: "debug" | "verbose" | "normal" | "silent"
+log_level = "normal"
+
+# Disable ANSI color codes in log output (useful for file logging).
+disable_colors = false
+
+# ------------------------------------------------------------------------------
+# FakeTLS Record Sizing
+# Buffer small MTProto packets into larger TLS records to mimic real HTTPS.
+# Real HTTPS servers send records close to MTU size (~1400 bytes).
+# A stream of tiny TLS records is a strong DPI signal.
+# Set to 0 to disable. Set to 1400 for realistic HTTPS emulation.
+# ------------------------------------------------------------------------------
+
+fast_mode_min_tls_record = 1400
+
+# ------------------------------------------------------------------------------
+# Periodic Updates
+# ------------------------------------------------------------------------------
+
+# How often (seconds) to re-fetch ME server lists and proxy secrets
+# from core.telegram.org. Keeps your proxy in sync with Telegram infrastructure.
+update_every = 300
+
+# How often (seconds) to force a Hardswap even if the ME map is unchanged.
+# Shorter intervals mean shorter-lived TCP flows, harder for DPI to profile.
+me_reinit_every_secs = 600
+
+# ------------------------------------------------------------------------------
+# Hardswap Warmup Tuning
+# Fine-grained control over how the new pool is warmed up before traffic switch.
+# ------------------------------------------------------------------------------
+
+me_hardswap_warmup_delay_min_ms = 1000
+me_hardswap_warmup_delay_max_ms = 2000
+me_hardswap_warmup_extra_passes = 3
+me_hardswap_warmup_pass_backoff_base_ms = 500
+
+# ------------------------------------------------------------------------------
+# Config Update Debouncing
+# Telegram sometimes pushes transient/broken configs. Debouncing requires
+# N consecutive identical fetches before applying a change.
+# ------------------------------------------------------------------------------
+
+# ME server list must be identical for this many fetches before applying.
+me_config_stable_snapshots = 2
+
+# Minimum seconds between config applications.
+me_config_apply_cooldown_secs = 300
+
+# Proxy secret must be identical for this many fetches before applying.
+proxy_secret_stable_snapshots = 2
+
+# ------------------------------------------------------------------------------
+# Proxy Secret Rotation
+# ------------------------------------------------------------------------------
+
+# Apply newly downloaded secrets at runtime without restart.
+proxy_secret_rotate_runtime = true
+
+# Maximum acceptable secret length (bytes). Rejects abnormally large secrets.
+proxy_secret_len_max = 256
+
+# ------------------------------------------------------------------------------
+# Hardswap Drain Settings
+# Controls graceful shutdown of old ME connections during pool rotation.
+# ------------------------------------------------------------------------------
+
+# Seconds to keep old connections alive for in-flight data before force-closing.
+me_pool_drain_ttl_secs = 90
+
+# Minimum ratio of healthy connections in new pool before draining old pool.
+# 0.8 = at least 80% of new pool must be ready.
+me_pool_min_fresh_ratio = 0.8
+
+# Maximum seconds to wait for drain to complete before force-killing.
+me_reinit_drain_timeout_secs = 120
+
+# ------------------------------------------------------------------------------
+# NTP Clock Check
+# MTProto uses timestamps. Clock drift > 30 seconds breaks handshakes.
+# Telemt checks on startup and warns if out of sync.
+# ------------------------------------------------------------------------------
+
+ntp_check = true
+ntp_servers = ["pool.ntp.org"]
+
+# ------------------------------------------------------------------------------
+# Auto-Degradation
+# If ME servers become completely unreachable (ISP blocking),
+# automatically fall back to Direct Mode so users stay connected.
+# ------------------------------------------------------------------------------
+
+auto_degradation_enabled = true
+
+# Number of DC groups that must be unreachable before triggering fallback.
+degradation_min_unavailable_dc_groups = 2
+
+
+# ==============================================================================
+# ALLOWED CLIENT PROTOCOLS
+# Only enable what you need. In censored regions, TLS-only is safest.
+# ==============================================================================
+
+[general.modes]
+
+# Classic MTProto. Unobfuscated length prefixes. Trivially detected by DPI.
+# No reason to enable unless you have ancient clients.
+classic = false
+
+# Obfuscated MTProto with randomized padding. Better than classic, but
+# still detectable by statistical analysis of packet sizes.
+secure = false
+
+# FakeTLS (ee-secrets). Wraps MTProto in TLS 1.3 framing.
+# To DPI, it looks like a normal HTTPS connection.
+# This should be the ONLY enabled mode in censored environments.
+tls = true
+
+
+# ==============================================================================
+# STARTUP LINK GENERATION
+# Controls what tg:// invite links are printed to console on startup.
+# ==============================================================================
+
+[general.links]
+
+# Which users to generate links for.
+# "*" = all users, or an array like ["alice", "bob"].
+show = "*"
+
+# IP or domain to embed in the tg:// link.
+# If omitted, Telemt uses STUN to auto-detect.
+# Set this to your server's public IP or domain for reliable links.
+# public_host = "proxy.example.com"
+
+# Port to embed in the tg:// link.
+# If omitted, uses [server].port.
+# public_port = 443
+
+
+# ==============================================================================
+# NETWORK & IP RESOLUTION
+# ==============================================================================
+
+[network]
+
+# Enable IPv4 for outbound connections to Telegram.
+ipv4 = true
+
+# Enable IPv6 for outbound connections to Telegram.
+ipv6 = false
+
+# Prefer IPv4 (4) or IPv6 (6) when both are available.
+prefer = 4
+
+# Experimental: use both IPv4 and IPv6 ME servers simultaneously.
+# May improve reliability but doubles connection count.
+multipath = false
+
+# STUN servers for external IP discovery.
+# Used for Middle-Proxy KDF (if nat_probe=true) and link generation.
+stun_servers = [
+    "stun.l.google.com:5349",
+    "stun1.l.google.com:3478",
+    "stun.gmx.net:3478",
+    "stun.l.google.com:19302"
+]
+
+# If UDP STUN is blocked, attempt TCP-based STUN as fallback.
+stun_tcp_fallback = true
+
+# If all STUN fails, use HTTP APIs to discover public IP.
+http_ip_detect_urls = [
+    "https://ifconfig.me/ip",
+    "https://api.ipify.org"
+]
+
+# Cache discovered public IP to this file to survive restarts.
+cache_public_ip_path = "cache/public_ip.txt"
+
+
+# ==============================================================================
+# SERVER BINDING & METRICS
+# ==============================================================================
+
+[server]
+
+# TCP port to listen on.
+# 443 is recommended (looks like normal HTTPS traffic).
+port = 443
+
+# IPv4 bind address. "0.0.0.0" = all interfaces.
+listen_addr_ipv4 = "0.0.0.0"
+
+# IPv6 bind address. "::" = all interfaces.
+listen_addr_ipv6 = "::"
+
+# Unix socket listener (for reverse proxy setups with Nginx/HAProxy).
+# listen_unix_sock = "/var/run/telemt.sock"
+# listen_unix_sock_perm = "0660"
+
+# Enable PROXY protocol header parsing.
+# Set true ONLY if Telemt is behind HAProxy/Nginx that injects PROXY headers.
+# If enabled without a proxy in front, clients will fail to connect.
+proxy_protocol = false
+
+# Prometheus metrics HTTP endpoint port.
+# Uncomment to enable. Access at http://your-server:9090/metrics
+# metrics_port = 9090
+
+# IP ranges allowed to access the metrics endpoint.
+metrics_whitelist = [
+    "127.0.0.1/32",
+    "::1/128"
+]
+
+# ------------------------------------------------------------------------------
+# Listener Overrides
+# Define explicit listeners with specific bind IPs and announce IPs.
+# The announce IP is what gets embedded in tg:// links and sent to ME servers.
+# You MUST set announce to your server's public IP for ME mode to work.
+# ------------------------------------------------------------------------------
+
+# [[server.listeners]]
+# ip = "0.0.0.0"
+# announce = "203.0.113.10"
+# reuse_allow = false
+
+
+# ==============================================================================
+# TIMEOUTS (seconds unless noted)
+# ==============================================================================
+
+[timeouts]
+
+# Maximum time for client to complete FakeTLS + MTProto handshake.
+client_handshake = 15
+
+# Maximum time to establish TCP connection to upstream Telegram DC.
+tg_connect = 10
+
+# TCP keepalive interval for client connections.
+client_keepalive = 60
+
+# Maximum client inactivity before dropping the connection.
+client_ack = 300
+
+# Instant retry count for a single ME endpoint before giving up on it.
+me_one_retry = 3
+
+# Timeout (milliseconds) for a single ME endpoint connection attempt.
+me_one_timeout_ms = 1500
+
+
+# ==============================================================================
+# ANTI-CENSORSHIP / FAKETLS / MASKING
+# This is where Telemt becomes invisible to Deep Packet Inspection.
+# ==============================================================================
+
+[censorship]
+
+# ------------------------------------------------------------------------------
+# TLS Domain Fronting
+# The SNI (Server Name Indication) your proxy presents to connecting clients.
+# Must be a popular, unblocked HTTPS website in your target country.
+# DPI sees traffic to this domain. Choose carefully.
+# Good choices: major CDNs, banks, government sites, search engines.
+# Bad choices: obscure sites, already-blocked domains.
+# ------------------------------------------------------------------------------
+
+tls_domain = "www.google.com"
+
+# ------------------------------------------------------------------------------
+# Active Probe Masking
+# When someone connects but fails the MTProto handshake (wrong secret),
+# they might be an ISP active prober testing if this is a proxy.
+#
+# mask = false: drop the connection (prober knows something is here)
+# mask = true:  transparently proxy them to mask_host (prober sees a real website)
+#
+# With mask enabled, your server is indistinguishable from a real web server
+# to anyone who doesn't have the correct secret.
+# ------------------------------------------------------------------------------
+
+mask = true
+
+# The real web server to forward failed handshakes to.
+# If omitted, defaults to tls_domain.
+# mask_host = "www.google.com"
+
+# Port on the mask host to connect to.
+mask_port = 443
+
+# Inject PROXY protocol header when forwarding to mask host.
+# 0 = disabled, 1 = v1, 2 = v2. Leave disabled unless mask_host expects it.
+# mask_proxy_protocol = 0
+
+# ------------------------------------------------------------------------------
+# TLS Certificate Emulation
+# ------------------------------------------------------------------------------
+
+# Size (bytes) of the locally generated fake TLS certificate.
+# Only used when tls_emulation is disabled.
+fake_cert_len = 2048
+
+# KILLER FEATURE: Real-Time TLS Emulation.
+# Telemt connects to tls_domain, fetches its actual TLS 1.3 certificate chain,
+# and exactly replicates the byte sizes of ServerHello and Certificate records.
+# Defeats DPI that uses TLS record length heuristics to detect proxies.
+# Strongly recommended in censored environments.
+tls_emulation = true
+
+# Directory to cache fetched TLS certificates.
+tls_front_dir = "tlsfront"
+
+# ------------------------------------------------------------------------------
+# ServerHello Timing
+# Real web servers take 30-150ms to respond to ClientHello due to network
+# latency and crypto processing. A proxy responding in <1ms is suspicious.
+# These settings add realistic delay to mimic genuine server behavior.
+# ------------------------------------------------------------------------------
+
+# Minimum delay before sending ServerHello (milliseconds).
+server_hello_delay_min_ms = 50
+
+# Maximum delay before sending ServerHello (milliseconds).
+server_hello_delay_max_ms = 150
+
+# ------------------------------------------------------------------------------
+# TLS Session Tickets
+# Real TLS 1.3 servers send 1-2 NewSessionTicket messages after handshake.
+# A server that sends zero tickets is anomalous and may trigger DPI flags.
+# Set this to match your tls_domain's behavior (usually 2).
+# ------------------------------------------------------------------------------
+
+# tls_new_session_tickets = 0
+
+# ------------------------------------------------------------------------------
+# Full Certificate Frequency
+# When tls_emulation is enabled, this controls how often (per client IP)
+# to send the complete emulated certificate chain.
+#
+# > 0: Subsequent connections within TTL seconds get a smaller cached version.
+#       Saves bandwidth but creates a detectable size difference between
+#       first and repeat connections.
+#
+# = 0: Every connection gets the full certificate. More bandwidth but
+#       perfectly consistent behavior, no anomalies for DPI to detect.
+# ------------------------------------------------------------------------------
+
+tls_full_cert_ttl_secs = 0
+
+# ------------------------------------------------------------------------------
+# ALPN Enforcement
+# Ensure ServerHello responds with the exact ALPN protocol the client requested.
+# Mismatched ALPN (e.g., client asks h2, server says http/1.1) is a DPI red flag.
+# ------------------------------------------------------------------------------
+
+alpn_enforce = true
+
+
+# ==============================================================================
+# ACCESS CONTROL & USERS
+# ==============================================================================
+
+[access]
+
+# ------------------------------------------------------------------------------
+# Replay Attack Protection
+# DPI can record a legitimate user's handshake and replay it later to probe
+# whether the server is a proxy. Telemt remembers recent handshake nonces
+# and rejects duplicates.
+# ------------------------------------------------------------------------------
+
+# Number of nonce slots in the replay detection buffer.
+replay_check_len = 65536
+
+# How long (seconds) to remember nonces before expiring them.
+replay_window_secs = 1800
+
+# Allow clients with incorrect system clocks to connect.
+# false = reject clients with significant time skew (more secure)
+# true  = accept anyone regardless of clock (more permissive)
+ignore_time_skew = false
+
+# ------------------------------------------------------------------------------
+# User Secrets
+# Each user needs a unique 32-character hex string as their secret.
+# Generate with: openssl rand -hex 16
+#
+# This secret is embedded in the tg:// link. Anyone with it can connect.
+# Format: username = "hex_secret"
+# ------------------------------------------------------------------------------
+
+[access.users]
+# alice = "0123456789abcdef0123456789abcdef"
+# bob = "fedcba9876543210fedcba9876543210"
+
+# ------------------------------------------------------------------------------
+# Per-User Connection Limits
+# Limits concurrent TCP connections per user to prevent secret sharing.
+# Uncomment and set for each user as needed.
+# ------------------------------------------------------------------------------
+
+[access.user_max_tcp_conns]
+# alice = 100
+# bob = 50
+
+# ------------------------------------------------------------------------------
+# Per-User Expiration Dates
+# Automatically revoke access after the specified date (ISO 8601 format).
+# ------------------------------------------------------------------------------
+
+[access.user_expirations]
+# alice = "2025-12-31T23:59:59Z"
+# bob = "2026-06-15T00:00:00Z"
+
+# ------------------------------------------------------------------------------
+# Per-User Data Quotas
+# Maximum total bytes transferred per user. Connection refused after limit.
+# ------------------------------------------------------------------------------
+
+[access.user_data_quota]
+# alice = 107374182400
+# bob = 53687091200
+
+# ------------------------------------------------------------------------------
+# Per-User Unique IP Limits
+# Maximum number of different IP addresses that can use this secret
+# at the same time. Highly effective against secret leaking/sharing.
+# Set to 1 for single-device, 2-3 for phone+desktop, etc.
+# ------------------------------------------------------------------------------
+
+[access.user_max_unique_ips]
+# alice = 3
+# bob = 2
+
+
+# ==============================================================================
+# UPSTREAM ROUTING
+# Controls how Telemt connects to Telegram servers (or ME servers).
+# If omitted entirely, uses the OS default route.
+# ==============================================================================
+
+# ------------------------------------------------------------------------------
+# Direct upstream: use the server's own network interface.
+# You can optionally bind to a specific interface or local IP.
+# ------------------------------------------------------------------------------
+
+# [[upstreams]]
+# type = "direct"
+# interface = "eth0"
+# bind_addresses = ["192.0.2.10"]
+# weight = 1
+# enabled = true
+# scopes = "*"
+
+# ------------------------------------------------------------------------------
+# SOCKS5 upstream: route Telegram traffic through a SOCKS5 proxy.
+# Useful if your server's IP is blocked from reaching Telegram DCs.
+# ------------------------------------------------------------------------------
+
+# [[upstreams]]
+# type = "socks5"
+# address = "198.51.100.30:1080"
+# username = "proxy-user"
+# password = "proxy-pass"
+# weight = 1
+# enabled = true
+
+
+# ==============================================================================
+# DATACENTER OVERRIDES
+# Force specific DC IDs to route to specific IP:Port combinations.
+# DC 203 (CDN) is auto-injected by Telemt if not specified here.
+# ==============================================================================
+
+# [dc_overrides]
+# "201" = "149.154.175.50:443"
+# "202" = ["149.154.167.51:443", "149.154.175.100:443"]

config.toml

@@ -1,13 +1,57 @@
-port = 443
+### Telemt Based Config.toml
+# We believe that these settings are sufficient for most scenarios 
+# where cutting-egde methods and parameters or special solutions are not needed
 
-[users]
-user1 = "00000000000000000000000000000000"
+# === General Settings ===
+[general]
+use_middle_proxy = false
+# Global ad_tag fallback when user has no per-user tag in [access.user_ad_tags]
+# ad_tag = "00000000000000000000000000000000"
+# Per-user ad_tag in [access.user_ad_tags] (32 hex from @MTProxybot)
 
-[modes]
-classic = true
-secure = true
+# === Log Level ===
+# Log level: debug | verbose | normal | silent
+# Can be overridden with --silent or --log-level CLI flags
+# RUST_LOG env var takes absolute priority over all of these
+log_level = "normal"
+
+[general.modes]
+classic = false
+secure = false
 tls = true
 
-tls_domain = "www.github.com"
-fast_mode = true
-prefer_ipv6 = false
+[general.links]
+show = "*"
+# show = ["alice", "bob"] # Only show links for alice and bob
+# show = "*"              # Show links for all users
+# public_host = "proxy.example.com"  # Host (IP or domain) for tg:// links
+# public_port = 443                  # Port for tg:// links (default: server.port)
+
+# === Server Binding ===
+[server]
+port = 443
+# proxy_protocol = false           # Enable if behind HAProxy/nginx with PROXY protocol
+# metrics_port = 9090
+# metrics_whitelist = ["127.0.0.1", "::1", "0.0.0.0/0"]
+
+[server.api]
+enabled = true
+listen = "0.0.0.0:9091"
+whitelist = ["127.0.0.0/8"]
+minimal_runtime_enabled = false
+minimal_runtime_cache_ttl_ms = 1000
+
+# Listen on multiple interfaces/IPs - IPv4
+[[server.listeners]]
+ip = "0.0.0.0"
+
+# === Anti-Censorship & Masking ===
+[censorship]
+tls_domain = "petrovich.ru"
+mask = true
+tls_emulation = true        # Fetch real cert lengths and emulate TLS records
+tls_front_dir = "tlsfront"   # Cache directory for TLS emulation
+
+[access.users]
+# format: "username" = "32_hex_chars_secret"
+hello = "00000000000000000000000000000000"

docker-compose.yml

@@ -0,0 +1,30 @@
+services:
+  telemt:
+    image: ghcr.io/telemt/telemt:latest
+    build: .
+    container_name: telemt
+    restart: unless-stopped
+    ports:
+      - "443:443"
+      - "127.0.0.1:9090:9090"
+    # Allow caching 'proxy-secret' in read-only container
+    working_dir: /run/telemt
+    volumes:
+      - ./config.toml:/run/telemt/config.toml:ro
+    tmpfs:
+      - /run/telemt:rw,mode=1777,size=1m
+    environment:
+      - RUST_LOG=info
+    # Uncomment this line if you want to use host network for IPv6, but bridge is default and usually better
+    # network_mode: host
+    cap_drop:
+      - ALL
+    cap_add:
+      - NET_BIND_SERVICE   # allow binding to port 443
+    read_only: true
+    security_opt:
+      - no-new-privileges:true
+    ulimits:
+      nofile:
+        soft: 65536
+        hard: 65536

docs/API.md

@@ -0,0 +1,548 @@
+# Telemt Control API
+
+## Purpose
+Control-plane HTTP API for runtime visibility and user/config management.
+Data-plane MTProto traffic is out of scope.
+
+## Runtime Configuration
+API runtime is configured in `[server.api]`.
+
+| Field | Type | Default | Description |
+| --- | --- | --- | --- |
+| `enabled` | `bool` | `false` | Enables REST API listener. |
+| `listen` | `string` (`IP:PORT`) | `127.0.0.1:9091` | API bind address. |
+| `whitelist` | `CIDR[]` | `127.0.0.1/32, ::1/128` | Source IP allowlist. Empty list means allow all. |
+| `auth_header` | `string` | `""` | Exact value for `Authorization` header. Empty disables header auth. |
+| `request_body_limit_bytes` | `usize` | `65536` | Maximum request body size. Must be `> 0`. |
+| `minimal_runtime_enabled` | `bool` | `false` | Enables runtime snapshot endpoints requiring ME pool read-lock aggregation. |
+| `minimal_runtime_cache_ttl_ms` | `u64` | `1000` | Cache TTL for minimal snapshots. `0` disables cache; valid range is `[0, 60000]`. |
+| `read_only` | `bool` | `false` | Disables mutating endpoints. |
+
+`server.admin_api` is accepted as an alias for backward compatibility.
+
+Runtime validation for API config:
+- `server.api.listen` must be a valid `IP:PORT`.
+- `server.api.request_body_limit_bytes` must be `> 0`.
+- `server.api.minimal_runtime_cache_ttl_ms` must be within `[0, 60000]`.
+
+## Protocol Contract
+
+| Item | Value |
+| --- | --- |
+| Transport | HTTP/1.1 |
+| Content type | `application/json; charset=utf-8` |
+| Prefix | `/v1` |
+| Optimistic concurrency | `If-Match: <revision>` on mutating requests (optional) |
+| Revision format | SHA-256 hex of current `config.toml` content |
+
+### Success Envelope
+```json
+{
+  "ok": true,
+  "data": {},
+  "revision": "sha256-hex"
+}
+```
+
+### Error Envelope
+```json
+{
+  "ok": false,
+  "error": {
+    "code": "machine_code",
+    "message": "human-readable"
+  },
+  "request_id": 1
+}
+```
+
+## Request Processing Order
+
+Requests are processed in this order:
+1. `api_enabled` gate (`503 api_disabled` if disabled).
+2. Source IP whitelist gate (`403 forbidden`).
+3. `Authorization` header gate when configured (`401 unauthorized`).
+4. Route and method matching (`404 not_found` or `405 method_not_allowed`).
+5. `read_only` gate for mutating routes (`403 read_only`).
+6. Request body read/limit/JSON decode (`413 payload_too_large`, `400 bad_request`).
+7. Business validation and config write path.
+
+Notes:
+- Whitelist is evaluated against the direct TCP peer IP (`SocketAddr::ip`), without `X-Forwarded-For` support.
+- `Authorization` check is exact string equality against configured `auth_header`.
+
+## Endpoint Matrix
+
+| Method | Path | Body | Success | `data` contract |
+| --- | --- | --- | --- | --- |
+| `GET` | `/v1/health` | none | `200` | `HealthData` |
+| `GET` | `/v1/stats/summary` | none | `200` | `SummaryData` |
+| `GET` | `/v1/stats/zero/all` | none | `200` | `ZeroAllData` |
+| `GET` | `/v1/stats/upstreams` | none | `200` | `UpstreamsData` |
+| `GET` | `/v1/stats/minimal/all` | none | `200` | `MinimalAllData` |
+| `GET` | `/v1/stats/me-writers` | none | `200` | `MeWritersData` |
+| `GET` | `/v1/stats/dcs` | none | `200` | `DcStatusData` |
+| `GET` | `/v1/stats/users` | none | `200` | `UserInfo[]` |
+| `GET` | `/v1/users` | none | `200` | `UserInfo[]` |
+| `POST` | `/v1/users` | `CreateUserRequest` | `201` | `CreateUserResponse` |
+| `GET` | `/v1/users/{username}` | none | `200` | `UserInfo` |
+| `PATCH` | `/v1/users/{username}` | `PatchUserRequest` | `200` | `UserInfo` |
+| `DELETE` | `/v1/users/{username}` | none | `200` | `string` (deleted username) |
+| `POST` | `/v1/users/{username}/rotate-secret` | `RotateSecretRequest` or empty body | `404` | `ErrorResponse` (`not_found`, current runtime behavior) |
+
+## Common Error Codes
+
+| HTTP | `error.code` | Trigger |
+| --- | --- | --- |
+| `400` | `bad_request` | Invalid JSON, validation failures, malformed request body. |
+| `401` | `unauthorized` | Missing/invalid `Authorization` when `auth_header` is configured. |
+| `403` | `forbidden` | Source IP is not allowed by whitelist. |
+| `403` | `read_only` | Mutating endpoint called while `read_only=true`. |
+| `404` | `not_found` | Unknown route, unknown user, or unsupported sub-route (including current `rotate-secret` route). |
+| `405` | `method_not_allowed` | Unsupported method for `/v1/users/{username}` route shape. |
+| `409` | `revision_conflict` | `If-Match` revision mismatch. |
+| `409` | `user_exists` | User already exists on create. |
+| `409` | `last_user_forbidden` | Attempt to delete last configured user. |
+| `413` | `payload_too_large` | Body exceeds `request_body_limit_bytes`. |
+| `500` | `internal_error` | Internal error (I/O, serialization, config load/save). |
+| `503` | `api_disabled` | API disabled in config. |
+
+## Routing and Method Edge Cases
+
+| Case | Behavior |
+| --- | --- |
+| Path matching | Exact match on `req.uri().path()`. Query string does not affect route matching. |
+| Trailing slash | Not normalized. Example: `/v1/users/` is `404`. |
+| Username route with extra slash | `/v1/users/{username}/...` is not treated as user route and returns `404`. |
+| `PUT /v1/users/{username}` | `405 method_not_allowed`. |
+| `POST /v1/users/{username}` | `404 not_found`. |
+| `POST /v1/users/{username}/rotate-secret` | `404 not_found` in current release due route matcher limitation. |
+
+## Body and JSON Semantics
+
+- Request body is read only for mutating routes that define a body contract.
+- Body size limit is enforced during streaming read (`413 payload_too_large`).
+- Invalid transport body frame returns `400 bad_request` (`Invalid request body`).
+- Invalid JSON returns `400 bad_request` (`Invalid JSON body`).
+- `Content-Type` is not required for JSON parsing.
+- Unknown JSON fields are ignored by deserialization.
+- `PATCH` updates only provided fields and does not support explicit clearing of optional fields.
+- `If-Match` supports both quoted and unquoted values; surrounding whitespace is trimmed.
+
+## Request Contracts
+
+### `CreateUserRequest`
+| Field | Type | Required | Description |
+| --- | --- | --- | --- |
+| `username` | `string` | yes | `[A-Za-z0-9_.-]`, length `1..64`. |
+| `secret` | `string` | no | Exactly 32 hex chars. If missing, generated automatically. |
+| `user_ad_tag` | `string` | no | Exactly 32 hex chars. |
+| `max_tcp_conns` | `usize` | no | Per-user concurrent TCP limit. |
+| `expiration_rfc3339` | `string` | no | RFC3339 expiration timestamp. |
+| `data_quota_bytes` | `u64` | no | Per-user traffic quota. |
+| `max_unique_ips` | `usize` | no | Per-user unique source IP limit. |
+
+### `PatchUserRequest`
+| Field | Type | Required | Description |
+| --- | --- | --- | --- |
+| `secret` | `string` | no | Exactly 32 hex chars. |
+| `user_ad_tag` | `string` | no | Exactly 32 hex chars. |
+| `max_tcp_conns` | `usize` | no | Per-user concurrent TCP limit. |
+| `expiration_rfc3339` | `string` | no | RFC3339 expiration timestamp. |
+| `data_quota_bytes` | `u64` | no | Per-user traffic quota. |
+| `max_unique_ips` | `usize` | no | Per-user unique source IP limit. |
+
+### `RotateSecretRequest`
+| Field | Type | Required | Description |
+| --- | --- | --- | --- |
+| `secret` | `string` | no | Exactly 32 hex chars. If missing, generated automatically. |
+
+Note: the request contract is defined, but the corresponding route currently returns `404` (see routing edge cases).
+
+## Response Data Contracts
+
+### `HealthData`
+| Field | Type | Description |
+| --- | --- | --- |
+| `status` | `string` | Always `"ok"`. |
+| `read_only` | `bool` | Mirrors current API `read_only` mode. |
+
+### `SummaryData`
+| Field | Type | Description |
+| --- | --- | --- |
+| `uptime_seconds` | `f64` | Process uptime in seconds. |
+| `connections_total` | `u64` | Total accepted client connections. |
+| `connections_bad_total` | `u64` | Failed/invalid client connections. |
+| `handshake_timeouts_total` | `u64` | Handshake timeout count. |
+| `configured_users` | `usize` | Number of configured users in config. |
+
+### `ZeroAllData`
+| Field | Type | Description |
+| --- | --- | --- |
+| `generated_at_epoch_secs` | `u64` | Snapshot time (Unix epoch seconds). |
+| `core` | `ZeroCoreData` | Core counters and telemetry policy snapshot. |
+| `upstream` | `ZeroUpstreamData` | Upstream connect counters/histogram buckets. |
+| `middle_proxy` | `ZeroMiddleProxyData` | ME protocol/health counters. |
+| `pool` | `ZeroPoolData` | ME pool lifecycle counters. |
+| `desync` | `ZeroDesyncData` | Frame desync counters. |
+
+#### `ZeroCoreData`
+| Field | Type | Description |
+| --- | --- | --- |
+| `uptime_seconds` | `f64` | Process uptime. |
+| `connections_total` | `u64` | Total accepted connections. |
+| `connections_bad_total` | `u64` | Failed/invalid connections. |
+| `handshake_timeouts_total` | `u64` | Handshake timeouts. |
+| `configured_users` | `usize` | Configured user count. |
+| `telemetry_core_enabled` | `bool` | Core telemetry toggle. |
+| `telemetry_user_enabled` | `bool` | User telemetry toggle. |
+| `telemetry_me_level` | `string` | ME telemetry level (`off|normal|verbose`). |
+
+#### `ZeroUpstreamData`
+| Field | Type | Description |
+| --- | --- | --- |
+| `connect_attempt_total` | `u64` | Total upstream connect attempts. |
+| `connect_success_total` | `u64` | Successful upstream connects. |
+| `connect_fail_total` | `u64` | Failed upstream connects. |
+| `connect_failfast_hard_error_total` | `u64` | Fail-fast hard errors. |
+| `connect_attempts_bucket_1` | `u64` | Connect attempts resolved in 1 try. |
+| `connect_attempts_bucket_2` | `u64` | Connect attempts resolved in 2 tries. |
+| `connect_attempts_bucket_3_4` | `u64` | Connect attempts resolved in 3-4 tries. |
+| `connect_attempts_bucket_gt_4` | `u64` | Connect attempts requiring more than 4 tries. |
+| `connect_duration_success_bucket_le_100ms` | `u64` | Successful connects <=100 ms. |
+| `connect_duration_success_bucket_101_500ms` | `u64` | Successful connects 101-500 ms. |
+| `connect_duration_success_bucket_501_1000ms` | `u64` | Successful connects 501-1000 ms. |
+| `connect_duration_success_bucket_gt_1000ms` | `u64` | Successful connects >1000 ms. |
+| `connect_duration_fail_bucket_le_100ms` | `u64` | Failed connects <=100 ms. |
+| `connect_duration_fail_bucket_101_500ms` | `u64` | Failed connects 101-500 ms. |
+| `connect_duration_fail_bucket_501_1000ms` | `u64` | Failed connects 501-1000 ms. |
+| `connect_duration_fail_bucket_gt_1000ms` | `u64` | Failed connects >1000 ms. |
+
+### `UpstreamsData`
+| Field | Type | Description |
+| --- | --- | --- |
+| `enabled` | `bool` | Runtime upstream snapshot availability according to API config. |
+| `reason` | `string?` | `feature_disabled` or `source_unavailable` when runtime snapshot is unavailable. |
+| `generated_at_epoch_secs` | `u64` | Snapshot generation time. |
+| `zero` | `ZeroUpstreamData` | Always available zero-cost upstream counters block. |
+| `summary` | `UpstreamSummaryData?` | Runtime upstream aggregate view, null when unavailable. |
+| `upstreams` | `UpstreamStatus[]?` | Per-upstream runtime status rows, null when unavailable. |
+
+#### `UpstreamSummaryData`
+| Field | Type | Description |
+| --- | --- | --- |
+| `configured_total` | `usize` | Total configured upstream entries. |
+| `healthy_total` | `usize` | Upstreams currently marked healthy. |
+| `unhealthy_total` | `usize` | Upstreams currently marked unhealthy. |
+| `direct_total` | `usize` | Number of direct upstream entries. |
+| `socks4_total` | `usize` | Number of SOCKS4 upstream entries. |
+| `socks5_total` | `usize` | Number of SOCKS5 upstream entries. |
+
+#### `UpstreamStatus`
+| Field | Type | Description |
+| --- | --- | --- |
+| `upstream_id` | `usize` | Runtime upstream index. |
+| `route_kind` | `string` | Upstream route kind: `direct`, `socks4`, `socks5`. |
+| `address` | `string` | Upstream address (`direct` for direct route kind). Authentication fields are intentionally omitted. |
+| `weight` | `u16` | Selection weight. |
+| `scopes` | `string` | Configured scope selector string. |
+| `healthy` | `bool` | Current health flag. |
+| `fails` | `u32` | Consecutive fail counter. |
+| `last_check_age_secs` | `u64` | Seconds since the last health-check update. |
+| `effective_latency_ms` | `f64?` | Effective upstream latency used by selector. |
+| `dc` | `UpstreamDcStatus[]` | Per-DC latency/IP preference snapshot. |
+
+#### `UpstreamDcStatus`
+| Field | Type | Description |
+| --- | --- | --- |
+| `dc` | `i16` | Telegram DC id. |
+| `latency_ema_ms` | `f64?` | Per-DC latency EMA value. |
+| `ip_preference` | `string` | Per-DC IP family preference: `unknown`, `prefer_v4`, `prefer_v6`, `both_work`, `unavailable`. |
+
+#### `ZeroMiddleProxyData`
+| Field | Type | Description |
+| --- | --- | --- |
+| `keepalive_sent_total` | `u64` | ME keepalive packets sent. |
+| `keepalive_failed_total` | `u64` | ME keepalive send failures. |
+| `keepalive_pong_total` | `u64` | Keepalive pong responses received. |
+| `keepalive_timeout_total` | `u64` | Keepalive timeout events. |
+| `rpc_proxy_req_signal_sent_total` | `u64` | RPC proxy activity signals sent. |
+| `rpc_proxy_req_signal_failed_total` | `u64` | RPC proxy activity signal failures. |
+| `rpc_proxy_req_signal_skipped_no_meta_total` | `u64` | Signals skipped due to missing metadata. |
+| `rpc_proxy_req_signal_response_total` | `u64` | RPC proxy signal responses received. |
+| `rpc_proxy_req_signal_close_sent_total` | `u64` | RPC proxy close signals sent. |
+| `reconnect_attempt_total` | `u64` | ME reconnect attempts. |
+| `reconnect_success_total` | `u64` | Successful reconnects. |
+| `handshake_reject_total` | `u64` | ME handshake rejects. |
+| `handshake_error_codes` | `ZeroCodeCount[]` | Handshake rejects grouped by code. |
+| `reader_eof_total` | `u64` | ME reader EOF events. |
+| `idle_close_by_peer_total` | `u64` | Idle closes initiated by peer. |
+| `route_drop_no_conn_total` | `u64` | Route drops due to missing bound connection. |
+| `route_drop_channel_closed_total` | `u64` | Route drops due to closed channel. |
+| `route_drop_queue_full_total` | `u64` | Route drops due to full queue (total). |
+| `route_drop_queue_full_base_total` | `u64` | Route drops in base queue mode. |
+| `route_drop_queue_full_high_total` | `u64` | Route drops in high queue mode. |
+| `socks_kdf_strict_reject_total` | `u64` | SOCKS KDF strict rejects. |
+| `socks_kdf_compat_fallback_total` | `u64` | SOCKS KDF compat fallbacks. |
+| `endpoint_quarantine_total` | `u64` | Endpoint quarantine activations. |
+| `kdf_drift_total` | `u64` | KDF drift detections. |
+| `kdf_port_only_drift_total` | `u64` | KDF port-only drift detections. |
+| `hardswap_pending_reuse_total` | `u64` | Pending hardswap reused events. |
+| `hardswap_pending_ttl_expired_total` | `u64` | Pending hardswap TTL expiry events. |
+| `single_endpoint_outage_enter_total` | `u64` | Entered single-endpoint outage mode. |
+| `single_endpoint_outage_exit_total` | `u64` | Exited single-endpoint outage mode. |
+| `single_endpoint_outage_reconnect_attempt_total` | `u64` | Reconnect attempts in outage mode. |
+| `single_endpoint_outage_reconnect_success_total` | `u64` | Reconnect successes in outage mode. |
+| `single_endpoint_quarantine_bypass_total` | `u64` | Quarantine bypasses in outage mode. |
+| `single_endpoint_shadow_rotate_total` | `u64` | Shadow writer rotations. |
+| `single_endpoint_shadow_rotate_skipped_quarantine_total` | `u64` | Shadow rotations skipped because of quarantine. |
+| `floor_mode_switch_total` | `u64` | Total floor mode switches. |
+| `floor_mode_switch_static_to_adaptive_total` | `u64` | Static -> adaptive switches. |
+| `floor_mode_switch_adaptive_to_static_total` | `u64` | Adaptive -> static switches. |
+
+#### `ZeroCodeCount`
+| Field | Type | Description |
+| --- | --- | --- |
+| `code` | `i32` | Handshake error code. |
+| `total` | `u64` | Events with this code. |
+
+#### `ZeroPoolData`
+| Field | Type | Description |
+| --- | --- | --- |
+| `pool_swap_total` | `u64` | Pool swap count. |
+| `pool_drain_active` | `u64` | Current active draining pools. |
+| `pool_force_close_total` | `u64` | Forced pool closes by timeout. |
+| `pool_stale_pick_total` | `u64` | Stale writer picks for binding. |
+| `writer_removed_total` | `u64` | Writer removals total. |
+| `writer_removed_unexpected_total` | `u64` | Unexpected writer removals. |
+| `refill_triggered_total` | `u64` | Refill triggers. |
+| `refill_skipped_inflight_total` | `u64` | Refill skipped because refill already in-flight. |
+| `refill_failed_total` | `u64` | Refill failures. |
+| `writer_restored_same_endpoint_total` | `u64` | Restores on same endpoint. |
+| `writer_restored_fallback_total` | `u64` | Restores on fallback endpoint. |
+
+#### `ZeroDesyncData`
+| Field | Type | Description |
+| --- | --- | --- |
+| `secure_padding_invalid_total` | `u64` | Invalid secure padding events. |
+| `desync_total` | `u64` | Desync events total. |
+| `desync_full_logged_total` | `u64` | Fully logged desync events. |
+| `desync_suppressed_total` | `u64` | Suppressed desync logs. |
+| `desync_frames_bucket_0` | `u64` | Desync frames bucket 0. |
+| `desync_frames_bucket_1_2` | `u64` | Desync frames bucket 1-2. |
+| `desync_frames_bucket_3_10` | `u64` | Desync frames bucket 3-10. |
+| `desync_frames_bucket_gt_10` | `u64` | Desync frames bucket >10. |
+
+### `MinimalAllData`
+| Field | Type | Description |
+| --- | --- | --- |
+| `enabled` | `bool` | Whether minimal runtime snapshots are enabled by config. |
+| `reason` | `string?` | `feature_disabled` or `source_unavailable` when applicable. |
+| `generated_at_epoch_secs` | `u64` | Snapshot generation time. |
+| `data` | `MinimalAllPayload?` | Null when disabled; fallback payload when source unavailable. |
+
+#### `MinimalAllPayload`
+| Field | Type | Description |
+| --- | --- | --- |
+| `me_writers` | `MeWritersData` | ME writer status block. |
+| `dcs` | `DcStatusData` | DC aggregate status block. |
+| `me_runtime` | `MinimalMeRuntimeData?` | Runtime ME control snapshot. |
+| `network_path` | `MinimalDcPathData[]` | Active IP path selection per DC. |
+
+#### `MinimalMeRuntimeData`
+| Field | Type | Description |
+| --- | --- | --- |
+| `active_generation` | `u64` | Active pool generation. |
+| `warm_generation` | `u64` | Warm pool generation. |
+| `pending_hardswap_generation` | `u64` | Pending hardswap generation. |
+| `pending_hardswap_age_secs` | `u64?` | Pending hardswap age in seconds. |
+| `hardswap_enabled` | `bool` | Hardswap mode toggle. |
+| `floor_mode` | `string` | Writer floor mode. |
+| `adaptive_floor_idle_secs` | `u64` | Idle threshold for adaptive floor. |
+| `adaptive_floor_min_writers_single_endpoint` | `u8` | Minimum writers for single-endpoint DC in adaptive mode. |
+| `adaptive_floor_recover_grace_secs` | `u64` | Grace period for floor recovery. |
+| `me_keepalive_enabled` | `bool` | ME keepalive toggle. |
+| `me_keepalive_interval_secs` | `u64` | Keepalive period. |
+| `me_keepalive_jitter_secs` | `u64` | Keepalive jitter. |
+| `me_keepalive_payload_random` | `bool` | Randomized keepalive payload toggle. |
+| `rpc_proxy_req_every_secs` | `u64` | Period for RPC proxy request signal. |
+| `me_reconnect_max_concurrent_per_dc` | `u32` | Reconnect concurrency per DC. |
+| `me_reconnect_backoff_base_ms` | `u64` | Base reconnect backoff. |
+| `me_reconnect_backoff_cap_ms` | `u64` | Max reconnect backoff. |
+| `me_reconnect_fast_retry_count` | `u32` | Fast retry attempts before normal backoff. |
+| `me_pool_drain_ttl_secs` | `u64` | Pool drain TTL. |
+| `me_pool_force_close_secs` | `u64` | Hard close timeout for draining writers. |
+| `me_pool_min_fresh_ratio` | `f32` | Minimum fresh ratio before swap. |
+| `me_bind_stale_mode` | `string` | Stale writer bind policy. |
+| `me_bind_stale_ttl_secs` | `u64` | Stale writer TTL. |
+| `me_single_endpoint_shadow_writers` | `u8` | Shadow writers for single-endpoint DCs. |
+| `me_single_endpoint_outage_mode_enabled` | `bool` | Outage mode toggle for single-endpoint DCs. |
+| `me_single_endpoint_outage_disable_quarantine` | `bool` | Quarantine behavior in outage mode. |
+| `me_single_endpoint_outage_backoff_min_ms` | `u64` | Outage mode min reconnect backoff. |
+| `me_single_endpoint_outage_backoff_max_ms` | `u64` | Outage mode max reconnect backoff. |
+| `me_single_endpoint_shadow_rotate_every_secs` | `u64` | Shadow rotation interval. |
+| `me_deterministic_writer_sort` | `bool` | Deterministic writer ordering toggle. |
+| `me_socks_kdf_policy` | `string` | Current SOCKS KDF policy mode. |
+| `quarantined_endpoints_total` | `usize` | Total quarantined endpoints. |
+| `quarantined_endpoints` | `MinimalQuarantineData[]` | Quarantine details. |
+
+#### `MinimalQuarantineData`
+| Field | Type | Description |
+| --- | --- | --- |
+| `endpoint` | `string` | Endpoint (`ip:port`). |
+| `remaining_ms` | `u64` | Remaining quarantine duration. |
+
+#### `MinimalDcPathData`
+| Field | Type | Description |
+| --- | --- | --- |
+| `dc` | `i16` | Telegram DC identifier. |
+| `ip_preference` | `string?` | Runtime IP family preference. |
+| `selected_addr_v4` | `string?` | Selected IPv4 endpoint for this DC. |
+| `selected_addr_v6` | `string?` | Selected IPv6 endpoint for this DC. |
+
+### `MeWritersData`
+| Field | Type | Description |
+| --- | --- | --- |
+| `middle_proxy_enabled` | `bool` | `false` when minimal runtime is disabled or source unavailable. |
+| `reason` | `string?` | `feature_disabled` or `source_unavailable` when not fully available. |
+| `generated_at_epoch_secs` | `u64` | Snapshot generation time. |
+| `summary` | `MeWritersSummary` | Coverage/availability summary. |
+| `writers` | `MeWriterStatus[]` | Per-writer statuses. |
+
+#### `MeWritersSummary`
+| Field | Type | Description |
+| --- | --- | --- |
+| `configured_dc_groups` | `usize` | Number of configured DC groups. |
+| `configured_endpoints` | `usize` | Total configured ME endpoints. |
+| `available_endpoints` | `usize` | Endpoints currently available. |
+| `available_pct` | `f64` | `available_endpoints / configured_endpoints * 100`. |
+| `required_writers` | `usize` | Required writers based on current floor policy. |
+| `alive_writers` | `usize` | Writers currently alive. |
+| `coverage_pct` | `f64` | `alive_writers / required_writers * 100`. |
+
+#### `MeWriterStatus`
+| Field | Type | Description |
+| --- | --- | --- |
+| `writer_id` | `u64` | Runtime writer identifier. |
+| `dc` | `i16?` | DC id if mapped. |
+| `endpoint` | `string` | Endpoint (`ip:port`). |
+| `generation` | `u64` | Pool generation owning this writer. |
+| `state` | `string` | Writer state (`warm`, `active`, `draining`). |
+| `draining` | `bool` | Draining flag. |
+| `degraded` | `bool` | Degraded flag. |
+| `bound_clients` | `usize` | Number of currently bound clients. |
+| `idle_for_secs` | `u64?` | Idle age in seconds if idle. |
+| `rtt_ema_ms` | `f64?` | RTT exponential moving average. |
+
+### `DcStatusData`
+| Field | Type | Description |
+| --- | --- | --- |
+| `middle_proxy_enabled` | `bool` | `false` when minimal runtime is disabled or source unavailable. |
+| `reason` | `string?` | `feature_disabled` or `source_unavailable` when not fully available. |
+| `generated_at_epoch_secs` | `u64` | Snapshot generation time. |
+| `dcs` | `DcStatus[]` | Per-DC status rows. |
+
+#### `DcStatus`
+| Field | Type | Description |
+| --- | --- | --- |
+| `dc` | `i16` | Telegram DC id. |
+| `endpoints` | `string[]` | Endpoints in this DC (`ip:port`). |
+| `available_endpoints` | `usize` | Endpoints currently available in this DC. |
+| `available_pct` | `f64` | `available_endpoints / endpoints_total * 100`. |
+| `required_writers` | `usize` | Required writer count for this DC. |
+| `alive_writers` | `usize` | Alive writers in this DC. |
+| `coverage_pct` | `f64` | `alive_writers / required_writers * 100`. |
+| `rtt_ms` | `f64?` | Aggregated RTT for DC. |
+| `load` | `usize` | Active client sessions bound to this DC. |
+
+### `UserInfo`
+| Field | Type | Description |
+| --- | --- | --- |
+| `username` | `string` | Username. |
+| `user_ad_tag` | `string?` | Optional ad tag (32 hex chars). |
+| `max_tcp_conns` | `usize?` | Optional max concurrent TCP limit. |
+| `expiration_rfc3339` | `string?` | Optional expiration timestamp. |
+| `data_quota_bytes` | `u64?` | Optional data quota. |
+| `max_unique_ips` | `usize?` | Optional unique IP limit. |
+| `current_connections` | `u64` | Current live connections. |
+| `active_unique_ips` | `usize` | Current active unique source IPs. |
+| `total_octets` | `u64` | Total traffic octets for this user. |
+| `links` | `UserLinks` | Active connection links derived from current config. |
+
+#### `UserLinks`
+| Field | Type | Description |
+| --- | --- | --- |
+| `classic` | `string[]` | Active `tg://proxy` links for classic mode. |
+| `secure` | `string[]` | Active `tg://proxy` links for secure/DD mode. |
+| `tls` | `string[]` | Active `tg://proxy` links for EE-TLS mode (for each host+TLS domain). |
+
+Link generation uses active config and enabled modes:
+- `[general.links].public_host/public_port` have priority.
+- If `public_host` is not set, startup-detected public IPs are used (`IPv4`, `IPv6`, or both when available).
+- Fallback host sources: listener `announce`, `announce_ip`, explicit listener `ip`.
+- Legacy fallback: `listen_addr_ipv4` and `listen_addr_ipv6` when routable.
+- Startup-detected IPs are fixed for process lifetime and refreshed on restart.
+- User rows are sorted by `username` in ascending lexical order.
+
+### `CreateUserResponse`
+| Field | Type | Description |
+| --- | --- | --- |
+| `user` | `UserInfo` | Created or updated user view. |
+| `secret` | `string` | Effective user secret. |
+
+## Mutation Semantics
+
+| Endpoint | Notes |
+| --- | --- |
+| `POST /v1/users` | Creates user and validates resulting config before atomic save. |
+| `PATCH /v1/users/{username}` | Partial update of provided fields only. Missing fields remain unchanged. |
+| `POST /v1/users/{username}/rotate-secret` | Currently returns `404` in runtime route matcher; request schema is reserved for intended behavior. |
+| `DELETE /v1/users/{username}` | Deletes user and related optional settings. Last user deletion is blocked. |
+
+All mutating endpoints:
+- Respect `read_only` mode.
+- Accept optional `If-Match` for optimistic concurrency.
+- Return new `revision` after successful write.
+- Use process-local mutation lock + atomic write (`tmp + rename`) for config persistence.
+
+## Runtime State Matrix
+
+| Endpoint | `minimal_runtime_enabled=false` | `minimal_runtime_enabled=true` + source unavailable | `minimal_runtime_enabled=true` + source available |
+| --- | --- | --- | --- |
+| `/v1/stats/minimal/all` | `enabled=false`, `reason=feature_disabled`, `data=null` | `enabled=true`, `reason=source_unavailable`, fallback `data` with disabled ME blocks | `enabled=true`, `reason` omitted, full payload |
+| `/v1/stats/me-writers` | `middle_proxy_enabled=false`, `reason=feature_disabled` | `middle_proxy_enabled=false`, `reason=source_unavailable` | `middle_proxy_enabled=true`, runtime snapshot |
+| `/v1/stats/dcs` | `middle_proxy_enabled=false`, `reason=feature_disabled` | `middle_proxy_enabled=false`, `reason=source_unavailable` | `middle_proxy_enabled=true`, runtime snapshot |
+| `/v1/stats/upstreams` | `enabled=false`, `reason=feature_disabled`, `summary/upstreams` omitted, `zero` still present | `enabled=true`, `reason=source_unavailable`, `summary/upstreams` omitted, `zero` present | `enabled=true`, `reason` omitted, `summary/upstreams` present, `zero` present |
+
+`source_unavailable` conditions:
+- ME endpoints: ME pool is absent (for example direct-only mode or failed ME initialization).
+- Upstreams endpoint: non-blocking upstream snapshot lock is unavailable at request time.
+
+## Serialization Rules
+
+- Success responses always include `revision`.
+- Error responses never include `revision`; they include `request_id`.
+- Optional fields with `skip_serializing_if` are omitted when absent.
+- Nullable payload fields may still be `null` where contract uses `?` (for example `UserInfo` option fields).
+- For `/v1/stats/upstreams`, authentication details of SOCKS upstreams are intentionally omitted.
+
+## Operational Notes
+
+| Topic | Details |
+| --- | --- |
+| API startup | API listener is spawned only when `[server.api].enabled=true`. |
+| `listen` port `0` | API spawn is skipped when parsed listen port is `0` (treated as disabled bind target). |
+| Bind failure | Failed API bind logs warning and API task exits (no auto-retry loop). |
+| ME runtime status endpoints | `/v1/stats/me-writers`, `/v1/stats/dcs`, `/v1/stats/minimal/all` require `[server.api].minimal_runtime_enabled=true`; otherwise they return disabled payload with `reason=feature_disabled`. |
+| Upstream runtime endpoint | `/v1/stats/upstreams` always returns `zero`, but runtime fields (`summary`, `upstreams`) require `[server.api].minimal_runtime_enabled=true`. |
+| Restart requirements | `server.api` changes are restart-required for predictable behavior. |
+| Hot-reload nuance | A pure `server.api`-only config change may not propagate through watcher broadcast; a mixed change (with hot fields) may propagate API flags while still warning that restart is required. |
+| Runtime apply path | Successful writes are picked up by existing config watcher/hot-reload path. |
+| Exposure | Built-in TLS/mTLS is not provided. Use loopback bind + reverse proxy if needed. |
+| Pagination | User list currently has no pagination/filtering. |
+| Serialization side effect | Config comments/manual formatting are not preserved on write. |
+
+## Known Limitations (Current Release)
+
+- `POST /v1/users/{username}/rotate-secret` is currently unreachable in route matcher and returns `404`.
+- API runtime controls under `server.api` are documented as restart-required; hot-reload behavior for these fields is not strictly uniform in all change combinations.

docs/FAQ.ru.md

@@ -0,0 +1,65 @@
+## Как настроить канал "спонсор прокси"
+
+1. Зайти в бота @MTProxybot.
+2. Ввести команду `/newproxy`
+3. Отправить IP и порт сервера. Например: 1.2.3.4:443
+4. Открыть конфиг `nano /etc/telemt.toml`.
+5. Скопировать и отправить боту секрет пользователя из раздела [access.users].
+6. Скопировать полученный tag у бота. Например 1234567890abcdef1234567890abcdef.
+> [!WARNING]
+> Ссылка, которую выдает бот, не будет работать. Не копируйте и не используйте её!
+7. Раскомментировать параметр ad_tag и вписать tag, полученный у бота.
+8. Раскомментировать/добавить параметр use_middle_proxy = true.
+
+Пример конфига:
+```toml
+[general]
+ad_tag = "1234567890abcdef1234567890abcdef"
+use_middle_proxy = true
+```
+9.  Сохранить конфиг. Ctrl+S -> Ctrl+X.
+10. Перезапустить telemt `systemctl restart telemt`.
+11. В боте отправить команду /myproxies и выбрать добавленный сервер.
+12. Нажать кнопку "Set promotion".
+13. Отправить **публичную ссылку** на канал. Приватный канал добавить нельзя!
+14. Подождать примерно 1 час, пока информация обновится на серверах Telegram.
+> [!WARNING]
+> У вас не будет отображаться "спонсор прокси" если вы уже подписаны на канал.
+
+## Сколько человек может пользоваться 1 ссылкой
+
+По умолчанию 1 ссылкой может пользоваться сколько угодно человек.  
+Вы можете ограничить число IP, использующих прокси.
+```toml
+[access.user_max_unique_ips]
+hello = 1
+```
+Этот параметр ограничивает, сколько уникальных IP может использовать 1 ссылку одновременно. Если один пользователь отключится, второй сможет подключиться. Также с одного IP может сидеть несколько пользователей.
+
+## Как сделать несколько разных ссылок
+
+1. Сгенерируйте нужное число секретов `openssl rand -hex 16`
+2. Открыть конфиг `nano /etc/telemt.toml`
+3. Добавить новых пользователей.
+```toml
+[access.users]
+user1 = "00000000000000000000000000000001"
+user2 = "00000000000000000000000000000002"
+user3 = "00000000000000000000000000000003"
+```
+4. Сохранить конфиг. Ctrl+S -> Ctrl+X. Перезапускать telemt не нужно.
+5. Получить ссылки через `journalctl -u telemt -n -g "links" --no-pager -o cat | tac`
+
+## Как посмотреть метрики
+
+1. Открыть конфиг `nano /etc/telemt.toml`
+2. Добавить следующие параметры
+```toml
+[server]
+metrics_port = 9090
+metrics_whitelist = ["127.0.0.1/32", "::1/128", "0.0.0.0/0"]
+```
+3. Сохранить конфиг. Ctrl+S -> Ctrl+X.
+4. Метрики доступны по адресу SERVER_IP:9090/metrics. 
+> [!WARNING]
+> "0.0.0.0/0" в metrics_whitelist открывает доступ с любого IP. Замените на свой ip. Например "1.2.3.4"

docs/MIDDLE-END-KDF.de.md

@@ -0,0 +1,40 @@
+# Middle-End Proxy
+
+## KDF-Adressierung — Implementierungs-FAQ
+
+### Benötigt die C-Referenzimplementierung sowohl externe IP-Adresse als auch Port für die KDF?
+
+Ja.
+
+In der C-Referenzimplementierung werden **sowohl IP-Adresse als auch Port in die KDF einbezogen** — auf beiden Seiten der Verbindung.
+
+In `aes_create_keys()` enthält der KDF-Input:
+
+- `server_ip + client_port`
+- `client_ip + server_port`
+- sowie Secret / Nonces
+
+Für IPv6:
+
+- IPv4-Felder werden auf 0 gesetzt
+- IPv6-Adressen werden ergänzt
+
+Die **Ports bleiben weiterhin Bestandteil der KDF**.
+
+> Wenn sich externe IP oder Port (z. B. durch NAT, SOCKS oder Proxy) von den erwarteten Werten unterscheiden, entstehen unterschiedliche Schlüssel — der Handshake schlägt fehl.
+
+---
+
+### Kann der Port aus der KDF ausgeschlossen werden (z. B. durch Port = 0)?
+
+**Nein!**
+
+Die C-Referenzimplementierung enthält **keine Möglichkeit, den Port zu ignorieren**:
+- `client_port` und `server_port` sind fester Bestandteil der KDF
+- Es werden immer reale Socket-Ports übergeben:
+  - `c->our_port`
+  - `c->remote_port`
+
+Falls ein Port den Wert `0` hat, wird er dennoch als `0` in die KDF übernommen.
+
+Eine „Port-Ignore“-Logik existiert nicht.

docs/MIDDLE-END-KDF.en.md

@@ -0,0 +1,41 @@
+# Middle-End Proxy
+
+## KDF Addressing — Implementation FAQ
+
+### Does the C-implementation require both external IP address and port for the KDF?
+
+**Yes!**
+
+In the C reference implementation, **both IP address and port are included in the KDF input** from both sides of the connection.
+
+Inside `aes_create_keys()`, the KDF input explicitly contains:
+
+- `server_ip + client_port`
+- `client_ip + server_port`
+- followed by shared secret / nonces
+
+For IPv6:
+
+- IPv4 fields are zeroed
+- IPv6 addresses are inserted
+
+However, **client_port and server_port remain part of the KDF regardless of IP version**.
+
+> If externally observed IP or port (e.g. due to NAT, SOCKS, or proxy traversal) differs from what the peer expects, the derived keys will not match and the handshake will fail.
+
+---
+
+### Can port be excluded from KDF (e.g. by using port = 0)?
+
+**No!**
+
+The C-implementation provides **no mechanism to ignore the port**:
+
+- `client_port` and `server_port` are explicitly included in the KDF input
+- Real socket ports are always passed:
+  - `c->our_port`
+  - `c->remote_port`
+
+If a port is `0`, it is still incorporated into the KDF as `0`.
+
+There is **no conditional logic to exclude ports**

docs/MIDDLE-END-KDF.ru.md

@@ -0,0 +1,41 @@
+# Middle-End Proxy
+
+## KDF Addressing — FAQ по реализации
+
+### Требует ли C-референсная реализация KDF внешний IP и порт?
+
+**Да**
+
+В C-референсе **в KDF участвуют и IP-адрес, и порт** — с обеих сторон соединения.
+
+В `aes_create_keys()` в строку KDF входят:
+
+- `server_ip + client_port`
+- `client_ip + server_port`
+- далее secret / nonces
+
+Для IPv6:
+
+- IPv4-поля заполняются нулями
+- добавляются IPv6-адреса
+
+Однако **порты client_port и server_port всё равно участвуют в KDF**.
+
+> Если внешний IP или порт (например, из-за NAT, SOCKS или прокси) не совпадает с ожидаемым другой стороной — ключи расходятся и handshake ломается.
+
+---
+
+### Можно ли исключить порт из KDF (например, установив порт = 0)?
+
+**Нет.**
+
+В C-референсе **нет механики отключения порта**.
+
+- `client_port` и `server_port` явно включены в KDF
+- Передаются реальные порты сокета:
+  - `c->our_port`
+  - `c->remote_port`
+
+Если порт равен `0`, он всё равно попадёт в KDF как `0`.
+
+Отдельной логики «игнорировать порт» не предусмотрено.

docs/QUICK_START_GUIDE.en.md

@@ -0,0 +1,153 @@
+# Telemt via Systemd
+
+## Installation
+
+This software is designed for Debian-based OS: in addition to Debian, these are Ubuntu, Mint, Kali, MX and many other Linux
+
+**1. Download**
+```bash
+wget -qO- "https://github.com/telemt/telemt/releases/latest/download/telemt-$(uname -m)-linux-$(ldd --version 2>&1 | grep -iq musl && echo musl || echo gnu).tar.gz" | tar -xz
+```
+**2. Move to the Bin folder**
+```bash
+mv telemt /bin
+```
+**3. Make the file executable**
+```bash
+chmod +x /bin/telemt
+```
+
+## How to use?
+
+**This guide "assumes" that you:**
+- logged in as root or executed `su -` / `sudo su`
+- Already have the "telemt" executable file in the /bin folder. Read the **[Installation](#Installation)** section.
+
+---
+
+**0. Check port and generate secrets**
+
+The port you have selected for use should be MISSING from the list, when:
+```bash
+netstat -lnp
+```
+
+Generate 16 bytes/32 characters HEX with OpenSSL or another way:
+```bash
+openssl rand -hex 16
+```
+OR
+```bash
+xxd -l 16 -p /dev/urandom
+```
+OR
+```bash
+python3 -c 'import os; print(os.urandom(16).hex())'
+```
+Save the obtained result somewhere. You will need it later!
+
+---
+
+**1. Place your config to /etc/telemt.toml**
+
+Open nano
+```bash
+nano /etc/telemt.toml
+```
+paste your config
+
+```toml
+# === General Settings ===
+[general]
+# ad_tag = "00000000000000000000000000000000"
+use_middle_proxy = false
+
+[general.modes]
+classic = false
+secure = false
+tls = true
+
+# === Anti-Censorship & Masking ===
+[censorship]
+tls_domain = "petrovich.ru"
+
+[access.users]
+# format: "username" = "32_hex_chars_secret"
+hello = "00000000000000000000000000000000"
+```
+then Ctrl+S -> Ctrl+X to save
+
+> [!WARNING]
+> Replace the value of the hello parameter with the value you obtained in step 0. 
+> Replace the value of the tls_domain parameter with another website.
+
+---
+
+**2. Create service on /etc/systemd/system/telemt.service**
+
+Open nano
+```bash
+nano /etc/systemd/system/telemt.service
+```
+
+paste this Systemd Module
+```bash
+[Unit]
+Description=Telemt
+After=network.target
+
+[Service]
+Type=simple
+WorkingDirectory=/bin
+ExecStart=/bin/telemt /etc/telemt.toml
+Restart=on-failure
+LimitNOFILE=65536
+
+[Install]
+WantedBy=multi-user.target
+```
+then Ctrl+S -> Ctrl+X to save
+
+
+**3.** To start it, enter the command `systemctl start telemt`
+
+**4.** To get status information, enter `systemctl status telemt`
+
+**5.** For automatic startup at system boot, enter `systemctl enable telemt`
+
+**6.** To get the links, enter `journalctl -u telemt -n -g "links" --no-pager -o cat | tac`
+
+---
+
+# Telemt via Docker Compose
+
+**1. Edit `config.toml` in repo root (at least: port, users secrets, tls_domain)**  
+**2. Start container:**
+```bash
+docker compose up -d --build
+```
+**3. Check logs:**
+```bash
+docker compose logs -f telemt
+```
+**4. Stop:**
+```bash
+docker compose down
+```
+> [!NOTE]
+> - `docker-compose.yml` maps `./config.toml` to `/app/config.toml` (read-only)
+> - By default it publishes `443:443` and runs with dropped capabilities (only `NET_BIND_SERVICE` is added)
+> - If you really need host networking (usually only for some IPv6 setups) uncomment `network_mode: host`
+
+**Run without Compose**
+```bash
+docker build -t telemt:local .
+docker run --name telemt --restart unless-stopped \
+  -p 443:443 \
+  -e RUST_LOG=info \
+  -v "$PWD/config.toml:/app/config.toml:ro" \
+  --read-only \
+  --cap-drop ALL --cap-add NET_BIND_SERVICE \
+  --ulimit nofile=65536:65536 \
+  telemt:local
+```

docs/QUICK_START_GUIDE.ru.md

@@ -0,0 +1,155 @@
+# Telemt через Systemd
+
+## Установка
+
+Это программное обеспечение разработано для ОС на базе Debian: помимо Debian, это Ubuntu, Mint, Kali, MX и многие другие Linux
+
+**1. Скачать**
+```bash
+wget -qO- "https://github.com/telemt/telemt/releases/latest/download/telemt-$(uname -m)-linux-$(ldd --version 2>&1 | grep -iq musl && echo musl || echo gnu).tar.gz" | tar -xz
+```
+**2. Переместить в папку Bin**
+```bash
+mv telemt /bin
+```
+**3. Сделать файл исполняемым**
+```bash
+chmod +x /bin/telemt
+```
+
+## Как правильно использовать?
+
+**Эта инструкция "предполагает", что вы:**
+- Авторизовались как пользователь root или выполнил `su -` / `sudo su`
+- У вас уже есть исполняемый файл "telemt" в папке /bin. Читайте раздел **[Установка](#установка)**
+
+---
+
+**0. Проверьте порт и сгенерируйте секреты**
+
+Порт, который вы выбрали для использования, должен отсутствовать в списке:
+```bash
+netstat -lnp
+```
+
+Сгенерируйте 16 bytes/32 символа в шестнадцатеричном формате с помощью OpenSSL или другим способом:
+```bash
+openssl rand -hex 16
+```
+ИЛИ
+```bash
+xxd -l 16 -p /dev/urandom
+```
+ИЛИ
+```bash
+python3 -c 'import os; print(os.urandom(16).hex())'
+```
+Полученный результат сохраняем где-нибудь. Он понадобиться вам дальше!
+
+---
+
+**1. Поместите свою конфигурацию в файл /etc/telemt.toml**
+
+Открываем nano
+```bash
+nano /etc/telemt.toml
+```
+Вставьте свою конфигурацию
+
+```toml
+# === General Settings ===
+[general]
+# ad_tag = "00000000000000000000000000000000"
+use_middle_proxy = false
+
+[general.modes]
+classic = false
+secure = false
+tls = true
+
+# === Anti-Censorship & Masking ===
+[censorship]
+tls_domain = "petrovich.ru"
+
+[access.users]
+# format: "username" = "32_hex_chars_secret"
+hello = "00000000000000000000000000000000"
+```
+Затем нажмите Ctrl+S -> Ctrl+X, чтобы сохранить
+
+> [!WARNING]
+> Замените значение параметра hello на значение, которое вы получили в пункте 0.  
+> Так же замените значение параметра tls_domain на другой сайт.
+
+---
+
+**2. Создайте службу в /etc/systemd/system/telemt.service**
+
+Открываем nano
+```bash
+nano /etc/systemd/system/telemt.service
+```
+
+Вставьте этот модуль Systemd
+```bash
+[Unit]
+Description=Telemt
+After=network.target
+
+[Service]
+Type=simple
+WorkingDirectory=/bin
+ExecStart=/bin/telemt /etc/telemt.toml
+Restart=on-failure
+LimitNOFILE=65536
+
+[Install]
+WantedBy=multi-user.target
+```
+Затем нажмите Ctrl+S -> Ctrl+X, чтобы сохранить
+
+
+**3.** Для запуска введите команду `systemctl start telemt`
+
+**4.** Для получения информации о статусе введите `systemctl status telemt`
+
+**5.** Для автоматического запуска при запуске системы в введите `systemctl enable telemt`
+
+**6.** Для получения ссылки введите `journalctl -u telemt -n -g "links" --no-pager -o cat | tac`
+> [!WARNING]
+> Рабочую ссылку может выдать только команда из 6 пункта. Не пытайтесь делать ее самостоятельно или копировать откуда-либо!
+
+---
+
+# Telemt через Docker Compose
+
+**1. Отредактируйте `config.toml` в корневом каталоге репозитория (как минимум: порт, пользовательские секреты, tls_domain)**  
+**2. Запустите контейнер:**
+```bash
+docker compose up -d --build
+```
+**3. Проверьте логи:**
+```bash
+docker compose logs -f telemt
+```
+**4. Остановите контейнер:**
+```bash
+docker compose down
+```
+> [!NOTE]
+> - В `docker-compose.yml` файл `./config.toml` монтируется в `/app/config.toml` (доступно только для чтения)  
+> - По умолчанию публикуются порты 443:443, а контейнер запускается со сброшенными привилегиями (добавлена только `NET_BIND_SERVICE`)  
+> - Если вам действительно нужна сеть хоста (обычно это требуется только для некоторых конфигураций IPv6), раскомментируйте `network_mode: host`
+
+**Запуск в Docker Compose**
+```bash
+docker build -t telemt:local .
+docker run --name telemt --restart unless-stopped \
+  -p 443:443 \
+  -e RUST_LOG=info \
+  -v "$PWD/config.toml:/app/config.toml:ro" \
+  --read-only \
+  --cap-drop ALL --cap-add NET_BIND_SERVICE \
+  --ulimit nofile=65536:65536 \
+  telemt:local
+```

docs/TUNING.de.md

@@ -0,0 +1,219 @@
+# Telemt Tuning-Leitfaden: Middle-End und Upstreams
+
+Dieses Dokument beschreibt das aktuelle Laufzeitverhalten für Middle-End (ME) und Upstream-Routing basierend auf:
+- `src/config/types.rs`
+- `src/config/defaults.rs`
+- `src/config/load.rs`
+- `src/transport/upstream.rs`
+
+Die unten angegebenen `Default`-Werte sind Code-Defaults (bei fehlendem Schlüssel), nicht zwingend die Werte aus `config.full.toml`.
+
+## Middle-End-Parameter
+
+### 1) ME-Grundmodus, NAT und STUN
+
+| Parameter | Typ | Default | Einschränkungen / Validierung | Laufzeiteffekt | Beispiel |
+|---|---|---:|---|---|---|
+| `general.use_middle_proxy` | `bool` | `true` | keine | Aktiviert den ME-Transportmodus. Bei `false` wird Direct-Modus verwendet. | `use_middle_proxy = true` |
+| `general.proxy_secret_path` | `Option<String>` | `"proxy-secret"` | Pfad kann `null` sein | Pfad zur Telegram-Infrastrukturdatei `proxy-secret`. | `proxy_secret_path = "proxy-secret"` |
+| `general.middle_proxy_nat_ip` | `Option<IpAddr>` | `null` | gültige IP bei gesetztem Wert | Manueller Override der öffentlichen NAT-IP für ME-Adressmaterial. | `middle_proxy_nat_ip = "203.0.113.10"` |
+| `general.middle_proxy_nat_probe` | `bool` | `true` | wird auf `true` erzwungen, wenn `use_middle_proxy=true` | Aktiviert NAT-Probing für ME. | `middle_proxy_nat_probe = true` |
+| `general.stun_nat_probe_concurrency` | `usize` | `8` | muss `> 0` sein | Maximale parallele STUN-Probes während NAT-Erkennung. | `stun_nat_probe_concurrency = 16` |
+| `network.stun_use` | `bool` | `true` | keine | Globaler STUN-Schalter. Bei `false` wird STUN deaktiviert. | `stun_use = true` |
+| `network.stun_servers` | `Vec<String>` | integrierter öffentlicher Pool | Duplikate/leer werden entfernt | Primäre STUN-Serverliste für NAT/Public-Endpoint-Erkennung. | `stun_servers = ["stun1.l.google.com:19302"]` |
+| `network.stun_tcp_fallback` | `bool` | `true` | keine | Aktiviert TCP-Fallback, wenn UDP-STUN blockiert ist. | `stun_tcp_fallback = true` |
+| `network.http_ip_detect_urls` | `Vec<String>` | `ifconfig.me` + `api.ipify.org` | keine | HTTP-Fallback zur öffentlichen IPv4-Erkennung, falls STUN ausfällt. | `http_ip_detect_urls = ["https://api.ipify.org"]` |
+| `general.stun_iface_mismatch_ignore` | `bool` | `false` | keine | Reserviertes Feld in der aktuellen Revision (derzeit kein aktiver Runtime-Verbrauch). | `stun_iface_mismatch_ignore = false` |
+| `timeouts.me_one_retry` | `u8` | `12` | keine | Anzahl schneller Reconnect-Versuche bei Single-Endpoint-DC-Fällen. | `me_one_retry = 6` |
+| `timeouts.me_one_timeout_ms` | `u64` | `1200` | keine | Timeout pro schnellem Einzelversuch (ms). | `me_one_timeout_ms = 1500` |
+
+### 2) Poolgröße, Keepalive und Reconnect-Policy
+
+| Parameter | Typ | Default | Einschränkungen / Validierung | Laufzeiteffekt | Beispiel |
+|---|---|---:|---|---|---|
+| `general.middle_proxy_pool_size` | `usize` | `8` | keine | Zielgröße des aktiven ME-Writer-Pools. | `middle_proxy_pool_size = 12` |
+| `general.middle_proxy_warm_standby` | `usize` | `16` | keine | Reserviertes Kompatibilitätsfeld in der aktuellen Revision (kein aktiver Runtime-Consumer). | `middle_proxy_warm_standby = 16` |
+| `general.me_keepalive_enabled` | `bool` | `true` | keine | Aktiviert periodischen ME-Keepalive/Ping-Traffic. | `me_keepalive_enabled = true` |
+| `general.me_keepalive_interval_secs` | `u64` | `25` | keine | Basisintervall für Keepalive (Sekunden). | `me_keepalive_interval_secs = 20` |
+| `general.me_keepalive_jitter_secs` | `u64` | `5` | keine | Keepalive-Jitter zur Vermeidung synchroner Peaks. | `me_keepalive_jitter_secs = 3` |
+| `general.me_keepalive_payload_random` | `bool` | `true` | keine | Randomisiert Keepalive-Payload-Bytes. | `me_keepalive_payload_random = true` |
+| `general.me_warmup_stagger_enabled` | `bool` | `true` | keine | Aktiviert gestaffeltes Warmup zusätzlicher ME-Verbindungen. | `me_warmup_stagger_enabled = true` |
+| `general.me_warmup_step_delay_ms` | `u64` | `500` | keine | Basisverzögerung zwischen Warmup-Schritten (ms). | `me_warmup_step_delay_ms = 300` |
+| `general.me_warmup_step_jitter_ms` | `u64` | `300` | keine | Zusätzlicher zufälliger Warmup-Jitter (ms). | `me_warmup_step_jitter_ms = 200` |
+| `general.me_reconnect_max_concurrent_per_dc` | `u32` | `8` | keine | Begrenzung paralleler Reconnect-Worker pro DC. | `me_reconnect_max_concurrent_per_dc = 12` |
+| `general.me_reconnect_backoff_base_ms` | `u64` | `500` | keine | Initiales Reconnect-Backoff (ms). | `me_reconnect_backoff_base_ms = 250` |
+| `general.me_reconnect_backoff_cap_ms` | `u64` | `30000` | keine | Maximales Reconnect-Backoff (ms). | `me_reconnect_backoff_cap_ms = 10000` |
+| `general.me_reconnect_fast_retry_count` | `u32` | `16` | keine | Budget für Sofort-Retries vor längerem Backoff. | `me_reconnect_fast_retry_count = 8` |
+
+### 3) Reinit/Hardswap, Secret-Rotation und Degradation
+
+| Parameter | Typ | Default | Einschränkungen / Validierung | Laufzeiteffekt | Beispiel |
+|---|---|---:|---|---|---|
+| `general.hardswap` | `bool` | `true` | keine | Aktiviert generation-basierte Hardswap-Strategie für den ME-Pool. | `hardswap = true` |
+| `general.me_reinit_every_secs` | `u64` | `900` | muss `> 0` sein | Intervall für periodische ME-Reinitialisierung. | `me_reinit_every_secs = 600` |
+| `general.me_hardswap_warmup_delay_min_ms` | `u64` | `1000` | muss `<= me_hardswap_warmup_delay_max_ms` sein | Untere Grenze für Warmup-Dial-Abstände. | `me_hardswap_warmup_delay_min_ms = 500` |
+| `general.me_hardswap_warmup_delay_max_ms` | `u64` | `2000` | muss `> 0` sein | Obere Grenze für Warmup-Dial-Abstände. | `me_hardswap_warmup_delay_max_ms = 1200` |
+| `general.me_hardswap_warmup_extra_passes` | `u8` | `3` | Bereich `[0,10]` | Zusätzliche Warmup-Pässe nach dem Basispass. | `me_hardswap_warmup_extra_passes = 2` |
+| `general.me_hardswap_warmup_pass_backoff_base_ms` | `u64` | `500` | muss `> 0` sein | Basis-Backoff zwischen zusätzlichen Warmup-Pässen. | `me_hardswap_warmup_pass_backoff_base_ms = 400` |
+| `general.me_config_stable_snapshots` | `u8` | `2` | muss `> 0` sein | Anzahl identischer ME-Config-Snapshots vor Apply. | `me_config_stable_snapshots = 3` |
+| `general.me_config_apply_cooldown_secs` | `u64` | `300` | keine | Cooldown zwischen angewendeten ME-Map-Updates. | `me_config_apply_cooldown_secs = 120` |
+| `general.proxy_secret_stable_snapshots` | `u8` | `2` | muss `> 0` sein | Anzahl identischer Secret-Snapshots vor Rotation. | `proxy_secret_stable_snapshots = 3` |
+| `general.proxy_secret_rotate_runtime` | `bool` | `true` | keine | Aktiviert Runtime-Rotation des Proxy-Secrets. | `proxy_secret_rotate_runtime = true` |
+| `general.proxy_secret_len_max` | `usize` | `256` | Bereich `[32,4096]` | Obergrenze für akzeptierte Secret-Länge. | `proxy_secret_len_max = 512` |
+| `general.update_every` | `Option<u64>` | `300` | wenn gesetzt: `> 0`; bei `null`: Legacy-Min-Fallback | Einheitliches Refresh-Intervall für ME-Config + Secret-Updater. | `update_every = 300` |
+| `general.me_pool_drain_ttl_secs` | `u64` | `90` | keine | Zeitraum, in dem stale Writer noch als Fallback zulässig sind. | `me_pool_drain_ttl_secs = 120` |
+| `general.me_pool_min_fresh_ratio` | `f32` | `0.8` | Bereich `[0.0,1.0]` | Coverage-Schwelle vor Drain der alten Generation. | `me_pool_min_fresh_ratio = 0.9` |
+| `general.me_reinit_drain_timeout_secs` | `u64` | `120` | `0` = kein Force-Close; wenn `>0 && < TTL`, dann auf TTL angehoben | Force-Close-Timeout für draining stale Writer. | `me_reinit_drain_timeout_secs = 0` |
+| `general.auto_degradation_enabled` | `bool` | `true` | keine | Reserviertes Kompatibilitätsfeld in aktueller Revision (kein aktiver Runtime-Consumer). | `auto_degradation_enabled = true` |
+| `general.degradation_min_unavailable_dc_groups` | `u8` | `2` | keine | Reservierter Kompatibilitäts-Schwellenwert in aktueller Revision (kein aktiver Runtime-Consumer). | `degradation_min_unavailable_dc_groups = 2` |
+
+## Deprecated / Legacy Parameter
+
+| Parameter | Status | Ersatz | Aktuelles Verhalten | Migrationshinweis |
+|---|---|---|---|---|
+| `general.middle_proxy_nat_stun` | Deprecated | `network.stun_servers` | Wird nur dann in `network.stun_servers` gemerged, wenn `network.stun_servers` nicht explizit gesetzt ist. | Wert nach `network.stun_servers` verschieben, Legacy-Key entfernen. |
+| `general.middle_proxy_nat_stun_servers` | Deprecated | `network.stun_servers` | Wird nur dann in `network.stun_servers` gemerged, wenn `network.stun_servers` nicht explizit gesetzt ist. | Werte nach `network.stun_servers` verschieben, Legacy-Key entfernen. |
+| `general.proxy_secret_auto_reload_secs` | Deprecated | `general.update_every` | Nur aktiv, wenn `update_every = null` (Legacy-Fallback). | `general.update_every` explizit setzen, Legacy-Key entfernen. |
+| `general.proxy_config_auto_reload_secs` | Deprecated | `general.update_every` | Nur aktiv, wenn `update_every = null` (Legacy-Fallback). | `general.update_every` explizit setzen, Legacy-Key entfernen. |
+
+## Wie Upstreams konfiguriert werden
+
+### Upstream-Schema
+
+| Feld | Gilt für | Typ | Pflicht | Default | Bedeutung |
+|---|---|---|---|---|---|
+| `[[upstreams]].type` | alle Upstreams | `"direct" \| "socks4" \| "socks5"` | ja | n/a | Upstream-Transporttyp. |
+| `[[upstreams]].weight` | alle Upstreams | `u16` | nein | `1` | Basisgewicht für weighted-random Auswahl. |
+| `[[upstreams]].enabled` | alle Upstreams | `bool` | nein | `true` | Deaktivierte Einträge werden beim Start ignoriert. |
+| `[[upstreams]].scopes` | alle Upstreams | `String` | nein | `""` | Komma-separierte Scope-Tags für Request-Routing. |
+| `interface` | `direct` | `Option<String>` | nein | `null` | Interface-Name (z. B. `eth0`) oder lokale Literal-IP. |
+| `bind_addresses` | `direct` | `Option<Vec<IpAddr>>` | nein | `null` | Explizite Source-IP-Kandidaten (strikter Vorrang vor `interface`). |
+| `address` | `socks4` | `String` | ja | n/a | SOCKS4-Server (`ip:port` oder `host:port`). |
+| `interface` | `socks4` | `Option<String>` | nein | `null` | Wird nur genutzt, wenn `address` als `ip:port` angegeben ist. |
+| `user_id` | `socks4` | `Option<String>` | nein | `null` | SOCKS4 User-ID für CONNECT. |
+| `address` | `socks5` | `String` | ja | n/a | SOCKS5-Server (`ip:port` oder `host:port`). |
+| `interface` | `socks5` | `Option<String>` | nein | `null` | Wird nur genutzt, wenn `address` als `ip:port` angegeben ist. |
+| `username` | `socks5` | `Option<String>` | nein | `null` | SOCKS5 Benutzername. |
+| `password` | `socks5` | `Option<String>` | nein | `null` | SOCKS5 Passwort. |
+
+### Runtime-Regeln (wichtig)
+
+1. Wenn `[[upstreams]]` fehlt, injiziert der Loader einen Default-`direct`-Upstream.
+2. Scope-Filterung basiert auf exaktem Token-Match:
+- mit Request-Scope -> nur Einträge, deren `scopes` genau dieses Token enthält;
+- ohne Request-Scope -> nur Einträge mit leerem `scopes`.
+3. Unter healthy Upstreams erfolgt die Auswahl per weighted random: `weight * latency_factor`.
+4. Gibt es im gefilterten Set keinen healthy Upstream, wird zufällig aus dem gefilterten Set gewählt.
+5. `direct`-Bind-Auflösung:
+- zuerst `bind_addresses` (nur gleiche IP-Familie wie Target);
+- bei `interface` (Name) + `bind_addresses` wird jede Candidate-IP gegen Interface-Adressen validiert;
+- ungültige Kandidaten werden mit `WARN` verworfen;
+- bleiben keine gültigen Kandidaten übrig, erfolgt unbound direct connect (`bind_ip=None`);
+- wenn `bind_addresses` nicht passt, wird `interface` verwendet (Literal-IP oder Interface-Primäradresse).
+6. Für `socks4/socks5` mit Hostname-`address` ist Interface-Binding nicht unterstützt und wird mit Warnung ignoriert.
+7. Runtime DNS Overrides werden für Hostname-Auflösung bei Upstream-Verbindungen genutzt.
+8. Im ME-Modus wird der gewählte Upstream auch für den ME-TCP-Dial-Pfad verwendet.
+9. Im ME-Modus ist bei `direct` mit bind/interface die STUN-Reflection bind-aware für KDF-Adressmaterial.
+10. Im ME-Modus werden bei SOCKS-Upstream `BND.ADDR/BND.PORT` für KDF verwendet, wenn gültig/öffentlich und gleiche IP-Familie.
+
+## Upstream-Konfigurationsbeispiele
+
+### Beispiel 1: Minimaler direct Upstream
+
+```toml
+[[upstreams]]
+type = "direct"
+weight = 1
+enabled = true
+```
+
+### Beispiel 2: direct mit Interface + expliziten bind IPs
+
+```toml
+[[upstreams]]
+type = "direct"
+interface = "eth0"
+bind_addresses = ["192.168.1.100", "192.168.1.101"]
+weight = 3
+enabled = true
+```
+
+### Beispiel 3: SOCKS5 Upstream mit Authentifizierung
+
+```toml
+[[upstreams]]
+type = "socks5"
+address = "198.51.100.30:1080"
+username = "proxy-user"
+password = "proxy-pass"
+weight = 2
+enabled = true
+```
+
+### Beispiel 4: Gemischte Upstreams mit Scopes
+
+```toml
+[[upstreams]]
+type = "direct"
+weight = 5
+enabled = true
+scopes = ""
+
+[[upstreams]]
+type = "socks5"
+address = "203.0.113.40:1080"
+username = "edge"
+password = "edgepass"
+weight = 3
+enabled = true
+scopes = "premium,me"
+```
+
+### Beispiel 5: ME-orientiertes Tuning-Profil
+
+```toml
+[general]
+use_middle_proxy = true
+proxy_secret_path = "proxy-secret"
+middle_proxy_nat_probe = true
+stun_nat_probe_concurrency = 16
+middle_proxy_pool_size = 12
+me_keepalive_enabled = true
+me_keepalive_interval_secs = 20
+me_keepalive_jitter_secs = 4
+me_reconnect_max_concurrent_per_dc = 12
+me_reconnect_backoff_base_ms = 300
+me_reconnect_backoff_cap_ms = 10000
+me_reconnect_fast_retry_count = 10
+hardswap = true
+me_reinit_every_secs = 600
+me_hardswap_warmup_delay_min_ms = 500
+me_hardswap_warmup_delay_max_ms = 1200
+me_hardswap_warmup_extra_passes = 2
+me_hardswap_warmup_pass_backoff_base_ms = 400
+me_config_stable_snapshots = 3
+me_config_apply_cooldown_secs = 120
+proxy_secret_stable_snapshots = 3
+proxy_secret_rotate_runtime = true
+proxy_secret_len_max = 512
+update_every = 300
+me_pool_drain_ttl_secs = 120
+me_pool_min_fresh_ratio = 0.9
+me_reinit_drain_timeout_secs = 180
+
+[timeouts]
+me_one_retry = 8
+me_one_timeout_ms = 1200
+
+[network]
+stun_use = true
+stun_tcp_fallback = true
+stun_servers = [
+  "stun1.l.google.com:19302",
+  "stun2.l.google.com:19302"
+]
+http_ip_detect_urls = [
+  "https://api.ipify.org",
+  "https://ifconfig.me/ip"
+]
+```

docs/TUNING.en.md

@@ -0,0 +1,219 @@
+# Telemt Tuning Guide: Middle-End and Upstreams
+
+This document describes the current runtime behavior for Middle-End (ME) and upstream routing based on:
+- `src/config/types.rs`
+- `src/config/defaults.rs`
+- `src/config/load.rs`
+- `src/transport/upstream.rs`
+
+Defaults below are code defaults (used when a key is omitted), not necessarily values from `config.full.toml` examples.
+
+## Middle-End Parameters
+
+### 1) Core ME mode, NAT, and STUN
+
+| Parameter | Type | Default | Constraints / validation | Runtime effect | Example |
+|---|---|---:|---|---|---|
+| `general.use_middle_proxy` | `bool` | `true` | none | Enables ME transport mode. If `false`, Direct mode is used. | `use_middle_proxy = true` |
+| `general.proxy_secret_path` | `Option<String>` | `"proxy-secret"` | path may be `null` | Path to Telegram infrastructure proxy-secret file. | `proxy_secret_path = "proxy-secret"` |
+| `general.middle_proxy_nat_ip` | `Option<IpAddr>` | `null` | valid IP when set | Manual public NAT IP override for ME address material. | `middle_proxy_nat_ip = "203.0.113.10"` |
+| `general.middle_proxy_nat_probe` | `bool` | `true` | auto-forced to `true` when `use_middle_proxy=true` | Enables ME NAT probing. | `middle_proxy_nat_probe = true` |
+| `general.stun_nat_probe_concurrency` | `usize` | `8` | must be `> 0` | Max parallel STUN probes during NAT discovery. | `stun_nat_probe_concurrency = 16` |
+| `network.stun_use` | `bool` | `true` | none | Global STUN switch. If `false`, STUN probing is disabled. | `stun_use = true` |
+| `network.stun_servers` | `Vec<String>` | built-in public pool | deduplicated + empty values removed | Primary STUN server list for NAT/public endpoint discovery. | `stun_servers = ["stun1.l.google.com:19302"]` |
+| `network.stun_tcp_fallback` | `bool` | `true` | none | Enables TCP fallback path when UDP STUN is blocked. | `stun_tcp_fallback = true` |
+| `network.http_ip_detect_urls` | `Vec<String>` | `ifconfig.me` + `api.ipify.org` | none | HTTP fallback for public IPv4 detection if STUN is unavailable. | `http_ip_detect_urls = ["https://api.ipify.org"]` |
+| `general.stun_iface_mismatch_ignore` | `bool` | `false` | none | Reserved flag in current revision (not consumed by runtime path). | `stun_iface_mismatch_ignore = false` |
+| `timeouts.me_one_retry` | `u8` | `12` | none | Fast reconnect attempts for single-endpoint DC cases. | `me_one_retry = 6` |
+| `timeouts.me_one_timeout_ms` | `u64` | `1200` | none | Timeout per quick single-endpoint attempt (ms). | `me_one_timeout_ms = 1500` |
+
+### 2) Pool size, keepalive, and reconnect policy
+
+| Parameter | Type | Default | Constraints / validation | Runtime effect | Example |
+|---|---|---:|---|---|---|
+| `general.middle_proxy_pool_size` | `usize` | `8` | none | Target active ME writer pool size. | `middle_proxy_pool_size = 12` |
+| `general.middle_proxy_warm_standby` | `usize` | `16` | none | Reserved compatibility field in current revision (no active runtime consumer). | `middle_proxy_warm_standby = 16` |
+| `general.me_keepalive_enabled` | `bool` | `true` | none | Enables periodic ME keepalive/ping traffic. | `me_keepalive_enabled = true` |
+| `general.me_keepalive_interval_secs` | `u64` | `25` | none | Base keepalive interval (seconds). | `me_keepalive_interval_secs = 20` |
+| `general.me_keepalive_jitter_secs` | `u64` | `5` | none | Keepalive jitter to avoid synchronization bursts. | `me_keepalive_jitter_secs = 3` |
+| `general.me_keepalive_payload_random` | `bool` | `true` | none | Randomizes keepalive payload bytes. | `me_keepalive_payload_random = true` |
+| `general.me_warmup_stagger_enabled` | `bool` | `true` | none | Staggers extra ME warmup dials to avoid spikes. | `me_warmup_stagger_enabled = true` |
+| `general.me_warmup_step_delay_ms` | `u64` | `500` | none | Base delay between warmup dial steps (ms). | `me_warmup_step_delay_ms = 300` |
+| `general.me_warmup_step_jitter_ms` | `u64` | `300` | none | Additional random delay for warmup steps (ms). | `me_warmup_step_jitter_ms = 200` |
+| `general.me_reconnect_max_concurrent_per_dc` | `u32` | `8` | none | Limits concurrent reconnect workers per DC in health recovery. | `me_reconnect_max_concurrent_per_dc = 12` |
+| `general.me_reconnect_backoff_base_ms` | `u64` | `500` | none | Initial reconnect backoff (ms). | `me_reconnect_backoff_base_ms = 250` |
+| `general.me_reconnect_backoff_cap_ms` | `u64` | `30000` | none | Maximum reconnect backoff (ms). | `me_reconnect_backoff_cap_ms = 10000` |
+| `general.me_reconnect_fast_retry_count` | `u32` | `16` | none | Immediate retry budget before long backoff behavior. | `me_reconnect_fast_retry_count = 8` |
+
+### 3) Reinit/hardswap, secret rotation, and degradation
+
+| Parameter | Type | Default | Constraints / validation | Runtime effect | Example |
+|---|---|---:|---|---|---|
+| `general.hardswap` | `bool` | `true` | none | Enables generation-based ME hardswap strategy. | `hardswap = true` |
+| `general.me_reinit_every_secs` | `u64` | `900` | must be `> 0` | Periodic ME reinit interval. | `me_reinit_every_secs = 600` |
+| `general.me_hardswap_warmup_delay_min_ms` | `u64` | `1000` | must be `<= me_hardswap_warmup_delay_max_ms` | Lower bound for hardswap warmup dial spacing. | `me_hardswap_warmup_delay_min_ms = 500` |
+| `general.me_hardswap_warmup_delay_max_ms` | `u64` | `2000` | must be `> 0` | Upper bound for hardswap warmup dial spacing. | `me_hardswap_warmup_delay_max_ms = 1200` |
+| `general.me_hardswap_warmup_extra_passes` | `u8` | `3` | must be within `[0,10]` | Additional warmup passes after base pass. | `me_hardswap_warmup_extra_passes = 2` |
+| `general.me_hardswap_warmup_pass_backoff_base_ms` | `u64` | `500` | must be `> 0` | Base backoff between extra warmup passes. | `me_hardswap_warmup_pass_backoff_base_ms = 400` |
+| `general.me_config_stable_snapshots` | `u8` | `2` | must be `> 0` | Number of identical ME config snapshots required before apply. | `me_config_stable_snapshots = 3` |
+| `general.me_config_apply_cooldown_secs` | `u64` | `300` | none | Cooldown between applied ME map updates. | `me_config_apply_cooldown_secs = 120` |
+| `general.proxy_secret_stable_snapshots` | `u8` | `2` | must be `> 0` | Number of identical proxy-secret snapshots required before rotation. | `proxy_secret_stable_snapshots = 3` |
+| `general.proxy_secret_rotate_runtime` | `bool` | `true` | none | Enables runtime proxy-secret rotation. | `proxy_secret_rotate_runtime = true` |
+| `general.proxy_secret_len_max` | `usize` | `256` | must be within `[32,4096]` | Upper limit for accepted proxy-secret length. | `proxy_secret_len_max = 512` |
+| `general.update_every` | `Option<u64>` | `300` | if set: must be `> 0`; if `null`: legacy min fallback | Unified refresh interval for ME config + secret updater. | `update_every = 300` |
+| `general.me_pool_drain_ttl_secs` | `u64` | `90` | none | Time window where stale writers remain fallback-eligible. | `me_pool_drain_ttl_secs = 120` |
+| `general.me_pool_min_fresh_ratio` | `f32` | `0.8` | must be within `[0.0,1.0]` | Coverage threshold before stale generation can be drained. | `me_pool_min_fresh_ratio = 0.9` |
+| `general.me_reinit_drain_timeout_secs` | `u64` | `120` | `0` means no force-close; if `>0 && < TTL` it is bumped to TTL | Force-close timeout for draining stale writers. | `me_reinit_drain_timeout_secs = 0` |
+| `general.auto_degradation_enabled` | `bool` | `true` | none | Reserved compatibility flag in current revision (no active runtime consumer). | `auto_degradation_enabled = true` |
+| `general.degradation_min_unavailable_dc_groups` | `u8` | `2` | none | Reserved compatibility threshold in current revision (no active runtime consumer). | `degradation_min_unavailable_dc_groups = 2` |
+
+## Deprecated / Legacy Parameters
+
+| Parameter | Status | Replacement | Current behavior | Migration recommendation |
+|---|---|---|---|---|
+| `general.middle_proxy_nat_stun` | Deprecated | `network.stun_servers` | Merged into `network.stun_servers` only when `network.stun_servers` is not explicitly set. | Move value into `network.stun_servers` and remove legacy key. |
+| `general.middle_proxy_nat_stun_servers` | Deprecated | `network.stun_servers` | Merged into `network.stun_servers` only when `network.stun_servers` is not explicitly set. | Move values into `network.stun_servers` and remove legacy key. |
+| `general.proxy_secret_auto_reload_secs` | Deprecated | `general.update_every` | Used only when `update_every = null` (legacy fallback path). | Set `general.update_every` explicitly and remove legacy key. |
+| `general.proxy_config_auto_reload_secs` | Deprecated | `general.update_every` | Used only when `update_every = null` (legacy fallback path). | Set `general.update_every` explicitly and remove legacy key. |
+
+## How Upstreams Are Configured
+
+### Upstream schema
+
+| Field | Applies to | Type | Required | Default | Meaning |
+|---|---|---|---|---|---|
+| `[[upstreams]].type` | all upstreams | `"direct" \| "socks4" \| "socks5"` | yes | n/a | Upstream transport type. |
+| `[[upstreams]].weight` | all upstreams | `u16` | no | `1` | Base weight for weighted-random selection. |
+| `[[upstreams]].enabled` | all upstreams | `bool` | no | `true` | Disabled entries are ignored at startup. |
+| `[[upstreams]].scopes` | all upstreams | `String` | no | `""` | Comma-separated scope tags for request-level routing. |
+| `interface` | `direct` | `Option<String>` | no | `null` | Interface name (e.g. `eth0`) or literal local IP for bind selection. |
+| `bind_addresses` | `direct` | `Option<Vec<IpAddr>>` | no | `null` | Explicit local source IP candidates (strict priority over `interface`). |
+| `address` | `socks4` | `String` | yes | n/a | SOCKS4 server endpoint (`ip:port` or `host:port`). |
+| `interface` | `socks4` | `Option<String>` | no | `null` | Used only for SOCKS server `ip:port` dial path. |
+| `user_id` | `socks4` | `Option<String>` | no | `null` | SOCKS4 user ID for CONNECT request. |
+| `address` | `socks5` | `String` | yes | n/a | SOCKS5 server endpoint (`ip:port` or `host:port`). |
+| `interface` | `socks5` | `Option<String>` | no | `null` | Used only for SOCKS server `ip:port` dial path. |
+| `username` | `socks5` | `Option<String>` | no | `null` | SOCKS5 username auth. |
+| `password` | `socks5` | `Option<String>` | no | `null` | SOCKS5 password auth. |
+
+### Runtime rules (important)
+
+1. If `[[upstreams]]` is omitted, loader injects one default `direct` upstream.
+2. Scope filtering is exact-token based:
+- when request scope is set -> only entries whose `scopes` contains that exact token;
+- when request scope is not set -> only entries with empty `scopes`.
+3. Healthy upstreams are selected by weighted random using: `weight * latency_factor`.
+4. If no healthy upstream exists in filtered set, random selection is used among filtered entries.
+5. `direct` bind resolution order:
+- `bind_addresses` candidates (same IP family as target) first;
+- if `interface` is an interface name and `bind_addresses` is set, each candidate IP is validated against addresses currently assigned to that interface;
+- invalid candidates are dropped with `WARN`;
+- if no valid candidate remains, connection falls back to unbound direct connect (`bind_ip=None`);
+- if no `bind_addresses` candidate, `interface` is used (literal IP or resolved interface primary IP).
+6. For `socks4/socks5` with `address` as hostname, interface binding is not supported and is ignored with warning.
+7. Runtime DNS overrides are used for upstream hostname resolution.
+8. In ME mode, the selected upstream is also used for ME TCP dial path.
+9. In ME mode for `direct` upstream with bind/interface, STUN reflection logic is bind-aware for KDF source material.
+10. In ME mode for SOCKS upstream, SOCKS `BND.ADDR/BND.PORT` is used for KDF when it is valid/public for the same family.
+
+## Upstream Configuration Examples
+
+### Example 1: Minimal direct upstream
+
+```toml
+[[upstreams]]
+type = "direct"
+weight = 1
+enabled = true
+```
+
+### Example 2: Direct with interface + explicit bind addresses
+
+```toml
+[[upstreams]]
+type = "direct"
+interface = "eth0"
+bind_addresses = ["192.168.1.100", "192.168.1.101"]
+weight = 3
+enabled = true
+```
+
+### Example 3: SOCKS5 upstream with authentication
+
+```toml
+[[upstreams]]
+type = "socks5"
+address = "198.51.100.30:1080"
+username = "proxy-user"
+password = "proxy-pass"
+weight = 2
+enabled = true
+```
+
+### Example 4: Mixed upstreams with scopes
+
+```toml
+[[upstreams]]
+type = "direct"
+weight = 5
+enabled = true
+scopes = ""
+
+[[upstreams]]
+type = "socks5"
+address = "203.0.113.40:1080"
+username = "edge"
+password = "edgepass"
+weight = 3
+enabled = true
+scopes = "premium,me"
+```
+
+### Example 5: ME-focused tuning profile
+
+```toml
+[general]
+use_middle_proxy = true
+proxy_secret_path = "proxy-secret"
+middle_proxy_nat_probe = true
+stun_nat_probe_concurrency = 16
+middle_proxy_pool_size = 12
+me_keepalive_enabled = true
+me_keepalive_interval_secs = 20
+me_keepalive_jitter_secs = 4
+me_reconnect_max_concurrent_per_dc = 12
+me_reconnect_backoff_base_ms = 300
+me_reconnect_backoff_cap_ms = 10000
+me_reconnect_fast_retry_count = 10
+hardswap = true
+me_reinit_every_secs = 600
+me_hardswap_warmup_delay_min_ms = 500
+me_hardswap_warmup_delay_max_ms = 1200
+me_hardswap_warmup_extra_passes = 2
+me_hardswap_warmup_pass_backoff_base_ms = 400
+me_config_stable_snapshots = 3
+me_config_apply_cooldown_secs = 120
+proxy_secret_stable_snapshots = 3
+proxy_secret_rotate_runtime = true
+proxy_secret_len_max = 512
+update_every = 300
+me_pool_drain_ttl_secs = 120
+me_pool_min_fresh_ratio = 0.9
+me_reinit_drain_timeout_secs = 180
+
+[timeouts]
+me_one_retry = 8
+me_one_timeout_ms = 1200
+
+[network]
+stun_use = true
+stun_tcp_fallback = true
+stun_servers = [
+  "stun1.l.google.com:19302",
+  "stun2.l.google.com:19302"
+]
+http_ip_detect_urls = [
+  "https://api.ipify.org",
+  "https://ifconfig.me/ip"
+]
+```

