fix(iam): persist and propagate scoped STS revocation

Enumerate durable STS identities for parent/type revocation, retain the
operation cutoff, and notify sibling caches for each committed deletion.
Repair missed notifications during periodic object/etcd IAM refresh.

Carry immutable STS tombstones and expiry through the existing IAM revision
protocol and deletion healing. Delayed creation cannot undo revocation,
while unrelated and newly issued keys stay valid. Keep partial failures
visible while propagating the successfully committed deletions.

Cover uncached credentials, warm sibling caches, dropped notifications,
object/etcd backends, reordered replay, concurrent issuance and partial
persistence failure. No new dependencies or authentication-path I/O.

Signed-off-by: Feng Ruohang <rh@vonng.com>

.github/workflows/go.yml

@@ -93,8 +93,16 @@ jobs:
       - name: Run conditional PUT tests under race detector
         run: go test -race ./cmd -run '^Test(PoolsConditionalPut|SinglePoolConditionalPutHTTP)' -count=1 -timeout=5m
 
-      - name: Run multipart listing and cancellation tests under race detector
+      - name: Run multipart listing, cancellation and cleanup tests under race detector
-        run: go test -race ./cmd -run '^Test(MultipartListing|MultipartAbort|PaginateMultipartUploads|ListMultipartUploads)' -count=1 -timeout=5m
+        env:
+          MINIO_API_REQUESTS_MAX: "10000"
+        run: go test -race ./cmd -run '^Test(Multipart|PaginateMultipartUploads|ListMultipartUploads|CleanupStaleUploads|ExtractModTime)' -count=1 -timeout=5m
+
+      - name: Run listing metadata recovery, pagination and cancellation tests under race detector
+        run: go test -race ./cmd -run '^Test(ListFallback|ListingCache|ListingMerge|ListingRecovery|ListingPending|ListingEntry|ListingSecondLock|StreamMetadataParts|LocalLockerListing|DistributedListing|MergeXLV2LatestNullCrossParity|CountXLV2MixedECGroups|ListCrossParity)' -count=1 -timeout=5m
+
+      - name: Run lock cancellation and late-grant cleanup tests under race detector
+        run: go test -race ./internal/dsync ./internal/lsync -run '^Test(DRWMutex|LRWMutex)' -count=1 -timeout=5m
 
       - name: Run CPU metrics tests under race detector
         run: go test -race ./cmd -run '^TestLoadCPUMetrics' -count=1 -timeout=5m

.github/workflows/release.yml

@@ -140,7 +140,7 @@ jobs:
       - name: Build Draft release with GoReleaser
         uses: goreleaser/goreleaser-action@f06c13b6b1a9625abc9e6e439d9c05a8f2190e94 # v7.2.3
         with:
-          version: v2.18.1
+          version: v2.18.3
           args: release --clean --skip=validate --config .github/goreleaser.yml
         env:
           GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}
@@ -156,7 +156,7 @@ jobs:
       - name: Install nFPM
         run: |
           set -euo pipefail
-          go install github.com/goreleaser/nfpm/v2/cmd/nfpm@v2.47.0
+          go install github.com/goreleaser/nfpm/v2/cmd/nfpm@v2.48.0
           echo "$(go env GOPATH)/bin" >> "${GITHUB_PATH}"
 
       - name: Build nFPM packages

.github/workflows/test-release.yml

@@ -97,7 +97,7 @@ jobs:
       - name: GoReleaser config check
         uses: goreleaser/goreleaser-action@f06c13b6b1a9625abc9e6e439d9c05a8f2190e94 # v7.2.3
         with:
-          version: v2.18.1
+          version: v2.18.3
           args: check --config .github/goreleaser.yml
 
       - name: Validate Helm chart and legacy upgrade identity
@@ -111,7 +111,7 @@ jobs:
       - name: Build snapshot artifacts
         uses: goreleaser/goreleaser-action@f06c13b6b1a9625abc9e6e439d9c05a8f2190e94 # v7.2.3
         with:
-          version: v2.18.1
+          version: v2.18.3
           # A pull-request snapshot has no trusted release identity. Exercise
           # the SBOM/checksum pipeline here, and reserve keyless signing for
           # the tag-triggered release workflow with GitHub OIDC.
@@ -143,7 +143,7 @@ jobs:
       - name: Install package validation tools
         run: |
           set -euo pipefail
-          go install github.com/goreleaser/nfpm/v2/cmd/nfpm@v2.47.0
+          go install github.com/goreleaser/nfpm/v2/cmd/nfpm@v2.48.0
           echo "$(go env GOPATH)/bin" >> "${GITHUB_PATH}"
           sudo apt-get update
           sudo apt-get install --yes rpm binutils

.github/workflows/vulncheck.yml

@@ -29,7 +29,7 @@ jobs:
 
       - name: Install govulncheck
         run: |
-          go install golang.org/x/vuln/cmd/govulncheck@v1.8.0
+          go install golang.org/x/vuln/cmd/govulncheck@v1.9.0
           echo "$(go env GOPATH)/bin" >> "${GITHUB_PATH}"
 
       - name: Run govulncheck

CHANGELOG.md

@@ -2,12 +2,34 @@
 
 ## Unreleased
 
-Preparation target: `RELEASE.2026-09-16T00-00-00Z` (package version
+Preparation target: `RELEASE.2026-10-18T00-00-00Z` (package version
-`20260916000000.0.0`). The entries below describe the candidate changes since
+`20261018000000.0.0`). **This is a tag and Draft preparation; it is not a public
-the latest published Server.
+Server release.** The latest published Server remains 20260916. Server 20261015
-**The latest published Server remains 20260903.** These changes are not in its
+also remains a Draft. Existing binaries, packages and images do not change.
-binaries, packages or images. See the [component matrix](https://silo.pgsty.com/compatibility/versions/)
+See the [component matrix](https://silo.pgsty.com/compatibility/versions/) and
-and [complete commit range](https://github.com/pgsty/silo/compare/RELEASE.2026-09-03T13-18-01Z...main).
+[changes since the public release](https://github.com/pgsty/silo/compare/RELEASE.2026-09-16T00-00-00Z...main).
+
+### October 18 dependency update
+
+- Select silo-pkg v3.14.3, upstream minio-go v7.4.1, published mcli 20261018
+  and Console v2.4.2 source (`ef2dd2c41d21`). The SDK includes cancellation cleanup, UTC retention
+  timestamps and preserved STS duration; parallel GET/Compose remain opt-in.
+- Adopt silo-pkg's faster admin-action classification during policy evaluation,
+  preserving authorization behavior and compatibility with existing clients.
+  Update the standalone `reorder-disks` debugging module to the same package version.
+- Build with Go 1.27.2 and refresh Go x/* modules. Apply go-jose v4.1.6,
+  LDAP v3.4.15, compress v1.20.3, msgp v1.6.5, etcd v3.7.2, fwd v1.2.1 and
+  cpuid v2.4.1 bug/security fixes. OIDC JWK RSA keys must not exceed 4096 bits.
+- Update the bundled mcli archive hashes, installation helper and Helm client
+  image to the published 20261018 release. Server Helm image defaults retain the
+  public 20260916 version until a Server image is published.
+- Refresh pinned UBI 9 images and use GoReleaser v2.18.3, nFPM v2.48.0,
+  golangci-lint v2.15.0 and govulncheck v1.9.0. Keep the NetBSD go-systemd
+  v22.6.0 compatibility replacement.
+
+The sections below retain the cumulative maintenance history, including changes
+already shipped in 20260916; the source comparison above identifies this
+candidate's exact delta.
 
 ### Authorization and security
 
@@ -54,35 +76,26 @@ and [complete commit range](https://github.com/pgsty/silo/compare/RELEASE.2026-0
 
 ### Object storage and replication
 
-- Make `ListMultipartUploads` discover quorum-valid uploads from durable state
+- Withdraw default persistent `ListMultipartUploads` scanning (#79). Restore
-  across pools, erasure sets and drives, then apply S3 prefix, delimiter,
+  node-local cache listing for an empty prefix and exact-key lookup otherwise;
-  marker, ordering and 1,000-entry pagination semantics globally (#198). New uploads
+  restart omissions, true-prefix and cache-pagination limitations remain.
-  store their canonical bucket and key as reserved fields in the existing
+  Preserve bucket/key identity metadata, path safety and stale-upload cleanup.
-  quorum-written `xl.meta`; completion removes those upload-only fields. Native
+  `MINIO_API_MULTIPART_LISTING` stays retired and ignored with a startup warning;
-  markers remain usable after their upload is completed or canceled. Strict
+  obsolete shared `api multipart_listing` keys remain tolerated. **Existing
-  listing returns a diagnostic 503 for legacy uploads or uncertain coverage;
+  opt-in strict users lose durable listing and strict cancellation.** The
-  the default remains the released exact-key/cache-based `legacy` behavior.
+  read-only admin `multipart-preflight` reports `mode=legacy`; its scan budgets
-  Opt into strict mode only through `MINIO_API_MULTIPART_LISTING=strict`, after
+  and readiness do not gate public listing or startup. Durable, complete,
-  upgrading every writer, draining old uploads, checking the read-only admin
+  scalable enumeration remains deferred. See the
-  `multipart-preflight` report and validating scan capacity. The process-only
+  [design and limits](https://silo.pgsty.com/blog/design/list-multipart-uploads/#implementation).
-  setting is not persisted into shared API configuration. Historical
+  Thanks to mr javad seydi (@mrjavadseydi) for the original metadata-and-scan
-  `multipart_listing` keys are ignored and can be removed with a targeted
+  implementation (#198).
-  `mcli admin config reset ALIAS api multipart_listing` before rollback. Per-process
+- Keep the default read-quorum, best-effort multipart cancellation and
-  admission, directory-entry, worker and time budgets bound scan scheduling;
+  first-matching-pool behavior when retiring the listing switch. The optional
-  each page still scans durable state. See [issue #79](https://github.com/pgsty/silo/issues/79)
+  strict cancellation branch is removed. Wrong-target protection, upload-ID
-  and its [design record](https://silo.pgsty.com/blog/design/list-multipart-uploads/).
+  path validation and peer notification with the request context remain.
-  Thanks to mr javad seydi (@mrjavadseydi) for the original implementation.
+  HTTP 204 for an absent upload remains idempotent and does not prove physical
-- Retain released read-quorum and best-effort multipart cancellation in default
+  cleanup. **Known boundary:** delayed creation writes can still restore a
-  legacy mode. Strict mode requires majority deletion acknowledgements and
+  readable upload after cancellation; this change adds no creation fence.
-  permits retries below read quorum. When most drives were already empty,
-  failed deletion of an observed remnant now returns 503 instead of being
-  masked by empty-drive successes. Wrong-key or wrong-bucket cancellation
-  preserves the valid upload's cache entry; successful cancellation notifies
-  peers with the request context after the distributed lock is released.
-  The HTTP response for an absent
-  upload remains 204; this is not proof of physical cleanup. **Known boundary:**
-  delayed creation writes can still restore an upload after cancellation;
-  this change does not add a durable creation fence.
 - Preserve object tags during multi-pool metadata reconciliation by reading the
   resolved tag field together with its revision (#189). Previously, reconciliation
   could replace existing tags with an empty value.

Dockerfile.goreleaser

@@ -17,9 +17,9 @@ ENV GOPATH=/go
 ENV CGO_ENABLED=0
 
 ARG MC_REPO=pgsty/mc
-ARG MC_VERSION=RELEASE.2026-09-16T00-00-00Z
+ARG MC_VERSION=RELEASE.2026-10-18T00-00-00Z
-ARG MC_AMD64_SHA256=4ba2814fd5507fbe6b4d237c359750b9119d28d7217495fa5b48002fcbd397ef
+ARG MC_AMD64_SHA256=b5392a642dbc9685b517ee901f032d1d5e88f41704436f0f08dde6060b577db3
-ARG MC_ARM64_SHA256=b7008ca2a1bc5735b6585981c59a3640a0daa152dc789df3d1a0d0438787de82
+ARG MC_ARM64_SHA256=17f3eb694ed34e155b331b1215832140749656d8d4b60e2e3a4d0573b8f320d7
 
 RUN apk add -U --no-cache \
     ca-certificates \
@@ -67,14 +67,14 @@ RUN apk add -U --no-cache \
     chmod +x /go/bin/mcli && \
     ln -sf mcli /go/bin/mc
 
-FROM registry.access.redhat.com/ubi9/ubi:latest@sha256:206b65b8ee0f04b992818c9a51b29081b14974630d4850bc358097d0c44ea156 AS certs
+FROM registry.access.redhat.com/ubi9/ubi:latest@sha256:16e9e43d99e181463513cc84a514859b1abb9460f1c4d9e208da4b4e51cd5cc3 AS certs
 RUN dnf -y install ca-certificates && \
     update-ca-trust && \
     cp /etc/pki/ca-trust/extracted/pem/tls-ca-bundle.pem /tmp/ca-certificates.crt && \
     dnf clean all && \
     rm -rf /var/cache/dnf
 
-FROM registry.access.redhat.com/ubi9/ubi-micro:latest@sha256:f332c99eb8f798a8486821c91937f10ad64ee83d7e739303be2df051040918f6
+FROM registry.access.redhat.com/ubi9/ubi-micro:latest@sha256:5be877a2a280ae36eb9ab80ca4efb28bbf77c57abbba38137cdc37a03b8465e6
 
 LABEL org.opencontainers.image.title="Silo" \
       org.opencontainers.image.description="S3-Interface Libre Object Storage" \

Makefile

@@ -4,7 +4,7 @@ LDFLAGS := $(shell go run buildscripts/gen-ldflags.go)
 
 GOOS ?= $(shell go env GOOS)
 GOARCH ?= $(shell go env GOARCH)
-GOLANGCI_VERSION ?= v2.14.0
+GOLANGCI_VERSION ?= v2.15.0
 
 VERSION ?= $(shell git describe --tags)
 REPO ?= docker.io/pgsty

buildscripts/install-mcli.sh

@@ -40,7 +40,7 @@ if [ -n "${MCLI_BIN:-}" ]; then
 	exit 0
 fi
 
-release=${MCLI_RELEASE:-RELEASE.2026-09-16T00-00-00Z}
+release=${MCLI_RELEASE:-RELEASE.2026-10-18T00-00-00Z}
 version_hyphen=${release#RELEASE.}
 package_version=$(printf '%s\n' "${version_hyphen}" | sed -E 's/^([0-9]{4})-([0-9]{2})-([0-9]{2})T([0-9]{2})-([0-9]{2})-([0-9]{2})Z$/\1\2\3\4\5\6.0.0/')
 if [ "${package_version}" = "${version_hyphen}" ]; then

buildscripts/verify-rebrand.sh

@@ -11,6 +11,33 @@ fail() {
 	exit 1
 }
 
+# Git is already a build prerequisite. --no-index also scans untracked files
+# and exported source trees. As before, honor .gitignore only in a worktree.
+search_ignore=--no-exclude-standard
+if git rev-parse --show-toplevel >/dev/null 2>&1; then
+	search_ignore=--exclude-standard
+fi
+search_errors="$(mktemp)"
+trap 'rm -f "${search_errors}"' EXIT
+
+search_text() {
+	local status=0
+	git --no-pager grep --no-index "${search_ignore}" --no-color -nIHE "$@" 2>"${search_errors}" || status=$?
+	# Git can return 1 for both no matches and a file read error. Check stderr
+	# as well, before callers filter matches through their allowlists.
+	if [ "${status}" -gt 1 ] || [ -s "${search_errors}" ]; then
+		cat "${search_errors}" >&2
+		fail "could not scan source files"
+	fi
+}
+
+# Keep hidden files and build output out of the repository-wide scans.
+source_excludes=(
+	':(exclude,glob)**/.*'
+	':(exclude,glob)**/.*/**'
+	':(exclude)dist/**'
+)
+
 require_file() {
 	[ -f "$1" ] || fail "missing required file: $1"
 }
@@ -152,7 +179,11 @@ if grep -Ev '^[[:space:]]*(#|$)' silo.env | grep -q '='; then
 	fail "silo.env must not contain active assignments that shadow /etc/default/minio"
 fi
 
-if rg -n 'pgsty/minio:' .github/workflows Dockerfile.goreleaser helm/silo; then
+delivery_hits="$(search_text 'pgsty/minio:' -- .github/workflows Dockerfile.goreleaser helm/silo \
+	':(exclude,glob).github/workflows/**/.*' ':(exclude,glob).github/workflows/**/.*/**' \
+	':(exclude,glob)helm/silo/**/.*' ':(exclude,glob)helm/silo/**/.*/**')"
+if [ -n "${delivery_hits}" ]; then
+	printf '%s\n' "${delivery_hits}" >&2
 	fail "an active delivery surface still publishes the frozen pgsty/minio image"
 fi
 
@@ -165,10 +196,10 @@ fi
 # So two rules. First, no live URL may resolve to the old repository anywhere,
 # READMEs and CONTRIBUTORS.md included.
 old_repo_pattern='pgsty/minio(\.git)?([^[:alnum:]_.-]|$)'
-stale_repo_url="$(rg -n -e "github\.com/${old_repo_pattern}" -e "hub\.docker\.com/r/${old_repo_pattern}" \
+stale_repo_url="$(search_text -e "github\.com/${old_repo_pattern}" -e "hub\.docker\.com/r/${old_repo_pattern}" -- . \
-	--glob '!.git/**' --glob '!dist/**' \
+	"${source_excludes[@]}" ':(exclude)SILO_REBRANDING_MIGRATION.md' \
-	--glob '!SILO_REBRANDING_MIGRATION.md' \
+	':(exclude)buildscripts/rebrand-guard/compat-baseline.json')"
-	--glob '!buildscripts/rebrand-guard/compat-baseline.json' . |
+stale_repo_url="$(printf '%s\n' "${stale_repo_url}" |
 	sed 's#^\./##' | grep -v '^buildscripts/verify-rebrand\.sh:' || true)"
 if [ -n "${stale_repo_url}" ]; then
 	printf '%s\n' "${stale_repo_url}" >&2
@@ -180,32 +211,32 @@ fi
 # legacy image, the two READMEs that document the rename and the archived
 # minio branch, and historical issue titles in CONTRIBUTORS.md.
 repo_guard_allowlist='^(buildscripts/minio-upgrade\.sh|buildscripts/verify-rebrand\.sh|buildscripts/helm-migration-guard/main\.go|README\.md|README_ZH\.md|CONTRIBUTORS\.md):'
-stale_repo="$(rg -n "${old_repo_pattern}" --glob '!.git/**' --glob '!dist/**' \
+stale_repo="$(search_text "${old_repo_pattern}" -- . "${source_excludes[@]}" \
-	--glob '!SILO_REBRANDING_MIGRATION.md' \
+	':(exclude)SILO_REBRANDING_MIGRATION.md' ':(exclude)buildscripts/rebrand-guard/compat-baseline.json')"
-	--glob '!buildscripts/rebrand-guard/compat-baseline.json' . |
+stale_repo="$(printf '%s\n' "${stale_repo}" |
 	sed 's#^\./##' | grep -Ev "${repo_guard_allowlist}" || true)"
 if [ -n "${stale_repo}" ]; then
 	printf '%s\n' "${stale_repo}" >&2
 	fail "a source reference still names the pre-rename pgsty/minio repository"
 fi
 
-stale_branch="$(rg -n 'pgsty/silo/(blob/|tree/|raw/)?master' \
+stale_branch="$(search_text 'pgsty/silo/(blob/|tree/|raw/)?master' -- . "${source_excludes[@]}")"
-	--glob '!.git/**' --glob '!dist/**' . || true)"
 if [ -n "${stale_branch}" ]; then
 	printf '%s\n' "${stale_branch}" >&2
 	fail "a link still targets the retired master branch; raw and Actions URLs do not follow a branch rename"
 fi
 
 for workflow in .github/workflows/go.yml .github/workflows/vulncheck.yml; do
-	if rg -q '^\s+- master$' "${workflow}"; then
+	stale_filter="$(search_text '^[[:space:]]+- master$' -- "${workflow}")"
+	if [ -n "${stale_filter}" ]; then
 		fail "${workflow} still filters on master and would go silently dormant on main"
 	fi
 	require_text "${workflow}" "      - main"
 done
 
-network_hits="$(rg -n --glob '*.go' --glob '!**/*_test.go' \
+network_hits="$(search_text \
-	'https?://[^"`[:space:]]*(dl\.min\.io|subnet\.min\.io|api\.min\.io|slack\.min\.io|play\.min\.io)' \
+	'https?://[^"`[:space:]]*(dl\.min\.io|subnet\.min\.io|api\.min\.io|slack\.min\.io|play\.min\.io)' -- \
-	cmd internal || true)"
+	':(glob)cmd/**/*.go' ':(glob)internal/**/*.go' ':(exclude,glob)**/*_test.go' "${source_excludes[@]}")"
 network_hits="$(printf '%s\n' "${network_hits}" | grep -Ev '^[^:]+:[0-9]+:[[:space:]]*//' || true)"
 if [ -n "${network_hits}" ]; then
 	printf '%s\n' "${network_hits}" >&2

cmd/admin-handlers-users-authz_test.go

@@ -11,6 +11,7 @@ import (
 	"io"
 	"net/http"
 	"net/http/httptest"
+	"net/url"
 	"strings"
 	"sync/atomic"
 	"testing"
@@ -248,8 +249,20 @@ func TestIAMAdminPolicylessSelf(t *testing.T) {
 	ctx, sys := iamAdminFixture(t)
 	caller := iamAdminUser(ctx, t, sys, "review-policyless", "")
 	svc := iamAdminServiceAccount(ctx, t, sys, caller.AccessKey, "rv-policyless-child", "")
+	other := iamAdminUser(ctx, t, sys, "bulk-other", "readwrite")
+	_ = iamAdminServiceAccount(ctx, t, sys, other.AccessKey, "bulk-other-child", "")
 	api := adminAPIHandlers{}
 	assertIAMAdminStatus(t, "list", iamAdminCall(t, caller, "GET", "/minio/admin/v3/list-service-accounts", nil, nil, api.ListServiceAccounts), 200)
+	rec := iamAdminCall(t, caller, "GET", "/minio/admin/v3/list-access-keys-bulk?listType=all", nil, nil, api.ListAccessKeysBulk)
+	assertIAMAdminStatus(t, "list bulk self", rec, 200)
+	data, err := madmin.DecryptData(caller.SecretKey, rec.Body)
+	mustIAM(t, err)
+	var listed map[string]madmin.ListAccessKeysResp
+	mustIAM(t, json.Unmarshal(data, &listed))
+	entry, ok := listed[caller.AccessKey]
+	if len(listed) != 1 || !ok || len(entry.ServiceAccounts) != 1 || entry.ServiceAccounts[0].AccessKey != svc.AccessKey || len(entry.STSKeys) != 0 {
+		t.Fatalf("policyless bulk self returned unexpected identities: %+v", listed)
+	}
 	assertIAMAdminStatus(t, "info service", iamAdminCall(t, caller, "GET", "/minio/admin/v3/info-service-account?accessKey="+svc.AccessKey, nil, map[string]string{"accessKey": svc.AccessKey}, api.InfoServiceAccount), 200)
 	assertIAMAdminStatus(t, "info access key", iamAdminCall(t, caller, "GET", "/minio/admin/v3/info-access-key?accessKey="+svc.AccessKey, nil, map[string]string{"accessKey": svc.AccessKey}, api.InfoAccessKey), 200)
 	assertIAMAdminStatus(t, "delete", iamAdminCall(t, caller, "DELETE", "/minio/admin/v3/delete-service-account?accessKey="+svc.AccessKey, nil, map[string]string{"accessKey": svc.AccessKey}, api.DeleteServiceAccount), 204)
@@ -384,6 +397,7 @@ func TestIAMAdminSelfDenials(t *testing.T) {
 			}
 			api := adminAPIHandlers{}
 			assertIAMAdminStatus(t, "list", iamAdminCall(t, caller, "GET", "/minio/admin/v3/list-service-accounts", nil, nil, api.ListServiceAccounts), 403)
+			assertIAMAdminStatus(t, "list bulk self", iamAdminCall(t, caller, "GET", "/minio/admin/v3/list-access-keys-bulk?listType=all", nil, nil, api.ListAccessKeysBulk), 403)
 			assertIAMAdminStatus(t, "info service", iamAdminCall(t, caller, "GET", "/minio/admin/v3/info-service-account?accessKey="+target.AccessKey, nil, map[string]string{"accessKey": target.AccessKey}, api.InfoServiceAccount), 403)
 			assertIAMAdminStatus(t, "info access key", iamAdminCall(t, caller, "GET", "/minio/admin/v3/info-access-key?accessKey="+target.AccessKey, nil, map[string]string{"accessKey": target.AccessKey}, api.InfoAccessKey), 403)
 			assertIAMAdminStatus(t, "delete", iamAdminCall(t, caller, "DELETE", "/minio/admin/v3/delete-service-account?accessKey="+target.AccessKey, nil, map[string]string{"accessKey": target.AccessKey}, api.DeleteServiceAccount), 403)
@@ -477,7 +491,10 @@ func TestIAMAdminListSelfAliases(t *testing.T) {
 			ctx, sys := iamAdminFixture(t)
 			parent := iamAdminUser(ctx, t, sys, "alias-parent", "readwrite")
 			target := iamAdminServiceAccount(ctx, t, sys, parent.AccessKey, "alias-target", "")
-			caller := iamAdminServiceAccount(ctx, t, sys, parent.AccessKey, "alias-caller", "")
+			serviceCaller := iamAdminServiceAccount(ctx, t, sys, parent.AccessKey, "alias-caller", "")
+			caller := serviceCaller
+			other := iamAdminUser(ctx, t, sys, "alias-other", "readwrite")
+			_ = iamAdminServiceAccount(ctx, t, sys, other.AccessKey, "alias-other-child", "")
 			if kind == "sts" {
 				var err error
 				caller, err = auth.GetNewCredentialsWithMetadata(map[string]any{parentClaim: parent.AccessKey, "exp": UTCNow().Add(time.Hour).Unix()}, globalActiveCred.SecretKey)
@@ -501,6 +518,38 @@ func TestIAMAdminListSelfAliases(t *testing.T) {
 				if !found {
 					t.Fatalf("self alias %q did not list the parent's service accounts", alias)
 				}
+
+				query := url.Values{"listType": {madmin.AccessKeyListAll}}
+				if alias != "" {
+					query.Set("users", alias)
+				}
+				bulk := iamAdminCall(t, caller, "GET", "/minio/admin/v3/list-access-keys-bulk?"+query.Encode(), nil, nil, api.ListAccessKeysBulk)
+				assertIAMAdminStatus(t, "list bulk self alias", bulk, 200)
+				data, err = madmin.DecryptData(caller.SecretKey, bulk.Body)
+				mustIAM(t, err)
+				var listed map[string]madmin.ListAccessKeysResp
+				mustIAM(t, json.Unmarshal(data, &listed))
+				entry, ok := listed[parent.AccessKey]
+				if len(listed) != 1 || !ok || len(entry.ServiceAccounts) != 2 {
+					t.Fatalf("bulk self alias %q returned unexpected identities: %+v", alias, listed)
+				}
+				want := map[string]bool{target.AccessKey: true, serviceCaller.AccessKey: true}
+				for _, account := range entry.ServiceAccounts {
+					if !want[account.AccessKey] {
+						t.Fatalf("bulk self alias %q returned unexpected or duplicate service account %s", alias, account.AccessKey)
+					}
+					delete(want, account.AccessKey)
+				}
+				if kind == "sts" {
+					if len(entry.STSKeys) != 1 || entry.STSKeys[0].AccessKey != caller.AccessKey {
+						t.Fatalf("bulk self alias %q returned unexpected STS accounts: %+v", alias, entry.STSKeys)
+					}
+				} else if len(entry.STSKeys) != 0 {
+					t.Fatalf("service caller returned unexpected STS accounts: %+v", entry.STSKeys)
+				}
+				if bytes.Contains(data, []byte(serviceCaller.SecretKey)) || bytes.Contains(data, []byte(`"secretKey"`)) || bytes.Contains(data, []byte(`"sessionToken"`)) {
+					t.Fatal("bulk listing returned credential secrets")
+				}
 			}
 		})
 	}

cmd/admin-handlers-users.go

@@ -1237,7 +1237,7 @@ func (a adminAPIHandlers) ListAccessKeysBulk(w http.ResponseWriter, r *http.Requ
 		}
 	}
 
-	if !globalIAMSys.IsAllowed(policy.Args{
+	if !isAllowedServiceAccountAction(cred, policy.Args{
 		AccountName:     cred.AccessKey,
 		Groups:          cred.Groups,
 		Action:          policy.ListServiceAccountsAdminAction,
@@ -1250,12 +1250,12 @@ func (a adminAPIHandlers) ListAccessKeysBulk(w http.ResponseWriter, r *http.Requ
 		return
 	}
 
-	if selfOnly && len(users) == 0 {
+	if selfOnly {
 		selfUser := cred.AccessKey
 		if cred.ParentUser != "" {
 			selfUser = cred.ParentUser
 		}
-		users = append(users, selfUser)
+		users = []string{selfUser}
 	}
 
 	var checkedUserList []string

cmd/api-errors.go

@@ -450,7 +450,7 @@ const (
 	ErrAdminNoSecretKey
 
 	ErrIAMNotInitialized
-	ErrMultipartListingLegacy
+	ErrMultipartListingLegacy // Reserved: legacy uploads no longer block listing.
 	ErrMultipartListingIdentity
 	ErrSlowDown
 
@@ -2191,8 +2191,6 @@ func toAPIErrorCode(ctx context.Context, err error) (apiErr APIErrorCode) {
 	err = unwrapAll(err)
 
 	switch err {
-	case errMultipartListingLegacy:
-		apiErr = ErrMultipartListingLegacy
 	case errMultipartListingIdentity:
 		apiErr = ErrMultipartListingIdentity
 	case errCompleteMultipartChecksumMismatch, errCompleteMultipartChecksumTypeMismatch:

cmd/common-main.go

@@ -56,6 +56,7 @@ import (
 	"github.com/minio/minio/internal/auth"
 	"github.com/minio/minio/internal/color"
 	"github.com/minio/minio/internal/config"
+	"github.com/minio/minio/internal/config/api"
 	"github.com/minio/minio/internal/handlers"
 	"github.com/minio/minio/internal/kms"
 	"github.com/minio/minio/internal/logger"
@@ -881,6 +882,9 @@ func serverHandleEnvVars() {
 		logger.Warning("%s=%s is ignored: Silo in-place updates are permanently disabled", config.EnvUpdate, updateSetting)
 	}
 	globalInplaceUpdateDisabled = true
+	if env.IsSet(api.EnvAPIMultipartListing) {
+		logger.Warning("%s is ignored: multipart listing uses the legacy cache/exact-key behavior; persistent listing is deferred", api.EnvAPIMultipartListing)
+	}
 
 	// Check if the supported credential env vars,
 	// "MINIO_ROOT_USER" and "MINIO_ROOT_PASSWORD" are provided

cmd/erasure-multipart-cleanup_test.go

@@ -126,8 +126,8 @@ func (d *multipartCleanupDisk) ReadVersion(context.Context, string, string, stri
 	return FileInfo{ModTime: d.modTime}, d.readErr
 }
 
-func (d *multipartCleanupDisk) StatVol(context.Context, string) (VolInfo, error) {
+func (d *multipartCleanupDisk) StatVol(ctx context.Context, volume string) (VolInfo, error) {
-	return VolInfo{Created: time.Now()}, nil
+	return (&xlStorage{drivePath: d.root}).StatVol(ctx, volume)
 }
 
 func TestCleanupStaleUploadsOnDisk(t *testing.T) {
@@ -157,7 +157,13 @@ func TestCleanupStaleUploadsOnDisk(t *testing.T) {
 			}
 			disk := &multipartCleanupDisk{root: t.TempDir(), modTime: modTime, readErr: tc.readErr}
 			uploadDir := filepath.Join(disk.root, minioMetaMultipartBucket, "sha", id)
-			if err := os.MkdirAll(uploadDir, 0o755); err != nil {
+			for _, dir := range []string{uploadDir, filepath.Join(disk.root, minioMetaTmpDeletedBucket)} {
+				if err := os.MkdirAll(dir, 0o755); err != nil {
+					t.Fatal(err)
+				}
+			}
+			// An old parent must not bypass a skipped upload's age or read error.
+			if err := os.Chtimes(filepath.Dir(uploadDir), modTime, modTime); err != nil {
 				t.Fatal(err)
 			}
 			erasureObjects{}.cleanupStaleUploadsOnDisk(t.Context(), disk)
@@ -169,9 +175,60 @@ func TestCleanupStaleUploadsOnDisk(t *testing.T) {
 				if !os.IsNotExist(err) {
 					t.Errorf("expired upload must be removed: %v", err)
 				}
+				if _, err := os.Stat(filepath.Dir(uploadDir)); !os.IsNotExist(err) {
+					t.Errorf("empty hash directory must be removed in the same pass: %v", err)
+				}
 			} else if err != nil {
 				t.Errorf("upload must be preserved: %v", err)
 			}
 		})
 	}
 }
+
+func TestCleanupStaleUploadsEmptyHashDirectory(t *testing.T) {
+	for _, age := range []time.Duration{0, 2 * globalAPIConfig.getStaleUploadsExpiry()} {
+		t.Run(age.String(), func(t *testing.T) {
+			disk := &multipartCleanupDisk{root: t.TempDir()}
+			shaDir := filepath.Join(disk.root, minioMetaMultipartBucket, "sha")
+			for _, dir := range []string{shaDir, filepath.Join(disk.root, minioMetaTmpDeletedBucket)} {
+				if err := os.MkdirAll(dir, 0o755); err != nil {
+					t.Fatal(err)
+				}
+			}
+			modTime := time.Now().Add(-age)
+			if err := os.Chtimes(shaDir, modTime, modTime); err != nil {
+				t.Fatal(err)
+			}
+			erasureObjects{}.cleanupStaleUploadsOnDisk(t.Context(), disk)
+			if _, err := os.Stat(shaDir); !os.IsNotExist(err) {
+				t.Fatalf("empty hash directory must be removed regardless of age: %v", err)
+			}
+		})
+	}
+}
+
+func TestCleanupStaleUploadsSharedHashDirectory(t *testing.T) {
+	disk := &multipartCleanupDisk{root: t.TempDir(), readErr: errFileNotFound}
+	shaDir := filepath.Join(disk.root, minioMetaMultipartBucket, "sha")
+	now := time.Now()
+	expiredID := fmt.Sprintf("d16ad42a-6fe5-4a3b-8092-69983142238fx%d", now.Add(-2*globalAPIConfig.getStaleUploadsExpiry()).UnixNano())
+	recentID := fmt.Sprintf("4eacd477-e733-4851-b821-6fd42785ff87x%d", now.UnixNano())
+	for _, id := range []string{expiredID, recentID} {
+		if err := os.MkdirAll(filepath.Join(shaDir, id), 0o755); err != nil {
+			t.Fatal(err)
+		}
+	}
+	partPath := filepath.Join(shaDir, recentID, "part.1")
+	const part = "recent upload part"
+	if err := os.WriteFile(partPath, []byte(part), 0o600); err != nil {
+		t.Fatal(err)
+	}
+
+	erasureObjects{}.cleanupStaleUploadsOnDisk(t.Context(), disk)
+	if _, err := os.Stat(filepath.Join(shaDir, expiredID)); !os.IsNotExist(err) {
+		t.Errorf("expired upload must be removed: %v", err)
+	}
+	if got, err := os.ReadFile(partPath); err != nil || string(got) != part {
+		t.Errorf("recent upload must remain intact: got %q, err %v", got, err)
+	}
+}

cmd/erasure-multipart-listing-benchmark_test.go

@@ -0,0 +1,170 @@
+// Copyright (c) 2026 Ruohang Feng
+// SPDX-License-Identifier: AGPL-3.0-or-later
+
+package cmd
+
+import (
+	"context"
+	"errors"
+	"fmt"
+	"sync/atomic"
+	"testing"
+)
+
+// Run explicitly with -benchtime=1x. Fixture creation is excluded from timing;
+// each request still discovers real on-disk metadata. The 16-drive/10k case
+// deliberately exceeds the existing aggregate directory-entry budget.
+func BenchmarkMultipartDiagnosticScanCapacity(b *testing.B) {
+	for _, tc := range []struct {
+		drives, uploads int
+		mixed, limited  bool
+	}{
+		{4, 1_000, false, false},
+		{4, 10_000, false, false},
+		{4, 10_000, true, false},
+		{16, 1_000, false, false},
+		{16, 10_000, false, true},
+	} {
+		b.Run(fmt.Sprintf("drives=%d/uploads=%d/mixed=%t", tc.drives, tc.uploads, tc.mixed), func(b *testing.B) {
+			obj, dirs, err := prepareErasure(b.Context(), tc.drives)
+			if err != nil {
+				b.Fatal(err)
+			}
+			z := obj.(*erasureServerPools)
+			b.Cleanup(func() { z.Shutdown(context.Background()); removeRoots(dirs) })
+			const bucket, other = "multipart-capacity", "multipart-unrelated"
+			for _, name := range []string{bucket, other} {
+				if err := z.MakeBucket(b.Context(), name, MakeBucketOptions{}); err != nil {
+					b.Fatal(err)
+				}
+			}
+			for n := range tc.uploads {
+				target := bucket
+				if tc.mixed && n%2 != 0 {
+					target = other
+				}
+				key := fmt.Sprintf("dir/%05d", n)
+				mp, err := z.NewMultipartUpload(b.Context(), target, key, ObjectOptions{})
+				if err != nil {
+					b.Fatal(err)
+				}
+				if tc.mixed && n%10 == 0 {
+					rewriteMultipartIdentity(b, z.serverPools[0].getHashedSet(key), target, key, mp.UploadID, removeMultipartIdentity)
+				}
+			}
+			z.mpCache.Clear()
+			var directoryCalls, metadataCalls atomic.Int64
+			for _, set := range z.serverPools[0].sets {
+				original := set.getDisks
+				set.getDisks = func() []StorageAPI {
+					disks := append([]StorageAPI(nil), original()...)
+					for i, disk := range disks {
+						disks[i] = multipartListingCountingDisk{disk, &directoryCalls, &metadataCalls}
+					}
+					return disks
+				}
+				b.Cleanup(func() { set.getDisks = original })
+			}
+			b.ResetTimer()
+			for range b.N {
+				got, err := z.scanMultipartUploadsForTest(b.Context(), bucket, "dir/", "", "", "", 1000)
+				if tc.limited {
+					var limited SlowDown
+					if !errors.As(err, &limited) || len(got.Uploads) != 0 {
+						b.Fatalf("overflow must return SlowDown without a partial page: %+v %v", got, err)
+					}
+				} else if err != nil || len(got.Uploads) != 1000 {
+					b.Fatalf("capacity listing: count=%d err=%v", len(got.Uploads), err)
+				}
+			}
+			b.StopTimer()
+			b.ReportMetric(float64(directoryCalls.Load())/float64(b.N), "ListDir/op")
+			b.ReportMetric(float64(metadataCalls.Load())/float64(b.N), "ReadVersion/op")
+		})
+	}
+}
+
+// The public path is measured with a populated cache as well as after cache
+// loss. An empty post-restart response alone is not capacity acceptance.
+func BenchmarkMultipartPublicListingCapacity(b *testing.B) {
+	for _, tc := range []struct {
+		drives, uploads int
+		sameKey         bool
+	}{
+		{4, 10_000, false},
+		{16, 3_126, false},
+		{16, 10_000, true},
+	} {
+		b.Run(fmt.Sprintf("drives=%d/uploads=%d/same-key=%t", tc.drives, tc.uploads, tc.sameKey), func(b *testing.B) {
+			obj, dirs, err := prepareErasure(b.Context(), tc.drives)
+			if err != nil {
+				b.Fatal(err)
+			}
+			z := obj.(*erasureServerPools)
+			b.Cleanup(func() { z.Shutdown(context.Background()); removeRoots(dirs) })
+			const bucket, other = "multipart-public-capacity", "multipart-public-other"
+			for _, name := range []string{bucket, other} {
+				if err := z.MakeBucket(b.Context(), name, MakeBucketOptions{}); err != nil {
+					b.Fatal(err)
+				}
+			}
+			wantCached, wantExact := 0, 1
+			for n := range tc.uploads {
+				target, key := bucket, fmt.Sprintf("dir/%05d", n)
+				if tc.sameKey {
+					key = "dir/00000"
+				} else if n%2 != 0 {
+					target = other
+				}
+				if target == bucket {
+					wantCached++
+				}
+				if _, err := z.NewMultipartUpload(b.Context(), target, key, ObjectOptions{}); err != nil {
+					b.Fatal(err)
+				}
+			}
+			if tc.sameKey {
+				wantExact = 1000
+			}
+			if tc.drives == 16 && !tc.sameKey {
+				_, err := z.scanMultipartUploadsForTest(b.Context(), bucket, "", "", "", "", 1000)
+				var limited SlowDown
+				if !errors.As(err, &limited) {
+					b.Fatalf("fixture did not reproduce the previous global budget failure: %v", err)
+				}
+			}
+			var directoryCalls, metadataCalls atomic.Int64
+			for _, set := range z.serverPools[0].sets {
+				original := set.getDisks
+				set.getDisks = func() []StorageAPI {
+					disks := append([]StorageAPI(nil), original()...)
+					for i, disk := range disks {
+						disks[i] = multipartListingCountingDisk{disk, &directoryCalls, &metadataCalls}
+					}
+					return disks
+				}
+				b.Cleanup(func() { set.getDisks = original })
+			}
+			measure := func(name, prefix string, want int) {
+				b.Run(name, func(b *testing.B) {
+					directoryCalls.Store(0)
+					metadataCalls.Store(0)
+					b.ResetTimer()
+					for range b.N {
+						got, err := z.ListMultipartUploads(b.Context(), bucket, prefix, "", "", "", 1000)
+						if err != nil || len(got.Uploads) != want {
+							b.Fatalf("public listing: count=%d want=%d err=%v", len(got.Uploads), want, err)
+						}
+					}
+					b.StopTimer()
+					b.ReportMetric(float64(directoryCalls.Load())/float64(b.N), "ListDir/op")
+					b.ReportMetric(float64(metadataCalls.Load())/float64(b.N), "ReadVersion/op")
+				})
+			}
+			measure("populated-cache", "", wantCached)
+			z.mpCache.Clear()
+			measure("lost-cache", "", 0)
+			measure("known-key", "dir/00000", wantExact)
+		})
+	}
+}

cmd/erasure-multipart-listing.go

@@ -23,7 +23,6 @@ const multipartScanEntryLimit = 100_000
 
 var (
 	multipartScanSlots          = make(chan struct{}, 2)
-	errMultipartListingLegacy   = errors.New("legacy multipart uploads require a coordinated upgrade and drain")
 	errMultipartListingIdentity = errors.New("multipart upload identity is invalid")
 )
 
@@ -133,6 +132,14 @@ func (er erasureObjects) multipartIdentity(fi FileInfo, shaDir string) (string,
 		IsValidObjectPrefix(object) && er.getMultipartSHADir(bucket, object) == shaDir
 }
 
+// Only the absence of both fields identifies the historical format. Empty,
+// partial or invalid identities must not hide a damaged new upload as legacy.
+func multipartLegacyIdentity(fi FileInfo) bool {
+	_, bucketPresent := fi.Metadata[multipartMetaBucket]
+	_, objectPresent := fi.Metadata[multipartMetaObject]
+	return !bucketPresent && !objectPresent
+}
+
 func (er erasureObjects) readMultipartUploadCandidate(s *multipartScan, bucket, candidate string, source StorageAPI) (MultipartInfo, bool, bool, error) {
 	if s.ctx.Err() != nil {
 		return MultipartInfo{}, false, false, SlowDown{}
@@ -162,8 +169,8 @@ func (er erasureObjects) readMultipartUploadCandidate(s *multipartScan, bucket,
 	disks := er.getDisks()
 	metadata, errs := readAllFileInfo(s.ctx, disks, bucket, minioMetaMultipartBucket, candidate, "", false, false)
 	if s.preflight {
-		// Readiness is stronger than listing liveness: even one readable legacy
+		// The diagnostic counts even a single readable legacy copy. It does not
-		// copy must be drained, and an unreadable copy cannot certify readiness.
+		// gate listing; unreadable copies still prevent a complete diagnosis.
 		var oldest MultipartInfo
 		legacy := false
 		_, nativeID := multipartUploadTime(uploadUUID)
@@ -176,10 +183,11 @@ func (er erasureObjects) readMultipartUploadCandidate(s *multipartScan, bucket,
 			}
 			fi := metadata[i]
 			storedBucket, storedObject, valid := er.multipartIdentity(fi, shaDir)
-			if storedBucket != "" && storedObject != "" && !valid {
+			isLegacy := multipartLegacyIdentity(fi)
+			if !isLegacy && (!valid || !nativeID) {
 				return MultipartInfo{}, false, false, errMultipartListingIdentity
 			}
-			if !valid || !nativeID {
+			if isLegacy {
 				info := multipartUploadInfo(storedBucket, storedObject, uploadUUID, fi.ModTime)
 				if !legacy || info.Initiated.Before(oldest.Initiated) {
 					oldest = info
@@ -203,16 +211,29 @@ func (er erasureObjects) readMultipartUploadCandidate(s *multipartScan, bucket,
 	if err != nil {
 		return MultipartInfo{}, false, false, err
 	}
+	// The generic metadata quorum does not compare these identity fields.
+	// Contradictory readable copies cannot certify a historical upload: picking
+	// a copy with missing fields must not silently hide a new-format record.
+	legacy := multipartLegacyIdentity(fi)
 	storedBucket, storedObject, valid := er.multipartIdentity(fi, shaDir)
+	for i, other := range metadata {
+		if errs[i] != nil || !other.IsValid() {
+			continue
+		}
+		if multipartLegacyIdentity(other) != legacy ||
+			(!legacy && (other.Metadata[multipartMetaBucket] != storedBucket || other.Metadata[multipartMetaObject] != storedObject)) {
+			return MultipartInfo{}, false, false, fmt.Errorf("%w: inconsistent copies at %s", errMultipartListingIdentity, candidate)
+		}
+	}
 	info := multipartUploadInfo(storedBucket, storedObject, uploadUUID, fi.ModTime)
-	if storedBucket == "" || storedObject == "" {
+	if legacy {
 		return info, false, true, nil
 	}
 	if !valid {
 		return MultipartInfo{}, false, false, fmt.Errorf("%w: %s", errMultipartListingIdentity, candidate)
 	}
 	if _, ok := multipartUploadTime(uploadUUID); !ok {
-		return info, false, true, nil
+		return MultipartInfo{}, false, false, fmt.Errorf("%w: %s", errMultipartListingIdentity, candidate)
 	}
 	if bucket != "" && bucket != storedBucket {
 		return MultipartInfo{}, false, false, nil
@@ -220,7 +241,7 @@ func (er erasureObjects) readMultipartUploadCandidate(s *multipartScan, bucket,
 	return info, true, false, nil
 }
 
-func (er erasureObjects) scanMultipartUploads(s *multipartScan, bucket string, poolIdx, setIdx int) ([]MultipartInfo, bool, error) {
+func (er erasureObjects) scanMultipartUploads(s *multipartScan, bucket string, poolIdx, setIdx int) ([]MultipartInfo, error) {
 	disks := er.getDisks()
 	report := multipartScanSet{Pool: poolIdx, Set: setIdx, Drives: er.setDriveCount}
 	var resultErr error
@@ -272,7 +293,7 @@ func (er erasureObjects) scanMultipartUploads(s *multipartScan, bucket string, p
 		resultErr = toObjectErr(errErasureReadQuorum, bucket)
 	}
 	if resultErr != nil {
-		return nil, false, resultErr
+		return nil, resultErr
 	}
 	type candidate struct {
 		path   string
@@ -325,25 +346,23 @@ dispatch:
 	if s.ctx.Err() != nil {
 		resultErr = SlowDown{}
 	}
-	return uploads, report.LegacyUploads != 0, resultErr
+	return uploads, resultErr
 }
 