docs/TUNING.ru.md

@@ -0,0 +1,219 @@
+# Руководство по тюнингу Telemt: Middle-End и Upstreams
+
+Документ описывает актуальное поведение Middle-End (ME) и маршрутизации через upstream на основе:
+- `src/config/types.rs`
+- `src/config/defaults.rs`
+- `src/config/load.rs`
+- `src/transport/upstream.rs`
+
+Значения `Default` ниже — это значения из кода при отсутствии ключа в конфиге, а не обязательно значения из примеров `config.full.toml`.
+
+## Параметры Middle-End
+
+### 1) Базовый режим ME, NAT и STUN
+
+| Параметр | Тип | Default | Ограничения / валидация | Влияние на runtime | Пример |
+|---|---|---:|---|---|---|
+| `general.use_middle_proxy` | `bool` | `true` | нет | Включает транспорт ME. При `false` используется Direct-режим. | `use_middle_proxy = true` |
+| `general.proxy_secret_path` | `Option<String>` | `"proxy-secret"` | путь может быть `null` | Путь к инфраструктурному proxy-secret Telegram. | `proxy_secret_path = "proxy-secret"` |
+| `general.middle_proxy_nat_ip` | `Option<IpAddr>` | `null` | валидный IP при задании | Ручной override публичного NAT IP для адресного материала ME. | `middle_proxy_nat_ip = "203.0.113.10"` |
+| `general.middle_proxy_nat_probe` | `bool` | `true` | авто-принудительно `true`, если `use_middle_proxy=true` | Включает NAT probing для ME. | `middle_proxy_nat_probe = true` |
+| `general.stun_nat_probe_concurrency` | `usize` | `8` | должно быть `> 0` | Максимум параллельных STUN-проб при NAT-детекте. | `stun_nat_probe_concurrency = 16` |
+| `network.stun_use` | `bool` | `true` | нет | Глобальный переключатель STUN. При `false` STUN отключен. | `stun_use = true` |
+| `network.stun_servers` | `Vec<String>` | встроенный публичный пул | удаляются дубликаты и пустые значения | Основной список STUN-серверов для NAT/public endpoint discovery. | `stun_servers = ["stun1.l.google.com:19302"]` |
+| `network.stun_tcp_fallback` | `bool` | `true` | нет | Включает TCP fallback, если UDP STUN недоступен. | `stun_tcp_fallback = true` |
+| `network.http_ip_detect_urls` | `Vec<String>` | `ifconfig.me` + `api.ipify.org` | нет | HTTP fallback для определения публичного IPv4 при недоступности STUN. | `http_ip_detect_urls = ["https://api.ipify.org"]` |
+| `general.stun_iface_mismatch_ignore` | `bool` | `false` | нет | Зарезервированный флаг в текущей ревизии (runtime его не использует). | `stun_iface_mismatch_ignore = false` |
+| `timeouts.me_one_retry` | `u8` | `12` | нет | Количество быстрых reconnect-попыток для DC с одним endpoint. | `me_one_retry = 6` |
+| `timeouts.me_one_timeout_ms` | `u64` | `1200` | нет | Таймаут одной быстрой попытки (мс). | `me_one_timeout_ms = 1500` |
+
+### 2) Размер пула, keepalive и reconnect-политика
+
+| Параметр | Тип | Default | Ограничения / валидация | Влияние на runtime | Пример |
+|---|---|---:|---|---|---|
+| `general.middle_proxy_pool_size` | `usize` | `8` | нет | Целевой размер активного пула ME-writer соединений. | `middle_proxy_pool_size = 12` |
+| `general.middle_proxy_warm_standby` | `usize` | `16` | нет | Зарезервированное поле совместимости в текущей ревизии (активного runtime-consumer нет). | `middle_proxy_warm_standby = 16` |
+| `general.me_keepalive_enabled` | `bool` | `true` | нет | Включает периодические keepalive/ping кадры ME. | `me_keepalive_enabled = true` |
+| `general.me_keepalive_interval_secs` | `u64` | `25` | нет | Базовый интервал keepalive (сек). | `me_keepalive_interval_secs = 20` |
+| `general.me_keepalive_jitter_secs` | `u64` | `5` | нет | Джиттер keepalive для предотвращения синхронных всплесков. | `me_keepalive_jitter_secs = 3` |
+| `general.me_keepalive_payload_random` | `bool` | `true` | нет | Рандомизирует payload keepalive-кадров. | `me_keepalive_payload_random = true` |
+| `general.me_warmup_stagger_enabled` | `bool` | `true` | нет | Включает staggered warmup дополнительных ME-коннектов. | `me_warmup_stagger_enabled = true` |
+| `general.me_warmup_step_delay_ms` | `u64` | `500` | нет | Базовая задержка между шагами warmup (мс). | `me_warmup_step_delay_ms = 300` |
+| `general.me_warmup_step_jitter_ms` | `u64` | `300` | нет | Дополнительный случайный warmup-джиттер (мс). | `me_warmup_step_jitter_ms = 200` |
+| `general.me_reconnect_max_concurrent_per_dc` | `u32` | `8` | нет | Ограничивает параллельные reconnect worker'ы на один DC. | `me_reconnect_max_concurrent_per_dc = 12` |
+| `general.me_reconnect_backoff_base_ms` | `u64` | `500` | нет | Начальный backoff reconnect (мс). | `me_reconnect_backoff_base_ms = 250` |
+| `general.me_reconnect_backoff_cap_ms` | `u64` | `30000` | нет | Верхняя граница backoff reconnect (мс). | `me_reconnect_backoff_cap_ms = 10000` |
+| `general.me_reconnect_fast_retry_count` | `u32` | `16` | нет | Бюджет быстрых retry до длинного backoff. | `me_reconnect_fast_retry_count = 8` |
+
+### 3) Reinit/hardswap, ротация секрета и деградация
+
+| Параметр | Тип | Default | Ограничения / валидация | Влияние на runtime | Пример |
+|---|---|---:|---|---|---|
+| `general.hardswap` | `bool` | `true` | нет | Включает generation-based стратегию hardswap для ME-пула. | `hardswap = true` |
+| `general.me_reinit_every_secs` | `u64` | `900` | должно быть `> 0` | Интервал периодического reinit ME-пула. | `me_reinit_every_secs = 600` |
+| `general.me_hardswap_warmup_delay_min_ms` | `u64` | `1000` | должно быть `<= me_hardswap_warmup_delay_max_ms` | Нижняя граница пауз между warmup dial попытками. | `me_hardswap_warmup_delay_min_ms = 500` |
+| `general.me_hardswap_warmup_delay_max_ms` | `u64` | `2000` | должно быть `> 0` | Верхняя граница пауз между warmup dial попытками. | `me_hardswap_warmup_delay_max_ms = 1200` |
+| `general.me_hardswap_warmup_extra_passes` | `u8` | `3` | диапазон `[0,10]` | Дополнительные warmup-проходы после базового. | `me_hardswap_warmup_extra_passes = 2` |
+| `general.me_hardswap_warmup_pass_backoff_base_ms` | `u64` | `500` | должно быть `> 0` | Базовый backoff между extra-pass в warmup. | `me_hardswap_warmup_pass_backoff_base_ms = 400` |
+| `general.me_config_stable_snapshots` | `u8` | `2` | должно быть `> 0` | Количество одинаковых snapshot перед применением ME map update. | `me_config_stable_snapshots = 3` |
+| `general.me_config_apply_cooldown_secs` | `u64` | `300` | нет | Cooldown между применёнными обновлениями ME map. | `me_config_apply_cooldown_secs = 120` |
+| `general.proxy_secret_stable_snapshots` | `u8` | `2` | должно быть `> 0` | Количество одинаковых snapshot перед runtime-rotation proxy-secret. | `proxy_secret_stable_snapshots = 3` |
+| `general.proxy_secret_rotate_runtime` | `bool` | `true` | нет | Включает runtime-ротацию proxy-secret. | `proxy_secret_rotate_runtime = true` |
+| `general.proxy_secret_len_max` | `usize` | `256` | диапазон `[32,4096]` | Верхний лимит длины принимаемого proxy-secret. | `proxy_secret_len_max = 512` |
+| `general.update_every` | `Option<u64>` | `300` | если задано: `> 0`; если `null`: fallback на legacy минимум | Единый интервал refresh для ME config + secret updater. | `update_every = 300` |
+| `general.me_pool_drain_ttl_secs` | `u64` | `90` | нет | Время, когда stale writer ещё может использоваться как fallback. | `me_pool_drain_ttl_secs = 120` |
+| `general.me_pool_min_fresh_ratio` | `f32` | `0.8` | диапазон `[0.0,1.0]` | Порог покрытия fresh-поколения перед drain старого поколения. | `me_pool_min_fresh_ratio = 0.9` |
+| `general.me_reinit_drain_timeout_secs` | `u64` | `120` | `0` = без force-close; если `>0 && < TTL`, поднимается до TTL | Таймаут force-close для draining stale writer. | `me_reinit_drain_timeout_secs = 0` |
+| `general.auto_degradation_enabled` | `bool` | `true` | нет | Зарезервированный флаг совместимости в текущей ревизии (активного runtime-consumer нет). | `auto_degradation_enabled = true` |
+| `general.degradation_min_unavailable_dc_groups` | `u8` | `2` | нет | Зарезервированный порог совместимости в текущей ревизии (активного runtime-consumer нет). | `degradation_min_unavailable_dc_groups = 2` |
+
+## Устаревшие / legacy параметры
+
+| Параметр | Статус | Замена | Текущее поведение | Рекомендация миграции |
+|---|---|---|---|---|
+| `general.middle_proxy_nat_stun` | Deprecated | `network.stun_servers` | Добавляется в `network.stun_servers`, только если `network.stun_servers` не задан явно. | Перенести значение в `network.stun_servers`, legacy-ключ удалить. |
+| `general.middle_proxy_nat_stun_servers` | Deprecated | `network.stun_servers` | Добавляется в `network.stun_servers`, только если `network.stun_servers` не задан явно. | Перенести значения в `network.stun_servers`, legacy-ключ удалить. |
+| `general.proxy_secret_auto_reload_secs` | Deprecated | `general.update_every` | Используется только если `update_every = null` (legacy fallback). | Явно задать `general.update_every`, legacy-ключ удалить. |
+| `general.proxy_config_auto_reload_secs` | Deprecated | `general.update_every` | Используется только если `update_every = null` (legacy fallback). | Явно задать `general.update_every`, legacy-ключ удалить. |
+
+## Как конфигурируются Upstreams
+
+### Схема upstream
+
+| Поле | Применимость | Тип | Обязательно | Default | Назначение |
+|---|---|---|---|---|---|
+| `[[upstreams]].type` | все upstream | `"direct" \| "socks4" \| "socks5"` | да | n/a | Тип upstream транспорта. |
+| `[[upstreams]].weight` | все upstream | `u16` | нет | `1` | Базовый вес в weighted-random выборе. |
+| `[[upstreams]].enabled` | все upstream | `bool` | нет | `true` | Выключенные записи игнорируются на старте. |
+| `[[upstreams]].scopes` | все upstream | `String` | нет | `""` | Список scope-токенов через запятую для маршрутизации. |
+| `interface` | `direct` | `Option<String>` | нет | `null` | Имя интерфейса (например `eth0`) или literal локальный IP. |
+| `bind_addresses` | `direct` | `Option<Vec<IpAddr>>` | нет | `null` | Явные кандидаты source IP (имеют приоритет над `interface`). |
+| `address` | `socks4` | `String` | да | n/a | Адрес SOCKS4 сервера (`ip:port` или `host:port`). |
+| `interface` | `socks4` | `Option<String>` | нет | `null` | Используется только если `address` задан как `ip:port`. |
+| `user_id` | `socks4` | `Option<String>` | нет | `null` | SOCKS4 user ID в CONNECT-запросе. |
+| `address` | `socks5` | `String` | да | n/a | Адрес SOCKS5 сервера (`ip:port` или `host:port`). |
+| `interface` | `socks5` | `Option<String>` | нет | `null` | Используется только если `address` задан как `ip:port`. |
+| `username` | `socks5` | `Option<String>` | нет | `null` | Логин SOCKS5 auth. |
+| `password` | `socks5` | `Option<String>` | нет | `null` | Пароль SOCKS5 auth. |
+
+### Runtime-правила
+
+1. Если `[[upstreams]]` отсутствует, loader добавляет один upstream `direct` по умолчанию.
+2. Scope-фильтрация — по точному совпадению токена:
+- если scope запроса задан -> используются только записи, где `scopes` содержит такой же токен;
+- если scope запроса не задан -> используются только записи с пустым `scopes`.
+3. Среди healthy upstream используется weighted-random выбор: `weight * latency_factor`.
+4. Если в отфильтрованном наборе нет healthy upstream, выбирается случайный из отфильтрованных.
+5. Порядок выбора bind для `direct`:
+- сначала `bind_addresses` (только IP нужного семейства);
+- если одновременно заданы `interface` (имя) и `bind_addresses`, каждый IP проверяется на принадлежность интерфейсу;
+- несовпадающие IP отбрасываются с `WARN`;
+- если валидных IP не осталось, используется unbound direct connect (`bind_ip=None`);
+- если `bind_addresses` не подходит, применяется `interface` (literal IP или адрес интерфейса).
+6. Для `socks4/socks5` с `address` в виде hostname интерфейсный bind не поддерживается и игнорируется с предупреждением.
+7. Runtime DNS overrides применяются к резолвингу hostname в upstream-подключениях.
+8. В ME-режиме выбранный upstream также используется для ME TCP dial path.
+9. В ME-режиме для `direct` upstream с bind/interface STUN-рефлексия выполняется bind-aware для KDF материала.
+10. В ME-режиме для SOCKS upstream используются `BND.ADDR/BND.PORT` для KDF, если адрес валиден/публичен и соответствует IP family.
+
+## Примеры конфигурации Upstreams
+
+### Пример 1: минимальный direct upstream
+
+```toml
+[[upstreams]]
+type = "direct"
+weight = 1
+enabled = true
+```
+
+### Пример 2: direct с interface + явными bind IP
+
+```toml
+[[upstreams]]
+type = "direct"
+interface = "eth0"
+bind_addresses = ["192.168.1.100", "192.168.1.101"]
+weight = 3
+enabled = true
+```
+
+### Пример 3: SOCKS5 upstream с аутентификацией
+
+```toml
+[[upstreams]]
+type = "socks5"
+address = "198.51.100.30:1080"
+username = "proxy-user"
+password = "proxy-pass"
+weight = 2
+enabled = true
+```
+
+### Пример 4: смешанные upstream с scopes
+
+```toml
+[[upstreams]]
+type = "direct"
+weight = 5
+enabled = true
+scopes = ""
+
+[[upstreams]]
+type = "socks5"
+address = "203.0.113.40:1080"
+username = "edge"
+password = "edgepass"
+weight = 3
+enabled = true
+scopes = "premium,me"
+```
+
+### Пример 5: профиль тюнинга под ME
+
+```toml
+[general]
+use_middle_proxy = true
+proxy_secret_path = "proxy-secret"
+middle_proxy_nat_probe = true
+stun_nat_probe_concurrency = 16
+middle_proxy_pool_size = 12
+me_keepalive_enabled = true
+me_keepalive_interval_secs = 20
+me_keepalive_jitter_secs = 4
+me_reconnect_max_concurrent_per_dc = 12
+me_reconnect_backoff_base_ms = 300
+me_reconnect_backoff_cap_ms = 10000
+me_reconnect_fast_retry_count = 10
+hardswap = true
+me_reinit_every_secs = 600
+me_hardswap_warmup_delay_min_ms = 500
+me_hardswap_warmup_delay_max_ms = 1200
+me_hardswap_warmup_extra_passes = 2
+me_hardswap_warmup_pass_backoff_base_ms = 400
+me_config_stable_snapshots = 3
+me_config_apply_cooldown_secs = 120
+proxy_secret_stable_snapshots = 3
+proxy_secret_rotate_runtime = true
+proxy_secret_len_max = 512
+update_every = 300
+me_pool_drain_ttl_secs = 120
+me_pool_min_fresh_ratio = 0.9
+me_reinit_drain_timeout_secs = 180
+
+[timeouts]
+me_one_retry = 8
+me_one_timeout_ms = 1200
+
+[network]
+stun_use = true
+stun_tcp_fallback = true
+stun_servers = [
+  "stun1.l.google.com:19302",
+  "stun2.l.google.com:19302"
+]
+http_ip_detect_urls = [
+  "https://api.ipify.org",
+  "https://ifconfig.me/ip"
+]
+```

docs/XRAY-SINGBOX-ROUTING.ru.md

@@ -0,0 +1,321 @@
+# SNI-маршрутизация в xray-core / sing-box + TLS-fronting
+
+## Термины (в контексте этого кейса)
+
+- **TLS-fronting домен** — домен, который фигурирует в TLS ClientHello как **SNI** (например, `petrovich.ru`): он используется как "маска" на L7 и как ключ маршрутизации в прокси-роутере.
+- **xray-core / sing-box** — локальный или удалённый L7/TLS-роутер (прокси), который:
+  1) принимает входящее TCP/TLS-соединение,
+  2) читает TLS ClientHello,
+  3) извлекает SNI,
+  4) по SNI выбирает outbound/апстрим,
+  5) устанавливает новое TCP-соединение к целевому хосту уже **от себя**.
+- **SNI (Server Name Indication)** — поле в TLS ClientHello, где клиент Telegram сообщает доменное имя для "маскировки"
+- **DNS-resolve на стороне L7-роутера** — если выходной адрес задан доменом (или роутер решил "всё равно идти по SNI"), то DNS резолвится **на стороне xray/sing-box**, а не на стороне Telegram-клиента
+
+---
+
+## Ключевая идея: куда на самом деле идёт соединение решает не то, что вы указали клиенту, а то как L7-роутер трактует SNI
+
+Механика:
+
+1) Telegram-клиенту вы можете указать **IP/домен telemt**,как "сервер".
+2) Между клиентом и telemt стоит xray-core/sing-box, который принимает TCP, читает TLS ClientHello и видит **SNI=petrovich.ru**
+3) Дальше роутер говорит: "Вижу SNI - направить на апстрим/маршрут N"
+4) И устанавливает исходящее соединение не "по тому IP, который пользователь подразумевал", а **по домену из SNI** (или по сопоставлению SNI→outbound), используя для определния его IP собственный DNS-кеш или резолвер
+5) `petrovich.ru` по A-записи указывает **не на IP telemt**, а значит при L7-маршрутизации трафик уйдёт на "оригинальный" сайт за этим доменом, а не в telemt: Telegram-клиент, естественно, не сможет получить ожидаемое поведение, потому что ответить с handshake на той стороне некому
+
+---
+
+## Схема №1 "Как это НЕ работает"
+
+```text
+Telegram Client
+   |
+   |  (указан IP/домен telemt)
+   v
+telemt instance
+````
+
+Ожидание: "я указал telemt -> значит трафик попадёт в telemt" - **нет!**
+
+---
+
+## Схема №2. "Как это реально работает с TLS/L7-роутером и SNI"
+
+```text
+Telegram Client
+   |
+   | 1) TCP/TLS connection: 
+   |    - ClientHello: 
+   |      - SNI=petrovich.ru
+   v
+xray-core / sing-box / любой L7 router
+   |
+   | 2) читает ClientHello -> вытаскивает SNI
+   | 3) выбирает маршрут по SNI
+   | 4) делает DNS для petrovich.ru 
+   | 5) подключается к полученному IP по TLS с этим SNI
+   v
+"Оригинальный" сайт, A-запись которого не на telemt
+   |
+   X  не telemt -> Telegram-клиент не коннектится как ожидалось
+```
+
+---
+
+## Почему указанный в клиенте IP/домен telemt "не спасает"
+
+Потому что в таком режиме xray/sing-box выступает как **точка терминации TCP/TLS**, можно сказать - TLS-инспектор на уровне ClientHello, это означает:
+
+* TCP-сессия от Telegram-клиента заканчивается на xray/sing-box
+* Дальше создаётся **новая** TCP-сессия "от имени" xray/sing-box к апстриму
+* Выбор апстрима делается правилами роутинга, а в TLS-сценариях самый удобный и распространённый ключ — **SNI**
+
+То есть, "куда идти дальше" определяется логикой L7-роутера:
+
+* либо правилами вида `if SNI == petrovich.ru -> outbound X`,
+* либо более "автоматическим" поведением: `подключаться к тому хосту, который указан в SNI`,
+* плюс кэш DNS и собственные резолверы роутера
+
+---
+
+## Что именно извлекается из TLS ClientHello и почему этого достаточно
+
+TLS ClientHello отправляется **в начале** TLS-сессии и, в классическом TLS без ECH, содержит SNI в открытом виде.
+
+Упрощённо:
+
+```text
+ClientHello:
+  - supported_versions
+  - cipher_suites
+  - extensions:
+      - server_name: petrovich.ru   <-- SNI
+      - alpn: h2/http1.1/...
+      - ...
+```
+
+Роутеру не нужно расшифровывать трафик и завершать TLS "как сервер" — часто достаточно просто прочитать первые пакеты и распарсить ClientHello, чтобы получить SNI и принять решение
+
+---
+
+## Типовой алгоритм SNI-роутинга
+
+1. Принять входящий TCP.
+2. Подождать первые байты.
+3. Определить протокол:
+
+   * если видим TLS ClientHello → парсим SNI/ALPN
+4. Применить route rules:
+
+   * match по `server_name` / `domain` / `tls.sni`
+5. Выбрать outbound:
+
+   * direct / proxy / specific upstream / detour
+6. Установить исходящее соединение:
+
+   * либо на фиксированный IP:порт,
+   * либо на домен через DNS-resolve на стороне роутера
+7. Начать проксирование данных между входом и выходом
+
+---
+
+## Почему "A-запись фронтинг-домена не на telemt" ломает кейс
+
+### Ситуация
+
+* В ClientHello: `SNI = petrovich.ru`
+* DNS: `petrovich.ru -> 203.0.113.77`  - "оригинальный" сайт
+* telemt живёт на: `198.51.100.10`
+
+### Что делает роутер
+
+* Видит SNI `petrovich.ru`
+* Либо:
+
+  * (а) напрямую коннектится к `petrovich.ru:443`, резолвя A-запись в `203.0.113.77`,
+  * либо:
+  * (б) выбирает outbound, который указывает на `petrovich.ru` как destination,
+  * либо:
+  * (в) делает sniffing/override destination по SNI
+
+В итоге исходящий коннект идёт на `203.0.113.77:443`, а не на telemt!
+Другой сервер, другой протокол, другая логика, где telemt не участвует
+
+---
+
+## "Где именно происходит подмена destination на SNI"
+
+Это зависит от конфигурации, но типовые варианты:
+
+### Вариант A: outbound задан доменом (и он совпадает с SNI)
+
+Правило по SNI выбирает outbound, у которого destination задан доменом фронтинга, 
+тогда DNS резолвится на стороне роутера и вы уходите на "оригинальный" хост
+
+### Вариант B: destination override / sniffing
+
+Роутер "снифает" SNI и **перезаписывает** destination на домен из SNI (даже если вход изначально был на IP telemt),
+это особенно коварно: пользователь видит "я подключаюсь к IP telemt", но роутер после sniffing решает иначе
+
+### Вариант C: split DNS / кеш / независимый резолвер
+
+Даже если клиент "где-то" резолвит иначе, это не важно: конечный DNS для исходящего коннекта — на стороне xray/sing-box,
+который может иметь:
+
+* свой DoH/DoT,
+* свой кеш,
+* свои правила fake-ip / system resolver,
+* и, как следствие, своя "карта" **домен/SNI -> IP**
+
+---
+
+## Признаки того, что трафик "утёк на оригинал", а не попал в telemt
+
+* На стороне telemt отсутствуют входящие соединения/логи
+* На стороне роутера видно, что destination — домен фронтинга, а IP соответствует публичному сайту
+* TLS-метрики/сертификат на выходе соответствует "оригинальному" сайту в записах трафика
+* Telegram-клиент получает неожиданный тип ответов/ошибку handshaking/timeout в debug-режиме
+
+---
+
+## Best-practice решение для этого кейса: свой домен фронтинга + заглушка на telemt + Let's Encrypt
+
+### Цель
+
+Сделать так, чтобы:
+
+* SNI (фронтинг-домен) **резолвился в IP telemt**,
+* на IP telemt реально был TLS-сервис с валидным сертификатом под этот домен,
+* даже если кто-то "попробует открыть домен как сайт", он увидит нормальную заглушку, а не "пустоту"
+
+### Что это даёт
+
+* xray/sing-box, маршрутизируя по SNI, будет неизбежно приходить на telemt, потому что DNS(SNI-домен) → IP telemt
+* Внешний вид будет правдоподобным: обычный домен с обычным сертификатом
+* Устойчивость: меньше сюрпризов от DNS-кеша/перерезолва/"умных" правил роутера
+
+---
+
+## Рекомендуемая схема (целевое состояние)
+
+```text
+Telegram Client
+   |
+   | TLS ClientHello: SNI = hello.example.com
+   v
+xray-core / sing-box
+   |
+   | Route by SNI -> outbound -> connect to hello.example.com:443
+   | DNS(hello.example.com) = IP telemt
+   v
+telemt instance (IP telemt)
+   |
+   | TLS cert for hello.example.com (Let's Encrypt)
+   | + сайт-заглушка / health endpoint
+   v
+OK
+```
+
+---
+
+## Практический чеклист (минимальный)
+
+1. Купить/иметь домен: `hello.example.com`
+2. В DNS:
+
+   * `A hello.example.com -> <IP telemt>`
+   * (опционально) AAAA, если используете IPv6 и он стабилен
+3. На telemt-хосте:
+
+   * поднять TLS endpoint на 443 с валидным сертификатом LE под `hello.example.com`
+   * отдать "заглушку" (например, статический сайт), чтобы домен выглядел как обычный веб-сервис
+4. В xray/sing-box правилах:
+
+   * маршрутизировать нужный трафик по SNI = `hello.example.com` в "правильный" outbound (к telemt)
+   * избегать конфигураций, где destination override уводит на чужой домен
+5. Важно:
+
+   * если вы используете кеш DNS на роутере — сбросить/обновить его после смены A-записи
+
+---
+
+## Пояснение про сайт-заглушку
+
+Для эмуляции TLS, telemt имеет подсистему TLS-F в `src/tls_front`:
+- её модуль - fetcher, собирает TLS-профили, чтоб максимально поведенчески корректно повторять TLS конкретно указанного сайта
+
+Когда вы указываете сайт, который не отвечает по TLS:
+- fetcher не может собрать TLS-профиль и происходит fallback на `fake_cert_len` - примитивный алгоритм,
+- он забивает служебную информацию TLS рандомными байтами,
+- простые системы DPI не распознают это
+- однако, продвинутые системы, такие как nEdge или Fraud Control в сетях мобильной связи легко заблокируют или замедлят такой трафик
+
+Создав сайт-заглушку с Let's Encrypt сертификатом, вы даёте TLS-F возможность получить данные сертификата и корректно его "повторять" в дальнейшем
+
+---
+
+## Вариант конфиг-подхода: "SNI строго привязываем к telemt - фиксированный IP"
+
+Чтобы полностью исключить зависимость от DNS если вам это нужно, можно сделать outbound, который ходит на **фиксированный IP telemt**, но при этом выставляет SNI/Host как `hello.example.com`.
+
+Идея:
+
+* destination: `IP:443`
+* SNI: `hello.example.com`
+* сертификат на telemt именно под `hello.example.com`
+
+Так вы получаете:
+
+* TLS выглядит корректно, ведь SNI совпадает с сертификатом,
+* а routing никогда не уйдёт на "оригинал", потому что A-запись указывает на telemt и контроллируется вами!
+
+Но в вашем описании проблема как раз в том, что роутер "сам решает по SNI и резолвит домен", поэтому самый универсальный вариант — сделать так, чтобы DNS всегда приводил в telemt
+
+---
+
+## Пример логики правил на псевдоконфиге L7-роутера
+
+```text
+if inbound is TLS and sni == "hello.example.com":
+    route -> outbound "telemt"
+else:
+    route -> outbound "default"
+```
+
+Outbound `telemt`:
+
+* destination: `hello.example.com:443`
+* TLS enabled
+* SNI: `hello.example.com`
+
+---
+
+## Отдельно: что может неожиданно сломать даже "правильный" DNS
+
+* **Кеширование DNS** на xray/sing-box или на системном резолвере, особенно при смене A-записи
+* **Split-horizon DNS**: разные ответы внутри/снаружи, попытки подмены/терминирования в других точках
+* **IPv6**: если есть AAAA и он указывает не туда, роутер может предпочесть IPv6: помните, что поддержка v6 нестабильна и не рекомендуется в prod
+* **DoH/DoT** на роутере: он может резолвить не тем резолвером, которым вы проверяли
+
+Минимальная гигиена:
+
+* контролировать A/AAAA,
+* держать TTL разумным,
+* проверять, каким резолвером пользуется именно роутер,
+* при необходимости отключить/ограничить destination override
+
+---
+
+## Итог
+
+В режиме TLS-fronting с xray-core/sing-box как L7/TLS-роутером **SNI становится приоритетным "source-of-truth" для маршрутизации**
+
+Если фронтинг-домен по DNS указывает не на IP telemt, роутер честно уводит трафик на "оригинальный" сайт, потому что он строит исходящее соединение "по SNI"
+
+Надёжное решение для этого кейса:
+
+* использовать **свой домен** для фронтинга,
+* направить его **A/AAAA** на IP telemt,
+* поднять на telemt **TLS-сервис с Let’s Encrypt сертификатом** под этот домен,
+* (желательно) держать **сайт-заглушку**, чтобы 443 выглядел как обычный HTTPS

install.sh

@@ -0,0 +1,73 @@
+sudo bash -c '
+set -e
+
+# --- Проверка на существующую установку ---
+if systemctl list-unit-files | grep -q telemt.service; then
+    # --- РЕЖИМ ОБНОВЛЕНИЯ ---
+    echo "--- Обнаружена существующая установка Telemt. Запускаю обновление... ---"
+
+    echo "[*] Остановка службы telemt..."
+    systemctl stop telemt || true # Игнорируем ошибку, если служба уже остановлена
+
+    echo "[1/2] Скачивание последней версии Telemt..."
+    wget -qO- "https://github.com/telemt/telemt/releases/latest/download/telemt-$(uname -m)-linux-$(ldd --version 2>&1 | grep -iq musl && echo musl || echo gnu).tar.gz" | tar -xz
+
+    echo "[1/2] Замена исполняемого файла в /usr/local/bin..."
+    mv telemt /usr/local/bin/telemt
+    chmod +x /usr/local/bin/telemt
+
+    echo "[2/2] Запуск службы..."
+    systemctl start telemt
+
+    echo "--- Обновление Telemt успешно завершено! ---"
+    echo
+    echo "Для проверки статуса службы выполните:"
+    echo "   systemctl status telemt"
+
+else
+    # --- РЕЖИМ НОВОЙ УСТАНОВКИ ---
+    echo "--- Начало автоматической установки Telemt ---"
+
+    # Шаг 1: Скачивание и установка бинарного файла
+    echo "[1/5] Скачивание последней версии Telemt..."
+    wget -qO- "https://github.com/telemt/telemt/releases/latest/download/telemt-$(uname -m)-linux-$(ldd --version 2>&1 | grep -iq musl && echo musl || echo gnu).tar.gz" | tar -xz
+
+    echo "[1/5] Перемещение исполняемого файла в /usr/local/bin и установка прав..."
+    mv telemt /usr/local/bin/telemt
+    chmod +x /usr/local/bin/telemt
+
+    # Шаг 2: Генерация секрета
+    echo "[2/5] Генерация секретного ключа..."
+    SECRET=$(openssl rand -hex 16)
+
+    # Шаг 3: Создание файла конфигурации
+    echo "[3/5] Создание файла конфигурации /etc/telemt.toml..."
+    printf "# === General Settings ===\n[general]\n[general.modes]\nclassic = false\nsecure = false\ntls = true\n\n# === Anti-Censorship & Masking ===\n[censorship]\n# !!! ВАЖНО: Замените на ваш домен или домен, который вы хотите использовать для маскировки !!!\ntls_domain = \"petrovich.ru\"\n\n[access.users]\nhello = \"%s\"\n" "$SECRET" > /etc/telemt.toml
+
+    # Шаг 4: Создание службы Systemd
+    echo "[4/5] Создание службы systemd..."
+    printf "[Unit]\nDescription=Telemt Proxy\nAfter=network.target\n\n[Service]\nType=simple\nExecStart=/usr/local/bin/telemt /etc/telemt.toml\nRestart=on-failure\nRestartSec=5\nLimitNOFILE=65536\n\n[Install]\nWantedBy=multi-user.target\n" > /etc/systemd/system/telemt.service
+
+    # Шаг 5: Запуск службы
+    echo "[5/5] Перезагрузка systemd, запуск и включение службы telemt..."
+    systemctl daemon-reload
+    systemctl start telemt
+    systemctl enable telemt
+
+    echo "--- Установка и запуск Telemt успешно завершены! ---"
+    echo
+    echo "ВАЖНАЯ ИНФОРМАЦИЯ:"
+    echo "==================="
+    echo "1. Вам НЕОБХОДИМО отредактировать файл /etc/telemt.toml и заменить '\''petrovich.ru'\'' на другой домен"
+    echo "   с помощью команды:"
+    echo "   nano /etc/telemt.toml"
+    echo "   После редактирования файла перезапустите службу командой:"
+    echo "   sudo systemctl restart telemt"
+    echo
+    echo "2. Для проверки статуса службы выполните команду:"
+    echo "   systemctl status telemt"
+    echo
+    echo "3. Для получения ссылок на подключение выполните команду:"
+    echo "   journalctl -u telemt -n -g '\''links'\'' --no-pager -o cat | tac"
+fi
+'

src/api/config_store.rs

@@ -0,0 +1,107 @@
+use std::io::Write;
+use std::path::{Path, PathBuf};
+
+use hyper::header::IF_MATCH;
+use sha2::{Digest, Sha256};
+
+use crate::config::ProxyConfig;
+
+use super::model::ApiFailure;
+
+pub(super) fn parse_if_match(headers: &hyper::HeaderMap) -> Option<String> {
+    headers
+        .get(IF_MATCH)
+        .and_then(|value| value.to_str().ok())
+        .map(str::trim)
+        .filter(|value| !value.is_empty())
+        .map(|value| value.trim_matches('"').to_string())
+}
+
+pub(super) async fn ensure_expected_revision(
+    config_path: &Path,
+    expected_revision: Option<&str>,
+) -> Result<(), ApiFailure> {
+    let Some(expected) = expected_revision else {
+        return Ok(());
+    };
+    let current = current_revision(config_path).await?;
+    if current != expected {
+        return Err(ApiFailure::new(
+            hyper::StatusCode::CONFLICT,
+            "revision_conflict",
+            "Config revision mismatch",
+        ));
+    }
+    Ok(())
+}
+
+pub(super) async fn current_revision(config_path: &Path) -> Result<String, ApiFailure> {
+    let content = tokio::fs::read_to_string(config_path)
+        .await
+        .map_err(|e| ApiFailure::internal(format!("failed to read config: {}", e)))?;
+    Ok(compute_revision(&content))
+}
+
+pub(super) fn compute_revision(content: &str) -> String {
+    let mut hasher = Sha256::new();
+    hasher.update(content.as_bytes());
+    hex::encode(hasher.finalize())
+}
+
+pub(super) async fn load_config_from_disk(config_path: &Path) -> Result<ProxyConfig, ApiFailure> {
+    let config_path = config_path.to_path_buf();
+    tokio::task::spawn_blocking(move || ProxyConfig::load(config_path))
+        .await
+        .map_err(|e| ApiFailure::internal(format!("failed to join config loader: {}", e)))?
+        .map_err(|e| ApiFailure::internal(format!("failed to load config: {}", e)))
+}
+
+pub(super) async fn save_config_to_disk(
+    config_path: &Path,
+    cfg: &ProxyConfig,
+) -> Result<String, ApiFailure> {
+    let serialized = toml::to_string_pretty(cfg)
+        .map_err(|e| ApiFailure::internal(format!("failed to serialize config: {}", e)))?;
+    write_atomic(config_path.to_path_buf(), serialized.clone()).await?;
+    Ok(compute_revision(&serialized))
+}
+
+async fn write_atomic(path: PathBuf, contents: String) -> Result<(), ApiFailure> {
+    tokio::task::spawn_blocking(move || write_atomic_sync(&path, &contents))
+        .await
+        .map_err(|e| ApiFailure::internal(format!("failed to join writer: {}", e)))?
+        .map_err(|e| ApiFailure::internal(format!("failed to write config: {}", e)))
+}
+
+fn write_atomic_sync(path: &Path, contents: &str) -> std::io::Result<()> {
+    let parent = path.parent().unwrap_or_else(|| Path::new("."));
+    std::fs::create_dir_all(parent)?;
+
+    let tmp_name = format!(
+        ".{}.tmp-{}",
+        path.file_name()
+            .and_then(|s| s.to_str())
+            .unwrap_or("config.toml"),
+        rand::random::<u64>()
+    );
+    let tmp_path = parent.join(tmp_name);
+
+    let write_result = (|| {
+        let mut file = std::fs::OpenOptions::new()
+            .create_new(true)
+            .write(true)
+            .open(&tmp_path)?;
+        file.write_all(contents.as_bytes())?;
+        file.sync_all()?;
+        std::fs::rename(&tmp_path, path)?;
+        if let Ok(dir) = std::fs::File::open(parent) {
+            let _ = dir.sync_all();
+        }
+        Ok(())
+    })();
+
+    if write_result.is_err() {
+        let _ = std::fs::remove_file(&tmp_path);
+    }
+    write_result
+}

src/api/mod.rs

@@ -0,0 +1,443 @@
+use std::convert::Infallible;
+use std::net::{IpAddr, SocketAddr};
+use std::path::PathBuf;
+use std::sync::Arc;
+use std::sync::atomic::{AtomicU64, Ordering};
+
+use http_body_util::{BodyExt, Full};
+use hyper::body::{Bytes, Incoming};
+use hyper::header::AUTHORIZATION;
+use hyper::server::conn::http1;
+use hyper::service::service_fn;
+use hyper::{Method, Request, Response, StatusCode};
+use serde::Serialize;
+use serde::de::DeserializeOwned;
+use tokio::net::TcpListener;
+use tokio::sync::{Mutex, watch};
+use tracing::{debug, info, warn};
+
+use crate::config::ProxyConfig;
+use crate::ip_tracker::UserIpTracker;
+use crate::stats::Stats;
+use crate::transport::middle_proxy::MePool;
+use crate::transport::UpstreamManager;
+
+mod config_store;
+mod model;
+mod runtime_stats;
+mod users;
+
+use config_store::{current_revision, parse_if_match};
+use model::{
+    ApiFailure, CreateUserRequest, ErrorBody, ErrorResponse, HealthData, PatchUserRequest,
+    RotateSecretRequest, SuccessResponse, SummaryData,
+};
+use runtime_stats::{
+    MinimalCacheEntry, build_dcs_data, build_me_writers_data, build_minimal_all_data,
+    build_upstreams_data, build_zero_all_data,
+};
+use users::{create_user, delete_user, patch_user, rotate_secret, users_from_config};
+
+#[derive(Clone)]
+pub(super) struct ApiShared {
+    pub(super) stats: Arc<Stats>,
+    pub(super) ip_tracker: Arc<UserIpTracker>,
+    pub(super) me_pool: Option<Arc<MePool>>,
+    pub(super) upstream_manager: Arc<UpstreamManager>,
+    pub(super) config_path: PathBuf,
+    pub(super) startup_detected_ip_v4: Option<IpAddr>,
+    pub(super) startup_detected_ip_v6: Option<IpAddr>,
+    pub(super) mutation_lock: Arc<Mutex<()>>,
+    pub(super) minimal_cache: Arc<Mutex<Option<MinimalCacheEntry>>>,
+    pub(super) request_id: Arc<AtomicU64>,
+}
+
+impl ApiShared {
+    fn next_request_id(&self) -> u64 {
+        self.request_id.fetch_add(1, Ordering::Relaxed)
+    }
+}
+
+pub async fn serve(
+    listen: SocketAddr,
+    stats: Arc<Stats>,
+    ip_tracker: Arc<UserIpTracker>,
+    me_pool: Option<Arc<MePool>>,
+    upstream_manager: Arc<UpstreamManager>,
+    config_rx: watch::Receiver<Arc<ProxyConfig>>,
+    config_path: PathBuf,
+    startup_detected_ip_v4: Option<IpAddr>,
+    startup_detected_ip_v6: Option<IpAddr>,
+) {
+    let listener = match TcpListener::bind(listen).await {
+        Ok(listener) => listener,
+        Err(error) => {
+            warn!(
+                error = %error,
+                listen = %listen,
+                "Failed to bind API listener"
+            );
+            return;
+        }
+    };
+
+    info!("API endpoint: http://{}/v1/*", listen);
+
+    let shared = Arc::new(ApiShared {
+        stats,
+        ip_tracker,
+        me_pool,
+        upstream_manager,
+        config_path,
+        startup_detected_ip_v4,
+        startup_detected_ip_v6,
+        mutation_lock: Arc::new(Mutex::new(())),
+        minimal_cache: Arc::new(Mutex::new(None)),
+        request_id: Arc::new(AtomicU64::new(1)),
+    });
+
+    loop {
+        let (stream, peer) = match listener.accept().await {
+            Ok(v) => v,
+            Err(error) => {
+                warn!(error = %error, "API accept error");
+                continue;
+            }
+        };
+
+        let shared_conn = shared.clone();
+        let config_rx_conn = config_rx.clone();
+        tokio::spawn(async move {
+            let svc = service_fn(move |req: Request<Incoming>| {
+                let shared_req = shared_conn.clone();
+                let config_rx_req = config_rx_conn.clone();
+                async move { handle(req, peer, shared_req, config_rx_req).await }
+            });
+            if let Err(error) = http1::Builder::new()
+                .serve_connection(hyper_util::rt::TokioIo::new(stream), svc)
+                .await
+            {
+                debug!(error = %error, "API connection error");
+            }
+        });
+    }
+}
+
+async fn handle(
+    req: Request<Incoming>,
+    peer: SocketAddr,
+    shared: Arc<ApiShared>,
+    config_rx: watch::Receiver<Arc<ProxyConfig>>,
+) -> Result<Response<Full<Bytes>>, Infallible> {
+    let request_id = shared.next_request_id();
+    let cfg = config_rx.borrow().clone();
+    let api_cfg = &cfg.server.api;
+
+    if !api_cfg.enabled {
+        return Ok(error_response(
+            request_id,
+            ApiFailure::new(
+                StatusCode::SERVICE_UNAVAILABLE,
+                "api_disabled",
+                "API is disabled",
+            ),
+        ));
+    }
+
+    if !api_cfg.whitelist.is_empty()
+        && !api_cfg
+            .whitelist
+            .iter()
+            .any(|net| net.contains(peer.ip()))
+    {
+        return Ok(error_response(
+            request_id,
+            ApiFailure::new(StatusCode::FORBIDDEN, "forbidden", "Source IP is not allowed"),
+        ));
+    }
+
+    if !api_cfg.auth_header.is_empty() {
+        let auth_ok = req
+            .headers()
+            .get(AUTHORIZATION)
+            .and_then(|v| v.to_str().ok())
+            .map(|v| v == api_cfg.auth_header)
+            .unwrap_or(false);
+        if !auth_ok {
+            return Ok(error_response(
+                request_id,
+                ApiFailure::new(
+                    StatusCode::UNAUTHORIZED,
+                    "unauthorized",
+                    "Missing or invalid Authorization header",
+                ),
+            ));
+        }
+    }
+
+    let method = req.method().clone();
+    let path = req.uri().path().to_string();
+    let body_limit = api_cfg.request_body_limit_bytes;
+
+    let result: Result<Response<Full<Bytes>>, ApiFailure> = async {
+        match (method.as_str(), path.as_str()) {
+            ("GET", "/v1/health") => {
+                let revision = current_revision(&shared.config_path).await?;
+                let data = HealthData {
+                    status: "ok",
+                    read_only: api_cfg.read_only,
+                };
+                Ok(success_response(StatusCode::OK, data, revision))
+            }
+            ("GET", "/v1/stats/summary") => {
+                let revision = current_revision(&shared.config_path).await?;
+                let data = SummaryData {
+                    uptime_seconds: shared.stats.uptime_secs(),
+                    connections_total: shared.stats.get_connects_all(),
+                    connections_bad_total: shared.stats.get_connects_bad(),
+                    handshake_timeouts_total: shared.stats.get_handshake_timeouts(),
+                    configured_users: cfg.access.users.len(),
+                };
+                Ok(success_response(StatusCode::OK, data, revision))
+            }
+            ("GET", "/v1/stats/zero/all") => {
+                let revision = current_revision(&shared.config_path).await?;
+                let data = build_zero_all_data(&shared.stats, cfg.access.users.len());
+                Ok(success_response(StatusCode::OK, data, revision))
+            }
+            ("GET", "/v1/stats/upstreams") => {
+                let revision = current_revision(&shared.config_path).await?;
+                let data = build_upstreams_data(shared.as_ref(), api_cfg);
+                Ok(success_response(StatusCode::OK, data, revision))
+            }
+            ("GET", "/v1/stats/minimal/all") => {
+                let revision = current_revision(&shared.config_path).await?;
+                let data = build_minimal_all_data(shared.as_ref(), api_cfg).await;
+                Ok(success_response(StatusCode::OK, data, revision))
+            }
+            ("GET", "/v1/stats/me-writers") => {
+                let revision = current_revision(&shared.config_path).await?;
+                let data = build_me_writers_data(shared.as_ref(), api_cfg).await;
+                Ok(success_response(StatusCode::OK, data, revision))
+            }
+            ("GET", "/v1/stats/dcs") => {
+                let revision = current_revision(&shared.config_path).await?;
+                let data = build_dcs_data(shared.as_ref(), api_cfg).await;
+                Ok(success_response(StatusCode::OK, data, revision))
+            }
+            ("GET", "/v1/stats/users") | ("GET", "/v1/users") => {
+                let revision = current_revision(&shared.config_path).await?;
+                let users = users_from_config(
+                    &cfg,
+                    &shared.stats,
+                    &shared.ip_tracker,
+                    shared.startup_detected_ip_v4,
+                    shared.startup_detected_ip_v6,
+                )
+                .await;
+                Ok(success_response(StatusCode::OK, users, revision))
+            }
+            ("POST", "/v1/users") => {
+                if api_cfg.read_only {
+                    return Ok(error_response(
+                        request_id,
+                        ApiFailure::new(
+                            StatusCode::FORBIDDEN,
+                            "read_only",
+                            "API runs in read-only mode",
+                        ),
+                    ));
+                }
+                let expected_revision = parse_if_match(req.headers());
+                let body = read_json::<CreateUserRequest>(req.into_body(), body_limit).await?;
+                let (data, revision) = create_user(body, expected_revision, &shared).await?;
+                Ok(success_response(StatusCode::CREATED, data, revision))
+            }
+            _ => {
+                if let Some(user) = path.strip_prefix("/v1/users/")
+                    && !user.is_empty()
+                    && !user.contains('/')
+                {
+                    if method == Method::GET {
+                        let revision = current_revision(&shared.config_path).await?;
+                        let users = users_from_config(
+                            &cfg,
+                            &shared.stats,
+                            &shared.ip_tracker,
+                            shared.startup_detected_ip_v4,
+                            shared.startup_detected_ip_v6,
+                        )
+                        .await;
+                        if let Some(user_info) = users.into_iter().find(|entry| entry.username == user)
+                        {
+                            return Ok(success_response(StatusCode::OK, user_info, revision));
+                        }
+                        return Ok(error_response(
+                            request_id,
+                            ApiFailure::new(StatusCode::NOT_FOUND, "not_found", "User not found"),
+                        ));
+                    }
+                    if method == Method::PATCH {
+                        if api_cfg.read_only {
+                            return Ok(error_response(
+                                request_id,
+                                ApiFailure::new(
+                                    StatusCode::FORBIDDEN,
+                                    "read_only",
+                                    "API runs in read-only mode",
+                                ),
+                            ));
+                        }
+                        let expected_revision = parse_if_match(req.headers());
+                        let body = read_json::<PatchUserRequest>(req.into_body(), body_limit).await?;
+                        let (data, revision) =
+                            patch_user(user, body, expected_revision, &shared).await?;
+                        return Ok(success_response(StatusCode::OK, data, revision));
+                    }
+                    if method == Method::DELETE {
+                        if api_cfg.read_only {
+                            return Ok(error_response(
+                                request_id,
+                                ApiFailure::new(
+                                    StatusCode::FORBIDDEN,
+                                    "read_only",
+                                    "API runs in read-only mode",
+                                ),
+                            ));
+                        }
+                        let expected_revision = parse_if_match(req.headers());
+                        let (deleted_user, revision) =
+                            delete_user(user, expected_revision, &shared).await?;
+                        return Ok(success_response(StatusCode::OK, deleted_user, revision));
+                    }
+                    if method == Method::POST
+                        && let Some(base_user) = user.strip_suffix("/rotate-secret")
+                        && !base_user.is_empty()
+                        && !base_user.contains('/')
+                    {
+                        if api_cfg.read_only {
+                            return Ok(error_response(
+                                request_id,
+                                ApiFailure::new(
+                                    StatusCode::FORBIDDEN,
+                                    "read_only",
+                                    "API runs in read-only mode",
+                                ),
+                            ));
+                        }
+                        let expected_revision = parse_if_match(req.headers());
+                        let body =
+                            read_optional_json::<RotateSecretRequest>(req.into_body(), body_limit)
+                                .await?;
+                        let (data, revision) =
+                            rotate_secret(base_user, body.unwrap_or_default(), expected_revision, &shared)
+                                .await?;
+                        return Ok(success_response(StatusCode::OK, data, revision));
+                    }
+                    if method == Method::POST {
+                        return Ok(error_response(
+                            request_id,
+                            ApiFailure::new(StatusCode::NOT_FOUND, "not_found", "Route not found"),
+                        ));
+                    }
+                    return Ok(error_response(
+                        request_id,
+                        ApiFailure::new(
+                            StatusCode::METHOD_NOT_ALLOWED,
+                            "method_not_allowed",
+                            "Unsupported HTTP method for this route",
+                        ),
+                    ));
+                }
+                Ok(error_response(
+                    request_id,
+                    ApiFailure::new(StatusCode::NOT_FOUND, "not_found", "Route not found"),
+                ))
+            }
+        }
+    }
+    .await;
+
+    match result {
+        Ok(resp) => Ok(resp),
+        Err(error) => Ok(error_response(request_id, error)),
+    }
+}
+
+fn success_response<T: Serialize>(
+    status: StatusCode,
+    data: T,
+    revision: String,
+) -> Response<Full<Bytes>> {
+    let payload = SuccessResponse {
+        ok: true,
+        data,
+        revision,
+    };
+    let body = serde_json::to_vec(&payload).unwrap_or_else(|_| b"{\"ok\":false}".to_vec());
+    Response::builder()
+        .status(status)
+        .header("content-type", "application/json; charset=utf-8")
+        .body(Full::new(Bytes::from(body)))
+        .unwrap()
+}
+
+fn error_response(request_id: u64, failure: ApiFailure) -> Response<Full<Bytes>> {
+    let payload = ErrorResponse {
+        ok: false,
+        error: ErrorBody {
+            code: failure.code,
+            message: failure.message,
+        },
+        request_id,
+    };
+    let body = serde_json::to_vec(&payload).unwrap_or_else(|_| {
+        format!(
+            "{{\"ok\":false,\"error\":{{\"code\":\"internal_error\",\"message\":\"serialization failed\"}},\"request_id\":{}}}",
+            request_id
+        )
+        .into_bytes()
+    });
+    Response::builder()
+        .status(failure.status)
+        .header("content-type", "application/json; charset=utf-8")
+        .body(Full::new(Bytes::from(body)))
+        .unwrap()
+}
+
+async fn read_json<T: DeserializeOwned>(body: Incoming, limit: usize) -> Result<T, ApiFailure> {
+    let bytes = read_body_with_limit(body, limit).await?;
+    serde_json::from_slice(&bytes).map_err(|_| ApiFailure::bad_request("Invalid JSON body"))
+}
+
+async fn read_optional_json<T: DeserializeOwned>(
+    body: Incoming,
+    limit: usize,
+) -> Result<Option<T>, ApiFailure> {
+    let bytes = read_body_with_limit(body, limit).await?;
+    if bytes.is_empty() {
+        return Ok(None);
+    }
+    serde_json::from_slice(&bytes)
+        .map(Some)
+        .map_err(|_| ApiFailure::bad_request("Invalid JSON body"))
+}
+
+async fn read_body_with_limit(body: Incoming, limit: usize) -> Result<Vec<u8>, ApiFailure> {
+    let mut collected = Vec::new();
+    let mut body = body;
+    while let Some(frame_result) = body.frame().await {
+        let frame = frame_result.map_err(|_| ApiFailure::bad_request("Invalid request body"))?;
+        if let Some(chunk) = frame.data_ref() {
+            if collected.len().saturating_add(chunk.len()) > limit {
+                return Err(ApiFailure::new(
+                    StatusCode::PAYLOAD_TOO_LARGE,
+                    "payload_too_large",
+                    format!("Body exceeds {} bytes", limit),
+                ));
+            }
+            collected.extend_from_slice(chunk);
+        }
+    }
+    Ok(collected)
+}

src/api/model.rs

@@ -0,0 +1,444 @@
+use std::net::IpAddr;
+
+use chrono::{DateTime, Utc};
+use hyper::StatusCode;
+use rand::Rng;
+use serde::{Deserialize, Serialize};
+
+const MAX_USERNAME_LEN: usize = 64;
+
+#[derive(Debug)]
+pub(super) struct ApiFailure {
+    pub(super) status: StatusCode,
+    pub(super) code: &'static str,
+    pub(super) message: String,
+}
+
+impl ApiFailure {
+    pub(super) fn new(status: StatusCode, code: &'static str, message: impl Into<String>) -> Self {
+        Self {
+            status,
+            code,
+            message: message.into(),
+        }
+    }
+
+    pub(super) fn internal(message: impl Into<String>) -> Self {
+        Self::new(StatusCode::INTERNAL_SERVER_ERROR, "internal_error", message)
+    }
+
+    pub(super) fn bad_request(message: impl Into<String>) -> Self {
+        Self::new(StatusCode::BAD_REQUEST, "bad_request", message)
+    }
+}
+
+#[derive(Serialize)]
+pub(super) struct ErrorBody {
+    pub(super) code: &'static str,
+    pub(super) message: String,
+}
+
+#[derive(Serialize)]
+pub(super) struct ErrorResponse {
+    pub(super) ok: bool,
+    pub(super) error: ErrorBody,
+    pub(super) request_id: u64,
+}
+
+#[derive(Serialize)]
+pub(super) struct SuccessResponse<T> {
+    pub(super) ok: bool,
+    pub(super) data: T,
+    pub(super) revision: String,
+}
+
+#[derive(Serialize)]
+pub(super) struct HealthData {
+    pub(super) status: &'static str,
+    pub(super) read_only: bool,
+}
+
+#[derive(Serialize)]
+pub(super) struct SummaryData {
+    pub(super) uptime_seconds: f64,
+    pub(super) connections_total: u64,
+    pub(super) connections_bad_total: u64,
+    pub(super) handshake_timeouts_total: u64,
+    pub(super) configured_users: usize,
+}
+
+#[derive(Serialize, Clone)]
+pub(super) struct ZeroCodeCount {
+    pub(super) code: i32,
+    pub(super) total: u64,
+}
+
+#[derive(Serialize, Clone)]
+pub(super) struct ZeroCoreData {
+    pub(super) uptime_seconds: f64,
+    pub(super) connections_total: u64,
+    pub(super) connections_bad_total: u64,
+    pub(super) handshake_timeouts_total: u64,
+    pub(super) configured_users: usize,
+    pub(super) telemetry_core_enabled: bool,
+    pub(super) telemetry_user_enabled: bool,
+    pub(super) telemetry_me_level: String,
+}
+
+#[derive(Serialize, Clone)]
+pub(super) struct ZeroUpstreamData {
+    pub(super) connect_attempt_total: u64,
+    pub(super) connect_success_total: u64,
+    pub(super) connect_fail_total: u64,
+    pub(super) connect_failfast_hard_error_total: u64,
+    pub(super) connect_attempts_bucket_1: u64,
+    pub(super) connect_attempts_bucket_2: u64,
+    pub(super) connect_attempts_bucket_3_4: u64,
+    pub(super) connect_attempts_bucket_gt_4: u64,
+    pub(super) connect_duration_success_bucket_le_100ms: u64,
+    pub(super) connect_duration_success_bucket_101_500ms: u64,
+    pub(super) connect_duration_success_bucket_501_1000ms: u64,
+    pub(super) connect_duration_success_bucket_gt_1000ms: u64,
+    pub(super) connect_duration_fail_bucket_le_100ms: u64,
+    pub(super) connect_duration_fail_bucket_101_500ms: u64,
+    pub(super) connect_duration_fail_bucket_501_1000ms: u64,
+    pub(super) connect_duration_fail_bucket_gt_1000ms: u64,
+}
+
+#[derive(Serialize, Clone)]
+pub(super) struct UpstreamDcStatus {
+    pub(super) dc: i16,
+    pub(super) latency_ema_ms: Option<f64>,
+    pub(super) ip_preference: &'static str,
+}
+
+#[derive(Serialize, Clone)]
+pub(super) struct UpstreamStatus {
+    pub(super) upstream_id: usize,
+    pub(super) route_kind: &'static str,
+    pub(super) address: String,
+    pub(super) weight: u16,
+    pub(super) scopes: String,
+    pub(super) healthy: bool,
+    pub(super) fails: u32,
+    pub(super) last_check_age_secs: u64,
+    pub(super) effective_latency_ms: Option<f64>,
+    pub(super) dc: Vec<UpstreamDcStatus>,
+}
+
+#[derive(Serialize, Clone)]
+pub(super) struct UpstreamSummaryData {
+    pub(super) configured_total: usize,
+    pub(super) healthy_total: usize,
+    pub(super) unhealthy_total: usize,
+    pub(super) direct_total: usize,
+    pub(super) socks4_total: usize,
+    pub(super) socks5_total: usize,
+}
+
+#[derive(Serialize, Clone)]
+pub(super) struct UpstreamsData {
+    pub(super) enabled: bool,
+    #[serde(skip_serializing_if = "Option::is_none")]
+    pub(super) reason: Option<&'static str>,
+    pub(super) generated_at_epoch_secs: u64,
+    pub(super) zero: ZeroUpstreamData,
+    #[serde(skip_serializing_if = "Option::is_none")]
+    pub(super) summary: Option<UpstreamSummaryData>,
+    #[serde(skip_serializing_if = "Option::is_none")]
+    pub(super) upstreams: Option<Vec<UpstreamStatus>>,
+}
+
+#[derive(Serialize, Clone)]
+pub(super) struct ZeroMiddleProxyData {
+    pub(super) keepalive_sent_total: u64,
+    pub(super) keepalive_failed_total: u64,
+    pub(super) keepalive_pong_total: u64,
+    pub(super) keepalive_timeout_total: u64,
+    pub(super) rpc_proxy_req_signal_sent_total: u64,
+    pub(super) rpc_proxy_req_signal_failed_total: u64,
+    pub(super) rpc_proxy_req_signal_skipped_no_meta_total: u64,
+    pub(super) rpc_proxy_req_signal_response_total: u64,
+    pub(super) rpc_proxy_req_signal_close_sent_total: u64,
+    pub(super) reconnect_attempt_total: u64,
+    pub(super) reconnect_success_total: u64,
+    pub(super) handshake_reject_total: u64,
+    pub(super) handshake_error_codes: Vec<ZeroCodeCount>,
+    pub(super) reader_eof_total: u64,
+    pub(super) idle_close_by_peer_total: u64,
+    pub(super) route_drop_no_conn_total: u64,
+    pub(super) route_drop_channel_closed_total: u64,
+    pub(super) route_drop_queue_full_total: u64,
+    pub(super) route_drop_queue_full_base_total: u64,
+    pub(super) route_drop_queue_full_high_total: u64,
+    pub(super) socks_kdf_strict_reject_total: u64,
+    pub(super) socks_kdf_compat_fallback_total: u64,
+    pub(super) endpoint_quarantine_total: u64,
+    pub(super) kdf_drift_total: u64,
+    pub(super) kdf_port_only_drift_total: u64,
+    pub(super) hardswap_pending_reuse_total: u64,
+    pub(super) hardswap_pending_ttl_expired_total: u64,
+    pub(super) single_endpoint_outage_enter_total: u64,
+    pub(super) single_endpoint_outage_exit_total: u64,
+    pub(super) single_endpoint_outage_reconnect_attempt_total: u64,
+    pub(super) single_endpoint_outage_reconnect_success_total: u64,
+    pub(super) single_endpoint_quarantine_bypass_total: u64,
+    pub(super) single_endpoint_shadow_rotate_total: u64,
+    pub(super) single_endpoint_shadow_rotate_skipped_quarantine_total: u64,
+    pub(super) floor_mode_switch_total: u64,
+    pub(super) floor_mode_switch_static_to_adaptive_total: u64,
+    pub(super) floor_mode_switch_adaptive_to_static_total: u64,
+}
+
+#[derive(Serialize, Clone)]
+pub(super) struct ZeroPoolData {
+    pub(super) pool_swap_total: u64,
+    pub(super) pool_drain_active: u64,
+    pub(super) pool_force_close_total: u64,
+    pub(super) pool_stale_pick_total: u64,
+    pub(super) writer_removed_total: u64,
+    pub(super) writer_removed_unexpected_total: u64,
+    pub(super) refill_triggered_total: u64,
+    pub(super) refill_skipped_inflight_total: u64,
+    pub(super) refill_failed_total: u64,
+    pub(super) writer_restored_same_endpoint_total: u64,
+    pub(super) writer_restored_fallback_total: u64,
+}
+
+#[derive(Serialize, Clone)]
+pub(super) struct ZeroDesyncData {
+    pub(super) secure_padding_invalid_total: u64,
+    pub(super) desync_total: u64,
+    pub(super) desync_full_logged_total: u64,
+    pub(super) desync_suppressed_total: u64,
+    pub(super) desync_frames_bucket_0: u64,
+    pub(super) desync_frames_bucket_1_2: u64,
+    pub(super) desync_frames_bucket_3_10: u64,
+    pub(super) desync_frames_bucket_gt_10: u64,
+}
+
+#[derive(Serialize, Clone)]
+pub(super) struct ZeroAllData {
+    pub(super) generated_at_epoch_secs: u64,
+    pub(super) core: ZeroCoreData,
+    pub(super) upstream: ZeroUpstreamData,
+    pub(super) middle_proxy: ZeroMiddleProxyData,
+    pub(super) pool: ZeroPoolData,
+    pub(super) desync: ZeroDesyncData,
+}
+
+#[derive(Serialize, Clone)]
+pub(super) struct MeWritersSummary {
+    pub(super) configured_dc_groups: usize,
+    pub(super) configured_endpoints: usize,
+    pub(super) available_endpoints: usize,
+    pub(super) available_pct: f64,
+    pub(super) required_writers: usize,
+    pub(super) alive_writers: usize,
+    pub(super) coverage_pct: f64,
+}
+
+#[derive(Serialize, Clone)]
+pub(super) struct MeWriterStatus {
+    pub(super) writer_id: u64,
+    pub(super) dc: Option<i16>,
+    pub(super) endpoint: String,
+    pub(super) generation: u64,
+    pub(super) state: &'static str,
+    pub(super) draining: bool,
+    pub(super) degraded: bool,
+    pub(super) bound_clients: usize,
+    pub(super) idle_for_secs: Option<u64>,
+    pub(super) rtt_ema_ms: Option<f64>,
+}
+
+#[derive(Serialize, Clone)]
+pub(super) struct MeWritersData {
+    pub(super) middle_proxy_enabled: bool,
+    #[serde(skip_serializing_if = "Option::is_none")]
+    pub(super) reason: Option<&'static str>,
+    pub(super) generated_at_epoch_secs: u64,
+    pub(super) summary: MeWritersSummary,
+    pub(super) writers: Vec<MeWriterStatus>,
+}
+
+#[derive(Serialize, Clone)]
+pub(super) struct DcStatus {
+    pub(super) dc: i16,
+    pub(super) endpoints: Vec<String>,
+    pub(super) available_endpoints: usize,
+    pub(super) available_pct: f64,
+    pub(super) required_writers: usize,
+    pub(super) alive_writers: usize,
+    pub(super) coverage_pct: f64,
+    pub(super) rtt_ms: Option<f64>,
+    pub(super) load: usize,
+}
+
+#[derive(Serialize, Clone)]
+pub(super) struct DcStatusData {
+    pub(super) middle_proxy_enabled: bool,
+    #[serde(skip_serializing_if = "Option::is_none")]
+    pub(super) reason: Option<&'static str>,
+    pub(super) generated_at_epoch_secs: u64,
+    pub(super) dcs: Vec<DcStatus>,
+}
+
+#[derive(Serialize, Clone)]
+pub(super) struct MinimalQuarantineData {
+    pub(super) endpoint: String,
+    pub(super) remaining_ms: u64,
+}
+
+#[derive(Serialize, Clone)]
+pub(super) struct MinimalDcPathData {
+    pub(super) dc: i16,
+    pub(super) ip_preference: Option<&'static str>,
+    pub(super) selected_addr_v4: Option<String>,
+    pub(super) selected_addr_v6: Option<String>,
+}
+
+#[derive(Serialize, Clone)]
+pub(super) struct MinimalMeRuntimeData {
+    pub(super) active_generation: u64,
+    pub(super) warm_generation: u64,
+    pub(super) pending_hardswap_generation: u64,
+    pub(super) pending_hardswap_age_secs: Option<u64>,
+    pub(super) hardswap_enabled: bool,
+    pub(super) floor_mode: &'static str,
+    pub(super) adaptive_floor_idle_secs: u64,
+    pub(super) adaptive_floor_min_writers_single_endpoint: u8,
+    pub(super) adaptive_floor_recover_grace_secs: u64,
+    pub(super) me_keepalive_enabled: bool,
+    pub(super) me_keepalive_interval_secs: u64,
+    pub(super) me_keepalive_jitter_secs: u64,
+    pub(super) me_keepalive_payload_random: bool,
+    pub(super) rpc_proxy_req_every_secs: u64,
+    pub(super) me_reconnect_max_concurrent_per_dc: u32,
+    pub(super) me_reconnect_backoff_base_ms: u64,
+    pub(super) me_reconnect_backoff_cap_ms: u64,
+    pub(super) me_reconnect_fast_retry_count: u32,
+    pub(super) me_pool_drain_ttl_secs: u64,
+    pub(super) me_pool_force_close_secs: u64,
+    pub(super) me_pool_min_fresh_ratio: f32,
+    pub(super) me_bind_stale_mode: &'static str,
+    pub(super) me_bind_stale_ttl_secs: u64,
+    pub(super) me_single_endpoint_shadow_writers: u8,
+    pub(super) me_single_endpoint_outage_mode_enabled: bool,
+    pub(super) me_single_endpoint_outage_disable_quarantine: bool,
+    pub(super) me_single_endpoint_outage_backoff_min_ms: u64,
+    pub(super) me_single_endpoint_outage_backoff_max_ms: u64,
+    pub(super) me_single_endpoint_shadow_rotate_every_secs: u64,
+    pub(super) me_deterministic_writer_sort: bool,
+    pub(super) me_socks_kdf_policy: &'static str,
+    pub(super) quarantined_endpoints_total: usize,
+    pub(super) quarantined_endpoints: Vec<MinimalQuarantineData>,
+}
+
+#[derive(Serialize, Clone)]
+pub(super) struct MinimalAllPayload {
+    pub(super) me_writers: MeWritersData,
+    pub(super) dcs: DcStatusData,
+    #[serde(skip_serializing_if = "Option::is_none")]
+    pub(super) me_runtime: Option<MinimalMeRuntimeData>,
+    pub(super) network_path: Vec<MinimalDcPathData>,
+}
+
+#[derive(Serialize, Clone)]
+pub(super) struct MinimalAllData {
+    pub(super) enabled: bool,
+    #[serde(skip_serializing_if = "Option::is_none")]
+    pub(super) reason: Option<&'static str>,
+    pub(super) generated_at_epoch_secs: u64,
+    #[serde(skip_serializing_if = "Option::is_none")]
+    pub(super) data: Option<MinimalAllPayload>,
+}
+
+#[derive(Serialize)]
+pub(super) struct UserLinks {
+    pub(super) classic: Vec<String>,
+    pub(super) secure: Vec<String>,
+    pub(super) tls: Vec<String>,
+}
+
+#[derive(Serialize)]
+pub(super) struct UserInfo {
+    pub(super) username: String,
+    pub(super) user_ad_tag: Option<String>,
+    pub(super) max_tcp_conns: Option<usize>,
+    pub(super) expiration_rfc3339: Option<String>,
+    pub(super) data_quota_bytes: Option<u64>,
+    pub(super) max_unique_ips: Option<usize>,
+    pub(super) current_connections: u64,
+    pub(super) active_unique_ips: usize,
+    pub(super) active_unique_ips_list: Vec<IpAddr>,
+    pub(super) recent_unique_ips: usize,
+    pub(super) recent_unique_ips_list: Vec<IpAddr>,
+    pub(super) total_octets: u64,
+    pub(super) links: UserLinks,
+}
+
+#[derive(Serialize)]
+pub(super) struct CreateUserResponse {
+    pub(super) user: UserInfo,
+    pub(super) secret: String,
+}
+
+#[derive(Deserialize)]
+pub(super) struct CreateUserRequest {
+    pub(super) username: String,
+    pub(super) secret: Option<String>,
+    pub(super) user_ad_tag: Option<String>,
+    pub(super) max_tcp_conns: Option<usize>,
+    pub(super) expiration_rfc3339: Option<String>,
+    pub(super) data_quota_bytes: Option<u64>,
+    pub(super) max_unique_ips: Option<usize>,
+}
+
+#[derive(Deserialize)]
+pub(super) struct PatchUserRequest {
+    pub(super) secret: Option<String>,
+    pub(super) user_ad_tag: Option<String>,
+    pub(super) max_tcp_conns: Option<usize>,
+    pub(super) expiration_rfc3339: Option<String>,
+    pub(super) data_quota_bytes: Option<u64>,
+    pub(super) max_unique_ips: Option<usize>,
+}
+
+#[derive(Default, Deserialize)]
+pub(super) struct RotateSecretRequest {
+    pub(super) secret: Option<String>,
+}
+
+pub(super) fn parse_optional_expiration(
+    value: Option<&str>,
+) -> Result<Option<DateTime<Utc>>, ApiFailure> {
+    let Some(raw) = value else {
+        return Ok(None);
+    };
+    let parsed = DateTime::parse_from_rfc3339(raw)
+        .map_err(|_| ApiFailure::bad_request("expiration_rfc3339 must be valid RFC3339"))?;
+    Ok(Some(parsed.with_timezone(&Utc)))
+}
+
+pub(super) fn is_valid_user_secret(secret: &str) -> bool {
+    secret.len() == 32 && secret.chars().all(|c| c.is_ascii_hexdigit())
+}
+
+pub(super) fn is_valid_ad_tag(tag: &str) -> bool {
+    tag.len() == 32 && tag.chars().all(|c| c.is_ascii_hexdigit())
+}
+
+pub(super) fn is_valid_username(user: &str) -> bool {
+    !user.is_empty()
+        && user.len() <= MAX_USERNAME_LEN
+        && user
+            .chars()
+            .all(|ch| ch.is_ascii_alphanumeric() || matches!(ch, '_' | '-' | '.'))
+}
+
+pub(super) fn random_user_secret() -> String {
+    let mut bytes = [0u8; 16];
+    rand::rng().fill(&mut bytes);
+    hex::encode(bytes)
+}

src/api/runtime_stats.rs

@@ -0,0 +1,484 @@
+use std::time::{Duration, Instant, SystemTime, UNIX_EPOCH};
+
+use crate::config::ApiConfig;
+use crate::stats::Stats;
+use crate::transport::upstream::IpPreference;
+use crate::transport::UpstreamRouteKind;
+
+use super::ApiShared;
+use super::model::{
+    DcStatus, DcStatusData, MeWriterStatus, MeWritersData, MeWritersSummary, MinimalAllData,
+    MinimalAllPayload, MinimalDcPathData, MinimalMeRuntimeData, MinimalQuarantineData,
+    UpstreamDcStatus, UpstreamStatus, UpstreamSummaryData, UpstreamsData, ZeroAllData,
+    ZeroCodeCount, ZeroCoreData, ZeroDesyncData, ZeroMiddleProxyData, ZeroPoolData,
+    ZeroUpstreamData,
+};
+
+const FEATURE_DISABLED_REASON: &str = "feature_disabled";
+const SOURCE_UNAVAILABLE_REASON: &str = "source_unavailable";
+
+#[derive(Clone)]
+pub(crate) struct MinimalCacheEntry {
+    pub(super) expires_at: Instant,
+    pub(super) payload: MinimalAllPayload,
+    pub(super) generated_at_epoch_secs: u64,
+}
+
+pub(super) fn build_zero_all_data(stats: &Stats, configured_users: usize) -> ZeroAllData {
+    let telemetry = stats.telemetry_policy();
+    let handshake_error_codes = stats
+        .get_me_handshake_error_code_counts()
+        .into_iter()
+        .map(|(code, total)| ZeroCodeCount { code, total })
+        .collect();
+
+    ZeroAllData {
+        generated_at_epoch_secs: now_epoch_secs(),
+        core: ZeroCoreData {
+            uptime_seconds: stats.uptime_secs(),
+            connections_total: stats.get_connects_all(),
+            connections_bad_total: stats.get_connects_bad(),
+            handshake_timeouts_total: stats.get_handshake_timeouts(),
+            configured_users,
+            telemetry_core_enabled: telemetry.core_enabled,
+            telemetry_user_enabled: telemetry.user_enabled,
+            telemetry_me_level: telemetry.me_level.to_string(),
+        },
+        upstream: build_zero_upstream_data(stats),
+        middle_proxy: ZeroMiddleProxyData {
+            keepalive_sent_total: stats.get_me_keepalive_sent(),
+            keepalive_failed_total: stats.get_me_keepalive_failed(),
+            keepalive_pong_total: stats.get_me_keepalive_pong(),
+            keepalive_timeout_total: stats.get_me_keepalive_timeout(),
+            rpc_proxy_req_signal_sent_total: stats.get_me_rpc_proxy_req_signal_sent_total(),
+            rpc_proxy_req_signal_failed_total: stats.get_me_rpc_proxy_req_signal_failed_total(),
+            rpc_proxy_req_signal_skipped_no_meta_total: stats
+                .get_me_rpc_proxy_req_signal_skipped_no_meta_total(),
+            rpc_proxy_req_signal_response_total: stats.get_me_rpc_proxy_req_signal_response_total(),
+            rpc_proxy_req_signal_close_sent_total: stats
+                .get_me_rpc_proxy_req_signal_close_sent_total(),
+            reconnect_attempt_total: stats.get_me_reconnect_attempts(),
+            reconnect_success_total: stats.get_me_reconnect_success(),
+            handshake_reject_total: stats.get_me_handshake_reject_total(),
+            handshake_error_codes,
+            reader_eof_total: stats.get_me_reader_eof_total(),
+            idle_close_by_peer_total: stats.get_me_idle_close_by_peer_total(),
+            route_drop_no_conn_total: stats.get_me_route_drop_no_conn(),
+            route_drop_channel_closed_total: stats.get_me_route_drop_channel_closed(),
+            route_drop_queue_full_total: stats.get_me_route_drop_queue_full(),
+            route_drop_queue_full_base_total: stats.get_me_route_drop_queue_full_base(),
+            route_drop_queue_full_high_total: stats.get_me_route_drop_queue_full_high(),
+            socks_kdf_strict_reject_total: stats.get_me_socks_kdf_strict_reject(),
+            socks_kdf_compat_fallback_total: stats.get_me_socks_kdf_compat_fallback(),
+            endpoint_quarantine_total: stats.get_me_endpoint_quarantine_total(),
+            kdf_drift_total: stats.get_me_kdf_drift_total(),
+            kdf_port_only_drift_total: stats.get_me_kdf_port_only_drift_total(),
+            hardswap_pending_reuse_total: stats.get_me_hardswap_pending_reuse_total(),
+            hardswap_pending_ttl_expired_total: stats.get_me_hardswap_pending_ttl_expired_total(),
+            single_endpoint_outage_enter_total: stats.get_me_single_endpoint_outage_enter_total(),
+            single_endpoint_outage_exit_total: stats.get_me_single_endpoint_outage_exit_total(),
+            single_endpoint_outage_reconnect_attempt_total: stats
+                .get_me_single_endpoint_outage_reconnect_attempt_total(),
+            single_endpoint_outage_reconnect_success_total: stats
+                .get_me_single_endpoint_outage_reconnect_success_total(),
+            single_endpoint_quarantine_bypass_total: stats
+                .get_me_single_endpoint_quarantine_bypass_total(),
+            single_endpoint_shadow_rotate_total: stats.get_me_single_endpoint_shadow_rotate_total(),
+            single_endpoint_shadow_rotate_skipped_quarantine_total: stats
+                .get_me_single_endpoint_shadow_rotate_skipped_quarantine_total(),
+            floor_mode_switch_total: stats.get_me_floor_mode_switch_total(),
+            floor_mode_switch_static_to_adaptive_total: stats
+                .get_me_floor_mode_switch_static_to_adaptive_total(),
+            floor_mode_switch_adaptive_to_static_total: stats
+                .get_me_floor_mode_switch_adaptive_to_static_total(),
+        },
+        pool: ZeroPoolData {
+            pool_swap_total: stats.get_pool_swap_total(),
+            pool_drain_active: stats.get_pool_drain_active(),
+            pool_force_close_total: stats.get_pool_force_close_total(),
+            pool_stale_pick_total: stats.get_pool_stale_pick_total(),
+            writer_removed_total: stats.get_me_writer_removed_total(),
+            writer_removed_unexpected_total: stats.get_me_writer_removed_unexpected_total(),
+            refill_triggered_total: stats.get_me_refill_triggered_total(),
+            refill_skipped_inflight_total: stats.get_me_refill_skipped_inflight_total(),
+            refill_failed_total: stats.get_me_refill_failed_total(),
+            writer_restored_same_endpoint_total: stats.get_me_writer_restored_same_endpoint_total(),
+            writer_restored_fallback_total: stats.get_me_writer_restored_fallback_total(),
+        },
+        desync: ZeroDesyncData {
+            secure_padding_invalid_total: stats.get_secure_padding_invalid(),
+            desync_total: stats.get_desync_total(),
+            desync_full_logged_total: stats.get_desync_full_logged(),
+            desync_suppressed_total: stats.get_desync_suppressed(),
+            desync_frames_bucket_0: stats.get_desync_frames_bucket_0(),
+            desync_frames_bucket_1_2: stats.get_desync_frames_bucket_1_2(),
+            desync_frames_bucket_3_10: stats.get_desync_frames_bucket_3_10(),
+            desync_frames_bucket_gt_10: stats.get_desync_frames_bucket_gt_10(),
+        },
+    }
+}
+
+fn build_zero_upstream_data(stats: &Stats) -> ZeroUpstreamData {
+    ZeroUpstreamData {
+        connect_attempt_total: stats.get_upstream_connect_attempt_total(),
+        connect_success_total: stats.get_upstream_connect_success_total(),
+        connect_fail_total: stats.get_upstream_connect_fail_total(),
+        connect_failfast_hard_error_total: stats.get_upstream_connect_failfast_hard_error_total(),
+        connect_attempts_bucket_1: stats.get_upstream_connect_attempts_bucket_1(),
+        connect_attempts_bucket_2: stats.get_upstream_connect_attempts_bucket_2(),
+        connect_attempts_bucket_3_4: stats.get_upstream_connect_attempts_bucket_3_4(),
+        connect_attempts_bucket_gt_4: stats.get_upstream_connect_attempts_bucket_gt_4(),
+        connect_duration_success_bucket_le_100ms: stats
+            .get_upstream_connect_duration_success_bucket_le_100ms(),
+        connect_duration_success_bucket_101_500ms: stats
+            .get_upstream_connect_duration_success_bucket_101_500ms(),
+        connect_duration_success_bucket_501_1000ms: stats
+            .get_upstream_connect_duration_success_bucket_501_1000ms(),
+        connect_duration_success_bucket_gt_1000ms: stats
+            .get_upstream_connect_duration_success_bucket_gt_1000ms(),
+        connect_duration_fail_bucket_le_100ms: stats.get_upstream_connect_duration_fail_bucket_le_100ms(),
+        connect_duration_fail_bucket_101_500ms: stats
+            .get_upstream_connect_duration_fail_bucket_101_500ms(),
+        connect_duration_fail_bucket_501_1000ms: stats
+            .get_upstream_connect_duration_fail_bucket_501_1000ms(),
+        connect_duration_fail_bucket_gt_1000ms: stats
+            .get_upstream_connect_duration_fail_bucket_gt_1000ms(),
+    }
+}
+
+pub(super) fn build_upstreams_data(shared: &ApiShared, api_cfg: &ApiConfig) -> UpstreamsData {
+    let generated_at_epoch_secs = now_epoch_secs();
+    let zero = build_zero_upstream_data(&shared.stats);
+    if !api_cfg.minimal_runtime_enabled {
+        return UpstreamsData {
+            enabled: false,
+            reason: Some(FEATURE_DISABLED_REASON),
+            generated_at_epoch_secs,
+            zero,
+            summary: None,
+            upstreams: None,
+        };
+    }
+
+    let Some(snapshot) = shared.upstream_manager.try_api_snapshot() else {
+        return UpstreamsData {
+            enabled: true,
+            reason: Some(SOURCE_UNAVAILABLE_REASON),
+            generated_at_epoch_secs,
+            zero,
+            summary: None,
+            upstreams: None,
+        };
+    };
+
+    let summary = UpstreamSummaryData {
+        configured_total: snapshot.summary.configured_total,
+        healthy_total: snapshot.summary.healthy_total,
+        unhealthy_total: snapshot.summary.unhealthy_total,
+        direct_total: snapshot.summary.direct_total,
+        socks4_total: snapshot.summary.socks4_total,
+        socks5_total: snapshot.summary.socks5_total,
+    };
+    let upstreams = snapshot
+        .upstreams
+        .into_iter()
+        .map(|upstream| UpstreamStatus {
+            upstream_id: upstream.upstream_id,
+            route_kind: map_route_kind(upstream.route_kind),
+            address: upstream.address,
+            weight: upstream.weight,
+            scopes: upstream.scopes,
+            healthy: upstream.healthy,
+            fails: upstream.fails,
+            last_check_age_secs: upstream.last_check_age_secs,
+            effective_latency_ms: upstream.effective_latency_ms,
+            dc: upstream
+                .dc
+                .into_iter()
+                .map(|dc| UpstreamDcStatus {
+                    dc: dc.dc,
+                    latency_ema_ms: dc.latency_ema_ms,
+                    ip_preference: map_ip_preference(dc.ip_preference),
+                })
+                .collect(),
+        })
+        .collect();
+
+    UpstreamsData {
+        enabled: true,
+        reason: None,
+        generated_at_epoch_secs,
+        zero,
+        summary: Some(summary),
+        upstreams: Some(upstreams),
+    }
+}
+
+pub(super) async fn build_minimal_all_data(
+    shared: &ApiShared,
+    api_cfg: &ApiConfig,
+) -> MinimalAllData {
+    let now = now_epoch_secs();
+    if !api_cfg.minimal_runtime_enabled {
+        return MinimalAllData {
+            enabled: false,
+            reason: Some(FEATURE_DISABLED_REASON),
+            generated_at_epoch_secs: now,
+            data: None,
+        };
+    }
+
+    let Some((generated_at_epoch_secs, payload)) =
+        get_minimal_payload_cached(shared, api_cfg.minimal_runtime_cache_ttl_ms).await
+    else {
+        return MinimalAllData {
+            enabled: true,
+            reason: Some(SOURCE_UNAVAILABLE_REASON),
+            generated_at_epoch_secs: now,
+            data: Some(MinimalAllPayload {
+                me_writers: disabled_me_writers(now, SOURCE_UNAVAILABLE_REASON),
+                dcs: disabled_dcs(now, SOURCE_UNAVAILABLE_REASON),
+                me_runtime: None,
+                network_path: Vec::new(),
+            }),
+        };
+    };
+
+    MinimalAllData {
+        enabled: true,
+        reason: None,
+        generated_at_epoch_secs,
+        data: Some(payload),
+    }
+}
+
+pub(super) async fn build_me_writers_data(
+    shared: &ApiShared,
+    api_cfg: &ApiConfig,
+) -> MeWritersData {
+    let now = now_epoch_secs();
+    if !api_cfg.minimal_runtime_enabled {
+        return disabled_me_writers(now, FEATURE_DISABLED_REASON);
+    }
+
+    let Some((_, payload)) =
+        get_minimal_payload_cached(shared, api_cfg.minimal_runtime_cache_ttl_ms).await
+    else {
+        return disabled_me_writers(now, SOURCE_UNAVAILABLE_REASON);
+    };
+    payload.me_writers
+}
+
+pub(super) async fn build_dcs_data(shared: &ApiShared, api_cfg: &ApiConfig) -> DcStatusData {
+    let now = now_epoch_secs();
+    if !api_cfg.minimal_runtime_enabled {
+        return disabled_dcs(now, FEATURE_DISABLED_REASON);
+    }
+
+    let Some((_, payload)) =
+        get_minimal_payload_cached(shared, api_cfg.minimal_runtime_cache_ttl_ms).await
+    else {
+        return disabled_dcs(now, SOURCE_UNAVAILABLE_REASON);
+    };
+    payload.dcs
+}
+
+async fn get_minimal_payload_cached(
+    shared: &ApiShared,
+    cache_ttl_ms: u64,
+) -> Option<(u64, MinimalAllPayload)> {
+    if cache_ttl_ms > 0 {
+        let now = Instant::now();
+        let cached = shared.minimal_cache.lock().await.clone();
+        if let Some(entry) = cached
+            && now < entry.expires_at
+        {
+            return Some((entry.generated_at_epoch_secs, entry.payload));
+        }
+    }
+
+    let pool = shared.me_pool.as_ref()?;
+    let status = pool.api_status_snapshot().await;
+    let runtime = pool.api_runtime_snapshot().await;
+    let generated_at_epoch_secs = status.generated_at_epoch_secs;
+
+    let me_writers = MeWritersData {
+        middle_proxy_enabled: true,
+        reason: None,
+        generated_at_epoch_secs,
+        summary: MeWritersSummary {
+            configured_dc_groups: status.configured_dc_groups,
+            configured_endpoints: status.configured_endpoints,
+            available_endpoints: status.available_endpoints,
+            available_pct: status.available_pct,
+            required_writers: status.required_writers,
+            alive_writers: status.alive_writers,
+            coverage_pct: status.coverage_pct,
+        },
+        writers: status
+            .writers
+            .into_iter()
+            .map(|entry| MeWriterStatus {
+                writer_id: entry.writer_id,
+                dc: entry.dc,
+                endpoint: entry.endpoint.to_string(),
+                generation: entry.generation,
+                state: entry.state,
+                draining: entry.draining,
+                degraded: entry.degraded,
+                bound_clients: entry.bound_clients,
+                idle_for_secs: entry.idle_for_secs,
+                rtt_ema_ms: entry.rtt_ema_ms,
+            })
+            .collect(),
+    };
+    let dcs = DcStatusData {
+        middle_proxy_enabled: true,
+        reason: None,
+        generated_at_epoch_secs,
+        dcs: status
+            .dcs
+            .into_iter()
+            .map(|entry| DcStatus {
+                dc: entry.dc,
+                endpoints: entry
+                    .endpoints
+                    .into_iter()
+                    .map(|value| value.to_string())
+                    .collect(),
+                available_endpoints: entry.available_endpoints,
+                available_pct: entry.available_pct,
+                required_writers: entry.required_writers,
+                alive_writers: entry.alive_writers,
+                coverage_pct: entry.coverage_pct,
+                rtt_ms: entry.rtt_ms,
+                load: entry.load,
+            })
+            .collect(),
+    };
+    let me_runtime = MinimalMeRuntimeData {
+        active_generation: runtime.active_generation,
+        warm_generation: runtime.warm_generation,
+        pending_hardswap_generation: runtime.pending_hardswap_generation,
+        pending_hardswap_age_secs: runtime.pending_hardswap_age_secs,
+        hardswap_enabled: runtime.hardswap_enabled,
+        floor_mode: runtime.floor_mode,
+        adaptive_floor_idle_secs: runtime.adaptive_floor_idle_secs,
+        adaptive_floor_min_writers_single_endpoint: runtime
+            .adaptive_floor_min_writers_single_endpoint,
+        adaptive_floor_recover_grace_secs: runtime.adaptive_floor_recover_grace_secs,
+        me_keepalive_enabled: runtime.me_keepalive_enabled,
+        me_keepalive_interval_secs: runtime.me_keepalive_interval_secs,
+        me_keepalive_jitter_secs: runtime.me_keepalive_jitter_secs,
+        me_keepalive_payload_random: runtime.me_keepalive_payload_random,
+        rpc_proxy_req_every_secs: runtime.rpc_proxy_req_every_secs,
+        me_reconnect_max_concurrent_per_dc: runtime.me_reconnect_max_concurrent_per_dc,
+        me_reconnect_backoff_base_ms: runtime.me_reconnect_backoff_base_ms,
+        me_reconnect_backoff_cap_ms: runtime.me_reconnect_backoff_cap_ms,
+        me_reconnect_fast_retry_count: runtime.me_reconnect_fast_retry_count,
+        me_pool_drain_ttl_secs: runtime.me_pool_drain_ttl_secs,
+        me_pool_force_close_secs: runtime.me_pool_force_close_secs,
+        me_pool_min_fresh_ratio: runtime.me_pool_min_fresh_ratio,
+        me_bind_stale_mode: runtime.me_bind_stale_mode,
+        me_bind_stale_ttl_secs: runtime.me_bind_stale_ttl_secs,
+        me_single_endpoint_shadow_writers: runtime.me_single_endpoint_shadow_writers,
+        me_single_endpoint_outage_mode_enabled: runtime.me_single_endpoint_outage_mode_enabled,
+        me_single_endpoint_outage_disable_quarantine: runtime
+            .me_single_endpoint_outage_disable_quarantine,
+        me_single_endpoint_outage_backoff_min_ms: runtime.me_single_endpoint_outage_backoff_min_ms,
+        me_single_endpoint_outage_backoff_max_ms: runtime.me_single_endpoint_outage_backoff_max_ms,
+        me_single_endpoint_shadow_rotate_every_secs: runtime
+            .me_single_endpoint_shadow_rotate_every_secs,
+        me_deterministic_writer_sort: runtime.me_deterministic_writer_sort,
+        me_socks_kdf_policy: runtime.me_socks_kdf_policy,
+        quarantined_endpoints_total: runtime.quarantined_endpoints.len(),
+        quarantined_endpoints: runtime
+            .quarantined_endpoints
+            .into_iter()
+            .map(|entry| MinimalQuarantineData {
+                endpoint: entry.endpoint.to_string(),
+                remaining_ms: entry.remaining_ms,
+            })
+            .collect(),
+    };
+    let network_path = runtime
+        .network_path
+        .into_iter()
+        .map(|entry| MinimalDcPathData {
+            dc: entry.dc,
+            ip_preference: entry.ip_preference,
+            selected_addr_v4: entry.selected_addr_v4.map(|value| value.to_string()),
+            selected_addr_v6: entry.selected_addr_v6.map(|value| value.to_string()),
+        })
+        .collect();
+
+    let payload = MinimalAllPayload {
+        me_writers,
+        dcs,
+        me_runtime: Some(me_runtime),
+        network_path,
+    };
+
+    if cache_ttl_ms > 0 {
+        let entry = MinimalCacheEntry {
+            expires_at: Instant::now() + Duration::from_millis(cache_ttl_ms),
+            payload: payload.clone(),
+            generated_at_epoch_secs,
+        };
+        *shared.minimal_cache.lock().await = Some(entry);
+    }
+
+    Some((generated_at_epoch_secs, payload))
+}
+
+fn disabled_me_writers(now_epoch_secs: u64, reason: &'static str) -> MeWritersData {
+    MeWritersData {
+        middle_proxy_enabled: false,
+        reason: Some(reason),
+        generated_at_epoch_secs: now_epoch_secs,
+        summary: MeWritersSummary {
+            configured_dc_groups: 0,
+            configured_endpoints: 0,
+            available_endpoints: 0,
+            available_pct: 0.0,
+            required_writers: 0,
+            alive_writers: 0,
+            coverage_pct: 0.0,
+        },
+        writers: Vec::new(),
+    }
+}
+
+fn disabled_dcs(now_epoch_secs: u64, reason: &'static str) -> DcStatusData {
+    DcStatusData {
+        middle_proxy_enabled: false,
+        reason: Some(reason),
+        generated_at_epoch_secs: now_epoch_secs,
+        dcs: Vec::new(),
+    }
+}
+
+fn map_route_kind(value: UpstreamRouteKind) -> &'static str {
+    match value {
+        UpstreamRouteKind::Direct => "direct",
+        UpstreamRouteKind::Socks4 => "socks4",
+        UpstreamRouteKind::Socks5 => "socks5",
+    }
+}
+
+fn map_ip_preference(value: IpPreference) -> &'static str {
+    match value {
+        IpPreference::Unknown => "unknown",
+        IpPreference::PreferV6 => "prefer_v6",
+        IpPreference::PreferV4 => "prefer_v4",
+        IpPreference::BothWork => "both_work",
+        IpPreference::Unavailable => "unavailable",
+    }
+}
+
+fn now_epoch_secs() -> u64 {
+    SystemTime::now()
+        .duration_since(UNIX_EPOCH)
+        .unwrap_or_default()
+        .as_secs()
+}

src/api/users.rs

@@ -0,0 +1,499 @@
+use std::net::IpAddr;
+
+use hyper::StatusCode;
+
+use crate::config::ProxyConfig;
+use crate::ip_tracker::UserIpTracker;
+use crate::stats::Stats;
+
+use super::ApiShared;
+use super::config_store::{
+    ensure_expected_revision, load_config_from_disk, save_config_to_disk,
+};
+use super::model::{
+    ApiFailure, CreateUserRequest, CreateUserResponse, PatchUserRequest, RotateSecretRequest,
+    UserInfo, UserLinks, is_valid_ad_tag, is_valid_user_secret, is_valid_username,
+    parse_optional_expiration, random_user_secret,
+};
+
+pub(super) async fn create_user(
+    body: CreateUserRequest,
+    expected_revision: Option<String>,
+    shared: &ApiShared,
+) -> Result<(CreateUserResponse, String), ApiFailure> {
+    if !is_valid_username(&body.username) {
+        return Err(ApiFailure::bad_request(
+            "username must match [A-Za-z0-9_.-] and be 1..64 chars",
+        ));
+    }
+
+    let secret = match body.secret {
+        Some(secret) => {
+            if !is_valid_user_secret(&secret) {
+                return Err(ApiFailure::bad_request(
+                    "secret must be exactly 32 hex characters",
+                ));
+            }
+            secret
+        }
+        None => random_user_secret(),
+    };
+
+    if let Some(ad_tag) = body.user_ad_tag.as_ref() && !is_valid_ad_tag(ad_tag) {
+        return Err(ApiFailure::bad_request(
+            "user_ad_tag must be exactly 32 hex characters",
+        ));
+    }
+
+    let expiration = parse_optional_expiration(body.expiration_rfc3339.as_deref())?;
+    let _guard = shared.mutation_lock.lock().await;
+    let mut cfg = load_config_from_disk(&shared.config_path).await?;
+    ensure_expected_revision(&shared.config_path, expected_revision.as_deref()).await?;
+
+    if cfg.access.users.contains_key(&body.username) {
+        return Err(ApiFailure::new(
+            StatusCode::CONFLICT,
+            "user_exists",
+            "User already exists",
+        ));
+    }
+
+    cfg.access.users.insert(body.username.clone(), secret.clone());
+    if let Some(ad_tag) = body.user_ad_tag {
+        cfg.access.user_ad_tags.insert(body.username.clone(), ad_tag);
+    }
+    if let Some(limit) = body.max_tcp_conns {
+        cfg.access.user_max_tcp_conns.insert(body.username.clone(), limit);
+    }
+    if let Some(expiration) = expiration {
+        cfg.access
+            .user_expirations
+            .insert(body.username.clone(), expiration);
+    }
+    if let Some(quota) = body.data_quota_bytes {
+        cfg.access.user_data_quota.insert(body.username.clone(), quota);
+    }
+
+    let updated_limit = body.max_unique_ips;
+    if let Some(limit) = updated_limit {
+        cfg.access
+            .user_max_unique_ips
+            .insert(body.username.clone(), limit);
+    }
+
+    cfg.validate()
+        .map_err(|e| ApiFailure::bad_request(format!("config validation failed: {}", e)))?;
+
+    let revision = save_config_to_disk(&shared.config_path, &cfg).await?;
+    drop(_guard);
+
+    if let Some(limit) = updated_limit {
+        shared.ip_tracker.set_user_limit(&body.username, limit).await;
+    }
+
+    let users = users_from_config(
+        &cfg,
+        &shared.stats,
+        &shared.ip_tracker,
+        shared.startup_detected_ip_v4,
+        shared.startup_detected_ip_v6,
+    )
+    .await;
+    let user = users
+        .into_iter()
+        .find(|entry| entry.username == body.username)
+        .unwrap_or(UserInfo {
+            username: body.username.clone(),
+            user_ad_tag: None,
+            max_tcp_conns: None,
+            expiration_rfc3339: None,
+            data_quota_bytes: None,
+            max_unique_ips: updated_limit,
+            current_connections: 0,
+            active_unique_ips: 0,
+            active_unique_ips_list: Vec::new(),
+            recent_unique_ips: 0,
+            recent_unique_ips_list: Vec::new(),
+            total_octets: 0,
+            links: build_user_links(
+                &cfg,
+                &secret,
+                shared.startup_detected_ip_v4,
+                shared.startup_detected_ip_v6,
+            ),
+        });
+
+    Ok((CreateUserResponse { user, secret }, revision))
+}
+
+pub(super) async fn patch_user(
+    user: &str,
+    body: PatchUserRequest,
+    expected_revision: Option<String>,
+    shared: &ApiShared,
+) -> Result<(UserInfo, String), ApiFailure> {
+    if let Some(secret) = body.secret.as_ref() && !is_valid_user_secret(secret) {
+        return Err(ApiFailure::bad_request(
+            "secret must be exactly 32 hex characters",
+        ));
+    }
+    if let Some(ad_tag) = body.user_ad_tag.as_ref() && !is_valid_ad_tag(ad_tag) {
+        return Err(ApiFailure::bad_request(
+            "user_ad_tag must be exactly 32 hex characters",
+        ));
+    }
+    let expiration = parse_optional_expiration(body.expiration_rfc3339.as_deref())?;
+    let _guard = shared.mutation_lock.lock().await;
+    let mut cfg = load_config_from_disk(&shared.config_path).await?;
+    ensure_expected_revision(&shared.config_path, expected_revision.as_deref()).await?;
+
+    if !cfg.access.users.contains_key(user) {
+        return Err(ApiFailure::new(
+            StatusCode::NOT_FOUND,
+            "not_found",
+            "User not found",
+        ));
+    }
+
+    if let Some(secret) = body.secret {
+        cfg.access.users.insert(user.to_string(), secret);
+    }
+    if let Some(ad_tag) = body.user_ad_tag {
+        cfg.access.user_ad_tags.insert(user.to_string(), ad_tag);
+    }
+    if let Some(limit) = body.max_tcp_conns {
+        cfg.access.user_max_tcp_conns.insert(user.to_string(), limit);
+    }
+    if let Some(expiration) = expiration {
+        cfg.access.user_expirations.insert(user.to_string(), expiration);
+    }
+    if let Some(quota) = body.data_quota_bytes {
+        cfg.access.user_data_quota.insert(user.to_string(), quota);
+    }
+
+    let mut updated_limit = None;
+    if let Some(limit) = body.max_unique_ips {
+        cfg.access.user_max_unique_ips.insert(user.to_string(), limit);
+        updated_limit = Some(limit);
+    }
+
+    cfg.validate()
+        .map_err(|e| ApiFailure::bad_request(format!("config validation failed: {}", e)))?;
+
+    let revision = save_config_to_disk(&shared.config_path, &cfg).await?;
+    drop(_guard);
+    if let Some(limit) = updated_limit {
+        shared.ip_tracker.set_user_limit(user, limit).await;
+    }
+    let users = users_from_config(
+        &cfg,
+        &shared.stats,
+        &shared.ip_tracker,
+        shared.startup_detected_ip_v4,
+        shared.startup_detected_ip_v6,
+    )
+    .await;
+    let user_info = users
+        .into_iter()
+        .find(|entry| entry.username == user)
+        .ok_or_else(|| ApiFailure::internal("failed to build updated user view"))?;
+
+    Ok((user_info, revision))
+}
+
+pub(super) async fn rotate_secret(
+    user: &str,
+    body: RotateSecretRequest,
+    expected_revision: Option<String>,
+    shared: &ApiShared,
+) -> Result<(CreateUserResponse, String), ApiFailure> {
+    let secret = body.secret.unwrap_or_else(random_user_secret);
+    if !is_valid_user_secret(&secret) {
+        return Err(ApiFailure::bad_request(
+            "secret must be exactly 32 hex characters",
+        ));
+    }
+
+    let _guard = shared.mutation_lock.lock().await;
+    let mut cfg = load_config_from_disk(&shared.config_path).await?;
+    ensure_expected_revision(&shared.config_path, expected_revision.as_deref()).await?;
+
+    if !cfg.access.users.contains_key(user) {
+        return Err(ApiFailure::new(
+            StatusCode::NOT_FOUND,
+            "not_found",
+            "User not found",
+        ));
+    }
+
+    cfg.access.users.insert(user.to_string(), secret.clone());
+    cfg.validate()
+        .map_err(|e| ApiFailure::bad_request(format!("config validation failed: {}", e)))?;
+    let revision = save_config_to_disk(&shared.config_path, &cfg).await?;
+    drop(_guard);
+
+    let users = users_from_config(
+        &cfg,
+        &shared.stats,
+        &shared.ip_tracker,
+        shared.startup_detected_ip_v4,
+        shared.startup_detected_ip_v6,
+    )
+    .await;
+    let user_info = users
+        .into_iter()
+        .find(|entry| entry.username == user)
+        .ok_or_else(|| ApiFailure::internal("failed to build updated user view"))?;
+
+    Ok((
+        CreateUserResponse {
+            user: user_info,
+            secret,
+        },
+        revision,
+    ))
+}
+
+pub(super) async fn delete_user(
+    user: &str,
+    expected_revision: Option<String>,
+    shared: &ApiShared,
+) -> Result<(String, String), ApiFailure> {
+    let _guard = shared.mutation_lock.lock().await;
+    let mut cfg = load_config_from_disk(&shared.config_path).await?;
+    ensure_expected_revision(&shared.config_path, expected_revision.as_deref()).await?;
+
+    if !cfg.access.users.contains_key(user) {
+        return Err(ApiFailure::new(
+            StatusCode::NOT_FOUND,
+            "not_found",
+            "User not found",
+        ));
+    }
+    if cfg.access.users.len() <= 1 {
+        return Err(ApiFailure::new(
+            StatusCode::CONFLICT,
+            "last_user_forbidden",
+            "Cannot delete the last configured user",
+        ));
+    }
+
+    cfg.access.users.remove(user);
+    cfg.access.user_ad_tags.remove(user);
+    cfg.access.user_max_tcp_conns.remove(user);
+    cfg.access.user_expirations.remove(user);
+    cfg.access.user_data_quota.remove(user);
+    cfg.access.user_max_unique_ips.remove(user);
+
+    cfg.validate()
+        .map_err(|e| ApiFailure::bad_request(format!("config validation failed: {}", e)))?;
+    let revision = save_config_to_disk(&shared.config_path, &cfg).await?;
+    drop(_guard);
+    shared.ip_tracker.remove_user_limit(user).await;
+    shared.ip_tracker.clear_user_ips(user).await;
+
+    Ok((user.to_string(), revision))
+}
+
+pub(super) async fn users_from_config(
+    cfg: &ProxyConfig,
+    stats: &Stats,
+    ip_tracker: &UserIpTracker,
+    startup_detected_ip_v4: Option<IpAddr>,
+    startup_detected_ip_v6: Option<IpAddr>,
+) -> Vec<UserInfo> {
+    let mut names = cfg.access.users.keys().cloned().collect::<Vec<_>>();
+    names.sort();
+    let active_ip_lists = ip_tracker.get_active_ips_for_users(&names).await;
+    let recent_ip_lists = ip_tracker.get_recent_ips_for_users(&names).await;
+
+    let mut users = Vec::with_capacity(names.len());
+    for username in names {
+        let active_ip_list = active_ip_lists
+            .get(&username)
+            .cloned()
+            .unwrap_or_else(Vec::new);
+        let recent_ip_list = recent_ip_lists
+            .get(&username)
+            .cloned()
+            .unwrap_or_else(Vec::new);
+        let links = cfg
+            .access
+            .users
+            .get(&username)
+            .map(|secret| {
+                build_user_links(
+                    cfg,
+                    secret,
+                    startup_detected_ip_v4,
+                    startup_detected_ip_v6,
+                )
+            })
+            .unwrap_or(UserLinks {
+                classic: Vec::new(),
+                secure: Vec::new(),
+                tls: Vec::new(),
+            });
+        users.push(UserInfo {
+            user_ad_tag: cfg.access.user_ad_tags.get(&username).cloned(),
+            max_tcp_conns: cfg.access.user_max_tcp_conns.get(&username).copied(),
+            expiration_rfc3339: cfg
+                .access
+                .user_expirations
+                .get(&username)
+                .map(chrono::DateTime::<chrono::Utc>::to_rfc3339),
+            data_quota_bytes: cfg.access.user_data_quota.get(&username).copied(),
+            max_unique_ips: cfg.access.user_max_unique_ips.get(&username).copied(),
+            current_connections: stats.get_user_curr_connects(&username),
+            active_unique_ips: active_ip_list.len(),
+            active_unique_ips_list: active_ip_list,
+            recent_unique_ips: recent_ip_list.len(),
+            recent_unique_ips_list: recent_ip_list,
+            total_octets: stats.get_user_total_octets(&username),
+            links,
+            username,
+        });
+    }
+    users
+}
+
+fn build_user_links(
+    cfg: &ProxyConfig,
+    secret: &str,
+    startup_detected_ip_v4: Option<IpAddr>,
+    startup_detected_ip_v6: Option<IpAddr>,
+) -> UserLinks {
+    let hosts = resolve_link_hosts(cfg, startup_detected_ip_v4, startup_detected_ip_v6);
+    let port = cfg.general.links.public_port.unwrap_or(cfg.server.port);
+    let tls_domains = resolve_tls_domains(cfg);
+
+    let mut classic = Vec::new();
+    let mut secure = Vec::new();
+    let mut tls = Vec::new();
+
+    for host in &hosts {
+        if cfg.general.modes.classic {
+            classic.push(format!(
+                "tg://proxy?server={}&port={}&secret={}",
+                host, port, secret
+            ));
+        }
+        if cfg.general.modes.secure {
+            secure.push(format!(
+                "tg://proxy?server={}&port={}&secret=dd{}",
+                host, port, secret
+            ));
+        }
+        if cfg.general.modes.tls {
+            for domain in &tls_domains {
+                let domain_hex = hex::encode(domain);
+                tls.push(format!(
+                    "tg://proxy?server={}&port={}&secret=ee{}{}",
+                    host, port, secret, domain_hex
+                ));
+            }
+        }
+    }
+
+    UserLinks {
+        classic,
+        secure,
+        tls,
+    }
+}
+
+fn resolve_link_hosts(
+    cfg: &ProxyConfig,
+    startup_detected_ip_v4: Option<IpAddr>,
+    startup_detected_ip_v6: Option<IpAddr>,
+) -> Vec<String> {
+    if let Some(host) = cfg
+        .general
+        .links
+        .public_host
+        .as_deref()
+        .map(str::trim)
+        .filter(|value| !value.is_empty())
+    {
+        return vec![host.to_string()];
+    }
+
+    let mut startup_hosts = Vec::new();
+    if let Some(ip) = startup_detected_ip_v4 {
+        push_unique_host(&mut startup_hosts, &ip.to_string());
+    }
+    if let Some(ip) = startup_detected_ip_v6 {
+        push_unique_host(&mut startup_hosts, &ip.to_string());
+    }
+    if !startup_hosts.is_empty() {
+        return startup_hosts;
+    }
+
+    let mut hosts = Vec::new();
+    for listener in &cfg.server.listeners {
+        if let Some(host) = listener
+            .announce
+            .as_deref()
+            .map(str::trim)
+            .filter(|value| !value.is_empty())
+        {
+            push_unique_host(&mut hosts, host);
+            continue;
+        }
+        if let Some(ip) = listener.announce_ip {
+            if !ip.is_unspecified() {
+                push_unique_host(&mut hosts, &ip.to_string());
+            }
+            continue;
+        }
+        if !listener.ip.is_unspecified() {
+            push_unique_host(&mut hosts, &listener.ip.to_string());
+        }
+    }
+
+    if hosts.is_empty() {
+        if let Some(host) = cfg.server.listen_addr_ipv4.as_deref() {
+            push_host_from_legacy_listen(&mut hosts, host);
+        }
+        if let Some(host) = cfg.server.listen_addr_ipv6.as_deref() {
+            push_host_from_legacy_listen(&mut hosts, host);
+        }
+    }
+
+    hosts
+}
+
+fn push_host_from_legacy_listen(hosts: &mut Vec<String>, raw: &str) {
+    let candidate = raw.trim();
+    if candidate.is_empty() {
+        return;
+    }
+
+    match candidate.parse::<IpAddr>() {
+        Ok(ip) if ip.is_unspecified() => {}
+        Ok(ip) => push_unique_host(hosts, &ip.to_string()),
+        Err(_) => push_unique_host(hosts, candidate),
+    }
+}
+
+fn push_unique_host(hosts: &mut Vec<String>, candidate: &str) {
+    if !hosts.iter().any(|existing| existing == candidate) {
+        hosts.push(candidate.to_string());
+    }
+}
+
+fn resolve_tls_domains(cfg: &ProxyConfig) -> Vec<&str> {
+    let mut domains = Vec::with_capacity(1 + cfg.censorship.tls_domains.len());
+    let primary = cfg.censorship.tls_domain.as_str();
+    if !primary.is_empty() {
+        domains.push(primary);
+    }
+    for domain in &cfg.censorship.tls_domains {
+        let value = domain.as_str();
+        if value.is_empty() || domains.contains(&value) {
+            continue;
+        }
+        domains.push(value);
+    }
+    domains
+}

src/cli.rs

@@ -0,0 +1,315 @@
+//! CLI commands: --init (fire-and-forget setup)
+
+use std::fs;
+use std::path::{Path, PathBuf};
+use std::process::Command;
+use rand::Rng;
+
+/// Options for the init command
+pub struct InitOptions {
+    pub port: u16,
+    pub domain: String,
+    pub secret: Option<String>,
+    pub username: String,
+    pub config_dir: PathBuf,
+    pub no_start: bool,
+}
+
+impl Default for InitOptions {
+    fn default() -> Self {
+        Self {
+            port: 443,
+            domain: "www.google.com".to_string(),
+            secret: None,
+            username: "user".to_string(),
+            config_dir: PathBuf::from("/etc/telemt"),
+            no_start: false,
+        }
+    }
+}
+
+/// Parse --init subcommand options from CLI args.
+///
+/// Returns `Some(InitOptions)` if `--init` was found, `None` otherwise.
+pub fn parse_init_args(args: &[String]) -> Option<InitOptions> {
+    if !args.iter().any(|a| a == "--init") {
+        return None;
+    }
+    
+    let mut opts = InitOptions::default();
+    let mut i = 0;
+    
+    while i < args.len() {
+        match args[i].as_str() {
+            "--port" => {
+                i += 1;
+                if i < args.len() {
+                    opts.port = args[i].parse().unwrap_or(443);
+                }
+            }
+            "--domain" => {
+                i += 1;
+                if i < args.len() {
+                    opts.domain = args[i].clone();
+                }
+            }
+            "--secret" => {
+                i += 1;
+                if i < args.len() {
+                    opts.secret = Some(args[i].clone());
+                }
+            }
+            "--user" => {
+                i += 1;
+                if i < args.len() {
+                    opts.username = args[i].clone();
+                }
+            }
+            "--config-dir" => {
+                i += 1;
+                if i < args.len() {
+                    opts.config_dir = PathBuf::from(&args[i]);
+                }
+            }
+            "--no-start" => {
+                opts.no_start = true;
+            }
+            _ => {}
+        }
+        i += 1;
+    }
+    
+    Some(opts)
+}
+
+/// Run the fire-and-forget setup.
+pub fn run_init(opts: InitOptions) -> Result<(), Box<dyn std::error::Error>> {
+    eprintln!("[telemt] Fire-and-forget setup");
+    eprintln!();
+    
+    // 1. Generate or validate secret
+    let secret = match opts.secret {
+        Some(s) => {
+            if s.len() != 32 || !s.chars().all(|c| c.is_ascii_hexdigit()) {
+                eprintln!("[error] Secret must be exactly 32 hex characters");
+                std::process::exit(1);
+            }
+            s
+        }
+        None => generate_secret(),
+    };
+    
+    eprintln!("[+] Secret: {}", secret);
+    eprintln!("[+] User:   {}", opts.username);
+    eprintln!("[+] Port:   {}", opts.port);
+    eprintln!("[+] Domain: {}", opts.domain);
+    
+    // 2. Create config directory
+    fs::create_dir_all(&opts.config_dir)?;
+    let config_path = opts.config_dir.join("config.toml");
+    
+    // 3. Write config
+    let config_content = generate_config(&opts.username, &secret, opts.port, &opts.domain);
+    fs::write(&config_path, &config_content)?;
+    eprintln!("[+] Config written to {}", config_path.display());
+    
+    // 4. Write systemd unit
+    let exe_path = std::env::current_exe()
+        .unwrap_or_else(|_| PathBuf::from("/usr/local/bin/telemt"));
+    
+    let unit_path = Path::new("/etc/systemd/system/telemt.service");
+    let unit_content = generate_systemd_unit(&exe_path, &config_path);
+    
+    match fs::write(unit_path, &unit_content) {
+        Ok(()) => {
+            eprintln!("[+] Systemd unit written to {}", unit_path.display());
+        }
+        Err(e) => {
+            eprintln!("[!] Cannot write systemd unit (run as root?): {}", e);
+            eprintln!("[!] Manual unit file content:");
+            eprintln!("{}", unit_content);
+            
+            // Still print links and config
+            print_links(&opts.username, &secret, opts.port, &opts.domain);
+            return Ok(());
+        }
+    }
+    
+    // 5. Reload systemd
+    run_cmd("systemctl", &["daemon-reload"]);
+    
+    // 6. Enable service
+    run_cmd("systemctl", &["enable", "telemt.service"]);
+    eprintln!("[+] Service enabled");
+    
+    // 7. Start service (unless --no-start)
+    if !opts.no_start {
+        run_cmd("systemctl", &["start", "telemt.service"]);
+        eprintln!("[+] Service started");
+        
+        // Brief delay then check status
+        std::thread::sleep(std::time::Duration::from_secs(1));
+        let status = Command::new("systemctl")
+            .args(["is-active", "telemt.service"])
+            .output();
+        
+        match status {
+            Ok(out) if out.status.success() => {
+                eprintln!("[+] Service is running");
+            }
+            _ => {
+                eprintln!("[!] Service may not have started correctly");
+                eprintln!("[!] Check: journalctl -u telemt.service -n 20");
+            }
+        }
+    } else {
+        eprintln!("[+] Service not started (--no-start)");
+        eprintln!("[+] Start manually: systemctl start telemt.service");
+    }
+    
+    eprintln!();
+    
+    // 8. Print links
+    print_links(&opts.username, &secret, opts.port, &opts.domain);
+    
+    Ok(())
+}
+
+fn generate_secret() -> String {
+    let mut rng = rand::rng();
+    let bytes: Vec<u8> = (0..16).map(|_| rng.random::<u8>()).collect();
+    hex::encode(bytes)
+}
+
+fn generate_config(username: &str, secret: &str, port: u16, domain: &str) -> String {
+    format!(
+r#"# Telemt MTProxy — auto-generated config
+# Re-run `telemt --init` to regenerate
+
+show_link = ["{username}"]
+
+[general]
+# prefer_ipv6 is deprecated; use [network].prefer
+prefer_ipv6 = false
+fast_mode = true
+use_middle_proxy = false
+log_level = "normal"
+desync_all_full = false
+update_every = 43200
+hardswap = false
+me_pool_drain_ttl_secs = 90
+me_pool_min_fresh_ratio = 0.8
+me_reinit_drain_timeout_secs = 120
+
+[network]
+ipv4 = true
+ipv6 = true
+prefer = 4
+multipath = false
+
+[general.modes]
+classic = false
+secure = false
+tls = true
+
+[server]
+port = {port}
+listen_addr_ipv4 = "0.0.0.0"
+listen_addr_ipv6 = "::"
+
+[[server.listeners]]
+ip = "0.0.0.0"
+# reuse_allow = false # Set true only when intentionally running multiple telemt instances on same port
+
+[[server.listeners]]
+ip = "::"
+
+[timeouts]
+client_handshake = 15
+tg_connect = 10
+client_keepalive = 60
+client_ack = 300
+
+[censorship]
+tls_domain = "{domain}"
+mask = true
+mask_port = 443
+fake_cert_len = 2048
+tls_full_cert_ttl_secs = 90
+
+[access]
+replay_check_len = 65536
+replay_window_secs = 1800
+ignore_time_skew = false
+
+[access.users]
+{username} = "{secret}"
+
+[[upstreams]]
+type = "direct"
+enabled = true
+weight = 10
+"#,
+        username = username,
+        secret = secret,
+        port = port,
+        domain = domain,
+    )
+}
+
+fn generate_systemd_unit(exe_path: &Path, config_path: &Path) -> String {
+    format!(
+r#"[Unit]
+Description=Telemt MTProxy
+Documentation=https://github.com/nicepkg/telemt
+After=network-online.target
+Wants=network-online.target
+
+[Service]
+Type=simple
+ExecStart={exe} {config}
+Restart=always
+RestartSec=5
+LimitNOFILE=65535
+# Security hardening
+NoNewPrivileges=true
+ProtectSystem=strict
+ProtectHome=true
+ReadWritePaths=/etc/telemt
+PrivateTmp=true
+
+[Install]
+WantedBy=multi-user.target
+"#,
+        exe = exe_path.display(),
+        config = config_path.display(),
+    )
+}
+
+fn run_cmd(cmd: &str, args: &[&str]) {
+    match Command::new(cmd).args(args).output() {
+        Ok(output) => {
+            if !output.status.success() {
+                let stderr = String::from_utf8_lossy(&output.stderr);
+                eprintln!("[!] {} {} failed: {}", cmd, args.join(" "), stderr.trim());
+            }
+        }
+        Err(e) => {
+            eprintln!("[!] Failed to run {} {}: {}", cmd, args.join(" "), e);
+        }
+    }
+}
+
+fn print_links(username: &str, secret: &str, port: u16, domain: &str) {
+    let domain_hex = hex::encode(domain);
+    
+    println!("=== Proxy Links ===");
+    println!("[{}]", username);
+    println!("  EE-TLS:  tg://proxy?server=YOUR_SERVER_IP&port={}&secret=ee{}{}", 
+        port, secret, domain_hex);
+    println!();
+    println!("Replace YOUR_SERVER_IP with your server's public IP.");
+    println!("The proxy will auto-detect and display the correct link on startup.");
+    println!("Check: journalctl -u telemt.service | head -30");
+    println!("===================");
+}

src/config/defaults.rs

@@ -0,0 +1,532 @@
+use std::collections::HashMap;
+use ipnetwork::IpNetwork;
+use serde::Deserialize;
+
+// Helper defaults kept private to the config module.
+const DEFAULT_NETWORK_IPV6: Option<bool> = Some(false);
+const DEFAULT_STUN_TCP_FALLBACK: bool = true;
+const DEFAULT_MIDDLE_PROXY_WARM_STANDBY: usize = 16;
+const DEFAULT_ME_RECONNECT_MAX_CONCURRENT_PER_DC: u32 = 8;
+const DEFAULT_ME_RECONNECT_FAST_RETRY_COUNT: u32 = 16;
+const DEFAULT_ME_SINGLE_ENDPOINT_SHADOW_WRITERS: u8 = 2;
+const DEFAULT_ME_ADAPTIVE_FLOOR_IDLE_SECS: u64 = 90;
+const DEFAULT_ME_ADAPTIVE_FLOOR_MIN_WRITERS_SINGLE_ENDPOINT: u8 = 1;
+const DEFAULT_ME_ADAPTIVE_FLOOR_RECOVER_GRACE_SECS: u64 = 180;
+const DEFAULT_USER_MAX_UNIQUE_IPS_WINDOW_SECS: u64 = 30;
+const DEFAULT_UPSTREAM_CONNECT_RETRY_ATTEMPTS: u32 = 2;
+const DEFAULT_UPSTREAM_UNHEALTHY_FAIL_THRESHOLD: u32 = 5;
+const DEFAULT_LISTEN_ADDR_IPV6: &str = "::";
+const DEFAULT_ACCESS_USER: &str = "default";
+const DEFAULT_ACCESS_SECRET: &str = "00000000000000000000000000000000";
+
+pub(crate) fn default_true() -> bool {
+    true
+}
+
+pub(crate) fn default_port() -> u16 {
+    443
+}
+
+pub(crate) fn default_tls_domain() -> String {
+    "petrovich.ru".to_string()
+}
+
+pub(crate) fn default_mask_port() -> u16 {
+    443
+}
+
+pub(crate) fn default_fake_cert_len() -> usize {
+    2048
+}
+
+pub(crate) fn default_tls_front_dir() -> String {
+    "tlsfront".to_string()
+}
+
+pub(crate) fn default_replay_check_len() -> usize {
+    65_536
+}
+
+pub(crate) fn default_replay_window_secs() -> u64 {
+    1800
+}
+
+pub(crate) fn default_handshake_timeout() -> u64 {
+    30
+}
+
+pub(crate) fn default_connect_timeout() -> u64 {
+    10
+}
+
+pub(crate) fn default_keepalive() -> u64 {
+    60
+}
+
+pub(crate) fn default_ack_timeout() -> u64 {
+    300
+}
+pub(crate) fn default_me_one_retry() -> u8 {
+    12
+}
+
+pub(crate) fn default_me_one_timeout() -> u64 {
+    1200
+}
+
+pub(crate) fn default_listen_addr() -> String {
+    "0.0.0.0".to_string()
+}
+
+pub(crate) fn default_listen_addr_ipv4() -> Option<String> {
+    Some(default_listen_addr())
+}
+
+pub(crate) fn default_weight() -> u16 {
+    1
+}
+
+pub(crate) fn default_metrics_whitelist() -> Vec<IpNetwork> {
+    vec![
+        "127.0.0.1/32".parse().unwrap(),
+        "::1/128".parse().unwrap(),
+    ]
+}
+
+pub(crate) fn default_api_listen() -> String {
+    "127.0.0.1:9091".to_string()
+}
+
+pub(crate) fn default_api_whitelist() -> Vec<IpNetwork> {
+    default_metrics_whitelist()
+}
+
+pub(crate) fn default_api_request_body_limit_bytes() -> usize {
+    64 * 1024
+}
+
+pub(crate) fn default_api_minimal_runtime_enabled() -> bool {
+    false
+}
+
+pub(crate) fn default_api_minimal_runtime_cache_ttl_ms() -> u64 {
+    1000
+}
+
+pub(crate) fn default_prefer_4() -> u8 {
+    4
+}
+
+pub(crate) fn default_network_ipv6() -> Option<bool> {
+    DEFAULT_NETWORK_IPV6
+}
+
+pub(crate) fn default_stun_tcp_fallback() -> bool {
+    DEFAULT_STUN_TCP_FALLBACK
+}
+
+pub(crate) fn default_unknown_dc_log_path() -> Option<String> {
+    Some("unknown-dc.txt".to_string())
+}
+
+pub(crate) fn default_unknown_dc_file_log_enabled() -> bool {
+    false
+}
+
+pub(crate) fn default_pool_size() -> usize {
+    8
+}
+
+pub(crate) fn default_proxy_secret_path() -> Option<String> {
+    Some("proxy-secret".to_string())
+}
+
+pub(crate) fn default_proxy_config_v4_cache_path() -> Option<String> {
+    Some("cache/proxy-config-v4.txt".to_string())
+}
+
+pub(crate) fn default_proxy_config_v6_cache_path() -> Option<String> {
+    Some("cache/proxy-config-v6.txt".to_string())
+}
+
+pub(crate) fn default_middle_proxy_nat_stun() -> Option<String> {
+    None
+}
+
+pub(crate) fn default_middle_proxy_nat_stun_servers() -> Vec<String> {
+    Vec::new()
+}
+
+pub(crate) fn default_stun_nat_probe_concurrency() -> usize {
+    8
+}
+
+pub(crate) fn default_middle_proxy_warm_standby() -> usize {
+    DEFAULT_MIDDLE_PROXY_WARM_STANDBY
+}
+
+pub(crate) fn default_me_init_retry_attempts() -> u32 {
+    0
+}
+
+pub(crate) fn default_me2dc_fallback() -> bool {
+    true
+}
+
+pub(crate) fn default_keepalive_interval() -> u64 {
+    8
+}
+
+pub(crate) fn default_keepalive_jitter() -> u64 {
+    2
+}
+
+pub(crate) fn default_warmup_step_delay_ms() -> u64 {
+    500
+}
+
+pub(crate) fn default_warmup_step_jitter_ms() -> u64 {
+    300
+}
+
+pub(crate) fn default_reconnect_backoff_base_ms() -> u64 {
+    500
+}
+
+pub(crate) fn default_reconnect_backoff_cap_ms() -> u64 {
+    30_000
+}
+
+pub(crate) fn default_me_reconnect_max_concurrent_per_dc() -> u32 {
+    DEFAULT_ME_RECONNECT_MAX_CONCURRENT_PER_DC
+}
+
+pub(crate) fn default_me_reconnect_fast_retry_count() -> u32 {
+    DEFAULT_ME_RECONNECT_FAST_RETRY_COUNT
+}
+
+pub(crate) fn default_me_single_endpoint_shadow_writers() -> u8 {
+    DEFAULT_ME_SINGLE_ENDPOINT_SHADOW_WRITERS
+}
+
+pub(crate) fn default_me_single_endpoint_outage_mode_enabled() -> bool {
+    true
+}
+
+pub(crate) fn default_me_single_endpoint_outage_disable_quarantine() -> bool {
+    true
+}
+
+pub(crate) fn default_me_single_endpoint_outage_backoff_min_ms() -> u64 {
+    250
+}
+
+pub(crate) fn default_me_single_endpoint_outage_backoff_max_ms() -> u64 {
+    3000
+}
+
+pub(crate) fn default_me_single_endpoint_shadow_rotate_every_secs() -> u64 {
+    900
+}
+
+pub(crate) fn default_me_adaptive_floor_idle_secs() -> u64 {
+    DEFAULT_ME_ADAPTIVE_FLOOR_IDLE_SECS
+}
+
+pub(crate) fn default_me_adaptive_floor_min_writers_single_endpoint() -> u8 {
+    DEFAULT_ME_ADAPTIVE_FLOOR_MIN_WRITERS_SINGLE_ENDPOINT
+}
+
+pub(crate) fn default_me_adaptive_floor_recover_grace_secs() -> u64 {
+    DEFAULT_ME_ADAPTIVE_FLOOR_RECOVER_GRACE_SECS
+}
+
+pub(crate) fn default_upstream_connect_retry_attempts() -> u32 {
+    DEFAULT_UPSTREAM_CONNECT_RETRY_ATTEMPTS
+}
+
+pub(crate) fn default_upstream_connect_retry_backoff_ms() -> u64 {
+    100
+}
+
+pub(crate) fn default_upstream_unhealthy_fail_threshold() -> u32 {
+    DEFAULT_UPSTREAM_UNHEALTHY_FAIL_THRESHOLD
+}
+
+pub(crate) fn default_upstream_connect_failfast_hard_errors() -> bool {
+    false
+}
+
+pub(crate) fn default_rpc_proxy_req_every() -> u64 {
+    0
+}
+
+pub(crate) fn default_crypto_pending_buffer() -> usize {
+    256 * 1024
+}
+
+pub(crate) fn default_max_client_frame() -> usize {
+    16 * 1024 * 1024
+}
+
+pub(crate) fn default_desync_all_full() -> bool {
+    false
+}
+
+pub(crate) fn default_me_route_backpressure_base_timeout_ms() -> u64 {
+    25
+}
+
+pub(crate) fn default_me_route_backpressure_high_timeout_ms() -> u64 {
+    120
+}
+
+pub(crate) fn default_me_route_backpressure_high_watermark_pct() -> u8 {
+    80
+}
+
+pub(crate) fn default_me_route_no_writer_wait_ms() -> u64 {
+    250
+}
+
+pub(crate) fn default_me_route_inline_recovery_attempts() -> u32 {
+    3
+}
+
+pub(crate) fn default_me_route_inline_recovery_wait_ms() -> u64 {
+    3000
+}
+
+pub(crate) fn default_beobachten_minutes() -> u64 {
+    10
+}
+
+pub(crate) fn default_beobachten_flush_secs() -> u64 {
+    15
+}
+
+pub(crate) fn default_beobachten_file() -> String {
+    "cache/beobachten.txt".to_string()
+}
+
+pub(crate) fn default_tls_new_session_tickets() -> u8 {
+    0
+}
+
+pub(crate) fn default_tls_full_cert_ttl_secs() -> u64 {
+    90
+}
+
+pub(crate) fn default_server_hello_delay_min_ms() -> u64 {
+    0
+}
+
+pub(crate) fn default_server_hello_delay_max_ms() -> u64 {
+    0
+}
+
+pub(crate) fn default_alpn_enforce() -> bool {
+    true
+}
+
+pub(crate) fn default_stun_servers() -> Vec<String> {
+    vec![
+        "stun.l.google.com:5349".to_string(),
+        "stun1.l.google.com:3478".to_string(),
+        "stun.gmx.net:3478".to_string(),
+        "stun.l.google.com:19302".to_string(),
+        "stun.1und1.de:3478".to_string(),
+        "stun1.l.google.com:19302".to_string(),
+        "stun2.l.google.com:19302".to_string(),
+        "stun3.l.google.com:19302".to_string(),
+        "stun4.l.google.com:19302".to_string(),
+        "stun.services.mozilla.com:3478".to_string(),
+        "stun.stunprotocol.org:3478".to_string(),
+        "stun.nextcloud.com:3478".to_string(),
+        "stun.voip.eutelia.it:3478".to_string(),
+    ]
+}
+
+pub(crate) fn default_http_ip_detect_urls() -> Vec<String> {
+    vec![
+        "https://ifconfig.me/ip".to_string(),
+        "https://api.ipify.org".to_string(),
+    ]
+}
+
+pub(crate) fn default_cache_public_ip_path() -> String {
+    "cache/public_ip.txt".to_string()
+}
+
+pub(crate) fn default_proxy_secret_reload_secs() -> u64 {
+    60 * 60
+}
+
+pub(crate) fn default_proxy_config_reload_secs() -> u64 {
+    60 * 60
+}
+
+pub(crate) fn default_update_every_secs() -> u64 {
+    5 * 60
+}
+
+pub(crate) fn default_update_every() -> Option<u64> {
+    Some(default_update_every_secs())
+}
+
+pub(crate) fn default_me_reinit_every_secs() -> u64 {
+    15 * 60
+}
+
+pub(crate) fn default_me_reinit_singleflight() -> bool {
+    true
+}
+
+pub(crate) fn default_me_reinit_trigger_channel() -> usize {
+    64
+}
+
+pub(crate) fn default_me_reinit_coalesce_window_ms() -> u64 {
+    200
+}
+
+pub(crate) fn default_me_hardswap_warmup_delay_min_ms() -> u64 {
+    1000
+}
+
+pub(crate) fn default_me_hardswap_warmup_delay_max_ms() -> u64 {
+    2000
+}
+
+pub(crate) fn default_me_hardswap_warmup_extra_passes() -> u8 {
+    3
+}
+
+pub(crate) fn default_me_hardswap_warmup_pass_backoff_base_ms() -> u64 {
+    500
+}
+
+pub(crate) fn default_me_config_stable_snapshots() -> u8 {
+    2
+}
+
+pub(crate) fn default_me_config_apply_cooldown_secs() -> u64 {
+    300
+}
+
+pub(crate) fn default_me_snapshot_require_http_2xx() -> bool {
+    true
+}
+
+pub(crate) fn default_me_snapshot_reject_empty_map() -> bool {
+    true
+}
+
+pub(crate) fn default_me_snapshot_min_proxy_for_lines() -> u32 {
+    1
+}
+
+pub(crate) fn default_proxy_secret_stable_snapshots() -> u8 {
+    2
+}
+
+pub(crate) fn default_proxy_secret_rotate_runtime() -> bool {
+    true
+}
+
+pub(crate) fn default_me_secret_atomic_snapshot() -> bool {
+    true
+}
+
+pub(crate) fn default_proxy_secret_len_max() -> usize {
+    256
+}
+
+pub(crate) fn default_me_reinit_drain_timeout_secs() -> u64 {
+    120
+}
+
+pub(crate) fn default_me_pool_drain_ttl_secs() -> u64 {
+    90
+}
+
+pub(crate) fn default_me_bind_stale_ttl_secs() -> u64 {
+    default_me_pool_drain_ttl_secs()
+}
+
+pub(crate) fn default_me_pool_min_fresh_ratio() -> f32 {
+    0.8
+}
+
+pub(crate) fn default_me_deterministic_writer_sort() -> bool {
+    true
+}
+
+pub(crate) fn default_hardswap() -> bool {
+    true
+}
+
+pub(crate) fn default_ntp_check() -> bool {
+    true
+}
+
+pub(crate) fn default_ntp_servers() -> Vec<String> {
+    vec!["pool.ntp.org".to_string()]
+}
+
+pub(crate) fn default_fast_mode_min_tls_record() -> usize {
+    0
+}
+
+pub(crate) fn default_degradation_min_unavailable_dc_groups() -> u8 {
+    2
+}
+
+pub(crate) fn default_listen_addr_ipv6() -> String {
+    DEFAULT_LISTEN_ADDR_IPV6.to_string()
+}
+
+pub(crate) fn default_listen_addr_ipv6_opt() -> Option<String> {
+    Some(default_listen_addr_ipv6())
+}
+
+pub(crate) fn default_access_users() -> HashMap<String, String> {
+    HashMap::from([(
+        DEFAULT_ACCESS_USER.to_string(),
+        DEFAULT_ACCESS_SECRET.to_string(),
+    )])
+}
+
+pub(crate) fn default_user_max_unique_ips_window_secs() -> u64 {
+    DEFAULT_USER_MAX_UNIQUE_IPS_WINDOW_SECS
+}
+
+// Custom deserializer helpers
+
+#[derive(Deserialize)]
+#[serde(untagged)]
+pub(crate) enum OneOrMany {
+    One(String),
+    Many(Vec<String>),
+}
+
+pub(crate) fn deserialize_dc_overrides<'de, D>(
+    deserializer: D,
+) -> std::result::Result<HashMap<String, Vec<String>>, D::Error>
+where
+    D: serde::de::Deserializer<'de>,
+{
+    let raw: HashMap<String, OneOrMany> = HashMap::deserialize(deserializer)?;
+    let mut out = HashMap::new();
+    for (dc, val) in raw {
+        let mut addrs = match val {
+            OneOrMany::One(s) => vec![s],
+            OneOrMany::Many(v) => v,
+        };
+        addrs.retain(|s| !s.trim().is_empty());
+        if !addrs.is_empty() {
+            out.insert(dc, addrs);
+        }
+    }
+    Ok(out)
+}

src/config/mod.rs

@@ -1,287 +1,9 @@
-//! Configuration
+//! Configuration.
 
-use std::collections::HashMap;
-use std::net::IpAddr;
-use std::path::Path;
-use chrono::{DateTime, Utc};
-use serde::{Deserialize, Serialize};
-use crate::error::{ProxyError, Result};
+pub(crate) mod defaults;
+mod types;
+mod load;
+pub mod hot_reload;
 
-#[derive(Debug, Clone, Serialize, Deserialize)]
-pub struct ProxyModes {
-    #[serde(default)]
-    pub classic: bool,
-    #[serde(default)]
-    pub secure: bool,
-    #[serde(default = "default_true")]
-    pub tls: bool,
-}
-
-fn default_true() -> bool { true }
-fn default_weight() -> u16 { 1 }
-
-impl Default for ProxyModes {
-    fn default() -> Self {
-        Self { classic: true, secure: true, tls: true }
-    }
-}
-
-#[derive(Debug, Clone, Serialize, Deserialize, PartialEq)]
-#[serde(tag = "type", rename_all = "lowercase")]
-pub enum UpstreamType {
-    Direct {
-        #[serde(default)]
-        interface: Option<String>, // Bind to specific IP/Interface
-    },
-    Socks4 {
-        address: String, // IP:Port of SOCKS server
-        #[serde(default)]
-        interface: Option<String>, // Bind to specific IP/Interface for connection to SOCKS
-        #[serde(default)]
-        user_id: Option<String>,
-    },
-    Socks5 {
-        address: String,
-        #[serde(default)]
-        interface: Option<String>,
-        #[serde(default)]
-        username: Option<String>,
-        #[serde(default)]
-        password: Option<String>,
-    },
-}
-
-#[derive(Debug, Clone, Serialize, Deserialize)]
-pub struct UpstreamConfig {
-    #[serde(flatten)]
-    pub upstream_type: UpstreamType,
-    #[serde(default = "default_weight")]
-    pub weight: u16,
-    #[serde(default = "default_true")]
-    pub enabled: bool,
-}
-
-#[derive(Debug, Clone, Serialize, Deserialize)]
-pub struct ListenerConfig {
-    pub ip: IpAddr,
-    #[serde(default)]
-    pub announce_ip: Option<IpAddr>, // IP to show in tg:// links
-}
-
-#[derive(Debug, Clone, Serialize, Deserialize)]
-pub struct ProxyConfig {
-    #[serde(default = "default_port")]
-    pub port: u16,
-    
-    #[serde(default)]
-    pub users: HashMap<String, String>,
-    
-    #[serde(default)]
-    pub ad_tag: Option<String>,
-    
-    #[serde(default)]
-    pub modes: ProxyModes,
-    
-    #[serde(default = "default_tls_domain")]
-    pub tls_domain: String,
-    
-    #[serde(default = "default_true")]
-    pub mask: bool,
-    
-    #[serde(default)]
-    pub mask_host: Option<String>,
-    
-    #[serde(default = "default_mask_port")]
-    pub mask_port: u16,
-    
-    #[serde(default)]
-    pub prefer_ipv6: bool,
-    
-    #[serde(default = "default_true")]
-    pub fast_mode: bool,
-    
-    #[serde(default)]
-    pub use_middle_proxy: bool,
-    
-    #[serde(default)]
-    pub user_max_tcp_conns: HashMap<String, usize>,
-    
-    #[serde(default)]
-    pub user_expirations: HashMap<String, DateTime<Utc>>,
-    
-    #[serde(default)]
-    pub user_data_quota: HashMap<String, u64>,
-    
-    #[serde(default = "default_replay_check_len")]
-    pub replay_check_len: usize,
-    
-    #[serde(default)]
-    pub ignore_time_skew: bool,
-    
-    #[serde(default = "default_handshake_timeout")]
-    pub client_handshake_timeout: u64,
-    
-    #[serde(default = "default_connect_timeout")]
-    pub tg_connect_timeout: u64,
-    
-    #[serde(default = "default_keepalive")]
-    pub client_keepalive: u64,
-    
-    #[serde(default = "default_ack_timeout")]
-    pub client_ack_timeout: u64,
-    
-    #[serde(default = "default_listen_addr")]
-    pub listen_addr_ipv4: String,
-    
-    #[serde(default)]
-    pub listen_addr_ipv6: Option<String>,
-    
-    #[serde(default)]
-    pub listen_unix_sock: Option<String>,
-    
-    #[serde(default)]
-    pub metrics_port: Option<u16>,
-    
-    #[serde(default = "default_metrics_whitelist")]
-    pub metrics_whitelist: Vec<IpAddr>,
-    
-    #[serde(default = "default_fake_cert_len")]
-    pub fake_cert_len: usize,
-
-    // New fields
-    #[serde(default)]
-    pub upstreams: Vec<UpstreamConfig>,
-
-    #[serde(default)]
-    pub listeners: Vec<ListenerConfig>,
-
-    #[serde(default)]
-    pub show_link: Vec<String>,
-}
-
-fn default_port() -> u16 { 443 }
-fn default_tls_domain() -> String { "www.google.com".to_string() }
-fn default_mask_port() -> u16 { 443 }
-fn default_replay_check_len() -> usize { 65536 }
-fn default_handshake_timeout() -> u64 { 10 }
-fn default_connect_timeout() -> u64 { 10 }
-fn default_keepalive() -> u64 { 600 }
-fn default_ack_timeout() -> u64 { 300 }
-fn default_listen_addr() -> String { "0.0.0.0".to_string() }
-fn default_fake_cert_len() -> usize { 2048 }
-
-fn default_metrics_whitelist() -> Vec<IpAddr> {
-    vec![
-        "127.0.0.1".parse().unwrap(),
-        "::1".parse().unwrap(),
-    ]
-}
-
-impl Default for ProxyConfig {
-    fn default() -> Self {
-        let mut users = HashMap::new();
-        users.insert("default".to_string(), "00000000000000000000000000000000".to_string());
-        
-        Self {
-            port: default_port(),
-            users,
-            ad_tag: None,
-            modes: ProxyModes::default(),
-            tls_domain: default_tls_domain(),
-            mask: true,
-            mask_host: None,
-            mask_port: default_mask_port(),
-            prefer_ipv6: false,
-            fast_mode: true,
-            use_middle_proxy: false,
-            user_max_tcp_conns: HashMap::new(),
-            user_expirations: HashMap::new(),
-            user_data_quota: HashMap::new(),
-            replay_check_len: default_replay_check_len(),
-            ignore_time_skew: false,
-            client_handshake_timeout: default_handshake_timeout(),
-            tg_connect_timeout: default_connect_timeout(),
-            client_keepalive: default_keepalive(),
-            client_ack_timeout: default_ack_timeout(),
-            listen_addr_ipv4: default_listen_addr(),
-            listen_addr_ipv6: Some("::".to_string()),
-            listen_unix_sock: None,
-            metrics_port: None,
-            metrics_whitelist: default_metrics_whitelist(),
-            fake_cert_len: default_fake_cert_len(),
-            upstreams: Vec::new(),
-            listeners: Vec::new(),
-            show_link: Vec::new(),
-        }
-    }
-}
-
-impl ProxyConfig {
-    pub fn load<P: AsRef<Path>>(path: P) -> Result<Self> {
-        let content = std::fs::read_to_string(path)
-            .map_err(|e| ProxyError::Config(e.to_string()))?;
-        
-        let mut config: ProxyConfig = toml::from_str(&content)
-            .map_err(|e| ProxyError::Config(e.to_string()))?;
-        
-        // Validate secrets
-        for (user, secret) in &config.users {
-            if !secret.chars().all(|c| c.is_ascii_hexdigit()) || secret.len() != 32 {
-                return Err(ProxyError::InvalidSecret {
-                    user: user.clone(),
-                    reason: "Must be 32 hex characters".to_string(),
-                });
-            }
-        }
-        
-        // Default mask_host
-        if config.mask_host.is_none() {
-            config.mask_host = Some(config.tls_domain.clone());
-        }
-        
-        // Random fake_cert_len
-        use rand::Rng;
-        config.fake_cert_len = rand::thread_rng().gen_range(1024..4096);
-        
-        // Migration: Populate listeners if empty
-        if config.listeners.is_empty() {
-            if let Ok(ipv4) = config.listen_addr_ipv4.parse::<IpAddr>() {
-                config.listeners.push(ListenerConfig {
-                    ip: ipv4,
-                    announce_ip: None,
-                });
-            }
-            if let Some(ipv6_str) = &config.listen_addr_ipv6 {
-                 if let Ok(ipv6) = ipv6_str.parse::<IpAddr>() {
-                    config.listeners.push(ListenerConfig {
-                        ip: ipv6,
-                        announce_ip: None,
-                    });
-                }
-            }
-        }
-
-        // Migration: Populate upstreams if empty (Default Direct)
-        if config.upstreams.is_empty() {
-             config.upstreams.push(UpstreamConfig {
-                upstream_type: UpstreamType::Direct { interface: None },
-                weight: 1,
-                enabled: true,
-            });
-        }
-        
-        Ok(config)
-    }
-    
-    pub fn validate(&self) -> Result<()> {
-        if self.users.is_empty() {
-            return Err(ProxyError::Config("No users configured".to_string()));
-        }
-        
-        if !self.modes.classic && !self.modes.secure && !self.modes.tls {
-            return Err(ProxyError::Config("No modes enabled".to_string()));
-        }
-        
-        Ok(())
-    }
-}
+pub use load::ProxyConfig;
+pub use types::*;

src/crypto/aes.rs

@@ -1,9 +1,21 @@
 //! AES encryption implementations
 //!
 //! Provides AES-256-CTR and AES-256-CBC modes for MTProto encryption.
+//!
+//! ## Zeroize policy
+//!
+//! - `AesCbc` stores raw key/IV bytes and zeroizes them on drop.
+//! - `AesCtr` wraps an opaque `Aes256Ctr` cipher from the `ctr` crate.
+//!   The expanded key schedule lives inside that type and cannot be
+//!   zeroized from outside. Callers that hold raw key material (e.g.
+//!   `HandshakeSuccess`, `ObfuscationParams`) are responsible for
+//!   zeroizing their own copies.
+
+#![allow(dead_code)]
 
 use aes::Aes256;
 use ctr::{Ctr128BE, cipher::{KeyIvInit, StreamCipher}};
+use zeroize::Zeroize;
 use crate::error::{ProxyError, Result};
 
 type Aes256Ctr = Ctr128BE<Aes256>;
@@ -11,8 +23,13 @@ type Aes256Ctr = Ctr128BE<Aes256>;
 // ============= AES-256-CTR =============
 
 /// AES-256-CTR encryptor/decryptor
-/// 
-/// CTR mode is symmetric - encryption and decryption are the same operation.
+///
+/// CTR mode is symmetric — encryption and decryption are the same operation.
+///
+/// **Zeroize note:** The inner `Aes256Ctr` cipher state (expanded key schedule
+///     + counter) is opaque and cannot be zeroized. If you need to protect key
+///     material, zeroize the `[u8; 32]` key and `u128` IV at the call site
+///     before dropping them.
 pub struct AesCtr {
     cipher: Aes256Ctr,
 }
@@ -62,14 +79,23 @@ impl AesCtr {
 
 /// AES-256-CBC cipher with proper chaining
 ///
-/// Unlike CTR mode, CBC is NOT symmetric - encryption and decryption
+/// Unlike CTR mode, CBC is NOT symmetric — encryption and decryption
 /// are different operations. This implementation handles CBC chaining
 /// correctly across multiple blocks.
+///
+/// Key and IV are zeroized on drop.
 pub struct AesCbc {
     key: [u8; 32],
     iv: [u8; 16],
 }
 
+impl Drop for AesCbc {
+    fn drop(&mut self) {
+        self.key.zeroize();
+        self.iv.zeroize();
+    }
+}
+
 impl AesCbc {
     /// AES block size
     const BLOCK_SIZE: usize = 16;
@@ -123,7 +149,7 @@ impl AesCbc {
     ///
     /// CBC Encryption: C[i] = AES_Encrypt(P[i] XOR C[i-1]), where C[-1] = IV
     pub fn encrypt(&self, data: &[u8]) -> Result<Vec<u8>> {
-        if data.len() % Self::BLOCK_SIZE != 0 {
+        if !data.len().is_multiple_of(Self::BLOCK_SIZE) {
             return Err(ProxyError::Crypto(
                 format!("CBC data must be aligned to 16 bytes, got {}", data.len())
             ));
@@ -141,17 +167,9 @@ impl AesCbc {
         
         for chunk in data.chunks(Self::BLOCK_SIZE) {
             let plaintext: [u8; 16] = chunk.try_into().unwrap();
-            
-            // XOR plaintext with previous ciphertext (or IV for first block)
             let xored = Self::xor_blocks(&plaintext, &prev_ciphertext);
-            
-            // Encrypt the XORed block
             let ciphertext = self.encrypt_block(&xored, &key_schedule);
-            
-            // Save for next iteration
             prev_ciphertext = ciphertext;
-            
-            // Append to result
             result.extend_from_slice(&ciphertext);
         }
         
@@ -162,7 +180,7 @@ impl AesCbc {
     ///
     /// CBC Decryption: P[i] = AES_Decrypt(C[i]) XOR C[i-1], where C[-1] = IV
     pub fn decrypt(&self, data: &[u8]) -> Result<Vec<u8>> {
-        if data.len() % Self::BLOCK_SIZE != 0 {
+        if !data.len().is_multiple_of(Self::BLOCK_SIZE) {
             return Err(ProxyError::Crypto(
                 format!("CBC data must be aligned to 16 bytes, got {}", data.len())
             ));
@@ -180,17 +198,9 @@ impl AesCbc {
         
         for chunk in data.chunks(Self::BLOCK_SIZE) {
             let ciphertext: [u8; 16] = chunk.try_into().unwrap();
-            
-            // Decrypt the block
             let decrypted = self.decrypt_block(&ciphertext, &key_schedule);
-            
-            // XOR with previous ciphertext (or IV for first block)
             let plaintext = Self::xor_blocks(&decrypted, &prev_ciphertext);
-            
-            // Save current ciphertext for next iteration
             prev_ciphertext = ciphertext;
-            
-            // Append to result
             result.extend_from_slice(&plaintext);
         }
         
@@ -199,7 +209,7 @@ impl AesCbc {
     
     /// Encrypt data in-place
     pub fn encrypt_in_place(&self, data: &mut [u8]) -> Result<()> {
-        if data.len() % Self::BLOCK_SIZE != 0 {
+        if !data.len().is_multiple_of(Self::BLOCK_SIZE) {
             return Err(ProxyError::Crypto(
                 format!("CBC data must be aligned to 16 bytes, got {}", data.len())
             ));
@@ -217,16 +227,13 @@ impl AesCbc {
         for i in (0..data.len()).step_by(Self::BLOCK_SIZE) {
             let block = &mut data[i..i + Self::BLOCK_SIZE];
             
-            // XOR with previous ciphertext
             for j in 0..Self::BLOCK_SIZE {
                 block[j] ^= prev_ciphertext[j];
             }
             
-            // Encrypt in-place
             let block_array: &mut [u8; 16] = block.try_into().unwrap();
             *block_array = self.encrypt_block(block_array, &key_schedule);
             
-            // Save for next iteration
             prev_ciphertext = *block_array;
         }
         
@@ -235,7 +242,7 @@ impl AesCbc {
     
     /// Decrypt data in-place
     pub fn decrypt_in_place(&self, data: &mut [u8]) -> Result<()> {
-        if data.len() % Self::BLOCK_SIZE != 0 {
+        if !data.len().is_multiple_of(Self::BLOCK_SIZE) {
             return Err(ProxyError::Crypto(
                 format!("CBC data must be aligned to 16 bytes, got {}", data.len())
             ));
@@ -248,26 +255,20 @@ impl AesCbc {
         use aes::cipher::KeyInit;
         let key_schedule = aes::Aes256::new((&self.key).into());
         
-        // For in-place decryption, we need to save ciphertext blocks
-        // before we overwrite them
         let mut prev_ciphertext = self.iv;
         
         for i in (0..data.len()).step_by(Self::BLOCK_SIZE) {
             let block = &mut data[i..i + Self::BLOCK_SIZE];
             
-            // Save current ciphertext before modifying
             let current_ciphertext: [u8; 16] = block.try_into().unwrap();
             
-            // Decrypt in-place
             let block_array: &mut [u8; 16] = block.try_into().unwrap();
             *block_array = self.decrypt_block(block_array, &key_schedule);
             
-            // XOR with previous ciphertext
             for j in 0..Self::BLOCK_SIZE {
                 block[j] ^= prev_ciphertext[j];
             }
             
-            // Save for next iteration
             prev_ciphertext = current_ciphertext;
         }
         
@@ -347,10 +348,8 @@ mod tests {
         let mut cipher = AesCtr::new(&key, iv);
         cipher.apply(&mut data);
         
-        // Encrypted should be different
         assert_ne!(&data[..], original);
         
-        // Decrypt with fresh cipher
         let mut cipher = AesCtr::new(&key, iv);
         cipher.apply(&mut data);
         
@@ -364,7 +363,7 @@ mod tests {
         let key = [0u8; 32];
         let iv = [0u8; 16];
         
-        let original = [0u8; 32]; // 2 blocks
+        let original = [0u8; 32];
         
         let cipher = AesCbc::new(key, iv);
         let encrypted = cipher.encrypt(&original).unwrap();
@@ -375,31 +374,25 @@ mod tests {
     
     #[test]
     fn test_aes_cbc_chaining_works() {
-        // This is the key test - verify CBC chaining is correct
         let key = [0x42u8; 32];
         let iv = [0x00u8; 16];
         
-        // Two IDENTICAL plaintext blocks
         let plaintext = [0xAAu8; 32];
         
         let cipher = AesCbc::new(key, iv);
         let ciphertext = cipher.encrypt(&plaintext).unwrap();
         
-        // With proper CBC, identical plaintext blocks produce DIFFERENT ciphertext
         let block1 = &ciphertext[0..16];
         let block2 = &ciphertext[16..32];
         
         assert_ne!(
             block1, block2,
-            "CBC chaining broken: identical plaintext blocks produced identical ciphertext. \
-             This indicates ECB mode, not CBC!"
+            "CBC chaining broken: identical plaintext blocks produced identical ciphertext"
         );
     }
     
     #[test]
     fn test_aes_cbc_known_vector() {
-        // Test with known NIST test vector
-        // AES-256-CBC with zero key and zero IV
         let key = [0u8; 32];
         let iv = [0u8; 16];
         let plaintext = [0u8; 16];
@@ -407,11 +400,9 @@ mod tests {
         let cipher = AesCbc::new(key, iv);
         let ciphertext = cipher.encrypt(&plaintext).unwrap();
         
-        // Decrypt and verify roundtrip
         let decrypted = cipher.decrypt(&ciphertext).unwrap();
         assert_eq!(plaintext.as_slice(), decrypted.as_slice());
         
-        // Ciphertext should not be all zeros
         assert_ne!(ciphertext.as_slice(), plaintext.as_slice());
     }
     
@@ -420,7 +411,6 @@ mod tests {
         let key = [0x12u8; 32];
         let iv = [0x34u8; 16];
         
-        // 5 blocks = 80 bytes
         let plaintext: Vec<u8> = (0..80).collect();
         
         let cipher = AesCbc::new(key, iv);
@@ -435,7 +425,7 @@ mod tests {
         let key = [0x12u8; 32];
         let iv = [0x34u8; 16];
         
-        let original = [0x56u8; 48]; // 3 blocks
+        let original = [0x56u8; 48];
         let mut buffer = original;
         
         let cipher = AesCbc::new(key, iv);
@@ -462,41 +452,33 @@ mod tests {
     fn test_aes_cbc_unaligned_error() {
         let cipher = AesCbc::new([0u8; 32], [0u8; 16]);
         
-        // 15 bytes - not aligned to block size
         let result = cipher.encrypt(&[0u8; 15]);
         assert!(result.is_err());
         
-        // 17 bytes - not aligned
         let result = cipher.encrypt(&[0u8; 17]);
         assert!(result.is_err());
     }
     
     #[test]
     fn test_aes_cbc_avalanche_effect() {
-        // Changing one bit in plaintext should change entire ciphertext block
-        // and all subsequent blocks (due to chaining)
         let key = [0xAB; 32];
         let iv = [0xCD; 16];
         
-        let mut plaintext1 = [0u8; 32];
+        let plaintext1 = [0u8; 32];
         let mut plaintext2 = [0u8; 32];
-        plaintext2[0] = 0x01; // Single bit difference in first block
+        plaintext2[0] = 0x01;
         
         let cipher = AesCbc::new(key, iv);
         
         let ciphertext1 = cipher.encrypt(&plaintext1).unwrap();
         let ciphertext2 = cipher.encrypt(&plaintext2).unwrap();
         
-        // First blocks should be different
         assert_ne!(&ciphertext1[0..16], &ciphertext2[0..16]);
-        
-        // Second blocks should ALSO be different (chaining effect)
         assert_ne!(&ciphertext1[16..32], &ciphertext2[16..32]);
     }
     
     #[test]
     fn test_aes_cbc_iv_matters() {
-        // Same plaintext with different IVs should produce different ciphertext
         let key = [0x55; 32];
         let plaintext = [0x77u8; 16];
         
@@ -511,7 +493,6 @@ mod tests {
     
     #[test]
     fn test_aes_cbc_deterministic() {
-        // Same key, IV, plaintext should always produce same ciphertext
         let key = [0x99; 32];
         let iv = [0x88; 16];
         let plaintext = [0x77u8; 32];
@@ -524,6 +505,23 @@ mod tests {
         assert_eq!(ciphertext1, ciphertext2);
     }
     
+    // ============= Zeroize Tests =============
+    
+    #[test]
+    fn test_aes_cbc_zeroize_on_drop() {
+        let key = [0xAA; 32];
+        let iv = [0xBB; 16];
+        
+        let cipher = AesCbc::new(key, iv);
+        // Verify key/iv are set
+        assert_eq!(cipher.key, [0xAA; 32]);
+        assert_eq!(cipher.iv, [0xBB; 16]);
+        
+        drop(cipher);
+        // After drop, key/iv are zeroized (can't observe directly,
+        // but the Drop impl runs without panic)
+    }
+    
     // ============= Error Handling Tests =============
     
     #[test]

src/crypto/hash.rs

@@ -1,3 +1,16 @@
+//! Cryptographic hash functions
+//!
+//! ## Protocol-required algorithms
+//!
+//! This module exposes MD5 and SHA-1 alongside SHA-256. These weaker
+//! hash functions are **required by the Telegram Middle Proxy protocol**
+//! (`derive_middleproxy_keys`) and cannot be replaced without breaking
+//! compatibility. They are NOT used for any security-sensitive purpose
+//! outside of that specific key derivation scheme mandated by Telegram.
+//!
+//! Static analysis tools (CodeQL, cargo-audit) may flag them — the
+//! usages are intentional and protocol-mandated.
+
 use hmac::{Hmac, Mac};
 use sha2::Sha256;
 use md5::Md5;
@@ -21,14 +34,16 @@ pub fn sha256_hmac(key: &[u8], data: &[u8]) -> [u8; 32] {
     mac.finalize().into_bytes().into()
 }
 
-/// SHA-1
+/// SHA-1 — **protocol-required** by Telegram Middle Proxy key derivation.
+/// Not used for general-purpose hashing.
 pub fn sha1(data: &[u8]) -> [u8; 20] {
     let mut hasher = Sha1::new();
     hasher.update(data);
     hasher.finalize().into()
 }
 
-/// MD5
+/// MD5 — **protocol-required** by Telegram Middle Proxy key derivation.
+/// Not used for general-purpose hashing.
 pub fn md5(data: &[u8]) -> [u8; 16] {
     let mut hasher = Md5::new();
     hasher.update(data);
@@ -40,7 +55,61 @@ pub fn crc32(data: &[u8]) -> u32 {
     crc32fast::hash(data)
 }
 
-/// Middle Proxy Keygen
+/// CRC32C (Castagnoli)
+pub fn crc32c(data: &[u8]) -> u32 {
+    crc32c::crc32c(data)
+}
+
+/// Build the exact prekey buffer used by Telegram Middle Proxy KDF.
+///
+/// Returned buffer layout (IPv4):
+/// nonce_srv | nonce_clt | clt_ts | srv_ip | clt_port | purpose | clt_ip | srv_port | secret | nonce_srv | [clt_v6 | srv_v6] | nonce_clt
+#[allow(clippy::too_many_arguments)]
+pub fn build_middleproxy_prekey(
+    nonce_srv: &[u8; 16],
+    nonce_clt: &[u8; 16],
+    clt_ts: &[u8; 4],
+    srv_ip: Option<&[u8]>,
+    clt_port: &[u8; 2],
+    purpose: &[u8],
+    clt_ip: Option<&[u8]>,
+    srv_port: &[u8; 2],
+    secret: &[u8],
+    clt_ipv6: Option<&[u8; 16]>,
+    srv_ipv6: Option<&[u8; 16]>,
+) -> Vec<u8> {
+    const EMPTY_IP: [u8; 4] = [0, 0, 0, 0];
+
+    let srv_ip = srv_ip.unwrap_or(&EMPTY_IP);
+    let clt_ip = clt_ip.unwrap_or(&EMPTY_IP);
+
+    let mut s = Vec::with_capacity(256);
+    s.extend_from_slice(nonce_srv);
+    s.extend_from_slice(nonce_clt);
+    s.extend_from_slice(clt_ts);
+    s.extend_from_slice(srv_ip);
+    s.extend_from_slice(clt_port);
+    s.extend_from_slice(purpose);
+    s.extend_from_slice(clt_ip);
+    s.extend_from_slice(srv_port);
+    s.extend_from_slice(secret);
+    s.extend_from_slice(nonce_srv);
+
+    if let (Some(clt_v6), Some(srv_v6)) = (clt_ipv6, srv_ipv6) {
+        s.extend_from_slice(clt_v6);
+        s.extend_from_slice(srv_v6);
+    }
+
+    s.extend_from_slice(nonce_clt);
+    s
+}
+
+/// Middle Proxy key derivation
+///
+/// Uses MD5 + SHA-1 as mandated by the Telegram Middle Proxy protocol.
+/// These algorithms are NOT replaceable here — changing them would break
+/// interoperability with Telegram's middle proxy infrastructure.
+#[allow(clippy::too_many_arguments)]
 pub fn derive_middleproxy_keys(
     nonce_srv: &[u8; 16],
     nonce_clt: &[u8; 16],
@@ -54,30 +123,20 @@ pub fn derive_middleproxy_keys(
     clt_ipv6: Option<&[u8; 16]>,
     srv_ipv6: Option<&[u8; 16]>,
 ) -> ([u8; 32], [u8; 16]) {
-    const EMPTY_IP: [u8; 4] = [0, 0, 0, 0];
-    
-    let srv_ip = srv_ip.unwrap_or(&EMPTY_IP);
-    let clt_ip = clt_ip.unwrap_or(&EMPTY_IP);
-    
-    let mut s = Vec::with_capacity(256);
-    s.extend_from_slice(nonce_srv);
-    s.extend_from_slice(nonce_clt);
-    s.extend_from_slice(clt_ts);
-    s.extend_from_slice(srv_ip);
-    s.extend_from_slice(clt_port);
-    s.extend_from_slice(purpose);
-    s.extend_from_slice(clt_ip);
-    s.extend_from_slice(srv_port);
-    s.extend_from_slice(secret);
-    s.extend_from_slice(nonce_srv);
-    
-    if let (Some(clt_v6), Some(srv_v6)) = (clt_ipv6, srv_ipv6) {
-        s.extend_from_slice(clt_v6);
-        s.extend_from_slice(srv_v6);
-    }
-    
-    s.extend_from_slice(nonce_clt);
-    
+    let s = build_middleproxy_prekey(
+        nonce_srv,
+        nonce_clt,
+        clt_ts,
+        srv_ip,
+        clt_port,
+        purpose,
+        clt_ip,
+        srv_port,
+        secret,
+        clt_ipv6,
+        srv_ipv6,
+    );
+
     let md5_1 = md5(&s[1..]);
     let sha1_sum = sha1(&s);
     let md5_2 = md5(&s[2..]);
@@ -87,4 +146,40 @@ pub fn derive_middleproxy_keys(
     key[12..].copy_from_slice(&sha1_sum);
     
     (key, md5_2)
-}
+}
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+
+    #[test]
+    fn middleproxy_prekey_sha_is_stable() {
+        let nonce_srv = [0x11u8; 16];
+        let nonce_clt = [0x22u8; 16];
+        let clt_ts = 0x44332211u32.to_le_bytes();
+        let srv_ip = Some([149u8, 154, 175, 50].as_ref());
+        let clt_ip = Some([10u8, 0, 0, 1].as_ref());
+        let clt_port = 0x1f90u16.to_le_bytes(); // 8080
+        let srv_port = 0x22b8u16.to_le_bytes(); // 8888
+        let secret = vec![0x55u8; 128];
+
+        let prekey = build_middleproxy_prekey(
+            &nonce_srv,
+            &nonce_clt,
+            &clt_ts,
+            srv_ip,
+            &clt_port,
+            b"CLIENT",
+            clt_ip,
+            &srv_port,
+            &secret,
+            None,
+            None,
+        );
+        let digest = sha256(&prekey);
+        assert_eq!(
+            hex::encode(digest),
+            "934f5facdafd65a44d5c2df90d2f35ddc81faaaeb337949dfeef817c8a7c1e00"
+        );
+    }
+}

src/crypto/mod.rs

@@ -5,5 +5,7 @@ pub mod hash;
 pub mod random;
 
 pub use aes::{AesCtr, AesCbc};
-pub use hash::{sha256, sha256_hmac, sha1, md5, crc32};
-pub use random::{SecureRandom, SECURE_RANDOM};
+pub use hash::{
+    build_middleproxy_prekey, crc32, crc32c, derive_middleproxy_keys, sha256, sha256_hmac,
+};
+pub use random::SecureRandom;

src/crypto/random.rs

@@ -1,55 +1,83 @@
 //! Pseudorandom
 
+#![allow(deprecated)]
+#![allow(dead_code)]
+
 use rand::{Rng, RngCore, SeedableRng};
 use rand::rngs::StdRng;
 use parking_lot::Mutex;
+use zeroize::Zeroize;
 use crate::crypto::AesCtr;
-use once_cell::sync::Lazy;
-
-/// Global secure random instance
-pub static SECURE_RANDOM: Lazy<SecureRandom> = Lazy::new(SecureRandom::new);
 
 /// Cryptographically secure PRNG with AES-CTR
 pub struct SecureRandom {
     inner: Mutex<SecureRandomInner>,
 }
 
+unsafe impl Send for SecureRandom {}
+unsafe impl Sync for SecureRandom {}
+
 struct SecureRandomInner {
     rng: StdRng,
     cipher: AesCtr,
     buffer: Vec<u8>,
 }
 
+impl Drop for SecureRandomInner {
+    fn drop(&mut self) {
+        self.buffer.zeroize();
+    }
+}
+
 impl SecureRandom {
     pub fn new() -> Self {
-        let mut rng = StdRng::from_entropy();
+        let mut seed_source = rand::rng();
+        let mut rng = StdRng::from_rng(&mut seed_source);
         
         let mut key = [0u8; 32];
         rng.fill_bytes(&mut key);
-        let iv: u128 = rng.gen();
+        let iv: u128 = rng.random();
+        
+        let cipher = AesCtr::new(&key, iv);
+        
+        // Zeroize local key copy — cipher already consumed it
+        key.zeroize();
         
         Self {
             inner: Mutex::new(SecureRandomInner {
                 rng,
-                cipher: AesCtr::new(&key, iv),
+                cipher,
                 buffer: Vec::with_capacity(1024),
             }),
         }
     }
     
-    /// Generate random bytes
-    pub fn bytes(&self, len: usize) -> Vec<u8> {
+    /// Fill a caller-provided buffer with random bytes.
+    pub fn fill(&self, out: &mut [u8]) {
         let mut inner = self.inner.lock();
         const CHUNK_SIZE: usize = 512;
-        
-        while inner.buffer.len() < len {
-            let mut chunk = vec![0u8; CHUNK_SIZE];
-            inner.rng.fill_bytes(&mut chunk);
-            inner.cipher.apply(&mut chunk);
-            inner.buffer.extend_from_slice(&chunk);
+
+        let mut written = 0usize;
+        while written < out.len() {
+            if inner.buffer.is_empty() {
+                let mut chunk = vec![0u8; CHUNK_SIZE];
+                inner.rng.fill_bytes(&mut chunk);
+                inner.cipher.apply(&mut chunk);
+                inner.buffer.extend_from_slice(&chunk);
+            }
+
+            let take = (out.len() - written).min(inner.buffer.len());
+            out[written..written + take].copy_from_slice(&inner.buffer[..take]);
+            inner.buffer.drain(..take);
+            written += take;
         }
-        
-        inner.buffer.drain(..len).collect()
+    }
+
+    /// Generate random bytes
+    pub fn bytes(&self, len: usize) -> Vec<u8> {
+        let mut out = vec![0u8; len];
+        self.fill(&mut out);
+        out
     }
     
     /// Generate random number in range [0, max)
@@ -67,7 +95,7 @@ impl SecureRandom {
             return 0;
         }
         
-        let bytes_needed = (k + 7) / 8;
+        let bytes_needed = k.div_ceil(8);
         let bytes = self.bytes(bytes_needed.min(8));
         
         let mut result = 0u64;
@@ -78,7 +106,6 @@ impl SecureRandom {
             result |= (b as u64) << (i * 8);
         }
         
-        // Mask extra bits
         if k < 64 {
             result &= (1u64 << k) - 1;
         }
@@ -107,13 +134,13 @@ impl SecureRandom {
     /// Generate random u32
     pub fn u32(&self) -> u32 {
         let mut inner = self.inner.lock();
-        inner.rng.gen()
+        inner.rng.random()
     }
     
     /// Generate random u64
     pub fn u64(&self) -> u64 {
         let mut inner = self.inner.lock();
-        inner.rng.gen()
+        inner.rng.random()
     }
 }
 
@@ -162,12 +189,10 @@ mod tests {
     fn test_bits() {
         let rng = SecureRandom::new();
         
-        // Single bit should be 0 or 1
         for _ in 0..100 {
             assert!(rng.bits(1) <= 1);
         }
         
-        // 8 bits should be 0-255
         for _ in 0..100 {
             assert!(rng.bits(8) <= 255);
         }
@@ -185,10 +210,8 @@ mod tests {
             }
         }
         
-        // Should have seen all items
         assert_eq!(seen.len(), 5);
         
-        // Empty slice should return None
         let empty: Vec<i32> = vec![];
         assert!(rng.choose(&empty).is_none());
     }
@@ -201,12 +224,10 @@ mod tests {
         let mut shuffled = original.clone();
         rng.shuffle(&mut shuffled);
         
-        // Should contain same elements
         let mut sorted = shuffled.clone();
         sorted.sort();
         assert_eq!(sorted, original);
         
-        // Should be different order (with very high probability)
         assert_ne!(shuffled, original);
     }
-}
+}

src/error.rs

@@ -1,5 +1,7 @@
 //! Error Types
 
+#![allow(dead_code)]
+
 use std::fmt;
 use std::net::SocketAddr;
 use thiserror::Error;
@@ -89,7 +91,7 @@ impl From<StreamError> for std::io::Error {
                 std::io::Error::new(std::io::ErrorKind::UnexpectedEof, err)
             }
             StreamError::Poisoned { .. } => {
-                std::io::Error::new(std::io::ErrorKind::Other, err)
+                std::io::Error::other(err)
             }
             StreamError::BufferOverflow { .. } => {
                 std::io::Error::new(std::io::ErrorKind::OutOfMemory, err)
@@ -98,7 +100,7 @@ impl From<StreamError> for std::io::Error {
                 std::io::Error::new(std::io::ErrorKind::InvalidData, err)
             }
             StreamError::PartialRead { .. } | StreamError::PartialWrite { .. } => {
-                std::io::Error::new(std::io::ErrorKind::Other, err)
+                std::io::Error::other(err)
             }
         }
     }
@@ -118,16 +120,13 @@ pub trait Recoverable {
 impl Recoverable for StreamError {
     fn is_recoverable(&self) -> bool {
         match self {
-            // Partial operations can be retried
             Self::PartialRead { .. } | Self::PartialWrite { .. } => true,
-            // I/O errors depend on kind
             Self::Io(e) => matches!(
                 e.kind(),
                 std::io::ErrorKind::WouldBlock 
                 | std::io::ErrorKind::Interrupted
                 | std::io::ErrorKind::TimedOut
             ),
-            // These are not recoverable
             Self::Poisoned { .. } 
             | Self::BufferOverflow { .. }
             | Self::InvalidFrame { .. }
@@ -136,16 +135,7 @@ impl Recoverable for StreamError {
     }
     
     fn can_continue(&self) -> bool {
-        match self {
-            // Poisoned stream cannot be used
-            Self::Poisoned { .. } => false,
-            // EOF means stream is done
-            Self::UnexpectedEof => false,
-            // Buffer overflow is fatal
-            Self::BufferOverflow { .. } => false,
-            // Others might allow continuation
-            _ => true,
-        }
+        !matches!(self, Self::Poisoned { .. } | Self::UnexpectedEof | Self::BufferOverflow { .. })
     }
 }
 
@@ -297,16 +287,16 @@ pub type StreamResult<T> = std::result::Result<T, StreamError>;
 
 /// Result with optional bad client handling
 #[derive(Debug)]
-pub enum HandshakeResult<T> {
+pub enum HandshakeResult<T, R, W> {
     /// Handshake succeeded
     Success(T),
-    /// Client failed validation, needs masking
-    BadClient,
+    /// Client failed validation, needs masking. Returns ownership of streams.
+    BadClient { reader: R, writer: W },
     /// Error occurred
     Error(ProxyError),
 }
 
-impl<T> HandshakeResult<T> {
+impl<T, R, W> HandshakeResult<T, R, W> {
     /// Check if successful
     pub fn is_success(&self) -> bool {
         matches!(self, HandshakeResult::Success(_))
@@ -314,49 +304,32 @@ impl<T> HandshakeResult<T> {
     
     /// Check if bad client
     pub fn is_bad_client(&self) -> bool {
-        matches!(self, HandshakeResult::BadClient)
-    }
-    
-    /// Convert to Result, treating BadClient as error
-    pub fn into_result(self) -> Result<T> {
-        match self {
-            HandshakeResult::Success(v) => Ok(v),
-            HandshakeResult::BadClient => Err(ProxyError::InvalidHandshake("Bad client".into())),
-            HandshakeResult::Error(e) => Err(e),
-        }
+        matches!(self, HandshakeResult::BadClient { .. })
     }
     
     /// Map the success value
-    pub fn map<U, F: FnOnce(T) -> U>(self, f: F) -> HandshakeResult<U> {
+    pub fn map<U, F: FnOnce(T) -> U>(self, f: F) -> HandshakeResult<U, R, W> {
         match self {
             HandshakeResult::Success(v) => HandshakeResult::Success(f(v)),
-            HandshakeResult::BadClient => HandshakeResult::BadClient,
+            HandshakeResult::BadClient { reader, writer } => HandshakeResult::BadClient { reader, writer },
             HandshakeResult::Error(e) => HandshakeResult::Error(e),
         }
     }
-    
-    /// Convert success to Option
-    pub fn ok(self) -> Option<T> {
-        match self {
-            HandshakeResult::Success(v) => Some(v),
-            _ => None,
-        }
-    }
 }
 
-impl<T> From<ProxyError> for HandshakeResult<T> {
+impl<T, R, W> From<ProxyError> for HandshakeResult<T, R, W> {
     fn from(err: ProxyError) -> Self {
         HandshakeResult::Error(err)
     }
 }
 
-impl<T> From<std::io::Error> for HandshakeResult<T> {
+impl<T, R, W> From<std::io::Error> for HandshakeResult<T, R, W> {
     fn from(err: std::io::Error) -> Self {
         HandshakeResult::Error(ProxyError::Io(err))
     }
 }
 
-impl<T> From<StreamError> for HandshakeResult<T> {
+impl<T, R, W> From<StreamError> for HandshakeResult<T, R, W> {
     fn from(err: StreamError) -> Self {
         HandshakeResult::Error(ProxyError::Stream(err))
     }
@@ -400,18 +373,18 @@ mod tests {
     
     #[test]
     fn test_handshake_result() {
-        let success: HandshakeResult<i32> = HandshakeResult::Success(42);
+        let success: HandshakeResult<i32, (), ()> = HandshakeResult::Success(42);
         assert!(success.is_success());
         assert!(!success.is_bad_client());
         
-        let bad: HandshakeResult<i32> = HandshakeResult::BadClient;
+        let bad: HandshakeResult<i32, (), ()> = HandshakeResult::BadClient { reader: (), writer: () };
         assert!(!bad.is_success());
         assert!(bad.is_bad_client());
     }
     