 // multipartPreflight scans all pools, sets and drives, independent of caches.
-// A majority suffices for normal listing; upgrade readiness requires every
+// A majority suffices for normal listing; a complete diagnosis requires every
-// drive to have been inspected. The operator must also upgrade all writers.
+// drive to have been inspected. Ready describes the absence of legacy copies,
+// not whether listing is enabled.
 func (z *erasureServerPools) multipartPreflight(ctx context.Context) (multipartPreflightReport, error) {
 	s, err := startMultipartScan(ctx, true)
 	if err != nil {
 		return multipartPreflightReport{}, err
 	}
 	defer s.close()
-	report := multipartPreflightReport{Complete: true, Mode: "strict"}
+	report := multipartPreflightReport{Complete: true, Mode: "legacy"}
-	if globalAPIConfig.getMultipartListingLegacy() {
-		report.Mode = "legacy"
-	}
 	for p, pool := range z.serverPools {
 		for i, set := range pool.sets {
-			_, _, _ = set.scanMultipartUploads(s, "", p, i)
+			_, _ = set.scanMultipartUploads(s, "", p, i)
 		}
 	}
 	report.Sets = s.sets

cmd/erasure-multipart-listing_test.go

@@ -22,16 +22,59 @@ import (
 	"github.com/minio/minio/internal/config/storageclass"
 )
 
-// Check the real handler and storage path, including a successful abort between
+// The persistent scanner is retained for diagnostics. This test-only wrapper
-// pages. Choose IDs increasing in both initiation time and lexical order so the
+// preserves its pagination coverage without exposing it through public listing.
-// failure does not depend on differing interpretations of S3 marker ordering.
+func (z *erasureServerPools) scanMultipartUploadsForTest(ctx context.Context, bucket, prefix, keyMarker, uploadIDMarker, delimiter string, maxUploads int) (ListMultipartsInfo, error) {
-func TestMultipartListingAbortBetweenHTTPPages(t *testing.T) {
+	if err := checkListMultipartArgs(ctx, bucket, prefix, keyMarker, uploadIDMarker, delimiter); err != nil {
+		return ListMultipartsInfo{}, err
+	}
+	if _, err := z.GetBucketInfo(ctx, bucket, BucketOptions{}); err != nil {
+		return ListMultipartsInfo{}, toObjectErr(err, bucket)
+	}
+	scan, err := startMultipartScan(ctx, false)
+	if err != nil {
+		return ListMultipartsInfo{}, err
+	}
+	defer scan.close()
+
+	var uploads []MultipartInfo
+	for idx, pool := range z.serverPools {
+		if z.IsSuspended(idx) {
+			continue
+		}
+		poolUploads, err := pool.scanMultipartUploadsForTest(scan, bucket, idx)
+		if err != nil {
+			return ListMultipartsInfo{}, err
+		}
+		uploads = append(uploads, poolUploads...)
+	}
+
+	// Confirmed historical records lack a recoverable bucket/key and are
+	// omitted by the scan. They do not block listing identifiable uploads.
+	return paginateMultipartUploads(uploads, prefix, keyMarker, uploadIDMarker, delimiter, maxUploads), nil
+}
+
+func (s *erasureSets) scanMultipartUploadsForTest(scan *multipartScan, bucket string, poolIdx int) ([]MultipartInfo, error) {
+	var uploads []MultipartInfo
+	for i, set := range s.sets {
+		setUploads, err := set.scanMultipartUploads(scan, bucket, poolIdx, i)
+		if err != nil {
+			return nil, err
+		}
+		uploads = append(uploads, setUploads...)
+	}
+	return uploads, nil
+}
+
+// Check that a successful HTTP abort removes the cached upload. The restored
+// cache path still ignores page limits; this is a documented legacy limitation.
+func TestMultipartListingAbortHTTPUpdatesCache(t *testing.T) {
 	z, _ := consistencyPools(t)
 	bucket, router, err := initAPIHandlerTest(t.Context(), z, []string{"ListMultipartUploads", "AbortMultipart"}, MakeBucketOptions{})
 	if err != nil {
 		t.Fatal(err)
 	}
-	setMultipartListingTestMode(t, false)
+
 	var firstID, secondID string
 	for attempt := 0; attempt < 32; attempt++ {
 		one, err := z.NewMultipartUpload(t.Context(), bucket, "a", ObjectOptions{})
@@ -81,8 +124,8 @@ func TestMultipartListingAbortBetweenHTTPPages(t *testing.T) {
 		return result
 	}
 	first := list("", "", 1)
-	if len(first.Uploads) != 1 || first.Uploads[0].UploadID != firstID || !first.IsTruncated {
+	if len(first.Uploads) != 3 || first.Uploads[0].UploadID != firstID || first.IsTruncated {
-		t.Fatalf("unexpected first page: %+v", first)
+		t.Fatalf("unexpected legacy cache view: %+v", first)
 	}
 	rec := request(http.MethodDelete, getAbortMultipartUploadURL("", bucket, "a", firstID))
 	if rec.Code != http.StatusNoContent {
@@ -93,12 +136,12 @@ func TestMultipartListingAbortBetweenHTTPPages(t *testing.T) {
 	for _, upload := range rest.Uploads {
 		keys = append(keys, upload.Key)
 	}
-	t.Logf("GET first page=200; DELETE marker=204; GET next page=200 keys=%v truncated=%v", keys, rest.IsTruncated)
+	t.Logf("GET cache=200; DELETE upload=204; GET cache=200 keys=%v truncated=%v", keys, rest.IsTruncated)
 	if _, err := z.GetMultipartInfo(t.Context(), bucket, "a", secondID, ObjectOptions{}); err != nil {
 		t.Fatalf("remaining upload is not valid: %v", err)
 	}
 	if len(rest.Uploads) != 2 || rest.Uploads[0].UploadID != secondID {
-		t.Fatalf("valid remaining upload for key a is missing from continuation: %+v", rest.Uploads)
+		t.Fatalf("valid remaining upload for key a is missing from cache: %+v", rest.Uploads)
 	}
 }
 
@@ -148,28 +191,25 @@ func TestMultipartListingLegacyPreflight(t *testing.T) {
 		t.Fatal(err)
 	}
 	z.mpCache.Clear()
-	t.Run("legacy-upgrade", func(t *testing.T) {
+	for _, target := range []string{bucket, other} {
-		setMultipartListingTestMode(t, true)
+		got, err := z.scanMultipartUploadsForTest(t.Context(), target, "", "", "", "", 10)
-		exact, err := z.ListMultipartUploads(t.Context(), other, "old", "", "", "", 10)
 		if err != nil {
 			t.Fatal(err)
 		}
-		requireMultipartUploadKeys(t, exact, "old")
+		if target == bucket {
-		const empty = "multipart-legacy-empty"
+			requireMultipartUploadKeys(t, got, "a")
-		if err := z.MakeBucket(t.Context(), empty, MakeBucketOptions{}); err != nil {
+		} else {
-			t.Fatal(err)
+			requireMultipartUploadKeys(t, got)
 		}
-		got, err := z.ListMultipartUploads(t.Context(), empty, "", "", "", "", 10)
-		if err != nil || len(got.Uploads) != 0 {
-			t.Fatalf("old upload in another bucket broke legacy listing: %+v %v", got, err)
-		}
-	})
-	_, err = z.ListMultipartUploads(t.Context(), bucket, "", "", "", "", 10)
-	if !errors.Is(err, errMultipartListingLegacy) {
-		t.Fatalf("old upload in another bucket: %v", err)
 	}
+	// Diagnostic discovery cannot recover an unidentified historical key.
+	exact, err := z.scanMultipartUploadsForTest(t.Context(), other, "old", "", "", "", 10)
+	if err != nil {
+		t.Fatal(err)
+	}
+	requireMultipartUploadKeys(t, exact)
 	report, err := z.multipartPreflight(t.Context())
-	if err != nil || report.Ready || !report.Complete || report.LegacyUploads != 1 {
+	if err != nil || report.Ready || !report.Complete || report.LegacyUploads != 1 || report.Mode != "legacy" {
 		t.Fatalf("legacy preflight: %+v %v", report, err)
 	}
 	if err = z.AbortMultipartUpload(t.Context(), other, "old", old.UploadID, ObjectOptions{}); err != nil {
@@ -179,15 +219,15 @@ func TestMultipartListingLegacyPreflight(t *testing.T) {
 	if err != nil || !report.Ready || report.LegacyUploads != 0 {
 		t.Fatalf("drained preflight: %+v %v", report, err)
 	}
-	got, err := z.ListMultipartUploads(t.Context(), bucket, "", "", "", "", 10)
+	got, err := z.scanMultipartUploadsForTest(t.Context(), bucket, "", "", "", "", 10)
 	if err != nil {
 		t.Fatal(err)
 	}
 	requireMultipartUploadKeys(t, got, "a")
 }
 
-func TestMultipartListingIdentityFallback(t *testing.T) {
+func TestMultipartDiagnosticIdentityFallback(t *testing.T) {
-	for _, kind := range []string{"old", "corrupt", "read-failure", "wrong-bucket"} {
+	for _, kind := range []string{"old", "partial", "corrupt", "read-failure", "wrong-bucket"} {
 		t.Run(kind, func(t *testing.T) {
 			z, set, bucket := multipartListingFixture(t)
 			if _, err := z.NewMultipartUpload(t.Context(), bucket, "a", ObjectOptions{}); err != nil {
@@ -209,8 +249,11 @@ func TestMultipartListingIdentityFallback(t *testing.T) {
 								return FileInfo{}, errFileCorrupt
 							}
 							fi.Metadata = cloneMSS(fi.Metadata)
-							if kind == "old" {
+							if kind == "old" || kind == "partial" {
 								delete(fi.Metadata, multipartMetaBucket)
+								if kind == "old" {
+									delete(fi.Metadata, multipartMetaObject)
+								}
 							} else {
 								fi.Metadata[multipartMetaBucket] = "another-bucket"
 							}
@@ -220,7 +263,7 @@ func TestMultipartListingIdentityFallback(t *testing.T) {
 				}
 				return disks
 			}
-			got, err := z.ListMultipartUploads(t.Context(), bucket, "", "", "", "", 10)
+			got, err := z.scanMultipartUploadsForTest(t.Context(), bucket, "", "", "", "", 10)
 			if err != nil {
 				t.Fatal(err)
 			}
@@ -231,7 +274,7 @@ func TestMultipartListingIdentityFallback(t *testing.T) {
 
 func TestMultipartAbortPoolsAndRetry(t *testing.T) {
 	z, bucket := consistencyPools(t)
-	setMultipartListingTestMode(t, false)
+
 	mp, err := z.serverPools[1].NewMultipartUpload(t.Context(), bucket, "a", ObjectOptions{})
 	if err != nil {
 		t.Fatal(err)
@@ -252,10 +295,12 @@ func TestMultipartAbortPoolsAndRetry(t *testing.T) {
 	if err == nil || toAPIError(t.Context(), err).HTTPStatusCode != 503 {
 		t.Fatalf("unknown pool must not acknowledge cancellation: %v", err)
 	}
+	if _, err := z.serverPools[1].GetMultipartInfo(t.Context(), bucket, "a", mp.UploadID, ObjectOptions{}); err != nil {
+		t.Fatalf("earlier pool error changed the later pool: %v", err)
+	}
 	emptySet.getDisks = original
-	err = z.AbortMultipartUpload(t.Context(), bucket, "a", mp.UploadID, ObjectOptions{})
+	if err = z.AbortMultipartUpload(t.Context(), bucket, "a", mp.UploadID, ObjectOptions{}); err != nil {
-	if _, ok := err.(InvalidUploadID); !ok {
+		t.Fatalf("retry after pool recovery: %v", err)
-		t.Fatalf("retry after all pools confirm absence: %v", err)
 	}
 	mp, err = z.serverPools[1].NewMultipartUpload(t.Context(), bucket, "a", ObjectOptions{})
 	if err != nil {
@@ -289,12 +334,13 @@ func TestMultipartAbortRetryBelowReadQuorum(t *testing.T) {
 		}}
 		return disks
 	}
-	if err = z.AbortMultipartUpload(t.Context(), bucket, "a", mp.UploadID, ObjectOptions{}); err == nil {
+	if err = z.AbortMultipartUpload(t.Context(), bucket, "a", mp.UploadID, ObjectOptions{}); err != nil {
-		t.Fatal("lost acknowledgement must fail")
+		t.Fatalf("best-effort cleanup must retain read-quorum availability: %v", err)
 	}
 	set.getDisks = original
-	if err = z.AbortMultipartUpload(t.Context(), bucket, "a", mp.UploadID, ObjectOptions{}); err != nil {
+	var invalid InvalidUploadID
-		t.Fatalf("one remaining metadata copy prevented retry: %v", err)
+	if err = z.AbortMultipartUpload(t.Context(), bucket, "a", mp.UploadID, ObjectOptions{}); !errors.As(err, &invalid) {
+		t.Fatalf("minority remnant must retain absent-upload result: %v", err)
 	}
 }
 
@@ -304,7 +350,7 @@ func TestMultipartListingMarkerHTTP(t *testing.T) {
 	if err != nil {
 		t.Fatal(err)
 	}
-	setMultipartListingTestMode(t, false)
+
 	for _, tc := range []struct {
 		key, marker string
 		status      int
@@ -388,7 +434,7 @@ func TestMultipartAbortLateCreateBoundary(t *testing.T) {
 		t.Fatal(err)
 	}
 	z := obj.(*erasureServerPools)
-	setMultipartListingTestMode(t, false)
+
 	t.Cleanup(func() { z.Shutdown(context.Background()); removeRoots(dirs) })
 	saved := globalStorageClass
 	globalStorageClass.Update(storageclass.Config{Standard: storageclass.StorageClass{Parity: 8}})
@@ -417,9 +463,7 @@ func TestMultipartAbortLateCreateBoundary(t *testing.T) {
 	}
 	set.getDisks = func() []StorageAPI {
 		disks := append([]StorageAPI(nil), original()...)
-		for i := 2; i < 9; i++ {
+		disks[8] = nil
-			disks[i] = nil
-		}
 		return disks
 	}
 	if err = z.AbortMultipartUpload(t.Context(), bucket, "a", mp.UploadID, ObjectOptions{}); err != nil {
@@ -431,7 +475,7 @@ func TestMultipartAbortLateCreateBoundary(t *testing.T) {
 		}
 	}
 	set.getDisks = original
-	_, metadata, err := set.checkUploadIDExists(t.Context(), bucket, "a", mp.UploadID, true)
+	_, metadata, err := set.checkUploadIDExists(t.Context(), bucket, "a", mp.UploadID, false)
 	if err != nil {
 		t.Fatalf("late creation boundary changed; review and remove the documented limitation: %v", err)
 	}
@@ -441,10 +485,10 @@ func TestMultipartAbortLateCreateBoundary(t *testing.T) {
 			count++
 		}
 	}
-	if count != 14 {
+	if count != 8 {
-		t.Fatalf("expected fourteen resurrected copies, got %d", count)
+		t.Fatalf("expected eight resurrected copies, got %d", count)
 	}
-	t.Log("KNOWN UNRESOLVED BOUNDARY: seven delayed creation writes plus seven offline copies restore a writable upload after acknowledged cancellation")
+	t.Log("KNOWN UNRESOLVED BOUNDARY: seven delayed creation writes plus one offline copy restore a readable upload after acknowledged cancellation")
 }
 
 type multipartListingCountingDisk struct {
@@ -468,13 +512,13 @@ func (d multipartListingCountingDisk) ReadVersion(ctx context.Context, original,
 }
 
 // Counts storage API calls; this is not a deployment throughput benchmark.
-func TestMultipartListingScanCosts(t *testing.T) {
+func TestMultipartDiagnosticScanCosts(t *testing.T) {
 	obj, dirs, err := prepareErasure(t.Context(), 4)
 	if err != nil {
 		t.Fatal(err)
 	}
 	z := obj.(*erasureServerPools)
-	setMultipartListingTestMode(t, false)
+
 	t.Cleanup(func() { z.Shutdown(context.Background()); removeRoots(dirs) })
 	const bucket, otherBucket = "r9-scan-target", "r9-scan-unrelated"
 	for _, name := range []string{bucket, otherBucket} {
@@ -506,7 +550,7 @@ func TestMultipartListingScanCosts(t *testing.T) {
 		t.Helper()
 		directoryCalls.Store(0)
 		metadataCalls.Store(0)
-		result, err := z.ListMultipartUploads(t.Context(), bucket, "dir/", "", "", "", 1)
+		result, err := z.scanMultipartUploadsForTest(t.Context(), bucket, "dir/", "", "", "", 1)
 		if err != nil {
 			t.Fatal(err)
 		}
@@ -544,7 +588,7 @@ func multipartListingFixture(t *testing.T) (*erasureServerPools, *erasureObjects
 	if err := z.MakeBucket(t.Context(), bucket, MakeBucketOptions{}); err != nil {
 		t.Fatal(err)
 	}
-	setMultipartListingTestMode(t, false)
+
 	return z, z.serverPools[0].getHashedSet("a"), bucket
 }
 
@@ -588,46 +632,7 @@ func TestMultipartListingMissingMarkers(t *testing.T) {
 	}
 }
 
-func TestMultipartListingAbortRecovery(t *testing.T) {
+func TestMultipartDiagnosticCoverage(t *testing.T) {
-	for _, offline := range []int{1, 2} {
-		t.Run(fmt.Sprint(offline), func(t *testing.T) {
-			z, set, bucket := multipartListingFixture(t)
-			mp, err := z.NewMultipartUpload(t.Context(), bucket, "a", ObjectOptions{})
-			if err != nil {
-				t.Fatal(err)
-			}
-			original := set.getDisks
-			t.Cleanup(func() { set.getDisks = original })
-			set.getDisks = func() []StorageAPI {
-				disks := append([]StorageAPI(nil), original()...)
-				for i := 0; i < offline; i++ {
-					disks[i] = nil
-				}
-				return disks
-			}
-			err = z.AbortMultipartUpload(t.Context(), bucket, "a", mp.UploadID, ObjectOptions{})
-			if offline == 1 && err != nil {
-				t.Fatal(err)
-			}
-			if offline == 2 && (err == nil || toAPIError(t.Context(), err).HTTPStatusCode != 503) {
-				t.Fatalf("two offline drives must not acknowledge cancellation: %v", err)
-			}
-			set.getDisks = original
-			if offline == 2 {
-				// Retry a partially completed deletion after the original disks return.
-				if err = z.AbortMultipartUpload(t.Context(), bucket, "a", mp.UploadID, ObjectOptions{}); err != nil {
-					t.Fatal(err)
-				}
-			}
-			got, err := z.ListMultipartUploads(t.Context(), bucket, "", "", "", "", 100)
-			if err != nil || len(got.Uploads) != 0 {
-				t.Fatalf("canceled upload reappeared: %+v %v", got, err)
-			}
-		})
-	}
-}
-
-func TestMultipartListingCoverage(t *testing.T) {
 	z, set, bucket := multipartListingFixture(t)
 	mp, err := z.NewMultipartUpload(t.Context(), bucket, "a", ObjectOptions{})
 	if err != nil {
@@ -643,12 +648,12 @@ func TestMultipartListingCoverage(t *testing.T) {
 		}
 	}
 	set.getDisks = func() []StorageAPI { return []StorageAPI{disks[0], disks[1], nil, nil} }
-	_, err = z.ListMultipartUploads(t.Context(), bucket, "", "", "", "", 10)
+	_, err = z.scanMultipartUploadsForTest(t.Context(), bucket, "", "", "", "", 10)
 	if err == nil || toAPIError(t.Context(), err).HTTPStatusCode != 503 {
 		t.Fatalf("incomplete discovery returned success: %v", err)
 	}
 	set.getDisks = func() []StorageAPI { return []StorageAPI{nil, disks[1], disks[2], disks[3]} }
-	got, err := z.ListMultipartUploads(t.Context(), bucket, "", "", "", "", 10)
+	got, err := z.scanMultipartUploadsForTest(t.Context(), bucket, "", "", "", "", 10)
 	if err != nil {
 		t.Fatal(err)
 	}
@@ -659,7 +664,7 @@ func TestMultipartListingCoverage(t *testing.T) {
 	}
 }
 
-func TestMultipartListingBudgetAndAdmission(t *testing.T) {
+func TestMultipartDiagnosticBudgetAndAdmission(t *testing.T) {
 	_, set, bucket := multipartListingFixture(t)
 	if _, err := set.NewMultipartUpload(t.Context(), bucket, "a", ObjectOptions{}); err != nil {
 		t.Fatal(err)
@@ -670,7 +675,7 @@ func TestMultipartListingBudgetAndAdmission(t *testing.T) {
 	}
 	defer scan.close()
 	scan.remaining.Store(1)
-	_, _, err = set.scanMultipartUploads(scan, bucket, 0, 0)
+	_, err = set.scanMultipartUploads(scan, bucket, 0, 0)
 	var limited SlowDown
 	if !errors.As(err, &limited) {
 		t.Fatalf("budget: %v", err)
@@ -689,13 +694,13 @@ func TestMultipartListingBudgetAndAdmission(t *testing.T) {
 	}
 }
 
-func TestMultipartListingAdmissionHTTP(t *testing.T) {
+func TestMultipartPublicListingAdmissionHTTP(t *testing.T) {
 	z, _, _ := multipartListingFixture(t)
 	bucket, router, err := initAPIHandlerTest(t.Context(), z, []string{"ListMultipartUploads"}, MakeBucketOptions{})
 	if err != nil {
 		t.Fatal(err)
 	}
-	setMultipartListingTestMode(t, false)
+
 	for range cap(multipartScanSlots) {
 		scan, err := startMultipartScan(t.Context(), false)
 		if err != nil {
@@ -711,12 +716,12 @@ func TestMultipartListingAdmissionHTTP(t *testing.T) {
 	}
 	rec := httptest.NewRecorder()
 	router.ServeHTTP(rec, req)
-	var response APIErrorResponse
+	var response ListMultipartUploadsResponse
 	if err := xml.Unmarshal(rec.Body.Bytes(), &response); err != nil {
 		t.Fatal(err)
 	}
-	if rec.Code != http.StatusServiceUnavailable || response.Code != "SlowDown" {
+	if rec.Code != http.StatusOK || len(response.Uploads) != 0 {
-		t.Fatalf("admission returned %d %s", rec.Code, rec.Body.String())
+		t.Fatalf("diagnostic admission blocked public listing: %d %s", rec.Code, rec.Body.String())
 	}
 }
 
@@ -743,12 +748,12 @@ func TestMultipartListingPreflightMinorityLegacy(t *testing.T) {
 		}
 	}
 	report, err := z.multipartPreflight(t.Context())
-	if err != nil || report.Ready || !report.Complete || report.LegacyUploads != 1 {
+	if err != nil || report.Ready || !report.Complete || report.LegacyUploads != 1 || report.Mode != "legacy" {
 		t.Fatalf("minority legacy copy must prevent readiness: %+v %v", report, err)
 	}
 }
 
-func TestMultipartListingCancellationRetainsAdmission(t *testing.T) {
+func TestMultipartDiagnosticCancellationRetainsAdmission(t *testing.T) {
 	z, set, bucket := multipartListingFixture(t)
 	for i := range 40 {
 		if _, err := z.NewMultipartUpload(t.Context(), bucket, fmt.Sprintf("key-%02d", i), ObjectOptions{}); err != nil {
@@ -787,7 +792,7 @@ func TestMultipartListingCancellationRetainsAdmission(t *testing.T) {
 	defer cancel()
 	done := make(chan error, 1)
 	go func() {
-		_, err := z.ListMultipartUploads(ctx, bucket, "", "", "", "", 10)
+		_, err := z.scanMultipartUploadsForTest(ctx, bucket, "", "", "", "", 10)
 		done <- err
 	}()
 	for range 16 {

cmd/erasure-multipart-migration_test.go

@@ -12,41 +12,13 @@ import (
 	"sync/atomic"
 	"testing"
 
-	"github.com/minio/minio/internal/config/api"
 	"github.com/minio/minio/internal/config/storageclass"
 	"github.com/minio/minio/internal/dsync"
 	"github.com/minio/minio/internal/grid"
 	xnet "github.com/pgsty/silo-pkg/v3/net"
 )
 
-func setMultipartListingTestMode(t *testing.T, legacy bool) {
+func TestMultipartAbortDefaultAvailability(t *testing.T) {
-	t.Helper()
-	globalAPIConfig.mu.Lock()
-	previous := globalAPIConfig.multipartListingStrict
-	globalAPIConfig.multipartListingStrict = !legacy
-	globalAPIConfig.mu.Unlock()
-	t.Cleanup(func() {
-		globalAPIConfig.mu.Lock()
-		globalAPIConfig.multipartListingStrict = previous
-		globalAPIConfig.mu.Unlock()
-	})
-}
-
-func TestMultipartListingDefaultMode(t *testing.T) {
-	var uninitialized apiConfig
-	if !uninitialized.getMultipartListingLegacy() {
-		t.Fatal("runtime zero value enabled strict mode before config initialization")
-	}
-	for _, mode := range []string{"", "legacy", "strict"} {
-		var local apiConfig
-		local.init(api.Config{RequestsMax: 1, MultipartListing: mode}, []int{4}, false)
-		if local.getMultipartListingLegacy() != (mode != "strict") {
-			t.Fatalf("unexpected mode for %q", mode)
-		}
-	}
-}
-
-func TestMultipartAbortLegacyAvailability(t *testing.T) {
 	for _, drives := range []int{4, 16} {
 		for _, offline := range []int{0, 1, drives / 2} {
 			t.Run(fmt.Sprintf("drives=%d/offline=%d", drives, offline), func(t *testing.T) {
@@ -56,7 +28,7 @@ func TestMultipartAbortLegacyAvailability(t *testing.T) {
 				}
 				z := obj.(*erasureServerPools)
 				t.Cleanup(func() { z.Shutdown(context.Background()); removeRoots(dirs) })
-				setMultipartListingTestMode(t, true)
+
 				previous := globalStorageClass
 				globalStorageClass.Update(storageclass.Config{Standard: storageclass.StorageClass{Parity: drives / 2}})
 				t.Cleanup(func() { globalStorageClass.Update(previous) })
@@ -91,11 +63,11 @@ func TestMultipartAbortLegacyAvailability(t *testing.T) {
 	}
 }
 
-func TestMultipartAbortLegacyPoolOrder(t *testing.T) {
+func TestMultipartAbortDefaultPoolOrder(t *testing.T) {
 	for _, owner := range []int{0, 1} {
 		t.Run(fmt.Sprintf("owner=%d", owner), func(t *testing.T) {
 			z, bucket := consistencyPools(t)
-			setMultipartListingTestMode(t, true)
+
 			mp, err := z.serverPools[owner].NewMultipartUpload(t.Context(), bucket, "a", ObjectOptions{})
 			if err != nil {
 				t.Fatal(err)
@@ -106,7 +78,7 @@ func TestMultipartAbortLegacyPoolOrder(t *testing.T) {
 			emptySet.getDisks = func() []StorageAPI { return make([]StorageAPI, emptySet.setDriveCount) }
 			err = z.AbortMultipartUpload(t.Context(), bucket, "a", mp.UploadID, ObjectOptions{})
 			if owner == 0 && err != nil {
-				t.Fatalf("unrelated later pool blocked legacy cancellation: %v", err)
+				t.Fatalf("unrelated later pool blocked cancellation: %v", err)
 			}
 			if owner == 1 {
 				if err == nil || toAPIError(t.Context(), err).HTTPStatusCode != 503 {
@@ -124,123 +96,56 @@ func TestMultipartAbortLegacyPoolOrder(t *testing.T) {
 	}
 }
 
-func TestMultipartAbortMinorityLegacyCleanup(t *testing.T) {
-	for _, failDelete := range []bool{false, true} {
-		t.Run(fmt.Sprintf("delete-fails=%v", failDelete), func(t *testing.T) {
-			z, set, bucket := multipartListingFixture(t)
-			mp, err := z.NewMultipartUpload(t.Context(), bucket, "old", ObjectOptions{})
-			if err != nil {
-				t.Fatal(err)
-			}
-			path := set.getUploadIDDir(bucket, "old", mp.UploadID)
-			original := set.getDisks
-			disks := original()
-			fi, err := disks[0].ReadVersion(t.Context(), bucket, minioMetaMultipartBucket, path, "", ReadOptions{})
-			if err != nil {
-				t.Fatal(err)
-			}
-			delete(fi.Metadata, multipartMetaBucket)
-			delete(fi.Metadata, multipartMetaObject)
-			if err := disks[0].WriteMetadata(t.Context(), bucket, minioMetaMultipartBucket, path, fi); err != nil {
-				t.Fatal(err)
-			}
-			for _, d := range disks[1:] {
-				if err := d.Delete(t.Context(), minioMetaMultipartBucket, path, DeleteOptions{Recursive: true}); err != nil {
-					t.Fatal(err)
-				}
-			}
-			report, err := z.multipartPreflight(t.Context())
-			if err != nil || report.Ready || report.LegacyUploads != 1 {
-				t.Fatalf("invalid legacy fixture: %+v %v", report, err)
-			}
-			if failDelete {
-				set.getDisks = func() []StorageAPI {
-					wrapped := append([]StorageAPI(nil), disks...)
-					wrapped[0] = multipartListingFaultDisk{StorageAPI: disks[0], delete: func(context.Context, string, string, DeleteOptions) error { return errFaultyDisk }}
-					return wrapped
-				}
-				t.Cleanup(func() { set.getDisks = original })
-				err := z.AbortMultipartUpload(t.Context(), bucket, "old", mp.UploadID, ObjectOptions{})
-				if err == nil || toAPIError(t.Context(), err).HTTPStatusCode != 503 {
-					t.Fatalf("empty drives masked failed deletion of the observed replica: %v", err)
-				}
-				if _, present := z.mpCache.Load(mp.UploadID); !present {
-					t.Fatal("failed cancellation evicted upload from cache")
-				}
-				set.getDisks = original
-			}
-			if err := z.AbortMultipartUpload(t.Context(), bucket, "old", mp.UploadID, ObjectOptions{}); err != nil {
-				t.Fatal(err)
-			}
-			report, err = z.multipartPreflight(t.Context())
-			if err != nil || !report.Ready || report.LegacyUploads != 0 {
-				t.Fatalf("acknowledged cleanup left online legacy remnants: %+v %v", report, err)
-			}
-			if _, present := z.mpCache.Load(mp.UploadID); present {
-				t.Fatal("successful cancellation retained cache entry")
-			}
-		})
-	}
-}
-
 func TestMultipartAbortWrongTargetKeepsCache(t *testing.T) {
-	for _, legacy := range []bool{true, false} {
+	z, _, bucket := multipartListingFixture(t)
-		t.Run(fmt.Sprintf("legacy=%v", legacy), func(t *testing.T) {
+
-			z, _, bucket := multipartListingFixture(t)
+	const other = "multipart-other-bucket"
-			setMultipartListingTestMode(t, legacy)
+	if err := z.MakeBucket(t.Context(), other, MakeBucketOptions{}); err != nil {
-			const other = "multipart-other-bucket"
+		t.Fatal(err)
-			if err := z.MakeBucket(t.Context(), other, MakeBucketOptions{}); err != nil {
+	}
-				t.Fatal(err)
+	mp, err := z.NewMultipartUpload(t.Context(), bucket, "valid-key", ObjectOptions{})
-			}
+	if err != nil {
-			mp, err := z.NewMultipartUpload(t.Context(), bucket, "valid-key", ObjectOptions{})
+		t.Fatal(err)
-			if err != nil {
+	}
-				t.Fatal(err)
+	for _, target := range [][2]string{{bucket, "wrong-key"}, {other, "valid-key"}} {
-			}
+		err := z.AbortMultipartUpload(t.Context(), target[0], target[1], mp.UploadID, ObjectOptions{})
-			for _, target := range [][2]string{{bucket, "wrong-key"}, {other, "valid-key"}} {
+		var invalid InvalidUploadID
-				err := z.AbortMultipartUpload(t.Context(), target[0], target[1], mp.UploadID, ObjectOptions{})
+		if !errors.As(err, &invalid) {
-				var invalid InvalidUploadID
+			t.Fatalf("wrong target result: %v", err)
-				if !errors.As(err, &invalid) {
+		}
-					t.Fatalf("wrong target result: %v", err)
+		if _, present := z.mpCache.Load(mp.UploadID); !present {
-				}
+			t.Fatal("wrong-target abort evicted another upload")
-				if _, present := z.mpCache.Load(mp.UploadID); !present {
+		}
-					t.Fatal("wrong-target abort evicted another upload")
+		if _, err := z.GetMultipartInfo(t.Context(), bucket, "valid-key", mp.UploadID, ObjectOptions{}); err != nil {
-				}
+			t.Fatalf("wrong-target abort harmed valid upload: %v", err)
-				if _, err := z.GetMultipartInfo(t.Context(), bucket, "valid-key", mp.UploadID, ObjectOptions{}); err != nil {
+		}
-					t.Fatalf("wrong-target abort harmed valid upload: %v", err)
+	}
-				}
+	if err := z.AbortMultipartUpload(t.Context(), bucket, "valid-key", mp.UploadID, ObjectOptions{}); err != nil {
-			}
+		t.Fatal(err)
-			if err := z.AbortMultipartUpload(t.Context(), bucket, "valid-key", mp.UploadID, ObjectOptions{}); err != nil {
+	}
-				t.Fatal(err)
+	var invalid InvalidUploadID
-			}
+	if err := z.AbortMultipartUpload(t.Context(), bucket, "valid-key", mp.UploadID, ObjectOptions{}); !errors.As(err, &invalid) {
-			var invalid InvalidUploadID
+		t.Fatalf("repeat abort: %v", err)
-			if err := z.AbortMultipartUpload(t.Context(), bucket, "valid-key", mp.UploadID, ObjectOptions{}); !errors.As(err, &invalid) {
-				t.Fatalf("repeat abort: %v", err)
-			}
-		})
 	}
 }
 
 func TestMultipartAbortRejectsUnsafeID(t *testing.T) {
-	for _, legacy := range []bool{true, false} {
+	z, _, bucket := multipartListingFixture(t)
-		t.Run(fmt.Sprintf("legacy=%v", legacy), func(t *testing.T) {
+
-			z, _, bucket := multipartListingFixture(t)
+	mp, err := z.NewMultipartUpload(t.Context(), bucket, "keep", ObjectOptions{})
-			setMultipartListingTestMode(t, legacy)
+	if err != nil {
-			mp, err := z.NewMultipartUpload(t.Context(), bucket, "keep", ObjectOptions{})
+		t.Fatal(err)
-			if err != nil {
+	}
-				t.Fatal(err)
+	for _, suffix := range []string{".", "..", "../other", "a/b", "a\\b", ""} {
-			}
+		id := base64.RawURLEncoding.EncodeToString([]byte("deployment." + suffix))
-			for _, suffix := range []string{".", "..", "../other", "a/b", "a\\b", ""} {
+		err := z.AbortMultipartUpload(t.Context(), bucket, "keep", id, ObjectOptions{})
-				id := base64.RawURLEncoding.EncodeToString([]byte("deployment." + suffix))
+		var invalid InvalidUploadID
-				err := z.AbortMultipartUpload(t.Context(), bucket, "keep", id, ObjectOptions{})
+		if !errors.As(err, &invalid) {
-				var invalid InvalidUploadID
+			t.Fatalf("unsafe ID accepted: suffix=%q err=%v", suffix, err)
-				if !errors.As(err, &invalid) {
+		}
-					t.Fatalf("unsafe ID accepted: suffix=%q err=%v", suffix, err)
+		if _, err := z.GetMultipartInfo(t.Context(), bucket, "keep", mp.UploadID, ObjectOptions{}); err != nil {
-				}
+			t.Fatalf("valid upload harmed by rejected ID: %v", err)
-				if _, err := z.GetMultipartInfo(t.Context(), bucket, "keep", mp.UploadID, ObjectOptions{}); err != nil {
+		}
-					t.Fatalf("valid upload harmed by rejected ID: %v", err)
-				}
-			}
-		})
 	}
 }
 
@@ -275,34 +180,30 @@ func TestMultipartAbortPeerNotification(t *testing.T) {
 	set.nsMutex = newNSLock(true)
 	set.getLockers = func() ([]dsync.NetLocker, string) { return lockers, "multipart-test" }
 	t.Cleanup(func() { set.nsMutex, set.getLockers = oldMutex, oldLockers })
-	for _, legacy := range []bool{true, false} {
+
-		t.Run(fmt.Sprintf("legacy=%v", legacy), func(t *testing.T) {
+	for range 4 {
-			setMultipartListingTestMode(t, legacy)
+		mp, err := z.NewMultipartUpload(t.Context(), bucket, "valid", ObjectOptions{})
-			for range 4 {
+		if err != nil {
-				mp, err := z.NewMultipartUpload(t.Context(), bucket, "valid", ObjectOptions{})
+			t.Fatal(err)
-				if err != nil {
+		}
-					t.Fatal(err)
+		before := notifications.Load()
-				}
+		var invalid InvalidUploadID
-				before := notifications.Load()
+		if err := z.AbortMultipartUpload(t.Context(), bucket, "wrong", mp.UploadID, ObjectOptions{}); !errors.As(err, &invalid) {
-				var invalid InvalidUploadID
+			t.Fatalf("wrong target: %v", err)
-				if err := z.AbortMultipartUpload(t.Context(), bucket, "wrong", mp.UploadID, ObjectOptions{}); !errors.As(err, &invalid) {
+		}
-					t.Fatalf("wrong target: %v", err)
+		if notifications.Load() != before {
-				}
+			t.Fatal("wrong-target abort sent a destructive peer notification")
-				if notifications.Load() != before {
+		}
-					t.Fatal("wrong-target abort sent a destructive peer notification")
+		if err := z.AbortMultipartUpload(t.Context(), bucket, "valid", mp.UploadID, ObjectOptions{}); err != nil {
-				}
+			t.Fatal(err)
-				if err := z.AbortMultipartUpload(t.Context(), bucket, "valid", mp.UploadID, ObjectOptions{}); err != nil {
+		}
-					t.Fatal(err)
+		if notifications.Load() != before+1 {
-				}
+			t.Fatal("successful abort lost peer notification after unlocking")
-				if notifications.Load() != before+1 {
+		}
-					t.Fatal("successful abort lost peer notification after unlocking")
-				}
-			}
-		})
 	}
 }
 
-func TestMultipartAbortStrictMajority(t *testing.T) {
+func TestMultipartAbortBestEffortCleanup(t *testing.T) {
 	z, set, bucket := multipartListingFixture(t)
 	mp, err := z.NewMultipartUpload(t.Context(), bucket, "a", ObjectOptions{})
 	if err != nil {
@@ -317,15 +218,19 @@ func TestMultipartAbortStrictMajority(t *testing.T) {
 	}
 	t.Cleanup(func() { set.getDisks = original })
 	if err := z.AbortMultipartUpload(t.Context(), bucket, "a", mp.UploadID, ObjectOptions{}); err != nil {
-		t.Fatalf("ordinary majority cancellation was tightened: %v", err)
+		t.Fatalf("read-quorum cancellation was tightened: %v", err)
 	}
-	// One failed deletion is tolerated in the ordinary majority path. Retrying
+	// Best-effort deletion may leave a remnant. Retrying after recovery cleans
-	// after recovery must also clean the now-minority remnant.
+	// it, but the read quorum no longer recognizes it as an active upload.
 	if _, err := disks[0].ReadVersion(t.Context(), bucket, minioMetaMultipartBucket, set.getUploadIDDir(bucket, "a", mp.UploadID), "", ReadOptions{}); err != nil {
 		t.Fatalf("fault fixture did not leave a remnant: %v", err)
 	}
 	set.getDisks = original
-	if err := z.AbortMultipartUpload(t.Context(), bucket, "a", mp.UploadID, ObjectOptions{}); err != nil {
+	var invalid InvalidUploadID
-		t.Fatal(err)
+	if err := z.AbortMultipartUpload(t.Context(), bucket, "a", mp.UploadID, ObjectOptions{}); !errors.As(err, &invalid) {
+		t.Fatalf("minority retry must preserve default absent-upload result: %v", err)
+	}
+	if _, err := disks[0].ReadVersion(t.Context(), bucket, minioMetaMultipartBucket, set.getUploadIDDir(bucket, "a", mp.UploadID), "", ReadOptions{}); !errors.Is(err, errFileNotFound) {
+		t.Fatalf("best-effort retry did not clean remnant: %v", err)
 	}
 }

cmd/erasure-multipart-public_test.go

@@ -0,0 +1,106 @@
+// Copyright (c) 2026 Ruohang Feng
+// SPDX-License-Identifier: AGPL-3.0-or-later
+
+package cmd
+
+import (
+	"context"
+	"encoding/xml"
+	"net/http"
+	"net/http/httptest"
+	"sync/atomic"
+	"testing"
+)
+
+type multipartRootScanGuard struct {
+	StorageAPI
+	roots *atomic.Int64
+}
+
+func (d multipartRootScanGuard) ListDir(ctx context.Context, original, volume, dir string, count int) ([]string, error) {
+	if volume == minioMetaMultipartBucket && dir == "" {
+		d.roots.Add(1)
+		return nil, errFaultyDisk
+	}
+	return d.StorageAPI.ListDir(ctx, original, volume, dir, count)
+}
+
+// A busy or unreadable diagnostic namespace must not become a prerequisite for
+// ordinary listing, including the lower-level exact-object entry points.
+func TestMultipartPublicListingIndependentOfDiagnostics(t *testing.T) {
+	for _, busy := range []bool{false, true} {
+		t.Run(map[bool]string{false: "unreadable-root", true: "busy-diagnostics"}[busy], func(t *testing.T) {
+			z, _ := consistencyPools(t)
+			bucket, router, err := initAPIHandlerTest(t.Context(), z, []string{"ListMultipartUploads"}, MakeBucketOptions{})
+			if err != nil {
+				t.Fatal(err)
+			}
+			const key = "dir/known"
+			mp, err := z.NewMultipartUpload(t.Context(), bucket, key, ObjectOptions{})
+			if err != nil {
+				t.Fatal(err)
+			}
+			var roots atomic.Int64
+			for _, pool := range z.serverPools {
+				for _, set := range pool.sets {
+					original := set.getDisks
+					set.getDisks = func() []StorageAPI {
+						disks := append([]StorageAPI(nil), original()...)
+						for i, disk := range disks {
+							if disk != nil {
+								disks[i] = multipartRootScanGuard{disk, &roots}
+							}
+						}
+						return disks
+					}
+					t.Cleanup(func() { set.getDisks = original })
+				}
+			}
+			if busy {
+				for range cap(multipartScanSlots) {
+					scan, err := startMultipartScan(t.Context(), true)
+					if err != nil {
+						t.Fatal(err)
+					}
+					defer scan.close()
+				}
+			}
+			for _, prefix := range []string{"", key, "dir/"} {
+				req, err := newTestSignedRequestV4(http.MethodGet,
+					getListMultipartUploadsURLWithParams("", bucket, prefix, "", "", "", "1"),
+					0, nil, globalActiveCred.AccessKey, globalActiveCred.SecretKey, nil)
+				if err != nil {
+					t.Fatal(err)
+				}
+				rec := httptest.NewRecorder()
+				router.ServeHTTP(rec, req)
+				if rec.Code != http.StatusOK {
+					t.Fatalf("prefix %q: HTTP %d %s", prefix, rec.Code, rec.Body.String())
+				}
+				var got ListMultipartUploadsResponse
+				if err := xml.Unmarshal(rec.Body.Bytes(), &got); err != nil {
+					t.Fatal(err)
+				}
+				if prefix == "dir/" {
+					if len(got.Uploads) != 0 {
+						t.Fatalf("partial prefix must retain the documented exact-key limitation: %+v", got)
+					}
+				} else if len(got.Uploads) != 1 || got.Uploads[0].UploadID != mp.UploadID {
+					t.Fatalf("known upload missing: %+v", got)
+				}
+			}
+			for _, pool := range z.serverPools {
+				for _, list := range []func(context.Context, string, string, string, string, string, int) (ListMultipartsInfo, error){
+					pool.ListMultipartUploads, pool.getHashedSet(key).ListMultipartUploads,
+				} {
+					if _, err := list(t.Context(), bucket, key, "", "", "", 10); err != nil {
+						t.Fatalf("lower-level exact listing: %v", err)
+					}
+				}
+			}
+			if roots.Load() != 0 {
+				t.Fatalf("public listing attempted %d global multipart scans", roots.Load())
+			}
+		})
+	}
+}