     #[test]
     fn test_handshake_result_map() {
-        let success: HandshakeResult<i32> = HandshakeResult::Success(42);
+        let success: HandshakeResult<i32, (), ()> = HandshakeResult::Success(42);
         let mapped = success.map(|x| x * 2);
         
         match mapped {

src/ip_tracker.rs

@@ -0,0 +1,573 @@
+// IP address tracking and per-user unique IP limiting.
+
+#![allow(dead_code)]
+
+use std::collections::HashMap;
+use std::net::IpAddr;
+use std::sync::Arc;
+use std::time::{Duration, Instant};
+
+use tokio::sync::RwLock;
+
+use crate::config::UserMaxUniqueIpsMode;
+
+#[derive(Debug, Clone)]
+pub struct UserIpTracker {
+    active_ips: Arc<RwLock<HashMap<String, HashMap<IpAddr, usize>>>>,
+    recent_ips: Arc<RwLock<HashMap<String, HashMap<IpAddr, Instant>>>>,
+    max_ips: Arc<RwLock<HashMap<String, usize>>>,
+    limit_mode: Arc<RwLock<UserMaxUniqueIpsMode>>,
+    limit_window: Arc<RwLock<Duration>>,
+}
+
+impl UserIpTracker {
+    pub fn new() -> Self {
+        Self {
+            active_ips: Arc::new(RwLock::new(HashMap::new())),
+            recent_ips: Arc::new(RwLock::new(HashMap::new())),
+            max_ips: Arc::new(RwLock::new(HashMap::new())),
+            limit_mode: Arc::new(RwLock::new(UserMaxUniqueIpsMode::ActiveWindow)),
+            limit_window: Arc::new(RwLock::new(Duration::from_secs(30))),
+        }
+    }
+
+    pub async fn set_limit_policy(&self, mode: UserMaxUniqueIpsMode, window_secs: u64) {
+        {
+            let mut current_mode = self.limit_mode.write().await;
+            *current_mode = mode;
+        }
+        let mut current_window = self.limit_window.write().await;
+        *current_window = Duration::from_secs(window_secs.max(1));
+    }
+
+    pub async fn set_user_limit(&self, username: &str, max_ips: usize) {
+        let mut limits = self.max_ips.write().await;
+        limits.insert(username.to_string(), max_ips);
+    }
+
+    pub async fn remove_user_limit(&self, username: &str) {
+        let mut limits = self.max_ips.write().await;
+        limits.remove(username);
+    }
+
+    pub async fn load_limits(&self, limits: &HashMap<String, usize>) {
+        let mut max_ips = self.max_ips.write().await;
+        max_ips.clone_from(limits);
+    }
+
+    fn prune_recent(user_recent: &mut HashMap<IpAddr, Instant>, now: Instant, window: Duration) {
+        if user_recent.is_empty() {
+            return;
+        }
+        user_recent.retain(|_, seen_at| now.duration_since(*seen_at) <= window);
+    }
+
+    pub async fn check_and_add(&self, username: &str, ip: IpAddr) -> Result<(), String> {
+        let limit = {
+            let max_ips = self.max_ips.read().await;
+            max_ips.get(username).copied()
+        };
+        let mode = *self.limit_mode.read().await;
+        let window = *self.limit_window.read().await;
+        let now = Instant::now();
+
+        let mut active_ips = self.active_ips.write().await;
+        let user_active = active_ips
+            .entry(username.to_string())
+            .or_insert_with(HashMap::new);
+
+        let mut recent_ips = self.recent_ips.write().await;
+        let user_recent = recent_ips
+            .entry(username.to_string())
+            .or_insert_with(HashMap::new);
+        Self::prune_recent(user_recent, now, window);
+
+        if let Some(count) = user_active.get_mut(&ip) {
+            *count = count.saturating_add(1);
+            user_recent.insert(ip, now);
+            return Ok(());
+        }
+
+        if let Some(limit) = limit {
+            let active_limit_reached = user_active.len() >= limit;
+            let recent_limit_reached = user_recent.len() >= limit;
+            let deny = match mode {
+                UserMaxUniqueIpsMode::ActiveWindow => active_limit_reached,
+                UserMaxUniqueIpsMode::TimeWindow => recent_limit_reached,
+                UserMaxUniqueIpsMode::Combined => active_limit_reached || recent_limit_reached,
+            };
+
+            if deny {
+                return Err(format!(
+                    "IP limit reached for user '{}': active={}/{} recent={}/{} mode={:?}",
+                    username,
+                    user_active.len(),
+                    limit,
+                    user_recent.len(),
+                    limit,
+                    mode
+                ));
+            }
+        }
+
+        user_active.insert(ip, 1);
+        user_recent.insert(ip, now);
+        Ok(())
+    }
+
+    pub async fn remove_ip(&self, username: &str, ip: IpAddr) {
+        let mut active_ips = self.active_ips.write().await;
+        if let Some(user_ips) = active_ips.get_mut(username) {
+            if let Some(count) = user_ips.get_mut(&ip) {
+                if *count > 1 {
+                    *count -= 1;
+                } else {
+                    user_ips.remove(&ip);
+                }
+            }
+            if user_ips.is_empty() {
+                active_ips.remove(username);
+            }
+        }
+    }
+
+    pub async fn get_recent_counts_for_users(&self, users: &[String]) -> HashMap<String, usize> {
+        let window = *self.limit_window.read().await;
+        let now = Instant::now();
+        let recent_ips = self.recent_ips.read().await;
+
+        let mut counts = HashMap::with_capacity(users.len());
+        for user in users {
+            let count = if let Some(user_recent) = recent_ips.get(user) {
+                user_recent
+                    .values()
+                    .filter(|seen_at| now.duration_since(**seen_at) <= window)
+                    .count()
+            } else {
+                0
+            };
+            counts.insert(user.clone(), count);
+        }
+        counts
+    }
+
+    pub async fn get_active_ips_for_users(&self, users: &[String]) -> HashMap<String, Vec<IpAddr>> {
+        let active_ips = self.active_ips.read().await;
+        let mut out = HashMap::with_capacity(users.len());
+        for user in users {
+            let mut ips = active_ips
+                .get(user)
+                .map(|per_ip| per_ip.keys().copied().collect::<Vec<_>>())
+                .unwrap_or_else(Vec::new);
+            ips.sort();
+            out.insert(user.clone(), ips);
+        }
+        out
+    }
+
+    pub async fn get_recent_ips_for_users(&self, users: &[String]) -> HashMap<String, Vec<IpAddr>> {
+        let window = *self.limit_window.read().await;
+        let now = Instant::now();
+        let recent_ips = self.recent_ips.read().await;
+
+        let mut out = HashMap::with_capacity(users.len());
+        for user in users {
+            let mut ips = if let Some(user_recent) = recent_ips.get(user) {
+                user_recent
+                    .iter()
+                    .filter(|(_, seen_at)| now.duration_since(**seen_at) <= window)
+                    .map(|(ip, _)| *ip)
+                    .collect::<Vec<_>>()
+            } else {
+                Vec::new()
+            };
+            ips.sort();
+            out.insert(user.clone(), ips);
+        }
+        out
+    }
+
+    pub async fn get_active_ip_count(&self, username: &str) -> usize {
+        let active_ips = self.active_ips.read().await;
+        active_ips.get(username).map(|ips| ips.len()).unwrap_or(0)
+    }
+
+    pub async fn get_active_ips(&self, username: &str) -> Vec<IpAddr> {
+        let active_ips = self.active_ips.read().await;
+        active_ips
+            .get(username)
+            .map(|ips| ips.keys().copied().collect())
+            .unwrap_or_else(Vec::new)
+    }
+
+    pub async fn get_stats(&self) -> Vec<(String, usize, usize)> {
+        let active_ips = self.active_ips.read().await;
+        let max_ips = self.max_ips.read().await;
+
+        let mut stats = Vec::new();
+        for (username, user_ips) in active_ips.iter() {
+            let limit = max_ips.get(username).copied().unwrap_or(0);
+            stats.push((username.clone(), user_ips.len(), limit));
+        }
+
+        stats.sort_by(|a, b| a.0.cmp(&b.0));
+        stats
+    }
+
+    pub async fn clear_user_ips(&self, username: &str) {
+        let mut active_ips = self.active_ips.write().await;
+        active_ips.remove(username);
+        drop(active_ips);
+
+        let mut recent_ips = self.recent_ips.write().await;
+        recent_ips.remove(username);
+    }
+
+    pub async fn clear_all(&self) {
+        let mut active_ips = self.active_ips.write().await;
+        active_ips.clear();
+        drop(active_ips);
+
+        let mut recent_ips = self.recent_ips.write().await;
+        recent_ips.clear();
+    }
+
+    pub async fn is_ip_active(&self, username: &str, ip: IpAddr) -> bool {
+        let active_ips = self.active_ips.read().await;
+        active_ips
+            .get(username)
+            .map(|ips| ips.contains_key(&ip))
+            .unwrap_or(false)
+    }
+
+    pub async fn get_user_limit(&self, username: &str) -> Option<usize> {
+        let max_ips = self.max_ips.read().await;
+        max_ips.get(username).copied()
+    }
+
+    pub async fn format_stats(&self) -> String {
+        let stats = self.get_stats().await;
+
+        if stats.is_empty() {
+            return String::from("No active users");
+        }
+
+        let mut output = String::from("User IP Statistics:\n");
+        output.push_str("==================\n");
+
+        for (username, active_count, limit) in stats {
+            output.push_str(&format!(
+                "User: {:<20} Active IPs: {}/{}\n",
+                username,
+                active_count,
+                if limit > 0 {
+                    limit.to_string()
+                } else {
+                    "unlimited".to_string()
+                }
+            ));
+
+            let ips = self.get_active_ips(&username).await;
+            for ip in ips {
+                output.push_str(&format!("  - {}\n", ip));
+            }
+        }
+
+        output
+    }
+}
+
+impl Default for UserIpTracker {
+    fn default() -> Self {
+        Self::new()
+    }
+}
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+    use std::net::{IpAddr, Ipv4Addr, Ipv6Addr};
+
+    fn test_ipv4(oct1: u8, oct2: u8, oct3: u8, oct4: u8) -> IpAddr {
+        IpAddr::V4(Ipv4Addr::new(oct1, oct2, oct3, oct4))
+    }
+
+    fn test_ipv6() -> IpAddr {
+        IpAddr::V6(Ipv6Addr::new(0x2001, 0xdb8, 0, 0, 0, 0, 0, 1))
+    }
+
+    #[tokio::test]
+    async fn test_basic_ip_limit() {
+        let tracker = UserIpTracker::new();
+        tracker.set_user_limit("test_user", 2).await;
+
+        let ip1 = test_ipv4(192, 168, 1, 1);
+        let ip2 = test_ipv4(192, 168, 1, 2);
+        let ip3 = test_ipv4(192, 168, 1, 3);
+
+        assert!(tracker.check_and_add("test_user", ip1).await.is_ok());
+        assert!(tracker.check_and_add("test_user", ip2).await.is_ok());
+        assert!(tracker.check_and_add("test_user", ip3).await.is_err());
+
+        assert_eq!(tracker.get_active_ip_count("test_user").await, 2);
+    }
+
+    #[tokio::test]
+    async fn test_active_window_rejects_new_ip_and_keeps_existing_session() {
+        let tracker = UserIpTracker::new();
+        tracker.set_user_limit("test_user", 1).await;
+        tracker
+            .set_limit_policy(UserMaxUniqueIpsMode::ActiveWindow, 30)
+            .await;
+
+        let ip1 = test_ipv4(10, 10, 10, 1);
+        let ip2 = test_ipv4(10, 10, 10, 2);
+
+        assert!(tracker.check_and_add("test_user", ip1).await.is_ok());
+        assert!(tracker.is_ip_active("test_user", ip1).await);
+        assert!(tracker.check_and_add("test_user", ip2).await.is_err());
+
+        // Existing session remains active; only new unique IP is denied.
+        assert!(tracker.is_ip_active("test_user", ip1).await);
+        assert_eq!(tracker.get_active_ip_count("test_user").await, 1);
+    }
+
+    #[tokio::test]
+    async fn test_reconnection_from_same_ip() {
+        let tracker = UserIpTracker::new();
+        tracker.set_user_limit("test_user", 2).await;
+
+        let ip1 = test_ipv4(192, 168, 1, 1);
+
+        assert!(tracker.check_and_add("test_user", ip1).await.is_ok());
+        assert!(tracker.check_and_add("test_user", ip1).await.is_ok());
+        assert_eq!(tracker.get_active_ip_count("test_user").await, 1);
+    }
+
+    #[tokio::test]
+    async fn test_same_ip_disconnect_keeps_active_while_other_session_alive() {
+        let tracker = UserIpTracker::new();
+        tracker.set_user_limit("test_user", 2).await;
+
+        let ip1 = test_ipv4(192, 168, 1, 1);
+
+        assert!(tracker.check_and_add("test_user", ip1).await.is_ok());
+        assert!(tracker.check_and_add("test_user", ip1).await.is_ok());
+        assert_eq!(tracker.get_active_ip_count("test_user").await, 1);
+
+        tracker.remove_ip("test_user", ip1).await;
+        assert_eq!(tracker.get_active_ip_count("test_user").await, 1);
+
+        tracker.remove_ip("test_user", ip1).await;
+        assert_eq!(tracker.get_active_ip_count("test_user").await, 0);
+    }
+
+    #[tokio::test]
+    async fn test_ip_removal() {
+        let tracker = UserIpTracker::new();
+        tracker.set_user_limit("test_user", 2).await;
+
+        let ip1 = test_ipv4(192, 168, 1, 1);
+        let ip2 = test_ipv4(192, 168, 1, 2);
+        let ip3 = test_ipv4(192, 168, 1, 3);
+
+        assert!(tracker.check_and_add("test_user", ip1).await.is_ok());
+        assert!(tracker.check_and_add("test_user", ip2).await.is_ok());
+        assert!(tracker.check_and_add("test_user", ip3).await.is_err());
+
+        tracker.remove_ip("test_user", ip1).await;
+
+        assert!(tracker.check_and_add("test_user", ip3).await.is_ok());
+        assert_eq!(tracker.get_active_ip_count("test_user").await, 2);
+    }
+
+    #[tokio::test]
+    async fn test_no_limit() {
+        let tracker = UserIpTracker::new();
+
+        let ip1 = test_ipv4(192, 168, 1, 1);
+        let ip2 = test_ipv4(192, 168, 1, 2);
+        let ip3 = test_ipv4(192, 168, 1, 3);
+
+        assert!(tracker.check_and_add("test_user", ip1).await.is_ok());
+        assert!(tracker.check_and_add("test_user", ip2).await.is_ok());
+        assert!(tracker.check_and_add("test_user", ip3).await.is_ok());
+
+        assert_eq!(tracker.get_active_ip_count("test_user").await, 3);
+    }
+
+    #[tokio::test]
+    async fn test_multiple_users() {
+        let tracker = UserIpTracker::new();
+        tracker.set_user_limit("user1", 2).await;
+        tracker.set_user_limit("user2", 1).await;
+
+        let ip1 = test_ipv4(192, 168, 1, 1);
+        let ip2 = test_ipv4(192, 168, 1, 2);
+
+        assert!(tracker.check_and_add("user1", ip1).await.is_ok());
+        assert!(tracker.check_and_add("user1", ip2).await.is_ok());
+
+        assert!(tracker.check_and_add("user2", ip1).await.is_ok());
+        assert!(tracker.check_and_add("user2", ip2).await.is_err());
+    }
+
+    #[tokio::test]
+    async fn test_ipv6_support() {
+        let tracker = UserIpTracker::new();
+        tracker.set_user_limit("test_user", 2).await;
+
+        let ipv4 = test_ipv4(192, 168, 1, 1);
+        let ipv6 = test_ipv6();
+
+        assert!(tracker.check_and_add("test_user", ipv4).await.is_ok());
+        assert!(tracker.check_and_add("test_user", ipv6).await.is_ok());
+
+        assert_eq!(tracker.get_active_ip_count("test_user").await, 2);
+    }
+
+    #[tokio::test]
+    async fn test_get_active_ips() {
+        let tracker = UserIpTracker::new();
+        tracker.set_user_limit("test_user", 3).await;
+
+        let ip1 = test_ipv4(192, 168, 1, 1);
+        let ip2 = test_ipv4(192, 168, 1, 2);
+
+        tracker.check_and_add("test_user", ip1).await.unwrap();
+        tracker.check_and_add("test_user", ip2).await.unwrap();
+
+        let active_ips = tracker.get_active_ips("test_user").await;
+        assert_eq!(active_ips.len(), 2);
+        assert!(active_ips.contains(&ip1));
+        assert!(active_ips.contains(&ip2));
+    }
+
+    #[tokio::test]
+    async fn test_stats() {
+        let tracker = UserIpTracker::new();
+        tracker.set_user_limit("user1", 3).await;
+        tracker.set_user_limit("user2", 2).await;
+
+        let ip1 = test_ipv4(192, 168, 1, 1);
+        let ip2 = test_ipv4(192, 168, 1, 2);
+
+        tracker.check_and_add("user1", ip1).await.unwrap();
+        tracker.check_and_add("user2", ip2).await.unwrap();
+
+        let stats = tracker.get_stats().await;
+        assert_eq!(stats.len(), 2);
+
+        assert!(stats.iter().any(|(name, _, _)| name == "user1"));
+        assert!(stats.iter().any(|(name, _, _)| name == "user2"));
+    }
+
+    #[tokio::test]
+    async fn test_clear_user_ips() {
+        let tracker = UserIpTracker::new();
+        let ip1 = test_ipv4(192, 168, 1, 1);
+
+        tracker.check_and_add("test_user", ip1).await.unwrap();
+        assert_eq!(tracker.get_active_ip_count("test_user").await, 1);
+
+        tracker.clear_user_ips("test_user").await;
+        assert_eq!(tracker.get_active_ip_count("test_user").await, 0);
+    }
+
+    #[tokio::test]
+    async fn test_is_ip_active() {
+        let tracker = UserIpTracker::new();
+        let ip1 = test_ipv4(192, 168, 1, 1);
+        let ip2 = test_ipv4(192, 168, 1, 2);
+
+        tracker.check_and_add("test_user", ip1).await.unwrap();
+
+        assert!(tracker.is_ip_active("test_user", ip1).await);
+        assert!(!tracker.is_ip_active("test_user", ip2).await);
+    }
+
+    #[tokio::test]
+    async fn test_load_limits_from_config() {
+        let tracker = UserIpTracker::new();
+
+        let mut config_limits = HashMap::new();
+        config_limits.insert("user1".to_string(), 5);
+        config_limits.insert("user2".to_string(), 3);
+
+        tracker.load_limits(&config_limits).await;
+
+        assert_eq!(tracker.get_user_limit("user1").await, Some(5));
+        assert_eq!(tracker.get_user_limit("user2").await, Some(3));
+        assert_eq!(tracker.get_user_limit("user3").await, None);
+    }
+
+    #[tokio::test]
+    async fn test_load_limits_replaces_previous_map() {
+        let tracker = UserIpTracker::new();
+
+        let mut first = HashMap::new();
+        first.insert("user1".to_string(), 2);
+        first.insert("user2".to_string(), 3);
+        tracker.load_limits(&first).await;
+
+        let mut second = HashMap::new();
+        second.insert("user2".to_string(), 5);
+        tracker.load_limits(&second).await;
+
+        assert_eq!(tracker.get_user_limit("user1").await, None);
+        assert_eq!(tracker.get_user_limit("user2").await, Some(5));
+    }
+
+    #[tokio::test]
+    async fn test_time_window_mode_blocks_recent_ip_churn() {
+        let tracker = UserIpTracker::new();
+        tracker.set_user_limit("test_user", 1).await;
+        tracker
+            .set_limit_policy(UserMaxUniqueIpsMode::TimeWindow, 30)
+            .await;
+
+        let ip1 = test_ipv4(10, 0, 0, 1);
+        let ip2 = test_ipv4(10, 0, 0, 2);
+
+        assert!(tracker.check_and_add("test_user", ip1).await.is_ok());
+        tracker.remove_ip("test_user", ip1).await;
+        assert!(tracker.check_and_add("test_user", ip2).await.is_err());
+    }
+
+    #[tokio::test]
+    async fn test_combined_mode_enforces_active_and_recent_limits() {
+        let tracker = UserIpTracker::new();
+        tracker.set_user_limit("test_user", 1).await;
+        tracker
+            .set_limit_policy(UserMaxUniqueIpsMode::Combined, 30)
+            .await;
+
+        let ip1 = test_ipv4(10, 0, 1, 1);
+        let ip2 = test_ipv4(10, 0, 1, 2);
+
+        assert!(tracker.check_and_add("test_user", ip1).await.is_ok());
+        assert!(tracker.check_and_add("test_user", ip2).await.is_err());
+
+        tracker.remove_ip("test_user", ip1).await;
+        assert!(tracker.check_and_add("test_user", ip2).await.is_err());
+    }
+
+    #[tokio::test]
+    async fn test_time_window_expires() {
+        let tracker = UserIpTracker::new();
+        tracker.set_user_limit("test_user", 1).await;
+        tracker
+            .set_limit_policy(UserMaxUniqueIpsMode::TimeWindow, 1)
+            .await;
+
+        let ip1 = test_ipv4(10, 1, 0, 1);
+        let ip2 = test_ipv4(10, 1, 0, 2);
+
+        assert!(tracker.check_and_add("test_user", ip1).await.is_ok());
+        tracker.remove_ip("test_user", ip1).await;
+        assert!(tracker.check_and_add("test_user", ip2).await.is_err());
+
+        tokio::time::sleep(Duration::from_millis(1100)).await;
+        assert!(tracker.check_and_add("test_user", ip2).await.is_ok());
+    }
+}

src/network/dns_overrides.rs

@@ -0,0 +1,197 @@
+//! Runtime DNS overrides for `host:port` targets.
+
+use std::collections::HashMap;
+use std::net::{IpAddr, Ipv6Addr, SocketAddr};
+use std::sync::{OnceLock, RwLock};
+
+use crate::error::{ProxyError, Result};
+
+type OverrideMap = HashMap<(String, u16), IpAddr>;
+
+static DNS_OVERRIDES: OnceLock<RwLock<OverrideMap>> = OnceLock::new();
+
+fn overrides_store() -> &'static RwLock<OverrideMap> {
+    DNS_OVERRIDES.get_or_init(|| RwLock::new(HashMap::new()))
+}
+
+fn parse_ip_spec(ip_spec: &str) -> Result<IpAddr> {
+    if ip_spec.starts_with('[') && ip_spec.ends_with(']') {
+        let inner = &ip_spec[1..ip_spec.len() - 1];
+        let ipv6 = inner.parse::<Ipv6Addr>().map_err(|_| {
+            ProxyError::Config(format!(
+                "network.dns_overrides IPv6 override is invalid: '{ip_spec}'"
+            ))
+        })?;
+        return Ok(IpAddr::V6(ipv6));
+    }
+
+    let ip = ip_spec.parse::<IpAddr>().map_err(|_| {
+        ProxyError::Config(format!(
+            "network.dns_overrides IP is invalid: '{ip_spec}'"
+        ))
+    })?;
+    if matches!(ip, IpAddr::V6(_)) {
+        return Err(ProxyError::Config(format!(
+            "network.dns_overrides IPv6 must be bracketed: '{ip_spec}'"
+        )));
+    }
+    Ok(ip)
+}
+
+fn parse_entry(entry: &str) -> Result<((String, u16), IpAddr)> {
+    let trimmed = entry.trim();
+    if trimmed.is_empty() {
+        return Err(ProxyError::Config(
+            "network.dns_overrides entry cannot be empty".to_string(),
+        ));
+    }
+
+    let first_sep = trimmed.find(':').ok_or_else(|| {
+        ProxyError::Config(format!(
+            "network.dns_overrides entry must use host:port:ip format: '{trimmed}'"
+        ))
+    })?;
+    let second_sep = trimmed[first_sep + 1..]
+        .find(':')
+        .map(|idx| first_sep + 1 + idx)
+        .ok_or_else(|| {
+            ProxyError::Config(format!(
+                "network.dns_overrides entry must use host:port:ip format: '{trimmed}'"
+            ))
+        })?;
+
+    let host = trimmed[..first_sep].trim();
+    let port_str = trimmed[first_sep + 1..second_sep].trim();
+    let ip_str = trimmed[second_sep + 1..].trim();
+
+    if host.is_empty() {
+        return Err(ProxyError::Config(format!(
+            "network.dns_overrides host cannot be empty: '{trimmed}'"
+        )));
+    }
+    if host.contains(':') {
+        return Err(ProxyError::Config(format!(
+            "network.dns_overrides host must be a domain name without ':' in this format: '{trimmed}'"
+        )));
+    }
+
+    let port = port_str.parse::<u16>().map_err(|_| {
+        ProxyError::Config(format!(
+            "network.dns_overrides port is invalid: '{trimmed}'"
+        ))
+    })?;
+    let ip = parse_ip_spec(ip_str)?;
+
+    Ok(((host.to_ascii_lowercase(), port), ip))
+}
+
+fn parse_entries(entries: &[String]) -> Result<OverrideMap> {
+    let mut parsed = HashMap::new();
+    for entry in entries {
+        let (key, ip) = parse_entry(entry)?;
+        parsed.insert(key, ip);
+    }
+    Ok(parsed)
+}
+
+/// Validate `network.dns_overrides` entries without updating runtime state.
+pub fn validate_entries(entries: &[String]) -> Result<()> {
+    let _ = parse_entries(entries)?;
+    Ok(())
+}
+
+/// Replace runtime DNS overrides with a new validated snapshot.
+pub fn install_entries(entries: &[String]) -> Result<()> {
+    let parsed = parse_entries(entries)?;
+    let mut guard = overrides_store()
+        .write()
+        .map_err(|_| ProxyError::Config("network.dns_overrides runtime lock is poisoned".to_string()))?;
+    *guard = parsed;
+    Ok(())
+}
+
+/// Resolve a hostname override for `(host, port)` if present.
+pub fn resolve(host: &str, port: u16) -> Option<IpAddr> {
+    let key = (host.to_ascii_lowercase(), port);
+    overrides_store()
+        .read()
+        .ok()
+        .and_then(|guard| guard.get(&key).copied())
+}
+
+/// Resolve a hostname override and construct a socket address when present.
+pub fn resolve_socket_addr(host: &str, port: u16) -> Option<SocketAddr> {
+    resolve(host, port).map(|ip| SocketAddr::new(ip, port))
+}
+
+/// Parse a runtime endpoint in `host:port` format.
+///
+/// Supports:
+/// - `example.com:443`
+/// - `[2001:db8::1]:443`
+pub fn split_host_port(endpoint: &str) -> Option<(String, u16)> {
+    if endpoint.starts_with('[') {
+        let bracket_end = endpoint.find(']')?;
+        if endpoint.as_bytes().get(bracket_end + 1) != Some(&b':') {
+            return None;
+        }
+        let host = endpoint[1..bracket_end].trim();
+        let port = endpoint[bracket_end + 2..].trim().parse::<u16>().ok()?;
+        if host.is_empty() {
+            return None;
+        }
+        return Some((host.to_ascii_lowercase(), port));
+    }
+
+    let split_idx = endpoint.rfind(':')?;
+    let host = endpoint[..split_idx].trim();
+    let port = endpoint[split_idx + 1..].trim().parse::<u16>().ok()?;
+    if host.is_empty() || host.contains(':') {
+        return None;
+    }
+
+    Some((host.to_ascii_lowercase(), port))
+}
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+
+    #[test]
+    fn validate_accepts_ipv4_and_bracketed_ipv6() {
+        let entries = vec![
+            "example.com:443:127.0.0.1".to_string(),
+            "example.net:8443:[2001:db8::10]".to_string(),
+        ];
+        assert!(validate_entries(&entries).is_ok());
+    }
+
+    #[test]
+    fn validate_rejects_unbracketed_ipv6() {
+        let entries = vec!["example.net:443:2001:db8::10".to_string()];
+        let err = validate_entries(&entries).unwrap_err().to_string();
+        assert!(err.contains("must be bracketed"));
+    }
+
+    #[test]
+    fn install_and_resolve_are_case_insensitive_for_host() {
+        let entries = vec!["MyPetrovich.ru:8443:127.0.0.1".to_string()];
+        install_entries(&entries).unwrap();
+
+        let resolved = resolve("mypetrovich.ru", 8443);
+        assert_eq!(resolved, Some("127.0.0.1".parse().unwrap()));
+    }
+
+    #[test]
+    fn split_host_port_parses_supported_shapes() {
+        assert_eq!(
+            split_host_port("example.com:443"),
+            Some(("example.com".to_string(), 443))
+        );
+        assert_eq!(
+            split_host_port("[2001:db8::1]:443"),
+            Some(("2001:db8::1".to_string(), 443))
+        );
+        assert_eq!(split_host_port("2001:db8::1:443"), None);
+    }
+}

src/network/mod.rs

@@ -0,0 +1,5 @@
+pub mod dns_overrides;
+pub mod probe;
+pub mod stun;
+
+pub use stun::IpFamily;

src/network/probe.rs

@@ -0,0 +1,370 @@
+#![allow(dead_code)]
+
+use std::collections::HashMap;
+use std::net::{IpAddr, Ipv4Addr, Ipv6Addr, SocketAddr, UdpSocket};
+use std::time::Duration;
+
+use tokio::task::JoinSet;
+use tokio::time::timeout;
+use tracing::{debug, info, warn};
+
+use crate::config::NetworkConfig;
+use crate::error::Result;
+use crate::network::stun::{stun_probe_dual, DualStunResult, IpFamily, StunProbeResult};
+
+#[derive(Debug, Clone, Default)]
+pub struct NetworkProbe {
+    pub detected_ipv4: Option<Ipv4Addr>,
+    pub detected_ipv6: Option<Ipv6Addr>,
+    pub reflected_ipv4: Option<SocketAddr>,
+    pub reflected_ipv6: Option<SocketAddr>,
+    pub ipv4_is_bogon: bool,
+    pub ipv6_is_bogon: bool,
+    pub ipv4_nat_detected: bool,
+    pub ipv6_nat_detected: bool,
+    pub ipv4_usable: bool,
+    pub ipv6_usable: bool,
+}
+
+#[derive(Debug, Clone, Default)]
+pub struct NetworkDecision {
+    pub ipv4_dc: bool,
+    pub ipv6_dc: bool,
+    pub ipv4_me: bool,
+    pub ipv6_me: bool,
+    pub effective_prefer: u8,
+    pub effective_multipath: bool,
+}
+
+impl NetworkDecision {
+    pub fn prefer_ipv6(&self) -> bool {
+        self.effective_prefer == 6
+    }
+
+    pub fn me_families(&self) -> Vec<IpFamily> {
+        let mut res = Vec::new();
+        if self.ipv4_me {
+            res.push(IpFamily::V4);
+        }
+        if self.ipv6_me {
+            res.push(IpFamily::V6);
+        }
+        res
+    }
+}
+
+const STUN_BATCH_TIMEOUT: Duration = Duration::from_secs(5);
+
+pub async fn run_probe(
+    config: &NetworkConfig,
+    nat_probe: bool,
+    stun_nat_probe_concurrency: usize,
+) -> Result<NetworkProbe> {
+    let mut probe = NetworkProbe::default();
+
+    probe.detected_ipv4 = detect_local_ip_v4();
+    probe.detected_ipv6 = detect_local_ip_v6();
+
+    probe.ipv4_is_bogon = probe.detected_ipv4.map(is_bogon_v4).unwrap_or(false);
+    probe.ipv6_is_bogon = probe.detected_ipv6.map(is_bogon_v6).unwrap_or(false);
+
+    let stun_res = if nat_probe && config.stun_use {
+        let servers = collect_stun_servers(config);
+        if servers.is_empty() {
+            warn!("STUN probe is enabled but network.stun_servers is empty");
+            DualStunResult::default()
+        } else {
+            probe_stun_servers_parallel(
+                &servers,
+                stun_nat_probe_concurrency.max(1),
+            )
+            .await
+        }
+    } else if nat_probe {
+        info!("STUN probe is disabled by network.stun_use=false");
+        DualStunResult::default()
+    } else {
+        DualStunResult::default()
+    };
+    probe.reflected_ipv4 = stun_res.v4.map(|r| r.reflected_addr);
+    probe.reflected_ipv6 = stun_res.v6.map(|r| r.reflected_addr);
+
+    // If STUN is blocked but IPv4 is private, try HTTP public-IP fallback.
+    if nat_probe
+        && probe.reflected_ipv4.is_none()
+        && probe.detected_ipv4.map(is_bogon_v4).unwrap_or(false)
+    {
+        if let Some(public_ip) = detect_public_ipv4_http(&config.http_ip_detect_urls).await {
+            probe.reflected_ipv4 = Some(SocketAddr::new(IpAddr::V4(public_ip), 0));
+            info!(public_ip = %public_ip, "STUN unavailable, using HTTP public IPv4 fallback");
+        }
+    }
+
+    probe.ipv4_nat_detected = match (probe.detected_ipv4, probe.reflected_ipv4) {
+        (Some(det), Some(reflected)) => det != reflected.ip(),
+        _ => false,
+    };
+    probe.ipv6_nat_detected = match (probe.detected_ipv6, probe.reflected_ipv6) {
+        (Some(det), Some(reflected)) => det != reflected.ip(),
+        _ => false,
+    };
+
+    probe.ipv4_usable = config.ipv4
+        && probe.detected_ipv4.is_some()
+        && (!probe.ipv4_is_bogon || probe.reflected_ipv4.map(|r| !is_bogon(r.ip())).unwrap_or(false));
+
+    let ipv6_enabled = config.ipv6.unwrap_or(probe.detected_ipv6.is_some());
+    probe.ipv6_usable = ipv6_enabled
+        && probe.detected_ipv6.is_some()
+        && (!probe.ipv6_is_bogon || probe.reflected_ipv6.map(|r| !is_bogon(r.ip())).unwrap_or(false));
+
+    Ok(probe)
+}
+
+async fn detect_public_ipv4_http(urls: &[String]) -> Option<Ipv4Addr> {
+    let client = reqwest::Client::builder()
+        .timeout(Duration::from_secs(3))
+        .build()
+        .ok()?;
+
+    for url in urls {
+        let response = match client.get(url).send().await {
+            Ok(response) => response,
+            Err(_) => continue,
+        };
+
+        let body = match response.text().await {
+            Ok(body) => body,
+            Err(_) => continue,
+        };
+
+        let Ok(ip) = body.trim().parse::<Ipv4Addr>() else {
+            continue;
+        };
+        if !is_bogon_v4(ip) {
+            return Some(ip);
+        }
+    }
+
+    None
+}
+
+fn collect_stun_servers(config: &NetworkConfig) -> Vec<String> {
+    let mut out = Vec::new();
+    for s in &config.stun_servers {
+        if !s.is_empty() && !out.contains(s) {
+            out.push(s.clone());
+        }
+    }
+    out
+}
+
+async fn probe_stun_servers_parallel(
+    servers: &[String],
+    concurrency: usize,
+) -> DualStunResult {
+    let mut join_set = JoinSet::new();
+    let mut next_idx = 0usize;
+    let mut best_v4_by_ip: HashMap<IpAddr, (usize, StunProbeResult)> = HashMap::new();
+    let mut best_v6_by_ip: HashMap<IpAddr, (usize, StunProbeResult)> = HashMap::new();
+
+    while next_idx < servers.len() || !join_set.is_empty() {
+        while next_idx < servers.len() && join_set.len() < concurrency {
+            let stun_addr = servers[next_idx].clone();
+            next_idx += 1;
+            join_set.spawn(async move {
+                let res = timeout(STUN_BATCH_TIMEOUT, stun_probe_dual(&stun_addr)).await;
+                (stun_addr, res)
+            });
+        }
+
+        let Some(task) = join_set.join_next().await else {
+            break;
+        };
+
+        match task {
+            Ok((stun_addr, Ok(Ok(result)))) => {
+                if let Some(v4) = result.v4 {
+                    let entry = best_v4_by_ip.entry(v4.reflected_addr.ip()).or_insert((0, v4));
+                    entry.0 += 1;
+                }
+                if let Some(v6) = result.v6 {
+                    let entry = best_v6_by_ip.entry(v6.reflected_addr.ip()).or_insert((0, v6));
+                    entry.0 += 1;
+                }
+                if result.v4.is_some() || result.v6.is_some() {
+                    debug!(stun = %stun_addr, "STUN server responded within probe timeout");
+                }
+            }
+            Ok((stun_addr, Ok(Err(e)))) => {
+                debug!(error = %e, stun = %stun_addr, "STUN probe failed");
+            }
+            Ok((stun_addr, Err(_))) => {
+                debug!(stun = %stun_addr, "STUN probe timeout");
+            }
+            Err(e) => {
+                debug!(error = %e, "STUN probe task join failed");
+            }
+        }
+    }
+
+    let mut out = DualStunResult::default();
+    if let Some((_, best)) = best_v4_by_ip
+        .into_values()
+        .max_by_key(|(count, _)| *count)
+    {
+        info!("STUN-Quorum reached, IP: {}", best.reflected_addr.ip());
+        out.v4 = Some(best);
+    }
+    if let Some((_, best)) = best_v6_by_ip
+        .into_values()
+        .max_by_key(|(count, _)| *count)
+    {
+        info!("STUN-Quorum reached, IP: {}", best.reflected_addr.ip());
+        out.v6 = Some(best);
+    }
+    out
+}
+
+pub fn decide_network_capabilities(config: &NetworkConfig, probe: &NetworkProbe) -> NetworkDecision {
+    let ipv4_dc = config.ipv4 && probe.detected_ipv4.is_some();
+    let ipv6_dc = config.ipv6.unwrap_or(probe.detected_ipv6.is_some()) && probe.detected_ipv6.is_some();
+
+    let ipv4_me = config.ipv4
+        && probe.detected_ipv4.is_some()
+        && (!probe.ipv4_is_bogon || probe.reflected_ipv4.is_some());
+
+    let ipv6_enabled = config.ipv6.unwrap_or(probe.detected_ipv6.is_some());
+    let ipv6_me = ipv6_enabled
+        && probe.detected_ipv6.is_some()
+        && (!probe.ipv6_is_bogon || probe.reflected_ipv6.is_some());
+
+    let effective_prefer = match config.prefer {
+        6 if ipv6_me || ipv6_dc => 6,
+        4 if ipv4_me || ipv4_dc => 4,
+        6 => {
+            warn!("prefer=6 requested but IPv6 unavailable; falling back to IPv4");
+            4
+        }
+        _ => 4,
+    };
+
+    let me_families = ipv4_me as u8 + ipv6_me as u8;
+    let effective_multipath = config.multipath && me_families >= 2;
+
+    NetworkDecision {
+        ipv4_dc,
+        ipv6_dc,
+        ipv4_me,
+        ipv6_me,
+        effective_prefer,
+        effective_multipath,
+    }
+}
+
+fn detect_local_ip_v4() -> Option<Ipv4Addr> {
+    let socket = UdpSocket::bind("0.0.0.0:0").ok()?;
+    socket.connect("8.8.8.8:80").ok()?;
+    match socket.local_addr().ok()?.ip() {
+        IpAddr::V4(v4) => Some(v4),
+        _ => None,
+    }
+}
+
+fn detect_local_ip_v6() -> Option<Ipv6Addr> {
+    let socket = UdpSocket::bind("[::]:0").ok()?;
+    socket.connect("[2001:4860:4860::8888]:80").ok()?;
+    match socket.local_addr().ok()?.ip() {
+        IpAddr::V6(v6) => Some(v6),
+        _ => None,
+    }
+}
+
+pub fn is_bogon(ip: IpAddr) -> bool {
+    match ip {
+        IpAddr::V4(v4) => is_bogon_v4(v4),
+        IpAddr::V6(v6) => is_bogon_v6(v6),
+    }
+}
+
+pub fn is_bogon_v4(ip: Ipv4Addr) -> bool {
+    let octets = ip.octets();
+    if ip.is_private() || ip.is_loopback() || ip.is_link_local() {
+        return true;
+    }
+    if octets[0] == 0 {
+        return true;
+    }
+    if octets[0] == 100 && (octets[1] & 0xC0) == 64 {
+        return true;
+    }
+    if octets[0] == 192 && octets[1] == 0 && octets[2] == 0 {
+        return true;
+    }
+    if octets[0] == 192 && octets[1] == 0 && octets[2] == 2 {
+        return true;
+    }
+    if octets[0] == 198 && (octets[1] & 0xFE) == 18 {
+        return true;
+    }
+    if octets[0] == 198 && octets[1] == 51 && octets[2] == 100 {
+        return true;
+    }
+    if octets[0] == 203 && octets[1] == 0 && octets[2] == 113 {
+        return true;
+    }
+    if ip.is_multicast() {
+        return true;
+    }
+    if octets[0] >= 240 {
+        return true;
+    }
+    if ip.is_broadcast() {
+        return true;
+    }
+    false
+}
+
+pub fn is_bogon_v6(ip: Ipv6Addr) -> bool {
+    if ip.is_unspecified() || ip.is_loopback() || ip.is_unique_local() {
+        return true;
+    }
+    let segs = ip.segments();
+    if (segs[0] & 0xFFC0) == 0xFE80 {
+        return true;
+    }
+    if segs[0..5] == [0, 0, 0, 0, 0] && segs[5] == 0xFFFF {
+        return true;
+    }
+    if segs[0] == 0x0100 && segs[1..4] == [0, 0, 0] {
+        return true;
+    }
+    if segs[0] == 0x2001 && segs[1] == 0x0db8 {
+        return true;
+    }
+    if segs[0] == 0x2002 {
+        return true;
+    }
+    if ip.is_multicast() {
+        return true;
+    }
+    false
+}
+
+pub fn log_probe_result(probe: &NetworkProbe, decision: &NetworkDecision) {
+    info!(
+        ipv4 = probe.detected_ipv4.as_ref().map(|v| v.to_string()).unwrap_or_else(|| "-".into()),
+        ipv6 = probe.detected_ipv6.as_ref().map(|v| v.to_string()).unwrap_or_else(|| "-".into()),
+        reflected_v4 = probe.reflected_ipv4.as_ref().map(|v| v.ip().to_string()).unwrap_or_else(|| "-".into()),
+        reflected_v6 = probe.reflected_ipv6.as_ref().map(|v| v.ip().to_string()).unwrap_or_else(|| "-".into()),
+        ipv4_bogon = probe.ipv4_is_bogon,
+        ipv6_bogon = probe.ipv6_is_bogon,
+        ipv4_me = decision.ipv4_me,
+        ipv6_me = decision.ipv6_me,
+        ipv4_dc = decision.ipv4_dc,
+        ipv6_dc = decision.ipv6_dc,
+        prefer = decision.effective_prefer,
+        multipath = decision.effective_multipath,
+        "Network capabilities resolved"
+    );
+}

src/network/stun.rs

@@ -0,0 +1,234 @@
+#![allow(unreachable_code)]
+#![allow(dead_code)]
+
+use std::net::{IpAddr, Ipv4Addr, Ipv6Addr, SocketAddr};
+
+use tokio::net::{lookup_host, UdpSocket};
+use tokio::time::{timeout, Duration, sleep};
+
+use crate::error::{ProxyError, Result};
+use crate::network::dns_overrides::{resolve, split_host_port};
+
+#[derive(Debug, Clone, Copy, PartialEq, Eq, Hash)]
+pub enum IpFamily {
+    V4,
+    V6,
+}
+
+#[derive(Debug, Clone, Copy)]
+pub struct StunProbeResult {
+    pub local_addr: SocketAddr,
+    pub reflected_addr: SocketAddr,
+    pub family: IpFamily,
+}
+
+#[derive(Debug, Default, Clone)]
+pub struct DualStunResult {
+    pub v4: Option<StunProbeResult>,
+    pub v6: Option<StunProbeResult>,
+}
+
+pub async fn stun_probe_dual(stun_addr: &str) -> Result<DualStunResult> {
+    let (v4, v6) = tokio::join!(
+        stun_probe_family(stun_addr, IpFamily::V4),
+        stun_probe_family(stun_addr, IpFamily::V6),
+    );
+
+    Ok(DualStunResult {
+        v4: v4?,
+        v6: v6?,
+    })
+}
+
+pub async fn stun_probe_family(stun_addr: &str, family: IpFamily) -> Result<Option<StunProbeResult>> {
+    stun_probe_family_with_bind(stun_addr, family, None).await
+}
+
+pub async fn stun_probe_family_with_bind(
+    stun_addr: &str,
+    family: IpFamily,
+    bind_ip: Option<IpAddr>,
+) -> Result<Option<StunProbeResult>> {
+    use rand::RngCore;
+
+    let bind_addr = match (family, bind_ip) {
+        (IpFamily::V4, Some(IpAddr::V4(ip))) => SocketAddr::new(IpAddr::V4(ip), 0),
+        (IpFamily::V6, Some(IpAddr::V6(ip))) => SocketAddr::new(IpAddr::V6(ip), 0),
+        (IpFamily::V4, Some(IpAddr::V6(_))) | (IpFamily::V6, Some(IpAddr::V4(_))) => {
+            return Ok(None);
+        }
+        (IpFamily::V4, None) => SocketAddr::new(IpAddr::V4(Ipv4Addr::UNSPECIFIED), 0),
+        (IpFamily::V6, None) => SocketAddr::new(IpAddr::V6(Ipv6Addr::UNSPECIFIED), 0),
+    };
+
+    let socket = match UdpSocket::bind(bind_addr).await {
+        Ok(socket) => socket,
+        Err(_) if bind_ip.is_some() => return Ok(None),
+        Err(e) => return Err(ProxyError::Proxy(format!("STUN bind failed: {e}"))),
+    };
+
+    let target_addr = resolve_stun_addr(stun_addr, family).await?;
+    if let Some(addr) = target_addr {
+        match socket.connect(addr).await {
+            Ok(()) => {}
+            Err(e) if family == IpFamily::V6 && matches!(
+                e.kind(),
+                std::io::ErrorKind::NetworkUnreachable
+                | std::io::ErrorKind::HostUnreachable
+                | std::io::ErrorKind::Unsupported
+                | std::io::ErrorKind::NetworkDown
+            ) => return Ok(None),
+            Err(e) => return Err(ProxyError::Proxy(format!("STUN connect failed: {e}"))),
+        }
+    } else {
+        return Ok(None);
+    }
+
+    let mut req = [0u8; 20];
+    req[0..2].copy_from_slice(&0x0001u16.to_be_bytes()); // Binding Request
+    req[2..4].copy_from_slice(&0u16.to_be_bytes()); // length
+    req[4..8].copy_from_slice(&0x2112A442u32.to_be_bytes()); // magic cookie
+    rand::rng().fill_bytes(&mut req[8..20]); // transaction ID
+
+    let mut buf = [0u8; 256];
+    let mut attempt = 0;
+    let mut backoff = Duration::from_secs(1);
+    loop {
+        socket
+            .send(&req)
+            .await
+            .map_err(|e| ProxyError::Proxy(format!("STUN send failed: {e}")))?;
+
+        let recv_res = timeout(Duration::from_secs(3), socket.recv(&mut buf)).await;
+        let n = match recv_res {
+            Ok(Ok(n)) => n,
+            Ok(Err(e)) => return Err(ProxyError::Proxy(format!("STUN recv failed: {e}"))),
+            Err(_) => {
+                attempt += 1;
+                if attempt >= 3 {
+                    return Ok(None);
+                }
+                sleep(backoff).await;
+                backoff *= 2;
+                continue;
+            }
+        };
+
+        if n < 20 {
+            return Ok(None);
+        }
+
+        let magic = 0x2112A442u32.to_be_bytes();
+        let txid = &req[8..20];
+    let mut idx = 20;
+    while idx + 4 <= n {
+        let atype = u16::from_be_bytes(buf[idx..idx + 2].try_into().unwrap());
+        let alen = u16::from_be_bytes(buf[idx + 2..idx + 4].try_into().unwrap()) as usize;
+        idx += 4;
+        if idx + alen > n {
+            break;
+        }
+
+        match atype {
+            0x0020 /* XOR-MAPPED-ADDRESS */ | 0x0001 /* MAPPED-ADDRESS */ => {
+                if alen < 8 {
+                    break;
+                }
+                let family_byte = buf[idx + 1];
+                let port_bytes = [buf[idx + 2], buf[idx + 3]];
+                let len_check = match family_byte {
+                    0x01 => 4,
+                    0x02 => 16,
+                    _ => 0,
+                };
+                if len_check == 0 || alen < 4 + len_check {
+                    break;
+                }
+
+                let raw_ip = &buf[idx + 4..idx + 4 + len_check];
+                let mut port = u16::from_be_bytes(port_bytes);
+
+                let reflected_ip = if atype == 0x0020 {
+                    port ^= ((magic[0] as u16) << 8) | magic[1] as u16;
+                    match family_byte {
+                        0x01 => {
+                            let ip = [
+                                raw_ip[0] ^ magic[0],
+                                raw_ip[1] ^ magic[1],
+                                raw_ip[2] ^ magic[2],
+                                raw_ip[3] ^ magic[3],
+                            ];
+                            IpAddr::V4(Ipv4Addr::new(ip[0], ip[1], ip[2], ip[3]))
+                        }
+                        0x02 => {
+                            let mut ip = [0u8; 16];
+                            let xor_key = [magic.as_slice(), txid].concat();
+                            for (i, b) in raw_ip.iter().enumerate().take(16) {
+                                ip[i] = *b ^ xor_key[i];
+                            }
+                            IpAddr::V6(Ipv6Addr::from(ip))
+                        }
+                        _ => {
+                            idx += (alen + 3) & !3;
+                            continue;
+                        }
+                    }
+                } else {
+                    match family_byte {
+                        0x01 => IpAddr::V4(Ipv4Addr::new(raw_ip[0], raw_ip[1], raw_ip[2], raw_ip[3])),
+                        0x02 => IpAddr::V6(Ipv6Addr::from(<[u8; 16]>::try_from(raw_ip).unwrap())),
+                        _ => {
+                            idx += (alen + 3) & !3;
+                            continue;
+                        }
+                    }
+                };
+
+                let reflected_addr = SocketAddr::new(reflected_ip, port);
+                let local_addr = socket
+                    .local_addr()
+                    .map_err(|e| ProxyError::Proxy(format!("STUN local_addr failed: {e}")))?;
+
+                return Ok(Some(StunProbeResult {
+                    local_addr,
+                    reflected_addr,
+                    family,
+                }));
+            }
+            _ => {}
+        }
+
+        idx += (alen + 3) & !3;
+    }
+
+    }
+
+    Ok(None)
+}
+
+async fn resolve_stun_addr(stun_addr: &str, family: IpFamily) -> Result<Option<SocketAddr>> {
+    if let Ok(addr) = stun_addr.parse::<SocketAddr>() {
+        return Ok(match (addr.is_ipv4(), family) {
+            (true, IpFamily::V4) | (false, IpFamily::V6) => Some(addr),
+            _ => None,
+        });
+    }
+
+    if let Some((host, port)) = split_host_port(stun_addr)
+        && let Some(ip) = resolve(&host, port)
+    {
+        let addr = SocketAddr::new(ip, port);
+        return Ok(match (addr.is_ipv4(), family) {
+            (true, IpFamily::V4) | (false, IpFamily::V6) => Some(addr),
+            _ => None,
+        });
+    }
+
+    let mut addrs = lookup_host(stun_addr)
+        .await
+        .map_err(|e| ProxyError::Proxy(format!("STUN resolve failed: {e}")))?;
+
+    let target = addrs
+        .find(|a| matches!((a.is_ipv4(), family), (true, IpFamily::V4) | (false, IpFamily::V6)));
+    Ok(target)
+}

src/protocol/constants.rs

@@ -1,13 +1,17 @@
 //! Protocol constants and datacenter addresses
 
-use std::net::{IpAddr, Ipv4Addr, Ipv6Addr};
-use once_cell::sync::Lazy;
+#![allow(dead_code)]
+
+use std::net::{IpAddr, Ipv4Addr};
+
+use crate::crypto::SecureRandom;
+use std::sync::LazyLock;
 
 // ============= Telegram Datacenters =============
 
 pub const TG_DATACENTER_PORT: u16 = 443;
 
-pub static TG_DATACENTERS_V4: Lazy<Vec<IpAddr>> = Lazy::new(|| {
+pub static TG_DATACENTERS_V4: LazyLock<Vec<IpAddr>> = LazyLock::new(|| {
     vec![
         IpAddr::V4(Ipv4Addr::new(149, 154, 175, 50)),
         IpAddr::V4(Ipv4Addr::new(149, 154, 167, 51)),
@@ -17,7 +21,7 @@ pub static TG_DATACENTERS_V4: Lazy<Vec<IpAddr>> = Lazy::new(|| {
     ]
 });
 
-pub static TG_DATACENTERS_V6: Lazy<Vec<IpAddr>> = Lazy::new(|| {
+pub static TG_DATACENTERS_V6: LazyLock<Vec<IpAddr>> = LazyLock::new(|| {
     vec![
         IpAddr::V6("2001:b28:f23d:f001::a".parse().unwrap()),
         IpAddr::V6("2001:67c:04e8:f002::a".parse().unwrap()),
@@ -29,8 +33,8 @@ pub static TG_DATACENTERS_V6: Lazy<Vec<IpAddr>> = Lazy::new(|| {
 
 // ============= Middle Proxies (for advertising) =============
 
-pub static TG_MIDDLE_PROXIES_V4: Lazy<std::collections::HashMap<i32, Vec<(IpAddr, u16)>>> = 
-    Lazy::new(|| {
+pub static TG_MIDDLE_PROXIES_V4: LazyLock<std::collections::HashMap<i32, Vec<(IpAddr, u16)>>> = 
+    LazyLock::new(|| {
         let mut m = std::collections::HashMap::new();
         m.insert(1, vec![(IpAddr::V4(Ipv4Addr::new(149, 154, 175, 50)), 8888)]);
         m.insert(-1, vec![(IpAddr::V4(Ipv4Addr::new(149, 154, 175, 50)), 8888)]);
@@ -45,8 +49,8 @@ pub static TG_MIDDLE_PROXIES_V4: Lazy<std::collections::HashMap<i32, Vec<(IpAddr
         m
     });
 
-pub static TG_MIDDLE_PROXIES_V6: Lazy<std::collections::HashMap<i32, Vec<(IpAddr, u16)>>> = 
-    Lazy::new(|| {
+pub static TG_MIDDLE_PROXIES_V6: LazyLock<std::collections::HashMap<i32, Vec<(IpAddr, u16)>>> = 
+    LazyLock::new(|| {
         let mut m = std::collections::HashMap::new();
         m.insert(1, vec![(IpAddr::V6("2001:b28:f23d:f001::d".parse().unwrap()), 8888)]);
         m.insert(-1, vec![(IpAddr::V6("2001:b28:f23d:f001::d".parse().unwrap()), 8888)]);
@@ -151,7 +155,32 @@ pub const TLS_RECORD_ALERT: u8 = 0x15;
 /// Maximum TLS record size
 pub const MAX_TLS_RECORD_SIZE: usize = 16384;
 /// Maximum TLS chunk size (with overhead)
-pub const MAX_TLS_CHUNK_SIZE: usize = 16384 + 24;
+/// RFC 8446 §5.2 allows up to 16384 + 256 bytes of ciphertext
+pub const MAX_TLS_CHUNK_SIZE: usize = 16384 + 256;
+
+/// Secure Intermediate payload is expected to be 4-byte aligned.
+pub fn is_valid_secure_payload_len(data_len: usize) -> bool {
+    data_len.is_multiple_of(4)
+}
+
+/// Compute Secure Intermediate payload length from wire length.
+/// Secure mode strips up to 3 random tail bytes by truncating to 4-byte boundary.
+pub fn secure_payload_len_from_wire_len(wire_len: usize) -> Option<usize> {
+    if wire_len < 4 {
+        return None;
+    }
+    Some(wire_len - (wire_len % 4))
+}
+
+/// Generate padding length for Secure Intermediate protocol.
+/// Data must be 4-byte aligned; padding is 1..=3 so total is never divisible by 4.
+pub fn secure_padding_len(data_len: usize, rng: &SecureRandom) -> usize {
+    debug_assert!(
+        is_valid_secure_payload_len(data_len),
+        "Secure payload must be 4-byte aligned, got {data_len}"
+    );
+    rng.range(3) + 1
+}
 
 // ============= Timeouts =============
 
@@ -167,7 +196,8 @@ pub const DEFAULT_ACK_TIMEOUT_SECS: u64 = 300;
 // ============= Buffer Sizes =============
 
 /// Default buffer size
-pub const DEFAULT_BUFFER_SIZE: usize = 65536;
+pub const DEFAULT_BUFFER_SIZE: usize = 16384;
+
 /// Small buffer size for bad client handling
 pub const SMALL_BUFFER_SIZE: usize = 8192;
 
@@ -201,6 +231,16 @@ pub static RESERVED_NONCE_CONTINUES: &[[u8; 4]] = &[
 // ============= RPC Constants (for Middle Proxy) =============
 
 /// RPC Proxy Request
+/// RPC Flags (from Erlang mtp_rpc.erl)
+pub const RPC_FLAG_NOT_ENCRYPTED: u32 = 0x2;
+pub const RPC_FLAG_HAS_AD_TAG: u32    = 0x8;
+pub const RPC_FLAG_MAGIC: u32          = 0x1000;
+pub const RPC_FLAG_EXTMODE2: u32       = 0x20000;
+pub const RPC_FLAG_PAD: u32            = 0x8000000;
+pub const RPC_FLAG_INTERMEDIATE: u32   = 0x20000000;
+pub const RPC_FLAG_ABRIDGED: u32       = 0x40000000;
+pub const RPC_FLAG_QUICKACK: u32       = 0x80000000;
+
 pub const RPC_PROXY_REQ: [u8; 4] = [0xee, 0xf1, 0xce, 0x36];
 /// RPC Proxy Answer
 pub const RPC_PROXY_ANS: [u8; 4] = [0x0d, 0xda, 0x03, 0x44];
@@ -227,7 +267,60 @@ pub mod rpc_flags {
     pub const FLAG_QUICKACK: u32 = 0x80000000;
 }
 
-#[cfg(test)]
+
+    // ============= Middle-End Proxy Servers =============
+    pub const ME_PROXY_PORT: u16 = 8888;
+    
+    pub static TG_MIDDLE_PROXIES_FLAT_V4: LazyLock<Vec<(IpAddr, u16)>> = LazyLock::new(|| {
+        vec![
+            (IpAddr::V4(Ipv4Addr::new(149, 154, 175, 50)), 8888),
+            (IpAddr::V4(Ipv4Addr::new(149, 154, 161, 144)), 8888),
+            (IpAddr::V4(Ipv4Addr::new(149, 154, 175, 100)), 8888),
+            (IpAddr::V4(Ipv4Addr::new(91, 108, 4, 136)), 8888),
+            (IpAddr::V4(Ipv4Addr::new(91, 108, 56, 183)), 8888),
+        ]
+    });
+    
+    // ============= RPC Constants (u32 native endian) =============
+    // From mtproto-common.h + net-tcp-rpc-common.h + mtproto-proxy.c
+    
+    pub const RPC_NONCE_U32: u32           = 0x7acb87aa;
+    pub const RPC_HANDSHAKE_U32: u32       = 0x7682eef5;
+    pub const RPC_HANDSHAKE_ERROR_U32: u32 = 0x6a27beda;
+    pub const TL_PROXY_TAG_U32: u32        = 0xdb1e26ae;  // mtproto-proxy.c:121
+    
+    // mtproto-common.h
+    pub const RPC_PROXY_REQ_U32: u32       = 0x36cef1ee;
+    pub const RPC_PROXY_ANS_U32: u32       = 0x4403da0d;
+    pub const RPC_CLOSE_CONN_U32: u32      = 0x1fcf425d;
+    pub const RPC_CLOSE_EXT_U32: u32       = 0x5eb634a2;
+    pub const RPC_SIMPLE_ACK_U32: u32      = 0x3bac409b;
+    pub const RPC_PING_U32: u32            = 0x5730a2df;
+    pub const RPC_PONG_U32: u32            = 0x8430eaa7;
+    
+    pub const RPC_CRYPTO_NONE_U32: u32 = 0;
+    pub const RPC_CRYPTO_AES_U32: u32  = 1;
+    
+    pub mod proxy_flags {
+        pub const FLAG_HAS_AD_TAG: u32    = 1;
+        pub const FLAG_NOT_ENCRYPTED: u32 = 0x2;
+        pub const FLAG_HAS_AD_TAG2: u32   = 0x8;
+        pub const FLAG_MAGIC: u32         = 0x1000;
+        pub const FLAG_EXTMODE2: u32      = 0x20000;
+        pub const FLAG_PAD: u32           = 0x8000000;
+        pub const FLAG_INTERMEDIATE: u32  = 0x20000000;
+        pub const FLAG_ABRIDGED: u32      = 0x40000000;
+        pub const FLAG_QUICKACK: u32      = 0x80000000;
+    }
+
+    pub mod rpc_crypto_flags {
+        pub const USE_CRC32C: u32 = 0x800;
+    }
+    
+    pub const ME_CONNECT_TIMEOUT_SECS: u64 = 5;
+    pub const ME_HANDSHAKE_TIMEOUT_SECS: u64 = 10;
+    
+    #[cfg(test)]
 mod tests {
     use super::*;
     
@@ -258,4 +351,43 @@ mod tests {
         assert_eq!(TG_DATACENTERS_V4.len(), 5);
         assert_eq!(TG_DATACENTERS_V6.len(), 5);
     }
-}
+
+    #[test]
+    fn secure_padding_never_produces_aligned_total() {
+        let rng = SecureRandom::new();
+        for data_len in (0..1000).step_by(4) {
+            for _ in 0..100 {
+                let padding = secure_padding_len(data_len, &rng);
+                assert!(
+                    padding <= 3,
+                    "padding out of range: data_len={data_len}, padding={padding}"
+                );
+                assert_ne!(
+                    (data_len + padding) % 4,
+                    0,
+                    "invariant violated: data_len={data_len}, padding={padding}, total={}",
+                    data_len + padding
+                );
+            }
+        }
+    }
+
+    #[test]
+    fn secure_wire_len_roundtrip_for_aligned_payload() {
+        for payload_len in (4..4096).step_by(4) {
+            for padding in 0..=3usize {
+                let wire_len = payload_len + padding;
+                let recovered = secure_payload_len_from_wire_len(wire_len);
+                assert_eq!(recovered, Some(payload_len));
+            }
+        }
+    }
+
+    #[test]
+    fn secure_wire_len_rejects_too_short_frames() {
+        assert_eq!(secure_payload_len_from_wire_len(0), None);
+        assert_eq!(secure_payload_len_from_wire_len(1), None);
+        assert_eq!(secure_payload_len_from_wire_len(2), None);
+        assert_eq!(secure_payload_len_from_wire_len(3), None);
+    }
+}

src/protocol/frame.rs

@@ -1,5 +1,7 @@
 //! MTProto frame types and metadata
 
+#![allow(dead_code)]
+
 use std::collections::HashMap;
 
 /// Extra metadata associated with a frame
@@ -83,7 +85,7 @@ impl FrameMode {
 pub fn validate_message_length(len: usize) -> bool {
     use super::constants::{MIN_MSG_LEN, MAX_MSG_LEN, PADDING_FILLER};
     
-    len >= MIN_MSG_LEN && len <= MAX_MSG_LEN && len % PADDING_FILLER.len() == 0
+    (MIN_MSG_LEN..=MAX_MSG_LEN).contains(&len) && len.is_multiple_of(PADDING_FILLER.len())
 }
 
 #[cfg(test)]

src/protocol/mod.rs

@@ -5,7 +5,11 @@ pub mod frame;
 pub mod obfuscation;
 pub mod tls;
 
+#[allow(unused_imports)]
 pub use constants::*;
+#[allow(unused_imports)]
 pub use frame::*;
+#[allow(unused_imports)]
 pub use obfuscation::*;
+#[allow(unused_imports)]
 pub use tls::*;

src/protocol/obfuscation.rs

@@ -1,10 +1,14 @@
 //! MTProto Obfuscation
 
+#![allow(dead_code)]
+
+use zeroize::Zeroize;
 use crate::crypto::{sha256, AesCtr};
-use crate::error::Result;
 use super::constants::*;
 
 /// Obfuscation parameters from handshake
+///
+/// Key material is zeroized on drop.
 #[derive(Debug, Clone)]
 pub struct ObfuscationParams {
     /// Key for decrypting client -> proxy traffic
@@ -21,25 +25,31 @@ pub struct ObfuscationParams {
     pub dc_idx: i16,
 }
 
+impl Drop for ObfuscationParams {
+    fn drop(&mut self) {
+        self.decrypt_key.zeroize();
+        self.decrypt_iv.zeroize();
+        self.encrypt_key.zeroize();
+        self.encrypt_iv.zeroize();
+    }
+}
+
 impl ObfuscationParams {
     /// Parse obfuscation parameters from handshake bytes
     /// Returns None if handshake doesn't match any user secret
     pub fn from_handshake(
         handshake: &[u8; HANDSHAKE_LEN],
-        secrets: &[(String, Vec<u8>)], // (username, secret_bytes)
+        secrets: &[(String, Vec<u8>)],
     ) -> Option<(Self, String)> {
-        // Extract prekey and IV for decryption
         let dec_prekey_iv = &handshake[SKIP_LEN..SKIP_LEN + PREKEY_LEN + IV_LEN];
         let dec_prekey = &dec_prekey_iv[..PREKEY_LEN];
         let dec_iv_bytes = &dec_prekey_iv[PREKEY_LEN..];
         
-        // Reversed for encryption direction
         let enc_prekey_iv: Vec<u8> = dec_prekey_iv.iter().rev().copied().collect();
         let enc_prekey = &enc_prekey_iv[..PREKEY_LEN];
         let enc_iv_bytes = &enc_prekey_iv[PREKEY_LEN..];
         
         for (username, secret) in secrets {
-            // Derive decryption key
             let mut dec_key_input = Vec::with_capacity(PREKEY_LEN + secret.len());
             dec_key_input.extend_from_slice(dec_prekey);
             dec_key_input.extend_from_slice(secret);
@@ -47,26 +57,22 @@ impl ObfuscationParams {
             
             let decrypt_iv = u128::from_be_bytes(dec_iv_bytes.try_into().unwrap());
             
-            // Create decryptor and decrypt handshake
             let mut decryptor = AesCtr::new(&decrypt_key, decrypt_iv);
             let decrypted = decryptor.decrypt(handshake);
             
-            // Check protocol tag
             let tag_bytes: [u8; 4] = decrypted[PROTO_TAG_POS..PROTO_TAG_POS + 4]
                 .try_into()
                 .unwrap();
             
             let proto_tag = match ProtoTag::from_bytes(tag_bytes) {
                 Some(tag) => tag,
-                None => continue, // Try next secret
+                None => continue,
             };
             
-            // Extract DC index
             let dc_idx = i16::from_le_bytes(
                 decrypted[DC_IDX_POS..DC_IDX_POS + 2].try_into().unwrap()
             );
             
-            // Derive encryption key
             let mut enc_key_input = Vec::with_capacity(PREKEY_LEN + secret.len());
             enc_key_input.extend_from_slice(enc_prekey);
             enc_key_input.extend_from_slice(secret);
@@ -123,18 +129,15 @@ pub fn generate_nonce<R: FnMut(usize) -> Vec<u8>>(mut random_bytes: R) -> [u8; H
 
 /// Check if nonce is valid (not matching reserved patterns)
 pub fn is_valid_nonce(nonce: &[u8; HANDSHAKE_LEN]) -> bool {
-    // Check first byte
     if RESERVED_NONCE_FIRST_BYTES.contains(&nonce[0]) {
         return false;
     }
     
-    // Check first 4 bytes
     let first_four: [u8; 4] = nonce[..4].try_into().unwrap();
     if RESERVED_NONCE_BEGINNINGS.contains(&first_four) {
         return false;
     }
     
-    // Check bytes 4-7
     let continue_four: [u8; 4] = nonce[4..8].try_into().unwrap();
     if RESERVED_NONCE_CONTINUES.contains(&continue_four) {
         return false;
@@ -147,12 +150,10 @@ pub fn is_valid_nonce(nonce: &[u8; HANDSHAKE_LEN]) -> bool {
 pub fn prepare_tg_nonce(
     nonce: &mut [u8; HANDSHAKE_LEN],
     proto_tag: ProtoTag,
-    enc_key_iv: Option<&[u8]>, // For fast mode
+    enc_key_iv: Option<&[u8]>,
 ) {
-    // Set protocol tag
     nonce[PROTO_TAG_POS..PROTO_TAG_POS + 4].copy_from_slice(&proto_tag.to_bytes());
     
-    // For fast mode, copy the reversed enc_key_iv
     if let Some(key_iv) = enc_key_iv {
         let reversed: Vec<u8> = key_iv.iter().rev().copied().collect();
         nonce[SKIP_LEN..SKIP_LEN + KEY_LEN + IV_LEN].copy_from_slice(&reversed);
@@ -160,15 +161,19 @@ pub fn prepare_tg_nonce(
 }
 
 /// Encrypt the outgoing nonce for Telegram
+/// Legacy helper — **do not use**.
+/// WARNING: logic diverges from Python/C reference (SHA256 of 48 bytes, IV from head).
+/// Kept only to avoid breaking external callers; prefer `encrypt_tg_nonce_with_ciphers`.
+#[deprecated(
+    note = "Incorrect MTProto obfuscation KDF; use proxy::handshake::encrypt_tg_nonce_with_ciphers"
+)]
 pub fn encrypt_nonce(nonce: &[u8; HANDSHAKE_LEN]) -> Vec<u8> {
-    // Derive encryption key from the nonce itself
     let key_iv = &nonce[SKIP_LEN..SKIP_LEN + KEY_LEN + IV_LEN];
     let enc_key = sha256(key_iv);
     let enc_iv = u128::from_be_bytes(key_iv[..IV_LEN].try_into().unwrap());
     
     let mut encryptor = AesCtr::new(&enc_key, enc_iv);
     
-    // Only encrypt from PROTO_TAG_POS onwards
     let mut result = nonce.to_vec();
     let encrypted_part = encryptor.encrypt(&nonce[PROTO_TAG_POS..]);
     result[PROTO_TAG_POS..].copy_from_slice(&encrypted_part);
@@ -182,22 +187,18 @@ mod tests {
     
     #[test]
     fn test_is_valid_nonce() {
-        // Valid nonce
         let mut valid = [0x42u8; HANDSHAKE_LEN];
         valid[4..8].copy_from_slice(&[1, 2, 3, 4]);
         assert!(is_valid_nonce(&valid));
         
-        // Invalid: starts with 0xef
         let mut invalid = [0x00u8; HANDSHAKE_LEN];
         invalid[0] = 0xef;
         assert!(!is_valid_nonce(&invalid));
         
-        // Invalid: starts with HEAD
         let mut invalid = [0x00u8; HANDSHAKE_LEN];
         invalid[..4].copy_from_slice(b"HEAD");
         assert!(!is_valid_nonce(&invalid));
         
-        // Invalid: bytes 4-7 are zeros
         let mut invalid = [0x42u8; HANDSHAKE_LEN];
         invalid[4..8].copy_from_slice(&[0, 0, 0, 0]);
         assert!(!is_valid_nonce(&invalid));
@@ -214,4 +215,4 @@ mod tests {
         assert!(is_valid_nonce(&nonce));
         assert_eq!(nonce.len(), HANDSHAKE_LEN);
     }
-}
+}

src/protocol/tls.rs

@@ -4,10 +4,15 @@
 //! for domain fronting. The handshake looks like valid TLS 1.3 but
 //! actually carries MTProto authentication data.
 
-use crate::crypto::{sha256_hmac, random::SECURE_RANDOM};
-use crate::error::{ProxyError, Result};
+#![allow(dead_code)]
+
+use crate::crypto::{sha256_hmac, SecureRandom};
+#[cfg(test)]
+use crate::error::ProxyError;
 use super::constants::*;
 use std::time::{SystemTime, UNIX_EPOCH};
+use num_bigint::BigUint;
+use num_traits::One;
 
 // ============= Public Constants =============
 
@@ -30,6 +35,7 @@ pub const TIME_SKEW_MAX: i64 = 10 * 60;  // 10 minutes after
 mod extension_type {
     pub const KEY_SHARE: u16 = 0x0033;
     pub const SUPPORTED_VERSIONS: u16 = 0x002b;
+    pub const ALPN: u16 = 0x0010;
 }
 
 /// TLS Cipher Suites
@@ -60,6 +66,7 @@ pub struct TlsValidation {
 // ============= TLS Extension Builder =============
 
 /// Builder for TLS extensions with correct length calculation
+#[derive(Clone)]
 struct TlsExtensionBuilder {
     extensions: Vec<u8>,
 }
@@ -106,6 +113,27 @@ impl TlsExtensionBuilder {
         
         self
     }
+
+    /// Add ALPN extension with a single selected protocol.
+    fn add_alpn(&mut self, proto: &[u8]) -> &mut Self {
+        // Extension type: ALPN (0x0010)
+        self.extensions.extend_from_slice(&extension_type::ALPN.to_be_bytes());
+
+        // ALPN extension format:
+        // extension_data length (2 bytes)
+        //   protocols length (2 bytes)
+        //     protocol name length (1 byte)
+        //     protocol name bytes
+        let proto_len = proto.len() as u8;
+        let list_len: u16 = 1 + proto_len as u16;
+        let ext_len: u16 = 2 + list_len;
+
+        self.extensions.extend_from_slice(&ext_len.to_be_bytes());
+        self.extensions.extend_from_slice(&list_len.to_be_bytes());
+        self.extensions.push(proto_len);
+        self.extensions.extend_from_slice(proto);
+        self
+    }
     
     /// Build final extensions with length prefix
     fn build(self) -> Vec<u8> {
@@ -142,6 +170,8 @@ struct ServerHelloBuilder {
     compression: u8,
     /// Extensions
     extensions: TlsExtensionBuilder,
+    /// Selected ALPN protocol (if any)
+    alpn: Option<Vec<u8>>,
 }
 
 impl ServerHelloBuilder {
@@ -152,6 +182,7 @@ impl ServerHelloBuilder {
             cipher_suite: cipher_suite::TLS_AES_128_GCM_SHA256,
             compression: 0x00,
             extensions: TlsExtensionBuilder::new(),
+            alpn: None,
         }
     }
     
@@ -165,10 +196,19 @@ impl ServerHelloBuilder {
         self.extensions.add_supported_versions(0x0304);
         self
     }
+
+    fn with_alpn(mut self, proto: Option<Vec<u8>>) -> Self {
+        self.alpn = proto;
+        self
+    }
     
     /// Build ServerHello message (without record header)
     fn build_message(&self) -> Vec<u8> {
-        let extensions = self.extensions.extensions.clone();
+        let mut ext_builder = self.extensions.clone();
+        if let Some(ref alpn) = self.alpn {
+            ext_builder.add_alpn(alpn);
+        }
+        let extensions = ext_builder.extensions.clone();
         let extensions_len = extensions.len() as u16;
         
         // Calculate total length
@@ -295,7 +335,7 @@ pub fn validate_tls_handshake(
             // This is a quirk in some clients that use uptime instead of real time
             let is_boot_time = timestamp < 60 * 60 * 24 * 1000; // < ~2.7 years in seconds
             
-            if !is_boot_time && (time_diff < TIME_SKEW_MIN || time_diff > TIME_SKEW_MAX) {
+            if !is_boot_time && !(TIME_SKEW_MIN..=TIME_SKEW_MAX).contains(&time_diff) {
                 continue;
             }
         }
@@ -311,13 +351,27 @@ pub fn validate_tls_handshake(
     None
 }
 
+fn curve25519_prime() -> BigUint {
+    (BigUint::one() << 255) - BigUint::from(19u32)
+}
+
 /// Generate a fake X25519 public key for TLS
 ///
-/// This generates random bytes that look like a valid X25519 public key.
-/// Since we're not doing real TLS, the actual cryptographic properties don't matter.
-pub fn gen_fake_x25519_key() -> [u8; 32] {
-    let bytes = SECURE_RANDOM.bytes(32);
-    bytes.try_into().unwrap()
+/// Produces a quadratic residue mod p = 2^255 - 19 by computing n² mod p,
+/// which matches Python/C behavior and avoids DPI fingerprinting.
+pub fn gen_fake_x25519_key(rng: &SecureRandom) -> [u8; 32] {
+    let mut n_bytes = [0u8; 32];
+    n_bytes.copy_from_slice(&rng.bytes(32));
+
+    let n = BigUint::from_bytes_le(&n_bytes);
+    let p = curve25519_prime();
+    let pk = (&n * &n) % &p;
+
+    let mut out = pk.to_bytes_le();
+    out.resize(32, 0);
+    let mut result = [0u8; 32];
+    result.copy_from_slice(&out[..32]);
+    result
 }
 
 /// Build TLS ServerHello response
@@ -333,13 +387,20 @@ pub fn build_server_hello(
     client_digest: &[u8; TLS_DIGEST_LEN],
     session_id: &[u8],
     fake_cert_len: usize,
+    rng: &SecureRandom,
+    alpn: Option<Vec<u8>>,
+    new_session_tickets: u8,
 ) -> Vec<u8> {
-    let x25519_key = gen_fake_x25519_key();
+    const MIN_APP_DATA: usize = 64;
+    const MAX_APP_DATA: usize = 16640; // RFC 8446 §5.2 upper bound
+    let fake_cert_len = fake_cert_len.clamp(MIN_APP_DATA, MAX_APP_DATA);
+    let x25519_key = gen_fake_x25519_key(rng);
     
     // Build ServerHello
     let server_hello = ServerHelloBuilder::new(session_id.to_vec())
         .with_x25519_key(&x25519_key)
         .with_tls13_version()
+        .with_alpn(alpn)
         .build_record();
     
     // Build Change Cipher Spec record
@@ -351,20 +412,40 @@ pub fn build_server_hello(
     ];
     
     // Build fake certificate (Application Data record)
-    let fake_cert = SECURE_RANDOM.bytes(fake_cert_len);
+    let fake_cert = rng.bytes(fake_cert_len);
     let mut app_data_record = Vec::with_capacity(5 + fake_cert_len);
     app_data_record.push(TLS_RECORD_APPLICATION);
     app_data_record.extend_from_slice(&TLS_VERSION);
     app_data_record.extend_from_slice(&(fake_cert_len as u16).to_be_bytes());
+    // Fill ApplicationData with fully random bytes of desired length to avoid
+    // deterministic DPI fingerprints (fixed inner content type markers).
     app_data_record.extend_from_slice(&fake_cert);
     
+    // Build optional NewSessionTicket records (TLS 1.3 handshake messages are encrypted;
+    // here we mimic with opaque ApplicationData records of plausible size).
+    let mut tickets = Vec::new();
+    if new_session_tickets > 0 {
+        for _ in 0..new_session_tickets {
+            let ticket_len: usize = rng.range(48) + 48; // 48-95 bytes
+            let mut record = Vec::with_capacity(5 + ticket_len);
+            record.push(TLS_RECORD_APPLICATION);
+            record.extend_from_slice(&TLS_VERSION);
+            record.extend_from_slice(&(ticket_len as u16).to_be_bytes());
+            record.extend_from_slice(&rng.bytes(ticket_len));
+            tickets.push(record);
+        }
+    }
+
     // Combine all records
     let mut response = Vec::with_capacity(
-        server_hello.len() + change_cipher_spec.len() + app_data_record.len()
+        server_hello.len() + change_cipher_spec.len() + app_data_record.len() + tickets.iter().map(|r| r.len()).sum::<usize>()
     );
     response.extend_from_slice(&server_hello);
     response.extend_from_slice(&change_cipher_spec);
     response.extend_from_slice(&app_data_record);
+    for t in &tickets {
+        response.extend_from_slice(t);
+    }
     
     // Compute HMAC for the response
     let mut hmac_input = Vec::with_capacity(TLS_DIGEST_LEN + response.len());
@@ -380,6 +461,131 @@ pub fn build_server_hello(
     response
 }
 
+/// Extract SNI (server_name) from a TLS ClientHello.
+pub fn extract_sni_from_client_hello(handshake: &[u8]) -> Option<String> {
+    if handshake.len() < 43 || handshake[0] != TLS_RECORD_HANDSHAKE {
+        return None;
+    }
+
+    let mut pos = 5; // after record header
+    if handshake.get(pos).copied()? != 0x01 {
+        return None; // not ClientHello
+    }
+
+    // Handshake length bytes
+    pos += 4; // type + len (3)
+
+    // version (2) + random (32)
+    pos += 2 + 32;
+    if pos + 1 > handshake.len() {
+        return None;
+    }
+
+    let session_id_len = *handshake.get(pos)? as usize;
+    pos += 1 + session_id_len;
+    if pos + 2 > handshake.len() {
+        return None;
+    }
+
+    let cipher_suites_len = u16::from_be_bytes([handshake[pos], handshake[pos + 1]]) as usize;
+    pos += 2 + cipher_suites_len;
+    if pos + 1 > handshake.len() {
+        return None;
+    }
+
+    let comp_len = *handshake.get(pos)? as usize;
+    pos += 1 + comp_len;
+    if pos + 2 > handshake.len() {
+        return None;
+    }
+
+    let ext_len = u16::from_be_bytes([handshake[pos], handshake[pos + 1]]) as usize;
+    pos += 2;
+    let ext_end = pos + ext_len;
+    if ext_end > handshake.len() {
+        return None;
+    }
+
+    while pos + 4 <= ext_end {
+        let etype = u16::from_be_bytes([handshake[pos], handshake[pos + 1]]);
+        let elen = u16::from_be_bytes([handshake[pos + 2], handshake[pos + 3]]) as usize;
+        pos += 4;
+        if pos + elen > ext_end {
+            break;
+        }
+        if etype == 0x0000 && elen >= 5 {
+            // server_name extension
+            let list_len = u16::from_be_bytes([handshake[pos], handshake[pos + 1]]) as usize;
+            let mut sn_pos = pos + 2;
+            let sn_end = std::cmp::min(sn_pos + list_len, pos + elen);
+            while sn_pos + 3 <= sn_end {
+                let name_type = handshake[sn_pos];
+                let name_len = u16::from_be_bytes([handshake[sn_pos + 1], handshake[sn_pos + 2]]) as usize;
+                sn_pos += 3;
+                if sn_pos + name_len > sn_end {
+                    break;
+                }
+                if name_type == 0 && name_len > 0
+                    && let Ok(host) = std::str::from_utf8(&handshake[sn_pos..sn_pos + name_len])
+                {
+                    return Some(host.to_string());
+                }
+                sn_pos += name_len;
+            }
+        }
+        pos += elen;
+    }
+
+    None
+}
+
+/// Extract ALPN protocol list from ClientHello, return in offered order.
+pub fn extract_alpn_from_client_hello(handshake: &[u8]) -> Vec<Vec<u8>> {
+    let mut pos = 5; // after record header
+    if handshake.get(pos) != Some(&0x01) {
+        return Vec::new();
+    }
+    pos += 4; // type + len
+    pos += 2 + 32; // version + random
+    if pos >= handshake.len() { return Vec::new(); }
+    let session_id_len = *handshake.get(pos).unwrap_or(&0) as usize;
+    pos += 1 + session_id_len;
+    if pos + 2 > handshake.len() { return Vec::new(); }
+    let cipher_len = u16::from_be_bytes([handshake[pos], handshake[pos+1]]) as usize;
+    pos += 2 + cipher_len;
+    if pos >= handshake.len() { return Vec::new(); }
+    let comp_len = *handshake.get(pos).unwrap_or(&0) as usize;
+    pos += 1 + comp_len;
+    if pos + 2 > handshake.len() { return Vec::new(); }
+    let ext_len = u16::from_be_bytes([handshake[pos], handshake[pos+1]]) as usize;
+    pos += 2;
+    let ext_end = pos + ext_len;
+    if ext_end > handshake.len() { return Vec::new(); }
+    let mut out = Vec::new();
+    while pos + 4 <= ext_end {
+        let etype = u16::from_be_bytes([handshake[pos], handshake[pos+1]]);
+        let elen = u16::from_be_bytes([handshake[pos+2], handshake[pos+3]]) as usize;
+        pos += 4;
+        if pos + elen > ext_end { break; }
+        if etype == extension_type::ALPN && elen >= 3 {
+            let list_len = u16::from_be_bytes([handshake[pos], handshake[pos+1]]) as usize;
+            let mut lp = pos + 2;
+            let list_end = (pos + 2).saturating_add(list_len).min(pos + elen);
+            while lp < list_end {
+                let plen = handshake[lp] as usize;
+                lp += 1;
+                if lp + plen > list_end { break; }
+                out.push(handshake[lp..lp+plen].to_vec());
+                lp += plen;
+            }
+            break;
+        }
+        pos += elen;
+    }
+    out
+}
+
+
 /// Check if bytes look like a TLS ClientHello
 pub fn is_tls_handshake(first_bytes: &[u8]) -> bool {
     if first_bytes.len() < 3 {
@@ -410,7 +616,7 @@ pub fn parse_tls_record_header(header: &[u8; 5]) -> Option<(u8, u16)> {
 ///
 /// This is useful for testing that our ServerHello is well-formed.
 #[cfg(test)]
-fn validate_server_hello_structure(data: &[u8]) -> Result<()> {
+fn validate_server_hello_structure(data: &[u8]) -> Result<(), ProxyError> {
     if data.len() < 5 {
         return Err(ProxyError::InvalidTlsRecord {
             record_type: 0,
@@ -489,13 +695,25 @@ mod tests {
     
     #[test]
     fn test_gen_fake_x25519_key() {
-        let key1 = gen_fake_x25519_key();
-        let key2 = gen_fake_x25519_key();
+        let rng = SecureRandom::new();
+        let key1 = gen_fake_x25519_key(&rng);
+        let key2 = gen_fake_x25519_key(&rng);
         
         assert_eq!(key1.len(), 32);
         assert_eq!(key2.len(), 32);
         assert_ne!(key1, key2); // Should be random
     }
+
+    #[test]
+    fn test_fake_x25519_key_is_quadratic_residue() {
+        let rng = SecureRandom::new();
+        let key = gen_fake_x25519_key(&rng);
+        let p = curve25519_prime();
+        let k_num = BigUint::from_bytes_le(&key);
+        let exponent = (&p - BigUint::one()) >> 1;
+        let legendre = k_num.modpow(&exponent, &p);
+        assert_eq!(legendre, BigUint::one());
+    }
     
     #[test]
     fn test_tls_extension_builder() {
@@ -545,7 +763,8 @@ mod tests {
         let client_digest = [0x42u8; 32];
         let session_id = vec![0xAA; 32];
         
-        let response = build_server_hello(secret, &client_digest, &session_id, 2048);
+        let rng = SecureRandom::new();
+        let response = build_server_hello(secret, &client_digest, &session_id, 2048, &rng, None, 0);
         
         // Should have at least 3 records
         assert!(response.len() > 100);
@@ -577,8 +796,9 @@ mod tests {
         let client_digest = [0x42u8; 32];
         let session_id = vec![0xAA; 32];
         
-        let response1 = build_server_hello(secret, &client_digest, &session_id, 1024);
-        let response2 = build_server_hello(secret, &client_digest, &session_id, 1024);
+        let rng = SecureRandom::new();
+        let response1 = build_server_hello(secret, &client_digest, &session_id, 1024, &rng, None, 0);
+        let response2 = build_server_hello(secret, &client_digest, &session_id, 1024, &rng, None, 0);
         
         // Digest position should have non-zero data
         let digest1 = &response1[TLS_DIGEST_POS..TLS_DIGEST_POS + TLS_DIGEST_LEN];
@@ -637,4 +857,101 @@ mod tests {
         // Should return None (no match) but not panic
         assert!(result.is_none());
     }
-}
+
+    fn build_client_hello_with_exts(exts: Vec<(u16, Vec<u8>)>, host: &str) -> Vec<u8> {
+        let mut body = Vec::new();
+        body.extend_from_slice(&TLS_VERSION); // legacy version
+        body.extend_from_slice(&[0u8; 32]); // random
+        body.push(0); // session id len
+        body.extend_from_slice(&2u16.to_be_bytes()); // cipher suites len
+        body.extend_from_slice(&[0x13, 0x01]); // TLS_AES_128_GCM_SHA256
+        body.push(1); // compression len
+        body.push(0); // null compression
+
+        // Build SNI extension
+        let host_bytes = host.as_bytes();
+        let mut sni_ext = Vec::new();
+        sni_ext.extend_from_slice(&(host_bytes.len() as u16 + 3).to_be_bytes());
+        sni_ext.push(0);
+        sni_ext.extend_from_slice(&(host_bytes.len() as u16).to_be_bytes());
+        sni_ext.extend_from_slice(host_bytes);
+
+        let mut ext_blob = Vec::new();
+        for (typ, data) in exts {
+            ext_blob.extend_from_slice(&typ.to_be_bytes());
+            ext_blob.extend_from_slice(&(data.len() as u16).to_be_bytes());
+            ext_blob.extend_from_slice(&data);
+        }
+        // SNI last
+        ext_blob.extend_from_slice(&0x0000u16.to_be_bytes());
+        ext_blob.extend_from_slice(&(sni_ext.len() as u16).to_be_bytes());
+        ext_blob.extend_from_slice(&sni_ext);
+
+        body.extend_from_slice(&(ext_blob.len() as u16).to_be_bytes());
+        body.extend_from_slice(&ext_blob);
+
+        let mut handshake = Vec::new();
+        handshake.push(0x01); // ClientHello
+        let len_bytes = (body.len() as u32).to_be_bytes();
+        handshake.extend_from_slice(&len_bytes[1..4]);
+        handshake.extend_from_slice(&body);
+
+        let mut record = Vec::new();
+        record.push(TLS_RECORD_HANDSHAKE);
+        record.extend_from_slice(&[0x03, 0x01]);
+        record.extend_from_slice(&(handshake.len() as u16).to_be_bytes());
+        record.extend_from_slice(&handshake);
+        record
+    }
+
+    #[test]
+    fn test_extract_sni_with_grease_extension() {
+        // GREASE type 0x0a0a with zero length before SNI
+        let ch = build_client_hello_with_exts(vec![(0x0a0a, Vec::new())], "example.com");
+        let sni = extract_sni_from_client_hello(&ch);
+        assert_eq!(sni.as_deref(), Some("example.com"));
+    }
+
+    #[test]
+    fn test_extract_sni_tolerates_empty_unknown_extension() {
+        let ch = build_client_hello_with_exts(vec![(0x1234, Vec::new())], "test.local");
+        let sni = extract_sni_from_client_hello(&ch);
+        assert_eq!(sni.as_deref(), Some("test.local"));
+    }
+
+    #[test]
+    fn test_extract_alpn_single() {
+        let mut alpn_data = Vec::new();
+        // list length = 3 (1 length byte + "h2")
+        alpn_data.extend_from_slice(&3u16.to_be_bytes());
+        alpn_data.push(2);
+        alpn_data.extend_from_slice(b"h2");
+        let ch = build_client_hello_with_exts(vec![(0x0010, alpn_data)], "alpn.test");
+        let alpn = extract_alpn_from_client_hello(&ch);
+        let alpn_str: Vec<String> = alpn
+            .iter()
+            .map(|p| std::str::from_utf8(p).unwrap().to_string())
+            .collect();
+        assert_eq!(alpn_str, vec!["h2"]);
+    }
+
+    #[test]
+    fn test_extract_alpn_multiple() {
+        let mut alpn_data = Vec::new();
+        // list length = 11 (sum of per-proto lengths including length bytes)
+        alpn_data.extend_from_slice(&11u16.to_be_bytes());
+        alpn_data.push(2);
+        alpn_data.extend_from_slice(b"h2");
+        alpn_data.push(4);
+        alpn_data.extend_from_slice(b"spdy");
+        alpn_data.push(2);
+        alpn_data.extend_from_slice(b"h3");
+        let ch = build_client_hello_with_exts(vec![(0x0010, alpn_data)], "alpn.test");
+        let alpn = extract_alpn_from_client_hello(&ch);
+        let alpn_str: Vec<String> = alpn
+            .iter()
+            .map(|p| std::str::from_utf8(p).unwrap().to_string())
+            .collect();
+        assert_eq!(alpn_str, vec!["h2", "spdy", "h3"]);
+    }
+}

src/proxy/direct_relay.rs

@@ -0,0 +1,192 @@
+use std::fs::OpenOptions;
+use std::io::Write;
+use std::net::SocketAddr;
+use std::sync::Arc;
+
+use tokio::io::{AsyncRead, AsyncWrite, AsyncWriteExt};
+use tokio::net::TcpStream;
+use tracing::{debug, info, warn};
+
+use crate::config::ProxyConfig;
+use crate::crypto::SecureRandom;
+use crate::error::Result;
+use crate::protocol::constants::*;
+use crate::proxy::handshake::{HandshakeSuccess, encrypt_tg_nonce_with_ciphers, generate_tg_nonce};
+use crate::proxy::relay::relay_bidirectional;
+use crate::stats::Stats;
+use crate::stream::{BufferPool, CryptoReader, CryptoWriter};
+use crate::transport::UpstreamManager;
+
+pub(crate) async fn handle_via_direct<R, W>(
+    client_reader: CryptoReader<R>,
+    client_writer: CryptoWriter<W>,
+    success: HandshakeSuccess,
+    upstream_manager: Arc<UpstreamManager>,
+    stats: Arc<Stats>,
+    config: Arc<ProxyConfig>,
+    buffer_pool: Arc<BufferPool>,
+    rng: Arc<SecureRandom>,
+) -> Result<()>
+where
+    R: AsyncRead + Unpin + Send + 'static,
+    W: AsyncWrite + Unpin + Send + 'static,
+{
+    let user = &success.user;
+    let dc_addr = get_dc_addr_static(success.dc_idx, &config)?;
+
+    info!(
+        user = %user,
+        peer = %success.peer,
+        dc = success.dc_idx,
+        dc_addr = %dc_addr,
+        proto = ?success.proto_tag,
+        mode = "direct",
+        "Connecting to Telegram DC"
+    );
+
+    let tg_stream = upstream_manager
+        .connect(dc_addr, Some(success.dc_idx), user.strip_prefix("scope_").filter(|s| !s.is_empty()))
+        .await?;
+
+    debug!(peer = %success.peer, dc_addr = %dc_addr, "Connected, performing TG handshake");
+
+    let (tg_reader, tg_writer) =
+        do_tg_handshake_static(tg_stream, &success, &config, rng.as_ref()).await?;
+
+    debug!(peer = %success.peer, "TG handshake complete, starting relay");
+
+    stats.increment_user_connects(user);
+    stats.increment_user_curr_connects(user);
+
+    let relay_result = relay_bidirectional(
+        client_reader,
+        client_writer,
+        tg_reader,
+        tg_writer,
+        user,
+        Arc::clone(&stats),
+        buffer_pool,
+    )
+    .await;
+
+    stats.decrement_user_curr_connects(user);
+
+    match &relay_result {
+        Ok(()) => debug!(user = %user, "Direct relay completed"),
+        Err(e) => debug!(user = %user, error = %e, "Direct relay ended with error"),
+    }
+
+    relay_result
+}
+
+fn get_dc_addr_static(dc_idx: i16, config: &ProxyConfig) -> Result<SocketAddr> {
+    let prefer_v6 = config.network.prefer == 6 && config.network.ipv6.unwrap_or(true);
+    let datacenters = if prefer_v6 {
+        &*TG_DATACENTERS_V6
+    } else {
+        &*TG_DATACENTERS_V4
+    };
+
+    let num_dcs = datacenters.len();
+
+    let dc_key = dc_idx.to_string();
+    if let Some(addrs) = config.dc_overrides.get(&dc_key) {
+        let mut parsed = Vec::new();
+        for addr_str in addrs {
+            match addr_str.parse::<SocketAddr>() {
+                Ok(addr) => parsed.push(addr),
+                Err(_) => warn!(dc_idx = dc_idx, addr_str = %addr_str, "Invalid DC override address in config, ignoring"),
+            }
+        }
+
+        if let Some(addr) = parsed
+            .iter()
+            .find(|a| a.is_ipv6() == prefer_v6)
+            .or_else(|| parsed.first())
+            .copied()
+        {
+            debug!(dc_idx = dc_idx, addr = %addr, count = parsed.len(), "Using DC override from config");
+            return Ok(addr);
+        }
+    }
+
+    let abs_dc = dc_idx.unsigned_abs() as usize;
+    if abs_dc >= 1 && abs_dc <= num_dcs {
+        return Ok(SocketAddr::new(datacenters[abs_dc - 1], TG_DATACENTER_PORT));
+    }
+
+    // Unknown DC requested by client without override: log and fall back.
+    if !config.dc_overrides.contains_key(&dc_key) {
+        warn!(dc_idx = dc_idx, "Requested non-standard DC with no override; falling back to default cluster");
+        if config.general.unknown_dc_file_log_enabled
+            && let Some(path) = &config.general.unknown_dc_log_path
+            && let Ok(handle) = tokio::runtime::Handle::try_current()
+        {
+            let path = path.clone();
+            handle.spawn_blocking(move || {
+                if let Ok(mut file) = OpenOptions::new().create(true).append(true).open(path) {
+                    let _ = writeln!(file, "dc_idx={dc_idx}");
+                }
+            });
+        }
+    }
+
+    let default_dc = config.default_dc.unwrap_or(2) as usize;
+    let fallback_idx = if default_dc >= 1 && default_dc <= num_dcs {
+        default_dc - 1
+    } else {
+        1
+    };
+
+    info!(
+        original_dc = dc_idx,
+        fallback_dc = (fallback_idx + 1) as u16,
+        fallback_addr = %datacenters[fallback_idx],
+        "Special DC ---> default_cluster"
+    );
+
+    Ok(SocketAddr::new(
+        datacenters[fallback_idx],
+        TG_DATACENTER_PORT,
+    ))
+}
+
+async fn do_tg_handshake_static(
+    mut stream: TcpStream,
+    success: &HandshakeSuccess,
+    config: &ProxyConfig,
+    rng: &SecureRandom,
+) -> Result<(
+    CryptoReader<tokio::net::tcp::OwnedReadHalf>,
+    CryptoWriter<tokio::net::tcp::OwnedWriteHalf>,
+)> {
+    let (nonce, _tg_enc_key, _tg_enc_iv, _tg_dec_key, _tg_dec_iv) = generate_tg_nonce(
+        success.proto_tag,
+        success.dc_idx,
+        &success.dec_key,
+        success.dec_iv,
+        &success.enc_key,
+        success.enc_iv,
+        rng,
+        config.general.fast_mode,
+    );
+
+    let (encrypted_nonce, tg_encryptor, tg_decryptor) = encrypt_tg_nonce_with_ciphers(&nonce);
+
+    debug!(
+        peer = %success.peer,
+        nonce_head = %hex::encode(&nonce[..16]),
+        "Sending nonce to Telegram"
+    );
+
+    stream.write_all(&encrypted_nonce).await?;
+    stream.flush().await?;
+
+    let (read_half, write_half) = stream.into_split();
+
+    let max_pending = config.general.crypto_pending_buffer;
+    Ok((
+        CryptoReader::new(read_half, tg_decryptor),
+        CryptoWriter::new(write_half, tg_encryptor, max_pending),
+    ))
+}

src/proxy/handshake.rs

@@ -1,19 +1,28 @@
-//! MTProto Handshake Magics
+//! MTProto Handshake
+
+#![allow(dead_code)]
 
 use std::net::SocketAddr;
+use std::sync::Arc;
+use std::time::Duration;
 use tokio::io::{AsyncRead, AsyncWrite, AsyncWriteExt};
 use tracing::{debug, warn, trace, info};
+use zeroize::Zeroize;
 
-use crate::crypto::{sha256, AesCtr};
-use crate::crypto::random::SECURE_RANDOM;
+use crate::crypto::{sha256, AesCtr, SecureRandom};
+use rand::Rng;
 use crate::protocol::constants::*;
 use crate::protocol::tls;
 use crate::stream::{FakeTlsReader, FakeTlsWriter, CryptoReader, CryptoWriter};
 use crate::error::{ProxyError, HandshakeResult};
 use crate::stats::ReplayChecker;
 use crate::config::ProxyConfig;
+use crate::tls_front::{TlsFrontCache, emulator};
 
 /// Result of successful handshake
+///
+/// Key material (`dec_key`, `dec_iv`, `enc_key`, `enc_iv`) is
+/// zeroized on drop.
 #[derive(Debug, Clone)]
 pub struct HandshakeSuccess {
     /// Authenticated user name
@@ -34,6 +43,15 @@ pub struct HandshakeSuccess {
     pub is_tls: bool,
 }
 
+impl Drop for HandshakeSuccess {
+    fn drop(&mut self) {
+        self.dec_key.zeroize();
+        self.dec_iv.zeroize();
+        self.enc_key.zeroize();
+        self.enc_iv.zeroize();
+    }
+}
+
 /// Handle fake TLS handshake
 pub async fn handle_tls_handshake<R, W>(
     handshake: &[u8],
@@ -42,84 +60,153 @@ pub async fn handle_tls_handshake<R, W>(
     peer: SocketAddr,
     config: &ProxyConfig,
     replay_checker: &ReplayChecker,
-) -> HandshakeResult<(FakeTlsReader<R>, FakeTlsWriter<W>, String)>
+    rng: &SecureRandom,
+    tls_cache: Option<Arc<TlsFrontCache>>,
+) -> HandshakeResult<(FakeTlsReader<R>, FakeTlsWriter<W>, String), R, W>
 where
     R: AsyncRead + Unpin,
     W: AsyncWrite + Unpin,
 {
     debug!(peer = %peer, handshake_len = handshake.len(), "Processing TLS handshake");
-    
-    // Check minimum length
+
     if handshake.len() < tls::TLS_DIGEST_POS + tls::TLS_DIGEST_LEN + 1 {
         debug!(peer = %peer, "TLS handshake too short");
-        return HandshakeResult::BadClient;
+        return HandshakeResult::BadClient { reader, writer };
     }
-    
-    // Extract digest for replay check
+
     let digest = &handshake[tls::TLS_DIGEST_POS..tls::TLS_DIGEST_POS + tls::TLS_DIGEST_LEN];
     let digest_half = &digest[..tls::TLS_DIGEST_HALF_LEN];
-    
-    // Check for replay
-    if replay_checker.check_tls_digest(digest_half) {
-        warn!(peer = %peer, "TLS replay attack detected");
-        return HandshakeResult::BadClient;
+
+    if replay_checker.check_and_add_tls_digest(digest_half) {
+        warn!(peer = %peer, "TLS replay attack detected (duplicate digest)");
+        return HandshakeResult::BadClient { reader, writer };
     }
-    
-    // Build secrets list
-    let secrets: Vec<(String, Vec<u8>)> = config.users.iter()
+
+    let secrets: Vec<(String, Vec<u8>)> = config.access.users.iter()
         .filter_map(|(name, hex)| {
             hex::decode(hex).ok().map(|bytes| (name.clone(), bytes))
         })
         .collect();
-    
-    debug!(peer = %peer, num_users = secrets.len(), "Validating TLS handshake against users");
-    
-    // Validate handshake
+
     let validation = match tls::validate_tls_handshake(
         handshake,
         &secrets,
-        config.ignore_time_skew,
+        config.access.ignore_time_skew,
     ) {
         Some(v) => v,
         None => {
-            debug!(peer = %peer, "TLS handshake validation failed - no matching user");
-            return HandshakeResult::BadClient;
+            debug!(
+                peer = %peer, 
+                ignore_time_skew = config.access.ignore_time_skew,
+                "TLS handshake validation failed - no matching user or time skew"
+            );
+            return HandshakeResult::BadClient { reader, writer };
         }
     };
-    
-    // Get secret for response
+
     let secret = match secrets.iter().find(|(name, _)| *name == validation.user) {
         Some((_, s)) => s,
-        None => return HandshakeResult::BadClient,
+        None => return HandshakeResult::BadClient { reader, writer },
     };
-    
-    // Build and send response
-    let response = tls::build_server_hello(
-        secret,
-        &validation.digest,
-        &validation.session_id,
-        config.fake_cert_len,
-    );
-    
+
+    let cached = if config.censorship.tls_emulation {
+        if let Some(cache) = tls_cache.as_ref() {
+            let selected_domain = if let Some(sni) = tls::extract_sni_from_client_hello(handshake) {
+                if cache.contains_domain(&sni).await {
+                    sni
+                } else {
+                    config.censorship.tls_domain.clone()
+                }
+            } else {
+                config.censorship.tls_domain.clone()
+            };
+            let cached_entry = cache.get(&selected_domain).await;
+            let use_full_cert_payload = cache
+                .take_full_cert_budget_for_ip(
+                    peer.ip(),
+                    Duration::from_secs(config.censorship.tls_full_cert_ttl_secs),
+                )
+                .await;
+            Some((cached_entry, use_full_cert_payload))
+        } else {
+            None
+        }
+    } else {
+        None
+    };
+
+    let alpn_list = if config.censorship.alpn_enforce {
+        tls::extract_alpn_from_client_hello(handshake)
+    } else {
+        Vec::new()
+    };
+    let selected_alpn = if config.censorship.alpn_enforce {
+        if alpn_list.iter().any(|p| p == b"h2") {
+            Some(b"h2".to_vec())
+        } else if alpn_list.iter().any(|p| p == b"http/1.1") {
+            Some(b"http/1.1".to_vec())
+        } else {
+            None
+        }
+    } else {
+        None
+    };
+
+    let response = if let Some((cached_entry, use_full_cert_payload)) = cached {
+        emulator::build_emulated_server_hello(
+            secret,
+            &validation.digest,
+            &validation.session_id,
+            &cached_entry,
+            use_full_cert_payload,
+            rng,
+            selected_alpn.clone(),
+            config.censorship.tls_new_session_tickets,
+        )
+    } else {
+        tls::build_server_hello(
+            secret,
+            &validation.digest,
+            &validation.session_id,
+            config.censorship.fake_cert_len,
+            rng,
+            selected_alpn.clone(),
+            config.censorship.tls_new_session_tickets,
+        )
+    };
+
+    // Optional anti-fingerprint delay before sending ServerHello.
+    if config.censorship.server_hello_delay_max_ms > 0 {
+        let min = config.censorship.server_hello_delay_min_ms;
+        let max = config.censorship.server_hello_delay_max_ms.max(min);
+        let delay_ms = if max == min {
+            max
+        } else {
+            rand::rng().random_range(min..=max)
+        };
+        if delay_ms > 0 {
+            tokio::time::sleep(std::time::Duration::from_millis(delay_ms)).await;
+        }
+    }
+
     debug!(peer = %peer, response_len = response.len(), "Sending TLS ServerHello");
-    
+
     if let Err(e) = writer.write_all(&response).await {
+        warn!(peer = %peer, error = %e, "Failed to write TLS ServerHello");
         return HandshakeResult::Error(ProxyError::Io(e));
     }
-    
+
     if let Err(e) = writer.flush().await {
+        warn!(peer = %peer, error = %e, "Failed to flush TLS ServerHello");
         return HandshakeResult::Error(ProxyError::Io(e));
     }
-    
-    // Record for replay protection
-    replay_checker.add_tls_digest(digest_half);
-    
+
     info!(
         peer = %peer,
         user = %validation.user,
         "TLS handshake successful"
     );
-    
+
     HandshakeResult::Success((
         FakeTlsReader::new(reader),
         FakeTlsWriter::new(writer),
@@ -136,111 +223,82 @@ pub async fn handle_mtproto_handshake<R, W>(
     config: &ProxyConfig,
     replay_checker: &ReplayChecker,
     is_tls: bool,
-) -> HandshakeResult<(CryptoReader<R>, CryptoWriter<W>, HandshakeSuccess)>
+) -> HandshakeResult<(CryptoReader<R>, CryptoWriter<W>, HandshakeSuccess), R, W>
 where
     R: AsyncRead + Unpin + Send,
     W: AsyncWrite + Unpin + Send,
 {
     trace!(peer = %peer, handshake = ?hex::encode(handshake), "MTProto handshake bytes");
-    
-    // Extract prekey and IV
+
     let dec_prekey_iv = &handshake[SKIP_LEN..SKIP_LEN + PREKEY_LEN + IV_LEN];
-    
-    debug!(
-        peer = %peer,
-        dec_prekey_iv = %hex::encode(dec_prekey_iv),
-        "Extracted prekey+IV from handshake"
-    );
-    
-    // Check for replay
-    if replay_checker.check_handshake(dec_prekey_iv) {
+
+    if replay_checker.check_and_add_handshake(dec_prekey_iv) {
         warn!(peer = %peer, "MTProto replay attack detected");
-        return HandshakeResult::BadClient;
+        return HandshakeResult::BadClient { reader, writer };
     }
-    
-    // Reversed for encryption direction
+
     let enc_prekey_iv: Vec<u8> = dec_prekey_iv.iter().rev().copied().collect();
-    
-    // Try each user's secret
-    for (user, secret_hex) in &config.users {
+
+    for (user, secret_hex) in &config.access.users {
         let secret = match hex::decode(secret_hex) {
             Ok(s) => s,
             Err(_) => continue,
         };
-        
-        // Derive decryption key
+
         let dec_prekey = &dec_prekey_iv[..PREKEY_LEN];
         let dec_iv_bytes = &dec_prekey_iv[PREKEY_LEN..];
-        
+
         let mut dec_key_input = Vec::with_capacity(PREKEY_LEN + secret.len());
         dec_key_input.extend_from_slice(dec_prekey);
         dec_key_input.extend_from_slice(&secret);
         let dec_key = sha256(&dec_key_input);
-        
+
         let dec_iv = u128::from_be_bytes(dec_iv_bytes.try_into().unwrap());
-        
-        // Decrypt handshake to check protocol tag
+
         let mut decryptor = AesCtr::new(&dec_key, dec_iv);
         let decrypted = decryptor.decrypt(handshake);
-        
-        trace!(
-            peer = %peer,
-            user = %user,
-            decrypted_tail = %hex::encode(&decrypted[PROTO_TAG_POS..]),
-            "Decrypted handshake tail"
-        );
-        
-        // Check protocol tag
+
         let tag_bytes: [u8; 4] = decrypted[PROTO_TAG_POS..PROTO_TAG_POS + 4]
             .try_into()
             .unwrap();
-        
+
         let proto_tag = match ProtoTag::from_bytes(tag_bytes) {
             Some(tag) => tag,
-            None => {
-                trace!(peer = %peer, user = %user, tag = %hex::encode(tag_bytes), "Invalid proto tag");
-                continue;
-            }
+            None => continue,
         };
-        
-        debug!(peer = %peer, user = %user, proto = ?proto_tag, "Found valid proto tag");
-        
-        // Check if mode is enabled
+
         let mode_ok = match proto_tag {
             ProtoTag::Secure => {
-                if is_tls { config.modes.tls } else { config.modes.secure }
+                if is_tls {
+                    config.general.modes.tls || config.general.modes.secure
+                } else {
+                    config.general.modes.secure || config.general.modes.tls
+                }
             }
-            ProtoTag::Intermediate | ProtoTag::Abridged => config.modes.classic,
+            ProtoTag::Intermediate | ProtoTag::Abridged => config.general.modes.classic,
         };
-        
+
         if !mode_ok {
             debug!(peer = %peer, user = %user, proto = ?proto_tag, "Mode not enabled");
             continue;
         }
-        
-        // Extract DC index
+
         let dc_idx = i16::from_le_bytes(
             decrypted[DC_IDX_POS..DC_IDX_POS + 2].try_into().unwrap()
         );
-        
-        // Derive encryption key
+
         let enc_prekey = &enc_prekey_iv[..PREKEY_LEN];
         let enc_iv_bytes = &enc_prekey_iv[PREKEY_LEN..];
-        
+
         let mut enc_key_input = Vec::with_capacity(PREKEY_LEN + secret.len());
         enc_key_input.extend_from_slice(enc_prekey);
         enc_key_input.extend_from_slice(&secret);
         let enc_key = sha256(&enc_key_input);
-        
+
         let enc_iv = u128::from_be_bytes(enc_iv_bytes.try_into().unwrap());
-        
-        // Record for replay protection
-        replay_checker.add_handshake(dec_prekey_iv);
-        
-        // Create new cipher instances
-        let decryptor = AesCtr::new(&dec_key, dec_iv);
+
         let encryptor = AesCtr::new(&enc_key, enc_iv);
-        
+
         let success = HandshakeSuccess {
             user: user.clone(),
             dc_idx,
@@ -252,7 +310,7 @@ where
             peer,
             is_tls,
         };
-        
+
         info!(
             peer = %peer,
             user = %user,
@@ -261,151 +319,170 @@ where
             tls = is_tls,
             "MTProto handshake successful"
         );
-        
+
+        let max_pending = config.general.crypto_pending_buffer;
         return HandshakeResult::Success((
             CryptoReader::new(reader, decryptor),
-            CryptoWriter::new(writer, encryptor),
+            CryptoWriter::new(writer, encryptor, max_pending),
             success,
         ));
     }
-    
+
     debug!(peer = %peer, "MTProto handshake: no matching user found");
-    HandshakeResult::BadClient
+    HandshakeResult::BadClient { reader, writer }
 }
 
 /// Generate nonce for Telegram connection
-/// 
-/// In FAST MODE: we use the same keys for TG as for client, but reversed.
-/// This means: client's enc_key becomes TG's dec_key and vice versa.
 pub fn generate_tg_nonce(
     proto_tag: ProtoTag, 
-    client_dec_key: &[u8; 32],
-    client_dec_iv: u128,
+    dc_idx: i16,
+    _client_dec_key: &[u8; 32],
+    _client_dec_iv: u128,
+    client_enc_key: &[u8; 32],
+    client_enc_iv: u128,
+    rng: &SecureRandom,
     fast_mode: bool,
 ) -> ([u8; HANDSHAKE_LEN], [u8; 32], u128, [u8; 32], u128) {
     loop {
-        let bytes = SECURE_RANDOM.bytes(HANDSHAKE_LEN);
+        let bytes = rng.bytes(HANDSHAKE_LEN);
         let mut nonce: [u8; HANDSHAKE_LEN] = bytes.try_into().unwrap();
-        
-        // Check reserved patterns
-        if RESERVED_NONCE_FIRST_BYTES.contains(&nonce[0]) {
-            continue;
-        }
-        
+
+        if RESERVED_NONCE_FIRST_BYTES.contains(&nonce[0]) { continue; }
+
         let first_four: [u8; 4] = nonce[..4].try_into().unwrap();
-        if RESERVED_NONCE_BEGINNINGS.contains(&first_four) {
-            continue;
-        }
-        
+        if RESERVED_NONCE_BEGINNINGS.contains(&first_four) { continue; }
+
         let continue_four: [u8; 4] = nonce[4..8].try_into().unwrap();
-        if RESERVED_NONCE_CONTINUES.contains(&continue_four) {
-            continue;
-        }
-        
-        // Set protocol tag
+        if RESERVED_NONCE_CONTINUES.contains(&continue_four) { continue; }
+
         nonce[PROTO_TAG_POS..PROTO_TAG_POS + 4].copy_from_slice(&proto_tag.to_bytes());
-        
-        // Fast mode: copy client's dec_key+iv (this becomes TG's enc direction)
-        // In fast mode, we make TG use the same keys as client but swapped:
-        // - What we decrypt FROM TG = what we encrypt TO client (so no re-encryption needed)
-        // - What we encrypt TO TG = what we decrypt FROM client
+        // CRITICAL: write dc_idx so upstream DC knows where to route
+        nonce[DC_IDX_POS..DC_IDX_POS + 2].copy_from_slice(&dc_idx.to_le_bytes());
+
         if fast_mode {
-            // Put client's dec_key + dec_iv into nonce[8:56]
-            // This will be used by TG for encryption TO us
-            nonce[SKIP_LEN..SKIP_LEN + KEY_LEN].copy_from_slice(client_dec_key);
-            nonce[SKIP_LEN + KEY_LEN..SKIP_LEN + KEY_LEN + IV_LEN]
-                .copy_from_slice(&client_dec_iv.to_be_bytes());
+            let mut key_iv = Vec::with_capacity(KEY_LEN + IV_LEN);
+            key_iv.extend_from_slice(client_enc_key);
+            key_iv.extend_from_slice(&client_enc_iv.to_be_bytes());
+            key_iv.reverse(); // Python/C behavior: reversed enc_key+enc_iv in nonce
+            nonce[SKIP_LEN..SKIP_LEN + KEY_LEN + IV_LEN].copy_from_slice(&key_iv);
         }
-        
-        // Now compute what keys WE will use for TG connection
-        // enc_key_iv = nonce[8:56] (for encrypting TO TG)
-        // dec_key_iv = nonce[8:56] reversed (for decrypting FROM TG)
+
         let enc_key_iv = &nonce[SKIP_LEN..SKIP_LEN + KEY_LEN + IV_LEN];
         let dec_key_iv: Vec<u8> = enc_key_iv.iter().rev().copied().collect();
-        
+
         let tg_enc_key: [u8; 32] = enc_key_iv[..KEY_LEN].try_into().unwrap();
         let tg_enc_iv = u128::from_be_bytes(enc_key_iv[KEY_LEN..].try_into().unwrap());
-        
+
         let tg_dec_key: [u8; 32] = dec_key_iv[..KEY_LEN].try_into().unwrap();
         let tg_dec_iv = u128::from_be_bytes(dec_key_iv[KEY_LEN..].try_into().unwrap());
-        
-        debug!(
-            fast_mode = fast_mode,
-            tg_enc_key = %hex::encode(&tg_enc_key[..8]),
-            tg_dec_key = %hex::encode(&tg_dec_key[..8]),
-            "Generated TG nonce"
-        );
-        
+
         return (nonce, tg_enc_key, tg_enc_iv, tg_dec_key, tg_dec_iv);
     }
 }
 
-/// Encrypt nonce for sending to Telegram
-/// 
-/// Only the part from PROTO_TAG_POS onwards is encrypted.
-/// The encryption key is derived from enc_key_iv in the nonce itself.
-pub fn encrypt_tg_nonce(nonce: &[u8; HANDSHAKE_LEN]) -> Vec<u8> {
-    // enc_key_iv is at nonce[8:56]
+/// Encrypt nonce for sending to Telegram and return cipher objects with correct counter state
+pub fn encrypt_tg_nonce_with_ciphers(nonce: &[u8; HANDSHAKE_LEN]) -> (Vec<u8>, AesCtr, AesCtr) {
     let enc_key_iv = &nonce[SKIP_LEN..SKIP_LEN + KEY_LEN + IV_LEN];
-    
-    // Key for encrypting is just the first 32 bytes of enc_key_iv
-    let key: [u8; 32] = enc_key_iv[..KEY_LEN].try_into().unwrap();
-    let iv = u128::from_be_bytes(enc_key_iv[KEY_LEN..].try_into().unwrap());
-    
-    let mut encryptor = AesCtr::new(&key, iv);
-    
-    // Encrypt the entire nonce first, then take only the encrypted tail
-    let encrypted_full = encryptor.encrypt(nonce);
-    
-    // Result: unencrypted head + encrypted tail
+    let dec_key_iv: Vec<u8> = enc_key_iv.iter().rev().copied().collect();
+
+    let enc_key: [u8; 32] = enc_key_iv[..KEY_LEN].try_into().unwrap();
+    let enc_iv = u128::from_be_bytes(enc_key_iv[KEY_LEN..].try_into().unwrap());
+
+    let dec_key: [u8; 32] = dec_key_iv[..KEY_LEN].try_into().unwrap();
+    let dec_iv = u128::from_be_bytes(dec_key_iv[KEY_LEN..].try_into().unwrap());
+
+    let mut encryptor = AesCtr::new(&enc_key, enc_iv);
+    let encrypted_full = encryptor.encrypt(nonce);  // counter: 0 → 4
+
     let mut result = nonce[..PROTO_TAG_POS].to_vec();
     result.extend_from_slice(&encrypted_full[PROTO_TAG_POS..]);
-    
-    trace!(
-        original = %hex::encode(&nonce[PROTO_TAG_POS..]),
-        encrypted = %hex::encode(&result[PROTO_TAG_POS..]),
-        "Encrypted nonce tail"
-    );
-    
-    result
+
+    let decryptor = AesCtr::new(&dec_key, dec_iv);
+
+    (result, encryptor, decryptor)
+}
+
+/// Encrypt nonce for sending to Telegram (legacy function for compatibility)
+pub fn encrypt_tg_nonce(nonce: &[u8; HANDSHAKE_LEN]) -> Vec<u8> {
+    let (encrypted, _, _) = encrypt_tg_nonce_with_ciphers(nonce);
+    encrypted
 }
 
 #[cfg(test)]
 mod tests {
     use super::*;
-    
+
     #[test]
     fn test_generate_tg_nonce() {
         let client_dec_key = [0x42u8; 32];
         let client_dec_iv = 12345u128;
-        
-        let (nonce, tg_enc_key, tg_enc_iv, tg_dec_key, tg_dec_iv) = 
-            generate_tg_nonce(ProtoTag::Secure, &client_dec_key, client_dec_iv, false);
-        
-        // Check length
+        let client_enc_key = [0x24u8; 32];
+        let client_enc_iv = 54321u128;
+
+        let rng = SecureRandom::new();
+        let (nonce, _tg_enc_key, _tg_enc_iv, _tg_dec_key, _tg_dec_iv) = 
+            generate_tg_nonce(
+                ProtoTag::Secure,
+                2,
+                &client_dec_key,
+                client_dec_iv,
+                &client_enc_key,
+                client_enc_iv,
+                &rng,
+                false,
+            );
+
         assert_eq!(nonce.len(), HANDSHAKE_LEN);
-        
-        // Check proto tag is set
+
         let tag_bytes: [u8; 4] = nonce[PROTO_TAG_POS..PROTO_TAG_POS + 4].try_into().unwrap();
         assert_eq!(ProtoTag::from_bytes(tag_bytes), Some(ProtoTag::Secure));
     }
-    
+
     #[test]
     fn test_encrypt_tg_nonce() {
         let client_dec_key = [0x42u8; 32];
         let client_dec_iv = 12345u128;
-        
+        let client_enc_key = [0x24u8; 32];
+        let client_enc_iv = 54321u128;
+
+        let rng = SecureRandom::new();
         let (nonce, _, _, _, _) = 
-            generate_tg_nonce(ProtoTag::Secure, &client_dec_key, client_dec_iv, false);
-        
+            generate_tg_nonce(
+                ProtoTag::Secure,
+                2,
+                &client_dec_key,
+                client_dec_iv,
+                &client_enc_key,
+                client_enc_iv,
+                &rng,
+                false,
+            );
+
         let encrypted = encrypt_tg_nonce(&nonce);
-        
+
         assert_eq!(encrypted.len(), HANDSHAKE_LEN);
-        
-        // First PROTO_TAG_POS bytes should be unchanged
         assert_eq!(&encrypted[..PROTO_TAG_POS], &nonce[..PROTO_TAG_POS]);
-        
-        // Rest should be different (encrypted)
         assert_ne!(&encrypted[PROTO_TAG_POS..], &nonce[PROTO_TAG_POS..]);
     }
-}
+
+    #[test]
+    fn test_handshake_success_zeroize_on_drop() {
+        let success = HandshakeSuccess {
+            user: "test".to_string(),
+            dc_idx: 2,
+            proto_tag: ProtoTag::Secure,
+            dec_key: [0xAA; 32],
+            dec_iv: 0xBBBBBBBB,
+            enc_key: [0xCC; 32],
+            enc_iv: 0xDDDDDDDD,
+            peer: "127.0.0.1:1234".parse().unwrap(),
+            is_tls: true,
+        };
+
+        assert_eq!(success.dec_key, [0xAA; 32]);
+        assert_eq!(success.enc_key, [0xCC; 32]);
+
+        drop(success);
+        // Drop impl zeroizes key material without panic
+    }
+}

src/proxy/masking.rs

@@ -1,72 +1,214 @@
 //! Masking - forward unrecognized traffic to mask host
 
+use std::str;
+use std::net::SocketAddr;
 use std::time::Duration;
 use tokio::net::TcpStream;
-use tokio::io::{AsyncReadExt, AsyncWriteExt};
+#[cfg(unix)]
+use tokio::net::UnixStream;
+use tokio::io::{AsyncRead, AsyncWrite, AsyncReadExt, AsyncWriteExt};
 use tokio::time::timeout;
 use tracing::debug;
 use crate::config::ProxyConfig;
-use crate::transport::set_linger_zero;
+use crate::network::dns_overrides::resolve_socket_addr;
+use crate::stats::beobachten::BeobachtenStore;
+use crate::transport::proxy_protocol::{ProxyProtocolV1Builder, ProxyProtocolV2Builder};
 
 const MASK_TIMEOUT: Duration = Duration::from_secs(5);
+/// Maximum duration for the entire masking relay.
+/// Limits resource consumption from slow-loris attacks and port scanners.
+const MASK_RELAY_TIMEOUT: Duration = Duration::from_secs(60);
 const MASK_BUFFER_SIZE: usize = 8192;
 
+/// Detect client type based on initial data
+fn detect_client_type(data: &[u8]) -> &'static str {
+    // Check for HTTP request
+    if data.len() > 4
+        && (data.starts_with(b"GET ") || data.starts_with(b"POST") ||
+           data.starts_with(b"HEAD") || data.starts_with(b"PUT ") ||
+           data.starts_with(b"DELETE") || data.starts_with(b"OPTIONS"))
+    {
+        return "HTTP";
+    }
+
+    // Check for TLS ClientHello (0x16 = handshake, 0x03 0x01-0x03 = TLS version)
+    if data.len() > 3 && data[0] == 0x16 && data[1] == 0x03 {
+        return "TLS-scanner";
+    }
+
+    // Check for SSH
+    if data.starts_with(b"SSH-") {
+        return "SSH";
+    }
+
+    // Port scanner (very short data)
+    if data.len() < 10 {
+        return "port-scanner";
+    }
+
+    "unknown"
+}
+
 /// Handle a bad client by forwarding to mask host
-pub async fn handle_bad_client(
-    client: TcpStream,
+pub async fn handle_bad_client<R, W>(
+    reader: R,
+    writer: W,
     initial_data: &[u8],
+    peer: SocketAddr,
+    local_addr: SocketAddr,
     config: &ProxyConfig,
-) {
-    if !config.mask {
+    beobachten: &BeobachtenStore,
+)
+where
+    R: AsyncRead + Unpin + Send + 'static,
+    W: AsyncWrite + Unpin + Send + 'static,
+{
+    let client_type = detect_client_type(initial_data);
+    if config.general.beobachten {
+        let ttl = Duration::from_secs(config.general.beobachten_minutes.saturating_mul(60));
+        beobachten.record(client_type, peer.ip(), ttl);
+    }
+
+    if !config.censorship.mask {
         // Masking disabled, just consume data
-        consume_client_data(client).await;
+        consume_client_data(reader).await;
         return;
     }
-    
-    let mask_host = config.mask_host.as_deref()
-        .unwrap_or(&config.tls_domain);
-    let mask_port = config.mask_port;
-    
+
+    // Connect via Unix socket or TCP
+    #[cfg(unix)]
+    if let Some(ref sock_path) = config.censorship.mask_unix_sock {
+        debug!(
+            client_type = client_type,
+            sock = %sock_path,
+            data_len = initial_data.len(),
+            "Forwarding bad client to mask unix socket"
+        );
+
+        let connect_result = timeout(MASK_TIMEOUT, UnixStream::connect(sock_path)).await;
+        match connect_result {
+            Ok(Ok(stream)) => {
+                let (mask_read, mut mask_write) = stream.into_split();
+                let proxy_header: Option<Vec<u8>> = match config.censorship.mask_proxy_protocol {
+                    0 => None,
+                    version => {
+                        let header = match version {
+                            2 => ProxyProtocolV2Builder::new().with_addrs(peer, local_addr).build(),
+                            _ => match (peer, local_addr) {
+                                (SocketAddr::V4(src), SocketAddr::V4(dst)) =>
+                                    ProxyProtocolV1Builder::new().tcp4(src.into(), dst.into()).build(),
+                                (SocketAddr::V6(src), SocketAddr::V6(dst)) =>
+                                    ProxyProtocolV1Builder::new().tcp6(src.into(), dst.into()).build(),
+                                _ =>
+                                    ProxyProtocolV1Builder::new().build(),
+                            },
+                        };
+                        Some(header)
+                    }
+                };
+                if let Some(header) = proxy_header {
+                    if mask_write.write_all(&header).await.is_err() {
+                        return;
+                    }
+                }
+                if timeout(MASK_RELAY_TIMEOUT, relay_to_mask(reader, writer, mask_read, mask_write, initial_data)).await.is_err() {
+                    debug!("Mask relay timed out (unix socket)");
+                }
+            }
+            Ok(Err(e)) => {
+                debug!(error = %e, "Failed to connect to mask unix socket");
+                consume_client_data(reader).await;
+            }
+            Err(_) => {
+                debug!("Timeout connecting to mask unix socket");
+                consume_client_data(reader).await;
+            }
+        }
+        return;
+    }
+
+    let mask_host = config.censorship.mask_host.as_deref()
+        .unwrap_or(&config.censorship.tls_domain);
+    let mask_port = config.censorship.mask_port;
+
     debug!(
+        client_type = client_type,
         host = %mask_host,
         port = mask_port,
+        data_len = initial_data.len(),
         "Forwarding bad client to mask host"
     );
-    
-    // Connect to mask host
-    let mask_addr = format!("{}:{}", mask_host, mask_port);
-    let connect_result = timeout(
-        MASK_TIMEOUT,
-        TcpStream::connect(&mask_addr)
-    ).await;
-    
-    let mut mask_stream = match connect_result {
-        Ok(Ok(s)) => s,
+
+    // Apply runtime DNS override for mask target when configured.
+    let mask_addr = resolve_socket_addr(mask_host, mask_port)
+        .map(|addr| addr.to_string())
+        .unwrap_or_else(|| format!("{}:{}", mask_host, mask_port));
+    let connect_result = timeout(MASK_TIMEOUT, TcpStream::connect(&mask_addr)).await;
+    match connect_result {
+        Ok(Ok(stream)) => {
+            let proxy_header: Option<Vec<u8>> = match config.censorship.mask_proxy_protocol {
+                0 => None,
+                version => {
+                    let header = match version {
+                        2 => ProxyProtocolV2Builder::new().with_addrs(peer, local_addr).build(),
+                        _ => match (peer, local_addr) {
+                            (SocketAddr::V4(src), SocketAddr::V4(dst)) =>
+                                ProxyProtocolV1Builder::new().tcp4(src.into(), dst.into()).build(),
+                            (SocketAddr::V6(src), SocketAddr::V6(dst)) =>
+                                ProxyProtocolV1Builder::new().tcp6(src.into(), dst.into()).build(),
+                            _ =>
+                                ProxyProtocolV1Builder::new().build(),
+                        },
+                    };
+                    Some(header)
+                }
+            };
+
+            let (mask_read, mut mask_write) = stream.into_split();
+            if let Some(header) = proxy_header {
+                if mask_write.write_all(&header).await.is_err() {
+                    return;
+                }
+            }
+            if timeout(MASK_RELAY_TIMEOUT, relay_to_mask(reader, writer, mask_read, mask_write, initial_data)).await.is_err() {
+                debug!("Mask relay timed out");
+            }
+        }
         Ok(Err(e)) => {
             debug!(error = %e, "Failed to connect to mask host");
-            consume_client_data(client).await;
-            return;
+            consume_client_data(reader).await;
         }
         Err(_) => {
             debug!("Timeout connecting to mask host");
-            consume_client_data(client).await;
-            return;
+            consume_client_data(reader).await;
         }
-    };
-    
+    }
+}
+
+/// Relay traffic between client and mask backend
+async fn relay_to_mask<R, W, MR, MW>(
+    mut reader: R,
+    mut writer: W,
+    mut mask_read: MR,
+    mut mask_write: MW,
+    initial_data: &[u8],
+)
+where
+    R: AsyncRead + Unpin + Send + 'static,
+    W: AsyncWrite + Unpin + Send + 'static,
+    MR: AsyncRead + Unpin + Send + 'static,
+    MW: AsyncWrite + Unpin + Send + 'static,
+{
     // Send initial data to mask host
-    if mask_stream.write_all(initial_data).await.is_err() {
+    if mask_write.write_all(initial_data).await.is_err() {
         return;
     }
-    
+
     // Relay traffic
-    let (mut client_read, mut client_write) = client.into_split();
-    let (mut mask_read, mut mask_write) = mask_stream.into_split();
-    
     let c2m = tokio::spawn(async move {
         let mut buf = vec![0u8; MASK_BUFFER_SIZE];
         loop {
-            match client_read.read(&mut buf).await {
+            match reader.read(&mut buf).await {
                 Ok(0) | Err(_) => {
                     let _ = mask_write.shutdown().await;
                     break;
@@ -79,24 +221,24 @@ pub async fn handle_bad_client(
             }
         }
     });
-    
+
     let m2c = tokio::spawn(async move {
         let mut buf = vec![0u8; MASK_BUFFER_SIZE];
         loop {
             match mask_read.read(&mut buf).await {
                 Ok(0) | Err(_) => {
-                    let _ = client_write.shutdown().await;
+                    let _ = writer.shutdown().await;
                     break;
                 }
                 Ok(n) => {
-                    if client_write.write_all(&buf[..n]).await.is_err() {
+                    if writer.write_all(&buf[..n]).await.is_err() {
                         break;
                     }
                 }
             }
         }
     });
-    
+
     // Wait for either to complete
     tokio::select! {
         _ = c2m => {}
@@ -105,11 +247,11 @@ pub async fn handle_bad_client(
 }
 
 /// Just consume all data from client without responding
-async fn consume_client_data(mut client: TcpStream) {
+async fn consume_client_data<R: AsyncRead + Unpin>(mut reader: R) {
     let mut buf = vec![0u8; MASK_BUFFER_SIZE];
-    while let Ok(n) = client.read(&mut buf).await {
+    while let Ok(n) = reader.read(&mut buf).await {
         if n == 0 {
             break;
         }
     }
-}
+}

src/proxy/middle_relay.rs

@@ -0,0 +1,776 @@
+use std::collections::HashMap;
+use std::collections::hash_map::DefaultHasher;
+use std::hash::{Hash, Hasher};
+use std::net::{IpAddr, SocketAddr};
+use std::sync::atomic::{AtomicU64, Ordering};
+use std::sync::{Arc, Mutex, OnceLock};
+use std::time::{Duration, Instant};
+
+use tokio::io::{AsyncRead, AsyncReadExt, AsyncWrite, AsyncWriteExt};
+use tokio::sync::{mpsc, oneshot};
+use tracing::{debug, info, trace, warn};
+
+use crate::config::ProxyConfig;
+use crate::crypto::SecureRandom;
+use crate::error::{ProxyError, Result};
+use crate::protocol::constants::{*, secure_padding_len};
+use crate::proxy::handshake::HandshakeSuccess;
+use crate::stats::Stats;
+use crate::stream::{BufferPool, CryptoReader, CryptoWriter};
+use crate::transport::middle_proxy::{MePool, MeResponse, proto_flags_for_tag};
+
+enum C2MeCommand {
+    Data { payload: Vec<u8>, flags: u32 },
+    Close,
+}
+
+const DESYNC_DEDUP_WINDOW: Duration = Duration::from_secs(60);
+const DESYNC_ERROR_CLASS: &str = "frame_too_large_crypto_desync";
+const C2ME_CHANNEL_CAPACITY: usize = 1024;
+const C2ME_SOFT_PRESSURE_MIN_FREE_SLOTS: usize = 64;
+const C2ME_SENDER_FAIRNESS_BUDGET: usize = 32;
+static DESYNC_DEDUP: OnceLock<Mutex<HashMap<u64, Instant>>> = OnceLock::new();
+
+struct RelayForensicsState {
+    trace_id: u64,
+    conn_id: u64,
+    user: String,
+    peer: SocketAddr,
+    peer_hash: u64,
+    started_at: Instant,
+    bytes_c2me: u64,
+    bytes_me2c: Arc<AtomicU64>,
+    desync_all_full: bool,
+}
+
+fn hash_value<T: Hash>(value: &T) -> u64 {
+    let mut hasher = DefaultHasher::new();
+    value.hash(&mut hasher);
+    hasher.finish()
+}
+
+fn hash_ip(ip: IpAddr) -> u64 {
+    hash_value(&ip)
+}
+
+fn should_emit_full_desync(key: u64, all_full: bool, now: Instant) -> bool {
+    if all_full {
+        return true;
+    }
+
+    let dedup = DESYNC_DEDUP.get_or_init(|| Mutex::new(HashMap::new()));
+    let mut guard = dedup.lock().expect("desync dedup mutex poisoned");
+    guard.retain(|_, seen_at| now.duration_since(*seen_at) < DESYNC_DEDUP_WINDOW);
+
+    match guard.get_mut(&key) {
+        Some(seen_at) => {
+            if now.duration_since(*seen_at) >= DESYNC_DEDUP_WINDOW {
+                *seen_at = now;
+                true
+            } else {
+                false
+            }
+        }
+        None => {
+            guard.insert(key, now);
+            true
+        }
+    }
+}
+
+fn report_desync_frame_too_large(
+    state: &RelayForensicsState,
+    proto_tag: ProtoTag,
+    frame_counter: u64,
+    max_frame: usize,
+    len: usize,
+    raw_len_bytes: Option<[u8; 4]>,
+    stats: &Stats,
+) -> ProxyError {
+    let len_buf = raw_len_bytes.unwrap_or((len as u32).to_le_bytes());
+    let looks_like_tls = raw_len_bytes
+        .map(|b| b[0] == 0x16 && b[1] == 0x03)
+        .unwrap_or(false);
+    let looks_like_http = raw_len_bytes
+        .map(|b| matches!(b[0], b'G' | b'P' | b'H' | b'C' | b'D'))
+        .unwrap_or(false);
+    let now = Instant::now();
+    let dedup_key = hash_value(&(
+        state.user.as_str(),
+        state.peer_hash,
+        proto_tag,
+        DESYNC_ERROR_CLASS,
+    ));
+    let emit_full = should_emit_full_desync(dedup_key, state.desync_all_full, now);
+    let duration_ms = state.started_at.elapsed().as_millis() as u64;
+    let bytes_me2c = state.bytes_me2c.load(Ordering::Relaxed);
+
+    stats.increment_desync_total();
+    stats.observe_desync_frames_ok(frame_counter);
+    if emit_full {
+        stats.increment_desync_full_logged();
+        warn!(
+            trace_id = format_args!("0x{:016x}", state.trace_id),
+            conn_id = state.conn_id,
+            user = %state.user,
+            peer_hash = format_args!("0x{:016x}", state.peer_hash),
+            proto = ?proto_tag,
+            mode = "middle_proxy",
+            is_tls = true,
+            duration_ms,
+            bytes_c2me = state.bytes_c2me,
+            bytes_me2c,
+            raw_len = len,
+            raw_len_hex = format_args!("0x{:08x}", len),
+            raw_bytes = format_args!(
+                "{:02x} {:02x} {:02x} {:02x}",
+                len_buf[0], len_buf[1], len_buf[2], len_buf[3]
+            ),
+            max_frame,
+            tls_like = looks_like_tls,
+            http_like = looks_like_http,
+            frames_ok = frame_counter,
+            dedup_window_secs = DESYNC_DEDUP_WINDOW.as_secs(),
+            desync_all_full = state.desync_all_full,
+            full_reason = if state.desync_all_full { "desync_all_full" } else { "first_in_dedup_window" },
+            error_class = DESYNC_ERROR_CLASS,
+            "Frame too large — crypto desync forensics"
+        );
+        debug!(
+            trace_id = format_args!("0x{:016x}", state.trace_id),
+            conn_id = state.conn_id,
+            user = %state.user,
+            peer = %state.peer,
+            "Frame too large forensic peer detail"
+        );
+    } else {
+        stats.increment_desync_suppressed();
+        debug!(
+            trace_id = format_args!("0x{:016x}", state.trace_id),
+            conn_id = state.conn_id,
+            user = %state.user,
+            peer_hash = format_args!("0x{:016x}", state.peer_hash),
+            proto = ?proto_tag,
+            duration_ms,
+            bytes_c2me = state.bytes_c2me,
+            bytes_me2c,
+            raw_len = len,
+            frames_ok = frame_counter,
+            dedup_window_secs = DESYNC_DEDUP_WINDOW.as_secs(),
+            error_class = DESYNC_ERROR_CLASS,
+            "Frame too large — crypto desync forensic suppressed"
+        );
+    }
+
+    ProxyError::Proxy(format!(
+        "Frame too large: {len} (max {max_frame}), frames_ok={frame_counter}, conn_id={}, trace_id=0x{:016x}",
+        state.conn_id,
+        state.trace_id
+    ))
+}
+
+fn should_yield_c2me_sender(sent_since_yield: usize, has_backlog: bool) -> bool {
+    has_backlog && sent_since_yield >= C2ME_SENDER_FAIRNESS_BUDGET
+}
+
+async fn enqueue_c2me_command(
+    tx: &mpsc::Sender<C2MeCommand>,
+    cmd: C2MeCommand,
+) -> std::result::Result<(), mpsc::error::SendError<C2MeCommand>> {
+    match tx.try_send(cmd) {
+        Ok(()) => Ok(()),
+        Err(mpsc::error::TrySendError::Closed(cmd)) => Err(mpsc::error::SendError(cmd)),
+        Err(mpsc::error::TrySendError::Full(cmd)) => {
+            // Cooperative yield reduces burst catch-up when the per-conn queue is near saturation.
+            if tx.capacity() <= C2ME_SOFT_PRESSURE_MIN_FREE_SLOTS {
+                tokio::task::yield_now().await;
+            }
+            tx.send(cmd).await
+        }
+    }
+}
+
+pub(crate) async fn handle_via_middle_proxy<R, W>(
+    mut crypto_reader: CryptoReader<R>,
+    crypto_writer: CryptoWriter<W>,
+    success: HandshakeSuccess,
+    me_pool: Arc<MePool>,
+    stats: Arc<Stats>,
+    config: Arc<ProxyConfig>,
+    _buffer_pool: Arc<BufferPool>,
+    local_addr: SocketAddr,
+    rng: Arc<SecureRandom>,
+) -> Result<()>
+where
+    R: AsyncRead + Unpin + Send + 'static,
+    W: AsyncWrite + Unpin + Send + 'static,
+{
+    let user = success.user.clone();
+    let peer = success.peer;
+    let proto_tag = success.proto_tag;
+    let pool_generation = me_pool.current_generation();
+
+    info!(
+        user = %user,
+        peer = %peer,
+        dc = success.dc_idx,
+        proto = ?proto_tag,
+        mode = "middle_proxy",
+        pool_generation,
+        "Routing via Middle-End"
+    );
+
+    let (conn_id, me_rx) = me_pool.registry().register().await;
+    let trace_id = conn_id;
+    let bytes_me2c = Arc::new(AtomicU64::new(0));
+    let mut forensics = RelayForensicsState {
+        trace_id,
+        conn_id,
+        user: user.clone(),
+        peer,
+        peer_hash: hash_ip(peer.ip()),
+        started_at: Instant::now(),
+        bytes_c2me: 0,
+        bytes_me2c: bytes_me2c.clone(),
+        desync_all_full: config.general.desync_all_full,
+    };
+
+    stats.increment_user_connects(&user);
+    stats.increment_user_curr_connects(&user);
+
+    // Per-user ad_tag from access.user_ad_tags; fallback to general.ad_tag (hot-reloadable)
+    let user_tag: Option<Vec<u8>> = config
+        .access
+        .user_ad_tags
+        .get(&user)
+        .and_then(|s| hex::decode(s).ok())
+        .filter(|v| v.len() == 16);
+    let global_tag: Option<Vec<u8>> = config
+        .general
+        .ad_tag
+        .as_ref()
+        .and_then(|s| hex::decode(s).ok())
+        .filter(|v| v.len() == 16);
+    let effective_tag = user_tag.or(global_tag);
+
+    let proto_flags = proto_flags_for_tag(proto_tag, effective_tag.is_some());
+    debug!(
+        trace_id = format_args!("0x{:016x}", trace_id),
+        user = %user,
+        conn_id,
+        peer_hash = format_args!("0x{:016x}", forensics.peer_hash),
+        desync_all_full = forensics.desync_all_full,
+        proto_flags = format_args!("0x{:08x}", proto_flags),
+        pool_generation,
+        "ME relay started"
+    );
+
+    let translated_local_addr = me_pool.translate_our_addr(local_addr);
+
+    let frame_limit = config.general.max_client_frame;
+
+    let (c2me_tx, mut c2me_rx) = mpsc::channel::<C2MeCommand>(C2ME_CHANNEL_CAPACITY);
+    let me_pool_c2me = me_pool.clone();
+    let effective_tag = effective_tag;
+    let c2me_sender = tokio::spawn(async move {
+        let mut sent_since_yield = 0usize;
+        while let Some(cmd) = c2me_rx.recv().await {
+            match cmd {
+                C2MeCommand::Data { payload, flags } => {
+                    me_pool_c2me.send_proxy_req(
+                        conn_id,
+                        success.dc_idx,
+                        peer,
+                        translated_local_addr,
+                        &payload,
+                        flags,
+                        effective_tag.as_deref(),
+                    ).await?;
+                    sent_since_yield = sent_since_yield.saturating_add(1);
+                    if should_yield_c2me_sender(sent_since_yield, !c2me_rx.is_empty()) {
+                        sent_since_yield = 0;
+                        tokio::task::yield_now().await;
+                    }
+                }
+                C2MeCommand::Close => {
+                    let _ = me_pool_c2me.send_close(conn_id).await;
+                    return Ok(());
+                }
+            }
+        }
+        Ok(())
+    });
+
+    let (stop_tx, mut stop_rx) = oneshot::channel::<()>();
+    let mut me_rx_task = me_rx;
+    let stats_clone = stats.clone();
+    let rng_clone = rng.clone();
+    let user_clone = user.clone();
+    let bytes_me2c_clone = bytes_me2c.clone();
+    let me_writer = tokio::spawn(async move {
+        let mut writer = crypto_writer;
+        let mut frame_buf = Vec::with_capacity(16 * 1024);
+        loop {
+            tokio::select! {
+                msg = me_rx_task.recv() => {
+                    match msg {
+                        Some(MeResponse::Data { flags, data }) => {
+                            trace!(conn_id, bytes = data.len(), flags, "ME->C data");
+                            bytes_me2c_clone.fetch_add(data.len() as u64, Ordering::Relaxed);
+                            stats_clone.add_user_octets_to(&user_clone, data.len() as u64);
+                            write_client_payload(
+                                &mut writer,
+                                proto_tag,
+                                flags,
+                                &data,
+                                rng_clone.as_ref(),
+                                &mut frame_buf,
+                            )
+                            .await?;
+
+                            // Drain all immediately queued ME responses and flush once.
+                            while let Ok(next) = me_rx_task.try_recv() {
+                                match next {
+                                    MeResponse::Data { flags, data } => {
+                                        trace!(conn_id, bytes = data.len(), flags, "ME->C data (batched)");
+                                        bytes_me2c_clone.fetch_add(data.len() as u64, Ordering::Relaxed);
+                                        stats_clone.add_user_octets_to(&user_clone, data.len() as u64);
+                                        write_client_payload(
+                                            &mut writer,
+                                            proto_tag,
+                                            flags,
+                                            &data,
+                                            rng_clone.as_ref(),
+                                            &mut frame_buf,
+                                        ).await?;
+                                    }
+                                    MeResponse::Ack(confirm) => {
+                                        trace!(conn_id, confirm, "ME->C quickack (batched)");
+                                        write_client_ack(&mut writer, proto_tag, confirm).await?;
+                                    }
+                                    MeResponse::Close => {
+                                        debug!(conn_id, "ME sent close (batched)");
+                                        let _ = writer.flush().await;
+                                        return Ok(());
+                                    }
+                                }
+                            }
+
+                            writer.flush().await.map_err(ProxyError::Io)?;
+                        }
+                        Some(MeResponse::Ack(confirm)) => {
+                            trace!(conn_id, confirm, "ME->C quickack");
+                            write_client_ack(&mut writer, proto_tag, confirm).await?;
+                        }
+                        Some(MeResponse::Close) => {
+                            debug!(conn_id, "ME sent close");
+                            let _ = writer.flush().await;
+                            return Ok(());
+                        }
+                        None => {
+                            debug!(conn_id, "ME channel closed");
+                            return Err(ProxyError::Proxy("ME connection lost".into()));
+                        }
+                    }
+                }
+                _ = &mut stop_rx => {
+                    debug!(conn_id, "ME writer stop signal");
+                    return Ok(());
+                }
+            }
+        }
+    });
+
+    let mut main_result: Result<()> = Ok(());
+    let mut client_closed = false;
+    let mut frame_counter: u64 = 0;
+    loop {
+        match read_client_payload(
+            &mut crypto_reader,
+            proto_tag,
+            frame_limit,
+            &forensics,
+            &mut frame_counter,
+            &stats,
+        ).await {
+            Ok(Some((payload, quickack))) => {
+                trace!(conn_id, bytes = payload.len(), "C->ME frame");
+                forensics.bytes_c2me = forensics
+                    .bytes_c2me
+                    .saturating_add(payload.len() as u64);
+                stats.add_user_octets_from(&user, payload.len() as u64);
+                let mut flags = proto_flags;
+                if quickack {
+                    flags |= RPC_FLAG_QUICKACK;
+                }
+                if payload.len() >= 8 && payload[..8].iter().all(|b| *b == 0) {
+                    flags |= RPC_FLAG_NOT_ENCRYPTED;
+                }
+                // Keep client read loop lightweight: route heavy ME send path via a dedicated task.
+                if enqueue_c2me_command(&c2me_tx, C2MeCommand::Data { payload, flags })
+                    .await
+                    .is_err()
+                {
+                    main_result = Err(ProxyError::Proxy("ME sender channel closed".into()));
+                    break;
+                }
+            }
+            Ok(None) => {
+                debug!(conn_id, "Client EOF");
+                client_closed = true;
+                let _ = enqueue_c2me_command(&c2me_tx, C2MeCommand::Close).await;
+                break;
+            }
+            Err(e) => {
+                main_result = Err(e);
+                break;
+            }
+        }
+    }
+
+    drop(c2me_tx);
+    let c2me_result = c2me_sender
+        .await
+        .unwrap_or_else(|e| Err(ProxyError::Proxy(format!("ME sender join error: {e}"))));
+
+    let _ = stop_tx.send(());
+    let mut writer_result = me_writer
+        .await
+        .unwrap_or_else(|e| Err(ProxyError::Proxy(format!("ME writer join error: {e}"))));
+
+    // When client closes, but ME channel stopped as unregistered - it isnt error
+    if client_closed
+        && matches!(
+            writer_result,
+            Err(ProxyError::Proxy(ref msg)) if msg == "ME connection lost"
+        )
+    {
+        writer_result = Ok(());
+    }
+
+    let result = match (main_result, c2me_result, writer_result) {
+        (Ok(()), Ok(()), Ok(())) => Ok(()),
+        (Err(e), _, _) => Err(e),
+        (_, Err(e), _) => Err(e),
+        (_, _, Err(e)) => Err(e),
+    };
+
+    debug!(
+        user = %user,
+        conn_id,
+        trace_id = format_args!("0x{:016x}", trace_id),
+        duration_ms = forensics.started_at.elapsed().as_millis() as u64,
+        bytes_c2me = forensics.bytes_c2me,
+        bytes_me2c = forensics.bytes_me2c.load(Ordering::Relaxed),
+        frames_ok = frame_counter,
+        "ME relay cleanup"
+    );
+    me_pool.registry().unregister(conn_id).await;
+    stats.decrement_user_curr_connects(&user);
+    result
+}
+
+async fn read_client_payload<R>(
+    client_reader: &mut CryptoReader<R>,
+    proto_tag: ProtoTag,
+    max_frame: usize,
+    forensics: &RelayForensicsState,
+    frame_counter: &mut u64,
+    stats: &Stats,
+) -> Result<Option<(Vec<u8>, bool)>>
+where
+    R: AsyncRead + Unpin + Send + 'static,
+{
+    loop {
+        let (len, quickack, raw_len_bytes) = match proto_tag {
+            ProtoTag::Abridged => {
+                let mut first = [0u8; 1];
+                match client_reader.read_exact(&mut first).await {
+                    Ok(_) => {}
+                    Err(e) if e.kind() == std::io::ErrorKind::UnexpectedEof => return Ok(None),
+                    Err(e) => return Err(ProxyError::Io(e)),
+                }
+
+                let quickack = (first[0] & 0x80) != 0;
+                let len_words = if (first[0] & 0x7f) == 0x7f {
+                    let mut ext = [0u8; 3];
+                    client_reader
+                        .read_exact(&mut ext)
+                        .await
+                        .map_err(ProxyError::Io)?;
+                    u32::from_le_bytes([ext[0], ext[1], ext[2], 0]) as usize
+                } else {
+                    (first[0] & 0x7f) as usize
+                };
+
+                let len = len_words
+                    .checked_mul(4)
+                    .ok_or_else(|| ProxyError::Proxy("Abridged frame length overflow".into()))?;
+                (len, quickack, None)
+            }
+            ProtoTag::Intermediate | ProtoTag::Secure => {
+                let mut len_buf = [0u8; 4];
+                match client_reader.read_exact(&mut len_buf).await {
+                    Ok(_) => {}
+                    Err(e) if e.kind() == std::io::ErrorKind::UnexpectedEof => return Ok(None),
+                    Err(e) => return Err(ProxyError::Io(e)),
+                }
+                let quickack = (len_buf[3] & 0x80) != 0;
+                (
+                    (u32::from_le_bytes(len_buf) & 0x7fff_ffff) as usize,
+                    quickack,
+                    Some(len_buf),
+                )
+            }
+        };
+
+        if len == 0 {
+            continue;
+        }
+        if len < 4 && proto_tag != ProtoTag::Abridged {
+            warn!(
+                trace_id = format_args!("0x{:016x}", forensics.trace_id),
+                conn_id = forensics.conn_id,
+                user = %forensics.user,
+                len,
+                proto = ?proto_tag,
+                "Frame too small — corrupt or probe"
+            );
+            return Err(ProxyError::Proxy(format!("Frame too small: {len}")));
+        }
+
+        if len > max_frame {
+            return Err(report_desync_frame_too_large(
+                forensics,
+                proto_tag,
+                *frame_counter,
+                max_frame,
+                len,
+                raw_len_bytes,
+                stats,
+            ));
+        }
+
+        let secure_payload_len = if proto_tag == ProtoTag::Secure {
+            match secure_payload_len_from_wire_len(len) {
+                Some(payload_len) => payload_len,
+                None => {
+                    stats.increment_secure_padding_invalid();
+                    return Err(ProxyError::Proxy(format!(
+                        "Invalid secure frame length: {len}"
+                    )));
+                }
+            }
+        } else {
+            len
+        };
+
+        let mut payload = vec![0u8; len];
+        client_reader
+            .read_exact(&mut payload)
+            .await
+            .map_err(ProxyError::Io)?;
+
+        // Secure Intermediate: strip validated trailing padding bytes.
+        if proto_tag == ProtoTag::Secure {
+            payload.truncate(secure_payload_len);
+        }
+        *frame_counter += 1;
+        return Ok(Some((payload, quickack)));
+    }
+}
+
+async fn write_client_payload<W>(
+    client_writer: &mut CryptoWriter<W>,
+    proto_tag: ProtoTag,
+    flags: u32,
+    data: &[u8],
+    rng: &SecureRandom,
+    frame_buf: &mut Vec<u8>,
+) -> Result<()>
+where
+    W: AsyncWrite + Unpin + Send + 'static,
+{
+    let quickack = (flags & RPC_FLAG_QUICKACK) != 0;
+
+    match proto_tag {
+        ProtoTag::Abridged => {
+            if !data.len().is_multiple_of(4) {
+                return Err(ProxyError::Proxy(format!(
+                    "Abridged payload must be 4-byte aligned, got {}",
+                    data.len()
+                )));
+            }
+
+            let len_words = data.len() / 4;
+            if len_words < 0x7f {
+                let mut first = len_words as u8;
+                if quickack {
+                    first |= 0x80;
+                }
+                frame_buf.clear();
+                frame_buf.reserve(1 + data.len());
+                frame_buf.push(first);
+                frame_buf.extend_from_slice(data);
+                client_writer
+                    .write_all(frame_buf)
+                    .await
+                    .map_err(ProxyError::Io)?;
+            } else if len_words < (1 << 24) {
+                let mut first = 0x7fu8;
+                if quickack {
+                    first |= 0x80;
+                }
+                let lw = (len_words as u32).to_le_bytes();
+                frame_buf.clear();
+                frame_buf.reserve(4 + data.len());
+                frame_buf.extend_from_slice(&[first, lw[0], lw[1], lw[2]]);
+                frame_buf.extend_from_slice(data);
+                client_writer
+                    .write_all(frame_buf)
+                    .await
+                    .map_err(ProxyError::Io)?;
+            } else {
+                return Err(ProxyError::Proxy(format!(
+                    "Abridged frame too large: {}",
+                    data.len()
+                )));
+            }
+        }
+        ProtoTag::Intermediate | ProtoTag::Secure => {
+            let padding_len = if proto_tag == ProtoTag::Secure {
+                if !is_valid_secure_payload_len(data.len()) {
+                    return Err(ProxyError::Proxy(format!(
+                        "Secure payload must be 4-byte aligned, got {}",
+                        data.len()
+                    )));
+                }
+                secure_padding_len(data.len(), rng)
+            } else {
+                0
+            };
+            let mut len_val = (data.len() + padding_len) as u32;
+            if quickack {
+                len_val |= 0x8000_0000;
+            }
+            let total = 4 + data.len() + padding_len;
+            frame_buf.clear();
+            frame_buf.reserve(total);
+            frame_buf.extend_from_slice(&len_val.to_le_bytes());
+            frame_buf.extend_from_slice(data);
+            if padding_len > 0 {
+                let start = frame_buf.len();
+                frame_buf.resize(start + padding_len, 0);
+                rng.fill(&mut frame_buf[start..]);
+            }
+            client_writer
+                .write_all(frame_buf)
+                .await
+                .map_err(ProxyError::Io)?;
+        }
+    }
+
+    Ok(())
+}
+
+async fn write_client_ack<W>(
+    client_writer: &mut CryptoWriter<W>,
+    proto_tag: ProtoTag,
+    confirm: u32,
+) -> Result<()>
+where
+    W: AsyncWrite + Unpin + Send + 'static,
+{
+    let bytes = if proto_tag == ProtoTag::Abridged {
+        confirm.to_be_bytes()
+    } else {
+        confirm.to_le_bytes()
+    };
+    client_writer
+        .write_all(&bytes)
+        .await
+        .map_err(ProxyError::Io)?;
+    // ACK should remain low-latency.
+    client_writer.flush().await.map_err(ProxyError::Io)
+}
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+    use tokio::time::{Duration as TokioDuration, timeout};
+
+    #[test]
+    fn should_yield_sender_only_on_budget_with_backlog() {
+        assert!(!should_yield_c2me_sender(0, true));
+        assert!(!should_yield_c2me_sender(C2ME_SENDER_FAIRNESS_BUDGET - 1, true));
+        assert!(!should_yield_c2me_sender(C2ME_SENDER_FAIRNESS_BUDGET, false));
+        assert!(should_yield_c2me_sender(C2ME_SENDER_FAIRNESS_BUDGET, true));
+    }
+
+    #[tokio::test]
+    async fn enqueue_c2me_command_uses_try_send_fast_path() {
+        let (tx, mut rx) = mpsc::channel::<C2MeCommand>(2);
+        enqueue_c2me_command(
+            &tx,
+            C2MeCommand::Data {
+                payload: vec![1, 2, 3],
+                flags: 0,
+            },
+        )
+        .await
+        .unwrap();
+
+        let recv = timeout(TokioDuration::from_millis(50), rx.recv())
+            .await
+            .unwrap()
+            .unwrap();
+        match recv {
+            C2MeCommand::Data { payload, flags } => {
+                assert_eq!(payload, vec![1, 2, 3]);
+                assert_eq!(flags, 0);
+            }
+            C2MeCommand::Close => panic!("unexpected close command"),
+        }
+    }
+
+    #[tokio::test]
+    async fn enqueue_c2me_command_falls_back_to_send_when_queue_is_full() {
+        let (tx, mut rx) = mpsc::channel::<C2MeCommand>(1);
+        tx.send(C2MeCommand::Data {
+            payload: vec![9],
+            flags: 9,
+        })
+        .await
+        .unwrap();
+
+        let tx2 = tx.clone();
+        let producer = tokio::spawn(async move {
+            enqueue_c2me_command(
+                &tx2,
+                C2MeCommand::Data {
+                    payload: vec![7, 7],
+                    flags: 7,
+                },
+            )
+            .await
+            .unwrap();
+        });
+
+        let _ = timeout(TokioDuration::from_millis(100), rx.recv())
+            .await
+            .unwrap();
+        producer.await.unwrap();
+
+        let recv = timeout(TokioDuration::from_millis(100), rx.recv())
+            .await
+            .unwrap()
+            .unwrap();
+        match recv {
+            C2MeCommand::Data { payload, flags } => {
+                assert_eq!(payload, vec![7, 7]);
+                assert_eq!(flags, 7);
+            }
+            C2MeCommand::Close => panic!("unexpected close command"),
+        }
+    }
+}

src/proxy/mod.rs

@@ -1,11 +1,16 @@
 //! Proxy Defs
 
-pub mod handshake;
 pub mod client;
-pub mod relay;
+pub mod direct_relay;
+pub mod handshake;
 pub mod masking;
+pub mod middle_relay;
+pub mod relay;
 
-pub use handshake::*;
 pub use client::ClientHandler;
+#[allow(unused_imports)]
+pub use handshake::*;
+#[allow(unused_imports)]
+pub use masking::*;
+#[allow(unused_imports)]
 pub use relay::*;
-pub use masking::*;

src/proxy/relay.rs

@@ -1,22 +1,320 @@
-//! Bidirectional Relay
+//! Bidirectional Relay — poll-based, no head-of-line blocking
+//!
+//! ## What changed and why
+//!
+//! Previous implementation used a single-task `select! { biased; ... }` loop
+//! where each branch called `write_all()`. This caused head-of-line blocking:
+//! while `write_all()` waited for a slow writer (e.g. client on 3G downloading
+//! media), the entire loop was blocked — the other direction couldn't make progress.
+//!
+//! Symptoms observed in production:
+//! - Media loading at ~8 KB/s despite fast server connection
+//! - Stop-and-go pattern with 50–500ms gaps between chunks
+//! - `biased` select starving S→C direction
+//! - Some users unable to load media at all
+//!
+//! ## New architecture
+//!
+//! Uses `tokio::io::copy_bidirectional` which polls both directions concurrently
+//! in a single task via non-blocking `poll_read` / `poll_write` calls:
+//!
+//! Old (select! + write_all — BLOCKING):
+//!
+//!   loop {
+//!       select! {
+//!           biased;
+//!           data = client.read()  => { server.write_all(data).await; }  ← BLOCKS here
+//!           data = server.read()  => { client.write_all(data).await; }  ← can't run
+//!       }
+//!   }
+//!
+//! New (copy_bidirectional — CONCURRENT):
+//!
+//!   poll(cx) {
+//!       // Both directions polled in the same poll cycle
+//!       C→S: poll_read(client) → poll_write(server)   // non-blocking
+//!       S→C: poll_read(server) → poll_write(client)   // non-blocking
+//!       // If one writer is Pending, the other direction still progresses
+//!   }
+//!
+//! Benefits:
+//! - No head-of-line blocking: slow client download doesn't block uploads
+//! - No biased starvation: fair polling of both directions
+//! - Proper flush: `copy_bidirectional` calls `poll_flush` when reader stalls,
+//!   so CryptoWriter's pending ciphertext is always drained (fixes "stuck at 95%")
+//! - No deadlock risk: old write_all could deadlock when both TCP buffers filled;
+//!   poll-based approach lets TCP flow control work correctly
+//!
+//! Stats tracking:
+//! - `StatsIo` wraps client side, intercepts `poll_read` / `poll_write`
+//! - `poll_read` on client = C→S (client sending) → `octets_from`, `msgs_from`
+//! - `poll_write` on client = S→C (to client)     → `octets_to`, `msgs_to`
+//! - `SharedCounters` (atomics) let the watchdog read stats without locking
 
+use std::io;
+use std::pin::Pin;
 use std::sync::Arc;
-use tokio::io::{AsyncRead, AsyncWrite, AsyncReadExt, AsyncWriteExt};
+use std::sync::atomic::{AtomicU64, Ordering};
+use std::task::{Context, Poll};
+use std::time::Duration;
+use tokio::io::{AsyncRead, AsyncWrite, AsyncWriteExt, ReadBuf, copy_bidirectional};
+use tokio::time::Instant;
 use tracing::{debug, trace, warn};
 use crate::error::Result;
 use crate::stats::Stats;
-use std::sync::atomic::{AtomicU64, Ordering};
+use crate::stream::BufferPool;
 
-const BUFFER_SIZE: usize = 65536;
+// ============= Constants =============
 
-/// Relay data bidirectionally between client and server
+/// Activity timeout for iOS compatibility.
+///
+/// iOS keeps Telegram connections alive in background for up to 30 minutes.
+/// Closing earlier causes unnecessary reconnects and handshake overhead.
+const ACTIVITY_TIMEOUT: Duration = Duration::from_secs(1800);
+
+/// Watchdog check interval — also used for periodic rate logging.
+///
+/// 10 seconds gives responsive timeout detection (±10s accuracy)
+/// without measurable overhead from atomic reads.
+const WATCHDOG_INTERVAL: Duration = Duration::from_secs(10);
+
+// ============= CombinedStream =============
+
+/// Combines separate read and write halves into a single bidirectional stream.
+///
+/// `copy_bidirectional` requires `AsyncRead + AsyncWrite` on each side,
+/// but the handshake layer produces split reader/writer pairs
+/// (e.g. `CryptoReader<FakeTlsReader<OwnedReadHalf>>` + `CryptoWriter<...>`).
+///
+/// This wrapper reunifies them with zero overhead — each trait method
+/// delegates directly to the corresponding half. No buffering, no copies.
+///
+/// Safety: `poll_read` only touches `reader`, `poll_write` only touches `writer`,
+/// so there's no aliasing even though both are called on the same `&mut self`.
+struct CombinedStream<R, W> {
+    reader: R,
+    writer: W,
+}
+
+impl<R, W> CombinedStream<R, W> {
+    fn new(reader: R, writer: W) -> Self {
+        Self { reader, writer }
+    }
+}
+
+impl<R: AsyncRead + Unpin, W: Unpin> AsyncRead for CombinedStream<R, W> {
+    #[inline]
+    fn poll_read(
+        self: Pin<&mut Self>,
+        cx: &mut Context<'_>,
+        buf: &mut ReadBuf<'_>,
+    ) -> Poll<io::Result<()>> {
+        Pin::new(&mut self.get_mut().reader).poll_read(cx, buf)
+    }
+}
+
+impl<R: Unpin, W: AsyncWrite + Unpin> AsyncWrite for CombinedStream<R, W> {
+    #[inline]
+    fn poll_write(
+        self: Pin<&mut Self>,
+        cx: &mut Context<'_>,
+        buf: &[u8],
+    ) -> Poll<io::Result<usize>> {
+        Pin::new(&mut self.get_mut().writer).poll_write(cx, buf)
+    }
+
+    #[inline]
+    fn poll_flush(self: Pin<&mut Self>, cx: &mut Context<'_>) -> Poll<io::Result<()>> {
+        Pin::new(&mut self.get_mut().writer).poll_flush(cx)
+    }
+
+    #[inline]
+    fn poll_shutdown(self: Pin<&mut Self>, cx: &mut Context<'_>) -> Poll<io::Result<()>> {
+        Pin::new(&mut self.get_mut().writer).poll_shutdown(cx)
+    }
+}
+
+// ============= SharedCounters =============
+
+/// Atomic counters shared between the relay (via StatsIo) and the watchdog task.
+///
+/// Using `Relaxed` ordering is sufficient because:
+/// - Counters are monotonically increasing (no ABA problem)
+/// - Slight staleness in watchdog reads is harmless (±10s check interval anyway)
+/// - No ordering dependencies between different counters
+struct SharedCounters {
+    /// Bytes read from client (C→S direction)
+    c2s_bytes: AtomicU64,
+    /// Bytes written to client (S→C direction)
+    s2c_bytes: AtomicU64,
+    /// Number of poll_read completions (≈ C→S chunks)
+    c2s_ops: AtomicU64,
+    /// Number of poll_write completions (≈ S→C chunks)
+    s2c_ops: AtomicU64,
+    /// Milliseconds since relay epoch of last I/O activity
+    last_activity_ms: AtomicU64,
+}
+
+impl SharedCounters {
+    fn new() -> Self {
+        Self {
+            c2s_bytes: AtomicU64::new(0),
+            s2c_bytes: AtomicU64::new(0),
+            c2s_ops: AtomicU64::new(0),
+            s2c_ops: AtomicU64::new(0),
+            last_activity_ms: AtomicU64::new(0),
+        }
+    }
+
+    /// Record activity at this instant.
+    #[inline]
+    fn touch(&self, now: Instant, epoch: Instant) {
+        let ms = now.duration_since(epoch).as_millis() as u64;
+        self.last_activity_ms.store(ms, Ordering::Relaxed);
+    }
+
+    /// How long since last recorded activity.
+    fn idle_duration(&self, now: Instant, epoch: Instant) -> Duration {
+        let last_ms = self.last_activity_ms.load(Ordering::Relaxed);
+        let now_ms = now.duration_since(epoch).as_millis() as u64;
+        Duration::from_millis(now_ms.saturating_sub(last_ms))
+    }
+}
+
+// ============= StatsIo =============
+
+/// Transparent I/O wrapper that tracks per-user statistics and activity.
+///
+/// Wraps the **client** side of the relay. Direction mapping:
+///
+/// | poll method  | direction | stats updated                        |
+/// |-------------|-----------|--------------------------------------|
+/// | `poll_read`  | C→S       | `octets_from`, `msgs_from`, counters |
+/// | `poll_write` | S→C       | `octets_to`, `msgs_to`, counters     |
+///
+/// Both update the shared activity timestamp for the watchdog.
+///
+/// Note on message counts: the original code counted one `read()`/`write_all()`
+/// as one "message". Here we count `poll_read`/`poll_write` completions instead.
+/// Byte counts are identical; op counts may differ slightly due to different
+/// internal buffering in `copy_bidirectional`. This is fine for monitoring.
+struct StatsIo<S> {
+    inner: S,
+    counters: Arc<SharedCounters>,
+    stats: Arc<Stats>,
+    user: String,
+    epoch: Instant,
+}
+
+impl<S> StatsIo<S> {
+    fn new(
+        inner: S,
+        counters: Arc<SharedCounters>,
+        stats: Arc<Stats>,
+        user: String,
+        epoch: Instant,
+    ) -> Self {
+        // Mark initial activity so the watchdog doesn't fire before data flows
+        counters.touch(Instant::now(), epoch);
+        Self { inner, counters, stats, user, epoch }
+    }
+}
+
+impl<S: AsyncRead + Unpin> AsyncRead for StatsIo<S> {
+    fn poll_read(
+        self: Pin<&mut Self>,
+        cx: &mut Context<'_>,
+        buf: &mut ReadBuf<'_>,
+    ) -> Poll<io::Result<()>> {
+        let this = self.get_mut();
+        let before = buf.filled().len();
+
+        match Pin::new(&mut this.inner).poll_read(cx, buf) {
+            Poll::Ready(Ok(())) => {
+                let n = buf.filled().len() - before;
+                if n > 0 {
+                    // C→S: client sent data
+                    this.counters.c2s_bytes.fetch_add(n as u64, Ordering::Relaxed);
+                    this.counters.c2s_ops.fetch_add(1, Ordering::Relaxed);
+                    this.counters.touch(Instant::now(), this.epoch);
+
+                    this.stats.add_user_octets_from(&this.user, n as u64);
+                    this.stats.increment_user_msgs_from(&this.user);
+
+                    trace!(user = %this.user, bytes = n, "C->S");
+                }
+                Poll::Ready(Ok(()))
+            }
+            other => other,
+        }
+    }
+}
+
+impl<S: AsyncWrite + Unpin> AsyncWrite for StatsIo<S> {
+    fn poll_write(
+        self: Pin<&mut Self>,
+        cx: &mut Context<'_>,
+        buf: &[u8],
+    ) -> Poll<io::Result<usize>> {
+        let this = self.get_mut();
+
+        match Pin::new(&mut this.inner).poll_write(cx, buf) {
+            Poll::Ready(Ok(n)) => {
+                if n > 0 {
+                    // S→C: data written to client
+                    this.counters.s2c_bytes.fetch_add(n as u64, Ordering::Relaxed);
+                    this.counters.s2c_ops.fetch_add(1, Ordering::Relaxed);
+                    this.counters.touch(Instant::now(), this.epoch);
+
+                    this.stats.add_user_octets_to(&this.user, n as u64);
+                    this.stats.increment_user_msgs_to(&this.user);
+
+                    trace!(user = %this.user, bytes = n, "S->C");
+                }
+                Poll::Ready(Ok(n))
+            }
+            other => other,
+        }
+    }
+
+    #[inline]
+    fn poll_flush(self: Pin<&mut Self>, cx: &mut Context<'_>) -> Poll<io::Result<()>> {
+        Pin::new(&mut self.get_mut().inner).poll_flush(cx)
+    }
+
+    #[inline]
+    fn poll_shutdown(self: Pin<&mut Self>, cx: &mut Context<'_>) -> Poll<io::Result<()>> {
+        Pin::new(&mut self.get_mut().inner).poll_shutdown(cx)
+    }
+}
+
+// ============= Relay =============
+
+/// Relay data bidirectionally between client and server.
+///
+/// Uses `tokio::io::copy_bidirectional` for concurrent, non-blocking data transfer.
+///
+/// ## API compatibility
+///
+/// Signature is identical to the previous implementation. The `_buffer_pool`
+/// parameter is retained for call-site compatibility — `copy_bidirectional`
+/// manages its own internal buffers (8 KB per direction).
+///
+/// ## Guarantees preserved
+///
+/// - Activity timeout: 30 minutes of inactivity → clean shutdown
+/// - Per-user stats: bytes and ops counted per direction
+/// - Periodic rate logging: every 10 seconds when active
+/// - Clean shutdown: both write sides are shut down on exit
+/// - Error propagation: I/O errors are returned as `ProxyError::Io`
 pub async fn relay_bidirectional<CR, CW, SR, SW>(
-    mut client_reader: CR,
-    mut client_writer: CW,
-    mut server_reader: SR,
-    mut server_writer: SW,
+    client_reader: CR,
+    client_writer: CW,
+    server_reader: SR,
+    server_writer: SW,
     user: &str,
     stats: Arc<Stats>,
+    _buffer_pool: Arc<BufferPool>,
 ) -> Result<()>
 where
     CR: AsyncRead + Unpin + Send + 'static,
@@ -24,139 +322,145 @@ where
     SR: AsyncRead + Unpin + Send + 'static,
     SW: AsyncWrite + Unpin + Send + 'static,
 {
-    let user_c2s = user.to_string();
-    let user_s2c = user.to_string();
-    
-    // Используем Arc::clone вместо stats.clone()
-    let stats_c2s = Arc::clone(&stats);
-    let stats_s2c = Arc::clone(&stats);
-    
-    let c2s_bytes = Arc::new(AtomicU64::new(0));
-    let s2c_bytes = Arc::new(AtomicU64::new(0));
-    let c2s_bytes_clone = Arc::clone(&c2s_bytes);
-    let s2c_bytes_clone = Arc::clone(&s2c_bytes);
-    
-    // Client -> Server task
-    let c2s = tokio::spawn(async move {
-        let mut buf = vec![0u8; BUFFER_SIZE];
-        let mut total_bytes = 0u64;
-        let mut msg_count = 0u64;
-        
+    let epoch = Instant::now();
+    let counters = Arc::new(SharedCounters::new());
+    let user_owned = user.to_string();
+
+    // ── Combine split halves into bidirectional streams ──────────────
+    let client_combined = CombinedStream::new(client_reader, client_writer);
+    let mut server = CombinedStream::new(server_reader, server_writer);
+
+    // Wrap client with stats/activity tracking
+    let mut client = StatsIo::new(
+        client_combined,
+        Arc::clone(&counters),
+        Arc::clone(&stats),
+        user_owned.clone(),
+        epoch,
+    );
+
+    // ── Watchdog: activity timeout + periodic rate logging ──────────
+    let wd_counters = Arc::clone(&counters);
+    let wd_user = user_owned.clone();
+
+    let watchdog = async {
+        let mut prev_c2s: u64 = 0;
+        let mut prev_s2c: u64 = 0;
+
         loop {
-            match client_reader.read(&mut buf).await {
-                Ok(0) => {
-                    debug!(
-                        user = %user_c2s, 
-                        total_bytes = total_bytes,
-                        msgs = msg_count,
-                        "Client closed connection (C->S)"
-                    );
-                    let _ = server_writer.shutdown().await;
-                    break;
-                }
-                Ok(n) => {
-                    total_bytes += n as u64;
-                    msg_count += 1;
-                    c2s_bytes_clone.store(total_bytes, Ordering::Relaxed);
-                    
-                    stats_c2s.add_user_octets_from(&user_c2s, n as u64);
-                    stats_c2s.increment_user_msgs_from(&user_c2s);
-                    
-                    trace!(
-                        user = %user_c2s,
-                        bytes = n,
-                        total = total_bytes,
-                        data_preview = %hex::encode(&buf[..n.min(32)]),
-                        "C->S data"
-                    );
-                    
-                    if let Err(e) = server_writer.write_all(&buf[..n]).await {
-                        debug!(user = %user_c2s, error = %e, "Failed to write to server");
-                        break;
-                    }
-                    if let Err(e) = server_writer.flush().await {
-                        debug!(user = %user_c2s, error = %e, "Failed to flush to server");
-                        break;
-                    }
-                }
-                Err(e) => {
-                    debug!(user = %user_c2s, error = %e, total_bytes = total_bytes, "Client read error");
-                    break;
-                }
+            tokio::time::sleep(WATCHDOG_INTERVAL).await;
+
+            let now = Instant::now();
+            let idle = wd_counters.idle_duration(now, epoch);
+
+            // ── Activity timeout ────────────────────────────────────
+            if idle >= ACTIVITY_TIMEOUT {
+                let c2s = wd_counters.c2s_bytes.load(Ordering::Relaxed);
+                let s2c = wd_counters.s2c_bytes.load(Ordering::Relaxed);
+                warn!(
+                    user = %wd_user,
+                    c2s_bytes = c2s,
+                    s2c_bytes = s2c,
+                    idle_secs = idle.as_secs(),
+                    "Activity timeout"
+                );
+                return; // Causes select! to cancel copy_bidirectional
             }
+
+            // ── Periodic rate logging ───────────────────────────────
+            let c2s = wd_counters.c2s_bytes.load(Ordering::Relaxed);
+            let s2c = wd_counters.s2c_bytes.load(Ordering::Relaxed);
+            let c2s_delta = c2s - prev_c2s;
+            let s2c_delta = s2c - prev_s2c;
+
+            if c2s_delta > 0 || s2c_delta > 0 {
+                let secs = WATCHDOG_INTERVAL.as_secs_f64();
+                debug!(
+                    user = %wd_user,
+                    c2s_kbps = (c2s_delta as f64 / secs / 1024.0) as u64,
+                    s2c_kbps = (s2c_delta as f64 / secs / 1024.0) as u64,
+                    c2s_total = c2s,
+                    s2c_total = s2c,
+                    "Relay active"
+                );
+            }
+
+            prev_c2s = c2s;
+            prev_s2c = s2c;
         }
-    });
-    
-    // Server -> Client task
-    let s2c = tokio::spawn(async move {
-        let mut buf = vec![0u8; BUFFER_SIZE];
-        let mut total_bytes = 0u64;
-        let mut msg_count = 0u64;
-        
-        loop {
-            match server_reader.read(&mut buf).await {
-                Ok(0) => {
-                    debug!(
-                        user = %user_s2c,
-                        total_bytes = total_bytes,
-                        msgs = msg_count,
-                        "Server closed connection (S->C)"
-                    );
-                    let _ = client_writer.shutdown().await;
-                    break;
-                }
-                Ok(n) => {
-                    total_bytes += n as u64;
-                    msg_count += 1;
-                    s2c_bytes_clone.store(total_bytes, Ordering::Relaxed);
-                    
-                    stats_s2c.add_user_octets_to(&user_s2c, n as u64);
-                    stats_s2c.increment_user_msgs_to(&user_s2c);
-                    
-                    trace!(
-                        user = %user_s2c,
-                        bytes = n,
-                        total = total_bytes,
-                        data_preview = %hex::encode(&buf[..n.min(32)]),
-                        "S->C data"
-                    );
-                    
-                    if let Err(e) = client_writer.write_all(&buf[..n]).await {
-                        debug!(user = %user_s2c, error = %e, "Failed to write to client");
-                        break;
-                    }
-                    if let Err(e) = client_writer.flush().await {
-                        debug!(user = %user_s2c, error = %e, "Failed to flush to client");
-                        break;
-                    }
-                }
-                Err(e) => {
-                    debug!(user = %user_s2c, error = %e, total_bytes = total_bytes, "Server read error");
-                    break;
-                }
-            }
+    };
+
+    // ── Run bidirectional copy + watchdog concurrently ───────────────
+    //
+    // copy_bidirectional polls both directions in the same poll() call:
+    //   C→S: poll_read(client/StatsIo) → poll_write(server)
+    //   S→C: poll_read(server)         → poll_write(client/StatsIo)
+    //
+    // When one direction's writer returns Pending, the other direction
+    // continues — no head-of-line blocking.
+    //
+    // When the watchdog fires, select! drops the copy future,
+    // releasing the &mut borrows on client and server.
+    let copy_result = tokio::select! {
+        result = copy_bidirectional(&mut client, &mut server) => Some(result),
+        _ = watchdog => None, // Activity timeout — cancel relay
+    };
+
+    // ── Clean shutdown ──────────────────────────────────────────────
+    // After select!, the losing future is dropped, borrows released.
+    // Shut down both write sides for clean TCP FIN.
+    let _ = client.shutdown().await;
+    let _ = server.shutdown().await;
+
+    // ── Final logging ───────────────────────────────────────────────
+    let c2s_ops = counters.c2s_ops.load(Ordering::Relaxed);
+    let s2c_ops = counters.s2c_ops.load(Ordering::Relaxed);
+    let duration = epoch.elapsed();
+
+    match copy_result {
+        Some(Ok((c2s, s2c))) => {
+            // Normal completion — one side closed the connection
+            debug!(
+                user = %user_owned,
+                c2s_bytes = c2s,
+                s2c_bytes = s2c,
+                c2s_msgs = c2s_ops,
+                s2c_msgs = s2c_ops,
+                duration_secs = duration.as_secs(),
+                "Relay finished"
+            );
+            Ok(())
         }
-    });
-    
-    // Wait for either direction to complete
-    tokio::select! {
-        result = c2s => {
-            if let Err(e) = result {
-                warn!(error = %e, "C->S task panicked");
-            }
+        Some(Err(e)) => {
+            // I/O error in one of the directions
+            let c2s = counters.c2s_bytes.load(Ordering::Relaxed);
+            let s2c = counters.s2c_bytes.load(Ordering::Relaxed);
+            debug!(
+                user = %user_owned,
+                c2s_bytes = c2s,
+                s2c_bytes = s2c,
+                c2s_msgs = c2s_ops,
+                s2c_msgs = s2c_ops,
+                duration_secs = duration.as_secs(),
+                error = %e,
+                "Relay error"
+            );
+            Err(e.into())
         }
-        result = s2c => {
-            if let Err(e) = result {
-                warn!(error = %e, "S->C task panicked");
-            }
+        None => {
+            // Activity timeout (watchdog fired)
+            let c2s = counters.c2s_bytes.load(Ordering::Relaxed);
+            let s2c = counters.s2c_bytes.load(Ordering::Relaxed);
+            debug!(
+                user = %user_owned,
+                c2s_bytes = c2s,
+                s2c_bytes = s2c,
+                c2s_msgs = c2s_ops,
+                s2c_msgs = s2c_ops,
+                duration_secs = duration.as_secs(),
+                "Relay finished (activity timeout)"
+            );
+            Ok(())
         }
     }
-    
-    debug!(
-        c2s_bytes = c2s_bytes.load(Ordering::Relaxed),
-        s2c_bytes = s2c_bytes.load(Ordering::Relaxed),
-        "Relay finished"
-    );
-    
-    Ok(())
 }

src/stats/beobachten.rs

@@ -0,0 +1,117 @@
+//! Per-IP forensic buckets for scanner and handshake failure observation.
+
+use std::collections::{BTreeMap, HashMap};
+use std::net::IpAddr;
+use std::time::{Duration, Instant};
+
+use parking_lot::Mutex;
+
+const CLEANUP_INTERVAL: Duration = Duration::from_secs(30);
+
+#[derive(Default)]
+struct BeobachtenInner {
+    entries: HashMap<(String, IpAddr), BeobachtenEntry>,
+    last_cleanup: Option<Instant>,
+}
+
+#[derive(Clone, Copy)]
+struct BeobachtenEntry {
+    tries: u64,
+    last_seen: Instant,
+}
+
+/// In-memory, TTL-scoped per-IP counters keyed by source class.
+pub struct BeobachtenStore {
+    inner: Mutex<BeobachtenInner>,
+}
+
+impl Default for BeobachtenStore {
+    fn default() -> Self {
+        Self::new()
+    }
+}
+
+impl BeobachtenStore {
+    pub fn new() -> Self {
+        Self {
+            inner: Mutex::new(BeobachtenInner::default()),
+        }
+    }
+
+    pub fn record(&self, class: &str, ip: IpAddr, ttl: Duration) {
+        if class.is_empty() || ttl.is_zero() {
+            return;
+        }
+
+        let now = Instant::now();
+        let mut guard = self.inner.lock();
+        Self::cleanup_if_needed(&mut guard, now, ttl);
+
+        let key = (class.to_string(), ip);
+        let entry = guard.entries.entry(key).or_insert(BeobachtenEntry {
+            tries: 0,
+            last_seen: now,
+        });
+        entry.tries = entry.tries.saturating_add(1);
+        entry.last_seen = now;
+    }
+
+    pub fn snapshot_text(&self, ttl: Duration) -> String {
+        if ttl.is_zero() {
+            return "beobachten disabled\n".to_string();
+        }
+
+        let now = Instant::now();
+        let mut guard = self.inner.lock();
+        Self::cleanup(&mut guard, now, ttl);
+        guard.last_cleanup = Some(now);
+
+        let mut grouped = BTreeMap::<String, Vec<(IpAddr, u64)>>::new();
+        for ((class, ip), entry) in &guard.entries {
+            grouped
+                .entry(class.clone())
+                .or_default()
+                .push((*ip, entry.tries));
+        }
+
+        if grouped.is_empty() {
+            return "empty\n".to_string();
+        }
+
+        let mut out = String::with_capacity(grouped.len() * 64);
+        for (class, entries) in &mut grouped {
+            out.push('[');
+            out.push_str(class);
+            out.push_str("]\n");
+
+            entries.sort_by(|(ip_a, tries_a), (ip_b, tries_b)| {
+                tries_b
+                    .cmp(tries_a)
+                    .then_with(|| ip_a.to_string().cmp(&ip_b.to_string()))
+            });
+
+            for (ip, tries) in entries {
+                out.push_str(&format!("{ip}-{tries}\n"));
+            }
+        }
+
+        out
+    }
+
+    fn cleanup_if_needed(inner: &mut BeobachtenInner, now: Instant, ttl: Duration) {
+        let should_cleanup = match inner.last_cleanup {
+            Some(last) => now.saturating_duration_since(last) >= CLEANUP_INTERVAL,
+            None => true,
+        };
+        if should_cleanup {
+            Self::cleanup(inner, now, ttl);
+            inner.last_cleanup = Some(now);
+        }
+    }
+
+    fn cleanup(inner: &mut BeobachtenInner, now: Instant, ttl: Duration) {
+        inner.entries.retain(|_, entry| {
+            now.saturating_duration_since(entry.last_seen) <= ttl
+        });
+    }
+}

src/stats/telemetry.rs

@@ -0,0 +1,29 @@
+use crate::config::{MeTelemetryLevel, TelemetryConfig};
+
+/// Runtime telemetry policy used by hot-path counters.
+#[derive(Debug, Clone, Copy, PartialEq, Eq)]
+pub struct TelemetryPolicy {
+    pub core_enabled: bool,
+    pub user_enabled: bool,
+    pub me_level: MeTelemetryLevel,
+}
+
+impl Default for TelemetryPolicy {
+    fn default() -> Self {
+        Self {
+            core_enabled: true,
+            user_enabled: true,
+            me_level: MeTelemetryLevel::Normal,
+        }
+    }
+}
+
+impl TelemetryPolicy {
+    pub fn from_config(cfg: &TelemetryConfig) -> Self {
+        Self {
+            core_enabled: cfg.core_enabled,
+            user_enabled: cfg.user_enabled,
+            me_level: cfg.me_level,
+        }
+    }
+}

src/stream/buffer_pool.rs

@@ -3,6 +3,8 @@
 //! This module provides a thread-safe pool of BytesMut buffers
 //! that can be reused across connections to reduce allocation pressure.
 
+#![allow(dead_code)]
+
 use bytes::BytesMut;
 use crossbeam_queue::ArrayQueue;
 use std::ops::{Deref, DerefMut};
@@ -11,8 +13,9 @@ use std::sync::Arc;
 
 // ============= Configuration =============
 
-/// Default buffer size (64KB - good for MTProto)
-pub const DEFAULT_BUFFER_SIZE: usize = 64 * 1024;
+/// Default buffer size
+/// CHANGED: Reduced from 64KB to 16KB to match TLS record size and prevent bufferbloat.
+pub const DEFAULT_BUFFER_SIZE: usize = 16 * 1024;
 
 /// Default maximum number of pooled buffers
 pub const DEFAULT_MAX_BUFFERS: usize = 1024;
@@ -380,9 +383,14 @@ mod tests {
         // Add a buffer to pool
         pool.preallocate(1);
         
-        // Now try_get should succeed
-        assert!(pool.try_get().is_some());
+        // Now try_get should succeed once while the buffer is held
+        let buf = pool.try_get();
+        assert!(buf.is_some());
+        // While buffer is held, pool is empty
         assert!(pool.try_get().is_none());
+        // Drop buffer -> returns to pool, should be obtainable again
+        drop(buf);
+        assert!(pool.try_get().is_some());
     }
     
     #[test]
@@ -447,4 +455,4 @@ mod tests {
         // All buffers should be returned
         assert!(stats.pooled > 0);
     }
-}
+}

src/stream/frame.rs

@@ -3,10 +3,14 @@
 //! This module defines the common types and traits used by all
 //! frame encoding/decoding implementations.
 
+#![allow(dead_code)]
+
 use bytes::{Bytes, BytesMut};
 use std::io::Result;
+use std::sync::Arc;
 
 use crate::protocol::constants::ProtoTag;
+use crate::crypto::SecureRandom;
 
 // ============= Frame Types =============
 
@@ -147,11 +151,11 @@ pub trait FrameCodec: Send + Sync {
 // ============= Codec Factory =============
 
 /// Create a frame codec for the given protocol tag
-pub fn create_codec(proto_tag: ProtoTag) -> Box<dyn FrameCodec> {
+pub fn create_codec(proto_tag: ProtoTag, rng: Arc<SecureRandom>) -> Box<dyn FrameCodec> {
     match proto_tag {
         ProtoTag::Abridged => Box::new(crate::stream::frame_codec::AbridgedCodec::new()),
         ProtoTag::Intermediate => Box::new(crate::stream::frame_codec::IntermediateCodec::new()),
-        ProtoTag::Secure => Box::new(crate::stream::frame_codec::SecureCodec::new()),
+        ProtoTag::Secure => Box::new(crate::stream::frame_codec::SecureCodec::new(rng)),
     }
 }
 

src/stream/frame_codec.rs

@@ -3,11 +3,17 @@
 //! This module provides Encoder/Decoder implementations compatible
 //! with tokio-util's Framed wrapper for easy async frame I/O.
 
+#![allow(dead_code)]
+
 use bytes::{Bytes, BytesMut, BufMut};
 use std::io::{self, Error, ErrorKind};
+use std::sync::Arc;
 use tokio_util::codec::{Decoder, Encoder};
 
-use crate::protocol::constants::ProtoTag;
+use crate::protocol::constants::{
+    ProtoTag, is_valid_secure_payload_len, secure_padding_len, secure_payload_len_from_wire_len,
+};
+use crate::crypto::SecureRandom;
 use super::frame::{Frame, FrameMeta, FrameCodec as FrameCodecTrait};
 
 // ============= Unified Codec =============
@@ -21,14 +27,17 @@ pub struct FrameCodec {
     proto_tag: ProtoTag,
     /// Maximum allowed frame size
     max_frame_size: usize,
+    /// RNG for secure padding
+    rng: Arc<SecureRandom>,
 }
 
 impl FrameCodec {
     /// Create a new codec for the given protocol
-    pub fn new(proto_tag: ProtoTag) -> Self {
+    pub fn new(proto_tag: ProtoTag, rng: Arc<SecureRandom>) -> Self {
         Self {
             proto_tag,
             max_frame_size: 16 * 1024 * 1024, // 16MB default
+            rng,
         }
     }
     
@@ -64,7 +73,7 @@ impl Encoder<Frame> for FrameCodec {
         match self.proto_tag {
             ProtoTag::Abridged => encode_abridged(&frame, dst),
             ProtoTag::Intermediate => encode_intermediate(&frame, dst),
-            ProtoTag::Secure => encode_secure(&frame, dst),
+            ProtoTag::Secure => encode_secure(&frame, dst, &self.rng),
         }
     }
 }
@@ -130,7 +139,7 @@ fn encode_abridged(frame: &Frame, dst: &mut BytesMut) -> io::Result<()> {
     let data = &frame.data;
     
     // Validate alignment
-    if data.len() % 4 != 0 {
+    if !data.len().is_multiple_of(4) {
         return Err(Error::new(
             ErrorKind::InvalidInput,
             format!("abridged frame must be 4-byte aligned, got {} bytes", data.len())
@@ -269,13 +278,13 @@ fn decode_secure(src: &mut BytesMut, max_size: usize) -> io::Result<Option<Frame
         return Ok(None);
     }
     
-    // Calculate padding (indicated by length not divisible by 4)
-    let padding_len = len % 4;
-    let data_len = if padding_len != 0 {
-        len - padding_len
-    } else {
-        len
-    };
+    let data_len = secure_payload_len_from_wire_len(len).ok_or_else(|| {
+        Error::new(
+            ErrorKind::InvalidData,
+            format!("invalid secure frame length: {len}"),
+        )
+    })?;
+    let padding_len = len - data_len;
     
     meta.padding_len = padding_len as u8;
     
@@ -288,9 +297,7 @@ fn decode_secure(src: &mut BytesMut, max_size: usize) -> io::Result<Option<Frame
     Ok(Some(Frame::with_meta(data, meta)))
 }
 
-fn encode_secure(frame: &Frame, dst: &mut BytesMut) -> io::Result<()> {
-    use crate::crypto::random::SECURE_RANDOM;
-    
+fn encode_secure(frame: &Frame, dst: &mut BytesMut, rng: &SecureRandom) -> io::Result<()> {
     let data = &frame.data;
     
     // Simple ACK: just send data
@@ -300,14 +307,15 @@ fn encode_secure(frame: &Frame, dst: &mut BytesMut) -> io::Result<()> {
         return Ok(());
     }
     
-    // Generate padding to make length not divisible by 4
-    let padding_len = if data.len() % 4 == 0 {
-        // Add 1-3 bytes to make it non-aligned
-        (SECURE_RANDOM.range(3) + 1) as usize
-    } else {
-        // Already non-aligned, can add 0-3
-        SECURE_RANDOM.range(4) as usize
-    };
+    if !is_valid_secure_payload_len(data.len()) {
+        return Err(Error::new(
+            ErrorKind::InvalidData,
+            format!("secure payload must be 4-byte aligned, got {}", data.len()),
+        ));
+    }
+
+    // Generate padding that keeps total length non-divisible by 4.
+    let padding_len = secure_padding_len(data.len(), rng);
     
     let total_len = data.len() + padding_len;
     dst.reserve(4 + total_len);
@@ -321,7 +329,7 @@ fn encode_secure(frame: &Frame, dst: &mut BytesMut) -> io::Result<()> {
     dst.extend_from_slice(data);
     
     if padding_len > 0 {
-        let padding = SECURE_RANDOM.bytes(padding_len);
+        let padding = rng.bytes(padding_len);
         dst.extend_from_slice(&padding);
     }
     
@@ -445,19 +453,21 @@ impl FrameCodecTrait for IntermediateCodec {
 /// Secure Intermediate protocol codec
 pub struct SecureCodec {
     max_frame_size: usize,
+    rng: Arc<SecureRandom>,
 }
 
 impl SecureCodec {
-    pub fn new() -> Self {
+    pub fn new(rng: Arc<SecureRandom>) -> Self {
         Self {
             max_frame_size: 16 * 1024 * 1024,
+            rng,
         }
     }
 }
 
 impl Default for SecureCodec {
     fn default() -> Self {
-        Self::new()
+        Self::new(Arc::new(SecureRandom::new()))
     }
 }
 
@@ -474,7 +484,7 @@ impl Encoder<Frame> for SecureCodec {
     type Error = io::Error;
     
     fn encode(&mut self, frame: Frame, dst: &mut BytesMut) -> Result<(), Self::Error> {
-        encode_secure(&frame, dst)
+        encode_secure(&frame, dst, &self.rng)
     }
 }
 
@@ -485,7 +495,7 @@ impl FrameCodecTrait for SecureCodec {
     
     fn encode(&self, frame: &Frame, dst: &mut BytesMut) -> io::Result<usize> {
         let before = dst.len();
-        encode_secure(frame, dst)?;
+        encode_secure(frame, dst, &self.rng)?;
         Ok(dst.len() - before)
     }
     
@@ -506,6 +516,8 @@ mod tests {
     use tokio_util::codec::{FramedRead, FramedWrite};
     use tokio::io::duplex;
     use futures::{SinkExt, StreamExt};
+    use crate::crypto::SecureRandom;
+    use std::sync::Arc;
     
     #[tokio::test]
     async fn test_framed_abridged() {
@@ -541,8 +553,8 @@ mod tests {
     async fn test_framed_secure() {
         let (client, server) = duplex(4096);
         
-        let mut writer = FramedWrite::new(client, SecureCodec::new());
-        let mut reader = FramedRead::new(server, SecureCodec::new());
+        let mut writer = FramedWrite::new(client, SecureCodec::new(Arc::new(SecureRandom::new())));
+        let mut reader = FramedRead::new(server, SecureCodec::new(Arc::new(SecureRandom::new())));
         
         let original = Bytes::from_static(&[1, 2, 3, 4, 5, 6, 7, 8]);
         let frame = Frame::new(original.clone());
@@ -557,8 +569,8 @@ mod tests {
         for proto_tag in [ProtoTag::Abridged, ProtoTag::Intermediate, ProtoTag::Secure] {
             let (client, server) = duplex(4096);
             
-            let mut writer = FramedWrite::new(client, FrameCodec::new(proto_tag));
-            let mut reader = FramedRead::new(server, FrameCodec::new(proto_tag));
+            let mut writer = FramedWrite::new(client, FrameCodec::new(proto_tag, Arc::new(SecureRandom::new())));
+            let mut reader = FramedRead::new(server, FrameCodec::new(proto_tag, Arc::new(SecureRandom::new())));
             
             // Use 4-byte aligned data for abridged compatibility
             let original = Bytes::from_static(&[1, 2, 3, 4, 5, 6, 7, 8]);
@@ -607,7 +619,7 @@ mod tests {
     
     #[test]
     fn test_frame_too_large() {
-        let mut codec = FrameCodec::new(ProtoTag::Intermediate)
+        let mut codec = FrameCodec::new(ProtoTag::Intermediate, Arc::new(SecureRandom::new()))
             .with_max_frame_size(100);
         
         // Create a "frame" that claims to be very large
@@ -618,4 +630,4 @@ mod tests {
         let result = codec.decode(&mut buf);
         assert!(result.is_err());
     }
-}
+}

src/stream/frame_stream.rs

@@ -1,11 +1,13 @@
 //! MTProto frame stream wrappers
 
-use bytes::{Bytes, BytesMut};
+#![allow(dead_code)]
+
+use bytes::Bytes;
 use std::io::{Error, ErrorKind, Result};
 use tokio::io::{AsyncRead, AsyncWrite, AsyncReadExt, AsyncWriteExt};
 use crate::protocol::constants::*;
-use crate::crypto::crc32;
-use crate::crypto::random::SECURE_RANDOM;
+use crate::crypto::{crc32, SecureRandom};
+use std::sync::Arc;
 use super::traits::{FrameMeta, LayeredStream};
 
 // ============= Abridged (Compact) Frame =============
@@ -76,7 +78,7 @@ impl<W> AbridgedFrameWriter<W> {
 impl<W: AsyncWrite + Unpin> AbridgedFrameWriter<W> {
     /// Write a frame
     pub async fn write_frame(&mut self, data: &[u8], meta: &FrameMeta) -> Result<()> {
-        if data.len() % 4 != 0 {
+        if !data.len().is_multiple_of(4) {
             return Err(Error::new(
                 ErrorKind::InvalidInput,
                 format!("Abridged frame must be aligned to 4 bytes, got {}", data.len()),
@@ -232,11 +234,13 @@ impl<R: AsyncRead + Unpin> SecureIntermediateFrameReader<R> {
         let mut data = vec![0u8; len];
         self.upstream.read_exact(&mut data).await?;
         
-        // Strip padding (not aligned to 4)
-        if len % 4 != 0 {
-            let actual_len = len - (len % 4);
-            data.truncate(actual_len);
-        }
+        let payload_len = secure_payload_len_from_wire_len(len).ok_or_else(|| {
+            Error::new(
+                ErrorKind::InvalidData,
+                format!("Invalid secure frame length: {len}"),
+            )
+        })?;
+        data.truncate(payload_len);
         
         Ok((Bytes::from(data), meta))
     }
@@ -251,11 +255,12 @@ impl<R> LayeredStream<R> for SecureIntermediateFrameReader<R> {
 /// Writer for secure intermediate MTProto framing
 pub struct SecureIntermediateFrameWriter<W> {
     upstream: W,
+    rng: Arc<SecureRandom>,
 }
 
 impl<W> SecureIntermediateFrameWriter<W> {
-    pub fn new(upstream: W) -> Self {
-        Self { upstream }
+    pub fn new(upstream: W, rng: Arc<SecureRandom>) -> Self {
+        Self { upstream, rng }
     }
 }
 
@@ -266,9 +271,16 @@ impl<W: AsyncWrite + Unpin> SecureIntermediateFrameWriter<W> {
             return Ok(());
         }
         
-        // Add random padding (0-3 bytes)
-        let padding_len = SECURE_RANDOM.range(4);
-        let padding = SECURE_RANDOM.bytes(padding_len);
+        if !is_valid_secure_payload_len(data.len()) {
+            return Err(Error::new(
+                ErrorKind::InvalidData,
+                format!("Secure payload must be 4-byte aligned, got {}", data.len()),
+            ));
+        }
+
+        // Add padding so total length is never divisible by 4 (MTProto Secure)
+        let padding_len = secure_padding_len(data.len(), &self.rng);
+        let padding = self.rng.bytes(padding_len);
         
         let total_len = data.len() + padding_len;
         let len_bytes = (total_len as u32).to_le_bytes();
@@ -319,7 +331,7 @@ impl<R: AsyncRead + Unpin> MtprotoFrameReader<R> {
             }
             
             // Validate length
-            if len < MIN_MSG_LEN || len > MAX_MSG_LEN || len % PADDING_FILLER.len() != 0 {
+            if !(MIN_MSG_LEN..=MAX_MSG_LEN).contains(&len) || !len.is_multiple_of(PADDING_FILLER.len()) {
                 return Err(Error::new(
                     ErrorKind::InvalidData,
                     format!("Invalid message length: {}", len),
@@ -454,11 +466,11 @@ pub enum FrameWriterKind<W> {
 }
 
 impl<W: AsyncWrite + Unpin> FrameWriterKind<W> {
-    pub fn new(upstream: W, proto_tag: ProtoTag) -> Self {
+    pub fn new(upstream: W, proto_tag: ProtoTag, rng: Arc<SecureRandom>) -> Self {
         match proto_tag {
             ProtoTag::Abridged => FrameWriterKind::Abridged(AbridgedFrameWriter::new(upstream)),
             ProtoTag::Intermediate => FrameWriterKind::Intermediate(IntermediateFrameWriter::new(upstream)),
-            ProtoTag::Secure => FrameWriterKind::SecureIntermediate(SecureIntermediateFrameWriter::new(upstream)),
+            ProtoTag::Secure => FrameWriterKind::SecureIntermediate(SecureIntermediateFrameWriter::new(upstream, rng)),
         }
     }
     
@@ -483,6 +495,8 @@ impl<W: AsyncWrite + Unpin> FrameWriterKind<W> {
 mod tests {
     use super::*;
     use tokio::io::duplex;
+    use std::sync::Arc;
+    use crate::crypto::SecureRandom;
     
     #[tokio::test]
     async fn test_abridged_roundtrip() {
@@ -539,7 +553,7 @@ mod tests {
     async fn test_secure_intermediate_padding() {
         let (client, server) = duplex(1024);
         
-        let mut writer = SecureIntermediateFrameWriter::new(client);
+        let mut writer = SecureIntermediateFrameWriter::new(client, Arc::new(SecureRandom::new()));
         let mut reader = SecureIntermediateFrameReader::new(server);
         
         let data = vec![1u8, 2, 3, 4, 5, 6, 7, 8];
@@ -547,9 +561,7 @@ mod tests {
         writer.flush().await.unwrap();
         
         let (received, _meta) = reader.read_frame().await.unwrap();
-        // Received should have padding stripped to align to 4
-        let expected_len = (data.len() / 4) * 4;
-        assert_eq!(received.len(), expected_len);
+        assert_eq!(received.len(), data.len());
     }
     
     #[tokio::test]
@@ -572,7 +584,7 @@ mod tests {
     async fn test_frame_reader_kind() {
         let (client, server) = duplex(1024);
         
-        let mut writer = FrameWriterKind::new(client, ProtoTag::Intermediate);
+        let mut writer = FrameWriterKind::new(client, ProtoTag::Intermediate, Arc::new(SecureRandom::new()));
         let mut reader = FrameReaderKind::new(server, ProtoTag::Intermediate);
         
         let data = vec![1u8, 2, 3, 4];
@@ -582,4 +594,4 @@ mod tests {
         let (received, _) = reader.read_frame().await.unwrap();
         assert_eq!(&received[..], &data[..]);
     }
-}
+}

src/stream/mod.rs

@@ -12,32 +12,38 @@ pub mod frame_codec;
 pub mod frame_stream;
 
 // Re-export state machine types
+#[allow(unused_imports)]
 pub use state::{
     StreamState, Transition, PollResult,
     ReadBuffer, WriteBuffer, HeaderBuffer, YieldBuffer,
 };
 
 // Re-export buffer pool
+#[allow(unused_imports)]
 pub use buffer_pool::{BufferPool, PooledBuffer, PoolStats};
 
 // Re-export stream implementations
+#[allow(unused_imports)]
 pub use crypto_stream::{CryptoReader, CryptoWriter, PassthroughStream};
 pub use tls_stream::{FakeTlsReader, FakeTlsWriter};
 
 // Re-export frame types
+#[allow(unused_imports)]
 pub use frame::{Frame, FrameMeta, FrameCodec as FrameCodecTrait, create_codec};
 
-// Re-export tokio-util compatible codecs  
+// Re-export tokio-util compatible codecs
+#[allow(unused_imports)]
 pub use frame_codec::{
     FrameCodec,
     AbridgedCodec, IntermediateCodec, SecureCodec,
 };
 
 // Legacy re-exports for compatibility
+#[allow(unused_imports)]
 pub use frame_stream::{
     AbridgedFrameReader, AbridgedFrameWriter,
     IntermediateFrameReader, IntermediateFrameWriter,
     SecureIntermediateFrameReader, SecureIntermediateFrameWriter,
     MtprotoFrameReader, MtprotoFrameWriter,
     FrameReaderKind, FrameWriterKind,
-};
+};

src/stream/state.rs

@@ -3,6 +3,8 @@
 //! This module provides core types and traits for implementing
 //! stateful async streams with proper partial read/write handling.
 