cmd/erasure-multipart-unified_test.go

@@ -0,0 +1,256 @@
+// Copyright (c) 2026 Ruohang Feng
+// SPDX-License-Identifier: AGPL-3.0-or-later
+
+package cmd
+
+import (
+	"bytes"
+	"context"
+	"errors"
+	"net/http"
+	"slices"
+	"testing"
+)
+
+func rewriteMultipartIdentity(t testing.TB, set *erasureObjects, bucket, key, id string, change func(map[string]string)) {
+	t.Helper()
+	fi, metadata, err := set.checkUploadIDExists(t.Context(), bucket, key, id, true)
+	if err != nil {
+		t.Fatal(err)
+	}
+	for i := range metadata {
+		change(metadata[i].Metadata)
+	}
+	if _, err := writeAllMetadata(t.Context(), set.getDisks(), bucket, minioMetaMultipartBucket,
+		set.getUploadIDDir(bucket, key, id), metadata, fi.WriteQuorum(set.defaultWQuorum())); err != nil {
+		t.Fatal(err)
+	}
+}
+
+func removeMultipartIdentity(metadata map[string]string) {
+	delete(metadata, multipartMetaBucket)
+	delete(metadata, multipartMetaObject)
+}
+
+// The retained diagnostic scanner handles mixed records independently of a
+// stale cache. Public HTTP behavior is tested separately.
+func TestMultipartDiagnosticMixed(t *testing.T) {
+	z, set, bucket := multipartListingFixture(t)
+	for _, key := range []string{"dir/old", "dir/a", "dir/sub/b", "outside"} {
+		mp, err := z.NewMultipartUpload(t.Context(), bucket, key, ObjectOptions{})
+		if err != nil {
+			t.Fatal(err)
+		}
+		if key == "dir/old" {
+			rewriteMultipartIdentity(t, set, bucket, key, mp.UploadID, removeMultipartIdentity)
+		}
+	}
+	z.mpCache.Clear()
+	z.mpCache.Store("stale", MultipartInfo{Bucket: bucket, Object: "ghost", UploadID: "stale"})
+	list := func(prefix, keyMarker, uploadMarker, delimiter string) ListMultipartUploadsResponse {
+		t.Helper()
+		got, err := z.scanMultipartUploadsForTest(t.Context(), bucket, prefix, keyMarker, uploadMarker, delimiter, 1)
+		if err != nil {
+			t.Fatal(err)
+		}
+		return generateListMultipartUploadsResponse(bucket, got, "")
+	}
+	var keys []string
+	keyMarker, uploadMarker := "", ""
+	for page := 0; ; page++ {
+		if page >= 4 {
+			t.Fatal("pagination did not terminate")
+		}
+		got := list("", keyMarker, uploadMarker, "")
+		if len(got.Uploads) != 1 {
+			t.Fatalf("page size: %+v", got)
+		}
+		keys = append(keys, got.Uploads[0].Key)
+		if !got.IsTruncated {
+			break
+		}
+		keyMarker, uploadMarker = got.NextKeyMarker, got.NextUploadIDMarker
+	}
+	if !slices.Equal(keys, []string{"dir/a", "dir/sub/b", "outside"}) {
+		t.Fatalf("mixed full listing: %v", keys)
+	}
+	first := list("dir/", "", "", "/")
+	if len(first.Uploads) != 1 || first.Uploads[0].Key != "dir/a" || !first.IsTruncated {
+		t.Fatalf("prefix first page: %+v", first)
+	}
+	rest := list("dir/", first.NextKeyMarker, first.NextUploadIDMarker, "/")
+	if len(rest.Uploads) != 0 || rest.IsTruncated || len(rest.CommonPrefixes) != 1 || rest.CommonPrefixes[0].Prefix != "dir/sub/" {
+		t.Fatalf("prefix continuation: %+v", rest)
+	}
+	if got := list("dir/old", "", "", ""); len(got.Uploads) != 0 || got.IsTruncated {
+		t.Fatalf("historical exact-looking prefix: %+v", got)
+	}
+}
+
+func TestMultipartDiagnosticInvalidIdentity(t *testing.T) {
+	for _, kind := range []string{"missing-bucket", "missing-object", "empty-bucket", "empty-object", "empty-both", "wrong-hash", "read-failure"} {
+		t.Run(kind, func(t *testing.T) {
+			z, set, bucket := multipartListingFixture(t)
+			mp, err := z.NewMultipartUpload(t.Context(), bucket, "a", ObjectOptions{})
+			if err != nil {
+				t.Fatal(err)
+			}
+			rewriteMultipartIdentity(t, set, bucket, "a", mp.UploadID, func(metadata map[string]string) {
+				switch kind {
+				case "missing-bucket":
+					delete(metadata, multipartMetaBucket)
+				case "missing-object":
+					delete(metadata, multipartMetaObject)
+				case "empty-bucket":
+					metadata[multipartMetaBucket] = ""
+				case "empty-object":
+					metadata[multipartMetaObject] = ""
+				case "empty-both":
+					metadata[multipartMetaBucket], metadata[multipartMetaObject] = "", ""
+				case "wrong-hash":
+					metadata[multipartMetaObject] = "wrong-key"
+				case "read-failure":
+					removeMultipartIdentity(metadata)
+				}
+			})
+			if kind == "read-failure" {
+				original := set.getDisks
+				t.Cleanup(func() { set.getDisks = original })
+				set.getDisks = func() []StorageAPI {
+					disks := append([]StorageAPI(nil), original()...)
+					for i, disk := range disks {
+						disks[i] = multipartListingFaultDisk{StorageAPI: disk, read: func(context.Context, string, string, string, string, ReadOptions) (FileInfo, error) {
+							return FileInfo{}, errFaultyDisk
+						}}
+					}
+					return disks
+				}
+			}
+			got, err := z.scanMultipartUploadsForTest(t.Context(), bucket, "", "", "", "", 10)
+			if err == nil || len(got.Uploads) != 0 || toAPIError(t.Context(), err).HTTPStatusCode != http.StatusServiceUnavailable {
+				t.Fatalf("invalid/unreadable record returned a successful page: %+v %v", got, err)
+			}
+			if kind != "read-failure" && !errors.Is(err, errMultipartListingIdentity) {
+				t.Fatalf("identity error: %v", err)
+			}
+			report, err := z.multipartPreflight(t.Context())
+			if err != nil || report.Complete || report.Ready || report.LegacyUploads != 0 {
+				t.Fatalf("invalid/unreadable record counted as historical: %+v %v", report, err)
+			}
+		})
+	}
+}
+
+func TestMultipartListingRestart(t *testing.T) {
+	ctx, cancel := context.WithCancel(t.Context())
+	obj, dirs, err := prepareErasure(ctx, 4)
+	if err != nil {
+		cancel()
+		t.Fatal(err)
+	}
+	t.Cleanup(func() { cancel(); obj.Shutdown(context.Background()); removeRoots(dirs) })
+	const bucket, key = "multipart-restart", "dir/new"
+	if err := obj.MakeBucket(t.Context(), bucket, MakeBucketOptions{}); err != nil {
+		t.Fatal(err)
+	}
+	mp, err := obj.NewMultipartUpload(t.Context(), bucket, key, ObjectOptions{})
+	if err != nil {
+		t.Fatal(err)
+	}
+	cancel()
+	if err := obj.Shutdown(context.Background()); err != nil {
+		t.Fatal(err)
+	}
+	reopened, _, err := initObjectLayer(t.Context(), mustGetPoolEndpoints(0, dirs...))
+	if err != nil {
+		t.Fatal(err)
+	}
+	obj = reopened
+	if _, cached := obj.(*erasureServerPools).mpCache.Load(mp.UploadID); cached {
+		t.Fatal("restart unexpectedly preserved the upload cache")
+	}
+	for _, prefix := range []string{"", "dir/"} {
+		got, err := obj.ListMultipartUploads(t.Context(), bucket, prefix, "", "", "", 10)
+		if err != nil {
+			t.Fatal(err)
+		}
+		requireMultipartUploadKeys(t, got)
+	}
+	got, err := obj.ListMultipartUploads(t.Context(), bucket, key, "", "", "", 10)
+	if err != nil {
+		t.Fatal(err)
+	}
+	requireMultipartUploadKeys(t, got, key)
+	if got.Uploads[0].UploadID != mp.UploadID {
+		t.Fatal("restart changed upload identity")
+	}
+	part, err := obj.PutObjectPart(t.Context(), bucket, key, mp.UploadID, 1,
+		mustGetPutObjReader(t, bytes.NewBufferString("part"), 4, "", ""), ObjectOptions{})
+	if err != nil {
+		t.Fatalf("continue known upload after restart: %v", err)
+	}
+	if _, err := obj.CompleteMultipartUpload(t.Context(), bucket, key, mp.UploadID,
+		[]CompletePart{{PartNumber: 1, ETag: part.ETag}}, ObjectOptions{}); err != nil {
+		t.Fatalf("complete known upload after restart: %v", err)
+	}
+}
+
+func TestMultipartDiagnosticConflictingIdentityCopies(t *testing.T) {
+	for _, oldCopies := range []int{1, 3} {
+		z, set, bucket := multipartListingFixture(t)
+		mp, err := z.NewMultipartUpload(t.Context(), bucket, "a", ObjectOptions{})
+		if err != nil {
+			t.Fatal(err)
+		}
+		path := set.getUploadIDDir(bucket, "a", mp.UploadID)
+		for _, disk := range set.getDisks()[:oldCopies] {
+			fi, err := disk.ReadVersion(t.Context(), bucket, minioMetaMultipartBucket, path, "", ReadOptions{})
+			if err != nil {
+				t.Fatal(err)
+			}
+			removeMultipartIdentity(fi.Metadata)
+			if err := disk.WriteMetadata(t.Context(), bucket, minioMetaMultipartBucket, path, fi); err != nil {
+				t.Fatal(err)
+			}
+		}
+		got, err := z.scanMultipartUploadsForTest(t.Context(), bucket, "", "", "", "", 10)
+		if !errors.Is(err, errMultipartListingIdentity) || len(got.Uploads) != 0 {
+			t.Fatalf("%d contradictory copies returned a successful page: %+v %v", oldCopies, got, err)
+		}
+	}
+}
+
+func TestMultipartDiagnosticUnsupportedStoredID(t *testing.T) {
+	for _, historical := range []bool{false, true} {
+		z, set, bucket := multipartListingFixture(t)
+		mp, err := z.NewMultipartUpload(t.Context(), bucket, "a", ObjectOptions{})
+		if err != nil {
+			t.Fatal(err)
+		}
+		oldPath := set.getUploadIDDir(bucket, "a", mp.UploadID)
+		unsupportedPath := pathJoin(set.getMultipartSHADir(bucket, "a"), "unsupported-id")
+		for _, disk := range set.getDisks() {
+			fi, err := disk.ReadVersion(t.Context(), bucket, minioMetaMultipartBucket, oldPath, "", ReadOptions{})
+			if err != nil {
+				t.Fatal(err)
+			}
+			if historical {
+				removeMultipartIdentity(fi.Metadata)
+			}
+			if err := disk.WriteMetadata(t.Context(), bucket, minioMetaMultipartBucket, unsupportedPath, fi); err != nil {
+				t.Fatal(err)
+			}
+			if err := disk.Delete(t.Context(), minioMetaMultipartBucket, oldPath, DeleteOptions{Recursive: true}); err != nil {
+				t.Fatal(err)
+			}
+		}
+		got, err := z.scanMultipartUploadsForTest(t.Context(), bucket, "", "", "", "", 10)
+		if historical {
+			if err != nil || len(got.Uploads) != 0 {
+				t.Fatalf("unidentified historical ID: %+v %v", got, err)
+			}
+		} else if !errors.Is(err, errMultipartListingIdentity) {
+			t.Fatalf("invalid new ID silently treated as historical: %+v %v", got, err)
+		}
+	}
+}

cmd/erasure-multipart.go

@@ -197,6 +197,7 @@ func (er erasureObjects) cleanupStaleUploadsOnDisk(ctx context.Context, disk Sto
 	drivePath := disk.Endpoint().Path
 
 	readDirFn(pathJoin(drivePath, minioMetaMultipartBucket), func(shaDir string, typ os.FileMode) error {
+		hasSkippedUploads := false
 		readDirFn(pathJoin(drivePath, minioMetaMultipartBucket, shaDir), func(uploadIDDir string, typ os.FileMode) error {
 			uploadIDPath := pathJoin(shaDir, uploadIDDir)
 			modTime := extractModTimeFromUploadIDDir(uploadIDDir)
@@ -205,12 +206,14 @@ func (er erasureObjects) cleanupStaleUploadsOnDisk(ctx context.Context, disk Sto
 				wait := deleteMultipartCleanupSleeper.Timer(ctx)
 				fi, err := disk.ReadVersion(ctx, "", minioMetaMultipartBucket, uploadIDPath, "", ReadOptions{})
 				if err != nil {
+					hasSkippedUploads = true
 					return nil
 				}
 				modTime = fi.ModTime
 				wait()
 			}
 			if time.Since(modTime) < globalAPIConfig.getStaleUploadsExpiry() {
+				hasSkippedUploads = true
 				return nil
 			}
 			w := xioutil.NewDeadlineWorker(globalDriveConfig.GetMaxTimeout())
@@ -223,24 +226,15 @@ func (er erasureObjects) cleanupStaleUploadsOnDisk(ctx context.Context, disk Sto
 				return nil
 			})
 		})
-		// Get the modtime of the shaDir.
+		if hasSkippedUploads {
-		vi, err := disk.StatVol(ctx, pathJoin(minioMetaMultipartBucket, shaDir))
-		if err != nil {
-			return nil
-		}
-		// Modtime is returned in the Created field. See (*xlStorage).StatVol
-		if time.Since(vi.Created) < globalAPIConfig.getStaleUploadsExpiry() {
 			return nil
 		}
 		w := xioutil.NewDeadlineWorker(globalDriveConfig.GetMaxTimeout())
 		return w.Run(func() error {
 			wait := deleteMultipartCleanupSleeper.Timer(ctx)
-			pathUUID := mustGetUUID()
+			// Removing an upload updates the parent's mtime. Remove the parent
-			targetPath := pathJoin(drivePath, minioMetaTmpDeletedBucket, pathUUID)
+			// immediately if it is empty; other uploads keep it in place.
-
+			_ = Remove(pathJoin(drivePath, minioMetaMultipartBucket, shaDir))
-			// We are not deleting shaDir recursively here, if shaDir is empty
-			// and its older then we can happily delete it.
-			Rename(pathJoin(drivePath, minioMetaMultipartBucket, shaDir), targetPath)
 			wait()
 			return nil
 		})
@@ -414,7 +408,7 @@ func paginateMultipartUploads(uploads []MultipartInfo, prefix, keyMarker, upload
 }
 
 // listMultipartUploadsExact preserves the hashed exact-object lookup used by
-// multipart write placement and by rolling-upgrade legacy mode.
+// multipart write placement and nonempty public listing prefixes.
 func (er erasureObjects) listMultipartUploadsExact(ctx context.Context, bucket, object, keyMarker, uploadIDMarker, delimiter string, maxUploads int) (result ListMultipartsInfo, err error) {
 	auditObjectErasureSet(ctx, "ListMultipartUploads", object, &er)
 
@@ -522,19 +516,7 @@ func (er erasureObjects) ListMultipartUploads(ctx context.Context, bucket, prefi
 	if err := checkListMultipartArgs(ctx, bucket, prefix, keyMarker, uploadIDMarker, delimiter); err != nil {
 		return ListMultipartsInfo{}, err
 	}
-	scan, err := startMultipartScan(ctx, false)
+	return er.listMultipartUploadsExact(ctx, bucket, prefix, keyMarker, uploadIDMarker, delimiter, maxUploads)
-	if err != nil {
-		return ListMultipartsInfo{}, err
-	}
-	defer scan.close()
-	uploads, legacy, err := er.scanMultipartUploads(scan, bucket, 0, 0)
-	if err != nil {
-		return ListMultipartsInfo{}, err
-	}
-	if legacy {
-		return ListMultipartsInfo{}, errMultipartListingLegacy
-	}
-	return paginateMultipartUploads(uploads, prefix, keyMarker, uploadIDMarker, delimiter, maxUploads), nil
 }
 
 // newMultipartUpload - wrapper for initializing a new multipart
@@ -1788,11 +1770,9 @@ func (er erasureObjects) CompleteMultipartUpload(ctx context.Context, bucket str
 	return fi.ToObjectInfo(bucket, object, opts.Versioned || opts.VersionSuspended), nil
 }
 
-// abortMultipartUpload retains read-quorum validation and best-effort cleanup
+// abortMultipartUpload retains read-quorum validation and best-effort cleanup.
-// in legacy mode. Strict mode requires majority deletion acknowledgements and
+// It does not fence creation writes still executing after a storage timeout.
-// permits retrying remnants below read quorum. Neither mode fences creation
+func (er erasureObjects) abortMultipartUpload(ctx context.Context, bucket, object, uploadID string, opts ObjectOptions) (bool, error) {
-// writes still executing after a storage timeout.
-func (er erasureObjects) abortMultipartUpload(ctx context.Context, bucket, object, uploadID string, opts ObjectOptions, legacy bool) (bool, error) {
 	if !opts.NoAuditLog {
 		auditObjectErasureSet(ctx, "AbortMultipartUpload", object, &er)
 	}
@@ -1804,70 +1784,19 @@ func (er erasureObjects) abortMultipartUpload(ctx context.Context, bucket, objec
 	if !ok || internalID == "" || internalID == "." || internalID == ".." || strings.ContainsAny(internalID, "/\\") {
 		return false, InvalidUploadID{Bucket: bucket, Object: object, UploadID: uploadID}
 	}
-	if legacy {
+	defer er.deleteAll(ctx, minioMetaMultipartBucket, er.getUploadIDDir(bucket, object, uploadID))
-		// Keep the released read-quorum and best-effort cleanup behavior.
+	_, _, err = er.checkUploadIDExists(ctx, bucket, object, uploadID, false)
-		// The upload ID safety check above applies to both modes.
+	err = toObjectErr(err, bucket, object, uploadID)
-		defer er.deleteAll(ctx, minioMetaMultipartBucket, er.getUploadIDDir(bucket, object, uploadID))
+	if _, absent := err.(InvalidUploadID); absent {
-		_, _, err := er.checkUploadIDExists(ctx, bucket, object, uploadID, false)
-		err = toObjectErr(err, bucket, object, uploadID)
-		if _, absent := err.(InvalidUploadID); absent {
-			return false, nil
-		}
-		return err == nil, err
-	}
-	disks := er.getDisks()
-	uploadPath := er.getUploadIDDir(bucket, object, uploadID)
-	_, errs := readAllFileInfo(ctx, disks, bucket, minioMetaMultipartBucket, uploadPath, "", false, false)
-	quorum := er.setDriveCount/2 + 1
-	found, absent := false, 0
-	observed := make([]bool, len(disks))
-	for i, err := range errs {
-		switch {
-		case err == nil, errors.Is(err, errFileCorrupt):
-			found = true
-			observed[i] = true
-		case errors.Is(err, errFileNotFound), errors.Is(err, errFileVersionNotFound):
-			absent++
-		}
-	}
-	if absent >= quorum && !found {
 		return false, nil
 	}
-	if !found {
+	return err == nil, err
-		return false, toObjectErr(errErasureReadQuorum, bucket, object, uploadID)
-	}
-	g := errgroup.WithNErrs(len(disks))
-	for i, disk := range disks {
-		g.Go(func() error {
-			if disk == nil {
-				return errDiskNotFound
-			}
-			err := disk.Delete(ctx, minioMetaMultipartBucket, uploadPath, DeleteOptions{Recursive: true, Immediate: false})
-			if errors.Is(err, errFileNotFound) || errors.Is(err, errFileVersionNotFound) {
-				return nil
-			}
-			return err
-		}, i)
-	}
-	deleteErrs := g.Wait()
-	if err := reduceWriteQuorumErrs(ctx, deleteErrs, nil, quorum); err != nil {
-		return true, toObjectErr(err, bucket, object, uploadID)
-	}
-	if absent >= quorum {
-		// Missing disks must not mask a failed cleanup of known remnants.
-		for i, found := range observed {
-			if found && deleteErrs[i] != nil {
-				return true, toObjectErr(errErasureWriteQuorum, bucket, object, uploadID)
-			}
-		}
-	}
-	return true, nil
 }
 
-// AbortMultipartUpload cancels an upload using the configured mode. Offline
+// AbortMultipartUpload cancels an upload using read-quorum validation.
-// part data may still need stale-upload cleanup after its drives return.
+// Offline part data may still need stale-upload cleanup after its drives return.
 func (er erasureObjects) AbortMultipartUpload(ctx context.Context, bucket, object, uploadID string, opts ObjectOptions) (err error) {
-	found, err := er.abortMultipartUpload(ctx, bucket, object, uploadID, opts, globalAPIConfig.getMultipartListingLegacy())
+	found, err := er.abortMultipartUpload(ctx, bucket, object, uploadID, opts)
 	if err != nil {
 		return err
 	}

cmd/erasure-object.go

@@ -729,6 +729,57 @@ func readAllXL(ctx context.Context, disks []StorageAPI, bucket, object string, r
 	return pickLatestQuorumFilesInfo(ctx, rawFileInfos, errs, bucket, object, readData, inclFreeVers)
 }
 
+// readListingFileInfo is the read-only, latest-version metadata path for listing
+// recovery. Ordinary GET keeps its fast-read, healing and dangling semantics.
+// Absence is proved before decoding can replace a native storage error: every
+// configured slot must have returned FileNotFound in this read.
+func (er erasureObjects) readListingFileInfo(ctx context.Context, bucket, object string) (FileInfo, bool, error) {
+	disks := er.getDisks()
+	if len(disks) == 0 || len(disks) != er.setDriveCount {
+		return FileInfo{}, false, listingReadQuorumError(bucket, object)
+	}
+	// An offline slot is not evidence of absence, even if a test or transport
+	// implementation could return a cached not-found response for it.
+	online := append([]StorageAPI(nil), disks...)
+	for i, disk := range online {
+		if disk != nil && !disk.IsOnline() {
+			online[i] = nil
+		}
+	}
+	raw, errs := readAllRawFileInfo(ctx, online, bucket, object, false)
+	if err := ctx.Err(); err != nil {
+		return FileInfo{}, false, err
+	}
+	absent := true
+	for _, err := range errs {
+		absent = absent && err == errFileNotFound
+	}
+	if absent {
+		return FileInfo{}, true, nil
+	}
+
+	// Keep the same latest-version and actual object parity quorum as GET,
+	// without its opportunistic dangling deletion or generic NotFound mapping.
+	meta, errs := pickLatestQuorumFilesInfo(ctx, raw, errs, bucket, object, false, false)
+	quorum, _, err := objectQuorumFromMeta(ctx, meta, errs, er.defaultParityCount)
+	if err == nil {
+		err = reduceReadQuorumErrs(ctx, errs, objectOpIgnoredErrs, quorum)
+	}
+	if err != nil {
+		return FileInfo{}, false, listingReadQuorumError(bucket, object)
+	}
+	_, modTime, etag := listOnlineDisks(disks, meta, errs, quorum)
+	fi, err := pickValidFileInfo(ctx, meta, modTime, etag, quorum)
+	if err != nil || (!fi.Deleted && len(fi.Erasure.Distribution) != len(disks)) {
+		return FileInfo{}, false, listingReadQuorumError(bucket, object)
+	}
+	return fi, false, nil
+}
+
+func listingReadQuorumError(bucket, object string) error {
+	return InsufficientReadQuorum{Bucket: bucket, Object: decodeDirObject(object), Err: errErasureReadQuorum, Type: RQInconsistentMeta}
+}
+
 func (er erasureObjects) getObjectFileInfo(ctx context.Context, bucket, object string, opts ObjectOptions, readData bool) (FileInfo, []FileInfo, []StorageAPI, error) {
 	rawArr := make([]RawFileInfo, er.setDriveCount)
 	metaArr := make([]FileInfo, er.setDriveCount)
@@ -2144,6 +2195,13 @@ func (er erasureObjects) DeleteObject(ctx context.Context, bucket, object string
 		}
 	}
 
+	if opts.DataMovement && opts.DeleteMarker {
+		// Moving a null marker replaces any older null data/marker. Moving
+		// a UUID marker preserves it on retries instead of purging it.
+		deleteMarker = true
+		markDelete = opts.VersionID != nullVersionID
+	}
+
 	purge := opts.isVersionPurge()
 	if purge {
 		// A delete marker describes the version being removed; it is not an

cmd/erasure-server-pool-consistency.go

@@ -62,6 +62,125 @@ func (z *erasureServerPools) objectPoolInfos(ctx context.Context, bucket, object
 	return copies, nil
 }
 
+// checkDataMovementCopies rejects moving a version when another pool holds a
+// newer copy of it, which reads return instead. A cross-pool overwrite can
+// leave such a copy, and a client write can add one while the move copies.
+// Equal timestamps normally identify an interrupted move. Conflicting ties
+// retain the same pool-order winner as reads. While the caller holds the
+// pools-layer object lock, the returned resolver serves the set layer's
+// reconciliation of this version from the same reads. Without opts.NoLock
+// it takes the read lock itself.
+func (z *erasureServerPools) checkDataMovementCopies(ctx context.Context, bucket, object string, opts ObjectOptions) (GetObjectInfoFn, error) {
+	if opts.MTime.IsZero() {
+		// The set layer would stamp the moved copy as the newest one.
+		return nil, errInvalidArgument
+	}
+	if !opts.NoLock {
+		lk := z.NewNSLock(bucket, object)
+		lkctx, err := lk.GetRLock(ctx, globalOperationTimeout)
+		if err != nil {
+			return nil, err
+		}
+		ctx = lkctx.Context()
+		defer lk.RUnlock(lkctx)
+	}
+	versionID := opts.VersionID
+	if versionID == "" {
+		versionID = nullVersionID
+	}
+	copies, err := z.objectPoolInfos(ctx, bucket, object, ObjectOptions{VersionID: versionID, NoAuditLog: true})
+	if err != nil && !isErrObjectNotFound(err) && !isErrVersionNotFound(err) {
+		return nil, err
+	}
+	// The snapshot may have been purged or replaced after the reader opened.
+	// In that case the walker must leave the current source state alone.
+	sourcePresent := false
+	for _, copy := range copies {
+		if copy.Index == opts.SrcPoolIdx && copy.ObjInfo.ModTime.Equal(opts.MTime) && copy.ObjInfo.DeleteMarker == opts.DeleteMarker {
+			sourcePresent = true
+		}
+	}
+	if !sourcePresent {
+		return nil, VersionNotFound{Bucket: bucket, Object: decodeDirObject(object), VersionID: versionID}
+	}
+	for _, copy := range copies {
+		if copy.Index != opts.SrcPoolIdx && copy.ObjInfo.ModTime.After(opts.MTime) {
+			return nil, DataMovementOverwriteErr{Bucket: bucket, Object: object, VersionID: opts.VersionID, Err: errDataMovementSuperseded}
+		}
+	}
+	if winner := copies[0]; winner.Index != opts.SrcPoolIdx && winner.ObjInfo.ModTime.Equal(opts.MTime) {
+		for _, copy := range copies {
+			if copy.Index == opts.SrcPoolIdx && (copy.ObjInfo.ETag != winner.ObjInfo.ETag || copy.ObjInfo.DeleteMarker != winner.ObjInfo.DeleteMarker || copy.ObjInfo.Size != winner.ObjInfo.Size) {
+				return nil, DataMovementOverwriteErr{Bucket: bucket, Object: object, VersionID: opts.VersionID, Err: errDataMovementSuperseded}
+			}
+		}
+	}
+	return func(ctx context.Context, b, o string, ropts ObjectOptions) (ObjectInfo, error) {
+		if ropts.VersionID == "" {
+			ropts.VersionID = nullVersionID
+		}
+		if b != bucket || o != object || ropts.VersionID != versionID {
+			return z.replicaObjectInfo(ctx, b, o, ropts)
+		}
+		return replicaCopiesInfo(copies, err, bucket, object, versionID)
+	}, nil
+}
+
+// removeMovedVersions retires only the source versions in the walker's
+// snapshot. A concurrent delete can create a new marker in the source pool;
+// deleting the whole prefix would remove that marker and expose moved data.
+func (z *erasureServerPools) removeMovedVersions(ctx context.Context, bucket, object string, source int, versions []FileInfo) error {
+	object = encodeDirObject(object)
+	lk := z.NewNSLock(bucket, object)
+	lkctx, err := lk.GetLock(ctx, globalOperationTimeout)
+	if err != nil {
+		return err
+	}
+	ctx = lkctx.Context()
+	defer lk.Unlock(lkctx)
+	for _, version := range versions {
+		vid := version.VersionID
+		if vid == "" {
+			vid = nullVersionID
+		}
+		copies, err := z.objectPoolInfos(ctx, bucket, object, ObjectOptions{VersionID: vid, NoAuditLog: true})
+		if isErrObjectNotFound(err) || isErrVersionNotFound(err) {
+			continue
+		}
+		if err != nil {
+			return err
+		}
+		var src *ObjectInfo
+		var retained []PoolObjInfo
+		for _, copy := range copies {
+			if copy.Index == source {
+				src = &copy.ObjInfo
+			} else {
+				retained = append(retained, copy)
+			}
+		}
+		if src == nil {
+			continue
+		}
+		if !src.ModTime.Equal(version.ModTime) || src.DeleteMarker != version.Deleted {
+			return PreConditionFailed{}
+		}
+		// Successful movement is not enough if the destination changed again.
+		// Keep the source unless another copy still preserves its read result.
+		if len(retained) == 0 || retained[0].ObjInfo.ModTime.Before(src.ModTime) {
+			return PreConditionFailed{}
+		}
+		_, err = z.serverPools[source].DeleteObject(ctx, bucket, object, ObjectOptions{
+			VersionID: vid, NoLock: true, NoAuditLog: true,
+			SkipFreeVersion: true, // preserve the old prefix cleanup's tier-GC behavior
+		})
+		if err != nil && !isErrObjectNotFound(err) && !isErrVersionNotFound(err) {
+			return err
+		}
+	}
+	return nil
+}
+
 // Healing also commits object metadata. Both queued healing and drive healing
 // must use the same namespace as pooled writes, rather than racing them under
 // a destination set's independent namespace.
@@ -137,11 +256,15 @@ func (z *erasureServerPools) replicaObjectInfo(ctx context.Context, bucket, obje
 		opts.VersionID = nullVersionID
 	}
 	copies, err := z.objectPoolInfos(ctx, bucket, object, opts)
+	return replicaCopiesInfo(copies, err, bucket, object, opts.VersionID)
+}
+
+func replicaCopiesInfo(copies []PoolObjInfo, err error, bucket, object, versionID string) (ObjectInfo, error) {
 	if err != nil {
 		return ObjectInfo{}, err
 	}
 	if copies[0].ObjInfo.DeleteMarker {
-		return copies[0].ObjInfo, MethodNotAllowed{Bucket: bucket, Object: decodeDirObject(object), VersionID: opts.VersionID}
+		return copies[0].ObjInfo, MethodNotAllowed{Bucket: bucket, Object: decodeDirObject(object), VersionID: versionID}
 	}
 	return mergedPoolObjectInfo(copies), nil
 }

cmd/erasure-server-pool-decom.go

@@ -616,6 +616,12 @@ func (z *erasureServerPools) decommissionObject(ctx context.Context, idx int, bu
 	}
 
 	if objInfo.isMultipart() {
+		// Completion repeats this check under the object write lock.
+		if _, err := z.checkDataMovementCopies(ctx, bucket, encodeDirObject(objInfo.Name), ObjectOptions{
+			VersionID: objInfo.VersionID, MTime: objInfo.ModTime, SrcPoolIdx: idx,
+		}); err != nil {
+			return err
+		}
 		res, err := z.NewMultipartUpload(ctx, bucket, objInfo.Name, ObjectOptions{
 			VersionID:    objInfo.VersionID,
 			UserDefined:  migrationObjectMetadata(objInfo),
@@ -656,6 +662,7 @@ func (z *erasureServerPools) decommissionObject(ctx context.Context, idx int, bu
 			}
 		}
 		_, err = z.CompleteMultipartUpload(ctx, bucket, objInfo.Name, res.UploadID, parts, ObjectOptions{
+			VersionID:    objInfo.VersionID,
 			SrcPoolIdx:   idx,
 			DataMovement: true,
 			MTime:        objInfo.ModTime,
@@ -825,27 +832,17 @@ func (z *erasureServerPools) decommissionPool(ctx context.Context, idx int, pool
 			// to create the appropriate stack.
 			versionsSorter(fivs.Versions).reverse()
 
-			var decommissioned, expired int
+			var decommissioned int
 			for _, version := range fivs.Versions {
 				stopFn := globalDecommissionMetrics.log(decomMetricDecommissionObject, idx, bi.Name, version.Name, version.VersionID)
 				// Apply lifecycle rules on the objects that are expired.
 				if filterLifecycle(bi.Name, version.Name, version) {
-					expired++
 					decommissioned++
 					stopFn(version.Size, errors.New("ILM expired object/version will be skipped"))
 					continue
 				}
 
-				// any object with only single DEL marker we don't need
+				// Preserve every marker: older versions may live in another pool.
-				// to decommission, just skip it, this also includes
-				// any other versions that have already expired.
-				remainingVersions := len(fivs.Versions) - expired
-				if version.Deleted && remainingVersions == 1 && rcfg == nil {
-					decommissioned++
-					stopFn(version.Size, errors.New("DELETE marked object with no other non-current versions will be skipped"))
-					continue
-				}
-
 				versionID := version.VersionID
 				if versionID == "" {
 					versionID = nullVersionID
@@ -870,6 +867,9 @@ func (z *erasureServerPools) decommissionPool(ctx context.Context, idx int, pool
 							SkipDecommissioned: true, // make sure we skip the decommissioned pool
 							NoAuditLog:         true,
 						})
+					if isDataMovementSuperseded(err) {
+						err = nil
+					}
 					if err != nil {
 						// This can happen when rebalance stop races with ongoing rebalance workers.
 						// These rebalance failures can be ignored.
@@ -898,17 +898,19 @@ func (z *erasureServerPools) decommissionPool(ctx context.Context, idx int, pool
 				// gr.Close() is ensured by decommissionObject().
 				for range 3 {
 					if version.IsRemote() {
-						if err := z.DecomTieredObject(ctx, bi.Name, version.Name, version, ObjectOptions{
+						err := z.DecomTieredObject(ctx, bi.Name, version.Name, version, ObjectOptions{
 							VersionID:    versionID,
 							MTime:        version.ModTime,
 							UserDefined:  version.Metadata,
 							SrcPoolIdx:   idx,
 							DataMovement: true,
-						}); err != nil {
+						})
-							if isErrObjectNotFound(err) || isErrVersionNotFound(err) || isDataMovementOverWriteErr(err) {
+						if isDataMovementSuperseded(err) {
-								ignore = true
+							err = nil
-								stopFn(0, nil)
+						}
-							}
+						if isErrObjectNotFound(err) || isErrVersionNotFound(err) || isDataMovementOverWriteErr(err) {
+							ignore = true
+							stopFn(0, nil)
 						}
 						if !ignore {
 							stopFn(version.Size, err)
@@ -949,7 +951,7 @@ func (z *erasureServerPools) decommissionPool(ctx context.Context, idx int, pool
 						stopFn(version.Size, err)
 						continue
 					}
-					if err = z.decommissionObject(ctx, idx, bi.Name, gr); err != nil {
+					if err = z.decommissionObject(ctx, idx, bi.Name, gr); err != nil && !isDataMovementSuperseded(err) {
 						if isErrObjectNotFound(err) || isErrVersionNotFound(err) || isDataMovementOverWriteErr(err) {
 							ignore = true
 							stopFn(0, nil)
@@ -979,15 +981,7 @@ func (z *erasureServerPools) decommissionPool(ctx context.Context, idx int, pool
 			// if all versions were decommissioned, then we can delete the object versions.
 			if decommissioned == len(fivs.Versions) {
 				stopFn := globalDecommissionMetrics.log(decomMetricDecommissionRemoveObject, idx, bi.Name, entry.name)
-				_, err := set.DeleteObject(ctx,
+				err := z.removeMovedVersions(ctx, bi.Name, entry.name, idx, fivs.Versions)
-					bi.Name,
-					encodeDirObject(entry.name),
-					ObjectOptions{
-						DeletePrefix:       true, // use prefix delete to delete all versions at once.
-						DeletePrefixObject: true, // use prefix delete on exact object (this is an optimization to avoid fan-out calls)
-						NoAuditLog:         true,
-					},
-				)
 				stopFn(0, err)
 				auditLogDecom(ctx, "DecomDeleteObject", bi.Name, entry.name, "", err)
 				if err != nil {
@@ -1196,10 +1190,6 @@ func (z *erasureServerPools) checkAfterDecom(ctx context.Context, idx int) error
 						ignored++
 						continue
 					}
-					if version.Deleted {
-						ignored++
-						continue
-					}
 				}
 
 				versionsFound += len(fivs.Versions) - ignored

cmd/erasure-server-pool-migration-delete_test.go

@@ -0,0 +1,169 @@
+// Copyright (c) 2026 Feng Ruohang
+//
+// This file is part of Silo Object Storage stack
+//
+// This program is free software: you can redistribute it and/or modify
+// it under the terms of the GNU Affero General Public License as published by
+// the Free Software Foundation, either version 3 of the License, or
+// (at your option) any later version.
+//
+// This program is distributed in the hope that it will be useful
+// but WITHOUT ANY WARRANTY; without even the implied warranty of
+// MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
+// GNU Affero General Public License for more details.
+//
+// You should have received a copy of the GNU Affero General Public License
+// along with this program. If not, see <http://www.gnu.org/licenses/>.
+
+package cmd
+
+import (
+	"fmt"
+	"testing"
+	"time"
+)
+
+// Both walkers must preserve markers even when all older versions reside in
+// another pool. Local version counts cannot prove that a marker is redundant.
+func TestPoolsMigrationDeleteMarkers(t *testing.T) {
+	for _, kind := range []string{"rebalance", "decommission"} {
+		for _, pools := range []int{2, 3} {
+			for _, null := range []bool{false, true} {
+				for _, history := range []string{"none", "source", "destination"} {
+					t.Run(fmt.Sprintf("%s/pools=%d/null=%t/history=%s", kind, pools, null, history), func(t *testing.T) {
+						var z *erasureServerPools
+						var bucket string
+						if pools == 2 {
+							z, bucket = migrationTestPools(t)
+						} else {
+							z, bucket = migrationTestThreePools(t)
+						}
+						const object = "deleted"
+						old, stamp := UTCNow().Add(-time.Hour), UTCNow().Add(-time.Minute)
+						var prior ObjectInfo
+						if history != "none" {
+							pool := 0
+							if history == "destination" {
+								pool = 1
+							}
+							prior = putConsistencyObject(t, z, bucket, object, pool, "history", ObjectOptions{Versioned: true, MTime: old})
+						}
+						// A stale null copy is a normal consequence of cross-pool overwrites.
+						putConsistencyObject(t, z, bucket, object, 1, "stale", ObjectOptions{MTime: old})
+						vid := mustGetUUID()
+						if null {
+							vid = nullVersionID
+						}
+						opts := ObjectOptions{Versioned: true, VersionID: vid, DeleteMarker: true, MTime: stamp}
+						marker, err := z.serverPools[0].DeleteObject(t.Context(), bucket, object, opts)
+						mustIAM(t, err)
+						if !marker.DeleteMarker {
+							t.Fatal("fixture did not create marker")
+						}
+						if oi, err := z.GetObjectInfo(t.Context(), bucket, object, ObjectOptions{}); !isErrObjectNotFound(err) || !oi.DeleteMarker {
+							t.Fatalf("fixture visible: %+v %v", oi, err)
+						}
+						if failed := migrationTestWalk(t, z, bucket, 0, kind); failed != 0 {
+							t.Fatalf("failed moves: %d", failed)
+						}
+						for attempt := 0; attempt < 2; attempt++ {
+							if oi, err := z.GetObjectInfo(t.Context(), bucket, object, ObjectOptions{}); !isErrObjectNotFound(err) || !oi.DeleteMarker || !oi.ModTime.Equal(stamp) {
+								t.Fatalf("marker lost: %+v %v", oi, err)
+							}
+							if _, err := z.serverPools[0].GetObjectInfo(t.Context(), bucket, object, ObjectOptions{VersionID: migrationTestVersion(marker)}); !isErrObjectNotFound(err) && !isErrVersionNotFound(err) {
+								t.Fatalf("source marker retained: %v", err)
+							}
+							if prior.VersionID != "" && migrationTestBody(t, z.GetObjectNInfo, bucket, object, prior.VersionID) != "history" {
+								t.Fatal("lost historical version")
+							}
+							// Repeating a completed walk must preserve the same state.
+							migrationTestWalk(t, z, bucket, 0, kind)
+						}
+					})
+				}
+			}
+		}
+	}
+}
+
+// A reader opened before a public purge must not recreate the purged version.
+func TestPoolsMigrationDoesNotResurrectPurgedSource(t *testing.T) {
+	for _, kind := range []string{"rebalance", "decommission"} {
+		for _, multipart := range []bool{false, true} {
+			t.Run(fmt.Sprintf("%s/multipart=%t", kind, multipart), func(t *testing.T) {
+				z, bucket := migrationTestPools(t)
+				move := migrationTestMover(t, z, 0, kind)
+				oi := migrationTestObject(t, z, bucket, "purged", 0, multipart, []string{"payload"}, ObjectOptions{Versioned: true})
+				gr := migrationTestReader(t, z, 0, oi)
+				_, err := z.DeleteObject(t.Context(), bucket, oi.Name, ObjectOptions{VersionID: oi.VersionID, Versioned: true})
+				mustIAM(t, err)
+				err = move(t.Context(), 0, bucket, gr)
+				if err == nil {
+					t.Error("move accepted a purged source")
+				}
+				if oi, err := z.GetObjectInfo(t.Context(), bucket, oi.Name, ObjectOptions{}); !isErrObjectNotFound(err) {
+					t.Fatalf("purged version resurrected: %+v %v", oi, err)
+				}
+			})
+		}
+	}
+}
+
+// Cleanup must not erase a marker created after the data was copied. It must
+// also fail closed if the destination becomes unreadable before source removal.
+func TestPoolsMigrationCleanupRechecksSource(t *testing.T) {
+	for _, change := range []string{"delete-null", "delete-uuid", "unreadable-destination", "missing-destination"} {
+		t.Run(change, func(t *testing.T) {
+			z, bucket := migrationTestPools(t)
+			move := migrationTestMover(t, z, 0, "decommission")
+			oi := migrationTestObject(t, z, bucket, "cleanup", 0, false, []string{"payload"}, ObjectOptions{})
+			snapshot := []FileInfo{{VersionID: oi.VersionID, ModTime: oi.ModTime}}
+			mustIAM(t, move(t.Context(), 0, bucket, migrationTestReader(t, z, 0, oi)))
+			switch change {
+			case "delete-null", "delete-uuid":
+				_, err := z.DeleteObject(t.Context(), bucket, oi.Name, ObjectOptions{VersionSuspended: change == "delete-null", Versioned: change == "delete-uuid"})
+				mustIAM(t, err)
+			case "unreadable-destination":
+				restore := conditionalPutSwapDisks(z.serverPools[1], oi.Name, func(d StorageAPI) StorageAPI {
+					return consistencyReadFaultDisk{StorageAPI: d, bucket: bucket, object: oi.Name}
+				})
+				defer restore()
+			case "missing-destination":
+				_, err := z.serverPools[1].DeleteObject(t.Context(), bucket, oi.Name, ObjectOptions{VersionID: nullVersionID})
+				mustIAM(t, err)
+			}
+			err := z.removeMovedVersions(t.Context(), bucket, oi.Name, 0, snapshot)
+			if change != "delete-uuid" && err == nil {
+				t.Fatal("cleanup accepted changed or uncertain state")
+			}
+			if change == "delete-null" || change == "delete-uuid" {
+				if got, err := z.GetObjectInfo(t.Context(), bucket, oi.Name, ObjectOptions{}); !isErrObjectNotFound(err) || !got.DeleteMarker {
+					t.Fatalf("cleanup resurrected data: %v", err)
+				}
+			} else if _, err := z.serverPools[0].GetObjectInfo(t.Context(), bucket, oi.Name, ObjectOptions{}); err != nil {
+				t.Fatalf("cleanup lost source: %v", err)
+			}
+		})
+	}
+}
+
+func TestPoolsMigrationConflictingTimestampTie(t *testing.T) {
+	for _, kind := range []string{"rebalance", "decommission"} {
+		for source := range 2 {
+			t.Run(fmt.Sprintf("%s/source=%d", kind, source), func(t *testing.T) {
+				z, bucket := migrationTestPools(t)
+				stamp := UTCNow().Add(-time.Minute)
+				for i := range 2 {
+					putConsistencyObject(t, z, bucket, "tie", i, fmt.Sprintf("pool-%d", i), ObjectOptions{MTime: stamp})
+				}
+				if got := migrationTestBody(t, z.GetObjectNInfo, bucket, "tie", ""); got != "pool-0" {
+					t.Fatalf("fixture winner: %s", got)
+				}
+				migrationTestWalk(t, z, bucket, source, kind)
+				if got := migrationTestBody(t, z.GetObjectNInfo, bucket, "tie", ""); got != "pool-0" {
+					t.Fatalf("changed tie winner: %s", got)
+				}
+			})
+		}
+	}
+}

cmd/erasure-server-pool-migration_test.go

@@ -24,11 +24,13 @@ import (
 	"fmt"
 	"io"
 	"maps"
+	"net/http"
 	"strings"
 	"sync"
 	"testing"
 	"time"
 
+	"github.com/minio/minio/internal/bucket/lifecycle"
 	xhttp "github.com/minio/minio/internal/http"
 )
 