+#![allow(dead_code)]
+
 use bytes::{Bytes, BytesMut};
 use std::io;
 

src/stream/traits.rs

@@ -1,5 +1,7 @@
 //! Stream traits and common types
 
+#![allow(dead_code)]
+
 use bytes::Bytes;
 use std::io::Result;
 use std::pin::Pin;

src/tls_front/cache.rs

@@ -0,0 +1,255 @@
+use std::collections::HashMap;
+use std::net::IpAddr;
+use std::path::{Path, PathBuf};
+use std::sync::Arc;
+use std::time::{Duration, Instant, SystemTime};
+
+use tokio::sync::RwLock;
+use tokio::time::sleep;
+use tracing::{debug, warn, info};
+
+use crate::tls_front::types::{CachedTlsData, ParsedServerHello, TlsFetchResult};
+
+/// Lightweight in-memory + optional on-disk cache for TLS fronting data.
+#[derive(Debug)]
+pub struct TlsFrontCache {
+    memory: RwLock<HashMap<String, Arc<CachedTlsData>>>,
+    default: Arc<CachedTlsData>,
+    full_cert_sent: RwLock<HashMap<IpAddr, Instant>>,
+    disk_path: PathBuf,
+}
+
+#[allow(dead_code)]
+impl TlsFrontCache {
+    pub fn new(domains: &[String], default_len: usize, disk_path: impl AsRef<Path>) -> Self {
+        let default_template = ParsedServerHello {
+            version: [0x03, 0x03],
+            random: [0u8; 32],
+            session_id: Vec::new(),
+            cipher_suite: [0x13, 0x01],
+            compression: 0,
+            extensions: Vec::new(),
+        };
+
+        let default = Arc::new(CachedTlsData {
+            server_hello_template: default_template,
+            cert_info: None,
+            cert_payload: None,
+            app_data_records_sizes: vec![default_len],
+            total_app_data_len: default_len,
+            fetched_at: SystemTime::now(),
+            domain: "default".to_string(),
+        });
+
+        let mut map = HashMap::new();
+        for d in domains {
+            map.insert(d.clone(), default.clone());
+        }
+
+        Self {
+            memory: RwLock::new(map),
+            default,
+            full_cert_sent: RwLock::new(HashMap::new()),
+            disk_path: disk_path.as_ref().to_path_buf(),
+        }
+    }
+
+    pub async fn get(&self, sni: &str) -> Arc<CachedTlsData> {
+        let guard = self.memory.read().await;
+        guard.get(sni).cloned().unwrap_or_else(|| self.default.clone())
+    }
+
+    pub async fn contains_domain(&self, domain: &str) -> bool {
+        self.memory.read().await.contains_key(domain)
+    }
+
+    /// Returns true when full cert payload should be sent for client_ip
+    /// according to TTL policy.
+    pub async fn take_full_cert_budget_for_ip(
+        &self,
+        client_ip: IpAddr,
+        ttl: Duration,
+    ) -> bool {
+        if ttl.is_zero() {
+            self.full_cert_sent
+                .write()
+                .await
+                .insert(client_ip, Instant::now());
+            return true;
+        }
+
+        let now = Instant::now();
+        let mut guard = self.full_cert_sent.write().await;
+        guard.retain(|_, seen_at| now.duration_since(*seen_at) < ttl);
+
+        match guard.get_mut(&client_ip) {
+            Some(seen_at) => {
+                if now.duration_since(*seen_at) >= ttl {
+                    *seen_at = now;
+                    true
+                } else {
+                    false
+                }
+            }
+            None => {
+                guard.insert(client_ip, now);
+                true
+            }
+        }
+    }
+
+    pub async fn set(&self, domain: &str, data: CachedTlsData) {
+        let mut guard = self.memory.write().await;
+        guard.insert(domain.to_string(), Arc::new(data));
+    }
+
+    pub async fn load_from_disk(&self) {
+        let path = self.disk_path.clone();
+        if tokio::fs::create_dir_all(&path).await.is_err() {
+            return;
+        }
+        let mut loaded = 0usize;
+        if let Ok(mut dir) = tokio::fs::read_dir(&path).await {
+            while let Ok(Some(entry)) = dir.next_entry().await {
+                if let Ok(name) = entry.file_name().into_string() {
+                    if !name.ends_with(".json") {
+                        continue;
+                    }
+                    if let Ok(data) = tokio::fs::read(entry.path()).await
+                        && let Ok(mut cached) = serde_json::from_slice::<CachedTlsData>(&data)
+                    {
+                        if cached.domain.is_empty()
+                            || cached.domain.len() > 255
+                            || !cached.domain.chars().all(|c| c.is_ascii_alphanumeric() || c == '.' || c == '-')
+                        {
+                            warn!(file = %name, "Skipping TLS cache entry with invalid domain");
+                            continue;
+                        }
+                        // fetched_at is skipped during deserialization; approximate with file mtime if available.
+                        if let Ok(meta) = entry.metadata().await
+                            && let Ok(modified) = meta.modified()
+                        {
+                            cached.fetched_at = modified;
+                        }
+                        // Drop entries older than 72h
+                        if let Ok(age) = cached.fetched_at.elapsed()
+                            && age > Duration::from_secs(72 * 3600)
+                        {
+                            warn!(domain = %cached.domain, "Skipping stale TLS cache entry (>72h)");
+                            continue;
+                        }
+                        let domain = cached.domain.clone();
+                        self.set(&domain, cached).await;
+                        loaded += 1;
+                    }
+                }
+            }
+        }
+        if loaded > 0 {
+            info!(count = loaded, "Loaded TLS cache entries from disk");
+        }
+    }
+
+    async fn persist(&self, domain: &str, data: &CachedTlsData) {
+        if tokio::fs::create_dir_all(&self.disk_path).await.is_err() {
+            return;
+        }
+        let fname = format!("{}.json", domain.replace(['/', '\\'], "_"));
+        let path = self.disk_path.join(fname);
+        if let Ok(json) = serde_json::to_vec_pretty(data) {
+            // best-effort write
+            let _ = tokio::fs::write(path, json).await;
+        }
+    }
+
+    /// Spawn background updater that periodically refreshes cached domains using provided fetcher.
+    pub fn spawn_updater<F>(
+        self: Arc<Self>,
+        domains: Vec<String>,
+        interval: Duration,
+        fetcher: F,
+    ) where
+        F: Fn(String) -> tokio::task::JoinHandle<()> + Send + Sync + 'static,
+    {
+        tokio::spawn(async move {
+            loop {
+                for domain in &domains {
+                    let _ = fetcher(domain.clone()).await;
+                }
+                sleep(interval).await;
+            }
+        });
+    }
+
+    /// Replace cached entry from a fetch result.
+    pub async fn update_from_fetch(&self, domain: &str, fetched: TlsFetchResult) {
+        let data = CachedTlsData {
+            server_hello_template: fetched.server_hello_parsed,
+            cert_info: fetched.cert_info,
+            cert_payload: fetched.cert_payload,
+            app_data_records_sizes: fetched.app_data_records_sizes.clone(),
+            total_app_data_len: fetched.total_app_data_len,
+            fetched_at: SystemTime::now(),
+            domain: domain.to_string(),
+        };
+
+        self.set(domain, data.clone()).await;
+        self.persist(domain, &data).await;
+        debug!(domain = %domain, len = fetched.total_app_data_len, "TLS cache updated");
+    }
+
+    pub fn default_entry(&self) -> Arc<CachedTlsData> {
+        self.default.clone()
+    }
+
+    pub fn disk_path(&self) -> &Path {
+        &self.disk_path
+    }
+}
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+
+    #[tokio::test]
+    async fn test_take_full_cert_budget_for_ip_uses_ttl() {
+        let cache = TlsFrontCache::new(
+            &["example.com".to_string()],
+            1024,
+            "tlsfront-test-cache",
+        );
+        let ip: IpAddr = "127.0.0.1".parse().expect("ip");
+        let ttl = Duration::from_millis(80);
+
+        assert!(cache
+            .take_full_cert_budget_for_ip(ip, ttl)
+            .await);
+        assert!(!cache
+            .take_full_cert_budget_for_ip(ip, ttl)
+            .await);
+
+        tokio::time::sleep(Duration::from_millis(90)).await;
+
+        assert!(cache
+            .take_full_cert_budget_for_ip(ip, ttl)
+            .await);
+    }
+
+    #[tokio::test]
+    async fn test_take_full_cert_budget_for_ip_zero_ttl_always_allows_full_payload() {
+        let cache = TlsFrontCache::new(
+            &["example.com".to_string()],
+            1024,
+            "tlsfront-test-cache",
+        );
+        let ip: IpAddr = "127.0.0.1".parse().expect("ip");
+        let ttl = Duration::ZERO;
+
+        assert!(cache
+            .take_full_cert_budget_for_ip(ip, ttl)
+            .await);
+        assert!(cache
+            .take_full_cert_budget_for_ip(ip, ttl)
+            .await);
+    }
+}

src/tls_front/emulator.rs

@@ -0,0 +1,388 @@
+use crate::crypto::{sha256_hmac, SecureRandom};
+use crate::protocol::constants::{
+    TLS_RECORD_APPLICATION, TLS_RECORD_CHANGE_CIPHER, TLS_RECORD_HANDSHAKE, TLS_VERSION,
+};
+use crate::protocol::tls::{TLS_DIGEST_LEN, TLS_DIGEST_POS, gen_fake_x25519_key};
+use crate::tls_front::types::{CachedTlsData, ParsedCertificateInfo};
+
+const MIN_APP_DATA: usize = 64;
+const MAX_APP_DATA: usize = 16640; // RFC 8446 §5.2 allows up to 2^14 + 256
+
+fn jitter_and_clamp_sizes(sizes: &[usize], rng: &SecureRandom) -> Vec<usize> {
+    sizes
+        .iter()
+        .map(|&size| {
+            let base = size.clamp(MIN_APP_DATA, MAX_APP_DATA);
+            let jitter_range = ((base as f64) * 0.03).round() as i64;
+            if jitter_range == 0 {
+                return base;
+            }
+            let mut rand_bytes = [0u8; 2];
+            rand_bytes.copy_from_slice(&rng.bytes(2));
+            let span = 2 * jitter_range + 1;
+            let delta = (u16::from_le_bytes(rand_bytes) as i64 % span) - jitter_range;
+            let adjusted = (base as i64 + delta).clamp(MIN_APP_DATA as i64, MAX_APP_DATA as i64);
+            adjusted as usize
+        })
+        .collect()
+}
+
+fn app_data_body_capacity(sizes: &[usize]) -> usize {
+    sizes.iter().map(|&size| size.saturating_sub(17)).sum()
+}
+
+fn ensure_payload_capacity(mut sizes: Vec<usize>, payload_len: usize) -> Vec<usize> {
+    if payload_len == 0 {
+        return sizes;
+    }
+
+    let mut body_total = app_data_body_capacity(&sizes);
+    if body_total >= payload_len {
+        return sizes;
+    }
+
+    if let Some(last) = sizes.last_mut() {
+        let free = MAX_APP_DATA.saturating_sub(*last);
+        let grow = free.min(payload_len - body_total);
+        *last += grow;
+        body_total += grow;
+    }
+
+    while body_total < payload_len {
+        let remaining = payload_len - body_total;
+        let chunk = (remaining + 17).clamp(MIN_APP_DATA, MAX_APP_DATA);
+        sizes.push(chunk);
+        body_total += chunk.saturating_sub(17);
+    }
+
+    sizes
+}
+
+fn build_compact_cert_info_payload(cert_info: &ParsedCertificateInfo) -> Option<Vec<u8>> {
+    let mut fields = Vec::new();
+
+    if let Some(subject) = cert_info.subject_cn.as_deref() {
+        fields.push(format!("CN={subject}"));
+    }
+    if let Some(issuer) = cert_info.issuer_cn.as_deref() {
+        fields.push(format!("ISSUER={issuer}"));
+    }
+    if let Some(not_before) = cert_info.not_before_unix {
+        fields.push(format!("NB={not_before}"));
+    }
+    if let Some(not_after) = cert_info.not_after_unix {
+        fields.push(format!("NA={not_after}"));
+    }
+    if !cert_info.san_names.is_empty() {
+        let san = cert_info
+            .san_names
+            .iter()
+            .take(8)
+            .map(String::as_str)
+            .collect::<Vec<_>>()
+            .join(",");
+        fields.push(format!("SAN={san}"));
+    }
+
+    if fields.is_empty() {
+        return None;
+    }
+
+    let mut payload = fields.join(";").into_bytes();
+    if payload.len() > 512 {
+        payload.truncate(512);
+    }
+    Some(payload)
+}
+
+/// Build a ServerHello + CCS + ApplicationData sequence using cached TLS metadata.
+pub fn build_emulated_server_hello(
+    secret: &[u8],
+    client_digest: &[u8; TLS_DIGEST_LEN],
+    session_id: &[u8],
+    cached: &CachedTlsData,
+    use_full_cert_payload: bool,
+    rng: &SecureRandom,
+    alpn: Option<Vec<u8>>,
+    new_session_tickets: u8,
+) -> Vec<u8> {
+    // --- ServerHello ---
+    let mut extensions = Vec::new();
+    // KeyShare (x25519)
+    let key = gen_fake_x25519_key(rng);
+    extensions.extend_from_slice(&0x0033u16.to_be_bytes()); // key_share
+    extensions.extend_from_slice(&(2 + 2 + 32u16).to_be_bytes()); // len
+    extensions.extend_from_slice(&0x001du16.to_be_bytes()); // X25519
+    extensions.extend_from_slice(&(32u16).to_be_bytes());
+    extensions.extend_from_slice(&key);
+    // supported_versions (TLS1.3)
+    extensions.extend_from_slice(&0x002bu16.to_be_bytes());
+    extensions.extend_from_slice(&(2u16).to_be_bytes());
+    extensions.extend_from_slice(&0x0304u16.to_be_bytes());
+    if let Some(alpn_proto) = &alpn {
+        extensions.extend_from_slice(&0x0010u16.to_be_bytes());
+        let list_len: u16 = 1 + alpn_proto.len() as u16;
+        let ext_len: u16 = 2 + list_len;
+        extensions.extend_from_slice(&ext_len.to_be_bytes());
+        extensions.extend_from_slice(&list_len.to_be_bytes());
+        extensions.push(alpn_proto.len() as u8);
+        extensions.extend_from_slice(alpn_proto);
+    }
+
+    let extensions_len = extensions.len() as u16;
+
+    let body_len = 2 + // version
+        32 + // random
+        1 + session_id.len() + // session id
+        2 + // cipher
+        1 + // compression
+        2 + extensions.len(); // extensions
+
+    let mut message = Vec::with_capacity(4 + body_len);
+    message.push(0x02); // ServerHello
+    let len_bytes = (body_len as u32).to_be_bytes();
+    message.extend_from_slice(&len_bytes[1..4]);
+    message.extend_from_slice(&cached.server_hello_template.version); // 0x0303
+    message.extend_from_slice(&[0u8; 32]); // random placeholder
+    message.push(session_id.len() as u8);
+    message.extend_from_slice(session_id);
+    let cipher = if cached.server_hello_template.cipher_suite == [0, 0] {
+        [0x13, 0x01]
+    } else {
+        cached.server_hello_template.cipher_suite
+    };
+    message.extend_from_slice(&cipher);
+    message.push(cached.server_hello_template.compression);
+    message.extend_from_slice(&extensions_len.to_be_bytes());
+    message.extend_from_slice(&extensions);
+
+    let mut server_hello = Vec::with_capacity(5 + message.len());
+    server_hello.push(TLS_RECORD_HANDSHAKE);
+    server_hello.extend_from_slice(&TLS_VERSION);
+    server_hello.extend_from_slice(&(message.len() as u16).to_be_bytes());
+    server_hello.extend_from_slice(&message);
+
+    // --- ChangeCipherSpec ---
+    let change_cipher_spec = [
+        TLS_RECORD_CHANGE_CIPHER,
+        TLS_VERSION[0],
+        TLS_VERSION[1],
+        0x00,
+        0x01,
+        0x01,
+    ];
+
+    // --- ApplicationData (fake encrypted records) ---
+    // Use the same number and sizes of ApplicationData records as the cached server.
+    let mut sizes = cached.app_data_records_sizes.clone();
+    if sizes.is_empty() {
+        sizes.push(cached.total_app_data_len.max(1024));
+    }
+    let mut sizes = jitter_and_clamp_sizes(&sizes, rng);
+    let compact_payload = cached
+        .cert_info
+        .as_ref()
+        .and_then(build_compact_cert_info_payload);
+    let selected_payload: Option<&[u8]> = if use_full_cert_payload {
+        cached
+            .cert_payload
+            .as_ref()
+            .map(|payload| payload.certificate_message.as_slice())
+            .filter(|payload| !payload.is_empty())
+            .or(compact_payload.as_deref())
+    } else {
+        compact_payload.as_deref()
+    };
+
+    if let Some(payload) = selected_payload {
+        sizes = ensure_payload_capacity(sizes, payload.len());
+    }
+
+    let mut app_data = Vec::new();
+    let mut payload_offset = 0usize;
+    for size in sizes {
+        let mut rec = Vec::with_capacity(5 + size);
+        rec.push(TLS_RECORD_APPLICATION);
+        rec.extend_from_slice(&TLS_VERSION);
+        rec.extend_from_slice(&(size as u16).to_be_bytes());
+
+        if let Some(payload) = selected_payload {
+            if size > 17 {
+                let body_len = size - 17;
+                let remaining = payload.len().saturating_sub(payload_offset);
+                let copy_len = remaining.min(body_len);
+                if copy_len > 0 {
+                    rec.extend_from_slice(&payload[payload_offset..payload_offset + copy_len]);
+                    payload_offset += copy_len;
+                }
+                if body_len > copy_len {
+                    rec.extend_from_slice(&rng.bytes(body_len - copy_len));
+                }
+                rec.push(0x16); // inner content type marker (handshake)
+                rec.extend_from_slice(&rng.bytes(16)); // AEAD-like tag
+            } else {
+                rec.extend_from_slice(&rng.bytes(size));
+            }
+        } else if size > 17 {
+            let body_len = size - 17;
+            rec.extend_from_slice(&rng.bytes(body_len));
+            rec.push(0x16); // inner content type marker (handshake)
+            rec.extend_from_slice(&rng.bytes(16)); // AEAD-like tag
+        } else {
+            rec.extend_from_slice(&rng.bytes(size));
+        }
+        app_data.extend_from_slice(&rec);
+    }
+
+    // --- Combine ---
+    // Optional NewSessionTicket mimic records (opaque ApplicationData for fingerprint).
+    let mut tickets = Vec::new();
+    if new_session_tickets > 0 {
+        for _ in 0..new_session_tickets {
+            let ticket_len: usize = rng.range(48) + 48;
+            let mut rec = Vec::with_capacity(5 + ticket_len);
+            rec.push(TLS_RECORD_APPLICATION);
+            rec.extend_from_slice(&TLS_VERSION);
+            rec.extend_from_slice(&(ticket_len as u16).to_be_bytes());
+            rec.extend_from_slice(&rng.bytes(ticket_len));
+            tickets.extend_from_slice(&rec);
+        }
+    }
+
+    let mut response = Vec::with_capacity(server_hello.len() + change_cipher_spec.len() + app_data.len() + tickets.len());
+    response.extend_from_slice(&server_hello);
+    response.extend_from_slice(&change_cipher_spec);
+    response.extend_from_slice(&app_data);
+    response.extend_from_slice(&tickets);
+
+    // --- HMAC ---
+    let mut hmac_input = Vec::with_capacity(TLS_DIGEST_LEN + response.len());
+    hmac_input.extend_from_slice(client_digest);
+    hmac_input.extend_from_slice(&response);
+    let digest = sha256_hmac(secret, &hmac_input);
+    response[TLS_DIGEST_POS..TLS_DIGEST_POS + TLS_DIGEST_LEN].copy_from_slice(&digest);
+
+    response
+}
+
+#[cfg(test)]
+mod tests {
+    use std::time::SystemTime;
+
+    use crate::tls_front::types::{CachedTlsData, ParsedServerHello, TlsCertPayload};
+
+    use super::build_emulated_server_hello;
+    use crate::crypto::SecureRandom;
+    use crate::protocol::constants::{
+        TLS_RECORD_APPLICATION, TLS_RECORD_CHANGE_CIPHER, TLS_RECORD_HANDSHAKE,
+    };
+
+    fn first_app_data_payload(response: &[u8]) -> &[u8] {
+        let hello_len = u16::from_be_bytes([response[3], response[4]]) as usize;
+        let ccs_start = 5 + hello_len;
+        let ccs_len = u16::from_be_bytes([response[ccs_start + 3], response[ccs_start + 4]]) as usize;
+        let app_start = ccs_start + 5 + ccs_len;
+        let app_len = u16::from_be_bytes([response[app_start + 3], response[app_start + 4]]) as usize;
+        &response[app_start + 5..app_start + 5 + app_len]
+    }
+
+    fn make_cached(cert_payload: Option<TlsCertPayload>) -> CachedTlsData {
+        CachedTlsData {
+            server_hello_template: ParsedServerHello {
+                version: [0x03, 0x03],
+                random: [0u8; 32],
+                session_id: Vec::new(),
+                cipher_suite: [0x13, 0x01],
+                compression: 0,
+                extensions: Vec::new(),
+            },
+            cert_info: None,
+            cert_payload,
+            app_data_records_sizes: vec![64],
+            total_app_data_len: 64,
+            fetched_at: SystemTime::now(),
+            domain: "example.com".to_string(),
+        }
+    }
+
+    #[test]
+    fn test_build_emulated_server_hello_uses_cached_cert_payload() {
+        let cert_msg = vec![0x0b, 0x00, 0x00, 0x05, 0x00, 0xaa, 0xbb, 0xcc, 0xdd];
+        let cached = make_cached(Some(TlsCertPayload {
+            cert_chain_der: vec![vec![0x30, 0x01, 0x00]],
+            certificate_message: cert_msg.clone(),
+        }));
+        let rng = SecureRandom::new();
+        let response = build_emulated_server_hello(
+            b"secret",
+            &[0x11; 32],
+            &[0x22; 16],
+            &cached,
+            true,
+            &rng,
+            None,
+            0,
+        );
+
+        assert_eq!(response[0], TLS_RECORD_HANDSHAKE);
+        let hello_len = u16::from_be_bytes([response[3], response[4]]) as usize;
+        let ccs_start = 5 + hello_len;
+        assert_eq!(response[ccs_start], TLS_RECORD_CHANGE_CIPHER);
+        let app_start = ccs_start + 6;
+        assert_eq!(response[app_start], TLS_RECORD_APPLICATION);
+
+        let payload = first_app_data_payload(&response);
+        assert!(payload.starts_with(&cert_msg));
+    }
+
+    #[test]
+    fn test_build_emulated_server_hello_random_fallback_when_no_cert_payload() {
+        let cached = make_cached(None);
+        let rng = SecureRandom::new();
+        let response = build_emulated_server_hello(
+            b"secret",
+            &[0x22; 32],
+            &[0x33; 16],
+            &cached,
+            true,
+            &rng,
+            None,
+            0,
+        );
+
+        let payload = first_app_data_payload(&response);
+        assert!(payload.len() >= 64);
+        assert_eq!(payload[payload.len() - 17], 0x16);
+    }
+
+    #[test]
+    fn test_build_emulated_server_hello_uses_compact_payload_after_first() {
+        let cert_msg = vec![0x0b, 0x00, 0x00, 0x05, 0x00, 0xaa, 0xbb, 0xcc, 0xdd];
+        let mut cached = make_cached(Some(TlsCertPayload {
+            cert_chain_der: vec![vec![0x30, 0x01, 0x00]],
+            certificate_message: cert_msg,
+        }));
+        cached.cert_info = Some(crate::tls_front::types::ParsedCertificateInfo {
+            not_after_unix: Some(1_900_000_000),
+            not_before_unix: Some(1_700_000_000),
+            issuer_cn: Some("Issuer".to_string()),
+            subject_cn: Some("example.com".to_string()),
+            san_names: vec!["example.com".to_string(), "www.example.com".to_string()],
+        });
+
+        let rng = SecureRandom::new();
+        let response = build_emulated_server_hello(
+            b"secret",
+            &[0x44; 32],
+            &[0x55; 16],
+            &cached,
+            false,
+            &rng,
+            None,
+            0,
+        );
+
+        let payload = first_app_data_payload(&response);
+        assert!(payload.starts_with(b"CN=example.com"));
+    }
+}

src/tls_front/fetcher.rs

@@ -0,0 +1,756 @@
+use std::sync::Arc;
+use std::time::Duration;
+
+use anyhow::{Result, anyhow};
+use tokio::io::{AsyncRead, AsyncReadExt, AsyncWrite, AsyncWriteExt};
+use tokio::net::TcpStream;
+#[cfg(unix)]
+use tokio::net::UnixStream;
+use tokio::time::timeout;
+use tokio_rustls::client::TlsStream;
+use tokio_rustls::TlsConnector;
+use tracing::{debug, warn};
+
+use rustls::client::danger::{HandshakeSignatureValid, ServerCertVerified, ServerCertVerifier};
+use rustls::client::ClientConfig;
+use rustls::pki_types::{CertificateDer, ServerName, UnixTime};
+use rustls::{DigitallySignedStruct, Error as RustlsError};
+
+use x509_parser::prelude::FromDer;
+use x509_parser::certificate::X509Certificate;
+
+use crate::crypto::SecureRandom;
+use crate::network::dns_overrides::resolve_socket_addr;
+use crate::protocol::constants::{TLS_RECORD_APPLICATION, TLS_RECORD_HANDSHAKE};
+use crate::transport::proxy_protocol::{ProxyProtocolV1Builder, ProxyProtocolV2Builder};
+use crate::tls_front::types::{
+    ParsedCertificateInfo,
+    ParsedServerHello,
+    TlsCertPayload,
+    TlsExtension,
+    TlsFetchResult,
+};
+
+/// No-op verifier: accept any certificate (we only need lengths and metadata).
+#[derive(Debug)]
+struct NoVerify;
+
+impl ServerCertVerifier for NoVerify {
+    fn verify_server_cert(
+        &self,
+        _end_entity: &CertificateDer<'_>,
+        _intermediates: &[CertificateDer<'_>],
+        _server_name: &ServerName<'_>,
+        _ocsp: &[u8],
+        _now: UnixTime,
+    ) -> Result<ServerCertVerified, RustlsError> {
+        Ok(ServerCertVerified::assertion())
+    }
+
+    fn verify_tls12_signature(
+        &self,
+        _message: &[u8],
+        _cert: &CertificateDer<'_>,
+        _dss: &DigitallySignedStruct,
+    ) -> Result<HandshakeSignatureValid, RustlsError> {
+        Ok(HandshakeSignatureValid::assertion())
+    }
+
+    fn verify_tls13_signature(
+        &self,
+        _message: &[u8],
+        _cert: &CertificateDer<'_>,
+        _dss: &DigitallySignedStruct,
+    ) -> Result<HandshakeSignatureValid, RustlsError> {
+        Ok(HandshakeSignatureValid::assertion())
+    }
+
+    fn supported_verify_schemes(&self) -> Vec<rustls::SignatureScheme> {
+        use rustls::SignatureScheme::*;
+        vec![
+            RSA_PKCS1_SHA256,
+            RSA_PSS_SHA256,
+            ECDSA_NISTP256_SHA256,
+            ECDSA_NISTP384_SHA384,
+        ]
+    }
+}
+
+fn build_client_config() -> Arc<ClientConfig> {
+    let root = rustls::RootCertStore::empty();
+
+    let provider = rustls::crypto::ring::default_provider();
+    let mut config = ClientConfig::builder_with_provider(Arc::new(provider))
+        .with_protocol_versions(&[&rustls::version::TLS13, &rustls::version::TLS12])
+        .expect("protocol versions")
+        .with_root_certificates(root)
+        .with_no_client_auth();
+
+    config
+        .dangerous()
+        .set_certificate_verifier(Arc::new(NoVerify));
+
+    Arc::new(config)
+}
+
+fn build_client_hello(sni: &str, rng: &SecureRandom) -> Vec<u8> {
+    // === ClientHello body ===
+    let mut body = Vec::new();
+
+    // Legacy version (TLS 1.0) as in real ClientHello headers
+    body.extend_from_slice(&[0x03, 0x03]);
+
+    // Random
+    body.extend_from_slice(&rng.bytes(32));
+
+    // Session ID: empty
+    body.push(0);
+
+    // Cipher suites (common minimal set, TLS1.3 + a few 1.2 fallbacks)
+    let cipher_suites: [u8; 10] = [
+        0x13, 0x01, // TLS_AES_128_GCM_SHA256
+        0x13, 0x02, // TLS_AES_256_GCM_SHA384
+        0x13, 0x03, // TLS_CHACHA20_POLY1305_SHA256
+        0x00, 0x2f, // TLS_RSA_WITH_AES_128_CBC_SHA (legacy)
+        0x00, 0xff, // RENEGOTIATION_INFO_SCSV
+    ];
+    body.extend_from_slice(&(cipher_suites.len() as u16).to_be_bytes());
+    body.extend_from_slice(&cipher_suites);
+
+    // Compression methods: null only
+    body.push(1);
+    body.push(0);
+
+    // === Extensions ===
+    let mut exts = Vec::new();
+
+    // server_name (SNI)
+    let sni_bytes = sni.as_bytes();
+    let mut sni_ext = Vec::with_capacity(5 + sni_bytes.len());
+    sni_ext.extend_from_slice(&(sni_bytes.len() as u16 + 3).to_be_bytes());
+    sni_ext.push(0); // host_name
+    sni_ext.extend_from_slice(&(sni_bytes.len() as u16).to_be_bytes());
+    sni_ext.extend_from_slice(sni_bytes);
+    exts.extend_from_slice(&0x0000u16.to_be_bytes());
+    exts.extend_from_slice(&(sni_ext.len() as u16).to_be_bytes());
+    exts.extend_from_slice(&sni_ext);
+
+    // supported_groups
+    let groups: [u16; 2] = [0x001d, 0x0017]; // x25519, secp256r1
+    exts.extend_from_slice(&0x000au16.to_be_bytes());
+    exts.extend_from_slice(&((2 + groups.len() * 2) as u16).to_be_bytes());
+    exts.extend_from_slice(&(groups.len() as u16 * 2).to_be_bytes());
+    for g in groups { exts.extend_from_slice(&g.to_be_bytes()); }
+
+    // signature_algorithms
+    let sig_algs: [u16; 4] = [0x0804, 0x0805, 0x0403, 0x0503]; // rsa_pss_rsae_sha256/384, ecdsa_secp256r1_sha256, rsa_pkcs1_sha256
+    exts.extend_from_slice(&0x000du16.to_be_bytes());
+    exts.extend_from_slice(&((2 + sig_algs.len() * 2) as u16).to_be_bytes());
+    exts.extend_from_slice(&(sig_algs.len() as u16 * 2).to_be_bytes());
+    for a in sig_algs { exts.extend_from_slice(&a.to_be_bytes()); }
+
+    // supported_versions (TLS1.3 + TLS1.2)
+    let versions: [u16; 2] = [0x0304, 0x0303];
+    exts.extend_from_slice(&0x002bu16.to_be_bytes());
+    exts.extend_from_slice(&((1 + versions.len() * 2) as u16).to_be_bytes());
+    exts.push((versions.len() * 2) as u8);
+    for v in versions { exts.extend_from_slice(&v.to_be_bytes()); }
+
+    // key_share (x25519)
+    let key = gen_key_share(rng);
+    let mut keyshare = Vec::with_capacity(4 + key.len());
+    keyshare.extend_from_slice(&0x001du16.to_be_bytes()); // group
+    keyshare.extend_from_slice(&(key.len() as u16).to_be_bytes());
+    keyshare.extend_from_slice(&key);
+    exts.extend_from_slice(&0x0033u16.to_be_bytes());
+    exts.extend_from_slice(&((2 + keyshare.len()) as u16).to_be_bytes());
+    exts.extend_from_slice(&(keyshare.len() as u16).to_be_bytes());
+    exts.extend_from_slice(&keyshare);
+
+    // ALPN (http/1.1)
+    let alpn_proto = b"http/1.1";
+    exts.extend_from_slice(&0x0010u16.to_be_bytes());
+    exts.extend_from_slice(&((2 + 1 + alpn_proto.len()) as u16).to_be_bytes());
+    exts.extend_from_slice(&((1 + alpn_proto.len()) as u16).to_be_bytes());
+    exts.push(alpn_proto.len() as u8);
+    exts.extend_from_slice(alpn_proto);
+
+    // padding to reduce recognizability and keep length ~500 bytes
+    const TARGET_EXT_LEN: usize = 180;
+    if exts.len() < TARGET_EXT_LEN {
+        let remaining = TARGET_EXT_LEN - exts.len();
+        if remaining > 4 {
+            let pad_len = remaining - 4; // minus type+len
+            exts.extend_from_slice(&0x0015u16.to_be_bytes()); // padding extension
+            exts.extend_from_slice(&(pad_len as u16).to_be_bytes());
+            exts.resize(exts.len() + pad_len, 0);
+        }
+    }
+
+    // Extensions length prefix
+    body.extend_from_slice(&(exts.len() as u16).to_be_bytes());
+    body.extend_from_slice(&exts);
+
+    // === Handshake wrapper ===
+    let mut handshake = Vec::new();
+    handshake.push(0x01); // ClientHello
+    let len_bytes = (body.len() as u32).to_be_bytes();
+    handshake.extend_from_slice(&len_bytes[1..4]);
+    handshake.extend_from_slice(&body);
+
+    // === Record ===
+    let mut record = Vec::new();
+    record.push(TLS_RECORD_HANDSHAKE);
+    record.extend_from_slice(&[0x03, 0x01]); // legacy record version
+    record.extend_from_slice(&(handshake.len() as u16).to_be_bytes());
+    record.extend_from_slice(&handshake);
+
+    record
+}
+
+fn gen_key_share(rng: &SecureRandom) -> [u8; 32] {
+    let mut key = [0u8; 32];
+    key.copy_from_slice(&rng.bytes(32));
+    key
+}
+
+async fn read_tls_record<S>(stream: &mut S) -> Result<(u8, Vec<u8>)>
+where
+    S: AsyncRead + Unpin,
+{
+    let mut header = [0u8; 5];
+    stream.read_exact(&mut header).await?;
+    let len = u16::from_be_bytes([header[3], header[4]]) as usize;
+    let mut body = vec![0u8; len];
+    stream.read_exact(&mut body).await?;
+    Ok((header[0], body))
+}
+
+fn parse_server_hello(body: &[u8]) -> Option<ParsedServerHello> {
+    if body.len() < 4 || body[0] != 0x02 {
+        return None;
+    }
+
+    let msg_len = u32::from_be_bytes([0, body[1], body[2], body[3]]) as usize;
+    if msg_len + 4 > body.len() {
+        return None;
+    }
+
+    let mut pos = 4;
+    let version = [*body.get(pos)?, *body.get(pos + 1)?];
+    pos += 2;
+
+    let mut random = [0u8; 32];
+    random.copy_from_slice(body.get(pos..pos + 32)?);
+    pos += 32;
+
+    let session_len = *body.get(pos)? as usize;
+    pos += 1;
+    let session_id = body.get(pos..pos + session_len)?.to_vec();
+    pos += session_len;
+
+    let cipher_suite = [*body.get(pos)?, *body.get(pos + 1)?];
+    pos += 2;
+
+    let compression = *body.get(pos)?;
+    pos += 1;
+
+    let ext_len = u16::from_be_bytes([*body.get(pos)?, *body.get(pos + 1)?]) as usize;
+    pos += 2;
+    let ext_end = pos.checked_add(ext_len)?;
+    if ext_end > body.len() {
+        return None;
+    }
+
+    let mut extensions = Vec::new();
+    while pos + 4 <= ext_end {
+        let etype = u16::from_be_bytes([body[pos], body[pos + 1]]);
+        let elen = u16::from_be_bytes([body[pos + 2], body[pos + 3]]) as usize;
+        pos += 4;
+        let data = body.get(pos..pos + elen)?.to_vec();
+        pos += elen;
+        extensions.push(TlsExtension { ext_type: etype, data });
+    }
+
+    Some(ParsedServerHello {
+        version,
+        random,
+        session_id,
+        cipher_suite,
+        compression,
+        extensions,
+    })
+}
+
+fn parse_cert_info(certs: &[CertificateDer<'static>]) -> Option<ParsedCertificateInfo> {
+    let first = certs.first()?;
+    let (_rem, cert) = X509Certificate::from_der(first.as_ref()).ok()?;
+
+    let not_before = Some(cert.validity().not_before.to_datetime().unix_timestamp());
+    let not_after = Some(cert.validity().not_after.to_datetime().unix_timestamp());
+
+    let issuer_cn = cert
+        .issuer()
+        .iter_common_name()
+        .next()
+        .and_then(|cn| cn.as_str().ok())
+        .map(|s| s.to_string());
+
+    let subject_cn = cert
+        .subject()
+        .iter_common_name()
+        .next()
+        .and_then(|cn| cn.as_str().ok())
+        .map(|s| s.to_string());
+
+    let san_names = cert
+        .subject_alternative_name()
+        .ok()
+        .flatten()
+        .map(|san| {
+            san.value
+                .general_names
+                .iter()
+                .filter_map(|gn| match gn {
+                    x509_parser::extensions::GeneralName::DNSName(n) => Some(n.to_string()),
+                    _ => None,
+                })
+                .collect::<Vec<_>>()
+        })
+        .unwrap_or_default();
+
+    Some(ParsedCertificateInfo {
+        not_after_unix: not_after,
+        not_before_unix: not_before,
+        issuer_cn,
+        subject_cn,
+        san_names,
+    })
+}
+
+fn u24_bytes(value: usize) -> Option<[u8; 3]> {
+    if value > 0x00ff_ffff {
+        return None;
+    }
+    Some([
+        ((value >> 16) & 0xff) as u8,
+        ((value >> 8) & 0xff) as u8,
+        (value & 0xff) as u8,
+    ])
+}
+
+async fn connect_with_dns_override(
+    host: &str,
+    port: u16,
+    connect_timeout: Duration,
+) -> Result<TcpStream> {
+    if let Some(addr) = resolve_socket_addr(host, port) {
+        return Ok(timeout(connect_timeout, TcpStream::connect(addr)).await??);
+    }
+    Ok(timeout(connect_timeout, TcpStream::connect((host, port))).await??)
+}
+
+async fn connect_tcp_with_upstream(
+    host: &str,
+    port: u16,
+    connect_timeout: Duration,
+    upstream: Option<std::sync::Arc<crate::transport::UpstreamManager>>,
+) -> Result<TcpStream> {
+    if let Some(manager) = upstream {
+        if let Some(addr) = resolve_socket_addr(host, port) {
+            match manager.connect(addr, None, None).await {
+                Ok(stream) => return Ok(stream),
+                Err(e) => {
+                    warn!(
+                        host = %host,
+                        port = port,
+                        error = %e,
+                        "Upstream connect failed, using direct connect"
+                    );
+                }
+            }
+        } else if let Ok(mut addrs) = tokio::net::lookup_host((host, port)).await {
+            if let Some(addr) = addrs.find(|a| a.is_ipv4()) {
+                match manager.connect(addr, None, None).await {
+                    Ok(stream) => return Ok(stream),
+                    Err(e) => {
+                        warn!(
+                            host = %host,
+                            port = port,
+                            error = %e,
+                            "Upstream connect failed, using direct connect"
+                        );
+                    }
+                }
+            }
+        }
+    }
+    connect_with_dns_override(host, port, connect_timeout).await
+}
+
+fn encode_tls13_certificate_message(cert_chain_der: &[Vec<u8>]) -> Option<Vec<u8>> {
+    if cert_chain_der.is_empty() {
+        return None;
+    }
+
+    let mut certificate_list = Vec::new();
+    for cert in cert_chain_der {
+        if cert.is_empty() {
+            return None;
+        }
+        certificate_list.extend_from_slice(&u24_bytes(cert.len())?);
+        certificate_list.extend_from_slice(cert);
+        certificate_list.extend_from_slice(&0u16.to_be_bytes()); // cert_entry extensions
+    }
+
+    // Certificate = context_len(1) + certificate_list_len(3) + entries
+    let body_len = 1usize
+        .checked_add(3)?
+        .checked_add(certificate_list.len())?;
+
+    let mut message = Vec::with_capacity(4 + body_len);
+    message.push(0x0b); // HandshakeType::certificate
+    message.extend_from_slice(&u24_bytes(body_len)?);
+    message.push(0x00); // certificate_request_context length
+    message.extend_from_slice(&u24_bytes(certificate_list.len())?);
+    message.extend_from_slice(&certificate_list);
+    Some(message)
+}
+
+async fn fetch_via_raw_tls_stream<S>(
+    mut stream: S,
+    sni: &str,
+    connect_timeout: Duration,
+    proxy_protocol: u8,
+) -> Result<TlsFetchResult>
+where
+    S: AsyncRead + AsyncWrite + Unpin,
+{
+    let rng = SecureRandom::new();
+    let client_hello = build_client_hello(sni, &rng);
+    timeout(connect_timeout, async {
+        if proxy_protocol > 0 {
+            let header = match proxy_protocol {
+                2 => ProxyProtocolV2Builder::new().build(),
+                _ => ProxyProtocolV1Builder::new().build(),
+            };
+            stream.write_all(&header).await?;
+        }
+        stream.write_all(&client_hello).await?;
+        stream.flush().await?;
+        Ok::<(), std::io::Error>(())
+    })
+    .await??;
+
+    let mut records = Vec::new();
+    // Read up to 4 records: ServerHello, CCS, and up to two ApplicationData.
+    for _ in 0..4 {
+        match timeout(connect_timeout, read_tls_record(&mut stream)).await {
+            Ok(Ok(rec)) => records.push(rec),
+            Ok(Err(e)) => return Err(e),
+            Err(_) => break,
+        }
+        if records.len() >= 3 && records.iter().any(|(t, _)| *t == TLS_RECORD_APPLICATION) {
+            break;
+        }
+    }
+
+    let mut app_sizes = Vec::new();
+    let mut server_hello = None;
+    for (t, body) in &records {
+        if *t == TLS_RECORD_HANDSHAKE && server_hello.is_none() {
+            server_hello = parse_server_hello(body);
+        } else if *t == TLS_RECORD_APPLICATION {
+            app_sizes.push(body.len());
+        }
+    }
+
+    let parsed = server_hello.ok_or_else(|| anyhow!("ServerHello not received"))?;
+    let total_app_data_len = app_sizes.iter().sum::<usize>().max(1024);
+
+    Ok(TlsFetchResult {
+        server_hello_parsed: parsed,
+        app_data_records_sizes: if app_sizes.is_empty() {
+            vec![total_app_data_len]
+        } else {
+            app_sizes
+        },
+        total_app_data_len,
+        cert_info: None,
+        cert_payload: None,
+    })
+}
+
+async fn fetch_via_raw_tls(
+    host: &str,
+    port: u16,
+    sni: &str,
+    connect_timeout: Duration,
+    upstream: Option<std::sync::Arc<crate::transport::UpstreamManager>>,
+    proxy_protocol: u8,
+    unix_sock: Option<&str>,
+) -> Result<TlsFetchResult> {
+    #[cfg(unix)]
+    if let Some(sock_path) = unix_sock {
+        match timeout(connect_timeout, UnixStream::connect(sock_path)).await {
+            Ok(Ok(stream)) => {
+                debug!(
+                    sni = %sni,
+                    sock = %sock_path,
+                    "Raw TLS fetch using mask unix socket"
+                );
+                return fetch_via_raw_tls_stream(stream, sni, connect_timeout, proxy_protocol).await;
+            }
+            Ok(Err(e)) => {
+                warn!(
+                    sni = %sni,
+                    sock = %sock_path,
+                    error = %e,
+                    "Raw TLS unix socket connect failed, falling back to TCP"
+                );
+            }
+            Err(_) => {
+                warn!(
+                    sni = %sni,
+                    sock = %sock_path,
+                    "Raw TLS unix socket connect timed out, falling back to TCP"
+                );
+            }
+        }
+    }
+
+    #[cfg(not(unix))]
+    let _ = unix_sock;
+
+    let stream = connect_tcp_with_upstream(host, port, connect_timeout, upstream).await?;
+    fetch_via_raw_tls_stream(stream, sni, connect_timeout, proxy_protocol).await
+}
+
+async fn fetch_via_rustls_stream<S>(
+    mut stream: S,
+    host: &str,
+    sni: &str,
+    proxy_protocol: u8,
+) -> Result<TlsFetchResult>
+where
+    S: AsyncRead + AsyncWrite + Unpin,
+{
+    // rustls handshake path for certificate and basic negotiated metadata.
+    if proxy_protocol > 0 {
+        let header = match proxy_protocol {
+            2 => ProxyProtocolV2Builder::new().build(),
+            _ => ProxyProtocolV1Builder::new().build(),
+        };
+        stream.write_all(&header).await?;
+        stream.flush().await?;
+    }
+
+    let config = build_client_config();
+    let connector = TlsConnector::from(config);
+
+    let server_name = ServerName::try_from(sni.to_owned())
+        .or_else(|_| ServerName::try_from(host.to_owned()))
+        .map_err(|_| RustlsError::General("invalid SNI".into()))?;
+
+    let tls_stream: TlsStream<S> = connector.connect(server_name, stream).await?;
+
+    // Extract negotiated parameters and certificates
+    let (_io, session) = tls_stream.get_ref();
+    let cipher_suite = session
+        .negotiated_cipher_suite()
+        .map(|s| u16::from(s.suite()).to_be_bytes())
+        .unwrap_or([0x13, 0x01]);
+
+    let certs: Vec<CertificateDer<'static>> = session
+        .peer_certificates()
+        .map(|slice| slice.to_vec())
+        .unwrap_or_default();
+    let cert_chain_der: Vec<Vec<u8>> = certs.iter().map(|c| c.as_ref().to_vec()).collect();
+    let cert_payload = encode_tls13_certificate_message(&cert_chain_der).map(|certificate_message| {
+        TlsCertPayload {
+            cert_chain_der: cert_chain_der.clone(),
+            certificate_message,
+        }
+    });
+
+    let total_cert_len = cert_payload
+        .as_ref()
+        .map(|payload| payload.certificate_message.len())
+        .unwrap_or_else(|| cert_chain_der.iter().map(Vec::len).sum::<usize>())
+        .max(1024);
+    let cert_info = parse_cert_info(&certs);
+
+    // Heuristic: split across two records if large to mimic real servers a bit.
+    let app_data_records_sizes = if total_cert_len > 3000 {
+        vec![total_cert_len / 2, total_cert_len - total_cert_len / 2]
+    } else {
+        vec![total_cert_len]
+    };
+
+    let parsed = ParsedServerHello {
+        version: [0x03, 0x03],
+        random: [0u8; 32],
+        session_id: Vec::new(),
+        cipher_suite,
+        compression: 0,
+        extensions: Vec::new(),
+    };
+
+    debug!(
+        sni = %sni,
+        len = total_cert_len,
+        cipher = format!("0x{:04x}", u16::from_be_bytes(cipher_suite)),
+        has_cert_payload = cert_payload.is_some(),
+        "Fetched TLS metadata via rustls"
+    );
+
+    Ok(TlsFetchResult {
+        server_hello_parsed: parsed,
+        app_data_records_sizes: app_data_records_sizes.clone(),
+        total_app_data_len: app_data_records_sizes.iter().sum(),
+        cert_info,
+        cert_payload,
+    })
+}
+
+async fn fetch_via_rustls(
+    host: &str,
+    port: u16,
+    sni: &str,
+    connect_timeout: Duration,
+    upstream: Option<std::sync::Arc<crate::transport::UpstreamManager>>,
+    proxy_protocol: u8,
+    unix_sock: Option<&str>,
+) -> Result<TlsFetchResult> {
+    #[cfg(unix)]
+    if let Some(sock_path) = unix_sock {
+        match timeout(connect_timeout, UnixStream::connect(sock_path)).await {
+            Ok(Ok(stream)) => {
+                debug!(
+                    sni = %sni,
+                    sock = %sock_path,
+                    "Rustls fetch using mask unix socket"
+                );
+                return fetch_via_rustls_stream(stream, host, sni, proxy_protocol).await;
+            }
+            Ok(Err(e)) => {
+                warn!(
+                    sni = %sni,
+                    sock = %sock_path,
+                    error = %e,
+                    "Rustls unix socket connect failed, falling back to TCP"
+                );
+            }
+            Err(_) => {
+                warn!(
+                    sni = %sni,
+                    sock = %sock_path,
+                    "Rustls unix socket connect timed out, falling back to TCP"
+                );
+            }
+        }
+    }
+
+    #[cfg(not(unix))]
+    let _ = unix_sock;
+
+    let stream = connect_tcp_with_upstream(host, port, connect_timeout, upstream).await?;
+    fetch_via_rustls_stream(stream, host, sni, proxy_protocol).await
+}
+
+/// Fetch real TLS metadata for the given SNI.
+///
+/// Strategy:
+/// 1) Probe raw TLS for realistic ServerHello and ApplicationData record sizes.
+/// 2) Fetch certificate chain via rustls to build cert payload.
+/// 3) Merge both when possible; otherwise auto-fallback to whichever succeeded.
+pub async fn fetch_real_tls(
+    host: &str,
+    port: u16,
+    sni: &str,
+    connect_timeout: Duration,
+    upstream: Option<std::sync::Arc<crate::transport::UpstreamManager>>,
+    proxy_protocol: u8,
+    unix_sock: Option<&str>,
+) -> Result<TlsFetchResult> {
+    let raw_result = match fetch_via_raw_tls(
+        host,
+        port,
+        sni,
+        connect_timeout,
+        upstream.clone(),
+        proxy_protocol,
+        unix_sock,
+    )
+    .await
+    {
+        Ok(res) => Some(res),
+        Err(e) => {
+            warn!(sni = %sni, error = %e, "Raw TLS fetch failed");
+            None
+        }
+    };
+
+    match fetch_via_rustls(
+        host,
+        port,
+        sni,
+        connect_timeout,
+        upstream,
+        proxy_protocol,
+        unix_sock,
+    )
+    .await
+    {
+        Ok(rustls_result) => {
+            if let Some(mut raw) = raw_result {
+                raw.cert_info = rustls_result.cert_info;
+                raw.cert_payload = rustls_result.cert_payload;
+                debug!(sni = %sni, "Fetched TLS metadata via raw probe + rustls cert chain");
+                Ok(raw)
+            } else {
+                Ok(rustls_result)
+            }
+        }
+        Err(e) => {
+            if let Some(raw) = raw_result {
+                warn!(sni = %sni, error = %e, "Rustls cert fetch failed, using raw TLS metadata only");
+                Ok(raw)
+            } else {
+                Err(e)
+            }
+        }
+    }
+}
+
+#[cfg(test)]
+mod tests {
+    use super::encode_tls13_certificate_message;
+
+    fn read_u24(bytes: &[u8]) -> usize {
+        ((bytes[0] as usize) << 16) | ((bytes[1] as usize) << 8) | (bytes[2] as usize)
+    }
+
+    #[test]
+    fn test_encode_tls13_certificate_message_single_cert() {
+        let cert = vec![0x30, 0x03, 0x02, 0x01, 0x01];
+        let message = encode_tls13_certificate_message(&[cert.clone()]).expect("message");
+
+        assert_eq!(message[0], 0x0b);
+        assert_eq!(read_u24(&message[1..4]), message.len() - 4);
+        assert_eq!(message[4], 0x00);
+
+        let cert_list_len = read_u24(&message[5..8]);
+        assert_eq!(cert_list_len, cert.len() + 5);
+
+        let cert_len = read_u24(&message[8..11]);
+        assert_eq!(cert_len, cert.len());
+        assert_eq!(&message[11..11 + cert.len()], cert.as_slice());
+        assert_eq!(&message[11 + cert.len()..13 + cert.len()], &[0x00, 0x00]);
+    }
+
+    #[test]
+    fn test_encode_tls13_certificate_message_empty_chain() {
+        assert!(encode_tls13_certificate_message(&[]).is_none());
+    }
+}

src/tls_front/mod.rs

@@ -0,0 +1,8 @@
+pub mod types;
+pub mod cache;
+pub mod fetcher;
+pub mod emulator;
+
+pub use cache::TlsFrontCache;
+#[allow(unused_imports)]
+pub use types::{CachedTlsData, TlsFetchResult};

src/tls_front/types.rs

@@ -0,0 +1,68 @@
+use std::time::SystemTime;
+use serde::{Serialize, Deserialize};
+
+/// Parsed representation of an unencrypted TLS ServerHello.
+#[derive(Debug, Clone, Serialize, Deserialize)]
+pub struct ParsedServerHello {
+    pub version: [u8; 2],
+    pub random: [u8; 32],
+    pub session_id: Vec<u8>,
+    pub cipher_suite: [u8; 2],
+    pub compression: u8,
+    pub extensions: Vec<TlsExtension>,
+}
+
+/// Generic TLS extension container.
+#[derive(Debug, Clone, Serialize, Deserialize)]
+pub struct TlsExtension {
+    pub ext_type: u16,
+    pub data: Vec<u8>,
+}
+
+/// Basic certificate metadata (optional, informative).
+#[derive(Debug, Clone, Serialize, Deserialize)]
+pub struct ParsedCertificateInfo {
+    pub not_after_unix: Option<i64>,
+    pub not_before_unix: Option<i64>,
+    pub issuer_cn: Option<String>,
+    pub subject_cn: Option<String>,
+    pub san_names: Vec<String>,
+}
+
+/// TLS certificate payload captured from profiled upstream.
+///
+/// `certificate_message` stores an encoded TLS 1.3 Certificate handshake
+/// message body that can be replayed as opaque ApplicationData bytes in FakeTLS.
+#[derive(Debug, Clone, Serialize, Deserialize)]
+pub struct TlsCertPayload {
+    pub cert_chain_der: Vec<Vec<u8>>,
+    pub certificate_message: Vec<u8>,
+}
+
+/// Cached data per SNI used by the emulator.
+#[derive(Debug, Clone, Serialize, Deserialize)]
+pub struct CachedTlsData {
+    pub server_hello_template: ParsedServerHello,
+    pub cert_info: Option<ParsedCertificateInfo>,
+    #[serde(default)]
+    pub cert_payload: Option<TlsCertPayload>,
+    pub app_data_records_sizes: Vec<usize>,
+    pub total_app_data_len: usize,
+    #[serde(default = "now_system_time", skip_serializing, skip_deserializing)]
+    pub fetched_at: SystemTime,
+    pub domain: String,
+}
+
+fn now_system_time() -> SystemTime {
+    SystemTime::now()
+}
+
+/// Result of attempting to fetch real TLS artifacts.
+#[derive(Debug, Clone, Serialize, Deserialize)]
+pub struct TlsFetchResult {
+    pub server_hello_parsed: ParsedServerHello,
+    pub app_data_records_sizes: Vec<usize>,
+    pub total_app_data_len: usize,
+    pub cert_info: Option<ParsedCertificateInfo>,
+    pub cert_payload: Option<TlsCertPayload>,
+}

src/transport/middle_proxy/codec.rs

@@ -0,0 +1,249 @@
+use tokio::io::{AsyncReadExt, AsyncWriteExt};
+
+use crate::crypto::{AesCbc, crc32, crc32c};
+use crate::error::{ProxyError, Result};
+use crate::protocol::constants::*;
+
+/// Commands sent to dedicated writer tasks to avoid mutex contention on TCP writes.
+pub(crate) enum WriterCommand {
+    Data(Vec<u8>),
+    DataAndFlush(Vec<u8>),
+    Close,
+}
+
+#[derive(Debug, Clone, Copy, PartialEq, Eq)]
+pub(crate) enum RpcChecksumMode {
+    Crc32,
+    Crc32c,
+}
+
+impl RpcChecksumMode {
+    pub(crate) fn from_handshake_flags(flags: u32) -> Self {
+        if (flags & rpc_crypto_flags::USE_CRC32C) != 0 {
+            Self::Crc32c
+        } else {
+            Self::Crc32
+        }
+    }
+
+    pub(crate) fn advertised_flags(self) -> u32 {
+        match self {
+            Self::Crc32 => 0,
+            Self::Crc32c => rpc_crypto_flags::USE_CRC32C,
+        }
+    }
+}
+
+pub(crate) fn rpc_crc(mode: RpcChecksumMode, data: &[u8]) -> u32 {
+    match mode {
+        RpcChecksumMode::Crc32 => crc32(data),
+        RpcChecksumMode::Crc32c => crc32c(data),
+    }
+}
+
+pub(crate) fn build_rpc_frame(seq_no: i32, payload: &[u8], crc_mode: RpcChecksumMode) -> Vec<u8> {
+    let total_len = (4 + 4 + payload.len() + 4) as u32;
+    let mut frame = Vec::with_capacity(total_len as usize);
+    frame.extend_from_slice(&total_len.to_le_bytes());
+    frame.extend_from_slice(&seq_no.to_le_bytes());
+    frame.extend_from_slice(payload);
+    let c = rpc_crc(crc_mode, &frame);
+    frame.extend_from_slice(&c.to_le_bytes());
+    frame
+}
+
+pub(crate) async fn read_rpc_frame_plaintext(
+    rd: &mut (impl AsyncReadExt + Unpin),
+) -> Result<(i32, Vec<u8>)> {
+    let mut len_buf = [0u8; 4];
+    rd.read_exact(&mut len_buf).await.map_err(ProxyError::Io)?;
+    let total_len = u32::from_le_bytes(len_buf) as usize;
+
+    if !(12..=(1 << 24)).contains(&total_len) {
+        return Err(ProxyError::InvalidHandshake(format!(
+            "Bad RPC frame length: {total_len}"
+        )));
+    }
+
+    let mut rest = vec![0u8; total_len - 4];
+    rd.read_exact(&mut rest).await.map_err(ProxyError::Io)?;
+
+    let mut full = Vec::with_capacity(total_len);
+    full.extend_from_slice(&len_buf);
+    full.extend_from_slice(&rest);
+
+    let crc_offset = total_len - 4;
+    let expected_crc = u32::from_le_bytes(full[crc_offset..crc_offset + 4].try_into().unwrap());
+    let actual_crc = rpc_crc(RpcChecksumMode::Crc32, &full[..crc_offset]);
+    if expected_crc != actual_crc {
+        return Err(ProxyError::InvalidHandshake(format!(
+            "CRC mismatch: 0x{expected_crc:08x} vs 0x{actual_crc:08x}"
+        )));
+    }
+
+    let seq_no = i32::from_le_bytes(full[4..8].try_into().unwrap());
+    let payload = full[8..crc_offset].to_vec();
+    Ok((seq_no, payload))
+}
+
+pub(crate) fn build_nonce_payload(key_selector: u32, crypto_ts: u32, nonce: &[u8; 16]) -> [u8; 32] {
+    let mut p = [0u8; 32];
+    p[0..4].copy_from_slice(&RPC_NONCE_U32.to_le_bytes());
+    p[4..8].copy_from_slice(&key_selector.to_le_bytes());
+    p[8..12].copy_from_slice(&RPC_CRYPTO_AES_U32.to_le_bytes());
+    p[12..16].copy_from_slice(&crypto_ts.to_le_bytes());
+    p[16..32].copy_from_slice(nonce);
+    p
+}
+
+pub(crate) fn parse_nonce_payload(d: &[u8]) -> Result<(u32, u32, u32, [u8; 16])> {
+    if d.len() < 32 {
+        return Err(ProxyError::InvalidHandshake(format!(
+            "Nonce payload too short: {} bytes",
+            d.len()
+        )));
+    }
+
+    let t = u32::from_le_bytes(d[0..4].try_into().unwrap());
+    if t != RPC_NONCE_U32 {
+        return Err(ProxyError::InvalidHandshake(format!(
+            "Expected RPC_NONCE 0x{RPC_NONCE_U32:08x}, got 0x{t:08x}"
+        )));
+    }
+
+    let key_select = u32::from_le_bytes(d[4..8].try_into().unwrap());
+    let schema = u32::from_le_bytes(d[8..12].try_into().unwrap());
+    let ts = u32::from_le_bytes(d[12..16].try_into().unwrap());
+    let mut nonce = [0u8; 16];
+    nonce.copy_from_slice(&d[16..32]);
+    Ok((key_select, schema, ts, nonce))
+}
+
+pub(crate) fn build_handshake_payload(
+    our_ip: [u8; 4],
+    our_port: u16,
+    peer_ip: [u8; 4],
+    peer_port: u16,
+    flags: u32,
+) -> [u8; 32] {
+    let mut p = [0u8; 32];
+    p[0..4].copy_from_slice(&RPC_HANDSHAKE_U32.to_le_bytes());
+    p[4..8].copy_from_slice(&flags.to_le_bytes());
+
+    // process_id sender_pid
+    p[8..12].copy_from_slice(&our_ip);
+    p[12..14].copy_from_slice(&our_port.to_le_bytes());
+    p[14..16].copy_from_slice(&process_pid16().to_le_bytes());
+    p[16..20].copy_from_slice(&process_utime().to_le_bytes());
+
+    // process_id peer_pid
+    p[20..24].copy_from_slice(&peer_ip);
+    p[24..26].copy_from_slice(&peer_port.to_le_bytes());
+    p[26..28].copy_from_slice(&0u16.to_le_bytes());
+    p[28..32].copy_from_slice(&0u32.to_le_bytes());
+    p
+}
+
+pub(crate) fn parse_handshake_flags(payload: &[u8]) -> Result<u32> {
+    if payload.len() != 32 {
+        return Err(ProxyError::InvalidHandshake(format!(
+            "Bad handshake payload len: {}",
+            payload.len()
+        )));
+    }
+    let hs_type = u32::from_le_bytes(payload[0..4].try_into().unwrap());
+    if hs_type != RPC_HANDSHAKE_U32 {
+        return Err(ProxyError::InvalidHandshake(format!(
+            "Expected HANDSHAKE 0x{RPC_HANDSHAKE_U32:08x}, got 0x{hs_type:08x}"
+        )));
+    }
+    Ok(u32::from_le_bytes(payload[4..8].try_into().unwrap()))
+}
+
+fn process_pid16() -> u16 {
+    (std::process::id() & 0xffff) as u16
+}
+
+fn process_utime() -> u32 {
+    std::time::SystemTime::now()
+        .duration_since(std::time::UNIX_EPOCH)
+        .unwrap_or_default()
+        .as_secs() as u32
+}
+
+pub(crate) fn cbc_encrypt_padded(
+    key: &[u8; 32],
+    iv: &[u8; 16],
+    plaintext: &[u8],
+) -> Result<(Vec<u8>, [u8; 16])> {
+    let pad = (16 - (plaintext.len() % 16)) % 16;
+    let mut buf = plaintext.to_vec();
+    let pad_pattern: [u8; 4] = [0x04, 0x00, 0x00, 0x00];
+    for i in 0..pad {
+        buf.push(pad_pattern[i % 4]);
+    }
+
+    let cipher = AesCbc::new(*key, *iv);
+    cipher
+        .encrypt_in_place(&mut buf)
+        .map_err(|e| ProxyError::Crypto(format!("CBC encrypt: {e}")))?;
+
+    let mut new_iv = [0u8; 16];
+    if buf.len() >= 16 {
+        new_iv.copy_from_slice(&buf[buf.len() - 16..]);
+    }
+    Ok((buf, new_iv))
+}
+
+pub(crate) fn cbc_decrypt_inplace(
+    key: &[u8; 32],
+    iv: &[u8; 16],
+    data: &mut [u8],
+) -> Result<[u8; 16]> {
+    let mut new_iv = [0u8; 16];
+    if data.len() >= 16 {
+        new_iv.copy_from_slice(&data[data.len() - 16..]);
+    }
+
+    AesCbc::new(*key, *iv)
+        .decrypt_in_place(data)
+        .map_err(|e| ProxyError::Crypto(format!("CBC decrypt: {e}")))?;
+    Ok(new_iv)
+}
+
+pub(crate) struct RpcWriter {
+    pub(crate) writer: tokio::io::WriteHalf<tokio::net::TcpStream>,
+    pub(crate) key: [u8; 32],
+    pub(crate) iv: [u8; 16],
+    pub(crate) seq_no: i32,
+    pub(crate) crc_mode: RpcChecksumMode,
+}
+
+impl RpcWriter {
+    pub(crate) async fn send(&mut self, payload: &[u8]) -> Result<()> {
+        let frame = build_rpc_frame(self.seq_no, payload, self.crc_mode);
+        self.seq_no = self.seq_no.wrapping_add(1);
+
+        let pad = (16 - (frame.len() % 16)) % 16;
+        let mut buf = frame;
+        let pad_pattern: [u8; 4] = [0x04, 0x00, 0x00, 0x00];
+        for i in 0..pad {
+            buf.push(pad_pattern[i % 4]);
+        }
+
+        let cipher = AesCbc::new(self.key, self.iv);
+        cipher
+            .encrypt_in_place(&mut buf)
+            .map_err(|e| ProxyError::Crypto(format!("{e}")))?;
+
+        if buf.len() >= 16 {
+            self.iv.copy_from_slice(&buf[buf.len() - 16..]);
+        }
+        self.writer.write_all(&buf).await.map_err(ProxyError::Io)
+    }
+
+    pub(crate) async fn send_and_flush(&mut self, payload: &[u8]) -> Result<()> {
+        self.send(payload).await?;
+        self.writer.flush().await.map_err(ProxyError::Io)
+    }
+}

src/transport/middle_proxy/config_updater.rs

@@ -0,0 +1,592 @@
+use std::collections::HashMap;
+use std::hash::{DefaultHasher, Hash, Hasher};
+use std::net::IpAddr;
+use std::path::Path;
+use std::sync::Arc;
+use std::time::Duration;
+
+use httpdate;
+use tokio::sync::{mpsc, watch};
+use tracing::{debug, info, warn};
+
+use crate::config::ProxyConfig;
+use crate::error::Result;
+
+use super::MePool;
+use super::rotation::{MeReinitTrigger, enqueue_reinit_trigger};
+use super::secret::download_proxy_secret_with_max_len;
+use std::time::SystemTime;
+
+async fn retry_fetch(url: &str) -> Option<ProxyConfigData> {
+    let delays = [1u64, 5, 15];
+    for (i, d) in delays.iter().enumerate() {
+        match fetch_proxy_config(url).await {
+            Ok(cfg) => return Some(cfg),
+            Err(e) => {
+                if i == delays.len() - 1 {
+                    warn!(error = %e, url, "fetch_proxy_config failed");
+                } else {
+                    debug!(error = %e, url, "fetch_proxy_config retrying");
+                    tokio::time::sleep(Duration::from_secs(*d)).await;
+                }
+            }
+        }
+    }
+    None
+}
+
+#[derive(Debug, Clone, Default)]
+pub struct ProxyConfigData {
+    pub map: HashMap<i32, Vec<(IpAddr, u16)>>,
+    pub default_dc: Option<i32>,
+    pub http_status: u16,
+    pub proxy_for_lines: u32,
+}
+
+pub fn parse_proxy_config_text(text: &str, http_status: u16) -> ProxyConfigData {
+    let mut map: HashMap<i32, Vec<(IpAddr, u16)>> = HashMap::new();
+    let mut proxy_for_lines: u32 = 0;
+    for line in text.lines() {
+        if let Some((dc, ip, port)) = parse_proxy_line(line) {
+            map.entry(dc).or_default().push((ip, port));
+            proxy_for_lines = proxy_for_lines.saturating_add(1);
+        }
+    }
+
+    let default_dc = text.lines().find_map(|l| {
+        let t = l.trim();
+        if let Some(rest) = t.strip_prefix("default") {
+            return rest.trim().trim_end_matches(';').parse::<i32>().ok();
+        }
+        None
+    });
+
+    ProxyConfigData {
+        map,
+        default_dc,
+        http_status,
+        proxy_for_lines,
+    }
+}
+
+pub async fn load_proxy_config_cache(path: &str) -> Result<ProxyConfigData> {
+    let text = tokio::fs::read_to_string(path).await.map_err(|e| {
+        crate::error::ProxyError::Proxy(format!("read proxy-config cache '{path}' failed: {e}"))
+    })?;
+    Ok(parse_proxy_config_text(&text, 200))
+}
+
+pub async fn save_proxy_config_cache(path: &str, raw_text: &str) -> Result<()> {
+    if let Some(parent) = Path::new(path).parent()
+        && !parent.as_os_str().is_empty()
+    {
+        tokio::fs::create_dir_all(parent).await.map_err(|e| {
+            crate::error::ProxyError::Proxy(format!(
+                "create proxy-config cache dir '{}' failed: {e}",
+                parent.display()
+            ))
+        })?;
+    }
+
+    tokio::fs::write(path, raw_text).await.map_err(|e| {
+        crate::error::ProxyError::Proxy(format!("write proxy-config cache '{path}' failed: {e}"))
+    })?;
+    Ok(())
+}
+
+pub async fn fetch_proxy_config_with_raw(url: &str) -> Result<(ProxyConfigData, String)> {
+    let resp = reqwest::get(url)
+        .await
+        .map_err(|e| crate::error::ProxyError::Proxy(format!("fetch_proxy_config GET failed: {e}")))?
+        ;
+    let http_status = resp.status().as_u16();
+
+    if let Some(date) = resp.headers().get(reqwest::header::DATE)
+        && let Ok(date_str) = date.to_str()
+        && let Ok(server_time) = httpdate::parse_http_date(date_str)
+        && let Ok(skew) = SystemTime::now().duration_since(server_time).or_else(|e| {
+            server_time.duration_since(SystemTime::now()).map_err(|_| e)
+        })
+    {
+        let skew_secs = skew.as_secs();
+        if skew_secs > 60 {
+            warn!(skew_secs, "Time skew >60s detected from fetch_proxy_config Date header");
+        } else if skew_secs > 30 {
+            warn!(skew_secs, "Time skew >30s detected from fetch_proxy_config Date header");
+        }
+    }
+
+    let text = resp
+        .text()
+        .await
+        .map_err(|e| crate::error::ProxyError::Proxy(format!("fetch_proxy_config read failed: {e}")))?;
+    let parsed = parse_proxy_config_text(&text, http_status);
+    Ok((parsed, text))
+}
+
+#[derive(Debug, Default)]
+struct StableSnapshot {
+    candidate_hash: Option<u64>,
+    candidate_hits: u8,
+    applied_hash: Option<u64>,
+}
+
+impl StableSnapshot {
+    fn observe(&mut self, hash: u64) -> u8 {
+        if self.candidate_hash == Some(hash) {
+            self.candidate_hits = self.candidate_hits.saturating_add(1);
+        } else {
+            self.candidate_hash = Some(hash);
+            self.candidate_hits = 1;
+        }
+        self.candidate_hits
+    }
+
+    fn is_applied(&self, hash: u64) -> bool {
+        self.applied_hash == Some(hash)
+    }
+
+    fn mark_applied(&mut self, hash: u64) {
+        self.applied_hash = Some(hash);
+    }
+}
+
+#[derive(Debug, Default)]
+struct UpdaterState {
+    config_v4: StableSnapshot,
+    config_v6: StableSnapshot,
+    secret: StableSnapshot,
+    last_map_apply_at: Option<tokio::time::Instant>,
+}
+
+fn hash_proxy_config(cfg: &ProxyConfigData) -> u64 {
+    let mut hasher = DefaultHasher::new();
+    cfg.default_dc.hash(&mut hasher);
+
+    let mut by_dc: Vec<(i32, Vec<(IpAddr, u16)>)> =
+        cfg.map.iter().map(|(dc, addrs)| (*dc, addrs.clone())).collect();
+    by_dc.sort_by_key(|(dc, _)| *dc);
+    for (dc, mut addrs) in by_dc {
+        dc.hash(&mut hasher);
+        addrs.sort_unstable();
+        for (ip, port) in addrs {
+            ip.hash(&mut hasher);
+            port.hash(&mut hasher);
+        }
+    }
+
+    hasher.finish()
+}
+
+fn hash_secret(secret: &[u8]) -> u64 {
+    let mut hasher = DefaultHasher::new();
+    secret.hash(&mut hasher);
+    hasher.finish()
+}
+
+fn map_apply_cooldown_ready(
+    last_applied: Option<tokio::time::Instant>,
+    cooldown: Duration,
+) -> bool {
+    if cooldown.is_zero() {
+        return true;
+    }
+    match last_applied {
+        Some(ts) => ts.elapsed() >= cooldown,
+        None => true,
+    }
+}
+
+fn map_apply_cooldown_remaining_secs(
+    last_applied: tokio::time::Instant,
+    cooldown: Duration,
+) -> u64 {
+    if cooldown.is_zero() {
+        return 0;
+    }
+    cooldown
+        .checked_sub(last_applied.elapsed())
+        .map(|d| d.as_secs())
+        .unwrap_or(0)
+}
+
+fn parse_host_port(s: &str) -> Option<(IpAddr, u16)> {
+    if let Some(bracket_end) = s.rfind(']')
+        && s.starts_with('[')
+        && bracket_end + 1 < s.len()
+        && s.as_bytes().get(bracket_end + 1) == Some(&b':')
+    {
+        let host = &s[1..bracket_end];
+        let port_str = &s[bracket_end + 2..];
+        let ip = host.parse::<IpAddr>().ok()?;
+        let port = port_str.parse::<u16>().ok()?;
+        return Some((ip, port));
+    }
+
+    let idx = s.rfind(':')?;
+    let host = &s[..idx];
+    let port_str = &s[idx + 1..];
+    let ip = host.parse::<IpAddr>().ok()?;
+    let port = port_str.parse::<u16>().ok()?;
+    Some((ip, port))
+}
+
+fn parse_proxy_line(line: &str) -> Option<(i32, IpAddr, u16)> {
+    // Accepts lines like:
+    // proxy_for 4 91.108.4.195:8888;
+    // proxy_for 2 [2001:67c:04e8:f002::d]:80;
+    // proxy_for 2 2001:67c:04e8:f002::d:80;
+    let trimmed = line.trim();
+    if !trimmed.starts_with("proxy_for") {
+        return None;
+    }
+    // Capture everything between dc and trailing ';'
+    let without_prefix = trimmed.trim_start_matches("proxy_for").trim();
+    let mut parts = without_prefix.split_whitespace();
+    let dc_str = parts.next()?;
+    let rest = parts.next()?;
+    let host_port = rest.trim_end_matches(';');
+    let dc = dc_str.parse::<i32>().ok()?;
+    let (ip, port) = parse_host_port(host_port)?;
+    Some((dc, ip, port))
+}
+
+pub async fn fetch_proxy_config(url: &str) -> Result<ProxyConfigData> {
+    fetch_proxy_config_with_raw(url)
+        .await
+        .map(|(parsed, _raw)| parsed)
+}
+
+fn snapshot_passes_guards(
+    cfg: &ProxyConfig,
+    snapshot: &ProxyConfigData,
+    snapshot_name: &'static str,
+) -> bool {
+    if cfg.general.me_snapshot_require_http_2xx
+        && !(200..=299).contains(&snapshot.http_status)
+    {
+        warn!(
+            snapshot = snapshot_name,
+            http_status = snapshot.http_status,
+            "ME snapshot rejected by non-2xx HTTP status"
+        );
+        return false;
+    }
+
+    let min_proxy_for = cfg.general.me_snapshot_min_proxy_for_lines;
+    if snapshot.proxy_for_lines < min_proxy_for {
+        warn!(
+            snapshot = snapshot_name,
+            parsed_proxy_for_lines = snapshot.proxy_for_lines,
+            min_proxy_for_lines = min_proxy_for,
+            "ME snapshot rejected by proxy_for line floor"
+        );
+        return false;
+    }
+
+    true
+}
+
+async fn run_update_cycle(
+    pool: &Arc<MePool>,
+    cfg: &ProxyConfig,
+    state: &mut UpdaterState,
+    reinit_tx: &mpsc::Sender<MeReinitTrigger>,
+) {
+    pool.update_runtime_reinit_policy(
+        cfg.general.hardswap,
+        cfg.general.me_pool_drain_ttl_secs,
+        cfg.general.effective_me_pool_force_close_secs(),
+        cfg.general.me_pool_min_fresh_ratio,
+        cfg.general.me_hardswap_warmup_delay_min_ms,
+        cfg.general.me_hardswap_warmup_delay_max_ms,
+        cfg.general.me_hardswap_warmup_extra_passes,
+        cfg.general.me_hardswap_warmup_pass_backoff_base_ms,
+        cfg.general.me_bind_stale_mode,
+        cfg.general.me_bind_stale_ttl_secs,
+        cfg.general.me_secret_atomic_snapshot,
+        cfg.general.me_deterministic_writer_sort,
+        cfg.general.me_single_endpoint_shadow_writers,
+        cfg.general.me_single_endpoint_outage_mode_enabled,
+        cfg.general.me_single_endpoint_outage_disable_quarantine,
+        cfg.general.me_single_endpoint_outage_backoff_min_ms,
+        cfg.general.me_single_endpoint_outage_backoff_max_ms,
+        cfg.general.me_single_endpoint_shadow_rotate_every_secs,
+        cfg.general.me_floor_mode,
+        cfg.general.me_adaptive_floor_idle_secs,
+        cfg.general.me_adaptive_floor_min_writers_single_endpoint,
+        cfg.general.me_adaptive_floor_recover_grace_secs,
+    );
+
+    let required_cfg_snapshots = cfg.general.me_config_stable_snapshots.max(1);
+    let required_secret_snapshots = cfg.general.proxy_secret_stable_snapshots.max(1);
+    let apply_cooldown = Duration::from_secs(cfg.general.me_config_apply_cooldown_secs);
+    let mut maps_changed = false;
+
+    let mut ready_v4: Option<(ProxyConfigData, u64)> = None;
+    let cfg_v4 = retry_fetch("https://core.telegram.org/getProxyConfig").await;
+    if let Some(cfg_v4) = cfg_v4 {
+        if snapshot_passes_guards(cfg, &cfg_v4, "getProxyConfig") {
+            let cfg_v4_hash = hash_proxy_config(&cfg_v4);
+            let stable_hits = state.config_v4.observe(cfg_v4_hash);
+            if stable_hits < required_cfg_snapshots {
+                debug!(
+                    stable_hits,
+                    required_cfg_snapshots,
+                    snapshot = format_args!("0x{cfg_v4_hash:016x}"),
+                    "ME config v4 candidate observed"
+                );
+            } else if state.config_v4.is_applied(cfg_v4_hash) {
+                debug!(
+                    snapshot = format_args!("0x{cfg_v4_hash:016x}"),
+                    "ME config v4 stable snapshot already applied"
+                );
+            } else {
+                ready_v4 = Some((cfg_v4, cfg_v4_hash));
+            }
+        }
+    }
+
+    let mut ready_v6: Option<(ProxyConfigData, u64)> = None;
+    let cfg_v6 = retry_fetch("https://core.telegram.org/getProxyConfigV6").await;
+    if let Some(cfg_v6) = cfg_v6 {
+        if snapshot_passes_guards(cfg, &cfg_v6, "getProxyConfigV6") {
+            let cfg_v6_hash = hash_proxy_config(&cfg_v6);
+            let stable_hits = state.config_v6.observe(cfg_v6_hash);
+            if stable_hits < required_cfg_snapshots {
+                debug!(
+                    stable_hits,
+                    required_cfg_snapshots,
+                    snapshot = format_args!("0x{cfg_v6_hash:016x}"),
+                    "ME config v6 candidate observed"
+                );
+            } else if state.config_v6.is_applied(cfg_v6_hash) {
+                debug!(
+                    snapshot = format_args!("0x{cfg_v6_hash:016x}"),
+                    "ME config v6 stable snapshot already applied"
+                );
+            } else {
+                ready_v6 = Some((cfg_v6, cfg_v6_hash));
+            }
+        }
+    }
+
+    if ready_v4.is_some() || ready_v6.is_some() {
+        if map_apply_cooldown_ready(state.last_map_apply_at, apply_cooldown) {
+            let update_v4 = ready_v4
+                .as_ref()
+                .map(|(snapshot, _)| snapshot.map.clone())
+                .unwrap_or_default();
+            let update_v6 = ready_v6
+                .as_ref()
+                .map(|(snapshot, _)| snapshot.map.clone());
+            let update_is_empty =
+                update_v4.is_empty() && update_v6.as_ref().is_none_or(|v| v.is_empty());
+            let apply_outcome = if update_is_empty && !cfg.general.me_snapshot_reject_empty_map {
+                super::pool_config::SnapshotApplyOutcome::AppliedNoDelta
+            } else {
+                pool.update_proxy_maps(update_v4, update_v6).await
+            };
+
+            if matches!(
+                apply_outcome,
+                super::pool_config::SnapshotApplyOutcome::RejectedEmpty
+            ) {
+                warn!("ME config stable snapshot rejected (empty endpoint map)");
+            } else {
+                if let Some((snapshot, hash)) = ready_v4 {
+                    if let Some(dc) = snapshot.default_dc {
+                        pool.default_dc
+                            .store(dc, std::sync::atomic::Ordering::Relaxed);
+                    }
+                    state.config_v4.mark_applied(hash);
+                }
+
+                if let Some((_snapshot, hash)) = ready_v6 {
+                    state.config_v6.mark_applied(hash);
+                }
+
+                state.last_map_apply_at = Some(tokio::time::Instant::now());
+
+                if apply_outcome.changed() {
+                    maps_changed = true;
+                    info!("ME config update applied after stable-gate");
+                } else {
+                    debug!("ME config stable-gate applied with no map delta");
+                }
+            }
+        } else if let Some(last) = state.last_map_apply_at {
+            let wait_secs = map_apply_cooldown_remaining_secs(last, apply_cooldown);
+            debug!(
+                wait_secs,
+                "ME config stable snapshot deferred by cooldown"
+            );
+        }
+    }
+
+    if maps_changed {
+        enqueue_reinit_trigger(reinit_tx, MeReinitTrigger::MapChanged);
+    }
+
+    pool.reset_stun_state();
+
+    if cfg.general.proxy_secret_rotate_runtime {
+        match download_proxy_secret_with_max_len(cfg.general.proxy_secret_len_max).await {
+            Ok(secret) => {
+                let secret_hash = hash_secret(&secret);
+                let stable_hits = state.secret.observe(secret_hash);
+                if stable_hits < required_secret_snapshots {
+                    debug!(
+                        stable_hits,
+                        required_secret_snapshots,
+                        snapshot = format_args!("0x{secret_hash:016x}"),
+                        "proxy-secret candidate observed"
+                    );
+                } else if state.secret.is_applied(secret_hash) {
+                    debug!(
+                        snapshot = format_args!("0x{secret_hash:016x}"),
+                        "proxy-secret stable snapshot already applied"
+                    );
+                } else {
+                    let rotated = pool.update_secret(secret).await;
+                    state.secret.mark_applied(secret_hash);
+                    if rotated {
+                        info!("proxy-secret rotated after stable-gate");
+                    } else {
+                        debug!("proxy-secret stable snapshot confirmed as unchanged");
+                    }
+                }
+            }
+            Err(e) => warn!(error = %e, "proxy-secret update failed"),
+        }
+    } else {
+        debug!("proxy-secret runtime rotation disabled by config");
+    }
+}
+
+pub async fn me_config_updater(
+    pool: Arc<MePool>,
+    mut config_rx: watch::Receiver<Arc<ProxyConfig>>,
+    reinit_tx: mpsc::Sender<MeReinitTrigger>,
+) {
+    let mut state = UpdaterState::default();
+    let mut update_every_secs = config_rx
+        .borrow()
+        .general
+        .effective_update_every_secs()
+        .max(1);
+    let mut update_every = Duration::from_secs(update_every_secs);
+    let mut next_tick = tokio::time::Instant::now() + update_every;
+    info!(update_every_secs, "ME config updater started");
+
+    loop {
+        let sleep = tokio::time::sleep_until(next_tick);
+        tokio::pin!(sleep);
+
+        tokio::select! {
+            _ = &mut sleep => {
+                let cfg = config_rx.borrow().clone();
+                run_update_cycle(&pool, cfg.as_ref(), &mut state, &reinit_tx).await;
+                let refreshed_secs = cfg.general.effective_update_every_secs().max(1);
+                if refreshed_secs != update_every_secs {
+                    info!(
+                        old_update_every_secs = update_every_secs,
+                        new_update_every_secs = refreshed_secs,
+                        "ME config updater interval changed"
+                    );
+                    update_every_secs = refreshed_secs;
+                    update_every = Duration::from_secs(update_every_secs);
+                }
+                next_tick = tokio::time::Instant::now() + update_every;
+            }
+            changed = config_rx.changed() => {
+                if changed.is_err() {
+                    warn!("ME config updater stopped: config channel closed");
+                    break;
+                }
+                let cfg = config_rx.borrow().clone();
+                pool.update_runtime_reinit_policy(
+                    cfg.general.hardswap,
+                    cfg.general.me_pool_drain_ttl_secs,
+                    cfg.general.effective_me_pool_force_close_secs(),
+                    cfg.general.me_pool_min_fresh_ratio,
+                    cfg.general.me_hardswap_warmup_delay_min_ms,
+                    cfg.general.me_hardswap_warmup_delay_max_ms,
+                    cfg.general.me_hardswap_warmup_extra_passes,
+                    cfg.general.me_hardswap_warmup_pass_backoff_base_ms,
+                    cfg.general.me_bind_stale_mode,
+                    cfg.general.me_bind_stale_ttl_secs,
+                    cfg.general.me_secret_atomic_snapshot,
+                    cfg.general.me_deterministic_writer_sort,
+                    cfg.general.me_single_endpoint_shadow_writers,
+                    cfg.general.me_single_endpoint_outage_mode_enabled,
+                    cfg.general.me_single_endpoint_outage_disable_quarantine,
+                    cfg.general.me_single_endpoint_outage_backoff_min_ms,
+                    cfg.general.me_single_endpoint_outage_backoff_max_ms,
+                    cfg.general.me_single_endpoint_shadow_rotate_every_secs,
+                    cfg.general.me_floor_mode,
+                    cfg.general.me_adaptive_floor_idle_secs,
+                    cfg.general.me_adaptive_floor_min_writers_single_endpoint,
+                    cfg.general.me_adaptive_floor_recover_grace_secs,
+                );
+                let new_secs = cfg.general.effective_update_every_secs().max(1);
+                if new_secs == update_every_secs {
+                    continue;
+                }
+
+                if new_secs < update_every_secs {
+                    info!(
+                        old_update_every_secs = update_every_secs,
+                        new_update_every_secs = new_secs,
+                        "ME config updater interval decreased, running immediate refresh"
+                    );
+                    update_every_secs = new_secs;
+                    update_every = Duration::from_secs(update_every_secs);
+                    run_update_cycle(&pool, cfg.as_ref(), &mut state, &reinit_tx).await;
+                    next_tick = tokio::time::Instant::now() + update_every;
+                } else {
+                    info!(
+                        old_update_every_secs = update_every_secs,
+                        new_update_every_secs = new_secs,
+                        "ME config updater interval increased"
+                    );
+                    update_every_secs = new_secs;
+                    update_every = Duration::from_secs(update_every_secs);
+                    next_tick = tokio::time::Instant::now() + update_every;
+                }
+            }
+        }
+    }
+}
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+
+    #[test]
+    fn parse_ipv6_bracketed() {
+        let line = "proxy_for 2 [2001:67c:04e8:f002::d]:80;";
+        let res = parse_proxy_line(line).unwrap();
+        assert_eq!(res.0, 2);
+        assert_eq!(res.1, "2001:67c:04e8:f002::d".parse::<IpAddr>().unwrap());
+        assert_eq!(res.2, 80);
+    }
+
+    #[test]
+    fn parse_ipv6_plain() {
+        let line = "proxy_for 2 2001:67c:04e8:f002::d:80;";
+        let res = parse_proxy_line(line).unwrap();
+        assert_eq!(res.0, 2);
+        assert_eq!(res.1, "2001:67c:04e8:f002::d".parse::<IpAddr>().unwrap());
+        assert_eq!(res.2, 80);
+    }
+
+    #[test]
+    fn parse_ipv4() {
+        let line = "proxy_for 4 91.108.4.195:8888;";
+        let res = parse_proxy_line(line).unwrap();
+        assert_eq!(res.0, 4);
+        assert_eq!(res.1, "91.108.4.195".parse::<IpAddr>().unwrap());
+        assert_eq!(res.2, 8888);
+    }
+}

src/transport/middle_proxy/handshake.rs

@@ -0,0 +1,684 @@
+use std::net::{IpAddr, SocketAddr};
+use std::sync::atomic::Ordering;
+use std::time::{Duration, Instant};
+use std::collections::hash_map::DefaultHasher;
+use std::hash::{Hash, Hasher};
+use socket2::{SockRef, TcpKeepalive};
+#[cfg(target_os = "linux")]
+use libc;
+#[cfg(target_os = "linux")]
+use std::os::fd::{AsRawFd, RawFd};
+#[cfg(target_os = "linux")]
+use std::os::raw::c_int;
+
+use bytes::BytesMut;
+use tokio::io::{AsyncReadExt, AsyncWriteExt, ReadHalf, WriteHalf};
+use tokio::net::{TcpStream, TcpSocket};
+use tokio::time::timeout;
+use tracing::{debug, info, warn};
+
+use crate::config::MeSocksKdfPolicy;
+use crate::crypto::{SecureRandom, build_middleproxy_prekey, derive_middleproxy_keys, sha256};
+use crate::error::{ProxyError, Result};
+use crate::network::IpFamily;
+use crate::network::probe::is_bogon;
+use crate::protocol::constants::{
+    ME_CONNECT_TIMEOUT_SECS, ME_HANDSHAKE_TIMEOUT_SECS, RPC_CRYPTO_AES_U32,
+    RPC_HANDSHAKE_ERROR_U32, rpc_crypto_flags,
+};
+use crate::transport::{UpstreamEgressInfo, UpstreamRouteKind};
+
+use super::codec::{
+    RpcChecksumMode, build_handshake_payload, build_nonce_payload, build_rpc_frame,
+    cbc_decrypt_inplace, cbc_encrypt_padded, parse_handshake_flags, parse_nonce_payload,
+    read_rpc_frame_plaintext, rpc_crc,
+};
+use super::wire::{extract_ip_material, IpMaterial};
+use super::MePool;
+
+const ME_KDF_DRIFT_STRICT: bool = false;
+
+#[derive(Clone, Copy, Debug, PartialEq, Eq, Hash)]
+enum KdfClientPortSource {
+    LocalSocket = 0,
+    SocksBound = 1,
+}
+
+impl KdfClientPortSource {
+    fn from_socks_bound_port(socks_bound_port: Option<u16>) -> Self {
+        if socks_bound_port.is_some() {
+            Self::SocksBound
+        } else {
+            Self::LocalSocket
+        }
+    }
+}
+
+/// Result of a successful ME handshake with timings.
+pub(crate) struct HandshakeOutput {
+    pub rd: ReadHalf<TcpStream>,
+    pub wr: WriteHalf<TcpStream>,
+    pub read_key: [u8; 32],
+    pub read_iv: [u8; 16],
+    pub write_key: [u8; 32],
+    pub write_iv: [u8; 16],
+    pub crc_mode: RpcChecksumMode,
+    pub handshake_ms: f64,
+}
+
+impl MePool {
+    fn kdf_material_fingerprint(
+        local_ip_nat: IpAddr,
+        peer_addr_nat: SocketAddr,
+        reflected_ip: Option<IpAddr>,
+        socks_bound_ip: Option<IpAddr>,
+        client_port_source: KdfClientPortSource,
+    ) -> u64 {
+        let mut hasher = DefaultHasher::new();
+        local_ip_nat.hash(&mut hasher);
+        peer_addr_nat.hash(&mut hasher);
+        reflected_ip.hash(&mut hasher);
+        socks_bound_ip.hash(&mut hasher);
+        client_port_source.hash(&mut hasher);
+        hasher.finish()
+    }
+
+    async fn resolve_dc_idx_for_endpoint(&self, addr: SocketAddr) -> Option<i16> {
+        if addr.is_ipv4() {
+            let map = self.proxy_map_v4.read().await;
+            for (dc, addrs) in map.iter() {
+                if addrs
+                    .iter()
+                    .any(|(ip, port)| SocketAddr::new(*ip, *port) == addr)
+                {
+                    let abs_dc = dc.abs();
+                    if abs_dc > 0
+                        && let Ok(dc_idx) = i16::try_from(abs_dc)
+                    {
+                        return Some(dc_idx);
+                    }
+                }
+            }
+        } else {
+            let map = self.proxy_map_v6.read().await;
+            for (dc, addrs) in map.iter() {
+                if addrs
+                    .iter()
+                    .any(|(ip, port)| SocketAddr::new(*ip, *port) == addr)
+                {
+                    let abs_dc = dc.abs();
+                    if abs_dc > 0
+                        && let Ok(dc_idx) = i16::try_from(abs_dc)
+                    {
+                        return Some(dc_idx);
+                    }
+                }
+            }
+        }
+        None
+    }
+
+    fn direct_bind_ip_for_stun(
+        family: IpFamily,
+        upstream_egress: Option<UpstreamEgressInfo>,
+    ) -> Option<IpAddr> {
+        let info = upstream_egress?;
+        if info.route_kind != UpstreamRouteKind::Direct {
+            return None;
+        }
+        match (family, info.direct_bind_ip) {
+            (IpFamily::V4, Some(IpAddr::V4(ip))) => Some(IpAddr::V4(ip)),
+            (IpFamily::V6, Some(IpAddr::V6(ip))) => Some(IpAddr::V6(ip)),
+            _ => None,
+        }
+    }
+
+    fn select_socks_bound_addr(
+        family: IpFamily,
+        upstream_egress: Option<UpstreamEgressInfo>,
+    ) -> Option<SocketAddr> {
+        let info = upstream_egress?;
+        if !matches!(
+            info.route_kind,
+            UpstreamRouteKind::Socks4 | UpstreamRouteKind::Socks5
+        ) {
+            return None;
+        }
+        let bound = info.socks_bound_addr?;
+        let family_matches = matches!(
+            (family, bound.ip()),
+            (IpFamily::V4, IpAddr::V4(_)) | (IpFamily::V6, IpAddr::V6(_))
+        );
+        if !family_matches || is_bogon(bound.ip()) || bound.ip().is_unspecified() {
+            return None;
+        }
+        Some(bound)
+    }
+
+    fn is_socks_route(upstream_egress: Option<UpstreamEgressInfo>) -> bool {
+        matches!(
+            upstream_egress.map(|info| info.route_kind),
+            Some(UpstreamRouteKind::Socks4 | UpstreamRouteKind::Socks5)
+        )
+    }
+
+    /// TCP connect with timeout + return RTT in milliseconds.
+    pub(crate) async fn connect_tcp(
+        &self,
+        addr: SocketAddr,
+    ) -> Result<(TcpStream, f64, Option<UpstreamEgressInfo>)> {
+        let start = Instant::now();
+        let (stream, upstream_egress) = if let Some(upstream) = &self.upstream {
+            let dc_idx = self.resolve_dc_idx_for_endpoint(addr).await;
+            let (stream, egress) = upstream.connect_with_details(addr, dc_idx, None).await?;
+            (stream, Some(egress))
+        } else {
+            let connect_fut = async {
+                if addr.is_ipv6()
+                    && let Some(v6) = self.detected_ipv6
+                {
+                    match TcpSocket::new_v6() {
+                        Ok(sock) => {
+                            if let Err(e) = sock.bind(SocketAddr::new(IpAddr::V6(v6), 0)) {
+                                debug!(error = %e, bind_ip = %v6, "ME IPv6 bind failed, falling back to default bind");
+                            } else {
+                                match sock.connect(addr).await {
+                                    Ok(stream) => return Ok(stream),
+                                    Err(e) => debug!(error = %e, target = %addr, "ME IPv6 bound connect failed, retrying default connect"),
+                                }
+                            }
+                        }
+                        Err(e) => debug!(error = %e, "ME IPv6 socket creation failed, falling back to default connect"),
+                    }
+                }
+                TcpStream::connect(addr).await
+            };
+
+            let stream = timeout(Duration::from_secs(ME_CONNECT_TIMEOUT_SECS), connect_fut)
+                .await
+                .map_err(|_| ProxyError::ConnectionTimeout {
+                    addr: addr.to_string(),
+                })??;
+            (stream, None)
+        };
+
+        let connect_ms = start.elapsed().as_secs_f64() * 1000.0;
+        stream.set_nodelay(true).ok();
+        if let Err(e) = Self::configure_keepalive(&stream) {
+            warn!(error = %e, "ME keepalive setup failed");
+        }
+        #[cfg(target_os = "linux")]
+        if let Err(e) = Self::configure_user_timeout(stream.as_raw_fd()) {
+            warn!(error = %e, "ME TCP_USER_TIMEOUT setup failed");
+        }
+        Ok((stream, connect_ms, upstream_egress))
+    }
+
+    fn configure_keepalive(stream: &TcpStream) -> std::io::Result<()> {
+        let sock = SockRef::from(stream);
+        let ka = TcpKeepalive::new()
+            .with_time(Duration::from_secs(30))
+            .with_interval(Duration::from_secs(10))
+            .with_retries(3);
+        sock.set_tcp_keepalive(&ka)?;
+        sock.set_keepalive(true)?;
+        Ok(())
+    }
+
+    #[cfg(target_os = "linux")]
+    fn configure_user_timeout(fd: RawFd) -> std::io::Result<()> {
+        let timeout_ms: c_int = 30_000;
+        let rc = unsafe {
+            libc::setsockopt(
+                fd,
+                libc::IPPROTO_TCP,
+                libc::TCP_USER_TIMEOUT,
+                &timeout_ms as *const _ as *const libc::c_void,
+                std::mem::size_of_val(&timeout_ms) as libc::socklen_t,
+            )
+        };
+        if rc != 0 {
+            return Err(std::io::Error::last_os_error());
+        }
+        Ok(())
+    }
+
+    /// Perform full ME RPC handshake on an established TCP stream.
+    /// Returns cipher keys/ivs and split halves; does not register writer.
+    pub(crate) async fn handshake_only(
+        &self,
+        stream: TcpStream,
+        addr: SocketAddr,
+        upstream_egress: Option<UpstreamEgressInfo>,
+        rng: &SecureRandom,
+    ) -> Result<HandshakeOutput> {
+        let hs_start = Instant::now();
+
+        let local_addr = stream.local_addr().map_err(ProxyError::Io)?;
+        let transport_peer_addr = stream.peer_addr().map_err(ProxyError::Io)?;
+        let peer_addr = addr;
+
+        let _ = self.maybe_detect_nat_ip(local_addr.ip()).await;
+        let family = if local_addr.ip().is_ipv4() {
+            IpFamily::V4
+        } else {
+            IpFamily::V6
+        };
+        let is_socks_route = Self::is_socks_route(upstream_egress);
+        let socks_bound_addr = Self::select_socks_bound_addr(family, upstream_egress);
+        let reflected = if let Some(bound) = socks_bound_addr {
+            Some(bound)
+        } else if is_socks_route {
+            match self.socks_kdf_policy() {
+                MeSocksKdfPolicy::Strict => {
+                    self.stats.increment_me_socks_kdf_strict_reject();
+                    return Err(ProxyError::InvalidHandshake(
+                        "SOCKS route returned no valid BND.ADDR for ME KDF (strict policy)"
+                            .to_string(),
+                    ));
+                }
+                MeSocksKdfPolicy::Compat => {
+                    self.stats.increment_me_socks_kdf_compat_fallback();
+                    if self.nat_probe {
+                        let bind_ip = Self::direct_bind_ip_for_stun(family, upstream_egress);
+                        self.maybe_reflect_public_addr(family, bind_ip).await
+                    } else {
+                        None
+                    }
+                }
+            }
+        } else if self.nat_probe {
+            let bind_ip = Self::direct_bind_ip_for_stun(family, upstream_egress);
+            self.maybe_reflect_public_addr(family, bind_ip).await
+        } else {
+            None
+        };
+
+        let local_addr_nat = self.translate_our_addr_with_reflection(local_addr, reflected);
+        let peer_addr_nat = SocketAddr::new(self.translate_ip_for_nat(peer_addr.ip()), peer_addr.port());
+        let (mut rd, mut wr) = tokio::io::split(stream);
+
+        let my_nonce: [u8; 16] = rng.bytes(16).try_into().unwrap();
+        let crypto_ts = std::time::SystemTime::now()
+            .duration_since(std::time::UNIX_EPOCH)
+            .unwrap_or_default()
+            .as_secs() as u32;
+
+        let secret_atomic_snapshot = self.secret_atomic_snapshot.load(Ordering::Relaxed);
+        let (ks, secret) = if secret_atomic_snapshot {
+            let snapshot = self.secret_snapshot().await;
+            (snapshot.key_selector, snapshot.secret)
+        } else {
+            // Backward-compatible mode: key selector and secret may come from different updates.
+            let key_selector = self.key_selector().await;
+            let secret = self.secret_snapshot().await.secret;
+            (key_selector, secret)
+        };
+        let nonce_payload = build_nonce_payload(ks, crypto_ts, &my_nonce);
+        let nonce_frame = build_rpc_frame(-2, &nonce_payload, RpcChecksumMode::Crc32);
+        let dump = hex_dump(&nonce_frame[..nonce_frame.len().min(44)]);
+        debug!(
+            key_selector = format_args!("0x{ks:08x}"),
+            crypto_ts,
+            frame_len = nonce_frame.len(),
+            nonce_frame_hex = %dump,
+            "Sending ME nonce frame"
+        );
+        wr.write_all(&nonce_frame).await.map_err(ProxyError::Io)?;
+        wr.flush().await.map_err(ProxyError::Io)?;
+
+        let (srv_seq, srv_nonce_payload) = timeout(
+            Duration::from_secs(ME_HANDSHAKE_TIMEOUT_SECS),
+            read_rpc_frame_plaintext(&mut rd),
+        )
+        .await
+        .map_err(|_| ProxyError::TgHandshakeTimeout)??;
+
+        if srv_seq != -2 {
+            return Err(ProxyError::InvalidHandshake(format!("Expected seq=-2, got {srv_seq}")));
+        }
+
+        let (srv_key_select, schema, srv_ts, srv_nonce) = parse_nonce_payload(&srv_nonce_payload)?;
+        if schema != RPC_CRYPTO_AES_U32 {
+            warn!(schema = format_args!("0x{schema:08x}"), "Unsupported ME crypto schema");
+            return Err(ProxyError::InvalidHandshake(format!(
+                "Unsupported crypto schema: 0x{schema:x}"
+            )));
+        }
+
+        if srv_key_select != ks {
+            return Err(ProxyError::InvalidHandshake(format!(
+                "Server key_select 0x{srv_key_select:08x} != client 0x{ks:08x}"
+            )));
+        }
+
+        let skew = crypto_ts.abs_diff(srv_ts);
+        if skew > 30 {
+            return Err(ProxyError::InvalidHandshake(format!(
+                "nonce crypto_ts skew too large: client={crypto_ts}, server={srv_ts}, skew={skew}s"
+            )));
+        }
+
+        info!(
+            %local_addr,
+            %local_addr_nat,
+            reflected_ip = reflected.map(|r| r.ip()).as_ref().map(ToString::to_string),
+            %peer_addr,
+            %transport_peer_addr,
+            %peer_addr_nat,
+            socks_bound_addr = socks_bound_addr.map(|v| v.to_string()),
+            key_selector = format_args!("0x{ks:08x}"),
+            crypto_schema = format_args!("0x{schema:08x}"),
+            skew_secs = skew,
+            "ME key derivation parameters"
+        );
+
+        let ts_bytes = crypto_ts.to_le_bytes();
+        let server_port_bytes = peer_addr_nat.port().to_le_bytes();
+        let socks_bound_port = socks_bound_addr
+            .map(|bound| bound.port())
+            .filter(|port| *port != 0);
+        let client_port_for_kdf = socks_bound_port.unwrap_or(local_addr_nat.port());
+        let client_port_source = KdfClientPortSource::from_socks_bound_port(socks_bound_port);
+        let kdf_fingerprint = Self::kdf_material_fingerprint(
+            local_addr_nat.ip(),
+            peer_addr_nat,
+            reflected.map(|value| value.ip()),
+            socks_bound_addr.map(|value| value.ip()),
+            client_port_source,
+        );
+        let previous_kdf_fingerprint = {
+            let kdf_fingerprint_guard = self.kdf_material_fingerprint.read().await;
+            kdf_fingerprint_guard.get(&peer_addr_nat).copied()
+        };
+        if let Some((prev_fingerprint, prev_client_port)) = previous_kdf_fingerprint
+        {
+            if prev_fingerprint != kdf_fingerprint {
+                self.stats.increment_me_kdf_drift_total();
+                warn!(
+                    %peer_addr_nat,
+                    %local_addr_nat,
+                    client_port_for_kdf,
+                    client_port_source = ?client_port_source,
+                    "ME KDF material drift detected for endpoint"
+                );
+                if ME_KDF_DRIFT_STRICT {
+                    return Err(ProxyError::InvalidHandshake(
+                        "ME KDF material drift detected (strict mode)".to_string(),
+                    ));
+                }
+            } else if prev_client_port != client_port_for_kdf {
+                self.stats.increment_me_kdf_port_only_drift_total();
+                debug!(
+                    %peer_addr_nat,
+                    previous_client_port_for_kdf = prev_client_port,
+                    client_port_for_kdf,
+                    client_port_source = ?client_port_source,
+                    "ME KDF client port changed with stable material"
+                );
+            }
+        }
+        // Keep fingerprint updates eventually consistent for diagnostics while avoiding
+        // serializing all concurrent handshakes on a single async mutex.
+        let mut kdf_fingerprint_guard = self.kdf_material_fingerprint.write().await;
+        kdf_fingerprint_guard.insert(peer_addr_nat, (kdf_fingerprint, client_port_for_kdf));
+        drop(kdf_fingerprint_guard);
+
+        let client_port_bytes = client_port_for_kdf.to_le_bytes();
+
+        let server_ip = extract_ip_material(peer_addr_nat);
+        let client_ip = extract_ip_material(local_addr_nat);
+
+        let (srv_ip_opt, clt_ip_opt, clt_v6_opt, srv_v6_opt, hs_our_ip, hs_peer_ip) = match (server_ip, client_ip) {
+            (IpMaterial::V4(mut srv), IpMaterial::V4(mut clt)) => {
+                srv.reverse();
+                clt.reverse();
+                (Some(srv), Some(clt), None, None, clt, srv)
+            }
+            (IpMaterial::V6(srv), IpMaterial::V6(clt)) => {
+                let zero = [0u8; 4];
+                (None, None, Some(clt), Some(srv), zero, zero)
+            }
+            _ => {
+                return Err(ProxyError::InvalidHandshake(
+                    "mixed IPv4/IPv6 endpoints are not supported for ME key derivation".to_string(),
+                ));
+            }
+        };
+
+        let diag_level: u8 = std::env::var("ME_DIAG").ok().and_then(|v| v.parse().ok()).unwrap_or(0);
+
+        let prekey_client = build_middleproxy_prekey(
+            &srv_nonce,
+            &my_nonce,
+            &ts_bytes,
+            srv_ip_opt.as_ref().map(|x| &x[..]),
+            &client_port_bytes,
+            b"CLIENT",
+            clt_ip_opt.as_ref().map(|x| &x[..]),
+            &server_port_bytes,
+            &secret,
+            clt_v6_opt.as_ref(),
+            srv_v6_opt.as_ref(),
+        );
+        let prekey_server = build_middleproxy_prekey(
+            &srv_nonce,
+            &my_nonce,
+            &ts_bytes,
+            srv_ip_opt.as_ref().map(|x| &x[..]),
+            &client_port_bytes,
+            b"SERVER",
+            clt_ip_opt.as_ref().map(|x| &x[..]),
+            &server_port_bytes,
+            &secret,
+            clt_v6_opt.as_ref(),
+            srv_v6_opt.as_ref(),
+        );
+
+        let (wk, wi) = derive_middleproxy_keys(
+            &srv_nonce,
+            &my_nonce,
+            &ts_bytes,
+            srv_ip_opt.as_ref().map(|x| &x[..]),
+            &client_port_bytes,
+            b"CLIENT",
+            clt_ip_opt.as_ref().map(|x| &x[..]),
+            &server_port_bytes,
+            &secret,
+            clt_v6_opt.as_ref(),
+            srv_v6_opt.as_ref(),
+        );
+        let (rk, ri) = derive_middleproxy_keys(
+            &srv_nonce,
+            &my_nonce,
+            &ts_bytes,
+            srv_ip_opt.as_ref().map(|x| &x[..]),
+            &client_port_bytes,
+            b"SERVER",
+            clt_ip_opt.as_ref().map(|x| &x[..]),
+            &server_port_bytes,
+            &secret,
+            clt_v6_opt.as_ref(),
+            srv_v6_opt.as_ref(),
+        );
+
+        let requested_crc_mode = RpcChecksumMode::Crc32c;
+        let hs_payload = build_handshake_payload(
+            hs_our_ip,
+            local_addr.port(),
+            hs_peer_ip,
+            peer_addr.port(),
+            requested_crc_mode.advertised_flags(),
+        );
+        let hs_frame = build_rpc_frame(-1, &hs_payload, RpcChecksumMode::Crc32);
+        if diag_level >= 1 {
+            info!(
+                write_key = %hex_dump(&wk),
+                write_iv = %hex_dump(&wi),
+                read_key = %hex_dump(&rk),
+                read_iv = %hex_dump(&ri),
+                srv_ip = %srv_ip_opt.map(|ip| hex_dump(&ip)).unwrap_or_default(),
+                clt_ip = %clt_ip_opt.map(|ip| hex_dump(&ip)).unwrap_or_default(),
+                srv_port = %hex_dump(&server_port_bytes),
+                clt_port = %hex_dump(&client_port_bytes),
+                crypto_ts = %hex_dump(&ts_bytes),
+                nonce_srv = %hex_dump(&srv_nonce),
+                nonce_clt = %hex_dump(&my_nonce),
+                prekey_sha256_client = %hex_dump(&sha256(&prekey_client)),
+                prekey_sha256_server = %hex_dump(&sha256(&prekey_server)),
+                hs_plain = %hex_dump(&hs_frame),
+                proxy_secret_sha256 = %hex_dump(&sha256(&secret)),
+                "ME diag: derived keys and handshake plaintext"
+            );
+        }
+        if diag_level >= 2 {
+            info!(
+                prekey_client = %hex_dump(&prekey_client),
+                prekey_server = %hex_dump(&prekey_server),
+                "ME diag: full prekey buffers"
+            );
+        }
+
+        let (encrypted_hs, write_iv) = cbc_encrypt_padded(&wk, &wi, &hs_frame)?;
+        if diag_level >= 1 {
+            info!(
+                hs_cipher = %hex_dump(&encrypted_hs),
+                "ME diag: handshake ciphertext"
+            );
+        }
+        wr.write_all(&encrypted_hs).await.map_err(ProxyError::Io)?;
+        wr.flush().await.map_err(ProxyError::Io)?;
+
+        let deadline = Instant::now() + Duration::from_secs(ME_HANDSHAKE_TIMEOUT_SECS);
+        let mut enc_buf = BytesMut::with_capacity(256);
+        let mut dec_buf = BytesMut::with_capacity(256);
+        let mut read_iv = ri;
+        let mut negotiated_crc_mode = RpcChecksumMode::Crc32;
+        let mut handshake_ok = false;
+
+        while Instant::now() < deadline && !handshake_ok {
+            let remaining = deadline - Instant::now();
+            let mut tmp = [0u8; 256];
+            let n = match timeout(remaining, rd.read(&mut tmp)).await {
+                Ok(Ok(0)) => {
+                    return Err(ProxyError::Io(std::io::Error::new(
+                        std::io::ErrorKind::UnexpectedEof,
+                        "ME closed during handshake",
+                    )));
+                }
+                Ok(Ok(n)) => n,
+                Ok(Err(e)) => return Err(ProxyError::Io(e)),
+                Err(_) => return Err(ProxyError::TgHandshakeTimeout),
+            };
+
+            enc_buf.extend_from_slice(&tmp[..n]);
+
+            let blocks = enc_buf.len() / 16 * 16;
+            if blocks > 0 {
+                let mut chunk = vec![0u8; blocks];
+                chunk.copy_from_slice(&enc_buf[..blocks]);
+                read_iv = cbc_decrypt_inplace(&rk, &read_iv, &mut chunk)?;
+                dec_buf.extend_from_slice(&chunk);
+                let _ = enc_buf.split_to(blocks);
+            }
+
+            while dec_buf.len() >= 4 {
+                let fl = u32::from_le_bytes(dec_buf[0..4].try_into().unwrap()) as usize;
+
+                if fl == 4 {
+                    let _ = dec_buf.split_to(4);
+                    continue;
+                }
+                if !(12..=(1 << 24)).contains(&fl) {
+                    return Err(ProxyError::InvalidHandshake(format!(
+                        "Bad HS response frame len: {fl}"
+                    )));
+                }
+                if dec_buf.len() < fl {
+                    break;
+                }
+
+                let frame = dec_buf.split_to(fl);
+                let pe = fl - 4;
+                let ec = u32::from_le_bytes(frame[pe..pe + 4].try_into().unwrap());
+                let ac = rpc_crc(RpcChecksumMode::Crc32, &frame[..pe]);
+                if ec != ac {
+                    return Err(ProxyError::InvalidHandshake(format!(
+                        "HS CRC mismatch: 0x{ec:08x} vs 0x{ac:08x}"
+                    )));
+                }
+
+                let hs_payload = &frame[8..pe];
+                if hs_payload.len() < 4 {
+                    return Err(ProxyError::InvalidHandshake(
+                        "Handshake payload too short".to_string(),
+                    ));
+                }
+                let hs_type = u32::from_le_bytes(hs_payload[0..4].try_into().unwrap());
+                if hs_type == RPC_HANDSHAKE_ERROR_U32 {
+                    let err_code = if hs_payload.len() >= 8 {
+                        i32::from_le_bytes(hs_payload[4..8].try_into().unwrap())
+                    } else {
+                        -1
+                    };
+                    self.stats.increment_me_handshake_reject_total();
+                    self.stats.increment_me_handshake_error_code(err_code);
+                    return Err(ProxyError::InvalidHandshake(format!(
+                        "ME rejected handshake (error={err_code})"
+                    )));
+                }
+                let hs_flags = parse_handshake_flags(hs_payload)?;
+                if hs_flags & 0xff != 0 {
+                    return Err(ProxyError::InvalidHandshake(format!(
+                        "Unsupported handshake flags: 0x{hs_flags:08x}"
+                    )));
+                }
+                negotiated_crc_mode = if (hs_flags & requested_crc_mode.advertised_flags()) != 0 {
+                    RpcChecksumMode::from_handshake_flags(hs_flags)
+                } else if (hs_flags & rpc_crypto_flags::USE_CRC32C) != 0 {
+                    return Err(ProxyError::InvalidHandshake(format!(
+                        "Peer negotiated unsupported CRC flags: 0x{hs_flags:08x}"
+                    )));
+                } else {
+                    RpcChecksumMode::Crc32
+                };
+
+                handshake_ok = true;
+                break;
+            }
+        }
+
+        if !handshake_ok {
+            return Err(ProxyError::TgHandshakeTimeout);
+        }
+
+        let handshake_ms = hs_start.elapsed().as_secs_f64() * 1000.0;
+        info!(%addr, "RPC handshake OK");
+
+        Ok(HandshakeOutput {
+            rd,
+            wr,
+            read_key: rk,
+            read_iv,
+            write_key: wk,
+            write_iv,
+            crc_mode: negotiated_crc_mode,
+            handshake_ms,
+        })
+    }
+}
+
+fn hex_dump(data: &[u8]) -> String {
+    const MAX: usize = 64;
+    let mut out = String::with_capacity(data.len() * 2 + 3);
+    for (i, b) in data.iter().take(MAX).enumerate() {
+        if i > 0 {
+            out.push(' ');
+        }
+        out.push_str(&format!("{b:02x}"));
+    }
+    if data.len() > MAX {
+        out.push_str(" …");
+    }
+    out
+}

src/transport/middle_proxy/health.rs

@@ -0,0 +1,696 @@
+use std::collections::HashMap;
+use std::collections::HashSet;
+use std::net::SocketAddr;
+use std::sync::Arc;
+use std::time::{Duration, Instant};
+
+use rand::Rng;
+use tracing::{debug, info, warn};
+
+use crate::config::MeFloorMode;
+use crate::crypto::SecureRandom;
+use crate::network::IpFamily;
+
+use super::MePool;
+
+const HEALTH_INTERVAL_SECS: u64 = 1;
+const JITTER_FRAC_NUM: u64 = 2; // jitter up to 50% of backoff
+#[allow(dead_code)]
+const MAX_CONCURRENT_PER_DC_DEFAULT: usize = 1;
+const SHADOW_ROTATE_RETRY_SECS: u64 = 30;
+const IDLE_REFRESH_TRIGGER_BASE_SECS: u64 = 45;
+const IDLE_REFRESH_TRIGGER_JITTER_SECS: u64 = 5;
+const IDLE_REFRESH_RETRY_SECS: u64 = 8;
+const IDLE_REFRESH_SUCCESS_GUARD_SECS: u64 = 5;
+
+pub async fn me_health_monitor(pool: Arc<MePool>, rng: Arc<SecureRandom>, _min_connections: usize) {
+    let mut backoff: HashMap<(i32, IpFamily), u64> = HashMap::new();
+    let mut next_attempt: HashMap<(i32, IpFamily), Instant> = HashMap::new();
+    let mut inflight: HashMap<(i32, IpFamily), usize> = HashMap::new();
+    let mut outage_backoff: HashMap<(i32, IpFamily), u64> = HashMap::new();
+    let mut outage_next_attempt: HashMap<(i32, IpFamily), Instant> = HashMap::new();
+    let mut single_endpoint_outage: HashSet<(i32, IpFamily)> = HashSet::new();
+    let mut shadow_rotate_deadline: HashMap<(i32, IpFamily), Instant> = HashMap::new();
+    let mut idle_refresh_next_attempt: HashMap<(i32, IpFamily), Instant> = HashMap::new();
+    let mut adaptive_idle_since: HashMap<(i32, IpFamily), Instant> = HashMap::new();
+    let mut adaptive_recover_until: HashMap<(i32, IpFamily), Instant> = HashMap::new();
+    loop {
+        tokio::time::sleep(Duration::from_secs(HEALTH_INTERVAL_SECS)).await;
+        pool.prune_closed_writers().await;
+        check_family(
+            IpFamily::V4,
+            &pool,
+            &rng,
+            &mut backoff,
+            &mut next_attempt,
+            &mut inflight,
+            &mut outage_backoff,
+            &mut outage_next_attempt,
+            &mut single_endpoint_outage,
+            &mut shadow_rotate_deadline,
+            &mut idle_refresh_next_attempt,
+            &mut adaptive_idle_since,
+            &mut adaptive_recover_until,
+        )
+        .await;
+        check_family(
+            IpFamily::V6,
+            &pool,
+            &rng,
+            &mut backoff,
+            &mut next_attempt,
+            &mut inflight,
+            &mut outage_backoff,
+            &mut outage_next_attempt,
+            &mut single_endpoint_outage,
+            &mut shadow_rotate_deadline,
+            &mut idle_refresh_next_attempt,
+            &mut adaptive_idle_since,
+            &mut adaptive_recover_until,
+        )
+        .await;
+    }
+}
+
+async fn check_family(
+    family: IpFamily,
+    pool: &Arc<MePool>,
+    rng: &Arc<SecureRandom>,
+    backoff: &mut HashMap<(i32, IpFamily), u64>,
+    next_attempt: &mut HashMap<(i32, IpFamily), Instant>,
+    inflight: &mut HashMap<(i32, IpFamily), usize>,
+    outage_backoff: &mut HashMap<(i32, IpFamily), u64>,
+    outage_next_attempt: &mut HashMap<(i32, IpFamily), Instant>,
+    single_endpoint_outage: &mut HashSet<(i32, IpFamily)>,
+    shadow_rotate_deadline: &mut HashMap<(i32, IpFamily), Instant>,
+    idle_refresh_next_attempt: &mut HashMap<(i32, IpFamily), Instant>,
+    adaptive_idle_since: &mut HashMap<(i32, IpFamily), Instant>,
+    adaptive_recover_until: &mut HashMap<(i32, IpFamily), Instant>,
+) {
+    let enabled = match family {
+        IpFamily::V4 => pool.decision.ipv4_me,
+        IpFamily::V6 => pool.decision.ipv6_me,
+    };
+    if !enabled {
+        return;
+    }
+
+    let map = match family {
+        IpFamily::V4 => pool.proxy_map_v4.read().await.clone(),
+        IpFamily::V6 => pool.proxy_map_v6.read().await.clone(),
+    };
+
+    let mut dc_endpoints = HashMap::<i32, Vec<SocketAddr>>::new();
+    for (dc, addrs) in map {
+        let entry = dc_endpoints.entry(dc.abs()).or_default();
+        for (ip, port) in addrs {
+            entry.push(SocketAddr::new(ip, port));
+        }
+    }
+    for endpoints in dc_endpoints.values_mut() {
+        endpoints.sort_unstable();
+        endpoints.dedup();
+    }
+
+    if pool.floor_mode() == MeFloorMode::Static {
+        adaptive_idle_since.clear();
+        adaptive_recover_until.clear();
+    }
+
+    let mut live_addr_counts = HashMap::<SocketAddr, usize>::new();
+    let mut live_writer_ids_by_addr = HashMap::<SocketAddr, Vec<u64>>::new();
+    for writer in pool.writers.read().await.iter().filter(|w| {
+        !w.draining.load(std::sync::atomic::Ordering::Relaxed)
+    }) {
+        *live_addr_counts.entry(writer.addr).or_insert(0) += 1;
+        live_writer_ids_by_addr
+            .entry(writer.addr)
+            .or_default()
+            .push(writer.id);
+    }
+    let writer_idle_since = pool.registry.writer_idle_since_snapshot().await;
+
+    for (dc, endpoints) in dc_endpoints {
+        if endpoints.is_empty() {
+            continue;
+        }
+        let key = (dc, family);
+        let reduce_for_idle = should_reduce_floor_for_idle(
+            pool,
+            key,
+            &endpoints,
+            &live_writer_ids_by_addr,
+            adaptive_idle_since,
+            adaptive_recover_until,
+        )
+        .await;
+        let required = pool.required_writers_for_dc_with_floor_mode(endpoints.len(), reduce_for_idle);
+        let alive = endpoints
+            .iter()
+            .map(|addr| *live_addr_counts.get(addr).unwrap_or(&0))
+            .sum::<usize>();
+
+        if endpoints.len() == 1 && pool.single_endpoint_outage_mode_enabled() && alive == 0 {
+            if single_endpoint_outage.insert(key) {
+                pool.stats.increment_me_single_endpoint_outage_enter_total();
+                warn!(
+                    dc = %dc,
+                    ?family,
+                    required,
+                    endpoint_count = endpoints.len(),
+                    "Single-endpoint DC outage detected"
+                );
+            }
+
+            recover_single_endpoint_outage(
+                pool,
+                rng,
+                key,
+                endpoints[0],
+                required,
+                outage_backoff,
+                outage_next_attempt,
+            )
+            .await;
+            continue;
+        }
+
+        if single_endpoint_outage.remove(&key) {
+            pool.stats.increment_me_single_endpoint_outage_exit_total();
+            outage_backoff.remove(&key);
+            outage_next_attempt.remove(&key);
+            shadow_rotate_deadline.remove(&key);
+            idle_refresh_next_attempt.remove(&key);
+            adaptive_idle_since.remove(&key);
+            adaptive_recover_until.remove(&key);
+            info!(
+                dc = %dc,
+                ?family,
+                alive,
+                required,
+                endpoint_count = endpoints.len(),
+                "Single-endpoint DC outage recovered"
+            );
+        }
+
+        if alive >= required {
+            maybe_refresh_idle_writer_for_dc(
+                pool,
+                rng,
+                key,
+                dc,
+                family,
+                &endpoints,
+                alive,
+                required,
+                &live_writer_ids_by_addr,
+                &writer_idle_since,
+                idle_refresh_next_attempt,
+            )
+            .await;
+            maybe_rotate_single_endpoint_shadow(
+                pool,
+                rng,
+                key,
+                dc,
+                family,
+                &endpoints,
+                alive,
+                required,
+                &live_writer_ids_by_addr,
+                shadow_rotate_deadline,
+            )
+            .await;
+            continue;
+        }
+        let missing = required - alive;
+
+        let now = Instant::now();
+        if let Some(ts) = next_attempt.get(&key)
+            && now < *ts
+        {
+            continue;
+        }
+
+        let max_concurrent = pool.me_reconnect_max_concurrent_per_dc.max(1) as usize;
+        if *inflight.get(&key).unwrap_or(&0) >= max_concurrent {
+            continue;
+        }
+        if pool.has_refill_inflight_for_endpoints(&endpoints).await {
+            debug!(
+                dc = %dc,
+                ?family,
+                alive,
+                required,
+                endpoint_count = endpoints.len(),
+                "Skipping health reconnect: immediate refill is already in flight for this DC group"
+            );
+            continue;
+        }
+        *inflight.entry(key).or_insert(0) += 1;
+
+        let mut restored = 0usize;
+        for _ in 0..missing {
+            let res = tokio::time::timeout(
+                pool.me_one_timeout,
+                pool.connect_endpoints_round_robin(&endpoints, rng.as_ref()),
+            )
+            .await;
+            match res {
+                Ok(true) => {
+                    restored += 1;
+                    pool.stats.increment_me_reconnect_success();
+                }
+                Ok(false) => {
+                    pool.stats.increment_me_reconnect_attempt();
+                    debug!(dc = %dc, ?family, "ME round-robin reconnect failed")
+                }
+                Err(_) => {
+                    pool.stats.increment_me_reconnect_attempt();
+                    debug!(dc = %dc, ?family, "ME reconnect timed out");
+                }
+            }
+        }
+
+        let now_alive = alive + restored;
+        if now_alive >= required {
+            info!(
+                dc = %dc,
+                ?family,
+                alive = now_alive,
+                required,
+                endpoint_count = endpoints.len(),
+                "ME writer floor restored for DC"
+            );
+            backoff.insert(key, pool.me_reconnect_backoff_base.as_millis() as u64);
+            let jitter = pool.me_reconnect_backoff_base.as_millis() as u64 / JITTER_FRAC_NUM;
+            let wait = pool.me_reconnect_backoff_base
+                + Duration::from_millis(rand::rng().random_range(0..=jitter.max(1)));
+            next_attempt.insert(key, now + wait);
+        } else {
+            let curr = *backoff.get(&key).unwrap_or(&(pool.me_reconnect_backoff_base.as_millis() as u64));
+            let next_ms = (curr.saturating_mul(2)).min(pool.me_reconnect_backoff_cap.as_millis() as u64);
+            backoff.insert(key, next_ms);
+            let jitter = next_ms / JITTER_FRAC_NUM;
+            let wait = Duration::from_millis(next_ms)
+                + Duration::from_millis(rand::rng().random_range(0..=jitter.max(1)));
+            next_attempt.insert(key, now + wait);
+            if pool.is_runtime_ready() {
+                warn!(
+                    dc = %dc,
+                    ?family,
+                    alive = now_alive,
+                    required,
+                    endpoint_count = endpoints.len(),
+                    backoff_ms = next_ms,
+                    "DC writer floor is below required level, scheduled reconnect"
+                );
+            } else {
+                info!(
+                    dc = %dc,
+                    ?family,
+                    alive = now_alive,
+                    required,
+                    endpoint_count = endpoints.len(),
+                    backoff_ms = next_ms,
+                    "DC writer floor is below required level during startup, scheduled reconnect"
+                );
+            }
+        }
+        if let Some(v) = inflight.get_mut(&key) {
+            *v = v.saturating_sub(1);
+        }
+    }
+}
+
+async fn maybe_refresh_idle_writer_for_dc(
+    pool: &Arc<MePool>,
+    rng: &Arc<SecureRandom>,
+    key: (i32, IpFamily),
+    dc: i32,
+    family: IpFamily,
+    endpoints: &[SocketAddr],
+    alive: usize,
+    required: usize,
+    live_writer_ids_by_addr: &HashMap<SocketAddr, Vec<u64>>,
+    writer_idle_since: &HashMap<u64, u64>,
+    idle_refresh_next_attempt: &mut HashMap<(i32, IpFamily), Instant>,
+) {
+    if alive < required {
+        return;
+    }
+
+    let now = Instant::now();
+    if let Some(next) = idle_refresh_next_attempt.get(&key)
+        && now < *next
+    {
+        return;
+    }
+
+    let now_epoch_secs = MePool::now_epoch_secs();
+    let mut candidate: Option<(u64, SocketAddr, u64, u64)> = None;
+    for endpoint in endpoints {
+        let Some(writer_ids) = live_writer_ids_by_addr.get(endpoint) else {
+            continue;
+        };
+        for writer_id in writer_ids {
+            let Some(idle_since_epoch_secs) = writer_idle_since.get(writer_id).copied() else {
+                continue;
+            };
+            let idle_age_secs = now_epoch_secs.saturating_sub(idle_since_epoch_secs);
+            let threshold_secs = IDLE_REFRESH_TRIGGER_BASE_SECS
+                + (*writer_id % (IDLE_REFRESH_TRIGGER_JITTER_SECS + 1));
+            if idle_age_secs < threshold_secs {
+                continue;
+            }
+            if candidate
+                .as_ref()
+                .map(|(_, _, age, _)| idle_age_secs > *age)
+                .unwrap_or(true)
+            {
+                candidate = Some((*writer_id, *endpoint, idle_age_secs, threshold_secs));
+            }
+        }
+    }
+
+    let Some((old_writer_id, endpoint, idle_age_secs, threshold_secs)) = candidate else {
+        return;
+    };
+
+    let rotate_ok = match tokio::time::timeout(pool.me_one_timeout, pool.connect_one(endpoint, rng.as_ref())).await {
+        Ok(Ok(())) => true,
+        Ok(Err(error)) => {
+            debug!(
+                dc = %dc,
+                ?family,
+                %endpoint,
+                old_writer_id,
+                idle_age_secs,
+                threshold_secs,
+                %error,
+                "Idle writer pre-refresh connect failed"
+            );
+            false
+        }
+        Err(_) => {
+            debug!(
+                dc = %dc,
+                ?family,
+                %endpoint,
+                old_writer_id,
+                idle_age_secs,
+                threshold_secs,
+                "Idle writer pre-refresh connect timed out"
+            );
+            false
+        }
+    };
+
+    if !rotate_ok {
+        idle_refresh_next_attempt.insert(key, now + Duration::from_secs(IDLE_REFRESH_RETRY_SECS));
+        return;
+    }
+
+    pool.mark_writer_draining_with_timeout(old_writer_id, pool.force_close_timeout(), false)
+        .await;
+    idle_refresh_next_attempt.insert(
+        key,
+        now + Duration::from_secs(IDLE_REFRESH_SUCCESS_GUARD_SECS),
+    );
+    info!(
+        dc = %dc,
+        ?family,
+        %endpoint,
+        old_writer_id,
+        idle_age_secs,
+        threshold_secs,
+        alive,
+        required,
+        "Idle writer refreshed before upstream idle timeout"
+    );
+}
+
+async fn should_reduce_floor_for_idle(
+    pool: &Arc<MePool>,
+    key: (i32, IpFamily),
+    endpoints: &[SocketAddr],
+    live_writer_ids_by_addr: &HashMap<SocketAddr, Vec<u64>>,
+    adaptive_idle_since: &mut HashMap<(i32, IpFamily), Instant>,
+    adaptive_recover_until: &mut HashMap<(i32, IpFamily), Instant>,
+) -> bool {
+    if endpoints.len() != 1 || pool.floor_mode() != MeFloorMode::Adaptive {
+        adaptive_idle_since.remove(&key);
+        adaptive_recover_until.remove(&key);
+        return false;
+    }
+
+    let now = Instant::now();
+    let endpoint = endpoints[0];
+    let writer_ids = live_writer_ids_by_addr
+        .get(&endpoint)
+        .map(Vec::as_slice)
+        .unwrap_or(&[]);
+    let has_bound_clients = has_bound_clients_on_endpoint(pool, writer_ids).await;
+    if has_bound_clients {
+        adaptive_idle_since.remove(&key);
+        adaptive_recover_until.insert(key, now + pool.adaptive_floor_recover_grace_duration());
+        return false;
+    }
+
+    if let Some(recover_until) = adaptive_recover_until.get(&key)
+        && now < *recover_until
+    {
+        adaptive_idle_since.remove(&key);
+        return false;
+    }
+    adaptive_recover_until.remove(&key);
+
+    let idle_since = adaptive_idle_since.entry(key).or_insert(now);
+    now.saturating_duration_since(*idle_since) >= pool.adaptive_floor_idle_duration()
+}
+
+async fn has_bound_clients_on_endpoint(pool: &Arc<MePool>, writer_ids: &[u64]) -> bool {
+    for writer_id in writer_ids {
+        if !pool.registry.is_writer_empty(*writer_id).await {
+            return true;
+        }
+    }
+    false
+}
+
+async fn recover_single_endpoint_outage(
+    pool: &Arc<MePool>,
+    rng: &Arc<SecureRandom>,
+    key: (i32, IpFamily),
+    endpoint: SocketAddr,
+    required: usize,
+    outage_backoff: &mut HashMap<(i32, IpFamily), u64>,
+    outage_next_attempt: &mut HashMap<(i32, IpFamily), Instant>,
+) {
+    let now = Instant::now();
+    if let Some(ts) = outage_next_attempt.get(&key)
+        && now < *ts
+    {
+        return;
+    }
+
+    let (min_backoff_ms, max_backoff_ms) = pool.single_endpoint_outage_backoff_bounds_ms();
+    pool.stats
+        .increment_me_single_endpoint_outage_reconnect_attempt_total();
+
+    let bypass_quarantine = pool.single_endpoint_outage_disable_quarantine();
+    let attempt_ok = if bypass_quarantine {
+        pool.stats
+            .increment_me_single_endpoint_quarantine_bypass_total();
+        match tokio::time::timeout(pool.me_one_timeout, pool.connect_one(endpoint, rng.as_ref())).await {
+            Ok(Ok(())) => true,
+            Ok(Err(e)) => {
+                debug!(
+                    dc = %key.0,
+                    family = ?key.1,
+                    %endpoint,
+                    error = %e,
+                    "Single-endpoint outage reconnect failed (quarantine bypass path)"
+                );
+                false
+            }
+            Err(_) => {
+                debug!(
+                    dc = %key.0,
+                    family = ?key.1,
+                    %endpoint,
+                    "Single-endpoint outage reconnect timed out (quarantine bypass path)"
+                );
+                false
+            }
+        }
+    } else {
+        let one_endpoint = [endpoint];
+        match tokio::time::timeout(
+            pool.me_one_timeout,
+            pool.connect_endpoints_round_robin(&one_endpoint, rng.as_ref()),
+        )
+        .await
+        {
+            Ok(ok) => ok,
+            Err(_) => {
+                debug!(
+                    dc = %key.0,
+                    family = ?key.1,
+                    %endpoint,
+                    "Single-endpoint outage reconnect timed out"
+                );
+                false
+            }
+        }
+    };
+
+    if attempt_ok {
+        pool.stats
+            .increment_me_single_endpoint_outage_reconnect_success_total();
+        pool.stats.increment_me_reconnect_success();
+        outage_backoff.insert(key, min_backoff_ms);
+        let jitter = min_backoff_ms / JITTER_FRAC_NUM;
+        let wait = Duration::from_millis(min_backoff_ms)
+            + Duration::from_millis(rand::rng().random_range(0..=jitter.max(1)));
+        outage_next_attempt.insert(key, now + wait);
+        info!(
+            dc = %key.0,
+            family = ?key.1,
+            %endpoint,
+            required,
+            backoff_ms = min_backoff_ms,
+            "Single-endpoint outage reconnect succeeded"
+        );
+        return;
+    }
+
+    pool.stats.increment_me_reconnect_attempt();
+    let current_ms = *outage_backoff.get(&key).unwrap_or(&min_backoff_ms);
+    let next_ms = current_ms.saturating_mul(2).min(max_backoff_ms);
+    outage_backoff.insert(key, next_ms);
+    let jitter = next_ms / JITTER_FRAC_NUM;
+    let wait = Duration::from_millis(next_ms)
+        + Duration::from_millis(rand::rng().random_range(0..=jitter.max(1)));
+    outage_next_attempt.insert(key, now + wait);
+    warn!(
+        dc = %key.0,
+        family = ?key.1,
+        %endpoint,
+        required,
+        backoff_ms = next_ms,
+        "Single-endpoint outage reconnect scheduled"
+    );
+}
+
+async fn maybe_rotate_single_endpoint_shadow(
+    pool: &Arc<MePool>,
+    rng: &Arc<SecureRandom>,
+    key: (i32, IpFamily),
+    dc: i32,
+    family: IpFamily,
+    endpoints: &[SocketAddr],
+    alive: usize,
+    required: usize,
+    live_writer_ids_by_addr: &HashMap<SocketAddr, Vec<u64>>,
+    shadow_rotate_deadline: &mut HashMap<(i32, IpFamily), Instant>,
+) {
+    if endpoints.len() != 1 || alive < required {
+        return;
+    }
+
+    let Some(interval) = pool.single_endpoint_shadow_rotate_interval() else {
+        return;
+    };
+
+    let now = Instant::now();
+    if let Some(deadline) = shadow_rotate_deadline.get(&key)
+        && now < *deadline
+    {
+        return;
+    }
+
+    let endpoint = endpoints[0];
+    if pool.is_endpoint_quarantined(endpoint).await {
+        pool.stats
+            .increment_me_single_endpoint_shadow_rotate_skipped_quarantine_total();
+        shadow_rotate_deadline.insert(key, now + Duration::from_secs(SHADOW_ROTATE_RETRY_SECS));
+        debug!(
+            dc = %dc,
+            ?family,
+            %endpoint,
+            "Single-endpoint shadow rotation skipped: endpoint is quarantined"
+        );
+        return;
+    }
+
+    let Some(writer_ids) = live_writer_ids_by_addr.get(&endpoint) else {
+        shadow_rotate_deadline.insert(key, now + Duration::from_secs(SHADOW_ROTATE_RETRY_SECS));
+        return;
+    };
+
+    let mut candidate_writer_id = None;
+    for writer_id in writer_ids {
+        if pool.registry.is_writer_empty(*writer_id).await {
+            candidate_writer_id = Some(*writer_id);
+            break;
+        }
+    }
+
+    let Some(old_writer_id) = candidate_writer_id else {
+        shadow_rotate_deadline.insert(key, now + Duration::from_secs(SHADOW_ROTATE_RETRY_SECS));
+        debug!(
+            dc = %dc,
+            ?family,
+            %endpoint,
+            alive,
+            required,
+            "Single-endpoint shadow rotation skipped: no empty writer candidate"
+        );
+        return;
+    };
+
+    let rotate_ok = match tokio::time::timeout(pool.me_one_timeout, pool.connect_one(endpoint, rng.as_ref())).await {
+        Ok(Ok(())) => true,
+        Ok(Err(e)) => {
+            debug!(
+                dc = %dc,
+                ?family,
+                %endpoint,
+                error = %e,
+                "Single-endpoint shadow rotation connect failed"
+            );
+            false
+        }
+        Err(_) => {
+            debug!(
+                dc = %dc,
+                ?family,
+                %endpoint,
+                "Single-endpoint shadow rotation connect timed out"
+            );
+            false
+        }
+    };
+
+    if !rotate_ok {
+        shadow_rotate_deadline.insert(
+            key,
+            now + interval.min(Duration::from_secs(SHADOW_ROTATE_RETRY_SECS)),
+        );
+        return;
+    }
+
+    pool.mark_writer_draining_with_timeout(old_writer_id, pool.force_close_timeout(), false)
+        .await;
+    pool.stats.increment_me_single_endpoint_shadow_rotate_total();
+    shadow_rotate_deadline.insert(key, now + interval);
+    info!(
+        dc = %dc,
+        ?family,
+        %endpoint,
+        old_writer_id,
+        rotate_every_secs = interval.as_secs(),
+        "Single-endpoint shadow writer rotated"
+    );
+}

src/transport/middle_proxy/mod.rs

@@ -0,0 +1,46 @@
+//! Middle Proxy RPC transport.
+
+mod codec;
+mod config_updater;
+mod handshake;
+mod health;
+mod pool;
+mod pool_config;
+mod pool_init;
+mod pool_nat;
+mod pool_refill;
+mod pool_reinit;
+mod pool_writer;
+mod ping;
+mod reader;
+mod registry;
+mod rotation;
+mod send;
+mod secret;
+mod wire;
+mod pool_status;
+
+use bytes::Bytes;
+
+pub use health::me_health_monitor;
+#[allow(unused_imports)]
+pub use ping::{run_me_ping, format_sample_line, format_me_route, MePingReport, MePingSample, MePingFamily};
+pub use pool::MePool;
+#[allow(unused_imports)]
+pub use pool_nat::{stun_probe, detect_public_ip};
+pub use registry::ConnRegistry;
+pub use secret::fetch_proxy_secret;
+#[allow(unused_imports)]
+pub use config_updater::{
+    ProxyConfigData, fetch_proxy_config, fetch_proxy_config_with_raw, load_proxy_config_cache,
+    me_config_updater, save_proxy_config_cache,
+};
+pub use rotation::{MeReinitTrigger, me_reinit_scheduler, me_rotation_task};
+pub use wire::proto_flags_for_tag;
+
+#[derive(Debug)]
+pub enum MeResponse {
+    Data { flags: u32, data: Bytes },
+    Ack(u32),
+    Close,
+}

src/transport/middle_proxy/ping.rs

@@ -0,0 +1,386 @@
+use std::collections::HashMap;
+use std::net::{IpAddr, SocketAddr};
+use std::sync::Arc;
+
+use tokio::net::UdpSocket;
+
+use crate::config::{UpstreamConfig, UpstreamType};
+use crate::crypto::SecureRandom;
+use crate::error::ProxyError;
+use crate::transport::{UpstreamEgressInfo, UpstreamRouteKind};
+
+use super::MePool;
+
+#[derive(Debug, Clone, Copy, PartialEq, Eq)]
+pub enum MePingFamily {
+    V4,
+    V6,
+}
+
+#[derive(Debug, Clone)]
+pub struct MePingSample {
+    pub dc: i32,
+    pub addr: SocketAddr,
+    pub route: Option<String>,
+    pub connect_ms: Option<f64>,
+    pub handshake_ms: Option<f64>,
+    pub error: Option<String>,
+    pub family: MePingFamily,
+}
+
+#[derive(Debug, Clone)]
+#[allow(dead_code)]
+pub struct MePingReport {
+    pub dc: i32,
+    pub family: MePingFamily,
+    pub samples: Vec<MePingSample>,
+}
+
+pub fn format_sample_line(sample: &MePingSample) -> String {
+    let sign = if sample.dc >= 0 { "+" } else { "-" };
+    let addr = format!("{}:{}", sample.addr.ip(), sample.addr.port());
+
+    match (sample.connect_ms, sample.handshake_ms.as_ref(), sample.error.as_ref()) {
+        (Some(conn), Some(hs), None) => format!(
+            "     {sign} {addr}\tPing: {:.0} ms / RPC: {:.0} ms / OK",
+            conn, hs
+        ),
+        (Some(conn), None, Some(err)) => format!(
+            "     {sign} {addr}\tPing: {:.0} ms / RPC: FAIL ({err})",
+            conn
+        ),
+        (None, _, Some(err)) => format!("     {sign} {addr}\tPing: FAIL ({err})"),
+        (Some(conn), None, None) => format!("     {sign} {addr}\tPing: {:.0} ms / RPC: FAIL", conn),
+        _ => format!("     {sign} {addr}\tPing: FAIL"),
+    }
+}
+
+fn format_direct_with_config(
+    interface: &Option<String>,
+    bind_addresses: &Option<Vec<String>>,
+) -> Option<String> {
+    let mut direct_parts: Vec<String> = Vec::new();
+    if let Some(dev) = interface.as_deref().filter(|v| !v.is_empty()) {
+        direct_parts.push(format!("dev={dev}"));
+    }
+    if let Some(src) = bind_addresses.as_ref().filter(|v| !v.is_empty()) {
+        direct_parts.push(format!("src={}", src.join(",")));
+    }
+    if direct_parts.is_empty() {
+        None
+    } else {
+        Some(format!("direct {}", direct_parts.join(" ")))
+    }
+}
+
+fn pick_target_for_family(reports: &[MePingReport], family: MePingFamily) -> Option<SocketAddr> {
+    reports.iter().find_map(|report| {
+        if report.family != family {
+            return None;
+        }
+        report
+            .samples
+            .iter()
+            .find(|s| s.error.is_none() && s.handshake_ms.is_some())
+            .map(|s| s.addr)
+    })
+}
+
+fn route_from_egress(egress: Option<UpstreamEgressInfo>) -> Option<String> {
+    let info = egress?;
+    match info.route_kind {
+        UpstreamRouteKind::Direct => {
+            let src_ip = info
+                .direct_bind_ip
+                .or_else(|| info.local_addr.map(|addr| addr.ip()));
+            let ip = src_ip?;
+            let mut parts = Vec::new();
+            if let Some(dev) = detect_interface_for_ip(ip) {
+                parts.push(format!("dev={dev}"));
+            }
+            parts.push(format!("src={ip}"));
+            Some(format!("direct {}", parts.join(" ")))
+        }
+        UpstreamRouteKind::Socks4 => {
+            let route = info
+                .socks_proxy_addr
+                .map(|addr| format!("socks4://{addr}"))
+                .unwrap_or_else(|| "socks4://unknown".to_string());
+            Some(match info.socks_bound_addr {
+                Some(bound) => format!("{route} bnd={bound}"),
+                None => route,
+            })
+        }
+        UpstreamRouteKind::Socks5 => {
+            let route = info
+                .socks_proxy_addr
+                .map(|addr| format!("socks5://{addr}"))
+                .unwrap_or_else(|| "socks5://unknown".to_string());
+            Some(match info.socks_bound_addr {
+                Some(bound) => format!("{route} bnd={bound}"),
+                None => route,
+            })
+        }
+    }
+}
+
+#[cfg(unix)]
+fn detect_interface_for_ip(ip: IpAddr) -> Option<String> {
+    use nix::ifaddrs::getifaddrs;
+
+    if let Ok(addrs) = getifaddrs() {
+        for iface in addrs {
+            if let Some(address) = iface.address {
+                if let Some(v4) = address.as_sockaddr_in() {
+                    if IpAddr::V4(v4.ip()) == ip {
+                        return Some(iface.interface_name);
+                    }
+                } else if let Some(v6) = address.as_sockaddr_in6() {
+                    if IpAddr::V6(v6.ip()) == ip {
+                        return Some(iface.interface_name);
+                    }
+                }
+            }
+        }
+    }
+    None
+}
+
+#[cfg(not(unix))]
+fn detect_interface_for_ip(_ip: IpAddr) -> Option<String> {
+    None
+}
+
+async fn detect_direct_route_details(
+    reports: &[MePingReport],
+    prefer_ipv6: bool,
+    v4_ok: bool,
+    v6_ok: bool,
+) -> Option<String> {
+    let target_addr = if prefer_ipv6 && v6_ok {
+        pick_target_for_family(reports, MePingFamily::V6)
+            .or_else(|| pick_target_for_family(reports, MePingFamily::V4))
+    } else if v4_ok {
+        pick_target_for_family(reports, MePingFamily::V4)
+            .or_else(|| pick_target_for_family(reports, MePingFamily::V6))
+    } else {
+        pick_target_for_family(reports, MePingFamily::V6)
+            .or_else(|| pick_target_for_family(reports, MePingFamily::V4))
+    }?;
+
+    let local_ip = if target_addr.is_ipv4() {
+        let sock = UdpSocket::bind("0.0.0.0:0").await.ok()?;
+        sock.connect(target_addr).await.ok()?;
+        sock.local_addr().ok().map(|a| a.ip())
+    } else {
+        let sock = UdpSocket::bind("[::]:0").await.ok()?;
+        sock.connect(target_addr).await.ok()?;
+        sock.local_addr().ok().map(|a| a.ip())
+    };
+
+    let mut parts = Vec::new();
+    if let Some(ip) = local_ip {
+        if let Some(dev) = detect_interface_for_ip(ip) {
+            parts.push(format!("dev={dev}"));
+        }
+        parts.push(format!("src={ip}"));
+    }
+
+    if parts.is_empty() {
+        None
+    } else {
+        Some(format!("direct {}", parts.join(" ")))
+    }
+}
+
+pub async fn format_me_route(
+    upstreams: &[UpstreamConfig],
+    reports: &[MePingReport],
+    prefer_ipv6: bool,
+    v4_ok: bool,
+    v6_ok: bool,
+) -> String {
+    if let Some(route) = reports
+        .iter()
+        .flat_map(|report| report.samples.iter())
+        .find(|sample| sample.error.is_none() && sample.handshake_ms.is_some())
+        .and_then(|sample| sample.route.clone())
+    {
+        return route;
+    }
+
+    let enabled_upstreams: Vec<_> = upstreams.iter().filter(|u| u.enabled).collect();
+    if enabled_upstreams.is_empty() {
+        return detect_direct_route_details(reports, prefer_ipv6, v4_ok, v6_ok)
+            .await
+            .unwrap_or_else(|| "direct".to_string());
+    }
+
+    if enabled_upstreams.len() == 1 {
+        return match &enabled_upstreams[0].upstream_type {
+            UpstreamType::Direct {
+                interface,
+                bind_addresses,
+            } => {
+                if let Some(route) = format_direct_with_config(interface, bind_addresses) {
+                    route
+                } else {
+                    detect_direct_route_details(reports, prefer_ipv6, v4_ok, v6_ok)
+                        .await
+                        .unwrap_or_else(|| "direct".to_string())
+                }
+            }
+            UpstreamType::Socks4 { address, .. } => format!("socks4://{address}"),
+            UpstreamType::Socks5 { address, .. } => format!("socks5://{address}"),
+        };
+    }
+
+    let has_direct = enabled_upstreams
+        .iter()
+        .any(|u| matches!(u.upstream_type, UpstreamType::Direct { .. }));
+    let has_socks4 = enabled_upstreams
+        .iter()
+        .any(|u| matches!(u.upstream_type, UpstreamType::Socks4 { .. }));
+    let has_socks5 = enabled_upstreams
+        .iter()
+        .any(|u| matches!(u.upstream_type, UpstreamType::Socks5 { .. }));
+    let mut kinds = Vec::new();
+    if has_direct {
+        kinds.push("direct");
+    }
+    if has_socks4 {
+        kinds.push("socks4");
+    }
+    if has_socks5 {
+        kinds.push("socks5");
+    }
+    format!("mixed upstreams ({})", kinds.join(", "))
+}
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+    use std::net::{IpAddr, Ipv4Addr, SocketAddr};
+
+    fn sample(base: MePingSample) -> MePingSample {
+        base
+    }
+
+    #[test]
+    fn ok_line_contains_both_timings() {
+        let s = sample(MePingSample {
+            dc: 4,
+            addr: SocketAddr::new(IpAddr::V4(Ipv4Addr::new(1, 2, 3, 4)), 8888),
+            route: Some("direct src=1.2.3.4".to_string()),
+            connect_ms: Some(12.3),
+            handshake_ms: Some(34.7),
+            error: None,
+            family: MePingFamily::V4,
+        });
+        let line = format_sample_line(&s);
+        assert!(line.contains("Ping: 12 ms"));
+        assert!(line.contains("RPC: 35 ms"));
+        assert!(line.contains("OK"));
+    }
+
+    #[test]
+    fn error_line_mentions_reason() {
+        let s = sample(MePingSample {
+            dc: -5,
+            addr: SocketAddr::new(IpAddr::V4(Ipv4Addr::new(5, 6, 7, 8)), 80),
+            route: Some("socks5".to_string()),
+            connect_ms: Some(10.0),
+            handshake_ms: None,
+            error: Some("handshake timeout".to_string()),
+            family: MePingFamily::V4,
+        });
+        let line = format_sample_line(&s);
+        assert!(line.contains("- 5.6.7.8:80"));
+        assert!(line.contains("handshake timeout"));
+    }
+}
+
+pub async fn run_me_ping(pool: &Arc<MePool>, rng: &SecureRandom) -> Vec<MePingReport> {
+    let mut reports = Vec::new();
+
+    let v4_map = if pool.decision.ipv4_me {
+        pool.proxy_map_v4.read().await.clone()
+    } else {
+        HashMap::new()
+    };
+    let v6_map = if pool.decision.ipv6_me {
+        pool.proxy_map_v6.read().await.clone()
+    } else {
+        HashMap::new()
+    };
+
+    let mut grouped: Vec<(MePingFamily, i32, Vec<(IpAddr, u16)>)> = Vec::new();
+    for (dc, addrs) in v4_map {
+        grouped.push((MePingFamily::V4, dc, addrs));
+    }
+    for (dc, addrs) in v6_map {
+        grouped.push((MePingFamily::V6, dc, addrs));
+    }
+
+    for (family, dc, addrs) in grouped {
+        let mut samples = Vec::new();
+        for (ip, port) in addrs {
+            let addr = SocketAddr::new(ip, port);
+            let mut connect_ms = None;
+            let mut handshake_ms = None;
+            let mut error = None;
+            let mut route = None;
+
+            match pool.connect_tcp(addr).await {
+                Ok((stream, conn_rtt, upstream_egress)) => {
+                    connect_ms = Some(conn_rtt);
+                    route = route_from_egress(upstream_egress);
+                    match pool.handshake_only(stream, addr, upstream_egress, rng).await {
+                        Ok(hs) => {
+                            handshake_ms = Some(hs.handshake_ms);
+                            // drop halves to close
+                            drop(hs.rd);
+                            drop(hs.wr);
+                        }
+                        Err(e) => {
+                            error = Some(short_err(&e));
+                        }
+                    }
+                }
+                Err(e) => {
+                    error = Some(short_err(&e));
+                }
+            }
+
+            samples.push(MePingSample {
+                dc,
+                addr,
+                route,
+                connect_ms,
+                handshake_ms,
+                error,
+                family,
+            });
+        }
+
+        reports.push(MePingReport {
+            dc,
+            family,
+            samples,
+        });
+    }
+
+    reports
+}
+
+fn short_err(err: &ProxyError) -> String {
+    match err {
+        ProxyError::ConnectionTimeout { .. } => "connect timeout".to_string(),
+        ProxyError::TgHandshakeTimeout => "handshake timeout".to_string(),
+        ProxyError::InvalidHandshake(e) => format!("bad handshake: {e}"),
+        ProxyError::Crypto(e) => format!("crypto: {e}"),
+        ProxyError::Proxy(e) => format!("proxy: {e}"),
+        ProxyError::Io(e) => format!("io: {e}"),
+        _ => format!("{err}"),
+    }
+}

src/transport/middle_proxy/pool.rs

@@ -0,0 +1,633 @@
+use std::collections::{HashMap, HashSet};
+use std::net::{IpAddr, Ipv6Addr, SocketAddr};
+use std::sync::Arc;
+use std::sync::atomic::{AtomicBool, AtomicI32, AtomicU8, AtomicU32, AtomicU64, AtomicUsize, Ordering};
+use std::time::{Duration, Instant, SystemTime, UNIX_EPOCH};
+
+use tokio::sync::{Mutex, Notify, RwLock, mpsc};
+use tokio_util::sync::CancellationToken;
+
+use crate::config::{MeBindStaleMode, MeFloorMode, MeRouteNoWriterMode, MeSocksKdfPolicy};
+use crate::crypto::SecureRandom;
+use crate::network::IpFamily;
+use crate::network::probe::NetworkDecision;
+use crate::transport::UpstreamManager;
+
+use super::ConnRegistry;
+use super::codec::WriterCommand;
+
+#[derive(Debug, Clone, Copy, PartialEq, Eq, Hash)]
+pub(super) struct RefillDcKey {
+    pub dc: i32,
+    pub family: IpFamily,
+}
+
+#[derive(Clone)]
+pub struct MeWriter {
+    pub id: u64,
+    pub addr: SocketAddr,
+    pub generation: u64,
+    pub contour: Arc<AtomicU8>,
+    pub created_at: Instant,
+    pub tx: mpsc::Sender<WriterCommand>,
+    pub cancel: CancellationToken,
+    pub degraded: Arc<AtomicBool>,
+    pub draining: Arc<AtomicBool>,
+    pub draining_started_at_epoch_secs: Arc<AtomicU64>,
+    pub allow_drain_fallback: Arc<AtomicBool>,
+}
+
+#[derive(Debug, Clone, Copy, PartialEq, Eq)]
+#[repr(u8)]
+pub(super) enum WriterContour {
+    Warm = 0,
+    Active = 1,
+    Draining = 2,
+}
+
+impl WriterContour {
+    pub(super) fn as_u8(self) -> u8 {
+        self as u8
+    }
+
+    pub(super) fn from_u8(value: u8) -> Self {
+        match value {
+            0 => Self::Warm,
+            1 => Self::Active,
+            2 => Self::Draining,
+            _ => Self::Draining,
+        }
+    }
+}
+
+#[derive(Debug, Clone)]
+pub struct SecretSnapshot {
+    pub epoch: u64,
+    pub key_selector: u32,
+    pub secret: Vec<u8>,
+}
+
+#[allow(dead_code)]
+pub struct MePool {
+    pub(super) registry: Arc<ConnRegistry>,
+    pub(super) writers: Arc<RwLock<Vec<MeWriter>>>,
+    pub(super) rr: AtomicU64,
+    pub(super) decision: NetworkDecision,
+    pub(super) upstream: Option<Arc<UpstreamManager>>,
+    pub(super) rng: Arc<SecureRandom>,
+    pub(super) proxy_tag: Option<Vec<u8>>,
+    pub(super) proxy_secret: Arc<RwLock<SecretSnapshot>>,
+    pub(super) nat_ip_cfg: Option<IpAddr>,
+    pub(super) nat_ip_detected: Arc<RwLock<Option<IpAddr>>>,
+    pub(super) nat_probe: bool,
+    pub(super) nat_stun: Option<String>,
+    pub(super) nat_stun_servers: Vec<String>,
+    pub(super) nat_stun_live_servers: Arc<RwLock<Vec<String>>>,
+    pub(super) nat_probe_concurrency: usize,
+    pub(super) detected_ipv6: Option<Ipv6Addr>,
+    pub(super) nat_probe_attempts: std::sync::atomic::AtomicU8,
+    pub(super) nat_probe_disabled: std::sync::atomic::AtomicBool,
+    pub(super) stun_backoff_until: Arc<RwLock<Option<Instant>>>,
+    pub(super) me_one_retry: u8,
+    pub(super) me_one_timeout: Duration,
+    pub(super) me_keepalive_enabled: bool,
+    pub(super) me_keepalive_interval: Duration,
+    pub(super) me_keepalive_jitter: Duration,
+    pub(super) me_keepalive_payload_random: bool,
+    pub(super) rpc_proxy_req_every_secs: AtomicU64,
+    pub(super) me_warmup_stagger_enabled: bool,
+    pub(super) me_warmup_step_delay: Duration,
+    pub(super) me_warmup_step_jitter: Duration,
+    pub(super) me_reconnect_max_concurrent_per_dc: u32,
+    pub(super) me_reconnect_backoff_base: Duration,
+    pub(super) me_reconnect_backoff_cap: Duration,
+    pub(super) me_reconnect_fast_retry_count: u32,
+    pub(super) me_single_endpoint_shadow_writers: AtomicU8,
+    pub(super) me_single_endpoint_outage_mode_enabled: AtomicBool,
+    pub(super) me_single_endpoint_outage_disable_quarantine: AtomicBool,
+    pub(super) me_single_endpoint_outage_backoff_min_ms: AtomicU64,
+    pub(super) me_single_endpoint_outage_backoff_max_ms: AtomicU64,
+    pub(super) me_single_endpoint_shadow_rotate_every_secs: AtomicU64,
+    pub(super) me_floor_mode: AtomicU8,
+    pub(super) me_adaptive_floor_idle_secs: AtomicU64,
+    pub(super) me_adaptive_floor_min_writers_single_endpoint: AtomicU8,
+    pub(super) me_adaptive_floor_recover_grace_secs: AtomicU64,
+    pub(super) proxy_map_v4: Arc<RwLock<HashMap<i32, Vec<(IpAddr, u16)>>>>,
+    pub(super) proxy_map_v6: Arc<RwLock<HashMap<i32, Vec<(IpAddr, u16)>>>>,
+    pub(super) default_dc: AtomicI32,
+    pub(super) next_writer_id: AtomicU64,
+    pub(super) ping_tracker: Arc<Mutex<HashMap<i64, (std::time::Instant, u64)>>>,
+    pub(super) rtt_stats: Arc<Mutex<HashMap<u64, (f64, f64)>>>,
+    pub(super) nat_reflection_cache: Arc<Mutex<NatReflectionCache>>,
+    pub(super) writer_available: Arc<Notify>,
+    pub(super) refill_inflight: Arc<Mutex<HashSet<SocketAddr>>>,
+    pub(super) refill_inflight_dc: Arc<Mutex<HashSet<RefillDcKey>>>,
+    pub(super) conn_count: AtomicUsize,
+    pub(super) stats: Arc<crate::stats::Stats>,
+    pub(super) generation: AtomicU64,
+    pub(super) active_generation: AtomicU64,
+    pub(super) warm_generation: AtomicU64,
+    pub(super) pending_hardswap_generation: AtomicU64,
+    pub(super) pending_hardswap_started_at_epoch_secs: AtomicU64,
+    pub(super) pending_hardswap_map_hash: AtomicU64,
+    pub(super) hardswap: AtomicBool,
+    pub(super) endpoint_quarantine: Arc<Mutex<HashMap<SocketAddr, Instant>>>,
+    pub(super) kdf_material_fingerprint: Arc<RwLock<HashMap<SocketAddr, (u64, u16)>>>,
+    pub(super) me_pool_drain_ttl_secs: AtomicU64,
+    pub(super) me_pool_force_close_secs: AtomicU64,
+    pub(super) me_pool_min_fresh_ratio_permille: AtomicU32,
+    pub(super) me_hardswap_warmup_delay_min_ms: AtomicU64,
+    pub(super) me_hardswap_warmup_delay_max_ms: AtomicU64,
+    pub(super) me_hardswap_warmup_extra_passes: AtomicU32,
+    pub(super) me_hardswap_warmup_pass_backoff_base_ms: AtomicU64,
+    pub(super) me_bind_stale_mode: AtomicU8,
+    pub(super) me_bind_stale_ttl_secs: AtomicU64,
+    pub(super) secret_atomic_snapshot: AtomicBool,
+    pub(super) me_deterministic_writer_sort: AtomicBool,
+    pub(super) me_socks_kdf_policy: AtomicU8,
+    pub(super) me_route_no_writer_mode: AtomicU8,
+    pub(super) me_route_no_writer_wait: Duration,
+    pub(super) me_route_inline_recovery_attempts: u32,
+    pub(super) me_route_inline_recovery_wait: Duration,
+    pub(super) runtime_ready: AtomicBool,
+    pool_size: usize,
+}
+
+#[derive(Debug, Default)]
+pub struct NatReflectionCache {
+    pub v4: Option<(std::time::Instant, std::net::SocketAddr)>,
+    pub v6: Option<(std::time::Instant, std::net::SocketAddr)>,
+}
+
+impl MePool {
+    fn ratio_to_permille(ratio: f32) -> u32 {
+        let clamped = ratio.clamp(0.0, 1.0);
+        (clamped * 1000.0).round() as u32
+    }
+
+    pub(super) fn permille_to_ratio(permille: u32) -> f32 {
+        (permille.min(1000) as f32) / 1000.0
+    }
+
+    pub(super) fn now_epoch_secs() -> u64 {
+        SystemTime::now()
+            .duration_since(UNIX_EPOCH)
+            .unwrap_or_default()
+            .as_secs()
+    }
+
+    pub fn new(
+        proxy_tag: Option<Vec<u8>>,
+        proxy_secret: Vec<u8>,
+        nat_ip: Option<IpAddr>,
+        nat_probe: bool,
+        nat_stun: Option<String>,
+        nat_stun_servers: Vec<String>,
+        nat_probe_concurrency: usize,
+        detected_ipv6: Option<Ipv6Addr>,
+        me_one_retry: u8,
+        me_one_timeout_ms: u64,
+        proxy_map_v4: HashMap<i32, Vec<(IpAddr, u16)>>,
+        proxy_map_v6: HashMap<i32, Vec<(IpAddr, u16)>>,
+        default_dc: Option<i32>,
+        decision: NetworkDecision,
+        upstream: Option<Arc<UpstreamManager>>,
+        rng: Arc<SecureRandom>,
+        stats: Arc<crate::stats::Stats>,
+        me_keepalive_enabled: bool,
+        me_keepalive_interval_secs: u64,
+        me_keepalive_jitter_secs: u64,
+        me_keepalive_payload_random: bool,
+        rpc_proxy_req_every_secs: u64,
+        me_warmup_stagger_enabled: bool,
+        me_warmup_step_delay_ms: u64,
+        me_warmup_step_jitter_ms: u64,
+        me_reconnect_max_concurrent_per_dc: u32,
+        me_reconnect_backoff_base_ms: u64,
+        me_reconnect_backoff_cap_ms: u64,
+        me_reconnect_fast_retry_count: u32,
+        me_single_endpoint_shadow_writers: u8,
+        me_single_endpoint_outage_mode_enabled: bool,
+        me_single_endpoint_outage_disable_quarantine: bool,
+        me_single_endpoint_outage_backoff_min_ms: u64,
+        me_single_endpoint_outage_backoff_max_ms: u64,
+        me_single_endpoint_shadow_rotate_every_secs: u64,
+        me_floor_mode: MeFloorMode,
+        me_adaptive_floor_idle_secs: u64,
+        me_adaptive_floor_min_writers_single_endpoint: u8,
+        me_adaptive_floor_recover_grace_secs: u64,
+        hardswap: bool,
+        me_pool_drain_ttl_secs: u64,
+        me_pool_force_close_secs: u64,
+        me_pool_min_fresh_ratio: f32,
+        me_hardswap_warmup_delay_min_ms: u64,
+        me_hardswap_warmup_delay_max_ms: u64,
+        me_hardswap_warmup_extra_passes: u8,
+        me_hardswap_warmup_pass_backoff_base_ms: u64,
+        me_bind_stale_mode: MeBindStaleMode,
+        me_bind_stale_ttl_secs: u64,
+        me_secret_atomic_snapshot: bool,
+        me_deterministic_writer_sort: bool,
+        me_socks_kdf_policy: MeSocksKdfPolicy,
+        me_route_backpressure_base_timeout_ms: u64,
+        me_route_backpressure_high_timeout_ms: u64,
+        me_route_backpressure_high_watermark_pct: u8,
+        me_route_no_writer_mode: MeRouteNoWriterMode,
+        me_route_no_writer_wait_ms: u64,
+        me_route_inline_recovery_attempts: u32,
+        me_route_inline_recovery_wait_ms: u64,
+    ) -> Arc<Self> {
+        let registry = Arc::new(ConnRegistry::new());
+        registry.update_route_backpressure_policy(
+            me_route_backpressure_base_timeout_ms,
+            me_route_backpressure_high_timeout_ms,
+            me_route_backpressure_high_watermark_pct,
+        );
+        Arc::new(Self {
+            registry,
+            writers: Arc::new(RwLock::new(Vec::new())),
+            rr: AtomicU64::new(0),
+            decision,
+            upstream,
+            rng,
+            proxy_tag,
+            proxy_secret: Arc::new(RwLock::new(SecretSnapshot {
+                epoch: 1,
+                key_selector: if proxy_secret.len() >= 4 {
+                    u32::from_le_bytes([
+                        proxy_secret[0],
+                        proxy_secret[1],
+                        proxy_secret[2],
+                        proxy_secret[3],
+                    ])
+                } else {
+                    0
+                },
+                secret: proxy_secret,
+            })),
+            nat_ip_cfg: nat_ip,
+            nat_ip_detected: Arc::new(RwLock::new(None)),
+            nat_probe,
+            nat_stun,
+            nat_stun_servers,
+            nat_stun_live_servers: Arc::new(RwLock::new(Vec::new())),
+            nat_probe_concurrency: nat_probe_concurrency.max(1),
+            detected_ipv6,
+            nat_probe_attempts: std::sync::atomic::AtomicU8::new(0),
+            nat_probe_disabled: std::sync::atomic::AtomicBool::new(false),
+            stun_backoff_until: Arc::new(RwLock::new(None)),
+            me_one_retry,
+            me_one_timeout: Duration::from_millis(me_one_timeout_ms),
+            stats,
+            me_keepalive_enabled,
+            me_keepalive_interval: Duration::from_secs(me_keepalive_interval_secs),
+            me_keepalive_jitter: Duration::from_secs(me_keepalive_jitter_secs),
+            me_keepalive_payload_random,
+            rpc_proxy_req_every_secs: AtomicU64::new(rpc_proxy_req_every_secs),
+            me_warmup_stagger_enabled,
+            me_warmup_step_delay: Duration::from_millis(me_warmup_step_delay_ms),
+            me_warmup_step_jitter: Duration::from_millis(me_warmup_step_jitter_ms),
+            me_reconnect_max_concurrent_per_dc,
+            me_reconnect_backoff_base: Duration::from_millis(me_reconnect_backoff_base_ms),
+            me_reconnect_backoff_cap: Duration::from_millis(me_reconnect_backoff_cap_ms),
+            me_reconnect_fast_retry_count,
+            me_single_endpoint_shadow_writers: AtomicU8::new(me_single_endpoint_shadow_writers),
+            me_single_endpoint_outage_mode_enabled: AtomicBool::new(
+                me_single_endpoint_outage_mode_enabled,
+            ),
+            me_single_endpoint_outage_disable_quarantine: AtomicBool::new(
+                me_single_endpoint_outage_disable_quarantine,
+            ),
+            me_single_endpoint_outage_backoff_min_ms: AtomicU64::new(
+                me_single_endpoint_outage_backoff_min_ms,
+            ),
+            me_single_endpoint_outage_backoff_max_ms: AtomicU64::new(
+                me_single_endpoint_outage_backoff_max_ms,
+            ),
+            me_single_endpoint_shadow_rotate_every_secs: AtomicU64::new(
+                me_single_endpoint_shadow_rotate_every_secs,
+            ),
+            me_floor_mode: AtomicU8::new(me_floor_mode.as_u8()),
+            me_adaptive_floor_idle_secs: AtomicU64::new(me_adaptive_floor_idle_secs),
+            me_adaptive_floor_min_writers_single_endpoint: AtomicU8::new(
+                me_adaptive_floor_min_writers_single_endpoint,
+            ),
+            me_adaptive_floor_recover_grace_secs: AtomicU64::new(
+                me_adaptive_floor_recover_grace_secs,
+            ),
+            pool_size: 2,
+            proxy_map_v4: Arc::new(RwLock::new(proxy_map_v4)),
+            proxy_map_v6: Arc::new(RwLock::new(proxy_map_v6)),
+            default_dc: AtomicI32::new(default_dc.unwrap_or(0)),
+            next_writer_id: AtomicU64::new(1),
+            ping_tracker: Arc::new(Mutex::new(HashMap::new())),
+            rtt_stats: Arc::new(Mutex::new(HashMap::new())),
+            nat_reflection_cache: Arc::new(Mutex::new(NatReflectionCache::default())),
+            writer_available: Arc::new(Notify::new()),
+            refill_inflight: Arc::new(Mutex::new(HashSet::new())),
+            refill_inflight_dc: Arc::new(Mutex::new(HashSet::new())),
+            conn_count: AtomicUsize::new(0),
+            generation: AtomicU64::new(1),
+            active_generation: AtomicU64::new(1),
+            warm_generation: AtomicU64::new(0),
+            pending_hardswap_generation: AtomicU64::new(0),
+            pending_hardswap_started_at_epoch_secs: AtomicU64::new(0),
+            pending_hardswap_map_hash: AtomicU64::new(0),
+            hardswap: AtomicBool::new(hardswap),
+            endpoint_quarantine: Arc::new(Mutex::new(HashMap::new())),
+            kdf_material_fingerprint: Arc::new(RwLock::new(HashMap::new())),
+            me_pool_drain_ttl_secs: AtomicU64::new(me_pool_drain_ttl_secs),
+            me_pool_force_close_secs: AtomicU64::new(me_pool_force_close_secs),
+            me_pool_min_fresh_ratio_permille: AtomicU32::new(Self::ratio_to_permille(
+                me_pool_min_fresh_ratio,
+            )),
+            me_hardswap_warmup_delay_min_ms: AtomicU64::new(me_hardswap_warmup_delay_min_ms),
+            me_hardswap_warmup_delay_max_ms: AtomicU64::new(me_hardswap_warmup_delay_max_ms),
+            me_hardswap_warmup_extra_passes: AtomicU32::new(me_hardswap_warmup_extra_passes as u32),
+            me_hardswap_warmup_pass_backoff_base_ms: AtomicU64::new(
+                me_hardswap_warmup_pass_backoff_base_ms,
+            ),
+            me_bind_stale_mode: AtomicU8::new(me_bind_stale_mode.as_u8()),
+            me_bind_stale_ttl_secs: AtomicU64::new(me_bind_stale_ttl_secs),
+            secret_atomic_snapshot: AtomicBool::new(me_secret_atomic_snapshot),
+            me_deterministic_writer_sort: AtomicBool::new(me_deterministic_writer_sort),
+            me_socks_kdf_policy: AtomicU8::new(me_socks_kdf_policy.as_u8()),
+            me_route_no_writer_mode: AtomicU8::new(me_route_no_writer_mode.as_u8()),
+            me_route_no_writer_wait: Duration::from_millis(me_route_no_writer_wait_ms),
+            me_route_inline_recovery_attempts,
+            me_route_inline_recovery_wait: Duration::from_millis(me_route_inline_recovery_wait_ms),
+            runtime_ready: AtomicBool::new(false),
+        })
+    }
+
+    pub fn current_generation(&self) -> u64 {
+        self.active_generation.load(Ordering::Relaxed)
+    }
+
+    pub fn set_runtime_ready(&self, ready: bool) {
+        self.runtime_ready.store(ready, Ordering::Relaxed);
+    }
+
+    pub fn is_runtime_ready(&self) -> bool {
+        self.runtime_ready.load(Ordering::Relaxed)
+    }
+
+    pub fn update_runtime_reinit_policy(
+        &self,
+        hardswap: bool,
+        drain_ttl_secs: u64,
+        force_close_secs: u64,
+        min_fresh_ratio: f32,
+        hardswap_warmup_delay_min_ms: u64,
+        hardswap_warmup_delay_max_ms: u64,
+        hardswap_warmup_extra_passes: u8,
+        hardswap_warmup_pass_backoff_base_ms: u64,
+        bind_stale_mode: MeBindStaleMode,
+        bind_stale_ttl_secs: u64,
+        secret_atomic_snapshot: bool,
+        deterministic_writer_sort: bool,
+        single_endpoint_shadow_writers: u8,
+        single_endpoint_outage_mode_enabled: bool,
+        single_endpoint_outage_disable_quarantine: bool,
+        single_endpoint_outage_backoff_min_ms: u64,
+        single_endpoint_outage_backoff_max_ms: u64,
+        single_endpoint_shadow_rotate_every_secs: u64,
+        floor_mode: MeFloorMode,
+        adaptive_floor_idle_secs: u64,
+        adaptive_floor_min_writers_single_endpoint: u8,
+        adaptive_floor_recover_grace_secs: u64,
+    ) {
+        self.hardswap.store(hardswap, Ordering::Relaxed);
+        self.me_pool_drain_ttl_secs
+            .store(drain_ttl_secs, Ordering::Relaxed);
+        self.me_pool_force_close_secs
+            .store(force_close_secs, Ordering::Relaxed);
+        self.me_pool_min_fresh_ratio_permille
+            .store(Self::ratio_to_permille(min_fresh_ratio), Ordering::Relaxed);
+        self.me_hardswap_warmup_delay_min_ms
+            .store(hardswap_warmup_delay_min_ms, Ordering::Relaxed);
+        self.me_hardswap_warmup_delay_max_ms
+            .store(hardswap_warmup_delay_max_ms, Ordering::Relaxed);
+        self.me_hardswap_warmup_extra_passes
+            .store(hardswap_warmup_extra_passes as u32, Ordering::Relaxed);
+        self.me_hardswap_warmup_pass_backoff_base_ms
+            .store(hardswap_warmup_pass_backoff_base_ms, Ordering::Relaxed);
+        self.me_bind_stale_mode
+            .store(bind_stale_mode.as_u8(), Ordering::Relaxed);
+        self.me_bind_stale_ttl_secs
+            .store(bind_stale_ttl_secs, Ordering::Relaxed);
+        self.secret_atomic_snapshot
+            .store(secret_atomic_snapshot, Ordering::Relaxed);
+        self.me_deterministic_writer_sort
+            .store(deterministic_writer_sort, Ordering::Relaxed);
+        self.me_single_endpoint_shadow_writers
+            .store(single_endpoint_shadow_writers, Ordering::Relaxed);
+        self.me_single_endpoint_outage_mode_enabled
+            .store(single_endpoint_outage_mode_enabled, Ordering::Relaxed);
+        self.me_single_endpoint_outage_disable_quarantine
+            .store(single_endpoint_outage_disable_quarantine, Ordering::Relaxed);
+        self.me_single_endpoint_outage_backoff_min_ms
+            .store(single_endpoint_outage_backoff_min_ms, Ordering::Relaxed);
+        self.me_single_endpoint_outage_backoff_max_ms
+            .store(single_endpoint_outage_backoff_max_ms, Ordering::Relaxed);
+        self.me_single_endpoint_shadow_rotate_every_secs
+            .store(single_endpoint_shadow_rotate_every_secs, Ordering::Relaxed);
+        let previous_floor_mode = self.floor_mode();
+        self.me_floor_mode
+            .store(floor_mode.as_u8(), Ordering::Relaxed);
+        self.me_adaptive_floor_idle_secs
+            .store(adaptive_floor_idle_secs, Ordering::Relaxed);
+        self.me_adaptive_floor_min_writers_single_endpoint
+            .store(adaptive_floor_min_writers_single_endpoint, Ordering::Relaxed);
+        self.me_adaptive_floor_recover_grace_secs
+            .store(adaptive_floor_recover_grace_secs, Ordering::Relaxed);
+        if previous_floor_mode != floor_mode {
+            self.stats.increment_me_floor_mode_switch_total();
+            match (previous_floor_mode, floor_mode) {
+                (MeFloorMode::Static, MeFloorMode::Adaptive) => {
+                    self.stats
+                        .increment_me_floor_mode_switch_static_to_adaptive_total();
+                }
+                (MeFloorMode::Adaptive, MeFloorMode::Static) => {
+                    self.stats
+                        .increment_me_floor_mode_switch_adaptive_to_static_total();
+                }
+                _ => {}
+            }
+        }
+    }
+
+    pub fn reset_stun_state(&self) {
+        self.nat_probe_attempts.store(0, Ordering::Relaxed);
+        self.nat_probe_disabled.store(false, Ordering::Relaxed);
+        if let Ok(mut live) = self.nat_stun_live_servers.try_write() {
+            live.clear();
+        }
+    }
+
+    pub fn translate_our_addr(&self, addr: SocketAddr) -> SocketAddr {
+        let ip = self.translate_ip_for_nat(addr.ip());
+        SocketAddr::new(ip, addr.port())
+    }
+
+    pub fn registry(&self) -> &Arc<ConnRegistry> {
+        &self.registry
+    }
+
+    pub fn update_runtime_transport_policy(
+        &self,
+        socks_kdf_policy: MeSocksKdfPolicy,
+        route_backpressure_base_timeout_ms: u64,
+        route_backpressure_high_timeout_ms: u64,
+        route_backpressure_high_watermark_pct: u8,
+    ) {
+        self.me_socks_kdf_policy
+            .store(socks_kdf_policy.as_u8(), Ordering::Relaxed);
+        self.registry.update_route_backpressure_policy(
+            route_backpressure_base_timeout_ms,
+            route_backpressure_high_timeout_ms,
+            route_backpressure_high_watermark_pct,
+        );
+    }
+
+    pub(super) fn socks_kdf_policy(&self) -> MeSocksKdfPolicy {
+        MeSocksKdfPolicy::from_u8(self.me_socks_kdf_policy.load(Ordering::Relaxed))
+    }
+
+    pub(super) fn writers_arc(&self) -> Arc<RwLock<Vec<MeWriter>>> {
+        self.writers.clone()
+    }
+
+    pub(super) fn force_close_timeout(&self) -> Option<Duration> {
+        let secs = self.me_pool_force_close_secs.load(Ordering::Relaxed);
+        if secs == 0 {
+            None
+        } else {
+            Some(Duration::from_secs(secs))
+        }
+    }
+
+    pub(super) async fn key_selector(&self) -> u32 {
+        self.proxy_secret.read().await.key_selector
+    }
+
+    pub(super) async fn secret_snapshot(&self) -> SecretSnapshot {
+        self.proxy_secret.read().await.clone()
+    }
+
+    pub(super) fn bind_stale_mode(&self) -> MeBindStaleMode {
+        MeBindStaleMode::from_u8(self.me_bind_stale_mode.load(Ordering::Relaxed))
+    }
+
+    pub(super) fn required_writers_for_dc(&self, endpoint_count: usize) -> usize {
+        if endpoint_count == 0 {
+            return 0;
+        }
+        if endpoint_count == 1 {
+            let shadow = self
+                .me_single_endpoint_shadow_writers
+                .load(Ordering::Relaxed) as usize;
+            return (1 + shadow).max(3);
+        }
+        endpoint_count.max(3)
+    }
+
+    pub(super) fn floor_mode(&self) -> MeFloorMode {
+        MeFloorMode::from_u8(self.me_floor_mode.load(Ordering::Relaxed))
+    }
+
+    pub(super) fn adaptive_floor_idle_duration(&self) -> Duration {
+        Duration::from_secs(self.me_adaptive_floor_idle_secs.load(Ordering::Relaxed))
+    }
+
+    pub(super) fn adaptive_floor_recover_grace_duration(&self) -> Duration {
+        Duration::from_secs(
+            self.me_adaptive_floor_recover_grace_secs
+                .load(Ordering::Relaxed),
+        )
+    }
+
+    pub(super) fn required_writers_for_dc_with_floor_mode(
+        &self,
+        endpoint_count: usize,
+        reduce_for_idle: bool,
+    ) -> usize {
+        let base_required = self.required_writers_for_dc(endpoint_count);
+        if !reduce_for_idle {
+            return base_required;
+        }
+        if endpoint_count != 1 || self.floor_mode() != MeFloorMode::Adaptive {
+            return base_required;
+        }
+        let min_writers = (self
+            .me_adaptive_floor_min_writers_single_endpoint
+            .load(Ordering::Relaxed) as usize)
+            .max(1);
+        base_required.min(min_writers)
+    }
+
+    pub(super) fn single_endpoint_outage_mode_enabled(&self) -> bool {
+        self.me_single_endpoint_outage_mode_enabled
+            .load(Ordering::Relaxed)
+    }
+
+    pub(super) fn single_endpoint_outage_disable_quarantine(&self) -> bool {
+        self.me_single_endpoint_outage_disable_quarantine
+            .load(Ordering::Relaxed)
+    }
+
+    pub(super) fn single_endpoint_outage_backoff_bounds_ms(&self) -> (u64, u64) {
+        let min_ms = self
+            .me_single_endpoint_outage_backoff_min_ms
+            .load(Ordering::Relaxed);
+        let max_ms = self
+            .me_single_endpoint_outage_backoff_max_ms
+            .load(Ordering::Relaxed);
+        if min_ms <= max_ms {
+            (min_ms, max_ms)
+        } else {
+            (max_ms, min_ms)
+        }
+    }
+
+    pub(super) fn single_endpoint_shadow_rotate_interval(&self) -> Option<Duration> {
+        let secs = self
+            .me_single_endpoint_shadow_rotate_every_secs
+            .load(Ordering::Relaxed);
+        if secs == 0 {
+            None
+        } else {
+            Some(Duration::from_secs(secs))
+        }
+    }
+
+    pub(super) fn family_order(&self) -> Vec<IpFamily> {
+        let mut order = Vec::new();
+        if self.decision.prefer_ipv6() {
+            if self.decision.ipv6_me {
+                order.push(IpFamily::V6);
+            }
+            if self.decision.ipv4_me {
+                order.push(IpFamily::V4);
+            }
+        } else {
+            if self.decision.ipv4_me {
+                order.push(IpFamily::V4);
+            }
+            if self.decision.ipv6_me {
+                order.push(IpFamily::V6);
+            }
+        }
+        order
+    }
+
+    pub(super) async fn proxy_map_for_family(
+        &self,
+        family: IpFamily,
+    ) -> HashMap<i32, Vec<(IpAddr, u16)>> {
+        match family {
+            IpFamily::V4 => self.proxy_map_v4.read().await.clone(),
+            IpFamily::V6 => self.proxy_map_v6.read().await.clone(),
+        }
+    }
+}

src/transport/middle_proxy/pool_config.rs

@@ -0,0 +1,113 @@
+use std::collections::HashMap;
+use std::net::IpAddr;
+use std::sync::Arc;
+use std::time::Duration;
+
+use tracing::warn;
+
+use super::pool::MePool;
+
+#[derive(Debug, Clone, Copy, PartialEq, Eq)]
+pub enum SnapshotApplyOutcome {
+    AppliedChanged,
+    AppliedNoDelta,
+    RejectedEmpty,
+}
+
+impl SnapshotApplyOutcome {
+    pub fn changed(self) -> bool {
+        matches!(self, SnapshotApplyOutcome::AppliedChanged)
+    }
+}
+
+impl MePool {
+    pub async fn update_proxy_maps(
+        &self,
+        new_v4: HashMap<i32, Vec<(IpAddr, u16)>>,
+        new_v6: Option<HashMap<i32, Vec<(IpAddr, u16)>>>,
+    ) -> SnapshotApplyOutcome {
+        if new_v4.is_empty() && new_v6.as_ref().is_none_or(|v| v.is_empty()) {
+            return SnapshotApplyOutcome::RejectedEmpty;
+        }
+
+        let mut changed = false;
+        {
+            let mut guard = self.proxy_map_v4.write().await;
+            if !new_v4.is_empty() && *guard != new_v4 {
+                *guard = new_v4;
+                changed = true;
+            }
+        }
+        if let Some(v6) = new_v6 {
+            let mut guard = self.proxy_map_v6.write().await;
+            if !v6.is_empty() && *guard != v6 {
+                *guard = v6;
+                changed = true;
+            }
+        }
+        // Ensure negative DC entries mirror positives when absent (Telegram convention).
+        {
+            let mut guard = self.proxy_map_v4.write().await;
+            let keys: Vec<i32> = guard.keys().cloned().collect();
+            for k in keys.iter().cloned().filter(|k| *k > 0) {
+                if !guard.contains_key(&-k)
+                    && let Some(addrs) = guard.get(&k).cloned()
+                {
+                    guard.insert(-k, addrs);
+                }
+            }
+        }
+        {
+            let mut guard = self.proxy_map_v6.write().await;
+            let keys: Vec<i32> = guard.keys().cloned().collect();
+            for k in keys.iter().cloned().filter(|k| *k > 0) {
+                if !guard.contains_key(&-k)
+                    && let Some(addrs) = guard.get(&k).cloned()
+                {
+                    guard.insert(-k, addrs);
+                }
+            }
+        }
+        if changed {
+            SnapshotApplyOutcome::AppliedChanged
+        } else {
+            SnapshotApplyOutcome::AppliedNoDelta
+        }
+    }
+
+    pub async fn update_secret(self: &Arc<Self>, new_secret: Vec<u8>) -> bool {
+        if new_secret.len() < 32 {
+            warn!(len = new_secret.len(), "proxy-secret update ignored (too short)");
+            return false;
+        }
+        let mut guard = self.proxy_secret.write().await;
+        if guard.secret != new_secret {
+            guard.secret = new_secret;
+            guard.key_selector = if guard.secret.len() >= 4 {
+                u32::from_le_bytes([
+                    guard.secret[0],
+                    guard.secret[1],
+                    guard.secret[2],
+                    guard.secret[3],
+                ])
+            } else {
+                0
+            };
+            guard.epoch = guard.epoch.saturating_add(1);
+            drop(guard);
+            self.reconnect_all().await;
+            return true;
+        }
+        false
+    }
+
+    pub async fn reconnect_all(self: &Arc<Self>) {
+        let ws = self.writers.read().await.clone();
+        for w in ws {
+            if let Ok(()) = self.connect_one(w.addr, self.rng.as_ref()).await {
+                self.mark_writer_draining(w.id).await;
+                tokio::time::sleep(Duration::from_secs(2)).await;
+            }
+        }
+    }
+}