@@ -114,15 +116,18 @@ func migrationTestPools(t *testing.T) (*erasureServerPools, string) {
 	t.Helper()
 	z, bucket := consistencyPools(t)
 	for _, pool := range z.serverPools {
-		for _, set := range pool.sets {
+		pool.erasureDisksMu.Lock()
-			original := set.getDisks
+		previous := pool.erasureDisks
-			disks := append([]StorageAPI(nil), original()...)
+		wrapped := make([][]StorageAPI, len(previous))
-			for i := range disks {
+		for i, disks := range previous {
-				disks[i] = tagTestCapacityDisk{StorageAPI: disks[i]}
+			wrapped[i] = append([]StorageAPI(nil), disks...)
+			for j := range wrapped[i] {
+				wrapped[i][j] = tagTestCapacityDisk{StorageAPI: wrapped[i][j]}
 			}
-			set.getDisks = func() []StorageAPI { return disks }
-			t.Cleanup(func() { set.getDisks = original })
 		}
+		pool.erasureDisks = wrapped
+		pool.erasureDisksMu.Unlock()
+		t.Cleanup(func() { pool.erasureDisksMu.Lock(); pool.erasureDisks = previous; pool.erasureDisksMu.Unlock() })
 	}
 	return z, bucket
 }
@@ -131,7 +136,10 @@ func migrationTestMover(t *testing.T, z *erasureServerPools, source int, kind st
 	t.Helper()
 	if kind == "rebalance" {
 		z.rebalMu.Lock()
-		z.rebalMeta = &rebalanceMeta{PoolStats: []*rebalanceStats{{}, {}}}
+		z.rebalMeta = &rebalanceMeta{PoolStats: make([]*rebalanceStats, len(z.serverPools))}
+		for i := range z.rebalMeta.PoolStats {
+			z.rebalMeta.PoolStats[i] = &rebalanceStats{}
+		}
 		z.rebalMeta.PoolStats[source] = &rebalanceStats{Participating: true, Info: rebalanceInfo{Status: rebalStarted}}
 		z.rebalMu.Unlock()
 		t.Cleanup(func() { z.rebalMu.Lock(); z.rebalMeta = nil; z.rebalMu.Unlock() })
@@ -590,3 +598,331 @@ func TestPoolsMigrationUnreadablePool(t *testing.T) {
 		}
 	}
 }
+
+// Three pools give a move two candidate destinations.
+func migrationTestThreePools(t *testing.T) (*erasureServerPools, string) {
+	t.Helper()
+	ctx, cancel := context.WithCancel(t.Context())
+	dirs, err := getRandomDisks(12)
+	if err != nil {
+		cancel()
+		t.Fatal(err)
+	}
+	var endpoints EndpointServerPools
+	for i := range 3 {
+		endpoints = append(endpoints, mustGetPoolEndpoints(i, dirs[4*i:4*i+4]...)...)
+	}
+	obj, _, err := initObjectLayer(ctx, endpoints)
+	if err != nil {
+		cancel()
+		removeRoots(dirs)
+		t.Fatal(err)
+	}
+	z := obj.(*erasureServerPools)
+	previous := newObjectLayerFn()
+	setObjectLayer(z)
+	t.Cleanup(func() { cancel(); z.Shutdown(context.Background()); removeRoots(dirs); setObjectLayer(previous) })
+	bucket := "pool-migration"
+	if err := z.MakeBucket(t.Context(), bucket, MakeBucketOptions{}); err != nil {
+		t.Fatal(err)
+	}
+	return z, bucket
+}
+
+type migrationTestGetter func(context.Context, string, string, *HTTPRangeSpec, http.Header, ObjectOptions) (*GetObjectReader, error)
+
+// migrationTestBody returns the content read for a version, or the read error.
+func migrationTestBody(t *testing.T, get migrationTestGetter, bucket, object, versionID string) string {
+	t.Helper()
+	gr, err := get(t.Context(), bucket, object, nil, nil, ObjectOptions{VersionID: versionID})
+	if err != nil {
+		return fmt.Sprintf("<%v>", err)
+	}
+	defer gr.Close()
+	data, err := io.ReadAll(gr)
+	if err != nil {
+		t.Fatal(err)
+	}
+	return string(data)
+}
+
+// migrationTestWalk runs one bucket of a decommission or rebalance of source,
+// including the walker's accounting and source cleanup. It returns the number
+// of versions decommission counted as failed.
+func migrationTestWalk(t *testing.T, z *erasureServerPools, bucket string, source int, kind string) int64 {
+	t.Helper()
+	migrationTestMover(t, z, source, kind)
+	if kind == "rebalance" {
+		if err := z.rebalanceBucket(t.Context(), bucket, source); err != nil {
+			t.Fatal(err)
+		}
+		return 0
+	}
+	z.poolMetaMutex.Lock()
+	z.poolMeta.Pools[source].LastUpdate = UTCNow() // keep pool.bin and peers out of the walk
+	z.poolMetaMutex.Unlock()
+	if err := z.decommissionPool(t.Context(), source, z.serverPools[source], decomBucketInfo{Name: bucket}); err != nil {
+		t.Fatal(err)
+	}
+	z.poolMetaMutex.RLock()
+	defer z.poolMetaMutex.RUnlock()
+	return z.poolMeta.Pools[source].Decommission.ItemsDecommissionFailed
+}
+
+// migrationTestTiered stores a null version whose data is in a remote tier, in
+// the form decommission moves it.
+func migrationTestTiered(t *testing.T, z *erasureServerPools, bucket, object string, pool int, mtime time.Time) ObjectInfo {
+	t.Helper()
+	oi := putConsistencyObject(t, z, bucket, object, pool, "old", ObjectOptions{MTime: mtime})
+	infos, errs := readAllFileInfo(t.Context(), z.serverPools[pool].getHashedSet(oi.Name).getDisks(), "", bucket, oi.Name, nullVersionID, false, false)
+	if errs[0] != nil {
+		t.Fatal(errs[0])
+	}
+	fi := infos[0]
+	fi.TransitionStatus, fi.TransitionTier, fi.TransitionedObjName = lifecycle.TransitionComplete, "WARM", "remote"
+	if err := z.serverPools[pool].DecomTieredObject(t.Context(), bucket, oi.Name, fi, ObjectOptions{}); err != nil {
+		t.Fatal(err)
+	}
+	return oi
+}
+
+// A cross-pool overwrite can leave copies of one version in several pools.
+// Reads return the newest copy, and moving an older one must not replace it.
+// An equal ModTime marks the same write, as an interrupted move leaves it.
+func TestPoolsMigrationKeepsCurrentCopy(t *testing.T) {
+	z, bucket := migrationTestPools(t)
+	base := time.Date(2026, 9, 16, 8, 0, 0, 0, time.UTC)
+	for _, kind := range []string{"rebalance", "decommission"} {
+		for _, method := range []string{"put", "multipart"} {
+			for _, versioned := range []bool{false, true} {
+				for source := range 2 {
+					for _, offset := range []time.Duration{-time.Minute, 0, time.Minute} {
+						t.Run(fmt.Sprintf("%s/%s/versioned=%t/source=%d/offset=%s", kind, method, versioned, source, offset), func(t *testing.T) {
+							move := migrationTestMover(t, z, source, kind)
+							opts := ObjectOptions{Versioned: versioned, MTime: base.Add(offset)}
+							if versioned {
+								opts.VersionID = mustGetUUID()
+							}
+							oi := migrationTestObject(t, z, bucket, t.Name(), source, method == "multipart", []string{"source"}, opts)
+							opts.MTime = base
+							want, other := "source", "other"
+							if offset == 0 {
+								other = "source"
+								opts.PreserveETag = oi.ETag // interrupted copies preserve the original multipart ETag
+							} else if offset < 0 {
+								want = "other"
+							}
+							putConsistencyObject(t, z, bucket, t.Name(), 1-source, other, opts)
+							version := migrationTestVersion(oi)
+							if got := migrationTestBody(t, z.GetObjectNInfo, bucket, t.Name(), version); got != want {
+								t.Fatalf("fixture reads %q, want %q", got, want)
+							}
+							err := move(t.Context(), source, bucket, migrationTestReader(t, z, source, oi))
+							if want == "other" && !isDataMovementSuperseded(err) || want == "source" && err != nil {
+								t.Fatalf("move: %v", err)
+							}
+							if err != nil && strings.Contains(err.Error(), "CompleteMultipartUpload") {
+								t.Error("copied the parts of a superseded version")
+							}
+							if got := migrationTestBody(t, z.serverPools[1-source].GetObjectNInfo, bucket, t.Name(), version); got != want {
+								t.Errorf("destination holds %q, want %q", got, want)
+							}
+							if got := migrationTestBody(t, z.GetObjectNInfo, bucket, t.Name(), version); got != want {
+								t.Errorf("move changed the read result to %q, want %q", got, want)
+							}
+						})
+					}
+				}
+			}
+		}
+	}
+}
+
+// A client overwrite, or a replica of the same version, committed while a move
+// copies the old version is newer than the moved copy. The move must yield to
+// it, also at multipart completion.
+func TestPoolsMigrationYieldsToNewerWrite(t *testing.T) {
+	z, bucket := migrationTestPools(t)
+	for _, kind := range []string{"rebalance", "decommission"} {
+		for _, method := range []string{"put", "multipart"} {
+			for _, versioned := range []bool{false, true} {
+				for source := range 2 {
+					t.Run(fmt.Sprintf("%s/%s/versioned=%t/source=%d", kind, method, versioned, source), func(t *testing.T) {
+						migrationTestNewerWrite(t, z, bucket, kind, method, versioned, source)
+					})
+				}
+			}
+		}
+	}
+}
+
+func migrationTestNewerWrite(t *testing.T, z *erasureServerPools, bucket, kind, method string, versioned bool, source int) {
+	move := migrationTestMover(t, z, source, kind)
+	oi := migrationTestObject(t, z, bucket, t.Name(), source, method == "multipart", []string{"moved"}, ObjectOptions{Versioned: versioned, MTime: UTCNow().Add(-time.Hour)})
+	gr := migrationTestReader(t, z, source, oi)
+	ctx, cancel := context.WithTimeout(t.Context(), 30*time.Second)
+	defer cancel()
+	gate := newMigrationTestGate()
+	done, finished := make(chan error, 1), make(chan struct{})
+	if method == "multipart" {
+		// Hold the move after its initial check, while it copies parts.
+		gr.Reader = migrationTestGateReader{Reader: gr.Reader, ctx: ctx, gate: gate}
+		go func() { defer close(finished); done <- move(ctx, source, bucket, gr) }()
+		defer func() { gate.unblock(); cancel(); <-finished }()
+		select {
+		case <-gate.entered:
+		case err := <-done:
+			t.Fatalf("migration finished before the part barrier: %v", err)
+		case <-ctx.Done():
+			t.Fatal(ctx.Err())
+		}
+	}
+	// An ordinary PUT copies under the object lock, so this write
+	// lands between the mover's source read and its write.
+	if versioned {
+		putConsistencyObject(t, z, bucket, t.Name(), 1-source, "newer", ObjectOptions{Versioned: true, VersionID: oi.VersionID, MTime: UTCNow()})
+	} else if _, err := z.PutObject(ctx, bucket, t.Name(), mustGetPutObjReader(t, strings.NewReader("newer"), 5, "", ""), ObjectOptions{}); err != nil {
+		t.Fatal(err)
+	}
+	var err error
+	if method == "multipart" {
+		gate.unblock()
+		err = <-done
+	} else {
+		err = move(ctx, source, bucket, gr)
+	}
+	if !isDataMovementSuperseded(err) {
+		t.Fatalf("move over a newer write: %v", err)
+	}
+	if got := migrationTestBody(t, z.GetObjectNInfo, bucket, t.Name(), migrationTestVersion(oi)); got != "newer" {
+		t.Fatalf("newer write replaced by %q", got)
+	}
+}
+
+// With two candidate destinations a superseded copy must neither replace the
+// newer copy nor become a new duplicate in the third pool.
+func TestPoolsMigrationThreePools(t *testing.T) {
+	z, bucket := migrationTestThreePools(t)
+	stale, recent := UTCNow().Add(-time.Hour), UTCNow().Add(-time.Minute)
+	for _, kind := range []string{"rebalance", "decommission"} {
+		for _, newest := range []int{0, 1} {
+			t.Run(fmt.Sprintf("%s/newest=%d", kind, newest), func(t *testing.T) {
+				move := migrationTestMover(t, z, 0, kind)
+				mtime := func(pool int) time.Time {
+					if pool == newest {
+						return recent
+					}
+					return stale
+				}
+				oi := migrationTestObject(t, z, bucket, t.Name(), 0, false, []string{"pool-0"}, ObjectOptions{MTime: mtime(0)})
+				putConsistencyObject(t, z, bucket, t.Name(), 1, "pool-1", ObjectOptions{MTime: mtime(1)})
+				if newest == 0 {
+					putConsistencyObject(t, z, bucket, t.Name(), 2, "pool-2", ObjectOptions{MTime: stale})
+				}
+				want := fmt.Sprintf("pool-%d", newest)
+				err := move(t.Context(), 0, bucket, migrationTestReader(t, z, 0, oi))
+				if newest == 1 && !isDataMovementSuperseded(err) || newest == 0 && err != nil {
+					t.Fatalf("move: %v", err)
+				}
+				if got := migrationTestBody(t, z.GetObjectNInfo, bucket, t.Name(), ""); got != want {
+					t.Errorf("reads return %q, want %q", got, want)
+				}
+				if newest == 1 {
+					if _, err := z.serverPools[2].GetObjectInfo(t.Context(), bucket, t.Name(), ObjectOptions{}); !isErrObjectNotFound(err) {
+						t.Errorf("superseded copy duplicated into pool 2: %v", err)
+					}
+				}
+			})
+		}
+	}
+}
+
+// The walkers count a superseded version as moved, so its stale source copy is
+// removed with the object and the source pool drains.
+func TestPoolsMigrationRetiresSupersededCopy(t *testing.T) {
+	for _, kind := range []string{"rebalance", "decommission"} {
+		t.Run(kind, func(t *testing.T) {
+			z, bucket := migrationTestPools(t)
+			ctx := t.Context()
+			stale, recent := UTCNow().Add(-time.Hour), UTCNow().Add(-time.Minute)
+			put := func(object string, pool int, body string, multipart bool, opts ObjectOptions) ObjectInfo {
+				return migrationTestObject(t, z, bucket, object, pool, multipart, []string{body}, opts)
+			}
+			put("stale", 0, "old", false, ObjectOptions{MTime: stale})
+			put("stale", 1, "new", false, ObjectOptions{MTime: recent})
+			put("stale-multipart", 0, "old", true, ObjectOptions{MTime: stale})
+			put("stale-multipart", 1, "new", false, ObjectOptions{MTime: recent})
+			put("newer", 0, "new", false, ObjectOptions{MTime: recent})
+			put("newer", 1, "old", false, ObjectOptions{MTime: stale})
+			oldest := put("history", 0, "old", false, ObjectOptions{Versioned: true, MTime: stale})
+			put("history", 1, "new", false, ObjectOptions{Versioned: true, MTime: recent})
+			// A suspended bucket's older null delete marker must not hide a newer
+			// null version.
+			hidden := put("stale-marker", 0, "old", false, ObjectOptions{Versioned: true, MTime: stale.Add(-time.Minute)})
+			if _, err := z.serverPools[0].DeleteObject(ctx, bucket, "stale-marker", ObjectOptions{VersionSuspended: true, MTime: stale}); err != nil {
+				t.Fatal(err)
+			}
+			put("stale-marker", 1, "new", false, ObjectOptions{VersionSuspended: true, MTime: recent})
+			objects := []string{"stale", "stale-multipart", "newer", "history", "stale-marker"}
+			if kind == "decommission" {
+				// Rebalance leaves tiered versions in place; decommission moves
+				// their metadata.
+				migrationTestTiered(t, z, bucket, "stale-tiered", 0, stale)
+				put("stale-tiered", 1, "new", false, ObjectOptions{MTime: recent})
+				objects = append(objects, "stale-tiered")
+			}
+			want := map[string]string{}
+			for _, object := range objects {
+				want[object] = migrationTestBody(t, z.GetObjectNInfo, bucket, object, "")
+			}
+			if want["stale"] != "new" || want["stale-marker"] != "new" {
+				t.Fatalf("fixture reads %v", want)
+			}
+			if failed := migrationTestWalk(t, z, bucket, 0, kind); failed != 0 {
+				t.Fatalf("%d versions failed to move", failed)
+			}
+			for _, object := range objects {
+				if got := migrationTestBody(t, z.GetObjectNInfo, bucket, object, ""); got != want[object] {
+					t.Errorf("%s: reads %q after the move, want %q", object, got, want[object])
+				}
+			}
+			var retained []string
+			for _, set := range z.serverPools[0].sets {
+				if err := set.listObjectsToDecommission(ctx, decomBucketInfo{Name: bucket}, func(entry metaCacheEntry) {
+					retained = append(retained, entry.name)
+				}); err != nil {
+					t.Fatal(err)
+				}
+			}
+			if len(retained) > 0 {
+				t.Errorf("source pool retained %v", retained)
+			}
+			for _, oi := range []ObjectInfo{oldest, hidden} {
+				if got := migrationTestBody(t, z.GetObjectNInfo, bucket, oi.Name, oi.VersionID); got != "old" {
+					t.Errorf("%s: lost version %s: %q", oi.Name, oi.VersionID, got)
+				}
+			}
+		})
+	}
+}
+
+// A tiered move that fails, here because another pool cannot be read, must keep
+// its source like any other failed version.
+func TestPoolsMigrationTieredFailureKeepsSource(t *testing.T) {
+	z, bucket := migrationTestPools(t)
+	tiered := migrationTestTiered(t, z, bucket, t.Name(), 0, UTCNow().Add(-time.Hour))
+	set := z.serverPools[1].getHashedSet(tiered.Name)
+	original := set.getDisks
+	disks := append([]StorageAPI(nil), original()...)
+	for i := range disks {
+		disks[i] = consistencyReadFaultDisk{StorageAPI: disks[i], bucket: bucket, object: tiered.Name}
+	}
+	set.getDisks = func() []StorageAPI { return disks }
+	defer func() { set.getDisks = original }()
+	if failed := migrationTestWalk(t, z, bucket, 0, "decommission"); failed != 1 {
+		t.Fatalf("%d failed versions, want 1", failed)
+	}
+	if _, err := z.serverPools[0].GetObjectInfo(t.Context(), bucket, tiered.Name, ObjectOptions{VersionID: nullVersionID}); err != nil {
+		t.Fatalf("failed move removed its source: %v", err)
+	}
+}

cmd/erasure-server-pool-put-conditional_test.go

@@ -355,7 +355,7 @@ func TestPoolsConditionalPutVersionSelection(t *testing.T) {
 				opts.ReplicaLockReconcile, opts.ReplicationRequest = true, true
 				want = addressed
 			case "movement":
-				opts.DataMovement, opts.SrcPoolIdx = true, 1
+				opts.DataMovement, opts.SrcPoolIdx, opts.MTime = true, 1, addressed.ModTime
 				want = addressed
 			case "tie":
 				want = putConsistencyObject(t, z, bucket, object, 0, "tie-winner", ObjectOptions{Versioned: true, MTime: current.ModTime})

cmd/erasure-server-pool-rebalance.go

@@ -623,7 +623,7 @@ func (z *erasureServerPools) rebalanceBucket(ctx context.Context, bucket string,
 			// to create the appropriate stack.
 			versionsSorter(fivs.Versions).reverse()
 
-			var rebalanced, expired int
+			var rebalanced int
 			for _, version := range fivs.Versions {
 				stopFn := globalRebalanceMetrics.log(rebalanceMetricRebalanceObject, poolIdx, bucket, version.Name, version.VersionID)
 
@@ -635,21 +635,11 @@ func (z *erasureServerPools) rebalanceBucket(ctx context.Context, bucket string,
 
 				// Apply lifecycle rules on the objects that are expired.
 				if filterLifecycle(bucket, version.Name, version) {
-					expired++
 					stopFn(version.Size, errors.New("ILM expired object/version will be skipped"))
 					continue
 				}
 
-				// any object with only single DEL marker we don't need
+				// Preserve every marker: older versions may live in another pool.
-				// to rebalance, just skip it, this also includes
-				// any other versions that have already expired.
-				remainingVersions := len(fivs.Versions) - expired
-				if version.Deleted && remainingVersions == 1 {
-					rebalanced++
-					stopFn(version.Size, errors.New("DELETE marked object with no other non-current versions will be skipped"))
-					continue
-				}
-
 				versionID := version.VersionID
 				if versionID == "" {
 					versionID = nullVersionID
@@ -671,6 +661,9 @@ func (z *erasureServerPools) rebalanceBucket(ctx context.Context, bucket string,
 							SkipRebalancing:   true, // make sure we skip the decommissioned pool
 							NoAuditLog:        true,
 						})
+					if isDataMovementSuperseded(err) {
+						err = nil
+					}
 					// This can happen when rebalance stop races with ongoing rebalance workers.
 					// These rebalance failures can be ignored.
 					if err != nil {
@@ -719,7 +712,7 @@ func (z *erasureServerPools) rebalanceBucket(ctx context.Context, bucket string,
 						continue
 					}
 
-					if err = z.rebalanceObject(ctx, poolIdx, bucket, gr); err != nil {
+					if err = z.rebalanceObject(ctx, poolIdx, bucket, gr); err != nil && !isDataMovementSuperseded(err) {
 						// This can happen when rebalance stop races with ongoing rebalance workers.
 						// These rebalance failures can be ignored.
 						if isErrObjectNotFound(err) || isErrVersionNotFound(err) || isDataMovementOverWriteErr(err) {
@@ -750,15 +743,7 @@ func (z *erasureServerPools) rebalanceBucket(ctx context.Context, bucket string,
 			// if all versions were rebalanced, we can delete the object versions.
 			if rebalanced == len(fivs.Versions) {
 				stopFn := globalRebalanceMetrics.log(rebalanceMetricRebalanceRemoveObject, poolIdx, bucket, entry.name)
-				_, err := set.DeleteObject(ctx,
+				err := z.removeMovedVersions(ctx, bucket, entry.name, poolIdx, fivs.Versions)
-					bucket,
-					encodeDirObject(entry.name),
-					ObjectOptions{
-						DeletePrefix:       true, // use prefix delete to delete all versions at once.
-						DeletePrefixObject: true, // use prefix delete on exact object (this is an optimization to avoid fan-out calls)
-						NoAuditLog:         true,
-					},
-				)
 				stopFn(0, err)
 				auditLogRebalance(ctx, "Rebalance:DeleteObject", bucket, entry.name, "", err)
 				if err != nil {
@@ -863,6 +848,12 @@ func (z *erasureServerPools) rebalanceObject(ctx context.Context, poolIdx int, b
 	}
 
 	if oi.isMultipart() {
+		// Completion repeats this check under the object write lock.
+		if _, err := z.checkDataMovementCopies(ctx, bucket, encodeDirObject(oi.Name), ObjectOptions{
+			VersionID: oi.VersionID, MTime: oi.ModTime, SrcPoolIdx: poolIdx,
+		}); err != nil {
+			return err
+		}
 		res, err := z.NewMultipartUpload(ctx, bucket, oi.Name, ObjectOptions{
 			VersionID:    oi.VersionID,
 			UserDefined:  migrationObjectMetadata(oi),
@@ -900,6 +891,8 @@ func (z *erasureServerPools) rebalanceObject(ctx context.Context, poolIdx int, b
 			}
 		}
 		_, err = z.CompleteMultipartUpload(ctx, bucket, oi.Name, res.UploadID, parts, ObjectOptions{
+			VersionID:    oi.VersionID,
+			SrcPoolIdx:   poolIdx,
 			DataMovement: true,
 			MTime:        oi.ModTime,
 			NoAuditLog:   true,
@@ -1010,7 +1003,10 @@ const (
 	rebalanceMetricSaveMetadata
 )
 
-var errDataMovementSrcDstPoolSame = errors.New("source and destination pool are the same")
+var (
+	errDataMovementSrcDstPoolSame = errors.New("source and destination pool are the same")
+	errDataMovementSuperseded     = errors.New("another pool holds a newer copy")
+)
 
 func rebalanceTrace(r rebalanceMetric, poolIdx int, startTime time.Time, duration time.Duration, err error, path string, sz int64) madmin.TraceInfo {
 	var errStr string

cmd/erasure-server-pool.go

@@ -1183,6 +1183,14 @@ func (z *erasureServerPools) PutObject(ctx context.Context, bucket string, objec
 		opts.CheckPrecondFn = nil
 	}
 
+	if opts.DataMovement {
+		resolve, err := z.checkDataMovementCopies(ctx, bucket, object, opts)
+		if err != nil {
+			return ObjectInfo{}, err
+		}
+		opts.replicaObjectInfo = resolve
+	}
+
 	idx, err := z.getWritePoolIdx(ctx, bucket, object, data.Size(), true)
 	if err != nil {
 		return ObjectInfo{}, err
@@ -1202,7 +1210,7 @@ func (z *erasureServerPools) PutObject(ctx context.Context, bucket string, objec
 	}
 	// Public overwrites need all pools for Object Lock. Unaddressed replica
 	// callbacks retain their set-local reads unless reconciliation is needed.
-	if opts.ReplicaLockReconcile || !opts.ReplicationRequest {
+	if opts.replicaObjectInfo == nil && (opts.ReplicaLockReconcile || !opts.ReplicationRequest) {
 		opts.replicaObjectInfo = z.replicaObjectInfo
 	}
 	oi, err := z.serverPools[idx].PutObject(ctx, bucket, object, data, opts)
@@ -1243,6 +1251,25 @@ func (z *erasureServerPools) DeleteObject(ctx context.Context, bucket string, ob
 		return ObjectInfo{}, z.deletePrefix(ctx, bucket, object)
 	}
 
+	if opts.DataMovement {
+		opts.NoLock = true
+		if _, err := z.checkDataMovementCopies(ctx, bucket, object, opts); err != nil {
+			return objInfo, err
+		}
+		// A marker still needs a destination when it is the only version of
+		// this key outside the draining pool.
+		idx, err := z.getPoolIdxNoLock(ctx, bucket, object, 0)
+		if err != nil {
+			return objInfo, err
+		}
+		if idx == opts.SrcPoolIdx {
+			return objInfo, DataMovementOverwriteErr{Bucket: bucket, Object: object, VersionID: opts.VersionID, Err: errDataMovementSrcDstPoolSame}
+		}
+		objInfo, err = z.serverPools[idx].DeleteObject(ctx, bucket, object, opts)
+		objInfo.Name = decodeDirObject(object)
+		return objInfo, err
+	}
+
 	// Resolve a physical purge by its addressed version even on a replica
 	// receiver: latest-key routing can select a different pool or miss copies.
 	// Marker creation and specialized movement/scanner operations retain their
@@ -1328,22 +1355,6 @@ func (z *erasureServerPools) DeleteObject(ctx context.Context, bucket string, ob
 		return pinfo.ObjInfo, nil
 	}
 
-	// Datamovement must never be allowed on the same pool.
-	if opts.DataMovement && opts.SrcPoolIdx == pinfo.Index {
-		return pinfo.ObjInfo, DataMovementOverwriteErr{
-			Bucket:    bucket,
-			Object:    object,
-			VersionID: opts.VersionID,
-			Err:       errDataMovementSrcDstPoolSame,
-		}
-	}
-
-	if opts.DataMovement {
-		objInfo, err = z.serverPools[pinfo.Index].DeleteObject(ctx, bucket, object, opts)
-		objInfo.Name = decodeDirObject(object)
-		return objInfo, err
-	}
-
 	// Delete concurrently in all server pools with read quorum error for unversioned objects.
 	if len(noReadQuorumPools) > 0 && !opts.Versioned && !opts.VersionSuspended {
 		return z.deleteObjectFromAllPools(ctx, bucket, object, opts, noReadQuorumPools)
@@ -1916,50 +1927,17 @@ func (z *erasureServerPools) ListMultipartUploads(ctx context.Context, bucket, p
 	if _, err := z.GetBucketInfo(ctx, bucket, BucketOptions{}); err != nil {
 		return ListMultipartsInfo{}, toObjectErr(err, bucket)
 	}
-	if globalAPIConfig.getMultipartListingLegacy() {
-		return z.listMultipartUploadsLegacy(ctx, bucket, prefix, keyMarker, uploadIDMarker, delimiter, maxUploads)
-	}
-	scan, err := startMultipartScan(ctx, false)
-	if err != nil {
-		return ListMultipartsInfo{}, err
-	}
-	defer scan.close()
 
-	var uploads []MultipartInfo
-	var keyless bool
-	for idx, pool := range z.serverPools {
-		if z.IsSuspended(idx) {
-			continue
-		}
-		poolUploads, poolKeyless, err := pool.scanMultipartUploads(scan, bucket, idx)
-		if err != nil {
-			return ListMultipartsInfo{}, err
-		}
-		uploads = append(uploads, poolUploads...)
-		keyless = keyless || poolKeyless
-	}
-
-	// The old format cannot be enumerated authoritatively. Migration mode is
-	// explicit: another bucket must never silently change this API's semantics.
-	if keyless {
-		return ListMultipartsInfo{}, errMultipartListingLegacy
-	}
-	return paginateMultipartUploads(uploads, prefix, keyMarker, uploadIDMarker, delimiter, maxUploads), nil
-}
-
-func (z *erasureServerPools) listMultipartUploadsLegacy(ctx context.Context, bucket, prefix, keyMarker, uploadIDMarker, delimiter string, maxUploads int) (ListMultipartsInfo, error) {
 	poolResult := ListMultipartsInfo{}
 	poolResult.MaxUploads = maxUploads
 	poolResult.KeyMarker = keyMarker
 	poolResult.Prefix = prefix
 	poolResult.Delimiter = delimiter
 
-	// if no prefix provided, return the list from cache
+	// Preserve the legacy cache view, including its incomplete pagination.
+	// Persistent discovery remains an explicit admin diagnostic, not a
+	// prerequisite for public listing or known-upload operations.
 	if prefix == "" {
-		if _, err := z.GetBucketInfo(ctx, bucket, BucketOptions{}); err != nil {
-			return ListMultipartsInfo{}, toObjectErr(err, bucket)
-		}
-
 		z.mpCache.Range(func(_ string, mp MultipartInfo) bool {
 			if mp.Bucket == bucket {
 				poolResult.Uploads = append(poolResult.Uploads, mp)
@@ -2199,28 +2177,15 @@ func (z *erasureServerPools) AbortMultipartUpload(ctx context.Context, bucket, o
 	ctx = lkctx.Context()
 	defer lk.Unlock(lkctx)
 
-	legacy := globalAPIConfig.getMultipartListingLegacy()
-	found := false
-	var firstErr error
 	for idx, pool := range z.serverPools {
 		if z.IsSuspended(idx) {
 			continue
 		}
-		poolFound, err := pool.getHashedSet(object).abortMultipartUpload(ctx, bucket, object, uploadID, opts, legacy)
+		found, err := pool.getHashedSet(object).abortMultipartUpload(ctx, bucket, object, uploadID, opts)
-		if legacy && (poolFound || err != nil) {
+		if found || err != nil {
-			// Match the released first-matching-pool behavior.
+			// Preserve the default first-matching-pool behavior.
 			return err
 		}
-		found = found || poolFound
-		if err != nil && firstErr == nil {
-			firstErr = err
-		}
-	}
-	if firstErr != nil {
-		return firstErr
-	}
-	if found {
-		return nil
 	}
 	return InvalidUploadID{
 		Bucket:   bucket,
@@ -2294,6 +2259,13 @@ func (z *erasureServerPools) CompleteMultipartUpload(ctx context.Context, bucket
 			}
 			opts.CheckPrecondFn = nil
 		}
+		if opts.DataMovement {
+			resolve, err := z.checkDataMovementCopies(ctx, bucket, encodeDirObject(object), opts)
+			if err != nil {
+				return ObjectInfo{}, err
+			}
+			opts.replicaObjectInfo = resolve
+		}
 	}
 
 	// Hold write locks to verify uploaded parts, also disallows any
@@ -2313,7 +2285,7 @@ func (z *erasureServerPools) CompleteMultipartUpload(ctx context.Context, bucket
 	if opts.ReplicaLockReconcile || opts.DataMovement {
 		opts.ReplicaLockReconcile = true
 	}
-	if opts.ReplicaLockReconcile || !opts.ReplicationRequest {
+	if opts.replicaObjectInfo == nil && (opts.ReplicaLockReconcile || !opts.ReplicationRequest) {
 		opts.replicaObjectInfo = z.replicaObjectInfo
 	}
 
@@ -2651,9 +2623,9 @@ func (z *erasureServerPools) Walk(ctx context.Context, bucket, prefix string, re
 					minDisks:       listingQuorum,
 					reportNotFound: false,
 					agreed:         send,
-					partial: func(entries metaCacheEntries, _ []error) {
+					partial: func(entries metaCacheEntries, errs []error) {
-						entry, ok := entries.resolve(&resolver)
+						entry := set.resolveListEntry(entries, errs, &resolver)
-						if ok {
+						if entry != nil {
 							send(*entry)
 						}
 					},
@@ -2755,7 +2727,11 @@ func (z *erasureServerPools) Walk(ctx context.Context, bucket, prefix string, re
 		// Merge all entries from all disks.
 		// We leave quorum at 1, since entries are already resolved to have the desired quorum.
 		// mergeEntryChannels will close 'merged' channel upon completion or cancellation.
-		errCh <- mergeEntryChannels(ctx, entries, merged, 1)
+		errCh <- mergeEntryChannelsWithResolver(ctx, entries, merged, 1, func(int) error {
+			return context.Cause(ctx)
+		}, func(group []*metaCacheEntry) (*metaCacheEntry, error) {
+			return z.resolveListingGroup(ctx, bucket, group)
+		})
 	}()
 
 	return nil
@@ -3300,6 +3276,11 @@ func (z *erasureServerPools) DecomTieredObject(ctx context.Context, bucket, obje
 		defer ns.Unlock(lkctx)
 		opts.NoLock = true
 	}
+	if opts.DataMovement {
+		if _, err := z.checkDataMovementCopies(ctx, bucket, object, opts); err != nil {
+			return err
+		}
+	}
 	idx, err := z.getPoolIdxNoLock(ctx, bucket, object, fi.Size)
 	if err != nil {
 		return err

cmd/erasure-sets.go

@@ -886,36 +886,7 @@ func (s *erasureSets) CopyObject(ctx context.Context, srcBucket, srcObject, dstB
 }
 
 func (s *erasureSets) ListMultipartUploads(ctx context.Context, bucket, prefix, keyMarker, uploadIDMarker, delimiter string, maxUploads int) (result ListMultipartsInfo, err error) {
-	if err := checkListMultipartArgs(ctx, bucket, prefix, keyMarker, uploadIDMarker, delimiter); err != nil {
+	return s.getHashedSet(prefix).ListMultipartUploads(ctx, bucket, prefix, keyMarker, uploadIDMarker, delimiter, maxUploads)
-		return ListMultipartsInfo{}, err
-	}
-	scan, err := startMultipartScan(ctx, false)
-	if err != nil {
-		return ListMultipartsInfo{}, err
-	}
-	defer scan.close()
-	uploads, legacy, err := s.scanMultipartUploads(scan, bucket, 0)
-	if err != nil {
-		return ListMultipartsInfo{}, err
-	}
-	if legacy {
-		return ListMultipartsInfo{}, errMultipartListingLegacy
-	}
-	return paginateMultipartUploads(uploads, prefix, keyMarker, uploadIDMarker, delimiter, maxUploads), nil
-}
-
-func (s *erasureSets) scanMultipartUploads(scan *multipartScan, bucket string, poolIdx int) ([]MultipartInfo, bool, error) {
-	var uploads []MultipartInfo
-	var keyless bool
-	for i, set := range s.sets {
-		setUploads, setKeyless, err := set.scanMultipartUploads(scan, bucket, poolIdx, i)
-		if err != nil {
-			return nil, false, err
-		}
-		uploads = append(uploads, setUploads...)
-		keyless = keyless || setKeyless
-	}
-	return uploads, keyless, nil
 }
 
 func (s *erasureSets) listMultipartUploadsExact(ctx context.Context, bucket, object string) (ListMultipartsInfo, error) {

cmd/globals.go

@@ -319,6 +319,8 @@ var (
 
 	globalOperationTimeout       = newDynamicTimeout(10*time.Minute, 5*time.Minute) // default timeout for general ops
 	globalDeleteOperationTimeout = newDynamicTimeout(5*time.Minute, 1*time.Minute)  // default time for delete ops
+	// Listing recovery has its own failure history and a five-second caller budget.
+	globalListingOperationTimeout = newDynamicTimeout(5*time.Second, 5*time.Second)
 
 	globalBucketObjectLockSys *BucketObjectLockSys
 	globalBucketQuotaSys      *BucketQuotaSys

cmd/handler-api.go

@@ -50,7 +50,6 @@ type apiConfig struct {
 	transitionWorkers      int
 
 	staleUploadsExpiry          time.Duration
-	multipartListingStrict      bool
 	staleUploadsCleanupInterval time.Duration
 	deleteCleanupInterval       time.Duration
 	enableODirect               bool
@@ -182,7 +181,6 @@ func (t *apiConfig) init(cfg api.Config, setDriveCounts []int, legacy bool) {
 	t.transitionWorkers = cfg.TransitionWorkers
 
 	t.staleUploadsExpiry = cfg.StaleUploadsExpiry
-	t.multipartListingStrict = cfg.MultipartListing == "strict"
 	t.deleteCleanupInterval = cfg.DeleteCleanupInterval
 	t.enableODirect = cfg.EnableODirect
 	t.gzipObjects = cfg.GzipObjects
@@ -208,12 +206,6 @@ func (t *apiConfig) odirectEnabled() bool {
 	return t.enableODirect
 }
 
-func (t *apiConfig) getMultipartListingLegacy() bool {
-	t.mu.RLock()
-	defer t.mu.RUnlock()
-	return !t.multipartListingStrict
-}
-
 func (t *apiConfig) shouldGzipObjects() bool {
 	t.mu.RLock()
 	defer t.mu.RUnlock()

cmd/iam-replication-protocol.go

@@ -29,6 +29,7 @@ const (
 	iamRevisionPeerPath  = "/v3/site-replication/peer/iam-revisions"
 	iamUserBoundaryType  = "silo-user-revocation"
 	iamGroupBoundaryType = "silo-group-revocation"
+	iamSTSDeleteType     = "silo-sts-revocation"
 	maxIAMRevisionBatch  = 128
 )
 
@@ -49,6 +50,7 @@ type iamGroupBoundary struct {
 // before applying any change that would lose revocation or member metadata.
 type iamReplicationItem struct {
 	madmin.SRIAMItem
+	STSExpiresAt    time.Time            `json:"stsExpiresAt,omitempty"`
 	GroupGrants     map[string]time.Time `json:"groupGrants,omitempty"`
 	GroupSnapshot   bool                 `json:"groupSnapshot,omitempty"`
 	UserRevocation  *iamUserBoundary     `json:"userRevocation,omitempty"`
@@ -155,6 +157,13 @@ type (
 
 func (c *SiteReplicationSys) replicationItem(ctx context.Context, item madmin.SRIAMItem) (iamReplicationItem, error) {
 	out := iamReplicationItem{SRIAMItem: item}
+	if item.Type == iamSTSDeleteType {
+		r, err := loadIAMRevision(ctx, globalIAMSys.store, getUserIdentityPath(item.Name, stsUser))
+		if err != nil {
+			return out, err
+		}
+		out.STSExpiresAt = r.ExpiresAt
+	}
 	if item.Type == madmin.SRIAMItemSvcAcc && item.SvcAccChange != nil {
 		var key string
 		if item.SvcAccChange.Create != nil {
@@ -242,6 +251,15 @@ func applyIAMReplicationItem(ctx context.Context, item iamReplicationItem) error
 		ctx = context.WithValue(ctx, iamRecordBoundaryKey{}, item.RevokedBefore)
 	}
 	switch item.Type {
+	case iamSTSDeleteType:
+		if item.Name == "" || strings.ContainsAny(item.Name, "/\\") || item.UpdatedAt.IsZero() {
+			return errSRInvalidRequest(errInvalidArgument)
+		}
+		if err := globalIAMSys.store.revokeTempUser(withIAMReplicationTime(ctx, item.UpdatedAt), item.Name, item.STSExpiresAt); err != nil {
+			return iamReplicationError(err)
+		}
+		globalIAMSys.notifyForUser(ctx, item.Name, true)
+		return nil
 	case iamUserBoundaryType:
 		if item.UserRevocation == nil || item.UserRevocation.User == "" || item.UserRevocation.Before.IsZero() {
 			return errSRInvalidRequest(errInvalidArgument)

cmd/iam-revision.go

@@ -348,6 +348,13 @@ func saveIAMRevision(ctx context.Context, store IAMStorageAPI, path string, item
 		// STS access keys identify immutable tokens, not reusable user names.
 		return errIAMStaleUpdate
 	}
+	if strings.HasPrefix(path, iamConfigSTSPrefix) && deleted {
+		// STS keys are immutable. A delayed revocation must also defeat a
+		// replayed creation that was recorded with a newer event timestamp.
+		if origin, replicated := iamReplicationTime(ctx); replicated && previous.timestamp().After(origin) {
+			ctx = withIAMReplicationTime(ctx, previous.timestamp())
+		}
+	}
 	if origin, replicated := iamReplicationTime(ctx); replicated {
 		*at = origin
 		if previous.timestamp().After(origin) || (previous.Deleted && !deleted && !origin.After(previous.timestamp())) {
@@ -394,11 +401,18 @@ func saveIAMRevision(ctx context.Context, store IAMStorageAPI, path string, item
 			// Retain only the parent name for root-account exclusion during heal.
 			v.Credentials = auth.Credentials{ParentUser: previous.Credentials.ParentUser}
 			v.RevokedBefore = *at
-			if strings.HasPrefix(path, iamConfigSTSPrefix) && !previous.Credentials.Expiration.IsZero() && !previous.Credentials.Expiration.Equal(timeSentinel) {
+			if strings.HasPrefix(path, iamConfigSTSPrefix) {
-				// The signed STS token cannot authorize beyond this time, even
+				// Carry expiry when a revocation arrives before the token itself.
-				// if an offline site replays it with a newer event timestamp.
+				// It is safe to forget this key only after its signed lifetime.
-				v.ExpiresAt = previous.Credentials.Expiration.Add(globalMaxSkewTime)
+				if expiration := previous.Credentials.Expiration; !expiration.IsZero() && !expiration.Equal(timeSentinel) && expiration.Add(globalMaxSkewTime).After(v.ExpiresAt) {
-				opts = []options{{ttl: max(1, int64(time.Until(v.ExpiresAt).Seconds())+1)}}
+					v.ExpiresAt = expiration.Add(globalMaxSkewTime)
+				}
+				if previous.ExpiresAt.After(v.ExpiresAt) {
+					v.ExpiresAt = previous.ExpiresAt
+				}
+				if !v.ExpiresAt.IsZero() {
+					opts = []options{{ttl: max(1, int64(time.Until(v.ExpiresAt).Seconds())+1)}}
+				}
 			}
 		} else {
 			if v.Credentials.SessionToken != "" && v.Credentials.Claims == nil {
@@ -489,6 +503,12 @@ func iamDeletionItem(path string, r iamRevision) (item madmin.SRIAMItem, ok bool
 		name := strings.TrimSuffix(strings.TrimPrefix(path, iamConfigUsersPrefix), "/"+iamIdentityFile)
 		item.Type = madmin.SRIAMItemIAMUser
 		item.IAMUser = &madmin.SRIAMUser{AccessKey: name, IsDeleteReq: true}
+	case strings.HasPrefix(path, iamConfigSTSPrefix):
+		if !r.ExpiresAt.IsZero() && UTCNow().After(r.ExpiresAt) {
+			return item, false
+		}
+		item.Type = iamSTSDeleteType
+		item.Name = strings.TrimSuffix(strings.TrimPrefix(path, iamConfigSTSPrefix), "/"+iamIdentityFile)
 	case strings.HasPrefix(path, iamConfigServiceAccountsPrefix):
 		name := strings.TrimSuffix(strings.TrimPrefix(path, iamConfigServiceAccountsPrefix), "/"+iamIdentityFile)
 		if name == siteReplicatorSvcAcc || r.Credentials.ParentUser == globalActiveCred.AccessKey {
@@ -518,8 +538,7 @@ func iamDeletionItem(path string, r iamRevision) (item madmin.SRIAMItem, ok bool
 		item.Type = madmin.SRIAMItemPolicyMapping
 		item.PolicyMapping = &madmin.SRPolicyMapping{UserOrGroup: strings.TrimSuffix(name, ".json"), UserType: int(typ), IsGroup: prefix == "groups"}
 	default:
-		// Expired STS credentials are not replayed. Parent revocations and
+		// Other config paths do not represent replicated IAM deletions.
-		// their retained timestamp reject delayed copies of derived tokens.
 		return item, false
 	}
 	return item, true
@@ -527,6 +546,8 @@ func iamDeletionItem(path string, r iamRevision) (item madmin.SRIAMItem, ok bool
 
 func iamDeletionPath(item madmin.SRIAMItem) string {
 	switch item.Type {
+	case iamSTSDeleteType:
+		return getUserIdentityPath(item.Name, stsUser)
 	case madmin.SRIAMItemIAMUser:
 		if item.IAMUser != nil && item.IAMUser.IsDeleteReq {
 			return getUserIdentityPath(item.IAMUser.AccessKey, regUser)
@@ -582,6 +603,9 @@ func (c *SiteReplicationSys) healIAMDeletions(ctx context.Context) (err error) {
 			continue
 		}
 		out := iamReplicationItem{SRIAMItem: item}
+		if item.Type == iamSTSDeleteType {
+			out.STSExpiresAt = r.ExpiresAt
+		}
 		if item.Type == madmin.SRIAMItemIAMUser && !r.Deleted {
 			out.Type, out.IAMUser = iamUserBoundaryType, nil
 			out.UserRevocation = &iamUserBoundary{User: item.IAMUser.AccessKey, Before: r.RevokedBefore}
@@ -593,7 +617,7 @@ func (c *SiteReplicationSys) healIAMDeletions(ctx context.Context) (err error) {
 		byType[item.Type] = append(byType[item.Type], out)
 	}
 	var items []iamReplicationItem
-	for _, typ := range []string{madmin.SRIAMItemPolicyMapping, madmin.SRIAMItemIAMUser, madmin.SRIAMItemSvcAcc, madmin.SRIAMItemGroupInfo, madmin.SRIAMItemPolicy} {
+	for _, typ := range []string{madmin.SRIAMItemPolicyMapping, madmin.SRIAMItemIAMUser, madmin.SRIAMItemSvcAcc, iamSTSDeleteType, madmin.SRIAMItemGroupInfo, madmin.SRIAMItemPolicy} {
 		items = append(items, byType[typ]...)
 	}
 	if len(items) == 0 {

cmd/iam-store.go

@@ -711,6 +711,11 @@ func (store *IAMStoreSys) LoadIAMCache(ctx context.Context, firstTime bool) erro
 		if err := store.loadUsers(ctx, svcUser, newCache.iamUsersMap); err != nil {
 			return err
 		}
+		// Rebuild STS revisions as well: a missed etcd watch event must not
+		// leave a cached token valid or hide its tombstone after a restart.
+		if err := store.loadUsers(ctx, stsUser, newCache.iamSTSAccountsMap); err != nil {
+			return err
+		}
 
 		newCache.buildUserGroupMemberships()
 	}
@@ -754,6 +759,14 @@ func (store *IAMStoreSys) LoadIAMCache(ctx context.Context, firstTime bool) erro
 		cache.updatedAt = time.Now()
 	}
 
+	// Revision loading is independent of cache replacement. Even if ongoing
+	// IAM writes defer the full swap, a missed STS notification must converge.
+	for key := range cache.iamSTSAccountsMap {
+		if store.revisionIndex().get(getUserIdentityPath(key, stsUser)).Deleted {
+			delete(cache.iamSTSAccountsMap, key)
+		}
+	}
+
 	return nil
 }
 
@@ -2184,45 +2197,68 @@ func (store *IAMStoreSys) SetTempUser(ctx context.Context, accessKey string, cre
 
 // RevokeTokens - revokes all temporary credentials, or those with matching type,
 // associated with the parent user.
-func (store *IAMStoreSys) RevokeTokens(ctx context.Context, parentUser string, tokenRevokeType string) error {
+func (store *IAMStoreSys) RevokeTokens(ctx context.Context, parentUser string, tokenRevokeType string) ([]string, error) {
 	if parentUser == "" {
-		return errInvalidArgument
+		return nil, errInvalidArgument
 	}
-
+	before := UTCNow()
 	cache := store.lock()
 	defer store.unlock()
-
 	secret, err := getTokenSigningKey()
 	if err != nil {
-		return err
+		return nil, err
 	}
-
+	// Another node may have issued a token whose notification was lost.
-	var revoked bool
+	// Enumerate durable identities, retaining cached entries for retrying a
-	for _, ui := range cache.iamSTSAccountsMap {
+	// partially propagated revocation. Do not hold up cached authentication.
-		if ui.Credentials.ParentUser != parentUser {
+	accounts := maps.Clone(cache.iamSTSAccountsMap)
+	if err := store.withIAMStorage(ctx, func(ctx context.Context) error {
+		return store.loadUsers(ctx, stsUser, accounts)
+	}); err != nil {
+		return nil, err
+	}
+	var revoked []string
+	var failures []error
+	for key, ui := range accounts {
+		if ui.Credentials.ParentUser != parentUser || ui.UpdatedAt.After(before) || ui.Credentials.IsExpired() {
 			continue
 		}
 		if tokenRevokeType != "" {
 			claims, err := getClaimsFromTokenWithSecret(ui.Credentials.SessionToken, secret)
 			if err != nil {
-				continue // skip if token is invalid
+				continue
 			}
-			// skip if token type is given and does not match
 			if v, _ := claims.Lookup(tokenRevokeTypeClaim); v != tokenRevokeType {
 				continue
 			}
 		}
-		if err := store.deleteUserIdentity(ctx, ui.Credentials.AccessKey, stsUser); err != nil {
+		if err := store.deleteUserIdentity(ctx, key, stsUser); err != nil {
-			return err
+			failures = append(failures, err)
+			continue
 		}
-		delete(cache.iamSTSAccountsMap, ui.Credentials.AccessKey)
+		delete(cache.iamSTSAccountsMap, key)
-		revoked = true
+		revoked = append(revoked, key)
 	}
-
+	if len(revoked) != 0 {
-	if revoked {
 		cache.updatedAt = time.Now()
 	}
+	return revoked, errors.Join(failures...)
+}
 
+// revokeTempUser applies an immutable token's replicated tombstone without
+// touching the parent, its policies, or other (including newer) tokens.
+func (store *IAMStoreSys) revokeTempUser(ctx context.Context, key string, expiresAt time.Time) error {
+	cache := store.lock()
+	defer store.unlock()
+	u := UserIdentity{Version: 1, Deleted: true, ExpiresAt: expiresAt}
+	if err := store.saveUserIdentity(ctx, key, stsUser, &u); err != nil {
+		return err
+	}
+	delete(cache.iamSTSAccountsMap, key)
+	cache.updatedAt = time.Now()
+	if store.group != nil {
+		store.group.Forget(key)
+	}
 	return nil
 }
 

cmd/iam-sts-revocation_test.go

@@ -0,0 +1,254 @@
+// Copyright (c) 2026 Feng Ruohang
+// SPDX-License-Identifier: AGPL-3.0-or-later
+
+package cmd
+
+import (
+	"context"
+	"errors"
+	"fmt"
+	"strings"
+	"testing"
+	"time"
+
+	"github.com/minio/madmin-go/v3"
+	"github.com/minio/minio/internal/auth"
+	"github.com/minio/minio/internal/grid"
+	xnet "github.com/pgsty/silo-pkg/v3/net"
+)
+
+func TestIAMSTSRevocationDurableScope(t *testing.T) {
+	for _, backend := range []string{"object", "etcd"} {
+		for _, notify := range []bool{false, true} {
+			t.Run(fmt.Sprintf("%s/notify=%t", backend, notify), func(t *testing.T) {
+				ctx, sys, obj := prepareIAMRevisionFixture(t, backend)
+				for _, parent := range []string{"sts-parent", "other-parent"} {
+					_, err := sys.CreateUser(ctx, parent, madmin.AddOrUpdateUserReq{SecretKey: "synthetic-parent-password", Status: madmin.AccountEnabled})
+					mustIAM(t, err)
+				}
+				sibling := &IAMStoreSys{IAMStorageAPI: newIAMObjectStore(obj, MinIOUsersSysType)}
+				if backend == "etcd" {
+					sibling.IAMStorageAPI = newIAMEtcdStore(sys.store.IAMStorageAPI.(*IAMEtcdStore).client, MinIOUsersSysType)
+				}
+				mustIAM(t, sibling.LoadIAMCache(ctx, false))
+				secret, err := getTokenSigningKey()
+				mustIAM(t, err)
+				issue := func(store *IAMStoreSys, parent, typ string) auth.Credentials {
+					cred, err := auth.GetNewCredentialsWithMetadata(map[string]any{"exp": UTCNow().Add(time.Hour).Unix(), parentClaim: parent, tokenRevokeTypeClaim: typ}, secret)
+					mustIAM(t, err)
+					cred.ParentUser = parent
+					_, err = store.SetTempUser(ctx, cred.AccessKey, cred, "")
+					mustIAM(t, err)
+					return cred
+				}
+				first := issue(sys.store, "sts-parent", "selected")
+				uncached := issue(sibling, "sts-parent", "selected")
+				otherType := issue(sys.store, "sts-parent", "other-type")
+				otherUser := issue(sys.store, "other-parent", "selected")
+				if _, ok := sys.store.GetUser(uncached.AccessKey); ok {
+					t.Fatal("fixture cached the remote token")
+				}
+				for _, c := range []auth.Credentials{first, otherType, otherUser} {
+					mustIAM(t, sibling.UserNotificationHandler(ctx, c.AccessKey, stsUser))
+				}
+				prior := globalNotificationSys
+				globalNotificationSys = &NotificationSys{}
+				t.Cleanup(func() { globalNotificationSys = prior })
+				if notify {
+					tg, err := grid.SetupTestGrid(2)
+					mustIAM(t, err)
+					t.Cleanup(tg.Cleanup)
+					mustIAM(t, loadUserRPC.Register(tg.Managers[1], func(r *grid.MSS) (grid.NoPayload, *grid.RemoteErr) {
+						if err := sibling.UserNotificationHandler(ctx, r.Get(peerRESTUser), stsUser); err != nil {
+							return grid.NoPayload{}, grid.NewRemoteErr(err)
+						}
+						return grid.NoPayload{}, nil
+					}))
+					host, err := xnet.ParseHost(strings.TrimPrefix(tg.Hosts[1], "http://"))
+					mustIAM(t, err)
+					globalNotificationSys = &NotificationSys{peerClients: []*peerRESTClient{{host: host, gridConn: func() *grid.Connection { return tg.Managers[0].Connection(tg.Hosts[1]) }}}}
+				}
+				mustIAM(t, sys.RevokeTokens(ctx, "sts-parent", "selected"))
+				for _, c := range []auth.Credentials{first, uncached} {
+					r, err := loadIAMRevision(ctx, sys.store, getUserIdentityPath(c.AccessKey, stsUser))
+					mustIAM(t, err)
+					if !r.Deleted {
+						t.Errorf("durable token was omitted: %s", c.AccessKey)
+					}
+					if _, ok := iamDeletionItem(getUserIdentityPath(c.AccessKey, stsUser), r); !ok {
+						t.Error("STS revocation cannot be replayed")
+					}
+					if notify && backend == "object" {
+						if _, ok := sibling.GetUser(c.AccessKey); ok {
+							t.Error("notified sibling accepted revoked token")
+						}
+					}
+				}
+				fresh := issue(sibling, "sts-parent", "selected")
+				// A missed notification must converge through the normal cache refresh.
+				mustIAM(t, sibling.LoadIAMCache(ctx, false))
+				for _, c := range []auth.Credentials{first, uncached} {
+					if _, ok := sibling.GetUser(c.AccessKey); ok {
+						t.Error("refresh retained revoked token")
+					}
+				}
+				for _, c := range []auth.Credentials{otherType, otherUser, fresh} {
+					mustIAM(t, sibling.UserNotificationHandler(ctx, c.AccessKey, stsUser))
+					if _, ok := sibling.GetUser(c.AccessKey); !ok {
+						t.Error("revocation removed an out-of-scope or new token")
+					}
+				}
+				if _, ok := sibling.GetUser("sts-parent"); !ok {
+					t.Error("revocation removed parent")
+				}
+			})
+		}
+	}
+}
+
+// Deletion is keyed by the immutable STS access key, so delayed delivery cannot
+// affect a newly issued token and a delayed creation cannot undo revocation.
+func TestIAMSTSRevocationReorderedReplay(t *testing.T) {
+	for _, backend := range []string{"object", "etcd"} {
+		for _, missing := range []bool{false, true} {
+			t.Run(fmt.Sprintf("%s/missing=%t", backend, missing), func(t *testing.T) {
+				ctx, sys, _ := prepareIAMRevisionFixture(t, backend)
+				const parent = "sts-replay-parent"
+				_, err := sys.CreateUser(ctx, parent, madmin.AddOrUpdateUserReq{SecretKey: "synthetic-parent-password", Status: madmin.AccountEnabled})
+				mustIAM(t, err)
+				secret, err := getTokenSigningKey()
+				mustIAM(t, err)
+				cred, err := auth.GetNewCredentialsWithMetadata(map[string]any{"exp": UTCNow().Add(time.Hour).Unix(), parentClaim: parent}, secret)
+				mustIAM(t, err)
+				cred.ParentUser = parent
+				created := UTCNow().Add(-time.Minute)
+				peer := madmin.SRSTSCredential{AccessKey: cred.AccessKey, SecretKey: cred.SecretKey, SessionToken: cred.SessionToken, ParentUser: parent}
+				if !missing {
+					mustIAM(t, globalSiteReplicationSys.PeerSTSAccHandler(ctx, &peer, created))
+				}
+				fresh, err := auth.GetNewCredentialsWithMetadata(map[string]any{"exp": UTCNow().Add(time.Hour).Unix(), parentClaim: parent}, secret)
+				mustIAM(t, err)
+				fresh.ParentUser = parent
+				_, err = sys.SetTempUser(ctx, fresh.AccessKey, fresh, "")
+				mustIAM(t, err)
+				// Older than the live record on this peer: deletion must still win.
+				item := iamReplicationItem{SRIAMItem: madmin.SRIAMItem{Type: iamSTSDeleteType, Name: cred.AccessKey, UpdatedAt: created.Add(-time.Second)}, STSExpiresAt: cred.Expiration.Add(globalMaxSkewTime)}
+				for range 2 {
+					mustIAM(t, applyIAMReplicationItem(ctx, item))
+					mustIAM(t, globalSiteReplicationSys.PeerSTSAccHandler(ctx, &peer, UTCNow()))
+					mustIAM(t, sys.store.LoadIAMCache(ctx, false))
+					if _, ok := sys.GetUser(ctx, cred.AccessKey); ok {
+						t.Fatal("delayed STS creation undid revocation")
+					}
+					if _, ok := sys.GetUser(ctx, fresh.AccessKey); !ok {
+						t.Fatal("delayed revocation removed new token")
+					}
+				}
+				r, err := loadIAMRevision(ctx, sys.store, getUserIdentityPath(cred.AccessKey, stsUser))
+				mustIAM(t, err)
+				if !r.Deleted || !r.ExpiresAt.Equal(item.STSExpiresAt) || r.Credentials.SessionToken != "" || r.Credentials.SecretKey != "" {
+					t.Fatal("revocation lost expiry or retained secret material")
+				}
+				for _, status := range []madmin.AccountStatus{madmin.AccountDisabled, madmin.AccountEnabled} {
+					_, err = sys.SetUserStatus(ctx, parent, status)
+					mustIAM(t, err)
+				}
+				if _, ok := sys.GetUser(ctx, cred.AccessKey); ok {
+					t.Fatal("parent re-enable revived revoked token")
+				}
+			})
+		}
+	}
+}
+
+// Pause the durable scan while another node persists a new credential. The
+// revocation's original cutoff must survive the scan and leave that token live.
+type iamSTSScanGateStore struct {
+	IAMStorageAPI
+	entered chan struct{}
+	resume  chan struct{}
+}
+
+func (s *iamSTSScanGateStore) loadUsers(ctx context.Context, typ IAMUserType, users map[string]UserIdentity) error {
+	close(s.entered)
+	select {
+	case <-s.resume:
+	case <-ctx.Done():
+		return ctx.Err()
+	}
+	return s.IAMStorageAPI.loadUsers(ctx, typ, users)
+}
+
+func TestIAMSTSRevocationConcurrentIssuance(t *testing.T) {
+	ctx, sys, obj := prepareIAMRevisionFixture(t)
+	_, err := sys.CreateUser(ctx, "sts-race-parent", madmin.AddOrUpdateUserReq{SecretKey: "synthetic-parent-password", Status: madmin.AccountEnabled})
+	mustIAM(t, err)
+	sibling := &IAMStoreSys{IAMStorageAPI: newIAMObjectStore(obj, MinIOUsersSysType)}
+	secret, err := getTokenSigningKey()
+	mustIAM(t, err)
+	issue := func() auth.Credentials {
+		c, err := auth.GetNewCredentialsWithMetadata(map[string]any{"exp": UTCNow().Add(time.Hour).Unix(), parentClaim: "sts-race-parent"}, secret)
+		mustIAM(t, err)
+		c.ParentUser = "sts-race-parent"
+		_, err = sibling.SetTempUser(ctx, c.AccessKey, c, "")
+		mustIAM(t, err)
+		return c
+	}
+	old := issue()
+	gate := &iamSTSScanGateStore{IAMStorageAPI: sys.store.IAMStorageAPI, entered: make(chan struct{}), resume: make(chan struct{})}
+	sys.store.IAMStorageAPI = gate
+	done := make(chan error, 1)
+	go func() { done <- sys.RevokeTokens(ctx, "sts-race-parent", "") }()
+	<-gate.entered
+	fresh := issue()
+	close(gate.resume)
+	mustIAM(t, <-done)
+	mustIAM(t, sibling.UserNotificationHandler(ctx, old.AccessKey, stsUser))
+	if _, ok := sibling.GetUser(old.AccessKey); ok {
+		t.Fatal("old uncached token survived")
+	}
+	mustIAM(t, sibling.UserNotificationHandler(ctx, fresh.AccessKey, stsUser))
+	if _, ok := sibling.GetUser(fresh.AccessKey); !ok {
+		t.Fatal("new token was included in earlier revocation")
+	}
+}
+
+type iamSTSFailureStore struct {
+	IAMStorageAPI
+	failedPath string
+}
+
+func (s *iamSTSFailureStore) saveIAMConfig(ctx context.Context, item any, path string, opts ...options) error {
+	if path == s.failedPath {
+		return errIAMInjectedWrite
+	}
+	return s.IAMStorageAPI.saveIAMConfig(ctx, item, path, opts...)
+}
+
+func TestIAMSTSRevocationPartialFailure(t *testing.T) {
+	ctx, sys, _ := prepareIAMRevisionFixture(t)
+	_, err := sys.CreateUser(ctx, "sts-partial-parent", madmin.AddOrUpdateUserReq{SecretKey: "synthetic-parent-password", Status: madmin.AccountEnabled})
+	mustIAM(t, err)
+	secret, err := getTokenSigningKey()
+	mustIAM(t, err)
+	var keys []string
+	for range 2 {
+		c, err := auth.GetNewCredentialsWithMetadata(map[string]any{"exp": UTCNow().Add(time.Hour).Unix(), parentClaim: "sts-partial-parent"}, secret)
+		mustIAM(t, err)
+		c.ParentUser = "sts-partial-parent"
+		_, err = sys.SetTempUser(ctx, c.AccessKey, c, "")
+		mustIAM(t, err)
+		keys = append(keys, c.AccessKey)
+	}
+	sys.store.IAMStorageAPI = &iamSTSFailureStore{IAMStorageAPI: sys.store.IAMStorageAPI, failedPath: getUserIdentityPath(keys[0], stsUser)}
+	revoked, err := sys.store.RevokeTokens(ctx, "sts-partial-parent", "")
+	if !errors.Is(err, errIAMInjectedWrite) || len(revoked) != 1 || revoked[0] != keys[1] {
+		t.Fatalf("committed keys or failure lost: %v %v", revoked, err)
+	}
+	if _, ok := sys.store.GetUser(keys[1]); ok {
+		t.Fatal("committed revocation retained cache entry")
+	}
+	if _, ok := sys.store.GetUser(keys[0]); !ok {
+		t.Fatal("failed revocation claimed cache success")
+	}
+}

cmd/iam.go

@@ -717,7 +717,14 @@ func (sys *IAMSys) RevokeTokens(ctx context.Context, accessKey, tokenRevokeType
 		return errServerNotInitialized
 	}
 
-	return sys.store.RevokeTokens(ctx, accessKey, tokenRevokeType)
+	keys, err := sys.store.RevokeTokens(ctx, accessKey, tokenRevokeType)
+	// Publish every committed tombstone even if another token failed to save.
+	for _, key := range keys {
+		sys.notifyForUser(ctx, key, true)
+		// Offline peers recover this same record through IAM deletion healing.
+		iamLogIf(ctx, globalSiteReplicationSys.IAMChangeHook(ctx, madmin.SRIAMItem{Type: iamSTSDeleteType, Name: key}))
+	}
+	return err
 }
 
 // DeleteUser - delete user (only for long-term users not STS users).

cmd/list-multipart-uploads-compat_test.go

@@ -19,7 +19,6 @@ package cmd
 
 import (
 	"bytes"
-	"errors"
 	"fmt"
 	"slices"
 	"testing"
@@ -41,13 +40,13 @@ func requireMultipartUploadKeys(t *testing.T, got ListMultipartsInfo, want ...st
 	}
 }
 
-func TestListMultipartUploadsS3Compatibility(t *testing.T) {
+func TestMultipartDiagnosticScanPagination(t *testing.T) {
 	obj, dirs, err := prepareErasureSets32(t.Context())
 	if err != nil {
 		t.Fatal(err)
 	}
 	z := obj.(*erasureServerPools)
-	setMultipartListingTestMode(t, false)
+
 	t.Cleanup(func() {
 		z.Shutdown(t.Context())
 		removeRoots(dirs)
@@ -82,14 +81,14 @@ func TestListMultipartUploadsS3Compatibility(t *testing.T) {
 		uploadIDs[object] = mp.UploadID
 	}
 
-	// Durable multipart metadata, rather than this node-local cache, must be
+	// The diagnostic scanner must discover metadata independently of the cache.
-	// authoritative after a restart or when another node handles the request.
+	// These checks do not claim that public listing offers durable pagination.
 	z.mpCache.Range(func(uploadID string, _ MultipartInfo) bool {
 		z.mpCache.Delete(uploadID)
 		return true
 	})
 
-	all, err := z.ListMultipartUploads(t.Context(), bucket, "", "", "", "", 100)
+	all, err := z.scanMultipartUploadsForTest(t.Context(), bucket, "", "", "", "", 100)
 	if err != nil {
 		t.Fatal(err)
 	}
@@ -98,7 +97,7 @@ func TestListMultipartUploadsS3Compatibility(t *testing.T) {
 		t.Fatalf("complete listing has truncation state: %+v", all)
 	}
 
-	first, err := z.ListMultipartUploads(t.Context(), bucket, "", "", "", "", 1)
+	first, err := z.scanMultipartUploadsForTest(t.Context(), bucket, "", "", "", "", 1)
 	if err != nil {
 		t.Fatal(err)
 	}
@@ -109,31 +108,31 @@ func TestListMultipartUploadsS3Compatibility(t *testing.T) {
 			objects[0], uploadIDs[objects[0]])
 	}
 
-	rest, err := z.ListMultipartUploads(t.Context(), bucket, "", first.NextKeyMarker, first.NextUploadIDMarker, "", 100)
+	rest, err := z.scanMultipartUploadsForTest(t.Context(), bucket, "", first.NextKeyMarker, first.NextUploadIDMarker, "", 100)
 	if err != nil {
 		t.Fatal(err)
 	}
 	requireMultipartUploadKeys(t, rest, objects[1:]...)
 
-	afterKey, err := z.ListMultipartUploads(t.Context(), bucket, "", objects[1], "", "", 100)
+	afterKey, err := z.scanMultipartUploadsForTest(t.Context(), bucket, "", objects[1], "", "", 100)
 	if err != nil {
 		t.Fatal(err)
 	}
 	requireMultipartUploadKeys(t, afterKey, objects[2:]...)
 
-	prefixed, err := z.ListMultipartUploads(t.Context(), bucket, "t/", "", "", "", 100)
+	prefixed, err := z.scanMultipartUploadsForTest(t.Context(), bucket, "t/", "", "", "", 100)
 	if err != nil {
 		t.Fatal(err)
 	}
 	requireMultipartUploadKeys(t, prefixed, objects[:3]...)
 
-	nested, err := z.ListMultipartUploads(t.Context(), bucket, "t/a_b/", "", "", "", 100)
+	nested, err := z.scanMultipartUploadsForTest(t.Context(), bucket, "t/a_b/", "", "", "", 100)
 	if err != nil {
 		t.Fatal(err)
 	}
 	requireMultipartUploadKeys(t, nested, objects[:2]...)
 
-	grouped, err := z.ListMultipartUploads(t.Context(), bucket, "t/", "", "", SlashSeparator, 100)
+	grouped, err := z.scanMultipartUploadsForTest(t.Context(), bucket, "t/", "", "", SlashSeparator, 100)
 	if err != nil {
 		t.Fatal(err)
 	}
@@ -142,7 +141,7 @@ func TestListMultipartUploadsS3Compatibility(t *testing.T) {
 		t.Fatalf("common prefixes = %v, want %v", grouped.CommonPrefixes, want)
 	}
 
-	groupPage, err := z.ListMultipartUploads(t.Context(), bucket, "t/", "", "", SlashSeparator, 1)
+	groupPage, err := z.scanMultipartUploadsForTest(t.Context(), bucket, "t/", "", "", SlashSeparator, 1)
 	if err != nil {
 		t.Fatal(err)
 	}
@@ -154,7 +153,7 @@ func TestListMultipartUploadsS3Compatibility(t *testing.T) {
 			groupPage.NextKeyMarker, groupPage.NextUploadIDMarker, groupPage.IsTruncated)
 	}
 
-	groupRest, err := z.ListMultipartUploads(t.Context(), bucket, "t/", groupPage.NextKeyMarker, "", SlashSeparator, 1)
+	groupRest, err := z.scanMultipartUploadsForTest(t.Context(), bucket, "t/", groupPage.NextKeyMarker, "", SlashSeparator, 1)
 	if err != nil {
 		t.Fatal(err)
 	}
@@ -182,8 +181,8 @@ func TestListMultipartUploadsS3Compatibility(t *testing.T) {
 	}
 
 	// Simulate an upload written by a pre-upgrade server. Its key cannot be
-	// recovered by scanning the hashed namespace. Strict listing must fail;
+	// recovered by scanning the hashed namespace. Omit only that record, and
-	// the old path is available only through an explicit migration setting.
+	// preserve normal operations when its bucket, key and upload ID are known.
 	legacyObject := objects[1]
 	er := sets.getHashedSet(legacyObject)
 	fi, metadata, err := er.checkUploadIDExists(t.Context(), bucket, legacyObject, uploadIDs[legacyObject], true)
@@ -198,16 +197,25 @@ func TestListMultipartUploadsS3Compatibility(t *testing.T) {
 		er.getUploadIDDir(bucket, legacyObject, uploadIDs[legacyObject]), metadata, fi.WriteQuorum(er.defaultWQuorum())); err != nil {
 		t.Fatal(err)
 	}
-	_, err = z.ListMultipartUploads(t.Context(), bucket, legacyObject, "", "", "", 100)
+	legacy, err := z.scanMultipartUploadsForTest(t.Context(), bucket, legacyObject, "", "", "", 100)
-	if !errors.Is(err, errMultipartListingLegacy) {
-		t.Fatalf("legacy strict listing: %v", err)
-	}
-	setMultipartListingTestMode(t, true)
-	legacy, err := z.ListMultipartUploads(t.Context(), bucket, legacyObject, "", "", "", 100)
 	if err != nil {
 		t.Fatal(err)
 	}
-	requireMultipartUploadKeys(t, legacy, legacyObject)
+	requireMultipartUploadKeys(t, legacy)
+	remaining, err := z.scanMultipartUploadsForTest(t.Context(), bucket, "", "", "", "", 100)
+	if err != nil {
+		t.Fatal(err)
+	}
+	requireMultipartUploadKeys(t, remaining, objects[2:]...)
+	part, err = z.PutObjectPart(t.Context(), bucket, legacyObject, uploadIDs[legacyObject], 1,
+		mustGetPutObjReader(t, bytes.NewBufferString("old part"), 8, "", ""), ObjectOptions{})
+	if err != nil {
+		t.Fatalf("continue historical upload: %v", err)
+	}
+	if _, err = z.CompleteMultipartUpload(t.Context(), bucket, legacyObject, uploadIDs[legacyObject],
+		[]CompletePart{{PartNumber: 1, ETag: part.ETag}}, ObjectOptions{}); err != nil {
+		t.Fatalf("complete historical upload: %v", err)
+	}
 }
 
 func TestPaginateMultipartUploads(t *testing.T) {
@@ -260,7 +268,7 @@ func TestPaginateMultipartUploads(t *testing.T) {
 
 func TestListMultipartUploadsGlobalPageAcrossPools(t *testing.T) {
 	z, bucket := consistencyPools(t)
-	setMultipartListingTestMode(t, false)
+
 	objects := []string{"a/one", "b/two", "c/three", "d/four"}
 	for i, object := range objects {
 		if _, err := z.serverPools[i%len(z.serverPools)].NewMultipartUpload(t.Context(), bucket, object, ObjectOptions{}); err != nil {
@@ -272,7 +280,7 @@ func TestListMultipartUploadsGlobalPageAcrossPools(t *testing.T) {
 		return true
 	})
 
-	page, err := z.ListMultipartUploads(t.Context(), bucket, "", "", "", "", 2)
+	page, err := z.scanMultipartUploadsForTest(t.Context(), bucket, "", "", "", "", 2)
 	if err != nil {
 		t.Fatal(err)
 	}
@@ -281,7 +289,7 @@ func TestListMultipartUploadsGlobalPageAcrossPools(t *testing.T) {
 		t.Fatalf("first global page = %+v", page)
 	}
 
-	rest, err := z.ListMultipartUploads(t.Context(), bucket, "", page.NextKeyMarker, page.NextUploadIDMarker, "", 2)
+	rest, err := z.scanMultipartUploadsForTest(t.Context(), bucket, "", page.NextKeyMarker, page.NextUploadIDMarker, "", 2)
 	if err != nil {
 		t.Fatal(err)
 	}

cmd/local-locker-listing-reverse_test.go

@@ -0,0 +1,412 @@
+// Copyright (c) 2026 Feng Ruohang
+//
+// This file is part of Silo Object Storage stack
+//
+// This program is free software: you can redistribute it and/or modify
+// it under the terms of the GNU Affero General Public License as published by
+// the Free Software Foundation, either version 3 of the License, or
+// (at your option) any later version.
+//
+// This program is distributed in the hope that it will be useful,
+// but WITHOUT ANY WARRANTY; without even the implied warranty of
+// MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
+// GNU Affero General Public License for more details.
+//
+// You should have received a copy of the GNU Affero General Public License
+// along with this program. If not, see <http://www.gnu.org/licenses/>.
+
+package cmd
+
+import (
+	"context"
+	"fmt"
+	"sync/atomic"
+	"testing"
+	"testing/synctest"
+	"time"
+
+	"github.com/minio/minio/internal/dsync"
+)
+
+func listingLockerTestWrite(t *testing.T, l *localLocker, args dsync.LockArgs) {
+	t.Helper()
+	if ok, err := l.Lock(t.Context(), args); !ok || err != nil {
+		t.Fatalf("Lock = %v, %v", ok, err)
+	}
+}
+
+func listingLockerTestUnlock(t *testing.T, l *localLocker, args dsync.LockArgs) {
+	t.Helper()
+	if ok, err := l.Unlock(t.Context(), args); !ok || err != nil {
+		t.Fatalf("Unlock = %v, %v", ok, err)
+	}
+}
+
+func TestLocalLockerListingReverseWaitRechecks(t *testing.T) {
+	for _, tc := range []struct {
+		name   string
+		change func(*testing.T, *localLocker, dsync.LockArgs)
+		want   bool
+	}{
+		{"writer-released", func(t *testing.T, l *localLocker, w dsync.LockArgs) { listingLockerTestUnlock(t, l, w) }, true},
+		{"writer-still-held", func(*testing.T, *localLocker, dsync.LockArgs) {}, false},
+		{"replacement-writer", func(t *testing.T, l *localLocker, w dsync.LockArgs) {
+			listingLockerTestUnlock(t, l, w)
+			listingLockerTestWrite(t, l, listingLockerTestArgs("new-writer", "key", "writer"))
+		}, false},
+		{"ordinary-reader-arrives", func(t *testing.T, l *localLocker, w dsync.LockArgs) {
+			listingLockerTestUnlock(t, l, w)
+			listingLockerTestRead(t, l, listingLockerTestArgs("ordinary-reader", "key", "ordinary-reader"))
+		}, true},
+		{"listing-reader-arrives", func(t *testing.T, l *localLocker, w dsync.LockArgs) {
+			listingLockerTestUnlock(t, l, w)
+			listingLockerTestRead(t, l, listingLockerTestArgs("other-listing", "key", listingLockerTestSource))
+		}, true},
+		{"force-unlock", func(t *testing.T, l *localLocker, w dsync.LockArgs) {
+			if ok, err := l.ForceUnlock(t.Context(), w); !ok || err != nil {
+				t.Fatalf("ForceUnlock = %v, %v", ok, err)
+			}
+		}, true},
+		{"expiry", func(_ *testing.T, l *localLocker, _ dsync.LockArgs) { l.expireOldLocks(-time.Minute) }, true},
+	} {
+		t.Run(tc.name, func(t *testing.T) {
+			synctest.Test(t, func(t *testing.T) {
+				l := newLocker()
+				writer := listingLockerTestArgs("writer", "key", "writer")
+				reader := listingLockerTestArgs("listing", "key", listingLockerTestSource)
+				listingLockerTestWrite(t, l, writer)
+				var got bool
+				var err error
+				start := time.Now()
+				go func() { got, err = l.RLock(t.Context(), reader) }()
+				synctest.Wait()
+				if state := listingLockerTestSnapshot(l); state.waiting != 1 {
+					t.Fatalf("waiting = %d, want 1", state.waiting)
+				}
+				listingLockerTestWrite(t, l, listingLockerTestArgs("unrelated", "other-key", "writer"))
+				if elapsed := time.Since(start); elapsed != 0 {
+					t.Fatalf("unrelated key waited %v", elapsed)
+				}
+				time.Sleep(time.Millisecond)
+				tc.change(t, l, writer)
+				time.Sleep(time.Millisecond)
+				synctest.Wait()
+				if state := listingLockerTestSnapshot(l); got != tc.want || err != nil || state.waiting != 0 {
+					t.Fatalf("RLock = %v, %v, waiting = %d; want %v, nil, 0", got, err, state.waiting, tc.want)
+				}
+				if elapsed := time.Since(start); elapsed != 2*time.Millisecond {
+					t.Fatalf("one attempt elapsed %v, want 2ms", elapsed)
+				}
+				found := false
+				for _, holder := range l.DupLockMap()["key"] {
+					if holder.UID == reader.UID {
+						found = true
+						if holder.Writer || holder.Timestamp != UTCNow().UnixNano() || holder.TimeLastRefresh != holder.Timestamp {
+							t.Fatalf("invalid granted reader: %+v", holder)
+						}
+					}
+				}
+				if found != got {
+					t.Fatalf("reader present = %v, granted = %v", found, got)
+				}
+			})
+		})
+	}
+}
+
+func TestLocalLockerListingReverseWaitCancellation(t *testing.T) {
+	for _, when := range []string{"before-call", "during-wait", "deadline"} {
+		t.Run(when, func(t *testing.T) {
+			synctest.Test(t, func(t *testing.T) {
+				l := newLocker()
+				writer := listingLockerTestArgs("writer", "key", "writer")
+				listingLockerTestWrite(t, l, writer)
+				ctx, cancel := context.WithCancel(t.Context())
+				if when == "deadline" {
+					cancel()
+					ctx, cancel = context.WithTimeout(t.Context(), time.Millisecond)
+				}
+				defer cancel()
+				if when == "before-call" {
+					cancel()
+				}
+				var got bool
+				var err error
+				start := time.Now()
+				go func() { got, err = l.RLock(ctx, listingLockerTestArgs("listing", "key", listingLockerTestSource)) }()
+				synctest.Wait()
+				if when != "before-call" {
+					if state := listingLockerTestSnapshot(l); state.waiting != 1 {
+						t.Fatalf("waiting = %d, want 1", state.waiting)
+					}
+					time.Sleep(time.Millisecond)
+					// Removal must not turn the canceled request into a grant.
+					listingLockerTestUnlock(t, l, writer)
+					cancel()
+				} else {
+					listingLockerTestUnlock(t, l, writer)
+				}
+				synctest.Wait()
+				state := listingLockerTestSnapshot(l)
+				if got || err == nil || err != ctx.Err() || state != (listingLockerTestState{}) {
+					t.Fatalf("canceled RLock = %v, %v; context %v; state %+v", got, err, ctx.Err(), state)
+				}
+				if elapsed := time.Since(start); elapsed >= 2*time.Millisecond {
+					t.Fatalf("canceled request waited for full timer: %v", elapsed)
+				}
+			})
+		})
+	}
+}
+
+func TestLocalLockerListingReverseWaitExcluded(t *testing.T) {
+	for _, source := range []string{"ordinary-reader", "", "[<unknown>:0:()]", "listingRLock", "[metacache-server-pool.go:0:(*erasureServerPools).listingRLock()]"} {
+		t.Run(source, func(t *testing.T) {
+			synctest.Test(t, func(t *testing.T) {
+				l := newLocker()
+				listingLockerTestWrite(t, l, listingLockerTestArgs("writer", "key", "writer"))
+				start := time.Now()
+				if ok, err := l.RLock(t.Context(), listingLockerTestArgs("reader", "key", source)); ok || err != nil {
+					t.Fatalf("excluded RLock = %v, %v", ok, err)
+				}
+				if state := listingLockerTestSnapshot(l); time.Since(start) != 0 || state.waiting != 0 || state.uids != 1 {
+					t.Fatalf("excluded RLock changed timing or state: %+v", state)
+				}
+			})
+		})
+	}
+}
+
+func TestLocalLockerListingReverseWaitFastPaths(t *testing.T) {
+	synctest.Test(t, func(t *testing.T) {
+		l := newLocker()
+		start := time.Now()
+		for i, source := range []string{listingLockerTestSource, "ordinary-reader", ""} {
+			listingLockerTestRead(t, l, listingLockerTestArgs(fmt.Sprintf("shared-%d", i), "shared", source))
+			listingLockerTestRead(t, l, listingLockerTestArgs(fmt.Sprintf("free-%d", i), fmt.Sprintf("free-%d", i), source))
+		}
+		if state := listingLockerTestSnapshot(l); time.Since(start) != 0 || state.waiting != 0 || state.uids != 6 {
+			t.Fatalf("uncontended readers changed timing or state: %+v", state)
+		}
+	})
+}
+
+func TestLocalLockerListingReverseWaitOverloaded(t *testing.T) {
+	for _, when := range []string{"before-call", "during-wait"} {
+		t.Run(when, func(t *testing.T) {
+			synctest.Test(t, func(t *testing.T) {
+				l := newLocker()
+				writer := listingLockerTestArgs("writer", "key", "writer")
+				listingLockerTestWrite(t, l, writer)
+				defer l.waitMutex.Store(0)
+				if when == "before-call" {
+					l.waitMutex.Store(lockMutexWaitLimit + 1)
+				}
+				var got bool
+				var err error
+				go func() {
+					got, err = l.RLock(t.Context(), listingLockerTestArgs("listing", "key", listingLockerTestSource))
+				}()
+				synctest.Wait()
+				if when == "during-wait" {
+					if state := listingLockerTestSnapshot(l); state.waiting != 1 {
+						t.Fatalf("waiting = %d, want 1", state.waiting)
+					}
+					listingLockerTestUnlock(t, l, writer)
+					l.waitMutex.Store(lockMutexWaitLimit + 1)
+					time.Sleep(2 * time.Millisecond)
+					synctest.Wait()
+				} else {
+					listingLockerTestUnlock(t, l, writer)
+				}
+				if state := listingLockerTestSnapshot(l); got || err != nil || state != (listingLockerTestState{}) || l.locksOverloaded.Load() != 1 {
+					t.Fatalf("overloaded RLock = %v, %v; state %+v; overloads %d", got, err, state, l.locksOverloaded.Load())
+				}
+			})
+		})
+	}
+}
+
+func TestLocalLockerListingWaitSharedAdmission(t *testing.T) {
+	synctest.Test(t, func(t *testing.T) {
+		l := newLocker()
+		listingLockerTestRead(t, l, listingLockerTestArgs("reader", "read-key", listingLockerTestSource))
+		listingLockerTestWrite(t, l, listingLockerTestArgs("writer", "write-key", "writer"))
+		ctx, cancel := context.WithCancel(t.Context())
+		defer cancel()
+		results := make(chan error, listingLockWaitLimit)
+		for i := range listingLockWaitLimit {
+			go func() {
+				var ok bool
+				var err error
+				if i%2 == 0 {
+					ok, err = l.Lock(ctx, listingLockerTestArgs(fmt.Sprintf("writer-%d", i), "read-key", "writer"))
+				} else {
+					ok, err = l.RLock(ctx, listingLockerTestArgs(fmt.Sprintf("listing-%d", i), "write-key", listingLockerTestSource))
+				}
+				if ok {
+					err = fmt.Errorf("granted over a conflicting holder")
+				}
+				results <- err
+			}()
+		}
+		synctest.Wait()
+		if state := listingLockerTestSnapshot(l); state.waiting != listingLockWaitLimit {
+			t.Fatalf("combined waiting = %d, want %d", state.waiting, listingLockWaitLimit)
+		}
+		start := time.Now()
+		if ok, err := l.Lock(ctx, listingLockerTestArgs("excess-writer", "read-key", "writer")); ok || err != nil {
+			t.Fatalf("excess Lock = %v, %v", ok, err)
+		}
+		if ok, err := l.RLock(ctx, listingLockerTestArgs("excess-listing", "write-key", listingLockerTestSource)); ok || err != nil {
+			t.Fatalf("excess RLock = %v, %v", ok, err)
+		}
+		listingLockerTestWrite(t, l, listingLockerTestArgs("unrelated-writer", "unrelated-write", "writer"))
+		listingLockerTestRead(t, l, listingLockerTestArgs("unrelated-listing", "unrelated-read", listingLockerTestSource))
+		if state := listingLockerTestSnapshot(l); time.Since(start) != 0 || state.waiting != listingLockWaitLimit || state.uids != 4 {
+			t.Fatalf("combined cap changed immediate paths: %+v", state)
+		}
+		cancel()
+		synctest.Wait()
+		for range listingLockWaitLimit {
+			if err := <-results; err != context.Canceled {
+				t.Fatalf("canceled waiter: %v", err)
+			}
+		}
+		if state := listingLockerTestSnapshot(l); state.waiting != 0 || state.uids != 4 {
+			t.Fatalf("combined wait leaked: %+v", state)
+		}
+	})
+}
+
+type listingReverseCountLocker struct {
+	*localLocker
+	calls  atomic.Int32
+	grants atomic.Int32
+	reject bool
+}
+
+func (l *listingReverseCountLocker) RLock(ctx context.Context, args dsync.LockArgs) (bool, error) {
+	l.calls.Add(1)
+	if l.reject {
+		return false, nil
+	}
+	ok, err := l.localLocker.RLock(ctx, args)
+	if ok {
+		l.grants.Add(1)
+	}
+	return ok, err
+}
+
+func TestDistributedListingReverseWait(t *testing.T) {
+	for _, tc := range []struct {
+		name             string
+		free, held       int
+		immediateRejects int
+		cancel           bool
+		want             bool
+	}{
+		{name: "all-writers-release", want: true},
+		{name: "two-writers-remain", held: 2, want: true},
+		{name: "three-writers-remain", held: 3},
+		{name: "partial-grant-then-quorum", free: 1, want: true},
+		{name: "partial-grant-without-quorum", free: 1, held: 3},
+		{name: "late-grant-after-rejection", immediateRejects: 3},
+		{name: "partial-and-late-grants-after-cancel", free: 1, cancel: true},
+	} {
+		t.Run(tc.name, func(t *testing.T) {
+			synctest.Test(t, func(t *testing.T) {
+				clients := make([]dsync.NetLocker, 4)
+				writer := listingLockerTestArgs("writer", "key", "writer")
+				for i := range clients {
+					l := &listingReverseCountLocker{localLocker: newLocker(), reject: i < tc.immediateRejects}
+					clients[i] = l
+					if i >= tc.free {
+						listingLockerTestWrite(t, l.localLocker, writer)
+					}
+				}
+				ds := &dsync.Dsync{GetLockers: func() ([]dsync.NetLocker, string) { return clients, writer.Owner }, Timeouts: dsync.DefaultTimeouts}
+				dm := dsync.NewDRWMutex(ds, "key")
+				ctx, cancel := context.WithCancel(t.Context())
+				defer cancel()
+				result := make(chan bool, 1)
+				go func() {
+					result <- dm.GetRLock(ctx, nil, "listing", listingLockerTestSource, dsync.Options{Timeout: time.Second, RetryInterval: -1})
+				}()
+				synctest.Wait()
+				for i, c := range clients {
+					l := c.(*listingReverseCountLocker)
+					if i < tc.free && l.grants.Load() != 1 {
+						t.Fatal("uncontended partial grant did not arrive immediately")
+					}
+					if i >= tc.free && !l.reject && listingLockerTestSnapshot(l.localLocker).waiting != 1 {
+						t.Fatal("conflicting reader did not wait in its first RPC")
+					}
+				}
+				earlyResult := tc.immediateRejects == 3 || tc.cancel
+				if tc.cancel {
+					cancel()
+					synctest.Wait()
+				}
+				if earlyResult {
+					select {
+					case got := <-result:
+						if got {
+							t.Fatal("attempt succeeded before delayed replies")
+						}
+					default:
+						t.Fatal("attempt did not return before delayed replies")
+					}
+					for i := range tc.free {
+						if state := listingLockerTestSnapshot(clients[i].(*listingReverseCountLocker).localLocker); state.uids != 0 {
+							t.Fatal("early partial grant was not released")
+						}
+					}
+				}
+				time.Sleep(time.Millisecond)
+				for i, c := range clients {
+					if i >= tc.free+tc.held {
+						listingLockerTestUnlock(t, c.(*listingReverseCountLocker).localLocker, writer)
+					}
+				}
+				time.Sleep(time.Millisecond)
+				synctest.Wait()
+				if !earlyResult {
+					if got := <-result; got != tc.want {
+						t.Fatalf("GetRLock = %v, want %v", got, tc.want)
+					}
+				}
+				for i, c := range clients {
+					l := c.(*listingReverseCountLocker)
+					if l.calls.Load() != 1 || listingLockerTestSnapshot(l.localLocker).waiting != 0 {
+						t.Fatalf("peer %d calls = %d or leaked admission", i, l.calls.Load())
+					}
+					wantGrant := !l.reject && (i < tc.free || i >= tc.free+tc.held)
+					if (l.grants.Load() == 1) != wantGrant {
+						t.Fatalf("peer %d grants = %d, want granted = %v", i, l.grants.Load(), wantGrant)
+					}
+					if !tc.want {
+						for _, holder := range l.DupLockMap()["key"] {
+							if holder.UID == "listing" {
+								t.Fatalf("peer %d leaked partial or late read grant", i)
+							}
+						}
+					}
+				}
+				if tc.want {
+					dm.RUnlock(t.Context())
+					synctest.Wait()
+				}
+				for i, c := range clients {
+					l := c.(*listingReverseCountLocker)
+					if i >= tc.free && i < tc.free+tc.held {
+						listingLockerTestUnlock(t, l.localLocker, writer)
+					}
+					if state := listingLockerTestSnapshot(l.localLocker); state != (listingLockerTestState{}) {
+						t.Fatalf("peer %d final state is not empty: %+v", i, state)
+					}
+				}
+			})
+		})
+	}
+}

cmd/local-locker-listing_test.go

@@ -0,0 +1,432 @@
+// Copyright (c) 2026 Feng Ruohang
+//
+// This file is part of Silo Object Storage stack
+//
+// This program is free software: you can redistribute it and/or modify
+// it under the terms of the GNU Affero General Public License as published by
+// the Free Software Foundation, either version 3 of the License, or
+// (at your option) any later version.
+//
+// This program is distributed in the hope that it will be useful,
+// but WITHOUT ANY WARRANTY; without even the implied warranty of
+// MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
+// GNU Affero General Public License for more details.
+//
+// You should have received a copy of the GNU Affero General Public License
+// along with this program. If not, see <http://www.gnu.org/licenses/>.
+
+package cmd
+
+import (
+	"context"
+	"fmt"
+	"strings"
+	"sync/atomic"
+	"testing"
+	"testing/synctest"
+	"time"
+
+	"github.com/minio/minio/internal/dsync"
+)
+
+const listingLockerTestSource = "[metacache-server-pool.go:274:(*erasureServerPools).listingRLock()]"
+
+func listingLockerTestArgs(uid, resource, source string) dsync.LockArgs {
+	quorum := 3
+	return dsync.LockArgs{UID: uid, Resources: []string{resource}, Source: source, Owner: "listing-wait-test", Quorum: &quorum}
+}
+
+func listingLockerTestRead(t *testing.T, l *localLocker, args dsync.LockArgs) {
+	t.Helper()
+	if ok, err := l.RLock(t.Context(), args); !ok || err != nil {
+		t.Fatalf("RLock = %v, %v", ok, err)
+	}
+}
+
+func listingLockerTestRUnlock(t *testing.T, l *localLocker, args dsync.LockArgs) {
+	t.Helper()
+	if ok, err := l.RUnlock(t.Context(), args); !ok || err != nil {
+		t.Fatalf("RUnlock = %v, %v", ok, err)
+	}
+}
+
+type listingLockerTestState struct {
+	waiting, resources, uids int
+}
+
+func listingLockerTestSnapshot(l *localLocker) listingLockerTestState {
+	defer l.getMutex()()
+	return listingLockerTestState{waiting: l.listingWaiters, resources: len(l.lockMap), uids: len(l.lockUID)}
+}
+
+func TestLocalLockerListingSource(t *testing.T) {
+	for _, source := range []string{
+		"", "listingRLock", "[<unknown>:0:()]",
+		strings.Replace(listingLockerTestSource, ":274:", ":", 1),
+		strings.Replace(listingLockerTestSource, "274", "", 1),
+		strings.Replace(listingLockerTestSource, "274", "0", 1),
+		strings.Replace(listingLockerTestSource, "274", "1x", 1),
+		strings.Replace(listingLockerTestSource, "274", "-1", 1),
+		strings.Replace(listingLockerTestSource, "listingRLock", "GetObjectNInfo", 1),
+		strings.Replace(listingLockerTestSource, "listingRLock", "listingRLock.func1", 1),
+		strings.Replace(listingLockerTestSource, "metacache-server-pool.go", "other.go", 1),
+		listingLockerTestSource + "trailing",
+	} {
+		if isListingReadLock([]lockRequesterInfo{{Source: source}}) {
+			t.Errorf("unrecognized source accepted: %q", source)
+		}
+	}
+	if isListingReadLock(nil) || isListingReadLock([]lockRequesterInfo{{Source: listingLockerTestSource, Writer: true}}) {
+		t.Fatal("empty or write holders accepted")
+	}
+	if isListingReadLock([]lockRequesterInfo{{Source: listingLockerTestSource}, {Source: "ordinary-reader"}}) {
+		t.Fatal("mixed readers accepted")
+	}
+
+	// Check the real runtime-generated source, not only a hand-written label.
+	synctest.Test(t, func(t *testing.T) {
+		lockers := []dsync.NetLocker{newLocker(), newLocker(), newLocker(), newLocker()}
+		set := &erasureObjects{nsMutex: newNSLock(true), getLockers: func() ([]dsync.NetLocker, string) { return lockers, "listing-source-test" }}
+		z := &erasureServerPools{serverPools: []*erasureSets{{sets: []*erasureObjects{set}}, {sets: []*erasureObjects{set}}}}
+		_, unlock, err := z.listingRLock(t.Context(), "bucket", "object", set)
+		if err != nil {
+			t.Fatal(err)
+		}
+		for _, c := range lockers {
+			l := c.(*localLocker)
+			for _, holders := range l.DupLockMap() {
+				if !isListingReadLock(holders) {
+					t.Errorf("actual listingRLock source was not recognized: %#v", holders)
+				}
+			}
+		}
+		unlock()
+		synctest.Wait()
+	})
+}
+
+func FuzzListingLockSource(f *testing.F) {
+	f.Add(listingLockerTestSource)
+	f.Add("[metacache-server-pool.go:(*erasureServerPools).listingRLock()]")
+	f.Add("")
+	f.Fuzz(func(t *testing.T, source string) {
+		// Source is diagnostic text from a peer, including older versions. No
+		// malformed string may panic or make a writer eligible for this wait.
+		isListingReadLock([]lockRequesterInfo{{Source: source}})
+		if isListingReadLock([]lockRequesterInfo{{Source: source, Writer: true}}) {
+			t.Fatal("writer was accepted as a listing reader")
+		}
+	})
+}
+
+func TestLocalLockerListingWaitRechecks(t *testing.T) {
+	for _, tc := range []struct {
+		name   string
+		change func(*testing.T, *localLocker, dsync.LockArgs)
+		want   bool
+	}{
+		{"reader-released", func(t *testing.T, l *localLocker, r dsync.LockArgs) { listingLockerTestRUnlock(t, l, r) }, true},
+		{"reader-still-held", func(*testing.T, *localLocker, dsync.LockArgs) {}, false},
+		{"one-of-two-readers-remains", func(t *testing.T, l *localLocker, r dsync.LockArgs) {
+			other := r
+			other.UID = "second-reader"
+			listingLockerTestRead(t, l, other)
+			listingLockerTestRUnlock(t, l, r)
+		}, false},
+		{"ordinary-reader-arrives", func(t *testing.T, l *localLocker, r dsync.LockArgs) {
+			listingLockerTestRUnlock(t, l, r)
+			listingLockerTestRead(t, l, listingLockerTestArgs("new-reader", "key", "ordinary-reader"))
+		}, false},
+		{"other-writer-wins", func(t *testing.T, l *localLocker, r dsync.LockArgs) {
+			listingLockerTestRUnlock(t, l, r)
+			if ok, err := l.Lock(t.Context(), listingLockerTestArgs("other-writer", "key", "writer")); !ok || err != nil {
+				t.Fatalf("competing Lock = %v, %v", ok, err)
+			}
+		}, false},
+		{"force-unlock", func(t *testing.T, l *localLocker, r dsync.LockArgs) {
+			if ok, err := l.ForceUnlock(t.Context(), r); !ok || err != nil {
+				t.Fatalf("ForceUnlock = %v, %v", ok, err)
+			}
+		}, true},
+		{"expiry", func(_ *testing.T, l *localLocker, _ dsync.LockArgs) { l.expireOldLocks(-time.Minute) }, true},
+	} {
+		t.Run(tc.name, func(t *testing.T) {
+			synctest.Test(t, func(t *testing.T) {
+				l := newLocker()
+				reader := listingLockerTestArgs("reader", "key", listingLockerTestSource)
+				listingLockerTestRead(t, l, reader)
+				var got bool
+				var err error
+				start := time.Now()
+				go func() { got, err = l.Lock(t.Context(), listingLockerTestArgs("writer", "key", "writer")) }()
+				synctest.Wait()
+				if state := listingLockerTestSnapshot(l); state.waiting != 1 {
+					t.Fatalf("waiting = %d, want 1", state.waiting)
+				}
+				// The global mutex must be released throughout the timer wait.
+				if ok, err := l.Lock(t.Context(), listingLockerTestArgs("unrelated", "other-key", "writer")); !ok || err != nil {
+					t.Fatalf("unrelated key blocked: %v, %v", ok, err)
+				}
+				if elapsed := time.Since(start); elapsed != 0 {
+					t.Fatalf("unrelated key waited %v", elapsed)
+				}
+				time.Sleep(time.Millisecond)
+				tc.change(t, l, reader)
+				time.Sleep(time.Millisecond)
+				synctest.Wait()
+				if state := listingLockerTestSnapshot(l); got != tc.want || err != nil || state.waiting != 0 {
+					t.Fatalf("Lock = %v, %v, waiting = %d; want %v, nil, 0", got, err, state.waiting, tc.want)
+				}
+				if elapsed := time.Since(start); elapsed != 2*time.Millisecond {
+					t.Fatalf("one attempt elapsed %v, want 2ms", elapsed)
+				}
+			})
+		})
+	}
+}
+
+func TestLocalLockerListingWaitCancellation(t *testing.T) {
+	for _, when := range []string{"before-call", "during-wait", "deadline"} {
+		t.Run(when, func(t *testing.T) {
+			synctest.Test(t, func(t *testing.T) {
+				l := newLocker()
+				reader := listingLockerTestArgs("reader", "key", listingLockerTestSource)
+				listingLockerTestRead(t, l, reader)
+				ctx, cancel := context.WithCancel(t.Context())
+				if when == "deadline" {
+					cancel()
+					ctx, cancel = context.WithTimeout(t.Context(), time.Millisecond)
+				}
+				defer cancel()
+				if when == "before-call" {
+					cancel()
+				}
+				var got bool
+				var err error
+				go func() { got, err = l.Lock(ctx, listingLockerTestArgs("writer", "key", "writer")) }()
+				synctest.Wait()
+				if when != "before-call" {
+					if state := listingLockerTestSnapshot(l); state.waiting != 1 {
+						t.Fatalf("waiting = %d", state.waiting)
+					}
+					time.Sleep(time.Millisecond)
+					cancel()
+				}
+				synctest.Wait()
+				if state := listingLockerTestSnapshot(l); got || err != ctx.Err() || err == nil || state.waiting != 0 {
+					t.Fatalf("canceled Lock = %v, %v; context %v; waiting = %d", got, err, ctx.Err(), state.waiting)
+				}
+				listingLockerTestRUnlock(t, l, reader)
+				if state := listingLockerTestSnapshot(l); state.resources != 0 || state.uids != 0 {
+					t.Fatal("canceled wait left a lock")
+				}
+			})
+		})
+	}
+}
+
+func TestLocalLockerListingWaitExcluded(t *testing.T) {
+	for _, kind := range []string{"ordinary-reader", "unknown-reader", "mixed-readers", "writer", "bulk", "overloaded"} {
+		t.Run(kind, func(t *testing.T) {
+			synctest.Test(t, func(t *testing.T) {
+				l := newLocker()
+				reader := listingLockerTestArgs("holder", "key", listingLockerTestSource)
+				if kind == "ordinary-reader" {
+					reader.Source = "[erasure-object.go:100:(*erasureObjects).GetObjectNInfo()]"
+				}
+				if kind == "unknown-reader" {
+					reader.Source = ""
+				}
+				if kind == "writer" {
+					if ok, err := l.Lock(t.Context(), reader); !ok || err != nil {
+						t.Fatalf("initial write = %v, %v", ok, err)
+					}
+				} else {
+					listingLockerTestRead(t, l, reader)
+				}
+				if kind == "mixed-readers" {
+					listingLockerTestRead(t, l, listingLockerTestArgs("ordinary", "key", "ordinary-reader"))
+				}
+				writer := listingLockerTestArgs("writer", "key", "writer")
+				if kind == "bulk" {
+					writer.Resources = []string{"key", "other-key"}
+				}
+				if kind == "overloaded" {
+					l.waitMutex.Store(lockMutexWaitLimit + 1)
+				}
+				start := time.Now()
+				if ok, err := l.Lock(t.Context(), writer); ok || err != nil {
+					t.Fatalf("conflicting Lock = %v, %v", ok, err)
+				}
+				state := listingLockerTestSnapshot(l)
+				if elapsed := time.Since(start); elapsed != 0 || state.waiting != 0 {
+					t.Fatalf("excluded path waited %v; waiters = %d", elapsed, state.waiting)
+				}
+				if _, ok := l.DupLockMap()["other-key"]; ok {
+					t.Fatal("bulk request partially acquired an unrelated key")
+				}
+			})
+		})
+	}
+}
+
+func TestLocalLockerListingWaitAdmission(t *testing.T) {
+	synctest.Test(t, func(t *testing.T) {
+		l := newLocker()
+		reader := listingLockerTestArgs("reader", "key", listingLockerTestSource)
+		listingLockerTestRead(t, l, reader)
+		ctx, cancel := context.WithCancel(t.Context())
+		defer cancel()
+		results := make(chan error, listingLockWaitLimit)
+		for i := range listingLockWaitLimit {
+			go func() {
+				ok, err := l.Lock(ctx, listingLockerTestArgs(fmt.Sprintf("writer-%d", i), "key", "writer"))
+				if ok {
+					err = fmt.Errorf("granted over a live reader")
+				}
+				results <- err
+			}()
+		}
+		synctest.Wait()
+		if state := listingLockerTestSnapshot(l); state.waiting != listingLockWaitLimit {
+			t.Fatalf("waiting = %d, want %d", state.waiting, listingLockWaitLimit)
+		}
+		start := time.Now()
+		if ok, err := l.Lock(ctx, listingLockerTestArgs("excess", "key", "writer")); ok || err != nil {
+			t.Fatalf("excess waiter = %v, %v", ok, err)
+		}
+		if state := listingLockerTestSnapshot(l); time.Since(start) != 0 || state.waiting != listingLockWaitLimit {
+			t.Fatal("excess waiter consumed time or admission")
+		}
+		if ok, err := l.Lock(ctx, listingLockerTestArgs("unrelated", "other-key", "writer")); !ok || err != nil {
+			t.Fatalf("admission limit rejected an uncontended key: %v, %v", ok, err)
+		}
+		cancel()
+		synctest.Wait()
+		for range listingLockWaitLimit {
+			if err := <-results; err != context.Canceled {
+				t.Fatalf("canceled waiter: %v", err)
+			}
+		}
+		if state := listingLockerTestSnapshot(l); state.waiting != 0 {
+			t.Fatalf("admission leaked: %d", state.waiting)
+		}
+	})
+}
+
+func TestLocalLockerListingWaitContended(t *testing.T) {
+	for _, n := range []int{2, 8, 32} {
+		t.Run(fmt.Sprintf("%d-writers", n), func(t *testing.T) {
+			synctest.Test(t, func(t *testing.T) {
+				l := newLocker()
+				reader := listingLockerTestArgs("reader", "key", listingLockerTestSource)
+				listingLockerTestRead(t, l, reader)
+				results := make(chan bool, n)
+				for i := range n {
+					go func() {
+						ok, err := l.Lock(t.Context(), listingLockerTestArgs(fmt.Sprintf("writer-%d", i), "key", "writer"))
+						if err != nil {
+							t.Errorf("Lock: %v", err)
+						}
+						results <- ok
+					}()
+				}
+				synctest.Wait()
+				if state := listingLockerTestSnapshot(l); state.waiting != n {
+					t.Fatalf("waiting = %d, want %d", state.waiting, n)
+				}
+				time.Sleep(time.Millisecond)
+				listingLockerTestRUnlock(t, l, reader)
+				time.Sleep(time.Millisecond)
+				synctest.Wait()
+				granted := 0
+				for range n {
+					if <-results {
+						granted++
+					}
+				}
+				if state := listingLockerTestSnapshot(l); granted != 1 || state.uids != 1 || state.waiting != 0 {
+					t.Fatalf("grants = %d, indexed = %d, waiting = %d", granted, state.uids, state.waiting)
+				}
+			})
+		})
+	}
+}
+
+type listingCountLocker struct {
+	*localLocker
+	calls atomic.Int32
+}
+
+func (l *listingCountLocker) Lock(ctx context.Context, args dsync.LockArgs) (bool, error) {
+	l.calls.Add(1)
+	return l.localLocker.Lock(ctx, args)
+}
+
+// Exercise real DRWMutex quorum and late-response cleanup. The mixed case with
+// two immediate rejections fails before the upgraded lockers finish waiting.
+func TestDistributedListingWait(t *testing.T) {
+	for _, ordinaryReaders := range []int{0, 1, 2} {
+		t.Run(fmt.Sprintf("%d-immediate-rejections", ordinaryReaders), func(t *testing.T) {
+			synctest.Test(t, func(t *testing.T) {
+				clients := make([]dsync.NetLocker, 4)
+				reader := listingLockerTestArgs("reader", "key", listingLockerTestSource)
+				for i := range clients {
+					l := &listingCountLocker{localLocker: newLocker()}
+					clients[i] = l
+					r := reader
+					if i < ordinaryReaders {
+						r.Source = "old-node-unknown-source"
+					}
+					listingLockerTestRead(t, l.localLocker, r)
+				}
+				ds := &dsync.Dsync{GetLockers: func() ([]dsync.NetLocker, string) { return clients, reader.Owner }, Timeouts: dsync.DefaultTimeouts}
+				dm := dsync.NewDRWMutex(ds, "key")
+				var got bool
+				go func() {
+					got = dm.GetLock(t.Context(), nil, "writer", "writer", dsync.Options{Timeout: time.Second, RetryInterval: -1})
+				}()
+				synctest.Wait()
+				time.Sleep(time.Millisecond)
+				for _, c := range clients {
+					listingLockerTestRUnlock(t, c.(*listingCountLocker).localLocker, reader)
+				}
+				time.Sleep(time.Millisecond)
+				synctest.Wait()
+				if got != (ordinaryReaders < 2) {
+					t.Fatalf("GetLock = %v", got)
+				}
+				granted := 0
+				for _, c := range clients {
+					l := c.(*listingCountLocker)
+					if l.calls.Load() != 1 {
+						t.Fatalf("lock calls = %d, want one round", l.calls.Load())
+					}
+					state := listingLockerTestSnapshot(l.localLocker)
+					if state.resources != 0 {
+						granted++
+					}
+					if state.waiting != 0 {
+						t.Fatal("wait admission leaked")
+					}
+				}
+				if got && granted < 3 {
+					t.Fatalf("writer accepted without quorum: %d", granted)
+				}
+				if !got && granted != 0 {
+					t.Fatalf("late grants leaked after rejected attempt: %d", granted)
+				}
+				if got {
+					dm.Unlock(t.Context())
+					synctest.Wait()
+				}
+				for _, c := range clients {
+					l := c.(*listingCountLocker)
+					if state := listingLockerTestSnapshot(l.localLocker); state.resources != 0 || state.uids != 0 {
+						t.Fatal("final lock state not empty")
+					}
+				}
+			})
+		})
+	}
+}