src/transport/middle_proxy/pool_init.rs

@@ -0,0 +1,230 @@
+use std::collections::{HashMap, HashSet};
+use std::net::{IpAddr, SocketAddr};
+use std::sync::Arc;
+
+use rand::Rng;
+use rand::seq::SliceRandom;
+use tracing::{debug, info, warn};
+
+use crate::crypto::SecureRandom;
+use crate::error::{ProxyError, Result};
+
+use super::pool::MePool;
+
+impl MePool {
+    pub async fn init(self: &Arc<Self>, pool_size: usize, rng: &Arc<SecureRandom>) -> Result<()> {
+        let family_order = self.family_order();
+        let connect_concurrency = self.me_reconnect_max_concurrent_per_dc.max(1) as usize;
+        let ks = self.key_selector().await;
+        info!(
+            me_servers = self.proxy_map_v4.read().await.len(),
+            pool_size,
+            connect_concurrency,
+            key_selector = format_args!("0x{ks:08x}"),
+            secret_len = self.proxy_secret.read().await.secret.len(),
+            "Initializing ME pool"
+        );
+
+        for family in family_order {
+            let map = self.proxy_map_for_family(family).await;
+            let mut grouped_dc_addrs: HashMap<i32, Vec<(IpAddr, u16)>> = HashMap::new();
+            for (dc, addrs) in map {
+                if addrs.is_empty() {
+                    continue;
+                }
+                grouped_dc_addrs.entry(dc.abs()).or_default().extend(addrs);
+            }
+            let mut dc_addrs: Vec<(i32, Vec<(IpAddr, u16)>)> = grouped_dc_addrs
+                .into_iter()
+                .map(|(dc, mut addrs)| {
+                    addrs.sort_unstable();
+                    addrs.dedup();
+                    (dc, addrs)
+                })
+                .collect();
+            dc_addrs.sort_unstable_by_key(|(dc, _)| *dc);
+            dc_addrs.sort_by_key(|(_, addrs)| (addrs.len() != 1, addrs.len()));
+
+            // Stage 1: build base coverage for conditional-cast.
+            // Single-endpoint DCs are prefilled first; multi-endpoint DCs require one live writer.
+            let mut join = tokio::task::JoinSet::new();
+            for (dc, addrs) in dc_addrs.iter().cloned() {
+                if addrs.is_empty() {
+                    continue;
+                }
+                let target_writers = if addrs.len() == 1 {
+                    self.required_writers_for_dc_with_floor_mode(addrs.len(), false)
+                } else {
+                    1usize
+                };
+                let endpoints: HashSet<SocketAddr> = addrs
+                    .iter()
+                    .map(|(ip, port)| SocketAddr::new(*ip, *port))
+                    .collect();
+                if self.active_writer_count_for_endpoints(&endpoints).await >= target_writers {
+                    continue;
+                }
+                let pool = Arc::clone(self);
+                let rng_clone = Arc::clone(rng);
+                join.spawn(async move {
+                    pool.connect_primary_for_dc(
+                        dc,
+                        addrs,
+                        target_writers,
+                        rng_clone,
+                        connect_concurrency,
+                    )
+                    .await
+                });
+            }
+            while join.join_next().await.is_some() {}
+
+            let mut missing_dcs = Vec::new();
+            for (dc, addrs) in &dc_addrs {
+                let endpoints: HashSet<SocketAddr> = addrs
+                    .iter()
+                    .map(|(ip, port)| SocketAddr::new(*ip, *port))
+                    .collect();
+                if self.active_writer_count_for_endpoints(&endpoints).await == 0 {
+                    missing_dcs.push(*dc);
+                }
+            }
+            if !missing_dcs.is_empty() {
+                return Err(ProxyError::Proxy(format!(
+                    "ME init incomplete: no live writers for DC groups {missing_dcs:?}"
+                )));
+            }
+
+            // Stage 2: continue saturating multi-endpoint DC groups in background.
+            let pool = Arc::clone(self);
+            let rng_clone = Arc::clone(rng);
+            let dc_addrs_bg = dc_addrs.clone();
+            tokio::spawn(async move {
+                let mut join_bg = tokio::task::JoinSet::new();
+                for (dc, addrs) in dc_addrs_bg {
+                    if addrs.len() <= 1 {
+                        continue;
+                    }
+                    let target_writers = pool.required_writers_for_dc_with_floor_mode(addrs.len(), false);
+                    let pool_clone = Arc::clone(&pool);
+                    let rng_clone_local = Arc::clone(&rng_clone);
+                    join_bg.spawn(async move {
+                        pool_clone
+                            .connect_primary_for_dc(
+                                dc,
+                                addrs,
+                                target_writers,
+                                rng_clone_local,
+                                connect_concurrency,
+                            )
+                            .await
+                    });
+                }
+                while join_bg.join_next().await.is_some() {}
+                debug!(
+                    current_pool_size = pool.connection_count(),
+                    "Background ME saturation warmup finished"
+                );
+            });
+
+            if !self.decision.effective_multipath && self.connection_count() > 0 {
+                break;
+            }
+        }
+
+        if self.writers.read().await.is_empty() {
+            return Err(ProxyError::Proxy("No ME connections".into()));
+        }
+        info!(
+            active_writers = self.connection_count(),
+            "ME primary pool ready; reserve warmup continues in background"
+        );
+        Ok(())
+    }
+
+    async fn connect_primary_for_dc(
+        self: Arc<Self>,
+        dc: i32,
+        mut addrs: Vec<(IpAddr, u16)>,
+        target_writers: usize,
+        rng: Arc<SecureRandom>,
+        connect_concurrency: usize,
+    ) -> bool {
+        if addrs.is_empty() {
+            return false;
+        }
+        let target_writers = target_writers.max(1);
+        addrs.shuffle(&mut rand::rng());
+        let endpoints: Vec<SocketAddr> = addrs
+            .iter()
+            .map(|(ip, port)| SocketAddr::new(*ip, *port))
+            .collect();
+        let endpoint_set: HashSet<SocketAddr> = endpoints.iter().copied().collect();
+
+        loop {
+            let alive = self.active_writer_count_for_endpoints(&endpoint_set).await;
+            if alive >= target_writers {
+                info!(
+                    dc = %dc,
+                    alive,
+                    target_writers,
+                    "ME connected"
+                );
+                return true;
+            }
+
+            let missing = target_writers.saturating_sub(alive).max(1);
+            let concurrency = connect_concurrency.max(1).min(missing);
+            let mut join = tokio::task::JoinSet::new();
+            for _ in 0..concurrency {
+                let pool = Arc::clone(&self);
+                let rng_clone = Arc::clone(&rng);
+                let endpoints_clone = endpoints.clone();
+                join.spawn(async move {
+                    pool.connect_endpoints_round_robin(&endpoints_clone, rng_clone.as_ref())
+                        .await
+                });
+            }
+
+            let mut progress = false;
+            while let Some(res) = join.join_next().await {
+                match res {
+                    Ok(true) => {
+                        progress = true;
+                    }
+                    Ok(false) => {}
+                    Err(e) => {
+                        warn!(dc = %dc, error = %e, "ME connect task failed");
+                    }
+                }
+            }
+
+            let alive_after = self.active_writer_count_for_endpoints(&endpoint_set).await;
+            if alive_after >= target_writers {
+                info!(
+                    dc = %dc,
+                    alive = alive_after,
+                    target_writers,
+                    "ME connected"
+                );
+                return true;
+            }
+            if !progress {
+                warn!(
+                    dc = %dc,
+                    alive = alive_after,
+                    target_writers,
+                    "All ME servers for DC failed at init"
+                );
+                return false;
+            }
+
+            if self.me_warmup_stagger_enabled {
+                let jitter = rand::rng()
+                    .random_range(0..=self.me_warmup_step_jitter.as_millis() as u64);
+                let delay_ms = self.me_warmup_step_delay.as_millis() as u64 + jitter;
+                tokio::time::sleep(std::time::Duration::from_millis(delay_ms)).await;
+            }
+        }
+    }
+}

src/transport/middle_proxy/pool_nat.rs

@@ -0,0 +1,338 @@
+use std::collections::HashMap;
+use std::net::{IpAddr, Ipv4Addr};
+use std::time::Duration;
+
+use tokio::task::JoinSet;
+use tokio::time::timeout;
+use tracing::{debug, info, warn};
+
+use crate::error::{ProxyError, Result};
+use crate::network::probe::is_bogon;
+use crate::network::stun::{stun_probe_dual, stun_probe_family_with_bind, IpFamily};
+
+use super::MePool;
+use std::time::Instant;
+
+const STUN_BATCH_TIMEOUT: Duration = Duration::from_secs(5);
+
+#[allow(dead_code)]
+pub async fn stun_probe(stun_addr: Option<String>) -> Result<crate::network::stun::DualStunResult> {
+    let stun_addr = stun_addr.unwrap_or_else(|| {
+        crate::config::defaults::default_stun_servers()
+            .into_iter()
+            .next()
+            .unwrap_or_default()
+    });
+    if stun_addr.is_empty() {
+        return Err(ProxyError::Proxy("STUN server is not configured".to_string()));
+    }
+    stun_probe_dual(&stun_addr).await
+}
+
+#[allow(dead_code)]
+pub async fn detect_public_ip() -> Option<IpAddr> {
+    fetch_public_ipv4_with_retry().await.ok().flatten().map(IpAddr::V4)
+}
+
+impl MePool {
+    fn configured_stun_servers(&self) -> Vec<String> {
+        if !self.nat_stun_servers.is_empty() {
+            return self.nat_stun_servers.clone();
+        }
+        if let Some(s) = &self.nat_stun
+            && !s.trim().is_empty()
+        {
+            return vec![s.clone()];
+        }
+        Vec::new()
+    }
+
+    async fn probe_stun_batch_for_family(
+        &self,
+        servers: &[String],
+        family: IpFamily,
+        attempt: u8,
+        bind_ip: Option<IpAddr>,
+    ) -> (Vec<String>, Option<std::net::SocketAddr>) {
+        let mut join_set = JoinSet::new();
+        let mut next_idx = 0usize;
+        let mut live_servers = Vec::new();
+        let mut best_by_ip: HashMap<IpAddr, (usize, std::net::SocketAddr)> = HashMap::new();
+        let concurrency = self.nat_probe_concurrency.max(1);
+
+        while next_idx < servers.len() || !join_set.is_empty() {
+            while next_idx < servers.len() && join_set.len() < concurrency {
+                let stun_addr = servers[next_idx].clone();
+                next_idx += 1;
+                join_set.spawn(async move {
+                    let res = timeout(
+                        STUN_BATCH_TIMEOUT,
+                        stun_probe_family_with_bind(&stun_addr, family, bind_ip),
+                    )
+                    .await;
+                    (stun_addr, res)
+                });
+            }
+
+            let Some(task) = join_set.join_next().await else {
+                break;
+            };
+
+            match task {
+                Ok((stun_addr, Ok(Ok(picked)))) => {
+                    if let Some(result) = picked {
+                        live_servers.push(stun_addr.clone());
+                        let entry = best_by_ip
+                            .entry(result.reflected_addr.ip())
+                            .or_insert((0, result.reflected_addr));
+                        entry.0 += 1;
+                        debug!(
+                            local = %result.local_addr,
+                            reflected = %result.reflected_addr,
+                            family = ?family,
+                            stun = %stun_addr,
+                            "NAT probe: reflected address"
+                        );
+                    }
+                }
+                Ok((stun_addr, Ok(Err(e)))) => {
+                    debug!(
+                        error = %e,
+                        stun = %stun_addr,
+                        attempt = attempt + 1,
+                        "NAT probe failed, trying next server"
+                    );
+                }
+                Ok((stun_addr, Err(_))) => {
+                    debug!(
+                        stun = %stun_addr,
+                        attempt = attempt + 1,
+                        "NAT probe timeout, trying next server"
+                    );
+                }
+                Err(e) => {
+                    debug!(
+                        error = %e,
+                        attempt = attempt + 1,
+                        "NAT probe task join failed"
+                    );
+                }
+            }
+        }
+
+        live_servers.sort_unstable();
+        live_servers.dedup();
+        let best_reflected = best_by_ip
+            .into_values()
+            .max_by_key(|(count, _)| *count)
+            .map(|(_, addr)| addr);
+
+        (live_servers, best_reflected)
+    }
+
+    pub(super) fn translate_ip_for_nat(&self, ip: IpAddr) -> IpAddr {
+        let nat_ip = self
+            .nat_ip_cfg
+            .or_else(|| self.nat_ip_detected.try_read().ok().and_then(|g| *g));
+
+        let Some(nat_ip) = nat_ip else {
+            return ip;
+        };
+
+        match (ip, nat_ip) {
+            (IpAddr::V4(src), IpAddr::V4(dst))
+                if is_bogon(IpAddr::V4(src))
+                    || src.is_loopback()
+                    || src.is_unspecified() =>
+            {
+                IpAddr::V4(dst)
+            }
+            (IpAddr::V6(src), IpAddr::V6(dst)) if src.is_loopback() || src.is_unspecified() => {
+                IpAddr::V6(dst)
+            }
+            (orig, _) => orig,
+        }
+    }
+
+    pub(super) fn translate_our_addr_with_reflection(
+        &self,
+        addr: std::net::SocketAddr,
+        reflected: Option<std::net::SocketAddr>,
+    ) -> std::net::SocketAddr {
+        let ip = if let Some(r) = reflected {
+            // Use reflected IP (not port) only when local address is non-public.
+            if is_bogon(addr.ip()) || addr.ip().is_loopback() || addr.ip().is_unspecified() {
+                r.ip()
+            } else {
+                self.translate_ip_for_nat(addr.ip())
+            }
+        } else {
+            self.translate_ip_for_nat(addr.ip())
+        };
+
+        // Keep the kernel-assigned TCP source port; STUN port can differ.
+        std::net::SocketAddr::new(ip, addr.port())
+    }
+
+    pub(super) async fn maybe_detect_nat_ip(&self, local_ip: IpAddr) -> Option<IpAddr> {
+        if self.nat_ip_cfg.is_some() {
+            return self.nat_ip_cfg;
+        }
+
+        if !(is_bogon(local_ip) || local_ip.is_loopback() || local_ip.is_unspecified()) {
+            return None;
+        }
+
+        if let Some(ip) = *self.nat_ip_detected.read().await {
+            return Some(ip);
+        }
+
+        match fetch_public_ipv4_with_retry().await {
+            Ok(Some(ip)) => {
+                {
+                    let mut guard = self.nat_ip_detected.write().await;
+                    *guard = Some(IpAddr::V4(ip));
+                }
+                info!(public_ip = %ip, "Auto-detected public IP for NAT translation");
+                Some(IpAddr::V4(ip))
+            }
+            Ok(None) => None,
+            Err(e) => {
+                warn!(error = %e, "Failed to auto-detect public IP");
+                None
+            }
+        }
+    }
+
+    pub(super) async fn maybe_reflect_public_addr(
+        &self,
+        family: IpFamily,
+        bind_ip: Option<IpAddr>,
+    ) -> Option<std::net::SocketAddr> {
+        const STUN_CACHE_TTL: Duration = Duration::from_secs(600);
+        let use_shared_cache = bind_ip.is_none();
+        if !use_shared_cache {
+            match (family, bind_ip) {
+                (IpFamily::V4, Some(IpAddr::V4(_)))
+                | (IpFamily::V6, Some(IpAddr::V6(_)))
+                | (_, None) => {}
+                _ => return None,
+            }
+        }
+        // Backoff window
+        if use_shared_cache
+            && let Some(until) = *self.stun_backoff_until.read().await
+            && Instant::now() < until
+        {
+            if let Ok(cache) = self.nat_reflection_cache.try_lock() {
+                let slot = match family {
+                    IpFamily::V4 => cache.v4,
+                    IpFamily::V6 => cache.v6,
+                };
+                return slot.map(|(_, addr)| addr);
+            }
+            return None;
+        }
+
+        if use_shared_cache
+            && let Ok(mut cache) = self.nat_reflection_cache.try_lock()
+        {
+            let slot = match family {
+                IpFamily::V4 => &mut cache.v4,
+                IpFamily::V6 => &mut cache.v6,
+            };
+            if let Some((ts, addr)) = slot
+                && ts.elapsed() < STUN_CACHE_TTL
+            {
+                return Some(*addr);
+            }
+        }
+
+        let attempt = if use_shared_cache {
+            self.nat_probe_attempts.fetch_add(1, std::sync::atomic::Ordering::Relaxed)
+        } else {
+            0
+        };
+        let configured_servers = self.configured_stun_servers();
+        let live_snapshot = self.nat_stun_live_servers.read().await.clone();
+        let primary_servers = if live_snapshot.is_empty() {
+            configured_servers.clone()
+        } else {
+            live_snapshot
+        };
+
+        let (mut live_servers, mut selected_reflected) = self
+            .probe_stun_batch_for_family(&primary_servers, family, attempt, bind_ip)
+            .await;
+
+        if selected_reflected.is_none() && !configured_servers.is_empty() && primary_servers != configured_servers {
+            let (rediscovered_live, rediscovered_reflected) = self
+                .probe_stun_batch_for_family(&configured_servers, family, attempt, bind_ip)
+                .await;
+            live_servers = rediscovered_live;
+            selected_reflected = rediscovered_reflected;
+        }
+
+        let live_server_count = live_servers.len();
+        if !live_servers.is_empty() {
+            *self.nat_stun_live_servers.write().await = live_servers;
+        } else {
+            self.nat_stun_live_servers.write().await.clear();
+        }
+
+        if let Some(reflected_addr) = selected_reflected {
+            if use_shared_cache {
+                self.nat_probe_attempts.store(0, std::sync::atomic::Ordering::Relaxed);
+            }
+            info!(
+                family = ?family,
+                live_servers = live_server_count,
+                "STUN-Quorum reached, IP: {}",
+                reflected_addr.ip()
+            );
+            if use_shared_cache
+                && let Ok(mut cache) = self.nat_reflection_cache.try_lock()
+            {
+                let slot = match family {
+                    IpFamily::V4 => &mut cache.v4,
+                    IpFamily::V6 => &mut cache.v6,
+                };
+                *slot = Some((Instant::now(), reflected_addr));
+            }
+            return Some(reflected_addr);
+        }
+
+        if use_shared_cache {
+            let backoff = Duration::from_secs(60 * 2u64.pow((attempt as u32).min(6)));
+            *self.stun_backoff_until.write().await = Some(Instant::now() + backoff);
+        }
+        None
+    }
+}
+
+async fn fetch_public_ipv4_with_retry() -> Result<Option<Ipv4Addr>> {
+    let providers = [
+        "https://checkip.amazonaws.com",
+        "http://v4.ident.me",
+        "http://ipv4.icanhazip.com",
+    ];
+    for url in providers {
+        if let Ok(Some(ip)) = fetch_public_ipv4_once(url).await {
+            return Ok(Some(ip));
+        }
+    }
+    Ok(None)
+}
+
+async fn fetch_public_ipv4_once(url: &str) -> Result<Option<Ipv4Addr>> {
+    let res = reqwest::get(url).await.map_err(|e| {
+        ProxyError::Proxy(format!("public IP detection request failed: {e}"))
+    })?;
+
+    let text = res.text().await.map_err(|e| {
+        ProxyError::Proxy(format!("public IP detection read failed: {e}"))
+    })?;
+
+    let ip = text.trim().parse().ok();
+    Ok(ip)
+}

src/transport/middle_proxy/pool_refill.rs

@@ -0,0 +1,336 @@
+use std::collections::HashSet;
+use std::net::SocketAddr;
+use std::sync::Arc;
+use std::sync::atomic::Ordering;
+use std::time::{Duration, Instant};
+
+use tracing::{debug, info, warn};
+
+use crate::crypto::SecureRandom;
+use crate::network::IpFamily;
+
+use super::pool::{MePool, RefillDcKey, WriterContour};
+
+const ME_FLAP_UPTIME_THRESHOLD_SECS: u64 = 20;
+const ME_FLAP_QUARANTINE_SECS: u64 = 25;
+
+impl MePool {
+    pub(super) async fn maybe_quarantine_flapping_endpoint(
+        &self,
+        addr: SocketAddr,
+        uptime: Duration,
+    ) {
+        if uptime > Duration::from_secs(ME_FLAP_UPTIME_THRESHOLD_SECS) {
+            return;
+        }
+
+        let until = Instant::now() + Duration::from_secs(ME_FLAP_QUARANTINE_SECS);
+        let mut guard = self.endpoint_quarantine.lock().await;
+        guard.retain(|_, expiry| *expiry > Instant::now());
+        guard.insert(addr, until);
+        self.stats.increment_me_endpoint_quarantine_total();
+        warn!(
+            %addr,
+            uptime_ms = uptime.as_millis(),
+            quarantine_secs = ME_FLAP_QUARANTINE_SECS,
+            "ME endpoint temporarily quarantined due to rapid writer flap"
+        );
+    }
+
+    pub(super) async fn is_endpoint_quarantined(&self, addr: SocketAddr) -> bool {
+        let mut guard = self.endpoint_quarantine.lock().await;
+        let now = Instant::now();
+        guard.retain(|_, expiry| *expiry > now);
+        guard.contains_key(&addr)
+    }
+
+    async fn connectable_endpoints(&self, endpoints: &[SocketAddr]) -> Vec<SocketAddr> {
+        if endpoints.is_empty() {
+            return Vec::new();
+        }
+
+        let mut guard = self.endpoint_quarantine.lock().await;
+        let now = Instant::now();
+        guard.retain(|_, expiry| *expiry > now);
+
+        let mut ready = Vec::<SocketAddr>::with_capacity(endpoints.len());
+        let mut earliest_quarantine: Option<(SocketAddr, Instant)> = None;
+        for addr in endpoints {
+            if let Some(expiry) = guard.get(addr).copied() {
+                match earliest_quarantine {
+                    Some((_, current_expiry)) if current_expiry <= expiry => {}
+                    _ => earliest_quarantine = Some((*addr, expiry)),
+                }
+            } else {
+                ready.push(*addr);
+            }
+        }
+
+        if !ready.is_empty() {
+            return ready;
+        }
+
+        if let Some((addr, expiry)) = earliest_quarantine {
+            debug!(
+                %addr,
+                wait_ms = expiry.saturating_duration_since(now).as_millis(),
+                "All ME endpoints are quarantined for the DC group; retrying earliest one"
+            );
+            return vec![addr];
+        }
+
+        Vec::new()
+    }
+
+    pub(super) async fn has_refill_inflight_for_endpoints(&self, endpoints: &[SocketAddr]) -> bool {
+        if endpoints.is_empty() {
+            return false;
+        }
+
+        {
+            let guard = self.refill_inflight.lock().await;
+            if endpoints.iter().any(|addr| guard.contains(addr)) {
+                return true;
+            }
+        }
+
+        let dc_keys = self.resolve_refill_dc_keys_for_endpoints(endpoints).await;
+        if dc_keys.is_empty() {
+            return false;
+        }
+        let guard = self.refill_inflight_dc.lock().await;
+        dc_keys.iter().any(|key| guard.contains(key))
+    }
+
+    async fn resolve_refill_dc_key_for_addr(&self, addr: SocketAddr) -> Option<RefillDcKey> {
+        let family = if addr.is_ipv4() {
+            IpFamily::V4
+        } else {
+            IpFamily::V6
+        };
+        let map = self.proxy_map_for_family(family).await;
+        for (dc, endpoints) in map {
+            if endpoints
+                .into_iter()
+                .any(|(ip, port)| SocketAddr::new(ip, port) == addr)
+            {
+                return Some(RefillDcKey {
+                    dc: dc.abs(),
+                    family,
+                });
+            }
+        }
+        None
+    }
+
+    async fn resolve_refill_dc_keys_for_endpoints(
+        &self,
+        endpoints: &[SocketAddr],
+    ) -> HashSet<RefillDcKey> {
+        let mut out = HashSet::<RefillDcKey>::new();
+        for addr in endpoints {
+            if let Some(key) = self.resolve_refill_dc_key_for_addr(*addr).await {
+                out.insert(key);
+            }
+        }
+        out
+    }
+
+    pub(super) async fn connect_endpoints_round_robin(
+        self: &Arc<Self>,
+        endpoints: &[SocketAddr],
+        rng: &SecureRandom,
+    ) -> bool {
+        self.connect_endpoints_round_robin_with_generation_contour(
+            endpoints,
+            rng,
+            self.current_generation(),
+            WriterContour::Active,
+        )
+        .await
+    }
+
+    pub(super) async fn connect_endpoints_round_robin_with_generation_contour(
+        self: &Arc<Self>,
+        endpoints: &[SocketAddr],
+        rng: &SecureRandom,
+        generation: u64,
+        contour: WriterContour,
+    ) -> bool {
+        let candidates = self.connectable_endpoints(endpoints).await;
+        if candidates.is_empty() {
+            return false;
+        }
+        let start = (self.rr.fetch_add(1, Ordering::Relaxed) as usize) % candidates.len();
+        for offset in 0..candidates.len() {
+            let idx = (start + offset) % candidates.len();
+            let addr = candidates[idx];
+            match self
+                .connect_one_with_generation_contour(addr, rng, generation, contour)
+                .await
+            {
+                Ok(()) => return true,
+                Err(e) => debug!(%addr, error = %e, "ME connect failed during round-robin warmup"),
+            }
+        }
+        false
+    }
+
+    async fn endpoints_for_same_dc(&self, addr: SocketAddr) -> Vec<SocketAddr> {
+        let mut target_dc = HashSet::<i32>::new();
+        let mut endpoints = HashSet::<SocketAddr>::new();
+
+        if self.decision.ipv4_me {
+            let map = self.proxy_map_v4.read().await.clone();
+            for (dc, addrs) in &map {
+                if addrs
+                    .iter()
+                    .any(|(ip, port)| SocketAddr::new(*ip, *port) == addr)
+                {
+                    target_dc.insert(dc.abs());
+                }
+            }
+            for dc in &target_dc {
+                for key in [*dc, -*dc] {
+                    if let Some(addrs) = map.get(&key) {
+                        for (ip, port) in addrs {
+                            endpoints.insert(SocketAddr::new(*ip, *port));
+                        }
+                    }
+                }
+            }
+        }
+
+        if self.decision.ipv6_me {
+            let map = self.proxy_map_v6.read().await.clone();
+            for (dc, addrs) in &map {
+                if addrs
+                    .iter()
+                    .any(|(ip, port)| SocketAddr::new(*ip, *port) == addr)
+                {
+                    target_dc.insert(dc.abs());
+                }
+            }
+            for dc in &target_dc {
+                for key in [*dc, -*dc] {
+                    if let Some(addrs) = map.get(&key) {
+                        for (ip, port) in addrs {
+                            endpoints.insert(SocketAddr::new(*ip, *port));
+                        }
+                    }
+                }
+            }
+        }
+
+        let mut sorted: Vec<SocketAddr> = endpoints.into_iter().collect();
+        sorted.sort_unstable();
+        sorted
+    }
+
+    async fn refill_writer_after_loss(self: &Arc<Self>, addr: SocketAddr) -> bool {
+        let fast_retries = self.me_reconnect_fast_retry_count.max(1);
+        let same_endpoint_quarantined = self.is_endpoint_quarantined(addr).await;
+
+        if !same_endpoint_quarantined {
+            for attempt in 0..fast_retries {
+                self.stats.increment_me_reconnect_attempt();
+                match self.connect_one(addr, self.rng.as_ref()).await {
+                    Ok(()) => {
+                        self.stats.increment_me_reconnect_success();
+                        self.stats.increment_me_writer_restored_same_endpoint_total();
+                        info!(
+                            %addr,
+                            attempt = attempt + 1,
+                            "ME writer restored on the same endpoint"
+                        );
+                        return true;
+                    }
+                    Err(e) => {
+                        debug!(
+                            %addr,
+                            attempt = attempt + 1,
+                            error = %e,
+                            "ME immediate same-endpoint reconnect failed"
+                        );
+                    }
+                }
+            }
+        } else {
+            debug!(
+                %addr,
+                "Skipping immediate same-endpoint reconnect because endpoint is quarantined"
+            );
+        }
+
+        let dc_endpoints = self.endpoints_for_same_dc(addr).await;
+        if dc_endpoints.is_empty() {
+            self.stats.increment_me_refill_failed_total();
+            return false;
+        }
+
+        for attempt in 0..fast_retries {
+            self.stats.increment_me_reconnect_attempt();
+            if self
+                .connect_endpoints_round_robin(&dc_endpoints, self.rng.as_ref())
+                .await
+            {
+                self.stats.increment_me_reconnect_success();
+                self.stats.increment_me_writer_restored_fallback_total();
+                info!(
+                    %addr,
+                    attempt = attempt + 1,
+                    "ME writer restored via DC fallback endpoint"
+                );
+                return true;
+            }
+        }
+
+        self.stats.increment_me_refill_failed_total();
+        false
+    }
+
+    pub(crate) fn trigger_immediate_refill(self: &Arc<Self>, addr: SocketAddr) {
+        let pool = Arc::clone(self);
+        tokio::spawn(async move {
+            let dc_endpoints = pool.endpoints_for_same_dc(addr).await;
+            let dc_keys = pool.resolve_refill_dc_keys_for_endpoints(&dc_endpoints).await;
+
+            {
+                let mut guard = pool.refill_inflight.lock().await;
+                if !guard.insert(addr) {
+                    pool.stats.increment_me_refill_skipped_inflight_total();
+                    return;
+                }
+            }
+
+            if !dc_keys.is_empty() {
+                let mut dc_guard = pool.refill_inflight_dc.lock().await;
+                if dc_keys.iter().any(|key| dc_guard.contains(key)) {
+                    pool.stats.increment_me_refill_skipped_inflight_total();
+                    drop(dc_guard);
+                    let mut guard = pool.refill_inflight.lock().await;
+                    guard.remove(&addr);
+                    return;
+                }
+                dc_guard.extend(dc_keys.iter().copied());
+            }
+
+            pool.stats.increment_me_refill_triggered_total();
+
+            let restored = pool.refill_writer_after_loss(addr).await;
+            if !restored {
+                warn!(%addr, "ME immediate refill failed");
+            }
+
+            let mut guard = pool.refill_inflight.lock().await;
+            guard.remove(&addr);
+            drop(guard);
+            if !dc_keys.is_empty() {
+                let mut dc_guard = pool.refill_inflight_dc.lock().await;
+                for key in &dc_keys {
+                    dc_guard.remove(key);
+                }
+            }
+        });
+    }
+}

src/transport/middle_proxy/pool_reinit.rs

@@ -0,0 +1,493 @@
+use std::collections::{HashMap, HashSet};
+use std::hash::{Hash, Hasher};
+use std::net::SocketAddr;
+use std::sync::Arc;
+use std::sync::atomic::Ordering;
+use std::time::Duration;
+
+use rand::Rng;
+use rand::seq::SliceRandom;
+use tracing::{debug, info, warn};
+use std::collections::hash_map::DefaultHasher;
+
+use crate::crypto::SecureRandom;
+
+use super::pool::{MePool, WriterContour};
+
+const ME_HARDSWAP_PENDING_TTL_SECS: u64 = 1800;
+
+impl MePool {
+    fn desired_map_hash(desired_by_dc: &HashMap<i32, HashSet<SocketAddr>>) -> u64 {
+        let mut hasher = DefaultHasher::new();
+        let mut dcs: Vec<i32> = desired_by_dc.keys().copied().collect();
+        dcs.sort_unstable();
+        for dc in dcs {
+            dc.hash(&mut hasher);
+            let mut endpoints: Vec<SocketAddr> = desired_by_dc
+                .get(&dc)
+                .map(|set| set.iter().copied().collect())
+                .unwrap_or_default();
+            endpoints.sort_unstable();
+            for endpoint in endpoints {
+                endpoint.hash(&mut hasher);
+            }
+        }
+        hasher.finish()
+    }
+
+    fn clear_pending_hardswap_state(&self) {
+        self.pending_hardswap_generation.store(0, Ordering::Relaxed);
+        self.pending_hardswap_started_at_epoch_secs
+            .store(0, Ordering::Relaxed);
+        self.pending_hardswap_map_hash.store(0, Ordering::Relaxed);
+        self.warm_generation.store(0, Ordering::Relaxed);
+    }
+
+    async fn promote_warm_generation_to_active(&self, generation: u64) {
+        self.active_generation.store(generation, Ordering::Relaxed);
+        self.warm_generation.store(0, Ordering::Relaxed);
+
+        let ws = self.writers.read().await;
+        for writer in ws.iter() {
+            if writer.draining.load(Ordering::Relaxed) {
+                continue;
+            }
+            if writer.generation == generation {
+                writer
+                    .contour
+                    .store(WriterContour::Active.as_u8(), Ordering::Relaxed);
+            }
+        }
+    }
+
+    fn coverage_ratio(
+        desired_by_dc: &HashMap<i32, HashSet<SocketAddr>>,
+        active_writer_addrs: &HashSet<SocketAddr>,
+    ) -> (f32, Vec<i32>) {
+        if desired_by_dc.is_empty() {
+            return (1.0, Vec::new());
+        }
+
+        let mut missing_dc = Vec::<i32>::new();
+        let mut covered = 0usize;
+        for (dc, endpoints) in desired_by_dc {
+            if endpoints.is_empty() {
+                continue;
+            }
+            if endpoints
+                .iter()
+                .any(|addr| active_writer_addrs.contains(addr))
+            {
+                covered += 1;
+            } else {
+                missing_dc.push(*dc);
+            }
+        }
+
+        missing_dc.sort_unstable();
+        let total = desired_by_dc.len().max(1);
+        let ratio = (covered as f32) / (total as f32);
+        (ratio, missing_dc)
+    }
+
+    pub async fn reconcile_connections(self: &Arc<Self>, rng: &SecureRandom) {
+        let writers = self.writers.read().await;
+        let current: HashSet<SocketAddr> = writers
+            .iter()
+            .filter(|w| !w.draining.load(Ordering::Relaxed))
+            .map(|w| w.addr)
+            .collect();
+        drop(writers);
+
+        for family in self.family_order() {
+            let map = self.proxy_map_for_family(family).await;
+            for (_dc, addrs) in &map {
+                let dc_addrs: Vec<SocketAddr> = addrs
+                    .iter()
+                    .map(|(ip, port)| SocketAddr::new(*ip, *port))
+                    .collect();
+                if !dc_addrs.iter().any(|a| current.contains(a)) {
+                    let mut shuffled = dc_addrs.clone();
+                    shuffled.shuffle(&mut rand::rng());
+                    for addr in shuffled {
+                        if self.connect_one(addr, rng).await.is_ok() {
+                            break;
+                        }
+                    }
+                }
+            }
+            if !self.decision.effective_multipath && !current.is_empty() {
+                break;
+            }
+        }
+    }
+
+    async fn desired_dc_endpoints(&self) -> HashMap<i32, HashSet<SocketAddr>> {
+        let mut out: HashMap<i32, HashSet<SocketAddr>> = HashMap::new();
+
+        if self.decision.ipv4_me {
+            let map_v4 = self.proxy_map_v4.read().await.clone();
+            for (dc, addrs) in map_v4 {
+                let entry = out.entry(dc.abs()).or_default();
+                for (ip, port) in addrs {
+                    entry.insert(SocketAddr::new(ip, port));
+                }
+            }
+        }
+
+        if self.decision.ipv6_me {
+            let map_v6 = self.proxy_map_v6.read().await.clone();
+            for (dc, addrs) in map_v6 {
+                let entry = out.entry(dc.abs()).or_default();
+                for (ip, port) in addrs {
+                    entry.insert(SocketAddr::new(ip, port));
+                }
+            }
+        }
+
+        out
+    }
+
+    fn hardswap_warmup_connect_delay_ms(&self) -> u64 {
+        let min_ms = self.me_hardswap_warmup_delay_min_ms.load(Ordering::Relaxed);
+        let max_ms = self.me_hardswap_warmup_delay_max_ms.load(Ordering::Relaxed);
+        let (min_ms, max_ms) = if min_ms <= max_ms {
+            (min_ms, max_ms)
+        } else {
+            (max_ms, min_ms)
+        };
+        if min_ms == max_ms {
+            return min_ms;
+        }
+        rand::rng().random_range(min_ms..=max_ms)
+    }
+
+    fn hardswap_warmup_backoff_ms(&self, pass_idx: usize) -> u64 {
+        let base_ms = self
+            .me_hardswap_warmup_pass_backoff_base_ms
+            .load(Ordering::Relaxed);
+        let cap_ms = (self.me_reconnect_backoff_cap.as_millis() as u64).max(base_ms);
+        let shift = (pass_idx as u32).min(20);
+        let scaled = base_ms.saturating_mul(1u64 << shift);
+        let core = scaled.min(cap_ms);
+        let jitter = (core / 2).max(1);
+        core.saturating_add(rand::rng().random_range(0..=jitter))
+    }
+
+    async fn fresh_writer_count_for_endpoints(
+        &self,
+        generation: u64,
+        endpoints: &HashSet<SocketAddr>,
+    ) -> usize {
+        let ws = self.writers.read().await;
+        ws.iter()
+            .filter(|w| !w.draining.load(Ordering::Relaxed))
+            .filter(|w| w.generation == generation)
+            .filter(|w| endpoints.contains(&w.addr))
+            .count()
+    }
+
+    pub(super) async fn active_writer_count_for_endpoints(
+        &self,
+        endpoints: &HashSet<SocketAddr>,
+    ) -> usize {
+        let ws = self.writers.read().await;
+        ws.iter()
+            .filter(|w| !w.draining.load(Ordering::Relaxed))
+            .filter(|w| endpoints.contains(&w.addr))
+            .count()
+    }
+
+    async fn warmup_generation_for_all_dcs(
+        self: &Arc<Self>,
+        rng: &SecureRandom,
+        generation: u64,
+        desired_by_dc: &HashMap<i32, HashSet<SocketAddr>>,
+    ) {
+        let extra_passes = self
+            .me_hardswap_warmup_extra_passes
+            .load(Ordering::Relaxed)
+            .min(10) as usize;
+        let total_passes = 1 + extra_passes;
+
+        for (dc, endpoints) in desired_by_dc {
+            if endpoints.is_empty() {
+                continue;
+            }
+
+            let mut endpoint_list: Vec<SocketAddr> = endpoints.iter().copied().collect();
+            endpoint_list.sort_unstable();
+            let required = self.required_writers_for_dc(endpoint_list.len());
+            let mut completed = false;
+            let mut last_fresh_count = self
+                .fresh_writer_count_for_endpoints(generation, endpoints)
+                .await;
+
+            for pass_idx in 0..total_passes {
+                if last_fresh_count >= required {
+                    completed = true;
+                    break;
+                }
+
+                let missing = required.saturating_sub(last_fresh_count);
+                debug!(
+                    dc = *dc,
+                    pass = pass_idx + 1,
+                    total_passes,
+                    fresh_count = last_fresh_count,
+                    required,
+                    missing,
+                    endpoint_count = endpoint_list.len(),
+                    "ME hardswap warmup pass started"
+                );
+
+                for attempt_idx in 0..missing {
+                    let delay_ms = self.hardswap_warmup_connect_delay_ms();
+                    tokio::time::sleep(Duration::from_millis(delay_ms)).await;
+
+                    let connected = self
+                        .connect_endpoints_round_robin_with_generation_contour(
+                            &endpoint_list,
+                            rng,
+                            generation,
+                            WriterContour::Warm,
+                        )
+                        .await;
+                    debug!(
+                        dc = *dc,
+                        pass = pass_idx + 1,
+                        total_passes,
+                        attempt = attempt_idx + 1,
+                        delay_ms,
+                        connected,
+                        "ME hardswap warmup connect attempt finished"
+                    );
+                }
+
+                last_fresh_count = self
+                    .fresh_writer_count_for_endpoints(generation, endpoints)
+                    .await;
+                if last_fresh_count >= required {
+                    completed = true;
+                    info!(
+                        dc = *dc,
+                        pass = pass_idx + 1,
+                        total_passes,
+                        fresh_count = last_fresh_count,
+                        required,
+                        "ME hardswap warmup floor reached for DC"
+                    );
+                    break;
+                }
+
+                if pass_idx + 1 < total_passes {
+                    let backoff_ms = self.hardswap_warmup_backoff_ms(pass_idx);
+                    debug!(
+                        dc = *dc,
+                        pass = pass_idx + 1,
+                        total_passes,
+                        fresh_count = last_fresh_count,
+                        required,
+                        backoff_ms,
+                        "ME hardswap warmup pass incomplete, delaying next pass"
+                    );
+                    tokio::time::sleep(Duration::from_millis(backoff_ms)).await;
+                }
+            }
+
+            if !completed {
+                warn!(
+                    dc = *dc,
+                    fresh_count = last_fresh_count,
+                    required,
+                    endpoint_count = endpoint_list.len(),
+                    total_passes,
+                    "ME warmup stopped: unable to reach required writer floor for DC"
+                );
+            }
+        }
+    }
+
+    pub async fn zero_downtime_reinit_after_map_change(self: &Arc<Self>, rng: &SecureRandom) {
+        let desired_by_dc = self.desired_dc_endpoints().await;
+        if desired_by_dc.is_empty() {
+            warn!("ME endpoint map is empty; skipping stale writer drain");
+            return;
+        }
+
+        let desired_map_hash = Self::desired_map_hash(&desired_by_dc);
+        let now_epoch_secs = Self::now_epoch_secs();
+        let previous_generation = self.current_generation();
+        let hardswap = self.hardswap.load(Ordering::Relaxed);
+        let generation = if hardswap {
+            let pending_generation = self.pending_hardswap_generation.load(Ordering::Relaxed);
+            let pending_started_at = self
+                .pending_hardswap_started_at_epoch_secs
+                .load(Ordering::Relaxed);
+            let pending_map_hash = self.pending_hardswap_map_hash.load(Ordering::Relaxed);
+            let pending_age_secs = now_epoch_secs.saturating_sub(pending_started_at);
+            let pending_ttl_expired = pending_started_at > 0 && pending_age_secs > ME_HARDSWAP_PENDING_TTL_SECS;
+            let pending_matches_map = pending_map_hash != 0 && pending_map_hash == desired_map_hash;
+
+            if pending_generation != 0
+                && pending_generation >= previous_generation
+                && pending_matches_map
+                && !pending_ttl_expired
+            {
+                self.stats.increment_me_hardswap_pending_reuse_total();
+                debug!(
+                    previous_generation,
+                    generation = pending_generation,
+                    pending_age_secs,
+                    "ME hardswap continues with pending generation"
+                );
+                pending_generation
+            } else {
+                if pending_generation != 0 && pending_ttl_expired {
+                    self.stats.increment_me_hardswap_pending_ttl_expired_total();
+                    warn!(
+                        previous_generation,
+                        generation = pending_generation,
+                        pending_age_secs,
+                        pending_ttl_secs = ME_HARDSWAP_PENDING_TTL_SECS,
+                        "ME hardswap pending generation expired by TTL; starting fresh generation"
+                    );
+                }
+                let next_generation = self.generation.fetch_add(1, Ordering::Relaxed) + 1;
+                self.pending_hardswap_generation
+                    .store(next_generation, Ordering::Relaxed);
+                self.pending_hardswap_started_at_epoch_secs
+                    .store(now_epoch_secs, Ordering::Relaxed);
+                self.pending_hardswap_map_hash
+                    .store(desired_map_hash, Ordering::Relaxed);
+                self.warm_generation.store(next_generation, Ordering::Relaxed);
+                next_generation
+            }
+        } else {
+            self.clear_pending_hardswap_state();
+            self.generation.fetch_add(1, Ordering::Relaxed) + 1
+        };
+
+        if hardswap {
+            self.warm_generation.store(generation, Ordering::Relaxed);
+            self.warmup_generation_for_all_dcs(rng, generation, &desired_by_dc)
+                .await;
+        } else {
+            self.reconcile_connections(rng).await;
+        }
+
+        let writers = self.writers.read().await;
+        let active_writer_addrs: HashSet<SocketAddr> = writers
+            .iter()
+            .filter(|w| !w.draining.load(Ordering::Relaxed))
+            .map(|w| w.addr)
+            .collect();
+        let min_ratio = Self::permille_to_ratio(
+            self.me_pool_min_fresh_ratio_permille
+                .load(Ordering::Relaxed),
+        );
+        let (coverage_ratio, missing_dc) = Self::coverage_ratio(&desired_by_dc, &active_writer_addrs);
+        if !hardswap && coverage_ratio < min_ratio {
+            warn!(
+                previous_generation,
+                generation,
+                coverage_ratio = format_args!("{coverage_ratio:.3}"),
+                min_ratio = format_args!("{min_ratio:.3}"),
+                missing_dc = ?missing_dc,
+                "ME reinit coverage below threshold; keeping stale writers"
+            );
+            return;
+        }
+
+        if hardswap {
+            let mut fresh_missing_dc = Vec::<(i32, usize, usize)>::new();
+            for (dc, endpoints) in &desired_by_dc {
+                if endpoints.is_empty() {
+                    continue;
+                }
+                let required = self.required_writers_for_dc(endpoints.len());
+                let fresh_count = writers
+                    .iter()
+                    .filter(|w| !w.draining.load(Ordering::Relaxed))
+                    .filter(|w| w.generation == generation)
+                    .filter(|w| endpoints.contains(&w.addr))
+                    .count();
+                if fresh_count < required {
+                    fresh_missing_dc.push((*dc, fresh_count, required));
+                }
+            }
+            if !fresh_missing_dc.is_empty() {
+                warn!(
+                    previous_generation,
+                    generation,
+                    missing_dc = ?fresh_missing_dc,
+                    "ME hardswap pending: fresh generation coverage incomplete"
+                );
+                return;
+            }
+        } else if !missing_dc.is_empty() {
+            warn!(
+                missing_dc = ?missing_dc,
+                // Keep stale writers alive when fresh coverage is incomplete.
+                "ME reinit coverage incomplete; keeping stale writers"
+            );
+            return;
+        }
+
+        if hardswap {
+            self.promote_warm_generation_to_active(generation).await;
+        }
+
+        let desired_addrs: HashSet<SocketAddr> = desired_by_dc
+            .values()
+            .flat_map(|set| set.iter().copied())
+            .collect();
+
+        let stale_writer_ids: Vec<u64> = writers
+            .iter()
+            .filter(|w| !w.draining.load(Ordering::Relaxed))
+            .filter(|w| {
+                if hardswap {
+                    w.generation < generation
+                } else {
+                    !desired_addrs.contains(&w.addr)
+                }
+            })
+            .map(|w| w.id)
+            .collect();
+        drop(writers);
+
+        if stale_writer_ids.is_empty() {
+            if hardswap {
+                self.clear_pending_hardswap_state();
+            }
+            debug!("ME reinit cycle completed with no stale writers");
+            return;
+        }
+
+        let drain_timeout = self.force_close_timeout();
+        let drain_timeout_secs = drain_timeout.map(|d| d.as_secs()).unwrap_or(0);
+        info!(
+            stale_writers = stale_writer_ids.len(),
+            previous_generation,
+            generation,
+            hardswap,
+            coverage_ratio = format_args!("{coverage_ratio:.3}"),
+            min_ratio = format_args!("{min_ratio:.3}"),
+            drain_timeout_secs,
+            "ME reinit cycle covered; draining stale writers"
+        );
+        self.stats.increment_pool_swap_total();
+        for writer_id in stale_writer_ids {
+            self.mark_writer_draining_with_timeout(writer_id, drain_timeout, !hardswap)
+                .await;
+        }
+        if hardswap {
+            self.clear_pending_hardswap_state();
+        }
+    }
+
+    pub async fn zero_downtime_reinit_periodic(self: &Arc<Self>, rng: &SecureRandom) {
+        self.zero_downtime_reinit_after_map_change(rng).await;
+    }
+}

src/transport/middle_proxy/pool_status.rs

@@ -0,0 +1,552 @@
+use std::collections::{BTreeMap, BTreeSet, HashMap};
+use std::net::SocketAddr;
+use std::sync::atomic::Ordering;
+use std::time::Instant;
+
+use super::pool::{MePool, WriterContour};
+use crate::config::{MeBindStaleMode, MeFloorMode, MeSocksKdfPolicy};
+use crate::transport::upstream::IpPreference;
+
+#[derive(Clone, Debug)]
+pub(crate) struct MeApiWriterStatusSnapshot {
+    pub writer_id: u64,
+    pub dc: Option<i16>,
+    pub endpoint: SocketAddr,
+    pub generation: u64,
+    pub state: &'static str,
+    pub draining: bool,
+    pub degraded: bool,
+    pub bound_clients: usize,
+    pub idle_for_secs: Option<u64>,
+    pub rtt_ema_ms: Option<f64>,
+}
+
+#[derive(Clone, Debug)]
+pub(crate) struct MeApiDcStatusSnapshot {
+    pub dc: i16,
+    pub endpoints: Vec<SocketAddr>,
+    pub available_endpoints: usize,
+    pub available_pct: f64,
+    pub required_writers: usize,
+    pub alive_writers: usize,
+    pub coverage_pct: f64,
+    pub rtt_ms: Option<f64>,
+    pub load: usize,
+}
+
+#[derive(Clone, Debug)]
+pub(crate) struct MeApiStatusSnapshot {
+    pub generated_at_epoch_secs: u64,
+    pub configured_dc_groups: usize,
+    pub configured_endpoints: usize,
+    pub available_endpoints: usize,
+    pub available_pct: f64,
+    pub required_writers: usize,
+    pub alive_writers: usize,
+    pub coverage_pct: f64,
+    pub writers: Vec<MeApiWriterStatusSnapshot>,
+    pub dcs: Vec<MeApiDcStatusSnapshot>,
+}
+
+#[derive(Clone, Debug)]
+pub(crate) struct MeApiQuarantinedEndpointSnapshot {
+    pub endpoint: SocketAddr,
+    pub remaining_ms: u64,
+}
+
+#[derive(Clone, Debug)]
+pub(crate) struct MeApiDcPathSnapshot {
+    pub dc: i16,
+    pub ip_preference: Option<&'static str>,
+    pub selected_addr_v4: Option<SocketAddr>,
+    pub selected_addr_v6: Option<SocketAddr>,
+}
+
+#[derive(Clone, Debug)]
+pub(crate) struct MeApiRuntimeSnapshot {
+    pub active_generation: u64,
+    pub warm_generation: u64,
+    pub pending_hardswap_generation: u64,
+    pub pending_hardswap_age_secs: Option<u64>,
+    pub hardswap_enabled: bool,
+    pub floor_mode: &'static str,
+    pub adaptive_floor_idle_secs: u64,
+    pub adaptive_floor_min_writers_single_endpoint: u8,
+    pub adaptive_floor_recover_grace_secs: u64,
+    pub me_keepalive_enabled: bool,
+    pub me_keepalive_interval_secs: u64,
+    pub me_keepalive_jitter_secs: u64,
+    pub me_keepalive_payload_random: bool,
+    pub rpc_proxy_req_every_secs: u64,
+    pub me_reconnect_max_concurrent_per_dc: u32,
+    pub me_reconnect_backoff_base_ms: u64,
+    pub me_reconnect_backoff_cap_ms: u64,
+    pub me_reconnect_fast_retry_count: u32,
+    pub me_pool_drain_ttl_secs: u64,
+    pub me_pool_force_close_secs: u64,
+    pub me_pool_min_fresh_ratio: f32,
+    pub me_bind_stale_mode: &'static str,
+    pub me_bind_stale_ttl_secs: u64,
+    pub me_single_endpoint_shadow_writers: u8,
+    pub me_single_endpoint_outage_mode_enabled: bool,
+    pub me_single_endpoint_outage_disable_quarantine: bool,
+    pub me_single_endpoint_outage_backoff_min_ms: u64,
+    pub me_single_endpoint_outage_backoff_max_ms: u64,
+    pub me_single_endpoint_shadow_rotate_every_secs: u64,
+    pub me_deterministic_writer_sort: bool,
+    pub me_socks_kdf_policy: &'static str,
+    pub quarantined_endpoints: Vec<MeApiQuarantinedEndpointSnapshot>,
+    pub network_path: Vec<MeApiDcPathSnapshot>,
+}
+
+impl MePool {
+    pub(crate) async fn admission_ready_conditional_cast(&self) -> bool {
+        let mut endpoints_by_dc = BTreeMap::<i16, BTreeSet<SocketAddr>>::new();
+        if self.decision.ipv4_me {
+            let map = self.proxy_map_v4.read().await.clone();
+            for (dc, addrs) in map {
+                let abs_dc = dc.abs();
+                if abs_dc == 0 {
+                    continue;
+                }
+                let Ok(dc_idx) = i16::try_from(abs_dc) else {
+                    continue;
+                };
+                let entry = endpoints_by_dc.entry(dc_idx).or_default();
+                for (ip, port) in addrs {
+                    entry.insert(SocketAddr::new(ip, port));
+                }
+            }
+        }
+        if self.decision.ipv6_me {
+            let map = self.proxy_map_v6.read().await.clone();
+            for (dc, addrs) in map {
+                let abs_dc = dc.abs();
+                if abs_dc == 0 {
+                    continue;
+                }
+                let Ok(dc_idx) = i16::try_from(abs_dc) else {
+                    continue;
+                };
+                let entry = endpoints_by_dc.entry(dc_idx).or_default();
+                for (ip, port) in addrs {
+                    entry.insert(SocketAddr::new(ip, port));
+                }
+            }
+        }
+
+        if endpoints_by_dc.is_empty() {
+            return false;
+        }
+
+        let writers = self.writers.read().await.clone();
+        let mut live_writers_by_endpoint = HashMap::<SocketAddr, usize>::new();
+        for writer in writers {
+            if writer.draining.load(Ordering::Relaxed) {
+                continue;
+            }
+            *live_writers_by_endpoint.entry(writer.addr).or_insert(0) += 1;
+        }
+
+        for endpoints in endpoints_by_dc.values() {
+            let alive: usize = endpoints
+                .iter()
+                .map(|endpoint| live_writers_by_endpoint.get(endpoint).copied().unwrap_or(0))
+                .sum();
+            if alive == 0 {
+                return false;
+            }
+        }
+
+        true
+    }
+
+    #[allow(dead_code)]
+    pub(crate) async fn admission_ready_full_floor(&self) -> bool {
+        let mut endpoints_by_dc = BTreeMap::<i16, BTreeSet<SocketAddr>>::new();
+        if self.decision.ipv4_me {
+            let map = self.proxy_map_v4.read().await.clone();
+            for (dc, addrs) in map {
+                let abs_dc = dc.abs();
+                if abs_dc == 0 {
+                    continue;
+                }
+                let Ok(dc_idx) = i16::try_from(abs_dc) else {
+                    continue;
+                };
+                let entry = endpoints_by_dc.entry(dc_idx).or_default();
+                for (ip, port) in addrs {
+                    entry.insert(SocketAddr::new(ip, port));
+                }
+            }
+        }
+        if self.decision.ipv6_me {
+            let map = self.proxy_map_v6.read().await.clone();
+            for (dc, addrs) in map {
+                let abs_dc = dc.abs();
+                if abs_dc == 0 {
+                    continue;
+                }
+                let Ok(dc_idx) = i16::try_from(abs_dc) else {
+                    continue;
+                };
+                let entry = endpoints_by_dc.entry(dc_idx).or_default();
+                for (ip, port) in addrs {
+                    entry.insert(SocketAddr::new(ip, port));
+                }
+            }
+        }
+
+        if endpoints_by_dc.is_empty() {
+            return false;
+        }
+
+        let writers = self.writers.read().await.clone();
+        let mut live_writers_by_endpoint = HashMap::<SocketAddr, usize>::new();
+        for writer in writers {
+            if writer.draining.load(Ordering::Relaxed) {
+                continue;
+            }
+            *live_writers_by_endpoint.entry(writer.addr).or_insert(0) += 1;
+        }
+
+        for endpoints in endpoints_by_dc.values() {
+            let endpoint_count = endpoints.len();
+            if endpoint_count == 0 {
+                return false;
+            }
+            let required = self.required_writers_for_dc_with_floor_mode(endpoint_count, false);
+            let alive: usize = endpoints
+                .iter()
+                .map(|endpoint| live_writers_by_endpoint.get(endpoint).copied().unwrap_or(0))
+                .sum();
+            if alive < required {
+                return false;
+            }
+        }
+
+        true
+    }
+
+    pub(crate) async fn api_status_snapshot(&self) -> MeApiStatusSnapshot {
+        let now_epoch_secs = Self::now_epoch_secs();
+
+        let mut endpoints_by_dc = BTreeMap::<i16, BTreeSet<SocketAddr>>::new();
+        if self.decision.ipv4_me {
+            let map = self.proxy_map_v4.read().await.clone();
+            for (dc, addrs) in map {
+                let abs_dc = dc.abs();
+                if abs_dc == 0 {
+                    continue;
+                }
+                let Ok(dc_idx) = i16::try_from(abs_dc) else {
+                    continue;
+                };
+                let entry = endpoints_by_dc.entry(dc_idx).or_default();
+                for (ip, port) in addrs {
+                    entry.insert(SocketAddr::new(ip, port));
+                }
+            }
+        }
+        if self.decision.ipv6_me {
+            let map = self.proxy_map_v6.read().await.clone();
+            for (dc, addrs) in map {
+                let abs_dc = dc.abs();
+                if abs_dc == 0 {
+                    continue;
+                }
+                let Ok(dc_idx) = i16::try_from(abs_dc) else {
+                    continue;
+                };
+                let entry = endpoints_by_dc.entry(dc_idx).or_default();
+                for (ip, port) in addrs {
+                    entry.insert(SocketAddr::new(ip, port));
+                }
+            }
+        }
+
+        let mut endpoint_to_dc = HashMap::<SocketAddr, i16>::new();
+        for (dc, endpoints) in &endpoints_by_dc {
+            for endpoint in endpoints {
+                endpoint_to_dc.entry(*endpoint).or_insert(*dc);
+            }
+        }
+
+        let configured_dc_groups = endpoints_by_dc.len();
+        let configured_endpoints = endpoints_by_dc.values().map(BTreeSet::len).sum();
+
+        let required_writers = endpoints_by_dc
+            .values()
+            .map(|endpoints| self.required_writers_for_dc_with_floor_mode(endpoints.len(), false))
+            .sum();
+
+        let idle_since = self.registry.writer_idle_since_snapshot().await;
+        let activity = self.registry.writer_activity_snapshot().await;
+        let rtt = self.rtt_stats.lock().await.clone();
+        let writers = self.writers.read().await.clone();
+
+        let mut live_writers_by_endpoint = HashMap::<SocketAddr, usize>::new();
+        let mut live_writers_by_dc = HashMap::<i16, usize>::new();
+        let mut dc_rtt_agg = HashMap::<i16, (f64, u64)>::new();
+        let mut writer_rows = Vec::<MeApiWriterStatusSnapshot>::with_capacity(writers.len());
+
+        for writer in writers {
+            let endpoint = writer.addr;
+            let dc = endpoint_to_dc.get(&endpoint).copied();
+            let draining = writer.draining.load(Ordering::Relaxed);
+            let degraded = writer.degraded.load(Ordering::Relaxed);
+            let bound_clients = activity
+                .bound_clients_by_writer
+                .get(&writer.id)
+                .copied()
+                .unwrap_or(0);
+            let idle_for_secs = idle_since
+                .get(&writer.id)
+                .map(|idle_ts| now_epoch_secs.saturating_sub(*idle_ts));
+            let rtt_ema_ms = rtt.get(&writer.id).map(|(_, ema)| *ema);
+            let state = match WriterContour::from_u8(writer.contour.load(Ordering::Relaxed)) {
+                WriterContour::Warm => "warm",
+                WriterContour::Active => "active",
+                WriterContour::Draining => "draining",
+            };
+
+            if !draining {
+                *live_writers_by_endpoint.entry(endpoint).or_insert(0) += 1;
+                if let Some(dc_idx) = dc {
+                    *live_writers_by_dc.entry(dc_idx).or_insert(0) += 1;
+                    if let Some(ema_ms) = rtt_ema_ms {
+                        let entry = dc_rtt_agg.entry(dc_idx).or_insert((0.0, 0));
+                        entry.0 += ema_ms;
+                        entry.1 += 1;
+                    }
+                }
+            }
+
+            writer_rows.push(MeApiWriterStatusSnapshot {
+                writer_id: writer.id,
+                dc,
+                endpoint,
+                generation: writer.generation,
+                state,
+                draining,
+                degraded,
+                bound_clients,
+                idle_for_secs,
+                rtt_ema_ms,
+            });
+        }
+
+        writer_rows.sort_by_key(|row| (row.dc.unwrap_or(i16::MAX), row.endpoint, row.writer_id));
+
+        let mut dcs = Vec::<MeApiDcStatusSnapshot>::with_capacity(endpoints_by_dc.len());
+        let mut available_endpoints = 0usize;
+        let mut alive_writers = 0usize;
+        for (dc, endpoints) in endpoints_by_dc {
+            let endpoint_count = endpoints.len();
+            let dc_available_endpoints = endpoints
+                .iter()
+                .filter(|endpoint| live_writers_by_endpoint.contains_key(endpoint))
+                .count();
+            let dc_required_writers =
+                self.required_writers_for_dc_with_floor_mode(endpoint_count, false);
+            let dc_alive_writers = live_writers_by_dc.get(&dc).copied().unwrap_or(0);
+            let dc_load = activity
+                .active_sessions_by_target_dc
+                .get(&dc)
+                .copied()
+                .unwrap_or(0);
+            let dc_rtt_ms = dc_rtt_agg
+                .get(&dc)
+                .and_then(|(sum, count)| (*count > 0).then_some(*sum / (*count as f64)));
+
+            available_endpoints += dc_available_endpoints;
+            alive_writers += dc_alive_writers;
+
+            dcs.push(MeApiDcStatusSnapshot {
+                dc,
+                endpoints: endpoints.into_iter().collect(),
+                available_endpoints: dc_available_endpoints,
+                available_pct: ratio_pct(dc_available_endpoints, endpoint_count),
+                required_writers: dc_required_writers,
+                alive_writers: dc_alive_writers,
+                coverage_pct: ratio_pct(dc_alive_writers, dc_required_writers),
+                rtt_ms: dc_rtt_ms,
+                load: dc_load,
+            });
+        }
+
+        MeApiStatusSnapshot {
+            generated_at_epoch_secs: now_epoch_secs,
+            configured_dc_groups,
+            configured_endpoints,
+            available_endpoints,
+            available_pct: ratio_pct(available_endpoints, configured_endpoints),
+            required_writers,
+            alive_writers,
+            coverage_pct: ratio_pct(alive_writers, required_writers),
+            writers: writer_rows,
+            dcs,
+        }
+    }
+
+    pub(crate) async fn api_runtime_snapshot(&self) -> MeApiRuntimeSnapshot {
+        let now = Instant::now();
+        let now_epoch_secs = Self::now_epoch_secs();
+        let pending_started_at = self
+            .pending_hardswap_started_at_epoch_secs
+            .load(Ordering::Relaxed);
+        let pending_hardswap_age_secs = (pending_started_at > 0)
+            .then_some(now_epoch_secs.saturating_sub(pending_started_at));
+
+        let mut quarantined_endpoints = Vec::<MeApiQuarantinedEndpointSnapshot>::new();
+        {
+            let guard = self.endpoint_quarantine.lock().await;
+            for (endpoint, expires_at) in guard.iter() {
+                if *expires_at <= now {
+                    continue;
+                }
+                let remaining_ms = expires_at.duration_since(now).as_millis() as u64;
+                quarantined_endpoints.push(MeApiQuarantinedEndpointSnapshot {
+                    endpoint: *endpoint,
+                    remaining_ms,
+                });
+            }
+        }
+        quarantined_endpoints.sort_by_key(|entry| entry.endpoint);
+
+        let mut network_path = Vec::<MeApiDcPathSnapshot>::new();
+        if let Some(upstream) = &self.upstream {
+            for dc in 1..=5 {
+                let dc_idx = dc as i16;
+                let ip_preference = upstream
+                    .get_dc_ip_preference(dc_idx)
+                    .await
+                    .map(ip_preference_label);
+                let selected_addr_v4 = upstream.get_dc_addr(dc_idx, false).await;
+                let selected_addr_v6 = upstream.get_dc_addr(dc_idx, true).await;
+                network_path.push(MeApiDcPathSnapshot {
+                    dc: dc_idx,
+                    ip_preference,
+                    selected_addr_v4,
+                    selected_addr_v6,
+                });
+            }
+        }
+
+        MeApiRuntimeSnapshot {
+            active_generation: self.active_generation.load(Ordering::Relaxed),
+            warm_generation: self.warm_generation.load(Ordering::Relaxed),
+            pending_hardswap_generation: self.pending_hardswap_generation.load(Ordering::Relaxed),
+            pending_hardswap_age_secs,
+            hardswap_enabled: self.hardswap.load(Ordering::Relaxed),
+            floor_mode: floor_mode_label(self.floor_mode()),
+            adaptive_floor_idle_secs: self.me_adaptive_floor_idle_secs.load(Ordering::Relaxed),
+            adaptive_floor_min_writers_single_endpoint: self
+                .me_adaptive_floor_min_writers_single_endpoint
+                .load(Ordering::Relaxed),
+            adaptive_floor_recover_grace_secs: self
+                .me_adaptive_floor_recover_grace_secs
+                .load(Ordering::Relaxed),
+            me_keepalive_enabled: self.me_keepalive_enabled,
+            me_keepalive_interval_secs: self.me_keepalive_interval.as_secs(),
+            me_keepalive_jitter_secs: self.me_keepalive_jitter.as_secs(),
+            me_keepalive_payload_random: self.me_keepalive_payload_random,
+            rpc_proxy_req_every_secs: self.rpc_proxy_req_every_secs.load(Ordering::Relaxed),
+            me_reconnect_max_concurrent_per_dc: self.me_reconnect_max_concurrent_per_dc,
+            me_reconnect_backoff_base_ms: self.me_reconnect_backoff_base.as_millis() as u64,
+            me_reconnect_backoff_cap_ms: self.me_reconnect_backoff_cap.as_millis() as u64,
+            me_reconnect_fast_retry_count: self.me_reconnect_fast_retry_count,
+            me_pool_drain_ttl_secs: self.me_pool_drain_ttl_secs.load(Ordering::Relaxed),
+            me_pool_force_close_secs: self.me_pool_force_close_secs.load(Ordering::Relaxed),
+            me_pool_min_fresh_ratio: Self::permille_to_ratio(
+                self.me_pool_min_fresh_ratio_permille.load(Ordering::Relaxed),
+            ),
+            me_bind_stale_mode: bind_stale_mode_label(self.bind_stale_mode()),
+            me_bind_stale_ttl_secs: self.me_bind_stale_ttl_secs.load(Ordering::Relaxed),
+            me_single_endpoint_shadow_writers: self
+                .me_single_endpoint_shadow_writers
+                .load(Ordering::Relaxed),
+            me_single_endpoint_outage_mode_enabled: self
+                .me_single_endpoint_outage_mode_enabled
+                .load(Ordering::Relaxed),
+            me_single_endpoint_outage_disable_quarantine: self
+                .me_single_endpoint_outage_disable_quarantine
+                .load(Ordering::Relaxed),
+            me_single_endpoint_outage_backoff_min_ms: self
+                .me_single_endpoint_outage_backoff_min_ms
+                .load(Ordering::Relaxed),
+            me_single_endpoint_outage_backoff_max_ms: self
+                .me_single_endpoint_outage_backoff_max_ms
+                .load(Ordering::Relaxed),
+            me_single_endpoint_shadow_rotate_every_secs: self
+                .me_single_endpoint_shadow_rotate_every_secs
+                .load(Ordering::Relaxed),
+            me_deterministic_writer_sort: self
+                .me_deterministic_writer_sort
+                .load(Ordering::Relaxed),
+            me_socks_kdf_policy: socks_kdf_policy_label(self.socks_kdf_policy()),
+            quarantined_endpoints,
+            network_path,
+        }
+    }
+}
+
+fn ratio_pct(part: usize, total: usize) -> f64 {
+    if total == 0 {
+        return 0.0;
+    }
+    let pct = ((part as f64) / (total as f64)) * 100.0;
+    pct.clamp(0.0, 100.0)
+}
+
+fn floor_mode_label(mode: MeFloorMode) -> &'static str {
+    match mode {
+        MeFloorMode::Static => "static",
+        MeFloorMode::Adaptive => "adaptive",
+    }
+}
+
+fn bind_stale_mode_label(mode: MeBindStaleMode) -> &'static str {
+    match mode {
+        MeBindStaleMode::Never => "never",
+        MeBindStaleMode::Ttl => "ttl",
+        MeBindStaleMode::Always => "always",
+    }
+}
+
+fn socks_kdf_policy_label(policy: MeSocksKdfPolicy) -> &'static str {
+    match policy {
+        MeSocksKdfPolicy::Strict => "strict",
+        MeSocksKdfPolicy::Compat => "compat",
+    }
+}
+
+fn ip_preference_label(preference: IpPreference) -> &'static str {
+    match preference {
+        IpPreference::Unknown => "unknown",
+        IpPreference::PreferV6 => "prefer_v6",
+        IpPreference::PreferV4 => "prefer_v4",
+        IpPreference::BothWork => "both",
+        IpPreference::Unavailable => "unavailable",
+    }
+}
+
+#[cfg(test)]
+mod tests {
+    use super::ratio_pct;
+
+    #[test]
+    fn ratio_pct_is_zero_when_denominator_is_zero() {
+        assert_eq!(ratio_pct(1, 0), 0.0);
+    }
+
+    #[test]
+    fn ratio_pct_is_capped_at_100() {
+        assert_eq!(ratio_pct(7, 3), 100.0);
+    }
+
+    #[test]
+    fn ratio_pct_reports_expected_value() {
+        assert_eq!(ratio_pct(1, 4), 25.0);
+    }
+}

src/transport/middle_proxy/pool_writer.rs

@@ -0,0 +1,535 @@
+use std::net::SocketAddr;
+use std::sync::Arc;
+use std::sync::atomic::{AtomicBool, AtomicU8, AtomicU64, Ordering};
+use std::time::{Duration, Instant};
+use std::io::ErrorKind;
+
+use bytes::BytesMut;
+use rand::Rng;
+use tokio::sync::mpsc;
+use tokio_util::sync::CancellationToken;
+use tracing::{debug, info, warn};
+
+use crate::config::MeBindStaleMode;
+use crate::crypto::SecureRandom;
+use crate::error::{ProxyError, Result};
+use crate::protocol::constants::{RPC_CLOSE_EXT_U32, RPC_PING_U32};
+
+use super::codec::{RpcWriter, WriterCommand};
+use super::pool::{MePool, MeWriter, WriterContour};
+use super::reader::reader_loop;
+use super::registry::BoundConn;
+use super::wire::build_proxy_req_payload;
+
+const ME_ACTIVE_PING_SECS: u64 = 25;
+const ME_ACTIVE_PING_JITTER_SECS: i64 = 5;
+const ME_IDLE_KEEPALIVE_MAX_SECS: u64 = 5;
+const ME_RPC_PROXY_REQ_RESPONSE_WAIT_MS: u64 = 700;
+
+fn is_me_peer_closed_error(error: &ProxyError) -> bool {
+    matches!(error, ProxyError::Io(ioe) if ioe.kind() == ErrorKind::UnexpectedEof)
+}
+
+impl MePool {
+    pub(crate) async fn prune_closed_writers(self: &Arc<Self>) {
+        let closed_writer_ids: Vec<u64> = {
+            let ws = self.writers.read().await;
+            ws.iter().filter(|w| w.tx.is_closed()).map(|w| w.id).collect()
+        };
+        if closed_writer_ids.is_empty() {
+            return;
+        }
+
+        for writer_id in closed_writer_ids {
+            if self.registry.is_writer_empty(writer_id).await {
+                let _ = self.remove_writer_only(writer_id).await;
+            } else {
+                let _ = self.remove_writer_and_close_clients(writer_id).await;
+            }
+        }
+    }
+
+    pub(crate) async fn connect_one(self: &Arc<Self>, addr: SocketAddr, rng: &SecureRandom) -> Result<()> {
+        self.connect_one_with_generation_contour(
+            addr,
+            rng,
+            self.current_generation(),
+            WriterContour::Active,
+        )
+        .await
+    }
+
+    pub(super) async fn connect_one_with_generation_contour(
+        self: &Arc<Self>,
+        addr: SocketAddr,
+        rng: &SecureRandom,
+        generation: u64,
+        contour: WriterContour,
+    ) -> Result<()> {
+        let secret_len = self.proxy_secret.read().await.secret.len();
+        if secret_len < 32 {
+            return Err(ProxyError::Proxy("proxy-secret too short for ME auth".into()));
+        }
+
+        let (stream, _connect_ms, upstream_egress) = self.connect_tcp(addr).await?;
+        let hs = self.handshake_only(stream, addr, upstream_egress, rng).await?;
+
+        let writer_id = self.next_writer_id.fetch_add(1, Ordering::Relaxed);
+        let contour = Arc::new(AtomicU8::new(contour.as_u8()));
+        let cancel = CancellationToken::new();
+        let degraded = Arc::new(AtomicBool::new(false));
+        let draining = Arc::new(AtomicBool::new(false));
+        let draining_started_at_epoch_secs = Arc::new(AtomicU64::new(0));
+        let allow_drain_fallback = Arc::new(AtomicBool::new(false));
+        let (tx, mut rx) = mpsc::channel::<WriterCommand>(4096);
+        let mut rpc_writer = RpcWriter {
+            writer: hs.wr,
+            key: hs.write_key,
+            iv: hs.write_iv,
+            seq_no: 0,
+            crc_mode: hs.crc_mode,
+        };
+        let cancel_wr = cancel.clone();
+        tokio::spawn(async move {
+            loop {
+                tokio::select! {
+                    cmd = rx.recv() => {
+                        match cmd {
+                            Some(WriterCommand::Data(payload)) => {
+                                if rpc_writer.send(&payload).await.is_err() { break; }
+                            }
+                            Some(WriterCommand::DataAndFlush(payload)) => {
+                                if rpc_writer.send_and_flush(&payload).await.is_err() { break; }
+                            }
+                            Some(WriterCommand::Close) | None => break,
+                        }
+                    }
+                    _ = cancel_wr.cancelled() => break,
+                }
+            }
+        });
+        let writer = MeWriter {
+            id: writer_id,
+            addr,
+            generation,
+            contour: contour.clone(),
+            created_at: Instant::now(),
+            tx: tx.clone(),
+            cancel: cancel.clone(),
+            degraded: degraded.clone(),
+            draining: draining.clone(),
+            draining_started_at_epoch_secs: draining_started_at_epoch_secs.clone(),
+            allow_drain_fallback: allow_drain_fallback.clone(),
+        };
+        self.writers.write().await.push(writer.clone());
+        self.registry.mark_writer_idle(writer_id).await;
+        self.conn_count.fetch_add(1, Ordering::Relaxed);
+        self.writer_available.notify_one();
+
+        let reg = self.registry.clone();
+        let writers_arc = self.writers_arc();
+        let ping_tracker = self.ping_tracker.clone();
+        let ping_tracker_reader = ping_tracker.clone();
+        let rtt_stats = self.rtt_stats.clone();
+        let stats_reader = self.stats.clone();
+        let stats_reader_close = self.stats.clone();
+        let stats_ping = self.stats.clone();
+        let pool = Arc::downgrade(self);
+        let cancel_ping = cancel.clone();
+        let tx_ping = tx.clone();
+        let ping_tracker_ping = ping_tracker.clone();
+        let cleanup_done = Arc::new(AtomicBool::new(false));
+        let cleanup_for_reader = cleanup_done.clone();
+        let cleanup_for_ping = cleanup_done.clone();
+        let keepalive_enabled = self.me_keepalive_enabled;
+        let keepalive_interval = self.me_keepalive_interval;
+        let keepalive_jitter = self.me_keepalive_jitter;
+        let rpc_proxy_req_every_secs = self.rpc_proxy_req_every_secs.load(Ordering::Relaxed);
+        let tx_signal = tx.clone();
+        let stats_signal = self.stats.clone();
+        let cancel_signal = cancel.clone();
+        let cleanup_for_signal = cleanup_done.clone();
+        let pool_signal = Arc::downgrade(self);
+        let keepalive_jitter_signal = self.me_keepalive_jitter;
+        let cancel_reader_token = cancel.clone();
+        let cancel_ping_token = cancel_ping.clone();
+
+        tokio::spawn(async move {
+            let res = reader_loop(
+                hs.rd,
+                hs.read_key,
+                hs.read_iv,
+                hs.crc_mode,
+                reg.clone(),
+                BytesMut::new(),
+                BytesMut::new(),
+                tx.clone(),
+                ping_tracker_reader,
+                rtt_stats.clone(),
+                stats_reader,
+                writer_id,
+                degraded.clone(),
+                cancel_reader_token.clone(),
+            )
+            .await;
+            let idle_close_by_peer = if let Err(e) = res.as_ref() {
+                is_me_peer_closed_error(e) && reg.is_writer_empty(writer_id).await
+            } else {
+                false
+            };
+            if idle_close_by_peer {
+                stats_reader_close.increment_me_idle_close_by_peer_total();
+                info!(writer_id, "ME socket closed by peer on idle writer");
+            }
+            if let Some(pool) = pool.upgrade()
+                && cleanup_for_reader
+                    .compare_exchange(false, true, Ordering::AcqRel, Ordering::Relaxed)
+                    .is_ok()
+            {
+                pool.remove_writer_and_close_clients(writer_id).await;
+            }
+            if let Err(e) = res {
+                if !idle_close_by_peer {
+                    warn!(error = %e, "ME reader ended");
+                }
+            }
+            let mut ws = writers_arc.write().await;
+            ws.retain(|w| w.id != writer_id);
+            info!(remaining = ws.len(), "Dead ME writer removed from pool");
+        });
+
+        let pool_ping = Arc::downgrade(self);
+        tokio::spawn(async move {
+            let mut ping_id: i64 = rand::random::<i64>();
+            let idle_interval_cap = Duration::from_secs(ME_IDLE_KEEPALIVE_MAX_SECS);
+            // Per-writer jittered start to avoid phase sync.
+            let startup_jitter = if keepalive_enabled {
+                let mut interval = keepalive_interval;
+                if let Some(pool) = pool_ping.upgrade() {
+                    if pool.registry.is_writer_empty(writer_id).await {
+                        interval = interval.min(idle_interval_cap);
+                    }
+                } else {
+                    return;
+                }
+                let jitter_cap_ms = interval.as_millis() / 2;
+                let effective_jitter_ms = keepalive_jitter.as_millis().min(jitter_cap_ms).max(1);
+                Duration::from_millis(rand::rng().random_range(0..=effective_jitter_ms as u64))
+            } else {
+                let jitter = rand::rng().random_range(-ME_ACTIVE_PING_JITTER_SECS..=ME_ACTIVE_PING_JITTER_SECS);
+                let wait = (ME_ACTIVE_PING_SECS as i64 + jitter).max(5) as u64;
+                Duration::from_secs(wait)
+            };
+            tokio::select! {
+                _ = cancel_ping_token.cancelled() => return,
+                _ = tokio::time::sleep(startup_jitter) => {}
+            }
+            loop {
+                let wait = if keepalive_enabled {
+                    let mut interval = keepalive_interval;
+                    if let Some(pool) = pool_ping.upgrade() {
+                        if pool.registry.is_writer_empty(writer_id).await {
+                            interval = interval.min(idle_interval_cap);
+                        }
+                    } else {
+                        break;
+                    }
+                    let jitter_cap_ms = interval.as_millis() / 2;
+                    let effective_jitter_ms = keepalive_jitter.as_millis().min(jitter_cap_ms).max(1);
+                    interval + Duration::from_millis(rand::rng().random_range(0..=effective_jitter_ms as u64))
+                } else {
+                    let jitter = rand::rng().random_range(-ME_ACTIVE_PING_JITTER_SECS..=ME_ACTIVE_PING_JITTER_SECS);
+                    let secs = (ME_ACTIVE_PING_SECS as i64 + jitter).max(5) as u64;
+                    Duration::from_secs(secs)
+                };
+                tokio::select! {
+                    _ = cancel_ping_token.cancelled() => {
+                        break;
+                    }
+                    _ = tokio::time::sleep(wait) => {}
+                }
+                let sent_id = ping_id;
+                let mut p = Vec::with_capacity(12);
+                p.extend_from_slice(&RPC_PING_U32.to_le_bytes());
+                p.extend_from_slice(&sent_id.to_le_bytes());
+                {
+                    let mut tracker = ping_tracker_ping.lock().await;
+                    let before = tracker.len();
+                    tracker.retain(|_, (ts, _)| ts.elapsed() < Duration::from_secs(120));
+                    let expired = before.saturating_sub(tracker.len());
+                    if expired > 0 {
+                        stats_ping.increment_me_keepalive_timeout_by(expired as u64);
+                    }
+                    tracker.insert(sent_id, (std::time::Instant::now(), writer_id));
+                }
+                ping_id = ping_id.wrapping_add(1);
+                stats_ping.increment_me_keepalive_sent();
+                if tx_ping.send(WriterCommand::DataAndFlush(p)).await.is_err() {
+                    stats_ping.increment_me_keepalive_failed();
+                    debug!("ME ping failed, removing dead writer");
+                    cancel_ping.cancel();
+                    if let Some(pool) = pool_ping.upgrade()
+                        && cleanup_for_ping
+                            .compare_exchange(false, true, Ordering::AcqRel, Ordering::Relaxed)
+                            .is_ok()
+                    {
+                        pool.remove_writer_and_close_clients(writer_id).await;
+                    }
+                    break;
+                }
+            }
+        });
+
+        tokio::spawn(async move {
+            if rpc_proxy_req_every_secs == 0 {
+                return;
+            }
+
+            let interval = Duration::from_secs(rpc_proxy_req_every_secs);
+            let startup_jitter_ms = {
+                let jitter_cap_ms = interval.as_millis() / 2;
+                let effective_jitter_ms = keepalive_jitter_signal
+                    .as_millis()
+                    .min(jitter_cap_ms)
+                    .max(1);
+                rand::rng().random_range(0..=effective_jitter_ms as u64)
+            };
+
+            tokio::select! {
+                _ = cancel_signal.cancelled() => return,
+                _ = tokio::time::sleep(Duration::from_millis(startup_jitter_ms)) => {}
+            }
+
+            loop {
+                let wait = {
+                    let jitter_cap_ms = interval.as_millis() / 2;
+                    let effective_jitter_ms = keepalive_jitter_signal
+                        .as_millis()
+                        .min(jitter_cap_ms)
+                        .max(1);
+                    interval + Duration::from_millis(rand::rng().random_range(0..=effective_jitter_ms as u64))
+                };
+
+                tokio::select! {
+                    _ = cancel_signal.cancelled() => break,
+                    _ = tokio::time::sleep(wait) => {}
+                }
+
+                let Some(pool) = pool_signal.upgrade() else {
+                    break;
+                };
+
+                let Some(meta) = pool.registry.get_last_writer_meta(writer_id).await else {
+                    stats_signal.increment_me_rpc_proxy_req_signal_skipped_no_meta_total();
+                    continue;
+                };
+
+                let (conn_id, mut service_rx) = pool.registry.register().await;
+                pool.registry
+                    .bind_writer(conn_id, writer_id, tx_signal.clone(), meta.clone())
+                    .await;
+
+                let payload = build_proxy_req_payload(
+                    conn_id,
+                    meta.client_addr,
+                    meta.our_addr,
+                    &[],
+                    pool.proxy_tag.as_deref(),
+                    meta.proto_flags,
+                );
+
+                if tx_signal.send(WriterCommand::DataAndFlush(payload)).await.is_err() {
+                    stats_signal.increment_me_rpc_proxy_req_signal_failed_total();
+                    let _ = pool.registry.unregister(conn_id).await;
+                    cancel_signal.cancel();
+                    if cleanup_for_signal
+                        .compare_exchange(false, true, Ordering::AcqRel, Ordering::Relaxed)
+                        .is_ok()
+                    {
+                        pool.remove_writer_and_close_clients(writer_id).await;
+                    }
+                    break;
+                }
+
+                stats_signal.increment_me_rpc_proxy_req_signal_sent_total();
+
+                if matches!(
+                    tokio::time::timeout(
+                        Duration::from_millis(ME_RPC_PROXY_REQ_RESPONSE_WAIT_MS),
+                        service_rx.recv(),
+                    )
+                    .await,
+                    Ok(Some(_))
+                ) {
+                    stats_signal.increment_me_rpc_proxy_req_signal_response_total();
+                }
+
+                let mut close_payload = Vec::with_capacity(12);
+                close_payload.extend_from_slice(&RPC_CLOSE_EXT_U32.to_le_bytes());
+                close_payload.extend_from_slice(&conn_id.to_le_bytes());
+
+                if tx_signal
+                    .send(WriterCommand::DataAndFlush(close_payload))
+                    .await
+                    .is_err()
+                {
+                    stats_signal.increment_me_rpc_proxy_req_signal_failed_total();
+                    let _ = pool.registry.unregister(conn_id).await;
+                    cancel_signal.cancel();
+                    if cleanup_for_signal
+                        .compare_exchange(false, true, Ordering::AcqRel, Ordering::Relaxed)
+                        .is_ok()
+                    {
+                        pool.remove_writer_and_close_clients(writer_id).await;
+                    }
+                    break;
+                }
+
+                stats_signal.increment_me_rpc_proxy_req_signal_close_sent_total();
+                let _ = pool.registry.unregister(conn_id).await;
+            }
+        });
+
+        Ok(())
+    }
+
+    pub(crate) async fn remove_writer_and_close_clients(self: &Arc<Self>, writer_id: u64) {
+        let conns = self.remove_writer_only(writer_id).await;
+        for bound in conns {
+            let _ = self.registry.route(bound.conn_id, super::MeResponse::Close).await;
+            let _ = self.registry.unregister(bound.conn_id).await;
+        }
+    }
+
+    async fn remove_writer_only(self: &Arc<Self>, writer_id: u64) -> Vec<BoundConn> {
+        let mut close_tx: Option<mpsc::Sender<WriterCommand>> = None;
+        let mut removed_addr: Option<SocketAddr> = None;
+        let mut removed_uptime: Option<Duration> = None;
+        let mut trigger_refill = false;
+        {
+            let mut ws = self.writers.write().await;
+            if let Some(pos) = ws.iter().position(|w| w.id == writer_id) {
+                let w = ws.remove(pos);
+                let was_draining = w.draining.load(Ordering::Relaxed);
+                if was_draining {
+                    self.stats.decrement_pool_drain_active();
+                }
+                self.stats.increment_me_writer_removed_total();
+                w.cancel.cancel();
+                removed_addr = Some(w.addr);
+                removed_uptime = Some(w.created_at.elapsed());
+                trigger_refill = !was_draining;
+                if trigger_refill {
+                    self.stats.increment_me_writer_removed_unexpected_total();
+                }
+                close_tx = Some(w.tx.clone());
+                self.conn_count.fetch_sub(1, Ordering::Relaxed);
+            }
+        }
+        if let Some(tx) = close_tx {
+            let _ = tx.send(WriterCommand::Close).await;
+        }
+        if trigger_refill
+            && let Some(addr) = removed_addr
+        {
+            if let Some(uptime) = removed_uptime {
+                self.maybe_quarantine_flapping_endpoint(addr, uptime).await;
+            }
+            self.trigger_immediate_refill(addr);
+        }
+        self.rtt_stats.lock().await.remove(&writer_id);
+        self.registry.writer_lost(writer_id).await
+    }
+
+    pub(crate) async fn mark_writer_draining_with_timeout(
+        self: &Arc<Self>,
+        writer_id: u64,
+        timeout: Option<Duration>,
+        allow_drain_fallback: bool,
+    ) {
+        let timeout = timeout.filter(|d| !d.is_zero());
+        let found = {
+            let mut ws = self.writers.write().await;
+            if let Some(w) = ws.iter_mut().find(|w| w.id == writer_id) {
+                let already_draining = w.draining.swap(true, Ordering::Relaxed);
+                w.allow_drain_fallback
+                    .store(allow_drain_fallback, Ordering::Relaxed);
+                w.draining_started_at_epoch_secs
+                    .store(Self::now_epoch_secs(), Ordering::Relaxed);
+                if !already_draining {
+                    self.stats.increment_pool_drain_active();
+                }
+                w.contour
+                    .store(WriterContour::Draining.as_u8(), Ordering::Relaxed);
+                w.draining.store(true, Ordering::Relaxed);
+                true
+            } else {
+                false
+            }
+        };
+
+        if !found {
+            return;
+        }
+
+        let timeout_secs = timeout.map(|d| d.as_secs()).unwrap_or(0);
+        debug!(
+            writer_id,
+            timeout_secs,
+            allow_drain_fallback,
+            "ME writer marked draining"
+        );
+
+        let pool = Arc::downgrade(self);
+        tokio::spawn(async move {
+            let deadline = timeout.map(|t| Instant::now() + t);
+            while let Some(p) = pool.upgrade() {
+                if let Some(deadline_at) = deadline
+                    && Instant::now() >= deadline_at
+                {
+                    warn!(writer_id, "Drain timeout, force-closing");
+                    p.stats.increment_pool_force_close_total();
+                    let _ = p.remove_writer_and_close_clients(writer_id).await;
+                    break;
+                }
+                if p.registry.is_writer_empty(writer_id).await {
+                    let _ = p.remove_writer_only(writer_id).await;
+                    break;
+                }
+                tokio::time::sleep(Duration::from_secs(1)).await;
+            }
+        });
+    }
+
+    pub(crate) async fn mark_writer_draining(self: &Arc<Self>, writer_id: u64) {
+        self.mark_writer_draining_with_timeout(writer_id, Some(Duration::from_secs(300)), false)
+            .await;
+    }
+
+    pub(super) fn writer_accepts_new_binding(&self, writer: &MeWriter) -> bool {
+        if !writer.draining.load(Ordering::Relaxed) {
+            return true;
+        }
+        if !writer.allow_drain_fallback.load(Ordering::Relaxed) {
+            return false;
+        }
+
+        match self.bind_stale_mode() {
+            MeBindStaleMode::Never => false,
+            MeBindStaleMode::Always => true,
+            MeBindStaleMode::Ttl => {
+                let ttl_secs = self.me_bind_stale_ttl_secs.load(Ordering::Relaxed);
+                if ttl_secs == 0 {
+                    return true;
+                }
+
+                let started = writer.draining_started_at_epoch_secs.load(Ordering::Relaxed);
+                if started == 0 {
+                    return false;
+                }
+
+                Self::now_epoch_secs().saturating_sub(started) <= ttl_secs
+            }
+        }
+    }
+}

src/transport/middle_proxy/reader.rs

@@ -0,0 +1,226 @@
+use std::collections::HashMap;
+use std::io::ErrorKind;
+use std::sync::Arc;
+use std::sync::atomic::{AtomicBool, Ordering};
+use std::time::Instant;
+
+use bytes::{Bytes, BytesMut};
+use tokio::io::AsyncReadExt;
+use tokio::net::TcpStream;
+use tokio::sync::{Mutex, mpsc};
+use tokio_util::sync::CancellationToken;
+use tracing::{debug, trace, warn};
+
+use crate::crypto::AesCbc;
+use crate::error::{ProxyError, Result};
+use crate::protocol::constants::*;
+use crate::stats::Stats;
+
+use super::codec::{RpcChecksumMode, WriterCommand, rpc_crc};
+use super::registry::RouteResult;
+use super::{ConnRegistry, MeResponse};
+
+pub(crate) async fn reader_loop(
+    mut rd: tokio::io::ReadHalf<TcpStream>,
+    dk: [u8; 32],
+    mut div: [u8; 16],
+    crc_mode: RpcChecksumMode,
+    reg: Arc<ConnRegistry>,
+    enc_leftover: BytesMut,
+    mut dec: BytesMut,
+    tx: mpsc::Sender<WriterCommand>,
+    ping_tracker: Arc<Mutex<HashMap<i64, (Instant, u64)>>>,
+    rtt_stats: Arc<Mutex<HashMap<u64, (f64, f64)>>>,
+    stats: Arc<Stats>,
+    _writer_id: u64,
+    degraded: Arc<AtomicBool>,
+    cancel: CancellationToken,
+) -> Result<()> {
+    let mut raw = enc_leftover;
+    let mut expected_seq: i32 = 0;
+
+    loop {
+        let mut tmp = [0u8; 16_384];
+        let n = tokio::select! {
+            res = rd.read(&mut tmp) => res.map_err(ProxyError::Io)?,
+            _ = cancel.cancelled() => return Ok(()),
+        };
+        if n == 0 {
+            stats.increment_me_reader_eof_total();
+            return Err(ProxyError::Io(std::io::Error::new(
+                ErrorKind::UnexpectedEof,
+                "ME socket closed by peer",
+            )));
+        }
+        raw.extend_from_slice(&tmp[..n]);
+
+        let blocks = raw.len() / 16 * 16;
+        if blocks > 0 {
+            let mut new_iv = [0u8; 16];
+            new_iv.copy_from_slice(&raw[blocks - 16..blocks]);
+
+            let mut chunk = vec![0u8; blocks];
+            chunk.copy_from_slice(&raw[..blocks]);
+            AesCbc::new(dk, div)
+                .decrypt_in_place(&mut chunk)
+                .map_err(|e| ProxyError::Crypto(format!("{e}")))?;
+            div = new_iv;
+            dec.extend_from_slice(&chunk);
+            let _ = raw.split_to(blocks);
+        }
+
+        while dec.len() >= 12 {
+            let fl = u32::from_le_bytes(dec[0..4].try_into().unwrap()) as usize;
+            if fl == 4 {
+                let _ = dec.split_to(4);
+                continue;
+            }
+            if !(12..=(1 << 24)).contains(&fl) {
+                warn!(frame_len = fl, "Invalid RPC frame len");
+                dec.clear();
+                break;
+            }
+            if dec.len() < fl {
+                break;
+            }
+
+            let frame = dec.split_to(fl);
+            let pe = fl - 4;
+            let ec = u32::from_le_bytes(frame[pe..pe + 4].try_into().unwrap());
+            let actual_crc = rpc_crc(crc_mode, &frame[..pe]);
+            if actual_crc != ec {
+                stats.increment_me_crc_mismatch();
+                warn!(
+                    frame_len = fl,
+                    expected_crc = format_args!("0x{ec:08x}"),
+                    actual_crc = format_args!("0x{actual_crc:08x}"),
+                    "CRC mismatch — CBC crypto desync, aborting ME connection"
+                );
+                return Err(ProxyError::Proxy("CRC mismatch (crypto desync)".into()));
+            }
+
+            let seq_no = i32::from_le_bytes(frame[4..8].try_into().unwrap());
+            if seq_no != expected_seq {
+                stats.increment_me_seq_mismatch();
+                warn!(seq_no, expected = expected_seq, "ME RPC seq mismatch");
+                return Err(ProxyError::SeqNoMismatch {
+                    expected: expected_seq,
+                    got: seq_no,
+                });
+            }
+            expected_seq = expected_seq.wrapping_add(1);
+
+            let payload = &frame[8..pe];
+            if payload.len() < 4 {
+                continue;
+            }
+
+            let pt = u32::from_le_bytes(payload[0..4].try_into().unwrap());
+            let body = &payload[4..];
+
+            if pt == RPC_PROXY_ANS_U32 && body.len() >= 12 {
+                let flags = u32::from_le_bytes(body[0..4].try_into().unwrap());
+                let cid = u64::from_le_bytes(body[4..12].try_into().unwrap());
+                let data = Bytes::copy_from_slice(&body[12..]);
+                trace!(cid, flags, len = data.len(), "RPC_PROXY_ANS");
+
+                let routed = reg.route(cid, MeResponse::Data { flags, data }).await;
+                if !matches!(routed, RouteResult::Routed) {
+                    match routed {
+                        RouteResult::NoConn => stats.increment_me_route_drop_no_conn(),
+                        RouteResult::ChannelClosed => stats.increment_me_route_drop_channel_closed(),
+                        RouteResult::QueueFullBase => {
+                            stats.increment_me_route_drop_queue_full();
+                            stats.increment_me_route_drop_queue_full_base();
+                        }
+                        RouteResult::QueueFullHigh => {
+                            stats.increment_me_route_drop_queue_full();
+                            stats.increment_me_route_drop_queue_full_high();
+                        }
+                        RouteResult::Routed => {}
+                    }
+                    reg.unregister(cid).await;
+                    send_close_conn(&tx, cid).await;
+                }
+            } else if pt == RPC_SIMPLE_ACK_U32 && body.len() >= 12 {
+                let cid = u64::from_le_bytes(body[0..8].try_into().unwrap());
+                let cfm = u32::from_le_bytes(body[8..12].try_into().unwrap());
+                trace!(cid, cfm, "RPC_SIMPLE_ACK");
+
+                let routed = reg.route(cid, MeResponse::Ack(cfm)).await;
+                if !matches!(routed, RouteResult::Routed) {
+                    match routed {
+                        RouteResult::NoConn => stats.increment_me_route_drop_no_conn(),
+                        RouteResult::ChannelClosed => stats.increment_me_route_drop_channel_closed(),
+                        RouteResult::QueueFullBase => {
+                            stats.increment_me_route_drop_queue_full();
+                            stats.increment_me_route_drop_queue_full_base();
+                        }
+                        RouteResult::QueueFullHigh => {
+                            stats.increment_me_route_drop_queue_full();
+                            stats.increment_me_route_drop_queue_full_high();
+                        }
+                        RouteResult::Routed => {}
+                    }
+                    reg.unregister(cid).await;
+                    send_close_conn(&tx, cid).await;
+                }
+            } else if pt == RPC_CLOSE_EXT_U32 && body.len() >= 8 {
+                let cid = u64::from_le_bytes(body[0..8].try_into().unwrap());
+                debug!(cid, "RPC_CLOSE_EXT from ME");
+                reg.route(cid, MeResponse::Close).await;
+                reg.unregister(cid).await;
+            } else if pt == RPC_CLOSE_CONN_U32 && body.len() >= 8 {
+                let cid = u64::from_le_bytes(body[0..8].try_into().unwrap());
+                debug!(cid, "RPC_CLOSE_CONN from ME");
+                reg.route(cid, MeResponse::Close).await;
+                reg.unregister(cid).await;
+            } else if pt == RPC_PING_U32 && body.len() >= 8 {
+                let ping_id = i64::from_le_bytes(body[0..8].try_into().unwrap());
+                trace!(ping_id, "RPC_PING -> RPC_PONG");
+                let mut pong = Vec::with_capacity(12);
+                pong.extend_from_slice(&RPC_PONG_U32.to_le_bytes());
+                pong.extend_from_slice(&ping_id.to_le_bytes());
+                if tx.send(WriterCommand::DataAndFlush(pong)).await.is_err() {
+                    warn!("PONG send failed");
+                    break;
+                }
+            } else if pt == RPC_PONG_U32 && body.len() >= 8 {
+                let ping_id = i64::from_le_bytes(body[0..8].try_into().unwrap());
+                stats.increment_me_keepalive_pong();
+                if let Some((sent, wid)) = {
+                    let mut guard = ping_tracker.lock().await;
+                    guard.remove(&ping_id)
+                } {
+                    let rtt = sent.elapsed().as_secs_f64() * 1000.0;
+                    let mut stats = rtt_stats.lock().await;
+                    let entry = stats.entry(wid).or_insert((rtt, rtt));
+                    entry.1 = entry.1 * 0.8 + rtt * 0.2;
+                    if rtt < entry.0 {
+                        entry.0 = rtt;
+                    } else {
+                        // allow slow baseline drift upward to avoid stale minimum
+                        entry.0 = entry.0 * 0.99 + rtt * 0.01;
+                    }
+                    let degraded_now = entry.1 > entry.0 * 2.0;
+                    degraded.store(degraded_now, Ordering::Relaxed);
+                    trace!(writer_id = wid, rtt_ms = rtt, ema_ms = entry.1, base_ms = entry.0, degraded = degraded_now, "ME RTT sample");
+                }
+            } else {
+                debug!(
+                    rpc_type = format_args!("0x{pt:08x}"),
+                    len = body.len(),
+                    "Unknown RPC"
+                );
+            }
+        }
+    }
+}
+
+async fn send_close_conn(tx: &mpsc::Sender<WriterCommand>, conn_id: u64) {
+    let mut p = Vec::with_capacity(12);
+    p.extend_from_slice(&RPC_CLOSE_CONN_U32.to_le_bytes());
+    p.extend_from_slice(&conn_id.to_le_bytes());
+
+    let _ = tx.send(WriterCommand::DataAndFlush(p)).await;
+}