cmd/local-locker.go

@@ -23,6 +23,7 @@ import (
 	"context"
 	"fmt"
 	"strconv"
+	"strings"
 	"sync"
 	"sync/atomic"
 	"time"
@@ -38,6 +39,14 @@ import (
 // Rejected clients are expected to retry.
 const lockMutexWaitLimit = 1000
 
+// A short listing recovery read or conflicting writer may finish before the
+// rejected request retries. Recheck once in the same request, sharing a bound
+// for timers outside the mutex. This is not a queue and gives neither priority.
+const (
+	listingLockWait      = 2 * time.Millisecond
+	listingLockWaitLimit = 64
+)
+
 // lockRequesterInfo stores various info from the client for each lock that is requested.
 type lockRequesterInfo struct {
 	Name            string // name of the resource lock was requested for
@@ -65,6 +74,8 @@ type localLocker struct {
 	waitMutex atomic.Int32
 	lockMap   map[string][]lockRequesterInfo
 	lockUID   map[string]string // UUID -> resource map.
+	// Protected by mutex. These waiters are not included in waitMutex.
+	listingWaiters int
 
 	// the following are updated on every cleanup defined in lockValidityDuration
 	readers         atomic.Int32
@@ -96,6 +107,32 @@ func (l *localLocker) canTakeLock(resources ...string) bool {
 	return true
 }
 
+// isListingReadLock recognizes only the existing diagnostic source emitted by
+// listingRLock. Unknown or changed source formats retain immediate rejection;
+// Source is not a new wire contract or proof that the read will finish quickly.
+func isListingReadLock(holders []lockRequesterInfo) bool {
+	if len(holders) == 0 {
+		return false
+	}
+	const prefix = "[metacache-server-pool.go:"
+	const suffix = ":(*erasureServerPools).listingRLock()]"
+	for _, holder := range holders {
+		if holder.Writer || len(holder.Source) <= len(prefix)+len(suffix) || !strings.HasPrefix(holder.Source, prefix) || !strings.HasSuffix(holder.Source, suffix) {
+			return false
+		}
+		line := holder.Source[len(prefix) : len(holder.Source)-len(suffix)]
+		if len(line) == 0 || line[0] == '0' {
+			return false
+		}
+		for _, c := range line {
+			if c < '0' || c > '9' {
+				return false
+			}
+		}
+	}
+	return true
+}
+
 func (l *localLocker) Lock(ctx context.Context, args dsync.LockArgs) (reply bool, err error) {
 	if len(args.Resources) > maxDeleteList {
 		return false, fmt.Errorf("internal error: localLocker.Lock called with more than %d resources", maxDeleteList)
@@ -112,9 +149,33 @@ func (l *localLocker) Lock(ctx context.Context, args dsync.LockArgs) (reply bool
 		return false, ctx.Err()
 	}
 	if !l.canTakeLock(args.Resources...) {
-		// Not all locks can be taken on resources,
+		// Only a single-resource write blocked exclusively by listing recovery
-		// reject it completely.
+		// readers gets a short, bounded opportunity to succeed in this attempt.
-		return false, nil
+		if len(args.Resources) != 1 || l.listingWaiters >= listingLockWaitLimit || !isListingReadLock(l.lockMap[args.Resources[0]]) {
+			return false, nil
+		}
+		l.listingWaiters++
+		l.mutex.Unlock()
+		timer := time.NewTimer(listingLockWait)
+		select {
+		case <-ctx.Done():
+		case <-timer.C:
+		}
+		timer.Stop()
+		// Restore the mutex before returning to the existing deferred unlock.
+		// Readers or another writer may have entered while it was released.
+		l.getMutex()
+		l.listingWaiters--
+		if ctx.Err() != nil {
+			return false, ctx.Err()
+		}
+		if l.waitMutex.Load() > lockMutexWaitLimit {
+			l.locksOverloaded.Add(1)
+			return false, nil
+		}
+		if !l.canTakeLock(args.Resources...) {
+			return false, nil
+		}
 	}
 
 	// No locks held on the all resources, so claim write
@@ -206,6 +267,31 @@ func (l *localLocker) RLock(ctx context.Context, args dsync.LockArgs) (reply boo
 		return false, ctx.Err()
 	}
 	resource := args.Resources[0]
+	lri, ok := l.lockMap[resource]
+	if ok && isWriteLock(lri) && l.listingWaiters < listingLockWaitLimit && isListingReadLock([]lockRequesterInfo{{Source: args.Source}}) {
+		// Only a positively identified listing read blocked by a writer may
+		// wait once. A long or replacement writer still rejects normally.
+		l.listingWaiters++
+		l.mutex.Unlock()
+		timer := time.NewTimer(listingLockWait)
+		select {
+		case <-ctx.Done():
+		case <-timer.C:
+		}
+		timer.Stop()
+		// Restore the mutex before the existing deferred unlock, then use
+		// the current holders in the normal read-lock grant path below.
+		l.getMutex()
+		l.listingWaiters--
+		if ctx.Err() != nil {
+			return false, ctx.Err()
+		}
+		if l.waitMutex.Load() > lockMutexWaitLimit {
+			l.locksOverloaded.Add(1)
+			return false, nil
+		}
+		lri, ok = l.lockMap[resource]
+	}
 	now := UTCNow()
 	lrInfo := lockRequesterInfo{
 		Name:            resource,
@@ -217,7 +303,6 @@ func (l *localLocker) RLock(ctx context.Context, args dsync.LockArgs) (reply boo
 		TimeLastRefresh: now.UnixNano(),
 		Quorum:          *args.Quorum,
 	}
-	lri, ok := l.lockMap[resource]
 	if ok {
 		if reply = !isWriteLock(lri); reply {
 			// Unless there is a write lock

cmd/metacache-entries.go

@@ -37,6 +37,10 @@ type metaCacheEntry struct {
 	// Entries without metadata will only be present in non-recursive scans.
 	metadata []byte
 
+	// listingSet marks an eligible unresolved walk entry. It is consumed by the
+	// pool merger before any cache serialization or downstream filtering.
+	listingSet *erasureObjects
+
 	// cached contains the metadata if decoded.
 	cached *xlMetaV2
 
@@ -46,17 +50,17 @@ type metaCacheEntry struct {
 
 // isDir returns if the entry is representing a prefix directory.
 func (e metaCacheEntry) isDir() bool {
-	return len(e.metadata) == 0 && strings.HasSuffix(e.name, slashSeparator)
+	return e.listingSet == nil && len(e.metadata) == 0 && strings.HasSuffix(e.name, slashSeparator)
 }
 
 // isObject returns if the entry is representing an object.
 func (e metaCacheEntry) isObject() bool {
-	return len(e.metadata) > 0
+	return e.listingSet != nil || len(e.metadata) > 0
 }
 
 // isObjectDir returns if the entry is representing an object/
 func (e metaCacheEntry) isObjectDir() bool {
-	return len(e.metadata) > 0 && strings.HasSuffix(e.name, slashSeparator)
+	return (e.listingSet != nil || len(e.metadata) > 0) && strings.HasSuffix(e.name, slashSeparator)
 }
 
 // hasPrefix returns whether an entry has a specific prefix
@@ -670,26 +674,82 @@ func (m *metaCacheEntriesSorted) forwardPast(s string) {
 // If the context is canceled the function will return the error,
 // otherwise the function will return nil.
 func mergeEntryChannels(ctx context.Context, in []chan metaCacheEntry, out chan<- metaCacheEntry, readQuorum int) error {
+	return mergeEntryChannelsWithErrors(ctx, in, out, readQuorum, nil)
+}
+
+// terminalErr is inspected after each input closes and before the merger
+// advances beyond it. Callers must publish the error before closing the input.
+func mergeEntryChannelsWithErrors(ctx context.Context, in []chan metaCacheEntry, out chan<- metaCacheEntry, readQuorum int, terminalErr func(int) error) error {
+	return mergeEntryChannelsWithResolver(ctx, in, out, readQuorum, terminalErr, nil)
+}
+
+// resolve is called only for groups containing a pending listing entry. All
+// selected inputs are advanced (and their terminal errors checked) before it is
+// called, so recovery never holds an object lock while waiting for input.
+func mergeEntryChannelsWithResolver(ctx context.Context, in []chan metaCacheEntry, out chan<- metaCacheEntry, readQuorum int, terminalErr func(int) error, resolve func([]*metaCacheEntry) (*metaCacheEntry, error)) error {
 	defer xioutil.SafeClose(out)
 	top := make([]*metaCacheEntry, len(in))
 	nDone := 0
 	ctxDone := ctx.Done()
 
-	// Use simpler forwarder.
+	// Keep the single-input normal path allocation-free. Only a pending entry
+	// requires looking ahead to observe its stream's terminal error first.
 	if len(in) == 1 {
+		var buffered metaCacheEntry
+		haveBuffered, exhausted := false, false
 		for {
-			select {
+			var v metaCacheEntry
-			case <-ctxDone:
+			if haveBuffered {
-				return ctx.Err()
+				v, haveBuffered = buffered, false
-			case v, ok := <-in[0]:
+			} else {
-				if !ok {
+				select {
-					return nil
+				case <-ctxDone:
+					return ctx.Err()
+				case entry, ok := <-in[0]:
+					if !ok {
+						if terminalErr != nil {
+							return terminalErr(0)
+						}
+						return nil
+					}
+					v = entry
+				}
+			}
+			if v.listingSet != nil {
+				if resolve == nil {
+					return errInvalidArgument
 				}
 				select {
 				case <-ctxDone:
 					return ctx.Err()
-				case out <- v:
+				case entry, ok := <-in[0]:
+					buffered, haveBuffered, exhausted = entry, ok, !ok
+					if !ok && terminalErr != nil {
+						if err := terminalErr(0); err != nil {
+							return err
+						}
+					}
 				}
+				pending := v
+				resolved, err := resolve([]*metaCacheEntry{&pending})
+				if err != nil {
+					return err
+				}
+				if resolved == nil {
+					if exhausted {
+						return nil
+					}
+					continue
+				}
+				v = *resolved
+			}
+			select {
+			case <-ctxDone:
+				return ctx.Err()
+			case out <- v:
+			}
+			if exhausted {
+				return nil
 			}
 		}
 	}
@@ -700,6 +760,11 @@ func mergeEntryChannels(ctx context.Context, in []chan metaCacheEntry, out chan<
 			return ctx.Err()
 		case entry, ok := <-in[idx]:
 			if !ok {
+				if terminalErr != nil {
+					if err := terminalErr(idx); err != nil {
+						return err
+					}
+				}
 				top[idx] = nil
 				nDone++
 			} else {
@@ -716,6 +781,15 @@ func mergeEntryChannels(ctx context.Context, in []chan metaCacheEntry, out chan<
 	}
 	last := ""
 	var toMerge []int
+	var directoryCollisions []int
+	discardDirectories := func() error {
+		for _, idx := range directoryCollisions {
+			if err := selectFrom(idx); err != nil {
+				return err
+			}
+		}
+		return nil
+	}
 
 	// Choose the best to return.
 	for {
@@ -725,6 +799,7 @@ func mergeEntryChannels(ctx context.Context, in []chan metaCacheEntry, out chan<
 		best := top[0]
 		bestIdx := 0
 		toMerge = toMerge[:0]
+		directoryCollisions = directoryCollisions[:0]
 		for i, other := range top[1:] {
 			otherIdx := i + 1
 			if other == nil {
@@ -754,10 +829,9 @@ func mergeEntryChannels(ctx context.Context, in []chan metaCacheEntry, out chan<
 						if serverDebugLog {
 							console.Debugln("mergeEntryChannels: discarding directory", other.name, "for object", best.name)
 						}
-						// Discard the directory.
+						// A pending object can resolve to absence. Keep the
-						if err := selectFrom(otherIdx); err != nil {
+						// directory until the whole object group is confirmed.
-							return err
+						directoryCollisions = append(directoryCollisions, otherIdx)
-						}
 						continue
 					}
 					// Replace directory with object.
@@ -773,11 +847,54 @@ func mergeEntryChannels(ctx context.Context, in []chan metaCacheEntry, out chan<
 			}
 			if best.name > other.name {
 				toMerge = toMerge[:0]
+				directoryCollisions = directoryCollisions[:0]
 				best = other
 				bestIdx = otherIdx
 			}
 		}
 
+		pending := best.listingSet != nil
+		for _, idx := range toMerge {
+			pending = pending || top[idx].listingSet != nil
+		}
+		if pending {
+			if resolve == nil {
+				return errInvalidArgument
+			}
+			group := make([]*metaCacheEntry, 1, len(toMerge)+1)
+			group[0] = best
+			for _, idx := range toMerge {
+				group = append(group, top[idx])
+				if err := selectFrom(idx); err != nil {
+					return err
+				}
+			}
+			if err := selectFrom(bestIdx); err != nil {
+				return err
+			}
+			resolved, err := resolve(group)
+			if err != nil {
+				return err
+			}
+			if resolved != nil {
+				if err := discardDirectories(); err != nil {
+					return err
+				}
+			}
+			if resolved != nil && resolved.name > last {
+				select {
+				case <-ctxDone:
+					return ctx.Err()
+				case out <- *resolved:
+					last = resolved.name
+				}
+			}
+			continue
+		}
+		if err := discardDirectories(); err != nil {
+			return err
+		}
+
 		// Merge any unmerged
 		if len(toMerge) > 0 {
 			versions := make([][]xlMetaV2ShallowVersion, 0, len(toMerge)+1)

cmd/metacache-listing-cache_test.go

@@ -0,0 +1,403 @@
+// Copyright (c) 2026 Feng Ruohang
+//
+// This file is part of Silo Object Storage stack
+//
+// This program is free software: you can redistribute it and/or modify
+// it under the terms of the GNU Affero General Public License as published by
+// the Free Software Foundation, either version 3 of the License, or
+// (at your option) any later version.
+//
+// This program is distributed in the hope that it will be useful
+// but WITHOUT ANY WARRANTY; without even the implied warranty of
+// MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+// GNU Affero General Public License for more details.
+//
+// You should have received a copy of the GNU Affero General Public License
+// along with this program.  If not, see <http://www.gnu.org/licenses/>.
+
+package cmd
+
+import (
+	"bytes"
+	"context"
+	"encoding/xml"
+	"errors"
+	"net/http"
+	"net/http/httptest"
+	"net/url"
+	"slices"
+	"sync"
+	"sync/atomic"
+	"testing"
+	"time"
+
+	"github.com/minio/minio/internal/grid"
+)
+
+type listingReadFailDisk struct {
+	StorageAPI
+	object string
+	reads  *atomic.Int64
+}
+
+func (d *listingReadFailDisk) ReadXL(ctx context.Context, bucket, object string, readData bool) (RawFileInfo, error) {
+	if object == d.object {
+		d.reads.Add(1)
+		return RawFileInfo{}, errDiskNotFound
+	}
+	return d.StorageAPI.ReadXL(ctx, bucket, object, readData)
+}
+
+// Install before issuing requests: cache writers can outlive their HTTP
+// response, so disk faults must be switched without replacing getDisks.
+// Each published slice must remain immutable.
+func installListingDiskSwitch(set *erasureObjects) ([]StorageAPI, func([]StorageAPI)) {
+	disks := set.getDisks()
+	var current atomic.Value
+	current.Store(disks)
+	set.getDisks = func() []StorageAPI { return current.Load().([]StorageAPI) }
+	return disks, func(disks []StorageAPI) { current.Store(disks) }
+}
+
+func TestListingCachePreservesFallbackError(t *testing.T) {
+	for _, continuation := range []bool{false, true} {
+		name := "cold"
+		if continuation {
+			name = "continuation"
+		}
+		t.Run(name, func(t *testing.T) {
+			testListingCacheFallbackError(t, continuation, false)
+		})
+	}
+}
+
+func TestListingCacheCleanupKeepsOriginalID(t *testing.T) {
+	testListingCacheFallbackError(t, true, true)
+}
+
+// Use actual peer RPCs to verify the abandoned cache is updated after the
+// caller clears its own listing ID. The local-only path cannot observe this.
+func recordListingPeerCleanup(t *testing.T) func() {
+	t.Helper()
+	tg, err := grid.SetupTestGrid(2)
+	if err != nil {
+		t.Fatal(err)
+	}
+	previous := globalNotificationSys
+	t.Cleanup(func() {
+		tg.Cleanup()
+		globalNotificationSys = previous
+	})
+	getIDs := make(chan string, 8)
+	type cleanupResult struct {
+		id  string
+		err *grid.RemoteErr
+	}
+	updated := make(chan cleanupResult, 1)
+	server := &peerRESTServer{}
+	if err := getMetacacheListingRPC.Register(tg.Managers[1], func(o *listPathOptions) (*metacache, *grid.RemoteErr) {
+		getIDs <- o.ID
+		return server.GetMetacacheListingHandler(o)
+	}); err != nil {
+		t.Fatal(err)
+	}
+	if err := updateMetacacheListingRPC.Register(tg.Managers[1], func(m *metacache) (*metacache, *grid.RemoteErr) {
+		result, err := server.UpdateMetacacheListingHandler(m)
+		if m.error == "no longer used" {
+			updated <- cleanupResult{id: m.id, err: err}
+		}
+		return result, err
+	}); err != nil {
+		t.Fatal(err)
+	}
+	client := &peerRESTClient{gridConn: func() *grid.Connection { return tg.Managers[0].Connection(tg.Hosts[1]) }}
+	globalNotificationSys = &NotificationSys{peerClients: []*peerRESTClient{client}, allPeerClients: []*peerRESTClient{client}}
+	return func() {
+		ctx, cancel := context.WithTimeout(t.Context(), 5*time.Second)
+		defer cancel()
+		var ids []string
+		for range 2 {
+			select {
+			case id := <-getIDs:
+				ids = append(ids, id)
+			case <-ctx.Done():
+				t.Fatal("peer did not receive the failed cache cleanup lookup")
+			}
+		}
+		if ids[0] == "" || ids[1] != ids[0] {
+			t.Fatalf("cleanup lookup lost the original cache ID: %q", ids)
+		}
+		select {
+		case result := <-updated:
+			if result.err != nil || result.id != ids[0] {
+				t.Fatalf("cleanup update: id=%q err=%v, want id=%q", result.id, result.err, ids[0])
+			}
+			t.Logf("peer cleanup Get/Update retained cache ID %q", result.id)
+		case <-ctx.Done():
+			t.Fatal("peer did not finish the failed cache cleanup update")
+		}
+	}
+}
+
+func testListingCacheFallbackError(t *testing.T, continuation, peer bool) {
+	t.Helper()
+	z, bucket, router := nullQuorumBackend(t)
+	var waitPeerCleanup func()
+	if peer {
+		waitPeerCleanup = recordListingPeerCleanup(t)
+	}
+	set := z.serverPools[0].sets[0]
+	disks, switchDisks := installListingDiskSwitch(set)
+	body := bytes.Repeat([]byte{'a'}, 8192)
+	mtime := time.Now().UTC().Add(-time.Hour)
+	for _, object := range []string{"a", "b", "z"} {
+		nullQuorumPut(t, z, bucket, object, body, ObjectOptions{MTime: mtime})
+	}
+	old := make([][]byte, 3)
+	for i := range old {
+		old[i] = mustReadNullQuorumMeta(t, disks[i], bucket, "z")
+	}
+	nullQuorumPut(t, z, bucket, "z", body, ObjectOptions{MTime: mtime.Add(time.Minute)})
+	var token string
+	if continuation {
+		rec := nullQuorumRequest(t, router, http.MethodGet, getListObjectsV2URL("", bucket, "", "1", "", "", ""))
+		var list ListObjectsV2Response
+		if rec.Code != http.StatusOK || xml.Unmarshal(rec.Body.Bytes(), &list) != nil || !list.IsTruncated || list.NextContinuationToken == "" {
+			t.Fatalf("first page: %d %s", rec.Code, rec.Body.String())
+		}
+		token = list.NextContinuationToken
+	}
+	for i := range old {
+		if err := disks[i].WriteAll(t.Context(), bucket, "z/"+xlStorageFormatFile, old[i]); err != nil {
+			t.Fatal(err)
+		}
+	}
+	var reads atomic.Int64
+	walkers := slices.Clone(disks)
+	for i := range walkers {
+		walkers[i] = &listingReadFailDisk{StorageAPI: disks[i], object: "z", reads: &reads}
+		if i >= 8 {
+			walkers[i] = &failedMetadataWalkDisk{StorageAPI: walkers[i]}
+		}
+	}
+	switchDisks(walkers)
+	target := getListObjectsV2URL("", bucket, "", "1000", "", "", "")
+	if token != "" {
+		target += "&continuation-token=" + url.QueryEscape(token)
+	}
+	rec := nullQuorumRequest(t, router, http.MethodGet, target)
+	t.Logf("LIST=%d failed metadata reads=%d body=%s", rec.Code, reads.Load(), rec.Body.String())
+	if reads.Load() == 0 {
+		t.Fatal("fallback not exercised")
+	}
+	var apiErr APIErrorResponse
+	if rec.Code != http.StatusServiceUnavailable || xml.Unmarshal(rec.Body.Bytes(), &apiErr) != nil || apiErr.Code != "SlowDownRead" {
+		t.Fatalf("fallback failure lost its read-quorum error: %d %s", rec.Code, rec.Body.String())
+	}
+	if waitPeerCleanup != nil {
+		waitPeerCleanup()
+	}
+	if continuation {
+		caches, _ := localMetacacheMgr.getBucket(t.Context(), bucket).cloneCaches()
+		if len(caches) == 0 {
+			t.Fatal("continuation did not create a cache")
+		}
+		for _, cache := range caches {
+			if cache.status != scanStateError {
+				t.Fatalf("incomplete cache was not marked failed: %+v", cache)
+			}
+		}
+	}
+	switchDisks(disks)
+	retry := nullQuorumRequest(t, router, http.MethodGet, target)
+	var recovered ListObjectsV2Response
+	if retry.Code != http.StatusOK || xml.Unmarshal(retry.Body.Bytes(), &recovered) != nil {
+		t.Fatalf("retry after metadata recovers: %d %s", retry.Code, retry.Body.String())
+	}
+	var keys []string
+	for _, entry := range recovered.Contents {
+		keys = append(keys, entry.Key)
+	}
+	want := []string{"a", "b", "z"}
+	if continuation {
+		want = []string{"b", "z"}
+	}
+	if !slices.Equal(keys, want) || recovered.IsTruncated {
+		t.Fatalf("retry returned incomplete result: %+v", recovered)
+	}
+}
+
+func TestListingCachePagination(t *testing.T) {
+	z, bucket, router := nullQuorumBackend(t)
+	for _, key := range []string{"a", "b", "c", "d"} {
+		nullQuorumPut(t, z, bucket, key, []byte(key), ObjectOptions{})
+	}
+	var keys []string
+	var token string
+	var finished bool
+	for page := 0; page < 8; page++ {
+		target := getListObjectsV2URL("", bucket, "", "1", "", "", "")
+		if token != "" {
+			target += "&continuation-token=" + url.QueryEscape(token)
+		}
+		rec := nullQuorumRequest(t, router, http.MethodGet, target)
+		var list ListObjectsV2Response
+		if rec.Code != http.StatusOK || xml.Unmarshal(rec.Body.Bytes(), &list) != nil {
+			t.Fatalf("page %d: %d %s", page, rec.Code, rec.Body.String())
+		}
+		for _, entry := range list.Contents {
+			keys = append(keys, entry.Key)
+		}
+		if !list.IsTruncated {
+			finished = true
+			break
+		}
+		if list.NextContinuationToken == "" || list.NextContinuationToken == token {
+			t.Fatal("pagination did not advance")
+		}
+		token = list.NextContinuationToken
+	}
+	if !finished || !slices.Equal(keys, []string{"a", "b", "c", "d"}) {
+		t.Fatalf("pagination returned %v, finished=%v", keys, finished)
+	}
+}
+
+// A cached full page must remain streamable even if a later key is slow. The
+// completion barrier belongs at EOF, not before every page response.
+func TestListingCacheFullPageDoesNotWaitForTail(t *testing.T) {
+	z, bucket, router := nullQuorumBackend(t)
+	disks, switchDisks := installListingDiskSwitch(z.serverPools[0].sets[0])
+	for _, key := range []string{"a", "b", "c", "d", "e", "zz"} {
+		nullQuorumPut(t, z, bucket, key, []byte(key), ObjectOptions{})
+	}
+	first := nullQuorumRequest(t, router, http.MethodGet, getListObjectsV2URL("", bucket, "", "1", "", "", ""))
+	var page ListObjectsV2Response
+	if first.Code != http.StatusOK || xml.Unmarshal(first.Body.Bytes(), &page) != nil || !page.IsTruncated {
+		t.Fatalf("first page: %d %s", first.Code, first.Body.String())
+	}
+	resume := make(chan struct{})
+	release := sync.OnceFunc(func() { close(resume) })
+	defer release()
+	var blocked, active atomic.Int64
+	walkers := slices.Clone(disks)
+	for i, disk := range walkers {
+		walkers[i] = &listingPausedTailDisk{StorageAPI: disk, bucket: bucket, tail: "zz", resume: resume, blocked: &blocked, active: &active}
+	}
+	switchDisks(walkers)
+	ctx, cancel := context.WithTimeout(t.Context(), 10*time.Second)
+	defer cancel()
+	target := getListObjectsV2URL("", bucket, "", "1", "", "", "") + "&continuation-token=" + url.QueryEscape(page.NextContinuationToken)
+	req, err := newTestSignedRequestV4(http.MethodGet, target, 0, nil, globalActiveCred.AccessKey, globalActiveCred.SecretKey, nil)
+	if err != nil {
+		t.Fatal(err)
+	}
+	response := make(chan *httptest.ResponseRecorder, 1)
+	go func() {
+		rec := httptest.NewRecorder()
+		router.ServeHTTP(rec, req.WithContext(ctx))
+		response <- rec
+	}()
+	select {
+	case rec := <-response:
+		page = ListObjectsV2Response{}
+		if rec.Code != http.StatusOK || xml.Unmarshal(rec.Body.Bytes(), &page) != nil || !page.IsTruncated || len(page.Contents) != 1 || page.Contents[0].Key != "b" || page.NextContinuationToken == "" {
+			t.Fatalf("full page before tail: %d %s", rec.Code, rec.Body.String())
+		}
+		if blocked.Load() == 0 || active.Load() == 0 {
+			t.Fatal("full page did not overlap the paused tail")
+		}
+	case <-ctx.Done():
+		t.Fatal("full page waited for the paused tail")
+	}
+	release()
+	waitForListingCondition(t, "disk walkers did not finish", func() bool { return active.Load() == 0 })
+	target = getListObjectsV2URL("", bucket, "", "1000", "", "", "") + "&continuation-token=" + url.QueryEscape(page.NextContinuationToken)
+	rec := nullQuorumRequest(t, router, http.MethodGet, target)
+	page = ListObjectsV2Response{}
+	if rec.Code != http.StatusOK || xml.Unmarshal(rec.Body.Bytes(), &page) != nil || page.IsTruncated {
+		t.Fatalf("page after tail resumes: %d %s", rec.Code, rec.Body.String())
+	}
+	var keys []string
+	for _, entry := range page.Contents {
+		keys = append(keys, entry.Key)
+	}
+	if !slices.Equal(keys, []string{"c", "d", "e", "zz"}) {
+		t.Fatalf("cached remainder lost keys: %v", keys)
+	}
+}
+
+// Keep the error update pending at the actual peer RPC boundary. Other cache
+// operations, including readers and abandoned-cache cleanup, remain live.
+func listingCacheErrorPeer(t *testing.T, fail bool) (*peerRESTClient, <-chan struct{}, func()) {
+	t.Helper()
+	tg, err := grid.SetupTestGrid(2)
+	if err != nil {
+		t.Fatal(err)
+	}
+	ready, resume := make(chan struct{}), make(chan struct{})
+	release := sync.OnceFunc(func() { close(resume) })
+	t.Cleanup(func() {
+		release()
+		tg.Cleanup()
+	})
+	var entered sync.Once
+	server := &peerRESTServer{}
+	if err := getMetacacheListingRPC.Register(tg.Managers[1], server.GetMetacacheListingHandler); err != nil {
+		t.Fatal(err)
+	}
+	if err := updateMetacacheListingRPC.Register(tg.Managers[1], func(m *metacache) (*metacache, *grid.RemoteErr) {
+		if m.error != "" && m.error != "no longer used" {
+			entered.Do(func() { close(ready) })
+			select {
+			case <-resume:
+			case <-t.Context().Done():
+				return nil, grid.NewRemoteErr(t.Context().Err())
+			}
+			if fail {
+				return nil, grid.NewRemoteErr(errors.New("cache error update failed"))
+			}
+		}
+		return server.UpdateMetacacheListingHandler(m)
+	}); err != nil {
+		t.Fatal(err)
+	}
+	return &peerRESTClient{gridConn: func() *grid.Connection { return tg.Managers[0].Connection(tg.Hosts[1]) }}, ready, release
+}
+
+func TestListingCacheRetainsErrorAfterPeerFailure(t *testing.T) {
+	client, ready, release := listingCacheErrorPeer(t, true)
+	o := listPathOptions{ID: mustGetUUID(), Bucket: "bucket"}
+	m := o.newMetacache()
+	mc := metaCacheRPC{meta: &m, rpc: client, o: o}
+	done := make(chan struct{})
+	go func() {
+		defer close(done)
+		mc.setErr("terminal listing failure")
+	}()
+	t.Cleanup(func() {
+		release()
+		select {
+		case <-done:
+		case <-time.After(5 * time.Second):
+			t.Error("setErr worker did not finish before grid cleanup")
+		}
+	})
+	select {
+	case <-ready:
+	case <-time.After(5 * time.Second):
+		t.Fatal("error update did not reach peer")
+	}
+	release()
+	select {
+	case <-done:
+	case <-time.After(5 * time.Second):
+		t.Fatal("failed peer update did not return")
+	}
+	if m.id != o.ID || m.bucket != o.Bucket || m.status != scanStateError || m.error != "terminal listing failure" || mc.getErr() == nil {
+		t.Fatalf("peer failure erased local failure: %+v", m)
+	}
+	mc.setErr("later failure")
+	if m.error != "terminal listing failure" {
+		t.Fatalf("later failure replaced the terminal error: %+v", m)
+	}
+}

cmd/metacache-listing-delete_test.go

@@ -0,0 +1,133 @@
+// Copyright (c) 2026 Feng Ruohang
+//
+// This file is part of Silo Object Storage stack
+//
+// This program is free software: you can redistribute it and/or modify
+// it under the terms of the GNU Affero General Public License as published by
+// the Free Software Foundation, either version 3 of the License, or
+// (at your option) any later version.
+//
+// This program is distributed in the hope that it will be useful
+// but WITHOUT ANY WARRANTY; without even the implied warranty of
+// MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+// GNU Affero General Public License for more details.
+//
+// You should have received a copy of the GNU Affero General Public License
+// along with this program.  If not, see <http://www.gnu.org/licenses/>.
+
+package cmd
+
+import (
+	"bytes"
+	"encoding/xml"
+	"fmt"
+	"net/http"
+	"slices"
+	"testing"
+	"time"
+)
+
+func TestListFallbackPreservesPoolDeleteMarker(t *testing.T) {
+	for _, markerPool := range []int{0, 1} {
+		t.Run(fmt.Sprint(markerPool), func(t *testing.T) {
+			z, bucket, router := listingPoolsBackend(t)
+			ctx := t.Context()
+			live, deleted := z.serverPools[1-markerPool].sets[0], z.serverPools[markerPool].sets[0]
+			disks, switchDisks := installListingDiskSwitch(deleted)
+			const object = "gone"
+			body := bytes.Repeat([]byte{'a'}, 8192)
+			oldTime := time.Now().UTC().Add(-time.Hour)
+			for _, set := range []*erasureObjects{live, deleted} {
+				_, err := set.PutObject(ctx, bucket, object, mustGetPutObjReader(t, bytes.NewReader(body), int64(len(body)), "", ""), ObjectOptions{MTime: oldTime, VersionSuspended: true})
+				if err != nil {
+					t.Fatal(err)
+				}
+			}
+			old := make([][]byte, 3)
+			for i := range old {
+				old[i] = mustReadNullQuorumMeta(t, disks[i], bucket, object)
+			}
+			_, err := deleted.DeleteObject(ctx, bucket, object, ObjectOptions{VersionSuspended: true, DeleteMarker: true, MTime: oldTime.Add(time.Minute)})
+			if err != nil {
+				t.Fatal(err)
+			}
+			get := nullQuorumRequest(t, router, http.MethodGet, getGetObjectURL("", bucket, object))
+			if get.Code != http.StatusNotFound {
+				t.Fatalf("expected deleted GET, got %d %s", get.Code, get.Body.String())
+			}
+			first := nullQuorumRequest(t, router, http.MethodGet, getListObjectsV2URL("", bucket, "", "1000", "", "", ""))
+			var page ListObjectsV2Response
+			if first.Code != http.StatusOK || xml.Unmarshal(first.Body.Bytes(), &page) != nil || len(page.Contents) != 0 {
+				t.Fatalf("normal LIST marker: %d %s", first.Code, first.Body.String())
+			}
+			for i := range old {
+				if err := disks[i].WriteAll(ctx, bucket, object+"/"+xlStorageFormatFile, old[i]); err != nil {
+					t.Fatal(err)
+				}
+			}
+			walkers := slices.Clone(disks)
+			for i := 8; i < len(walkers); i++ {
+				walkers[i] = &failedMetadataWalkDisk{StorageAPI: disks[i]}
+			}
+			switchDisks(walkers)
+			fi, _, _, err := deleted.getObjectFileInfo(ctx, bucket, object, ObjectOptions{}, false)
+			if err != nil || !fi.Deleted {
+				t.Fatalf("read fallback marker: %+v %v", fi, err)
+			}
+			rec := nullQuorumRequest(t, router, http.MethodGet, getListObjectsV2URL("", bucket, "", "1000", "", "", ""))
+			page = ListObjectsV2Response{}
+			if rec.Code != http.StatusOK || xml.Unmarshal(rec.Body.Bytes(), &page) != nil || len(page.Contents) != 0 {
+				t.Fatalf("fallback resurrected old pool copy: %d %s", rec.Code, rec.Body.String())
+			}
+		})
+	}
+}
+
+// The regular metadata reader can translate a persistent generation split to
+// ObjectNotFound even though every drive still has valid metadata. LIST must
+// preserve that uncertainty instead of treating the key as deleted.
+func TestListFallbackRejectsSplitGenerations(t *testing.T) {
+	z, bucket, _ := nullQuorumBackend(t)
+	set := z.serverPools[0].sets[0]
+	disks := set.getDisks()
+	const object = "object"
+	mtime := time.Now().UTC().Add(-time.Hour)
+	var generations [3][][]byte
+	for generation := range generations {
+		body := bytes.Repeat([]byte{byte('a' + generation)}, 8192)
+		nullQuorumPut(t, z, bucket, object, body, ObjectOptions{MTime: mtime.Add(time.Duration(generation) * time.Minute)})
+		generations[generation] = make([][]byte, len(disks))
+		for i, disk := range disks {
+			generations[generation][i] = mustReadNullQuorumMeta(t, disk, bucket, object)
+		}
+	}
+	// All sixteen drives are online, but the 7/5/4 split leaves no generation
+	// with the eight matching snapshots required by the listing resolver.
+	for i, disk := range disks {
+		generation := 0
+		if i >= 7 {
+			generation = 1
+		}
+		if i >= 12 {
+			generation = 2
+		}
+		if err := disk.WriteAll(t.Context(), bucket, object+"/"+xlStorageFormatFile, generations[generation][i]); err != nil {
+			t.Fatal(err)
+		}
+	}
+	if _, _, _, err := set.getObjectFileInfo(t.Context(), bucket, object, ObjectOptions{}, false); !isErrObjectNotFound(err) {
+		t.Fatalf("split metadata did not exercise ambiguous not-found: %v", err)
+	}
+	entries := make(metaCacheEntries, len(disks))
+	for i, disk := range disks {
+		entries[i] = metaCacheEntry{name: object, metadata: mustReadNullQuorumMeta(t, disk, bucket, object)}
+		if _, err := entries[i].xlmeta(); err != nil {
+			t.Fatalf("drive %d no longer has valid metadata: %v", i, err)
+		}
+	}
+	resolver := metadataResolutionParams{dirQuorum: 8, objQuorum: 8, bucket: bucket, requestedVersions: 1}
+	entry, err := z.resolveListingGroup(t.Context(), bucket, []*metaCacheEntry{set.resolveListEntry(entries, nil, &resolver)})
+	if entry != nil || !isErrReadQuorum(err) {
+		t.Fatalf("split live metadata became absence: entry=%v err=%v", entry, err)
+	}
+}

cmd/metacache-listing-directory_test.go

@@ -0,0 +1,243 @@
+// Copyright (c) 2026 Ruohang Feng
+// SPDX-License-Identifier: AGPL-3.0-or-later
+
+package cmd
+
+import (
+	"bytes"
+	"context"
+	"encoding/xml"
+	"fmt"
+	"io"
+	"net/http"
+	"net/http/httptest"
+	"slices"
+	"testing"
+	"time"
+)
+
+type listingDirectoryWalkSnapshot struct {
+	StorageAPI
+	bucket   string
+	snapshot []byte
+}
+
+func (d listingDirectoryWalkSnapshot) WalkDir(ctx context.Context, opts WalkDirOptions, out io.Writer) error {
+	if opts.Bucket != d.bucket {
+		return d.StorageAPI.WalkDir(ctx, opts, out)
+	}
+	_, err := out.Write(d.snapshot)
+	return err
+}
+
+// Capture two actual directory-walk generations for a directory object, then
+// acknowledge its deletion. A different pool still contains dir/child. The
+// stale source walkers disagree, while every current source drive is absent.
+func TestListingRecoveryDeletedDirectoryKeepsCommonPrefix(t *testing.T) {
+	for _, marker := range []bool{false, true} {
+		for _, sourcePool := range []int{0, 1} {
+			t.Run(fmt.Sprintf("pool=%d/marker=%t", sourcePool, marker), func(t *testing.T) {
+				z, bucket, router := listingPoolsBackend(t)
+				ctx := t.Context()
+				source := z.serverPools[sourcePool].sets[0]
+				target := z.serverPools[1-sourcePool].sets[0]
+				childBody := []byte("surviving child")
+				put := func(set *erasureObjects, key string, tm time.Time) {
+					var body []byte
+					if key == "dir/child" {
+						body = childBody
+					}
+					_, err := set.PutObject(ctx, bucket, encodeDirObject(key), mustGetPutObjReader(t, bytes.NewReader(body), int64(len(body)), "", ""), ObjectOptions{MTime: tm, VersionSuspended: marker})
+					if err != nil {
+						t.Fatal(err)
+					}
+				}
+				tm := time.Now().UTC().Add(-time.Hour)
+				put(target, "dir/child", tm)
+				disks, switchDisks := installListingDiskSwitch(source)
+				snapshots := make([][]byte, len(disks))
+				for gen := range 3 {
+					if gen < 2 {
+						put(source, "dir/", tm.Add(time.Duration(gen)*time.Minute))
+					} else {
+						if _, err := source.DeleteObject(ctx, bucket, encodeDirObject("dir/"), ObjectOptions{VersionSuspended: marker, DeleteMarker: marker, MTime: tm.Add(2 * time.Minute)}); err != nil {
+							t.Fatal(err)
+						}
+					}
+					for i, disk := range disks {
+						wanted := 0
+						if i >= 6 {
+							wanted = 1
+						}
+						if i >= 12 {
+							wanted = 2
+						}
+						if wanted != gen {
+							continue
+						}
+						var b bytes.Buffer
+						if err := disk.WalkDir(ctx, WalkDirOptions{Bucket: bucket, Recursive: false}, &b); err != nil {
+							t.Fatal(err)
+						}
+						snapshots[i] = append([]byte(nil), b.Bytes()...)
+					}
+				}
+				for _, key := range []string{"dir/child", "dir/"} {
+					request, err := newTestSignedRequestV4(http.MethodGet, getGetObjectURL("", bucket, key), 0, nil, globalActiveCred.AccessKey, globalActiveCred.SecretKey, nil)
+					if err != nil {
+						t.Fatal(err)
+					}
+					response := httptest.NewRecorder()
+					router.ServeHTTP(response, request.WithContext(ctx))
+					t.Logf("GET %s HTTP %d", key, response.Code)
+					want := http.StatusOK
+					if key == "dir/" {
+						want = http.StatusNotFound
+					}
+					if response.Code != want {
+						t.Fatalf("GET %s: got %d want %d: %s", key, response.Code, want, response.Body.String())
+					}
+					if key == "dir/child" && !bytes.Equal(response.Body.Bytes(), childBody) {
+						t.Fatal("surviving child content changed")
+					}
+				}
+				wrapped := make([]StorageAPI, len(disks))
+				for i, disk := range disks {
+					wrapped[i] = listingDirectoryWalkSnapshot{disk, bucket, snapshots[i]}
+				}
+				switchDisks(wrapped)
+				t.Cleanup(func() { switchDisks(disks) })
+				req, err := newTestSignedRequestV4(http.MethodGet, getListObjectsV2URL("", bucket, "", "1000", "", "", "/"), 0, nil, globalActiveCred.AccessKey, globalActiveCred.SecretKey, nil)
+				if err != nil {
+					t.Fatal(err)
+				}
+				rec := httptest.NewRecorder()
+				router.ServeHTTP(rec, req.WithContext(ctx))
+				t.Logf("LIST HTTP %d body=%s", rec.Code, rec.Body.String())
+				var page ListObjectsV2Response
+				if rec.Code != http.StatusOK || xml.Unmarshal(rec.Body.Bytes(), &page) != nil {
+					t.Fatalf("LIST failed: %d %s", rec.Code, rec.Body.String())
+				}
+				if len(page.CommonPrefixes) != 1 || page.CommonPrefixes[0].Prefix != "dir/" {
+					t.Fatalf("acknowledged child lost its CommonPrefix: %+v", page)
+				}
+			})
+		}
+	}
+}
+
+// The directory must remain in the sorted input when an earlier object name
+// disappears. Emitting it immediately would skip an intervening name, and
+// keeping only one colliding directory would leave input order significant.
+func TestListingRecoveryDirectoryCollisionOrdering(t *testing.T) {
+	for _, object := range []string{"dir", "dir/"} {
+		for _, present := range []bool{false, true} {
+			if present && object == "dir" {
+				continue
+			}
+			for source := range 3 {
+				t.Run(fmt.Sprintf("object=%s/present=%t/source=%d", object, present, source), func(t *testing.T) {
+					in := make([]chan metaCacheEntry, 6)
+					for i := range in {
+						in[i] = make(chan metaCacheEntry, 1)
+						entry := metaCacheEntry{name: "dir/"}
+						if i == source {
+							entry = metaCacheEntry{name: object, listingSet: &erasureObjects{}}
+						}
+						if i >= 3 {
+							entry = metaCacheEntry{name: []string{"dir-", "dir.ext", "dir0"}[i-3], metadata: []byte{1}}
+						}
+						in[i] <- entry
+						close(in[i])
+					}
+					out := make(chan metaCacheEntry, len(in))
+					err := mergeEntryChannelsWithResolver(t.Context(), in, out, 1, nil, func(group []*metaCacheEntry) (*metaCacheEntry, error) {
+						if len(group) != 1 || group[0].listingSet == nil {
+							t.Fatalf("directory entered the object group: %+v", group)
+						}
+						if !present {
+							return nil, nil
+						}
+						return &metaCacheEntry{name: object, metadata: []byte{1}}, nil
+					})
+					if err != nil {
+						t.Fatal(err)
+					}
+					var names []string
+					for entry := range out {
+						names = append(names, entry.name)
+						if entry.name == "dir/" && entry.isObject() != present {
+							t.Fatalf("directory collision returned the wrong entry type: %+v", entry)
+						}
+					}
+					want := []string{"dir-", "dir.ext", "dir/", "dir0"}
+					if !slices.Equal(names, want) {
+						t.Fatalf("names=%v, want %v", names, want)
+					}
+				})
+			}
+		}
+	}
+}
+
+// A regular walk result may be stale when a pending peer forces current
+// recovery. Keep sibling prefixes until the complete object group is resolved.
+func TestListingRecoveryMixedDirectoryCollision(t *testing.T) {
+	orders := [][3]int{{0, 1, 2}, {0, 2, 1}, {1, 0, 2}, {1, 2, 0}, {2, 0, 1}, {2, 1, 0}}
+	for _, order := range orders {
+		for _, present := range []bool{false, true} {
+			t.Run(fmt.Sprintf("order=%v/present=%t", order, present), func(t *testing.T) {
+				entries := []metaCacheEntry{
+					{name: "dir/", metadata: []byte{1}},
+					{name: "dir/"},
+					{name: "dir/", listingSet: &erasureObjects{}},
+				}
+				in := make([]chan metaCacheEntry, len(order))
+				for i, index := range order {
+					in[i] = make(chan metaCacheEntry, 2)
+					in[i] <- entries[index]
+					in[i] <- metaCacheEntry{name: fmt.Sprintf("tail-%d", i), metadata: []byte{1}}
+					close(in[i])
+				}
+				out := make(chan metaCacheEntry, 6)
+				calls := 0
+				err := mergeEntryChannelsWithResolver(t.Context(), in, out, 1, nil, func(group []*metaCacheEntry) (*metaCacheEntry, error) {
+					calls++
+					if len(group) != 2 {
+						t.Fatalf("incomplete object group: %+v", group)
+					}
+					pending := 0
+					for _, entry := range group {
+						if entry.name != "dir/" || entry.isDir() {
+							t.Fatalf("unexpected recovery entry: %+v", entry)
+						}
+						if entry.listingSet != nil {
+							pending++
+						}
+					}
+					if pending != 1 {
+						t.Fatalf("pending entries=%d, want 1", pending)
+					}
+					if !present {
+						return nil, nil
+					}
+					return &metaCacheEntry{name: "dir/", metadata: []byte{1}}, nil
+				})
+				if err != nil || calls != 1 {
+					t.Fatalf("err=%v recovery calls=%d", err, calls)
+				}
+				var names []string
+				for entry := range out {
+					names = append(names, entry.name)
+					if entry.listingSet != nil || entry.name == "dir/" && entry.isObject() != present {
+						t.Fatalf("unexpected merged entry: %+v", entry)
+					}
+				}
+				want := []string{"dir/", "tail-0", "tail-1", "tail-2"}
+				if !slices.Equal(names, want) {
+					t.Fatalf("names=%v, want %v", names, want)
+				}
+			})
+		}
+	}
+}

cmd/metacache-listing-http_test.go

@@ -0,0 +1,101 @@
+// Copyright (c) 2026 Feng Ruohang
+//
+// This file is part of Silo Object Storage stack
+//
+// This program is free software: you can redistribute it and/or modify
+// it under the terms of the GNU Affero General Public License as published by
+// the Free Software Foundation, either version 3 of the License, or
+// (at your option) any later version.
+//
+// This program is distributed in the hope that it will be useful
+// but WITHOUT ANY WARRANTY; without even the implied warranty of
+// MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+// GNU Affero General Public License for more details.
+//
+// You should have received a copy of the GNU Affero General Public License
+// along with this program.  If not, see <http://www.gnu.org/licenses/>.
+
+package cmd
+
+import (
+	"bytes"
+	"encoding/xml"
+	"net/http"
+	"net/http/httptest"
+	"slices"
+	"testing"
+	"time"
+)
+
+// Exercise the actual LIST pipeline with the disputed entry in pool 1.
+func TestListFallbackMultiPoolHTTPReadLock(t *testing.T) {
+	for _, object := range []string{"object", "dir/"} {
+		t.Run(object, func(t *testing.T) {
+			z, bucket, router := listingPoolsBackend(t)
+			ctx := t.Context()
+			set := z.serverPools[1].sets[0]
+			disks := set.getDisks()
+			encoded := encodeDirObject(object)
+			body := bytes.Repeat([]byte{'a'}, 8192)
+			if object == "dir/" {
+				body = nil
+			}
+			mtime := time.Now().UTC().Add(-time.Hour)
+			put := func(tm time.Time) {
+				_, err := set.PutObject(ctx, bucket, encoded, mustGetPutObjReader(t, bytes.NewReader(body), int64(len(body)), "", ""), ObjectOptions{MTime: tm})
+				if err != nil {
+					t.Fatal(err)
+				}
+			}
+			put(mtime)
+			old := make([][]byte, 3)
+			for i := range old {
+				old[i] = mustReadNullQuorumMeta(t, disks[i], bucket, encoded)
+			}
+			put(mtime.Add(time.Minute))
+			for i := range old {
+				if err := disks[i].WriteAll(ctx, bucket, encoded+"/"+xlStorageFormatFile, old[i]); err != nil {
+					t.Fatal(err)
+				}
+			}
+			walkers := slices.Clone(disks)
+			for i := 8; i < len(walkers); i++ {
+				walkers[i] = &failedMetadataWalkDisk{StorageAPI: disks[i]}
+			}
+			set.getDisks = func() []StorageAPI { return walkers }
+			lock := z.NewNSLock(bucket, encoded)
+			lkctx, err := lock.GetLock(ctx, globalOperationTimeout)
+			if err != nil {
+				t.Fatal(err)
+			}
+			locked := true
+			defer func() {
+				if locked {
+					lock.Unlock(lkctx)
+				}
+			}()
+			target := getListObjectsV2URL("", bucket, "", "1000", "", "", "")
+			req, err := newTestSignedRequestV4(http.MethodGet, target, 0, nil, globalActiveCred.AccessKey, globalActiveCred.SecretKey, nil)
+			if err != nil {
+				t.Fatal(err)
+			}
+			rec := httptest.NewRecorder()
+			done := make(chan struct{})
+			go func() {
+				defer close(done)
+				router.ServeHTTP(rec, req.WithContext(ctx))
+			}()
+			waitForListingReadLock(t, z.serverPools[0].sets[0].nsMutex, bucket, encoded, done)
+			lock.Unlock(lkctx)
+			locked = false
+			<-done
+			var list ListObjectsV2Response
+			if rec.Code != http.StatusOK || xml.Unmarshal(rec.Body.Bytes(), &list) != nil {
+				t.Fatalf("LIST HTTP=%d body=%s", rec.Code, rec.Body.String())
+			}
+			if len(list.Contents) != 1 || list.Contents[0].Key != object {
+				t.Fatalf("LIST lost key: %+v", list)
+			}
+		})
+	}
+}

cmd/metacache-listing-lock-routes_test.go

@@ -0,0 +1,131 @@
+// Copyright (c) 2026 Feng Ruohang
+//
+// This file is part of Silo Object Storage stack
+//
+// This program is free software: you can redistribute it and/or modify
+// it under the terms of the GNU Affero General Public License as published by
+// the Free Software Foundation, either version 3 of the License, or
+// (at your option) any later version.
+//
+// This program is distributed in the hope that it will be useful
+// but WITHOUT ANY WARRANTY; without even the implied warranty of
+// MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+// GNU Affero General Public License for more details.
+//
+// You should have received a copy of the GNU Affero General Public License
+// along with this program.  If not, see <http://www.gnu.org/licenses/>.
+
+package cmd
+
+import (
+	"context"
+	"testing"
+	"time"
+
+	"github.com/minio/minio/internal/dsync"
+)
+
+// Use the real distributed-lock algorithm with in-process lock servers.
+// This is a protocol-level check, not a network or rolling-restart test.
+func TestListFallbackDistributedLockRoute(t *testing.T) {
+	for _, tc := range []struct {
+		name                     string
+		multiPool, sharedLockers bool
+	}{
+		{"multi-pool-shared-lockers", true, true},
+		{"multi-pool-disjoint-lockers", true, false},
+		{"single-pool-target-set-lockers", false, false},
+		{"single-pool-first-set-lockers", false, false},
+		{"single-pool-shared-ns-distinct-lockers", false, false},
+	} {
+		t.Run(tc.name, func(t *testing.T) {
+			a := []dsync.NetLocker{newLocker(), newLocker()}
+			b := []dsync.NetLocker{newLocker(), newLocker()}
+			if tc.sharedLockers {
+				b = a
+			}
+			set0 := &erasureObjects{nsMutex: newNSLock(true), getLockers: func() ([]dsync.NetLocker, string) { return a, "listing-test" }}
+			target := &erasureObjects{nsMutex: newNSLock(true), getLockers: func() ([]dsync.NetLocker, string) { return b, "listing-test" }}
+			if tc.name == "single-pool-shared-ns-distinct-lockers" {
+				target.nsMutex = set0.nsMutex
+			}
+			z := &erasureServerPools{serverPools: []*erasureSets{{sets: []*erasureObjects{set0, target}}}}
+			if tc.multiPool {
+				z.serverPools = []*erasureSets{{sets: []*erasureObjects{set0}}, {sets: []*erasureObjects{target}}}
+			}
+			writer := target.NewNSLock("bucket", "object")
+			if tc.multiPool || tc.name == "single-pool-first-set-lockers" {
+				writer = z.NewNSLock("bucket", "object")
+			}
+			wctx, err := writer.GetLock(t.Context(), globalOperationTimeout)
+			if err != nil {
+				t.Fatal(err)
+			}
+			locked := true
+			defer func() {
+				if locked {
+					writer.Unlock(wctx)
+				}
+			}()
+			ctx, cancel := context.WithTimeout(t.Context(), 80*time.Millisecond)
+			defer cancel()
+			_, unlock, err := z.listingRLock(ctx, "bucket", "object", target)
+			if err == nil {
+				unlock()
+				t.Fatal("fallback read lock acquired while ordinary writer lock is held")
+			}
+			writer.Unlock(wctx)
+			locked = false
+			ctx, cancel = context.WithTimeout(t.Context(), 5*time.Second)
+			defer cancel()
+			_, unlock, err = z.listingRLock(ctx, "bucket", "object", target)
+			if err != nil {
+				t.Fatalf("fallback read lock failed after writer unlock: %v", err)
+			}
+			unlock()
+		})
+	}
+}
+
+// Cancellation while waiting for the second domain must release the first.
+func TestListingSecondLockCancellationReleasesFirst(t *testing.T) {
+	first := &erasureObjects{nsMutex: newNSLock(false)}
+	target := &erasureObjects{nsMutex: newNSLock(false)}
+	z := &erasureServerPools{serverPools: []*erasureSets{{sets: []*erasureObjects{first, target}}}}
+	writer := target.NewNSLock("bucket", "object")
+	wctx, err := writer.GetLock(t.Context(), globalOperationTimeout)
+	if err != nil {
+		t.Fatal(err)
+	}
+	defer writer.Unlock(wctx)
+	ctx, cancel := context.WithCancel(t.Context())
+	defer cancel()
+	done := make(chan struct{})
+	var lockErr error
+	go func() {
+		defer close(done)
+		_, unlock, err := z.listingRLock(ctx, "bucket", "object", target)
+		lockErr = err
+		if err == nil {
+			unlock()
+		}
+	}()
+	waitForListingReadLock(t, target.nsMutex, "bucket", "object", done)
+	cancel()
+	select {
+	case <-done:
+	case <-time.After(time.Second):
+		t.Fatal("second lock did not cancel")
+	}
+	if lockErr == nil {
+		t.Fatal("canceled second lock succeeded")
+	}
+	poolWriter := z.NewNSLock("bucket", "object")
+	ctx, cancel = context.WithTimeout(t.Context(), time.Second)
+	defer cancel()
+	pctx, err := poolWriter.GetLock(ctx, globalOperationTimeout)
+	if err != nil {
+		t.Fatalf("first lock leaked after second failed: %v", err)
+	}
+	poolWriter.Unlock(pctx)
+}

cmd/metacache-listing-locks_test.go

@@ -0,0 +1,137 @@
+// Copyright (c) 2026 Feng Ruohang
+//
+// This file is part of Silo Object Storage stack
+//
+// This program is free software: you can redistribute it and/or modify
+// it under the terms of the GNU Affero General Public License as published by
+// the Free Software Foundation, either version 3 of the License, or
+// (at your option) any later version.
+//
+// This program is distributed in the hope that it will be useful
+// but WITHOUT ANY WARRANTY; without even the implied warranty of
+// MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+// GNU Affero General Public License for more details.
+//
+// You should have received a copy of the GNU Affero General Public License
+// along with this program.  If not, see <http://www.gnu.org/licenses/>.
+
+package cmd
+
+import (
+	"bytes"
+	"context"
+	"net/http"
+	"testing"
+	"time"
+)
+
+// The fallback on pool 1 must contend with the
+// pools-layer lock used by ordinary multi-pool writers (anchored at pool 0).
+func TestListFallbackMultiPoolReadLock(t *testing.T) {
+	z, bucket, _ := listingPoolsBackend(t)
+	ctx := t.Context()
+	set := z.serverPools[1].sets[0]
+	disks := set.getDisks()
+	const object = "object"
+	body := bytes.Repeat([]byte{'a'}, 8192)
+	mtime := time.Now().UTC().Add(-time.Hour)
+	put := func(tm time.Time) {
+		_, err := set.PutObject(ctx, bucket, object, mustGetPutObjReader(t, bytes.NewReader(body), int64(len(body)), "", ""), ObjectOptions{MTime: tm})
+		if err != nil {
+			t.Fatal(err)
+		}
+	}
+	put(mtime)
+	old := make([][]byte, len(disks))
+	for i := range old {
+		old[i] = mustReadNullQuorumMeta(t, disks[i], bucket, object)
+	}
+	put(mtime.Add(time.Minute))
+	entries := make(metaCacheEntries, len(disks))
+	errs := make([]error, len(disks))
+	for i := range 3 {
+		entries[i] = metaCacheEntry{name: object, metadata: old[i]}
+	}
+	for i := 3; i < 8; i++ {
+		entries[i] = metaCacheEntry{name: object, metadata: mustReadNullQuorumMeta(t, disks[i], bucket, object)}
+	}
+	for i := 8; i < len(disks); i++ {
+		errs[i] = errDiskNotFound
+	}
+	resolver := metadataResolutionParams{dirQuorum: 8, objQuorum: 8, bucket: bucket, requestedVersions: 1}
+	lock := z.NewNSLock(bucket, object)
+	lkctx, err := lock.GetLock(ctx, globalOperationTimeout)
+	if err != nil {
+		t.Fatal(err)
+	}
+	locked := true
+	defer func() {
+		if locked {
+			lock.Unlock(lkctx)
+		}
+	}()
+	done := make(chan struct{})
+	var entry *metaCacheEntry
+	var readErr error
+	go func() {
+		defer close(done)
+		entry, readErr = z.resolveListingGroup(ctx, bucket, []*metaCacheEntry{set.resolveListEntry(entries, errs, &resolver)})
+	}()
+	waitForListingReadLock(t, z.serverPools[0].sets[0].nsMutex, bucket, object, done)
+	lock.Unlock(lkctx)
+	locked = false
+	<-done
+	if readErr != nil || entry == nil || entry.name != object {
+		t.Fatalf("fallback after writer unlock: entry=%v err=%v", entry, readErr)
+	}
+}
+
+// Observe the reader registered on the writer's actual namespace lock before
+// releasing the writer. This tests mutual exclusion without a timed sleep.
+func waitForListingReadLock(t *testing.T, mutex *nsLockMap, bucket, object string, done <-chan struct{}) {
+	t.Helper()
+	timer := time.NewTimer(5 * time.Second)
+	defer timer.Stop()
+	tick := time.NewTicker(time.Millisecond)
+	defer tick.Stop()
+	for {
+		mutex.lockMapMutex.Lock()
+		lock := mutex.lockMap[pathJoin(bucket, object)]
+		waiting := lock != nil && lock.ref > 1
+		mutex.lockMapMutex.Unlock()
+		if waiting {
+			return
+		}
+		select {
+		case <-done:
+			t.Fatal("listing completed while the writer held its namespace lock")
+		case <-timer.C:
+			t.Fatal("listing did not wait on the writer's namespace lock")
+		case <-tick.C:
+		}
+	}
+}
+
+func listingPoolsBackend(t *testing.T) (*erasureServerPools, string, http.Handler) {
+	t.Helper()
+	ctx, cancel := context.WithCancel(t.Context())
+	obj, dirs, err := prepareErasurePoolsWithContext(ctx)
+	if err != nil {
+		cancel()
+		t.Fatal(err)
+	}
+	z := obj.(*erasureServerPools)
+	previous := newObjectLayerFn()
+	setObjectLayer(z)
+	t.Cleanup(func() {
+		cancel()
+		z.Shutdown(context.Background())
+		removeRoots(dirs)
+		setObjectLayer(previous)
+	})
+	bucket, router, err := initAPIHandlerTest(ctx, z, nil, MakeBucketOptions{})
+	if err != nil {
+		t.Fatal(err)
+	}
+	return z, bucket, router
+}

cmd/metacache-listing-merge_test.go

@@ -0,0 +1,244 @@
+// Copyright (c) 2026 Feng Ruohang
+//
+// This file is part of Silo Object Storage stack
+//
+// This program is free software: you can redistribute it and/or modify
+// it under the terms of the GNU Affero General Public License as published by
+// the Free Software Foundation, either version 3 of the License, or
+// (at your option) any later version.
+//
+// This program is distributed in the hope that it will be useful
+// but WITHOUT ANY WARRANTY; without even the implied warranty of
+// MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+// GNU Affero General Public License for more details.
+//
+// You should have received a copy of the GNU Affero General Public License
+// along with this program.  If not, see <http://www.gnu.org/licenses/>.
+
+package cmd
+
+import (
+	"bytes"
+	"context"
+	"encoding/xml"
+	"errors"
+	"io"
+	"net/http"
+	"net/http/httptest"
+	"net/url"
+	"slices"
+	"sync/atomic"
+	"testing"
+	"time"
+
+	"github.com/klauspost/compress/s2"
+	"github.com/tinylib/msgp/msgp"
+)
+
+// Replay real disk-walk metadata in its existing wire format, but pause before
+// the final key. This keeps the healthy pool's listing alive after the other
+// pool has already failed, without forging entries or listing errors.
+type listingPausedTailDisk struct {
+	StorageAPI
+	bucket  string
+	tail    string
+	resume  <-chan struct{}
+	blocked *atomic.Int64
+	active  *atomic.Int64
+}
+
+func (d *listingPausedTailDisk) WalkDir(ctx context.Context, opts WalkDirOptions, out io.Writer) (result error) {
+	if opts.Bucket != d.bucket {
+		return d.StorageAPI.WalkDir(ctx, opts, out)
+	}
+	if d.active != nil {
+		d.active.Add(1)
+		defer d.active.Add(-1)
+	}
+	var source bytes.Buffer
+	if err := d.StorageAPI.WalkDir(ctx, opts, &source); err != nil {
+		return err
+	}
+	r := newMetacacheReader(bytes.NewReader(source.Bytes()))
+	defer r.Close()
+	compressed := s2.NewWriter(out)
+	defer func() {
+		if err := compressed.Close(); result == nil {
+			result = err
+		}
+	}()
+	w := msgp.NewWriter(compressed)
+	if err := w.WriteByte(metacacheStreamVersion); err != nil {
+		return err
+	}
+	for {
+		entry, err := r.next()
+		if err == io.EOF {
+			if err = w.WriteBool(false); err != nil {
+				return err
+			}
+			if err = w.Flush(); err != nil {
+				return err
+			}
+			return nil
+		}
+		if err != nil {
+			return err
+		}
+		if entry.name == d.tail {
+			if err = w.Flush(); err != nil {
+				return err
+			}
+			if err = compressed.Flush(); err != nil {
+				return err
+			}
+			d.blocked.Add(1)
+			select {
+			case <-d.resume:
+			case <-ctx.Done():
+				return ctx.Err()
+			}
+		}
+		if err = w.WriteBool(true); err != nil {
+			return err
+		}
+		if err = w.WriteString(entry.name); err != nil {
+			return err
+		}
+		if err = w.WriteBytes(entry.metadata); err != nil {
+			return err
+		}
+	}
+}
+
+func TestListingMergeDoesNotSkipFailedPool(t *testing.T) {
+	ctx, cancel := context.WithTimeout(t.Context(), 20*time.Second)
+	defer cancel()
+	z, bucket, router := listingPoolsBackend(t)
+	broken, healthy := z.serverPools[0].sets[0], z.serverPools[1].sets[0]
+	brokenDisks, switchBrokenDisks := installListingDiskSwitch(broken)
+	healthyDisks, switchHealthyDisks := installListingDiskSwitch(healthy)
+	body := bytes.Repeat([]byte{'a'}, 8192)
+	mtime := time.Now().UTC().Add(-time.Hour)
+	put := func(set *erasureObjects, name string, tm time.Time) {
+		_, err := set.PutObject(ctx, bucket, name, mustGetPutObjReader(t, bytes.NewReader(body), int64(len(body)), "", ""), ObjectOptions{MTime: tm})
+		if err != nil {
+			t.Fatal(err)
+		}
+	}
+	put(broken, "a", mtime)
+	put(broken, "e", mtime)
+	for _, name := range []string{"f", "g", "h", "i", "j", "k", "zz"} {
+		put(healthy, name, mtime)
+	}
+	first := nullQuorumRequest(t, router, http.MethodGet, getListObjectsV2URL("", bucket, "", "1", "", "", ""))
+	var page ListObjectsV2Response
+	if first.Code != http.StatusOK || xml.Unmarshal(first.Body.Bytes(), &page) != nil || !page.IsTruncated || page.NextContinuationToken == "" || len(page.Contents) != 1 || page.Contents[0].Key != "a" {
+		t.Fatalf("first page HTTP=%d body=%s", first.Code, first.Body.String())
+	}
+	firstToken := page.NextContinuationToken
+	old := make([][]byte, 3)
+	for i := range old {
+		old[i] = mustReadNullQuorumMeta(t, brokenDisks[i], bucket, "e")
+	}
+	put(broken, "e", mtime.Add(time.Minute))
+	for i := range old {
+		if err := brokenDisks[i].WriteAll(ctx, bucket, "e/"+xlStorageFormatFile, old[i]); err != nil {
+			t.Fatal(err)
+		}
+	}
+	var reads, blocked, active atomic.Int64
+	brokenWalkers := slices.Clone(brokenDisks)
+	for i := range brokenWalkers {
+		brokenWalkers[i] = &listingReadFailDisk{StorageAPI: brokenDisks[i], object: "e", reads: &reads}
+		if i >= 8 {
+			brokenWalkers[i] = &failedMetadataWalkDisk{StorageAPI: brokenWalkers[i]}
+		}
+	}
+	switchBrokenDisks(brokenWalkers)
+	resume := make(chan struct{})
+	defer close(resume)
+	healthyWalkers := slices.Clone(healthyDisks)
+	for i, disk := range healthyWalkers {
+		healthyWalkers[i] = &listingPausedTailDisk{StorageAPI: disk, bucket: bucket, tail: "zz", resume: resume, blocked: &blocked, active: &active}
+	}
+	switchHealthyDisks(healthyWalkers)
+
+	target := getListObjectsV2URL("", bucket, "", "2", "", "", "") + "&continuation-token=" + url.QueryEscape(firstToken)
+	req, err := newTestSignedRequestV4(http.MethodGet, target, 0, nil, globalActiveCred.AccessKey, globalActiveCred.SecretKey, nil)
+	if err != nil {
+		t.Fatal(err)
+	}
+	response := make(chan *httptest.ResponseRecorder, 1)
+	go func() {
+		rec := httptest.NewRecorder()
+		router.ServeHTTP(rec, req.WithContext(ctx))
+		response <- rec
+	}()
+	select {
+	case rec := <-response:
+		t.Logf("continuation LIST=%d failed fallback ReadXL=%d paused healthy walkers=%d body=%s", rec.Code, reads.Load(), blocked.Load(), rec.Body.String())
+		if reads.Load() == 0 || blocked.Load() == 0 {
+			t.Fatal("schedule did not exercise fallback failure and unfinished healthy pool")
+		}
+		var apiErr APIErrorResponse
+		if rec.Code != http.StatusServiceUnavailable || xml.Unmarshal(rec.Body.Bytes(), &apiErr) != nil || apiErr.Code != "SlowDownRead" {
+			t.Fatalf("failed pool advanced the page instead of reporting read quorum: %d %s", rec.Code, rec.Body.String())
+		}
+		waitForListingCondition(t, "canceled disk walkers did not exit", func() bool { return active.Load() == 0 })
+	case <-ctx.Done():
+		t.Fatalf("no early response: %v reads=%d blocked=%d", ctx.Err(), reads.Load(), blocked.Load())
+	}
+}
+
+func TestListingMergeCancellation(t *testing.T) {
+	z, bucket, _ := nullQuorumBackend(t)
+	for _, key := range []string{"a", "b", "zz"} {
+		nullQuorumPut(t, z, bucket, key, []byte(key), ObjectOptions{})
+	}
+	set := z.serverPools[0].sets[0]
+	var blocked, active atomic.Int64
+	resume := make(chan struct{})
+	defer close(resume)
+	walkers := slices.Clone(set.getDisks())
+	for i, disk := range walkers {
+		walkers[i] = &listingPausedTailDisk{StorageAPI: disk, bucket: bucket, tail: "zz", resume: resume, blocked: &blocked, active: &active}
+	}
+	set.getDisks = func() []StorageAPI { return walkers }
+	ctx, cancel := context.WithCancel(t.Context())
+	defer cancel()
+	entries := make(chan metaCacheEntry, 10)
+	done := make(chan error, 1)
+	go func() {
+		done <- z.listMerged(ctx, listPathOptions{Bucket: bucket, Recursive: true, Limit: 10}, entries)
+	}()
+	waitForListingCondition(t, "disk walks did not reach the paused tail", func() bool { return blocked.Load() > 0 })
+	cancel()
+	select {
+	case err := <-done:
+		if !errors.Is(err, context.Canceled) {
+			t.Fatalf("canceled merge returned %v", err)
+		}
+	case <-time.After(5 * time.Second):
+		t.Fatal("canceled merge did not finish")
+	}
+	// listMerged must close its output and release every paused disk walk.
+	for range entries {
+	}
+	waitForListingCondition(t, "canceled disk walkers did not exit", func() bool { return active.Load() == 0 })
+}
+
+func waitForListingCondition(t *testing.T, failure string, ready func() bool) {
+	t.Helper()
+	timer := time.NewTimer(5 * time.Second)
+	defer timer.Stop()
+	tick := time.NewTicker(time.Millisecond)
+	defer tick.Stop()
+	for !ready() {
+		select {
+		case <-timer.C:
+			t.Fatal(failure)
+		case <-tick.C:
+		}
+	}
+}

cmd/metacache-listing-recovery_test.go

@@ -0,0 +1,408 @@
+// Copyright (c) 2026 Feng Ruohang
+//
+// This file is part of Silo Object Storage stack
+//
+// This program is free software: you can redistribute it and/or modify
+// it under the terms of the GNU Affero General Public License as published by
+// the Free Software Foundation, either version 3 of the License, or
+// (at your option) any later version.
+//
+// This program is distributed in the hope that it will be useful
+// but WITHOUT ANY WARRANTY; without even the implied warranty of
+// MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+// GNU Affero General Public License for more details.
+//
+// You should have received a copy of the GNU Affero General Public License
+// along with this program.  If not, see <http://www.gnu.org/licenses/>.
+
+package cmd
+
+import (
+	"bytes"
+	"context"
+	"errors"
+	"fmt"
+	"io"
+	"slices"
+	"sync/atomic"
+	"testing"
+	"time"
+	"unsafe"
+)
+
+type listingRecoveryDisk struct {
+	StorageAPI
+	reads        *atomic.Int64
+	err          error
+	emptySuccess bool
+}
+
+func (d *listingRecoveryDisk) ReadXL(ctx context.Context, bucket, object string, data bool) (RawFileInfo, error) {
+	if d.reads != nil {
+		d.reads.Add(1)
+	}
+	if d.emptySuccess {
+		return RawFileInfo{}, nil
+	}
+	if d.err != nil {
+		return RawFileInfo{}, d.err
+	}
+	return d.StorageAPI.ReadXL(ctx, bucket, object, data)
+}
+
+func TestListingRecoveryStrictAbsence(t *testing.T) {
+	z, bucket, _ := nullQuorumBackend(t)
+	set := z.serverPools[0].sets[0]
+	disks, switchDisks := installListingDiskSwitch(set)
+	for _, tc := range []struct {
+		name                   string
+		err                    error
+		offline, empty, absent bool
+	}{
+		{name: "all-native-not-found", absent: true},
+		{name: "offline", offline: true},
+		{name: "volume-not-found", err: errVolumeNotFound},
+		{name: "version-not-found", err: errFileVersionNotFound},
+		{name: "corrupt", err: errFileCorrupt},
+		{name: "decode-generated-not-found", empty: true},
+	} {
+		t.Run(tc.name, func(t *testing.T) {
+			next := slices.Clone(disks)
+			next[0] = &listingRecoveryDisk{StorageAPI: disks[0], err: tc.err, emptySuccess: tc.empty}
+			if tc.offline {
+				next[0] = &offlineNullQuorumDisk{StorageAPI: disks[0]}
+			}
+			switchDisks(next)
+			_, absent, err := set.readListingFileInfo(t.Context(), bucket, "missing")
+			if absent != tc.absent || (tc.absent && err != nil) || (!tc.absent && !isErrReadQuorum(err)) {
+				t.Fatalf("absent=%v err=%v, want absent=%v", absent, err, tc.absent)
+			}
+		})
+	}
+	switchDisks(disks)
+}
+
+// The target walk can predate the copy. Only current cross-pool reads can
+// distinguish this copy-before-delete schedule from a genuine object deletion.
+func TestListingRecoveryCurrentPools(t *testing.T) {
+	for _, tc := range []struct {
+		name                           string
+		target, marker, unknown, stale bool
+	}{
+		{name: "copied-before-source-delete", target: true},
+		{name: "deleted-with-stale-other-walk", stale: true},
+		{name: "current-marker-over-stale-live", target: true, marker: true, stale: true},
+		{name: "missing-source-unknown-target", unknown: true},
+	} {
+		t.Run(tc.name, func(t *testing.T) {
+			z, bucket, _ := listingPoolsBackend(t)
+			target, source := z.serverPools[0].sets[0], z.serverPools[1].sets[0]
+			targetDisks, switchTarget := installListingDiskSwitch(target)
+			sourceDisks, switchSource := installListingDiskSwitch(source)
+			const object = "moved"
+			body := bytes.Repeat([]byte{'a'}, 8192)
+			mtime := time.Now().UTC().Add(-time.Hour)
+			put := func(set *erasureObjects) {
+				_, err := set.PutObject(t.Context(), bucket, object, mustGetPutObjReader(t, bytes.NewReader(body), int64(len(body)), "", ""), ObjectOptions{MTime: mtime, VersionSuspended: true})
+				if err != nil {
+					t.Fatal(err)
+				}
+			}
+			put(source)
+			var stale *metaCacheEntry
+			if tc.stale || tc.target {
+				put(target)
+				stale = &metaCacheEntry{name: object, metadata: mustReadNullQuorumMeta(t, targetDisks[0], bucket, object)}
+			}
+			if tc.stale && !tc.target {
+				if _, err := target.DeleteObject(t.Context(), bucket, object, ObjectOptions{}); err != nil {
+					t.Fatal(err)
+				}
+			}
+			if tc.marker {
+				if _, err := target.DeleteObject(t.Context(), bucket, object, ObjectOptions{VersionSuspended: true, DeleteMarker: true, MTime: mtime.Add(time.Minute)}); err != nil {
+					t.Fatal(err)
+				}
+			}
+			// Actual copy is committed before actual source cleanup.
+			if _, err := source.DeleteObject(t.Context(), bucket, object, ObjectOptions{}); err != nil {
+				t.Fatal(err)
+			}
+			var sourceReads, targetReads atomic.Int64
+			nextSource, nextTarget := slices.Clone(sourceDisks), slices.Clone(targetDisks)
+			for i := range nextSource {
+				nextSource[i] = &listingRecoveryDisk{StorageAPI: sourceDisks[i], reads: &sourceReads}
+			}
+			for i := range nextTarget {
+				nextTarget[i] = &listingRecoveryDisk{StorageAPI: targetDisks[i], reads: &targetReads}
+			}
+			if tc.unknown {
+				nextTarget[0] = &listingRecoveryDisk{StorageAPI: targetDisks[0], reads: &targetReads, err: errDiskNotFound}
+			}
+			switchSource(nextSource)
+			switchTarget(nextTarget)
+			group := []*metaCacheEntry{{name: object, listingSet: source}}
+			if stale != nil && tc.stale {
+				group = append(group, stale)
+			}
+			entry, err := z.resolveListingGroup(t.Context(), bucket, group)
+			if sourceReads.Load() != int64(len(sourceDisks)) || targetReads.Load() != int64(len(targetDisks)) {
+				t.Fatalf("reads source=%d target=%d", sourceReads.Load(), targetReads.Load())
+			}
+			if tc.unknown {
+				if entry != nil || !isErrReadQuorum(err) {
+					t.Fatalf("unknown became success: entry=%v err=%v", entry, err)
+				}
+				return
+			}
+			if err != nil {
+				t.Fatal(err)
+			}
+			if !tc.target {
+				if entry != nil {
+					t.Fatalf("stale walk survived current absence: %+v", entry)
+				}
+				return
+			}
+			if entry == nil || entry.listingSet != nil {
+				t.Fatalf("missing current result: %+v", entry)
+			}
+			fi, err := entry.fileInfo(bucket)
+			if err != nil || fi.Deleted != tc.marker {
+				t.Fatalf("current result deleted=%v err=%v, want marker=%v", fi.Deleted, err, tc.marker)
+			}
+		})
+	}
+}
+
+func TestListingRecoveryNormalPathDoesNotReadOrLock(t *testing.T) {
+	z, bucket, _ := listingPoolsBackend(t)
+	const object = "normal"
+	set := z.serverPools[1].sets[0]
+	body := []byte("normal data")
+	if _, err := set.PutObject(t.Context(), bucket, object, mustGetPutObjReader(t, bytes.NewReader(body), int64(len(body)), "", ""), ObjectOptions{}); err != nil {
+		t.Fatal(err)
+	}
+	var reads atomic.Int64
+	for _, pool := range z.serverPools {
+		for _, set := range pool.sets {
+			disks, switchDisks := installListingDiskSwitch(set)
+			wrapped := slices.Clone(disks)
+			for i := range wrapped {
+				wrapped[i] = &listingRecoveryDisk{StorageAPI: disks[i], reads: &reads}
+			}
+			switchDisks(wrapped)
+		}
+	}
+	writer := z.NewNSLock(bucket, object)
+	lkctx, err := writer.GetLock(t.Context(), globalOperationTimeout)
+	if err != nil {
+		t.Fatal(err)
+	}
+	defer writer.Unlock(lkctx)
+	ctx, cancel := context.WithTimeout(t.Context(), 5*time.Second)
+	defer cancel()
+	out := make(chan metaCacheEntry, 10)
+	err = z.listMerged(ctx, listPathOptions{Bucket: bucket, Recursive: true, Limit: 10}, out)
+	if err != nil && err != io.EOF {
+		t.Fatalf("normal listing waited on writer or failed: %v", err)
+	}
+	var names []string
+	for entry := range out {
+		names = append(names, entry.name)
+		if entry.listingSet != nil {
+			t.Fatal("pending escaped")
+		}
+	}
+	if !slices.Equal(names, []string{object}) || reads.Load() != 0 {
+		t.Fatalf("normal names=%v additional ReadXL=%d", names, reads.Load())
+	}
+}
+
+func TestListingPendingDirectoryCollision(t *testing.T) {
+	for _, objectFirst := range []bool{false, true} {
+		for _, present := range []bool{true, false} {
+			t.Run(fmt.Sprintf("objectFirst=%v/present=%v", objectFirst, present), func(t *testing.T) {
+				in := []chan metaCacheEntry{make(chan metaCacheEntry, 2), make(chan metaCacheEntry, 2), make(chan metaCacheEntry, 2)}
+				objectIndex := 1
+				if objectFirst {
+					objectIndex = 0
+				}
+				in[objectIndex] <- metaCacheEntry{name: "dir/", listingSet: &erasureObjects{}}
+				in[1-objectIndex] <- metaCacheEntry{name: "dir/"}
+				in[2] <- metaCacheEntry{name: "a", metadata: []byte{1}}
+				for _, c := range in {
+					close(c)
+				}
+				out := make(chan metaCacheEntry, 4)
+				calls := 0
+				err := mergeEntryChannelsWithResolver(t.Context(), in, out, 1, nil, func(group []*metaCacheEntry) (*metaCacheEntry, error) {
+					calls++
+					if len(group) != 1 || group[0].listingSet == nil {
+						t.Fatalf("wrong pending group: %+v", group)
+					}
+					if !present {
+						return nil, nil
+					}
+					return &metaCacheEntry{name: "dir/", metadata: []byte{1}}, nil
+				})
+				if err != nil || calls != 1 || len(out) != 2 {
+					t.Fatalf("err=%v recoveries=%d outputs=%d", err, calls, len(out))
+				}
+				first := <-out
+				if first.name != "a" {
+					t.Fatalf("expected smaller key first, got %+v", first)
+				}
+				entry := <-out
+				if entry.name != "dir/" || entry.listingSet != nil {
+					t.Fatalf("unexpected recovered entry: %+v", entry)
+				}
+				if present && (entry.isDir() || !entry.isObjectDir()) {
+					t.Fatalf("object became directory: %+v", entry)
+				}
+				if !present && (!entry.isDir() || entry.name != "dir/") {
+					t.Fatalf("absent pending object dropped live directory prefix: %+v", entry)
+				}
+			})
+		}
+	}
+}
+
+func TestListingRecoveryRejectsUnexpectedSinglePoolSet(t *testing.T) {
+	sets := []*erasureObjects{{}, {}}
+	pool := &erasureSets{sets: sets, distributionAlgo: formatErasureVersionV3DistributionAlgoV3}
+	z := &erasureServerPools{serverPools: []*erasureSets{pool}}
+	for _, object := range []string{"key", "directory/"} {
+		target := pool.getHashedSet(encodeDirObject(object))
+		wrong := sets[0]
+		if wrong == target {
+			wrong = sets[1]
+		}
+		entry, err := z.resolveListingGroup(t.Context(), "bucket", []*metaCacheEntry{{name: object, listingSet: wrong}})
+		if entry != nil || !isErrReadQuorum(err) {
+			t.Fatalf("unexpected set became success: %v %v", entry, err)
+		}
+	}
+}
+
+func TestListingRecoveryReleasesBeforeOutput(t *testing.T) {
+	z, bucket, _ := nullQuorumBackend(t)
+	const object = "object/"
+	nullQuorumPut(t, z, bucket, object, []byte("data"), ObjectOptions{})
+	set := z.serverPools[0].sets[0]
+	input := make(chan metaCacheEntry, 1)
+	input <- metaCacheEntry{name: object, listingSet: set}
+	close(input)
+	out := make(chan metaCacheEntry)
+	resolved := make(chan struct{})
+	done := make(chan error, 1)
+	ctx, cancel := context.WithTimeout(t.Context(), 5*time.Second)
+	defer cancel()
+	go func() {
+		done <- mergeEntryChannelsWithResolver(ctx, []chan metaCacheEntry{input}, out, 1, nil, func(group []*metaCacheEntry) (*metaCacheEntry, error) {
+			entry, err := z.resolveListingGroup(ctx, bucket, group)
+			close(resolved)
+			return entry, err
+		})
+	}()
+	select {
+	case <-resolved:
+	case <-ctx.Done():
+		t.Fatal(ctx.Err())
+	}
+	lock := set.NewNSLock(bucket, encodeDirObject(object))
+	lkctx, err := lock.GetLock(ctx, globalOperationTimeout)
+	if err != nil {
+		t.Fatalf("writer blocked by output backpressure: %v", err)
+	}
+	lock.Unlock(lkctx)
+	entry := <-out
+	if entry.name != object || !entry.isObjectDir() || entry.isDir() || entry.listingSet != nil {
+		t.Fatalf("directory object identity lost: %+v", entry)
+	}
+	if err := <-done; err != nil {
+		t.Fatal(err)
+	}
+}
+
+func TestListingRecoveryTerminalErrorPrecedesLock(t *testing.T) {
+	for _, n := range []int{1, 2} {
+		t.Run(fmt.Sprint(n), func(t *testing.T) {
+			inputs := make([]chan metaCacheEntry, n)
+			for i := range inputs {
+				inputs[i] = make(chan metaCacheEntry, 1)
+				inputs[i] <- metaCacheEntry{name: "key", listingSet: &erasureObjects{}}
+				close(inputs[i])
+			}
+			out := make(chan metaCacheEntry, 1)
+			want := errors.New("walk failed")
+			called := false
+			err := mergeEntryChannelsWithResolver(t.Context(), inputs, out, 1, func(int) error { return want }, func([]*metaCacheEntry) (*metaCacheEntry, error) { called = true; return nil, nil })
+			if !errors.Is(err, want) || called || len(out) != 0 {
+				t.Fatalf("err=%v recovery=%v output=%d", err, called, len(out))
+			}
+		})
+	}
+}
+
+func TestListingPendingCannotBeSerialized(t *testing.T) {
+	entry := metaCacheEntry{name: "directory/", listingSet: &erasureObjects{}}
+	if entry.isDir() || !entry.isObjectDir() {
+		t.Fatal("pending object became a directory")
+	}
+	var buf bytes.Buffer
+	w := newMetacacheWriter(&buf, 0)
+	if err := w.write(entry); err == nil {
+		t.Fatal("pending entry serialized")
+	}
+	_ = w.Close()
+	buf.Reset()
+	w = newMetacacheWriter(&buf, 0)
+	stream, err := w.stream()
+	if err != nil {
+		t.Fatal(err)
+	}
+	stream <- entry
+	close(stream)
+	if err = w.Close(); err == nil {
+		t.Fatal("stream serialized pending entry")
+	}
+}
+
+func TestListingEntryRepresentationCost(t *testing.T) {
+	type before struct {
+		_ string    // name
+		_ []byte    // metadata
+		_ *xlMetaV2 // cached
+		_ bool      // reusable
+	}
+	t.Logf("metaCacheEntry before=%d after=%d bytes", unsafe.Sizeof(before{}), unsafe.Sizeof(metaCacheEntry{}))
+}
+
+func BenchmarkListingMergeNormal(b *testing.B) {
+	for _, n := range []int{1, 2} {
+		for _, resolver := range []bool{false, true} {
+			b.Run(fmt.Sprintf("inputs=%d/resolver=%v", n, resolver), func(b *testing.B) {
+				b.ReportAllocs()
+				for b.Loop() {
+					inputs := make([]chan metaCacheEntry, n)
+					for i := range inputs {
+						inputs[i] = make(chan metaCacheEntry, 100)
+						for j := range 100 {
+							inputs[i] <- metaCacheEntry{name: fmt.Sprintf("key-%03d-%d/", j, i)}
+						}
+						close(inputs[i])
+					}
+					out := make(chan metaCacheEntry, n*100)
+					var resolveGroup func([]*metaCacheEntry) (*metaCacheEntry, error)
+					if resolver {
+						resolveGroup = func([]*metaCacheEntry) (*metaCacheEntry, error) { return nil, io.ErrUnexpectedEOF }
+					}
+					if err := mergeEntryChannelsWithResolver(context.Background(), inputs, out, 1, nil, resolveGroup); err != nil {
+						b.Fatal(err)
+					}
+				}
+			})
+		}
+	}
+}

cmd/metacache-null-quorum_test.go

@@ -29,6 +29,7 @@ import (
 	"path/filepath"
 	"slices"
 	"sync"
+	"sync/atomic"
 	"testing"
 	"time"
 
@@ -52,6 +53,37 @@ type nullQuorumWalkDisk struct {
 	index    int
 }
 
+type failedMetadataWalkDisk struct {
+	StorageAPI
+}
+
+func (d *failedMetadataWalkDisk) WalkDir(context.Context, WalkDirOptions, io.Writer) error {
+	return errDiskNotFound
+}
+
+// offlineNullQuorumDisk models a drive of a node that is away.
+type offlineNullQuorumDisk struct {
+	StorageAPI
+}
+
+func (d *offlineNullQuorumDisk) IsOnline() bool { return false }
+
+func (d *offlineNullQuorumDisk) DiskInfo(context.Context, DiskInfoOptions) (DiskInfo, error) {
+	return DiskInfo{}, errDiskNotFound
+}
+
+func (d *offlineNullQuorumDisk) WalkDir(context.Context, WalkDirOptions, io.Writer) error {
+	return errDiskNotFound
+}
+
+func (d *offlineNullQuorumDisk) ReadXL(context.Context, string, string, bool) (RawFileInfo, error) {
+	return RawFileInfo{}, errDiskNotFound
+}
+
+func (d *offlineNullQuorumDisk) ReadVersion(context.Context, string, string, string, string, ReadOptions) (FileInfo, error) {
+	return FileInfo{}, errDiskNotFound
+}
+
 func (d *nullQuorumWalkDisk) WalkDir(ctx context.Context, opts WalkDirOptions, out io.Writer) error {
 	if opts.Bucket != d.schedule.bucket {
 		return d.StorageAPI.WalkDir(ctx, opts, out)
@@ -119,6 +151,30 @@ func nullQuorumRequest(t *testing.T, router http.Handler, method, target string)
 	return rec
 }
 
+func nullQuorumPut(t *testing.T, z *erasureServerPools, bucket, object string, body []byte, opts ObjectOptions) {
+	t.Helper()
+	if _, err := z.PutObject(t.Context(), bucket, object, mustGetPutObjReader(t, bytes.NewReader(body), int64(len(body)), "", ""), opts); err != nil {
+		t.Fatal(err)
+	}
+}
+
+func nullQuorumListV2Keys(t *testing.T, router http.Handler, bucket string) (int, []string) {
+	t.Helper()
+	rec := nullQuorumRequest(t, router, http.MethodGet, getListObjectsV2URL("", bucket, "", "1000", "", "", ""))
+	if rec.Code != http.StatusOK {
+		return rec.Code, nil
+	}
+	var list ListObjectsV2Response
+	if err := xml.Unmarshal(rec.Body.Bytes(), &list); err != nil {
+		t.Fatal(err)
+	}
+	keys := make([]string, 0, len(list.Contents))
+	for _, object := range list.Contents {
+		keys = append(keys, object.Key)
+	}
+	return rec.Code, keys
+}
+
 func TestListObjectsSingleNullQuorumHTTP(t *testing.T) {
 	z, bucket, router := nullQuorumBackend(t)
 	oldBody := bytes.Repeat([]byte{'a'}, 8192)
@@ -227,6 +283,400 @@ func TestListObjectsSingleNullQuorumHTTP(t *testing.T) {
 	}
 }
 
+func TestListObjectsFallsBackToReadableMetadata(t *testing.T) {
+	z, bucket, router := nullQuorumBackend(t)
+	set := z.serverPools[0].sets[0]
+	disks, switchDisks := installListingDiskSwitch(set)
+	const object = "object"
+	oldBody := bytes.Repeat([]byte{'a'}, 8192)
+	newBody := bytes.Repeat([]byte{'b'}, 8192)
+	oldTime := time.Now().UTC().Add(-time.Hour)
+
+	if _, err := z.PutObject(t.Context(), bucket, object, mustGetPutObjReader(t, bytes.NewReader(oldBody), int64(len(oldBody)), "", ""), ObjectOptions{MTime: oldTime}); err != nil {
+		t.Fatal(err)
+	}
+	oldMeta := make([][]byte, len(disks))
+	for i, disk := range disks {
+		oldMeta[i] = mustReadNullQuorumMeta(t, disk, bucket, object)
+	}
+	if _, err := z.PutObject(t.Context(), bucket, object, mustGetPutObjReader(t, bytes.NewReader(newBody), int64(len(newBody)), "", ""), ObjectOptions{MTime: oldTime.Add(time.Minute)}); err != nil {
+		t.Fatal(err)
+	}
+
+	// The eight readable walk streams are split 3/5 across two generations;
+	// the other eight streams fail. Neither generation is proven readable by
+	// the listing inputs, so returning HTTP 200 would silently omit the key.
+	for i := range 3 {
+		if err := disks[i].WriteAll(t.Context(), bucket, object+"/"+xlStorageFormatFile, oldMeta[i]); err != nil {
+			t.Fatal(err)
+		}
+	}
+	wrapped := make([]StorageAPI, len(disks))
+	for i, disk := range disks {
+		wrapped[i] = disk
+		if i >= 8 {
+			wrapped[i] = &failedMetadataWalkDisk{StorageAPI: disk}
+		}
+	}
+	switchDisks(wrapped)
+
+	rec := nullQuorumRequest(t, router, http.MethodGet, getListObjectsV2URL("", bucket, "", "1000", "", "", ""))
+	var list ListObjectsV2Response
+	if rec.Code != http.StatusOK || xml.Unmarshal(rec.Body.Bytes(), &list) != nil {
+		t.Fatalf("LIST failed to resolve readable metadata: %d %s", rec.Code, rec.Body.String())
+	}
+	if list.KeyCount != 1 || len(list.Contents) != 1 || list.Contents[0].Key != object {
+		t.Fatalf("LIST omitted the readable key: %+v", list)
+	}
+
+	// The object itself still has a read quorum outside the failed walk
+	// streams, demonstrating why the LIST must report uncertainty, not absence.
+	get := nullQuorumRequest(t, router, http.MethodGet, getGetObjectURL("", bucket, object))
+	if get.Code != http.StatusOK || !bytes.Equal(get.Body.Bytes(), newBody) {
+		t.Fatalf("GET lost the readable generation: %d %q", get.Code, get.Body.String())
+	}
+	// Walk uses a separate producer but must share the same latest-only pool
+	// recovery. Versioned walks retain their existing no-fallback behavior.
+	for _, latestOnly := range []bool{false, true} {
+		out := make(chan itemOrErr[ObjectInfo], 10)
+		if err := z.Walk(t.Context(), bucket, "", out, WalkOptions{LatestOnly: latestOnly, AskDisks: "strict"}); err != nil {
+			t.Fatal(err)
+		}
+		var objects []ObjectInfo
+		for result := range out {
+			if result.Err != nil {
+				t.Fatal(result.Err)
+			}
+			objects = append(objects, result.Item)
+		}
+		if latestOnly {
+			if len(objects) != 1 || objects[0].Name != object || !objects[0].ModTime.Equal(oldTime.Add(time.Minute)) {
+				t.Fatalf("Walk latest recovery: %+v", objects)
+			}
+		} else if len(objects) != 0 {
+			t.Fatalf("versioned Walk unexpectedly enabled recovery: %+v", objects)
+		}
+	}
+	failedDisks := slices.Clone(wrapped)
+	for i, disk := range failedDisks {
+		failedDisks[i] = &listingReadFailDisk{StorageAPI: disk, object: object, reads: &atomic.Int64{}}
+	}
+	switchDisks(failedDisks)
+	out := make(chan itemOrErr[ObjectInfo], 10)
+	if err := z.Walk(t.Context(), bucket, "", out, WalkOptions{LatestOnly: true, AskDisks: "strict"}); err != nil {
+		t.Fatal(err)
+	}
+	failed := false
+	for result := range out {
+		if result.Err == nil {
+			t.Fatalf("failed Walk yielded object: %+v", result.Item)
+		}
+		if isErrReadQuorum(result.Err) {
+			failed = true
+		}
+	}
+	if !failed {
+		t.Fatal("failed Walk recovery did not report quorum error")
+	}
+}
+
+// The fallback must read under the object read lock. A lock-free read of an
+// object being overwritten stably reports a live object as absent, which is
+// the omission this repair exists to remove. Hold the write lock and assert
+// the listing waits for it rather than resolving straight through.
+func TestListFallbackTakesObjectReadLock(t *testing.T) {
+	z, bucket, router := nullQuorumBackend(t)
+	set := z.serverPools[0].sets[0]
+	disks := set.getDisks()
+	const object = "object"
+	oldBody, newBody := bytes.Repeat([]byte{'a'}, 8192), bytes.Repeat([]byte{'b'}, 8192)
+	oldTime := time.Now().UTC().Add(-time.Hour)
+
+	if _, err := z.PutObject(t.Context(), bucket, object, mustGetPutObjReader(t, bytes.NewReader(oldBody), int64(len(oldBody)), "", ""), ObjectOptions{MTime: oldTime}); err != nil {
+		t.Fatal(err)
+	}
+	oldMeta := make([][]byte, len(disks))
+	for i, disk := range disks {
+		oldMeta[i] = mustReadNullQuorumMeta(t, disk, bucket, object)
+	}
+	if _, err := z.PutObject(t.Context(), bucket, object, mustGetPutObjReader(t, bytes.NewReader(newBody), int64(len(newBody)), "", ""), ObjectOptions{MTime: oldTime.Add(time.Minute)}); err != nil {
+		t.Fatal(err)
+	}
+	// Split the readable walk streams so the entry cannot be resolved from
+	// the snapshots and the fallback has to run.
+	for i := range 3 {
+		if err := disks[i].WriteAll(t.Context(), bucket, object+"/"+xlStorageFormatFile, oldMeta[i]); err != nil {
+			t.Fatal(err)
+		}
+	}
+	wrapped := make([]StorageAPI, len(disks))
+	for i, disk := range disks {
+		wrapped[i] = disk
+		if i >= 8 {
+			wrapped[i] = &failedMetadataWalkDisk{StorageAPI: disk}
+		}
+	}
+	set.getDisks = func() []StorageAPI { return wrapped }
+
+	lock := set.NewNSLock(bucket, object)
+	lkctx, err := lock.GetLock(t.Context(), globalOperationTimeout)
+	if err != nil {
+		t.Fatal(err)
+	}
+	locked := true
+	defer func() {
+		if locked {
+			lock.Unlock(lkctx)
+		}
+	}()
+	target := getListObjectsV2URL("", bucket, "", "1000", "", "", "")
+	req, err := newTestSignedRequestV4(http.MethodGet, target, 0, nil, globalActiveCred.AccessKey, globalActiveCred.SecretKey, nil)
+	if err != nil {
+		t.Fatal(err)
+	}
+	rec := httptest.NewRecorder()
+	done := make(chan struct{})
+	go func() {
+		defer close(done)
+		router.ServeHTTP(rec, req)
+	}()
+	waitForListingReadLock(t, set.nsMutex, bucket, object, done)
+	lock.Unlock(lkctx)
+	locked = false
+	<-done
+
+	var list ListObjectsV2Response
+	if rec.Code != http.StatusOK || xml.Unmarshal(rec.Body.Bytes(), &list) != nil {
+		t.Fatalf("LIST failed after the lock was released: %d %s", rec.Code, rec.Body.String())
+	}
+	if list.KeyCount != 1 || len(list.Contents) != 1 || list.Contents[0].Key != object {
+		t.Fatalf("LIST omitted the readable key: %+v", list)
+	}
+}
+
+// A truncated xl.meta is what a concurrent overwrite looks like on one drive,
+// and the walker skips that entry. Skipping is safe and deliberate: the key is
+// only omitted when no generation reaches quorum, which listPath now resolves
+// against the read path. Pin that, so the walker is not "hardened" into
+// failing a whole listing over a minority of unreadable drives.
+func TestListObjectsSurvivesTruncatedMetadataMinority(t *testing.T) {
+	z, bucket, router := nullQuorumBackend(t)
+	set := z.serverPools[0].sets[0]
+	disks := set.getDisks()
+	const object = "object"
+	body := bytes.Repeat([]byte{'a'}, 8192)
+
+	if _, err := z.PutObject(t.Context(), bucket, object, mustGetPutObjReader(t, bytes.NewReader(body), int64(len(body)), "", ""), ObjectOptions{}); err != nil {
+		t.Fatal(err)
+	}
+
+	// Truncate xl.meta on a minority of drives, as an overwrite in flight does.
+	for i := range 4 {
+		meta := mustReadNullQuorumMeta(t, disks[i], bucket, object)
+		if err := disks[i].WriteAll(t.Context(), bucket, object+"/"+xlStorageFormatFile, meta[:len(meta)/2]); err != nil {
+			t.Fatal(err)
+		}
+	}
+
+	rec := nullQuorumRequest(t, router, http.MethodGet, getListObjectsV2URL("", bucket, "", "1000", "", "", ""))
+	var list ListObjectsV2Response
+	if rec.Code != http.StatusOK || xml.Unmarshal(rec.Body.Bytes(), &list) != nil {
+		t.Fatalf("a truncated minority failed the listing: %d %s", rec.Code, rec.Body.String())
+	}
+	if list.KeyCount != 1 || len(list.Contents) != 1 || list.Contents[0].Key != object {
+		t.Fatalf("a truncated minority omitted the key: %+v", list)
+	}
+}
+
+func TestListStaleMinorityIgnored(t *testing.T) {
+	z, bucket, router := nullQuorumBackend(t)
+	disks := z.serverPools[0].sets[0].getDisks()
+	body := bytes.Repeat([]byte{'a'}, 8192)
+	nullQuorumPut(t, z, bucket, "gone", body, ObjectOptions{})
+	nullQuorumPut(t, z, bucket, "keep", body, ObjectOptions{})
+	for _, disk := range disks[4:] {
+		if err := disk.Delete(t.Context(), bucket, "gone", DeleteOptions{Recursive: true, Immediate: true}); err != nil {
+			t.Fatal(err)
+		}
+	}
+
+	if code, keys := nullQuorumListV2Keys(t, router, bucket); code != http.StatusOK || !slices.Equal(keys, []string{"keep"}) {
+		t.Errorf("ListObjectsV2: got %d %v; want 200 [keep]", code, keys)
+	}
+	results := make(chan itemOrErr[ObjectInfo], 16)
+	if err := z.Walk(t.Context(), bucket, "", results, WalkOptions{}); err != nil {
+		t.Fatal(err)
+	}
+	var names []string
+	for result := range results {
+		if result.Err != nil {
+			t.Fatal(result.Err)
+		}
+		names = append(names, result.Item.Name)
+	}
+	if !slices.Equal(names, []string{"keep"}) {
+		t.Errorf("Walk: got %v; want [keep]", names)
+	}
+
+	rec := nullQuorumRequest(t, router, http.MethodGet, getListObjectVersionsURL("", bucket, "", "1000", ""))
+	var list struct {
+		Versions []struct {
+			Key string `xml:"Key"`
+		} `xml:"Version"`
+	}
+	if rec.Code != http.StatusOK {
+		t.Errorf("ListObjectVersions: got %d; want 200", rec.Code)
+	} else if err := xml.Unmarshal(rec.Body.Bytes(), &list); err != nil {
+		t.Fatal(err)
+	} else if len(list.Versions) != 1 || list.Versions[0].Key != "keep" {
+		t.Errorf("ListObjectVersions: got %+v; want [keep]", list.Versions)
+	}
+
+	if _, err := z.DeleteObject(t.Context(), bucket, "keep", ObjectOptions{}); err != nil {
+		t.Fatal(err)
+	}
+	if err := z.DeleteBucket(t.Context(), bucket, DeleteBucketOptions{}); err != nil {
+		t.Errorf("DeleteBucket with only a stale minority: %v", err)
+	}
+}
+
+func TestIAMListingIgnoresStaleMinority(t *testing.T) {
+	z, _, _ := nullQuorumBackend(t)
+	body := []byte(`{"version":1}`)
+	alive := iamConfigUsersPrefix + "alive/identity.json"
+	ghost := iamConfigUsersPrefix + "ghost/identity.json"
+	nullQuorumPut(t, z, minioMetaBucket, alive, body, ObjectOptions{})
+	nullQuorumPut(t, z, minioMetaBucket, ghost, body, ObjectOptions{})
+	for _, disk := range z.serverPools[0].getHashedSet(ghost).getDisks()[4:] {
+		if err := disk.Delete(t.Context(), minioMetaBucket, ghost, DeleteOptions{Recursive: true, Immediate: true}); err != nil {
+			t.Fatal(err)
+		}
+	}
+
+	items, err := newIAMObjectStore(z, MinIOUsersSysType).listAllIAMConfigItems(t.Context())
+	if err != nil {
+		t.Fatalf("IAM listing failed on a stale minority (retriable=%v): %v", configRetriableErrors(err), err)
+	}
+	if got := items["users/"]; !slices.Equal(got, []string{"alive/identity.json"}) {
+		t.Errorf("IAM users: got %v; want [alive/identity.json]", got)
+	}
+}
+
+func TestListSplitDirectoryObject(t *testing.T) {
+	z, bucket, router := nullQuorumBackend(t)
+	set := z.serverPools[0].sets[0]
+	disks := set.getDisks()
+	const object = "dir/"
+	encoded := encodeDirObject(object)
+	mtime := time.Now().UTC().Add(-time.Hour)
+
+	nullQuorumPut(t, z, bucket, object, nil, ObjectOptions{MTime: mtime})
+	old := make([][]byte, 3)
+	for i := range old {
+		old[i] = mustReadNullQuorumMeta(t, disks[i], bucket, encoded)
+	}
+	nullQuorumPut(t, z, bucket, object, nil, ObjectOptions{MTime: mtime.Add(time.Minute)})
+	for i := range old {
+		if err := disks[i].WriteAll(t.Context(), bucket, encoded+"/"+xlStorageFormatFile, old[i]); err != nil {
+			t.Fatal(err)
+		}
+	}
+
+	walkers := slices.Clone(disks)
+	for i := 8; i < len(walkers); i++ {
+		walkers[i] = &failedMetadataWalkDisk{StorageAPI: disks[i]}
+	}
+	set.getDisks = func() []StorageAPI { return walkers }
+	if code, keys := nullQuorumListV2Keys(t, router, bucket); code != http.StatusOK || !slices.Equal(keys, []string{object}) {
+		t.Errorf("ListObjectsV2: got %d %v; want 200 [%s]", code, keys, object)
+	}
+}
+
+func TestListFallbackConfirmsCurrentAbsence(t *testing.T) {
+	z, bucket, _ := nullQuorumBackend(t)
+	set := z.serverPools[0].sets[0]
+	disks := set.getDisks()
+	const object = "object"
+	body := bytes.Repeat([]byte{'a'}, 8192)
+	mtime := time.Now().UTC().Add(-time.Hour)
+
+	nullQuorumPut(t, z, bucket, object, body, ObjectOptions{MTime: mtime})
+	old := make([][]byte, len(disks))
+	for i := range disks {
+		old[i] = mustReadNullQuorumMeta(t, disks[i], bucket, object)
+	}
+	nullQuorumPut(t, z, bucket, object, body, ObjectOptions{MTime: mtime.Add(time.Minute)})
+	newer := make([][]byte, len(disks))
+	for i := range disks {
+		newer[i] = mustReadNullQuorumMeta(t, disks[i], bucket, object)
+		if err := disks[i].Delete(t.Context(), bucket, object, DeleteOptions{Recursive: true, Immediate: true}); err != nil {
+			t.Fatal(err)
+		}
+	}
+
+	entries := make(metaCacheEntries, len(disks))
+	errs := make([]error, len(disks))
+	for i := range 3 {
+		entries[i] = metaCacheEntry{name: object, metadata: old[i]}
+	}
+	for i := 3; i < 8; i++ {
+		entries[i] = metaCacheEntry{name: object, metadata: newer[i]}
+	}
+	for i := 8; i < len(errs); i++ {
+		errs[i] = errDiskNotFound
+	}
+	resolver := metadataResolutionParams{dirQuorum: 8, objQuorum: 8, bucket: bucket, requestedVersions: 1}
+	entry, err := z.resolveListingGroup(t.Context(), bucket, []*metaCacheEntry{set.resolveListEntry(entries, errs, &resolver)})
+	if entry != nil || err != nil {
+		t.Fatalf("strict current absence: got entry=%v err=%v; want no entry or error", entry, err)
+	}
+}
+
+// G0 was written 8+8 while a node was away. G1 was later written 12+4 but
+// missed drive 0, which still holds G0. Then a node holding four G1 drives
+// goes away: eleven G1 and one G0 remain online. G1 exists but lacks one of
+// its twelve data shards until the node returns, so HEAD must not report 404
+// and LIST must still show the key.
+func TestListCrossParityMinorityStaysVisible(t *testing.T) {
+	z, bucket, router := nullQuorumBackend(t)
+	set := z.serverPools[0].sets[0]
+	disks, switchDisks := installListingDiskSwitch(set)
+	const object = "object"
+	mtime := time.Now().UTC().Add(-time.Hour)
+
+	nullQuorumPut(t, z, bucket, object, bytes.Repeat([]byte{'a'}, 256<<10), ObjectOptions{MTime: mtime, MaxParity: true})
+	g0 := mustReadNullQuorumMeta(t, disks[0], bucket, object)
+	var x xlMetaV2
+	if err := x.LoadOrConvert(g0); err != nil {
+		t.Fatal(err)
+	}
+	if h := x.versions[0].header; h.EcM != 8 || h.EcN != 8 {
+		t.Fatalf("G0 is %d+%d, want 8+8", h.EcM, h.EcN)
+	}
+	g1Body := bytes.Repeat([]byte{'b'}, 256<<10)
+	nullQuorumPut(t, z, bucket, object, g1Body, ObjectOptions{MTime: mtime.Add(time.Minute)})
+	if err := disks[0].WriteAll(t.Context(), bucket, object+"/"+xlStorageFormatFile, g0); err != nil {
+		t.Fatal(err)
+	}
+
+	online := slices.Clone(disks)
+	for i := 12; i < len(online); i++ {
+		online[i] = &offlineNullQuorumDisk{StorageAPI: disks[i]}
+	}
+	switchDisks(online)
+	if code := nullQuorumRequest(t, router, http.MethodHead, getHeadObjectURL("", bucket, object)).Code; code == http.StatusNotFound {
+		t.Errorf("HEAD with a node away: 404 for an existing object")
+	}
+	if code, keys := nullQuorumListV2Keys(t, router, bucket); code != http.StatusOK || !slices.Equal(keys, []string{object}) {
+		t.Errorf("ListObjectsV2 with a node away: got %d %v; want 200 [%s]", code, keys, object)
+	}
+
+	switchDisks(disks)
+	if get := nullQuorumRequest(t, router, http.MethodGet, getGetObjectURL("", bucket, object)); get.Code != http.StatusOK || !bytes.Equal(get.Body.Bytes(), g1Body) {
+		t.Fatalf("GET after the node returns: %d", get.Code)
+	}
+}
+
 func TestSingleNullQuorumReadAndConsumers(t *testing.T) {
 	z, bucket, router := nullQuorumBackend(t)
 	set := z.serverPools[0].sets[0]

cmd/metacache-parts-retry_test.go

@@ -0,0 +1,444 @@
+// Copyright (c) 2026 Feng Ruohang
+//
+// This file is part of Silo Object Storage stack
+//
+// This program is free software: you can redistribute it and/or modify
+// it under the terms of the GNU Affero General Public License as published by
+// the Free Software Foundation, either version 3 of the License, or
+// (at your option) any later version.
+//
+// This program is distributed in the hope that it will be useful
+// but WITHOUT ANY WARRANTY; without even the implied warranty of
+// MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
+// GNU Affero General Public License for more details.
+//
+// You should have received a copy of the GNU Affero General Public License
+// along with this program. If not, see <http://www.gnu.org/licenses/>.
+
+package cmd
+
+import (
+	"bytes"
+	"context"
+	"errors"
+	"fmt"
+	"io"
+	"slices"
+	"strings"
+	"sync"
+	"sync/atomic"
+	"testing"
+	"time"
+
+	"github.com/minio/minio/internal/grid"
+)
+
+// Read real cache objects and use an actual peer RPC, counting only the block
+// under test. Gates control publication, never replace quorum or EOF decisions.
+type listingCachePartsProbe struct {
+	state    atomic.Pointer[metacache]
+	rpcFails atomic.Bool
+	reads    atomic.Int64
+	mu       sync.Mutex
+	probes   []int
+	checks   []time.Time
+	onProbe  func(int) bool
+	onCheck  func(int)
+	target   string
+	readErr  error
+}
+
+func (p *listingCachePartsProbe) setState(o listPathOptions, failed bool) {
+	m := o.newMetacache()
+	m.lastUpdate = time.Now()
+	if failed {
+		m.status, m.error = scanStateError, "cache producer failed"
+	} else {
+		m.status = scanStateStarted
+	}
+	p.state.Store(&m)
+}
+
+func (p *listingCachePartsProbe) snapshot() ([]int, []time.Time) {
+	p.mu.Lock()
+	defer p.mu.Unlock()
+	return slices.Clone(p.probes), slices.Clone(p.checks)
+}
+
+type listingCachePartsDisk struct {
+	StorageAPI
+	index int
+	probe *listingCachePartsProbe
+}
+
+func (d *listingCachePartsDisk) ReadXL(ctx context.Context, bucket, object string, readData bool) (RawFileInfo, error) {
+	if bucket == minioMetaBucket && object == d.probe.target {
+		d.probe.reads.Add(1)
+		if d.probe.readErr != nil {
+			return RawFileInfo{}, d.probe.readErr
+		}
+	}
+	return d.StorageAPI.ReadXL(ctx, bucket, object, readData)
+}
+
+func (d *listingCachePartsDisk) ReadVersion(ctx context.Context, origVolume, volume, path, versionID string, opts ReadOptions) (FileInfo, error) {
+	if volume == minioMetaBucket && path == d.probe.target {
+		d.probe.mu.Lock()
+		d.probe.probes = append(d.probe.probes, d.index)
+		d.probe.mu.Unlock()
+		if d.probe.onProbe != nil && d.probe.onProbe(d.index) {
+			return FileInfo{}, errFileNotFound
+		}
+	}
+	return d.StorageAPI.ReadVersion(ctx, origVolume, volume, path, versionID, opts)
+}
+
+func listingCachePartsBackend(t *testing.T, part int) (*erasureObjects, listPathOptions, *listingCachePartsProbe, func([]StorageAPI)) {
+	t.Helper()
+	z, bucket, _ := nullQuorumBackend(t)
+	set := z.serverPools[0].sets[0]
+	o := listPathOptions{ID: mustGetUUID(), Bucket: bucket, Limit: 100, Recursive: true, IncludeDirectories: true}
+	p := &listingCachePartsProbe{target: o.objectPath(part)}
+	p.setState(o, false)
+	disks, switchDisks := installListingDiskSwitch(set)
+	wrapped := slices.Clone(disks)
+	for i, disk := range disks {
+		wrapped[i] = &listingCachePartsDisk{StorageAPI: disk, index: i, probe: p}
+	}
+	switchDisks(wrapped)
+
+	tg, err := grid.SetupTestGrid(2)
+	if err != nil {
+		t.Fatal(err)
+	}
+	previous := globalNotificationSys
+	t.Cleanup(func() {
+		globalNotificationSys = previous
+		tg.Cleanup()
+	})
+	if err := getMetacacheListingRPC.Register(tg.Managers[1], func(got *listPathOptions) (*metacache, *grid.RemoteErr) {
+		if got.ID != o.ID || got.Create {
+			return nil, grid.NewRemoteErr(errors.New("invalid cache state lookup"))
+		}
+		p.mu.Lock()
+		p.checks = append(p.checks, time.Now())
+		n := len(p.checks)
+		p.mu.Unlock()
+		if p.onCheck != nil {
+			p.onCheck(n)
+		}
+		if p.rpcFails.Load() {
+			return nil, grid.NewRemoteErr(errors.New("cache state RPC failed"))
+		}
+		m := *p.state.Load()
+		return &m, nil
+	}); err != nil {
+		t.Fatal(err)
+	}
+	client := &peerRESTClient{gridConn: func() *grid.Connection { return tg.Managers[0].Connection(tg.Hosts[1]) }}
+	globalNotificationSys = &NotificationSys{peerClients: []*peerRESTClient{client}, allPeerClients: []*peerRESTClient{client}}
+	return set, o, p, switchDisks
+}
+
+func putListingCachePart(t *testing.T, set *erasureObjects, o listPathOptions, part int, name string, eos bool) {
+	t.Helper()
+	var buf bytes.Buffer
+	w := newMetacacheWriter(&buf, 0)
+	if err := w.write(metaCacheEntry{name: name}); err != nil {
+		t.Fatal(err)
+	}
+	if err := w.Close(); err != nil {
+		t.Fatal(err)
+	}
+	block := metacacheBlock{n: part, First: name, Last: name, EOS: eos}
+	_, err := set.putMetacacheObject(t.Context(), o.objectPath(part), mustGetPutObjReader(t, bytes.NewReader(buf.Bytes()), int64(buf.Len()), "", ""), ObjectOptions{UserDefined: block.headerKV()})
+	if err != nil {
+		t.Fatal(err)
+	}
+}
+
+func TestStreamMetadataPartsHealthyNoStateRPC(t *testing.T) {
+	set, o, p, _ := listingCachePartsBackend(t, 1)
+	putListingCachePart(t, set, o, 0, "a/", false)
+	putListingCachePart(t, set, o, 1, "b/", true)
+	entries, err := set.streamMetadataParts(t.Context(), o)
+	if !errors.Is(err, io.EOF) || entries.len() != 2 || entries.o[0].name != "a/" || entries.o[1].name != "b/" {
+		t.Fatalf("healthy two-block read: entries=%v err=%v", entries.o, err)
+	}
+	probes, checks := p.snapshot()
+	if len(probes) != 0 || len(checks) != 0 {
+		t.Fatalf("healthy read added probes=%v state RPCs=%d", probes, len(checks))
+	}
+	t.Logf("healthy persisted blocks: state RPCs=0 retry probes=0 target ReadXL=%d", p.reads.Load())
+}
+
+func TestStreamMetadataPartsTerminalError(t *testing.T) {
+	for _, part := range []int{0, 1} {
+		t.Run(fmt.Sprintf("part=%d", part), func(t *testing.T) {
+			set, o, p, _ := listingCachePartsBackend(t, part)
+			if part > 0 {
+				putListingCachePart(t, set, o, 0, "a/", false)
+			}
+			p.setState(o, true)
+			start := time.Now()
+			entries, err := set.streamMetadataParts(t.Context(), o)
+			elapsed := time.Since(start)
+			if err == nil || errors.Is(err, io.EOF) || !strings.Contains(err.Error(), "cache producer failed") || entries.len() != part {
+				t.Fatalf("failed cache became successful: entries=%v err=%v", entries.o, err)
+			}
+			probes, checks := p.snapshot()
+			if len(checks) != 1 || len(probes) != 0 || elapsed > time.Second {
+				t.Fatalf("terminal error waited for disk retries: elapsed=%v state RPCs=%d probes=%v", elapsed, len(checks), probes)
+			}
+			t.Logf("terminal part=%d elapsed=%v state RPCs=%d probes=%d target ReadXL=%d", part, elapsed, len(checks), len(probes), p.reads.Load())
+		})
+	}
+}
+
+func TestStreamMetadataPartsLateTerminalError(t *testing.T) {
+	set, o, p, _ := listingCachePartsBackend(t, 1)
+	putListingCachePart(t, set, o, 0, "a/", false)
+	p.onProbe = func(int) bool {
+		p.setState(o, true)
+		return false
+	}
+	start := time.Now()
+	entries, err := set.streamMetadataParts(t.Context(), o)
+	elapsed := time.Since(start)
+	probes, checks := p.snapshot()
+	if err == nil || errors.Is(err, io.EOF) || !strings.Contains(err.Error(), "cache producer failed") || entries.len() != 1 {
+		t.Fatalf("late producer error lost: entries=%v err=%v", entries.o, err)
+	}
+	if elapsed > time.Second || len(probes) != 2 || !slices.Equal(probes, []int{0, 1}) || len(checks) != 2 {
+		t.Fatalf("late error retry work: elapsed=%v probes=%v state RPCs=%d", elapsed, probes, len(checks))
+	}
+	if checks[1].Sub(checks[0]) < 250*time.Millisecond || p.reads.Load() > 16 {
+		t.Fatalf("unbounded retry: state interval=%v target ReadXL=%d", checks[1].Sub(checks[0]), p.reads.Load())
+	}
+	t.Logf("late terminal elapsed=%v state RPCs=%d probes=%v target ReadXL=%d", elapsed, len(checks), probes, p.reads.Load())
+}
+
+type listingCachePartsResult struct {
+	entries metaCacheEntriesSorted
+	err     error
+}
+
+func startListingCacheParts(t *testing.T, set *erasureObjects, o listPathOptions) (<-chan listingCachePartsResult, context.CancelFunc) {
+	t.Helper()
+	ctx, cancel := context.WithCancel(t.Context())
+	done := make(chan listingCachePartsResult, 1)
+	joined := make(chan struct{})
+	go func() {
+		defer close(joined)
+		entries, err := set.streamMetadataParts(ctx, o)
+		done <- listingCachePartsResult{entries: entries, err: err}
+	}()
+	t.Cleanup(func() {
+		cancel()
+		select {
+		case <-joined:
+		case <-time.After(8 * time.Second):
+			t.Error("cache continuation worker did not finish")
+		}
+	})
+	return done, cancel
+}
+
+func receiveListingCacheParts(t *testing.T, done <-chan listingCachePartsResult) listingCachePartsResult {
+	t.Helper()
+	select {
+	case result := <-done:
+		return result
+	case <-time.After(8 * time.Second):
+		t.Fatal("cache continuation did not finish")
+		return listingCachePartsResult{}
+	}
+}
+
+func TestStreamMetadataPartsGeneratingThenPublished(t *testing.T) {
+	set, o, p, _ := listingCachePartsBackend(t, 1)
+	putListingCachePart(t, set, o, 0, "a/", false)
+	checked := make(chan struct{})
+	p.onCheck = func(n int) {
+		if n == 1 {
+			close(checked)
+		}
+	}
+	// The first selected replica is absent even after the producer publishes
+	// the block elsewhere. Rotation must reach a readable replica.
+	p.onProbe = func(index int) bool { return index == 0 }
+	done, _ := startListingCacheParts(t, set, o)
+	select {
+	case <-checked:
+	case <-time.After(5 * time.Second):
+		t.Fatal("missing block did not check producer state")
+	}
+	select {
+	case result := <-done:
+		t.Fatalf("still-generating cache finished early: %+v", result)
+	case <-time.After(100 * time.Millisecond):
+	}
+	putListingCachePart(t, set, o, 1, "b/", true)
+	result := receiveListingCacheParts(t, done)
+	if !errors.Is(result.err, io.EOF) || result.entries.len() != 2 || result.entries.o[0].name != "a/" || result.entries.o[1].name != "b/" {
+		t.Fatalf("published second block lost: entries=%v err=%v", result.entries.o, result.err)
+	}
+	probes, checks := p.snapshot()
+	if len(probes) < 2 || probes[0] != 0 || probes[1] != 1 {
+		t.Fatalf("retry did not rotate: probes=%v checks=%d", probes, len(checks))
+	}
+	t.Logf("later block recovered: state RPCs=%d probes=%v target ReadXL=%d", len(checks), probes, p.reads.Load())
+}
+
+func TestStreamMetadataPartsRPCFailure(t *testing.T) {
+	set, o, p, _ := listingCachePartsBackend(t, 1)
+	putListingCachePart(t, set, o, 0, "a/", false)
+	p.rpcFails.Store(true)
+	entries, err := set.streamMetadataParts(t.Context(), o)
+	if err == nil || errors.Is(err, io.EOF) || !strings.Contains(err.Error(), "cache state RPC failed") || entries.len() != 1 {
+		t.Fatalf("RPC failure became successful EOF: entries=%v err=%v", entries.o, err)
+	}
+	probes, checks := p.snapshot()
+	if len(checks) != 1 || len(probes) != 0 {
+		t.Fatalf("state RPC failure continued retrying: probes=%v checks=%d", probes, len(checks))
+	}
+}
+
+func TestStreamMetadataPartsMissingIndex(t *testing.T) {
+	set, o, p, _ := listingCachePartsBackend(t, 0)
+	putListingCachePart(t, set, o, 0, "a/", false)
+	o.Marker = "b/"
+	p.setState(o, true)
+	entries, err := set.streamMetadataParts(t.Context(), o)
+	if err == nil || errors.Is(err, io.EOF) || !strings.Contains(err.Error(), "cache producer failed") || entries.len() != 0 {
+		t.Fatalf("unpublished index became successful EOF: entries=%v err=%v", entries.o, err)
+	}
+	probes, checks := p.snapshot()
+	if len(checks) != 1 || len(probes) != 0 {
+		t.Fatalf("missing index delayed terminal state: probes=%v checks=%d", probes, len(checks))
+	}
+}
+
+func TestStreamMetadataPartsOtherErrorRetriesOnce(t *testing.T) {
+	set, o, p, _ := listingCachePartsBackend(t, 0)
+	p.readErr = errVolumeNotFound
+	entries, err := set.streamMetadataParts(t.Context(), o)
+	if err == nil || errors.Is(err, io.EOF) || !strings.Contains(err.Error(), "reading first part metadata:") || entries.len() != 0 {
+		t.Fatalf("unexpected metadata error result: entries=%v err=%v", entries.o, err)
+	}
+	probes, checks := p.snapshot()
+	if p.reads.Load() != 32 || len(probes) != 0 || len(checks) != 1 {
+		t.Fatalf("other metadata error did not get exactly one full retry: ReadXL=%d probes=%v checks=%d", p.reads.Load(), probes, len(checks))
+	}
+}
+
+func TestStreamMetadataPartsMissingBlockWorkBound(t *testing.T) {
+	set, o, p, _ := listingCachePartsBackend(t, 0)
+	checked := make(chan struct{})
+	p.onCheck = func(n int) {
+		if n == 3 {
+			close(checked)
+		}
+	}
+	start := time.Now()
+	done, cancel := startListingCacheParts(t, set, o)
+	select {
+	case <-checked:
+	case <-time.After(5 * time.Second):
+		t.Fatal("missing cache stopped checking producer state")
+	}
+	cancel()
+	result := receiveListingCacheParts(t, done)
+	probes, checks := p.snapshot()
+	if !errors.Is(result.err, context.Canceled) || result.entries.len() != 0 {
+		t.Fatalf("generating cache became successful EOF: %+v", result)
+	}
+	if len(probes) != 3 || !slices.Equal(probes, []int{0, 1, 2}) || p.reads.Load() != 16 || len(checks) != 3 {
+		t.Fatalf("missing block exceeded work bound: probes=%v target ReadXL=%d checks=%d", probes, p.reads.Load(), len(checks))
+	}
+	for i := 1; i < len(checks); i++ {
+		// Receive timestamps allow a small amount of peer scheduling jitter.
+		if checks[i].Sub(checks[i-1]) < 240*time.Millisecond {
+			t.Fatalf("cache state queried too often: %v", checks)
+		}
+	}
+	t.Logf("still missing for %v: state RPCs=%d probes=%v target ReadXL=%d", time.Since(start), len(checks), probes, p.reads.Load())
+}
+
+func TestStreamMetadataPartsCancelRetryWait(t *testing.T) {
+	set, o, p, _ := listingCachePartsBackend(t, 1)
+	putListingCachePart(t, set, o, 0, "a/", false)
+	probed := make(chan struct{})
+	p.onProbe = func(index int) bool {
+		if index == 1 {
+			close(probed)
+		}
+		return false
+	}
+	done, cancel := startListingCacheParts(t, set, o)
+	select {
+	case <-probed:
+	case <-time.After(5 * time.Second):
+		t.Fatal("retry probe did not run")
+	}
+	// Let the probe return and enter the regular 250 ms wait before canceling.
+	time.Sleep(20 * time.Millisecond)
+	start := time.Now()
+	cancel()
+	result := receiveListingCacheParts(t, done)
+	elapsed := time.Since(start)
+	if !errors.Is(result.err, context.Canceled) || elapsed > 150*time.Millisecond {
+		t.Fatalf("retry wait ignored cancellation: elapsed=%v err=%v", elapsed, result.err)
+	}
+	t.Logf("cancel during retry wait: elapsed=%v", elapsed)
+}
+
+func TestStreamMetadataPartsChangedDisks(t *testing.T) {
+	for _, empty := range []bool{false, true} {
+		t.Run(fmt.Sprintf("empty=%v", empty), func(t *testing.T) {
+			testStreamMetadataPartsChangedDisks(t, empty)
+		})
+	}
+}
+
+func testStreamMetadataPartsChangedDisks(t *testing.T, empty bool) {
+	set, o, p, switchDisks := listingCachePartsBackend(t, 1)
+	putListingCachePart(t, set, o, 0, "a/", false)
+	disks := set.getDisks()
+	probeDone, restore := make(chan struct{}), make(chan struct{})
+	p.onProbe = func(index int) bool {
+		if index == 0 {
+			// A later probe must safely skip an empty/offline disk snapshot.
+			if empty {
+				switchDisks([]StorageAPI{})
+			} else {
+				switchDisks([]StorageAPI{nil})
+			}
+			close(probeDone)
+			return true
+		}
+		return false
+	}
+	p.onCheck = func(n int) {
+		if n == 2 {
+			close(restore)
+		}
+	}
+	done, _ := startListingCacheParts(t, set, o)
+	select {
+	case <-probeDone:
+	case <-time.After(5 * time.Second):
+		t.Fatal("initial retry probe did not run")
+	}
+	select {
+	case <-restore:
+	case <-time.After(5 * time.Second):
+		t.Fatal("empty disk snapshot prevented cache state checks")
+	}
+	switchDisks(disks)
+	putListingCachePart(t, set, o, 1, "b/", true)
+	result := receiveListingCacheParts(t, done)
+	if !errors.Is(result.err, io.EOF) || result.entries.len() != 2 {
+		t.Fatalf("changed disks did not recover: entries=%v err=%v", result.entries.o, result.err)
+	}
+}

cmd/metacache-server-pool.go

@@ -191,19 +191,19 @@ func (z *erasureServerPools) listPath(ctx context.Context, o *listPathOptions) (
 			return entries, err
 		}
 		entries.truncate(0)
-		go func() {
+		rpc := globalNotificationSys.restClientFromHash(pathJoin(o.Bucket, o.Prefix))
-			rpc := globalNotificationSys.restClientFromHash(pathJoin(o.Bucket, o.Prefix))
+		go func(o listPathOptions) {
 			if rpc != nil {
 				ctx, cancel := context.WithTimeout(GlobalContext, 5*time.Second)
 				defer cancel()
-				c, err := rpc.GetMetacacheListing(ctx, *o)
+				c, err := rpc.GetMetacacheListing(ctx, o)
 				if err == nil {
 					c.error = "no longer used"
 					c.status = scanStateError
 					rpc.UpdateMetacacheListing(ctx, *c)
 				}
 			}
-		}()
+		}(*o)
 		o.ID = ""
 	}
 
@@ -264,6 +264,176 @@ func (z *erasureServerPools) listPathShortcutEOF(ctx context.Context, bucket str
 	return io.EOF
 }
 
+// listingRLock covers the lock domains used by normal same-key mutations.
+// Multi-pool writes use the pools lock. In a single pool, DELETE/COPY use the
+// first set while PUT and metadata updates use the destination set, so recovery
+// takes both in that order. No writer routing or wire protocol changes here.
+func (z *erasureServerPools) listingRLock(ctx context.Context, bucket, object string, source *erasureObjects) (LockContext, func(), error) {
+	firstSet := z.serverPools[0].sets[0]
+	first := z.NewNSLock(bucket, object)
+	firstCtx, err := first.GetRLock(ctx, globalListingOperationTimeout)
+	if err != nil {
+		return LockContext{}, nil, err
+	}
+	if !z.SinglePool() || source == firstSet || (!source.nsMutex.isDistErasure && source.nsMutex == firstSet.nsMutex) {
+		return firstCtx, func() { first.RUnlock(firstCtx) }, nil
+	}
+	// Shared nsMutex pointers do not prove shared distributed locker sets.
+	second := source.NewNSLock(bucket, object)
+	secondCtx, err := second.GetRLock(firstCtx.Context(), globalListingOperationTimeout)
+	if err != nil {
+		first.RUnlock(firstCtx)
+		return LockContext{}, nil, err
+	}
+	return secondCtx, func() {
+		second.RUnlock(secondCtx)
+		first.RUnlock(firstCtx)
+	}, nil
+}
+
+// resolveListingGroup recovers a complete same-name group, never a private
+// per-set contribution. A strictly absent source requires current reads from
+// every pool, since another pool's earlier walk may predate a migration copy.
+func (z *erasureServerPools) resolveListingGroup(ctx context.Context, bucket string, group []*metaCacheEntry) (*metaCacheEntry, error) {
+	var source *erasureObjects
+	for _, entry := range group {
+		if entry.listingSet != nil {
+			source = entry.listingSet
+			break
+		}
+	}
+	if source == nil {
+		return nil, errInvalidArgument
+	}
+	// Recovery consumes this group. Return reusable walk buffers only after
+	// their contents have been copied into the newly encoded result.
+	defer func() {
+		for _, entry := range group {
+			if entry.reusable {
+				metaDataPoolPut(entry.metadata)
+			}
+		}
+	}()
+	name := group[0].name
+	object := encodeDirObject(name)
+	if z.SinglePool() {
+		target := z.serverPools[0].getHashedSet(object)
+		for _, entry := range group {
+			if entry.listingSet != nil && entry.listingSet != target {
+				// Unexpected placement is not evidence of absence at the
+				// current write target. Do not extend recovery to arbitrary sets.
+				return nil, listingReadQuorumError(bucket, object)
+			}
+		}
+	}
+	type readResult struct {
+		fi     FileInfo
+		absent bool
+	}
+	reads := make(map[*erasureObjects]readResult)
+	currentOnly := false
+	var current *FileInfo
+	err := func() error {
+		lockCtx, cancel := context.WithTimeout(ctx, 5*time.Second)
+		defer cancel()
+		lkctx, unlock, err := z.listingRLock(lockCtx, bucket, object, source)
+		if err != nil {
+			return err
+		}
+		defer unlock()
+		read := func(set *erasureObjects) (readResult, error) {
+			if result, ok := reads[set]; ok {
+				return result, nil
+			}
+			fi, absent, err := set.readListingFileInfo(lkctx.Context(), bucket, object)
+			result := readResult{fi: fi, absent: absent}
+			if err == nil {
+				reads[set] = result
+			}
+			return result, err
+		}
+		for _, entry := range group {
+			if entry.listingSet == nil {
+				continue
+			}
+			result, err := read(entry.listingSet)
+			if err != nil {
+				return err
+			}
+			currentOnly = currentOnly || result.absent
+		}
+		if !currentOnly {
+			return lkctx.Context().Err()
+		}
+		// Migration publishes the replacement through the pooled write lock
+		// before source cleanup. While this read lock is held, new publication
+		// cannot pass these reads; any concurrent cleanup already has a copy.
+		// Unknown pools still fail rather than being counted as absent.
+		for _, pool := range z.serverPools {
+			result, err := read(pool.getHashedSet(object))
+			if err != nil {
+				return err
+			}
+			if result.absent {
+				continue
+			}
+			// Match latest GET: newest ModTime, then lowest pool index.
+			if current == nil || result.fi.ModTime.After(current.ModTime) {
+				fi := result.fi
+				current = &fi
+			}
+		}
+		return lkctx.Context().Err()
+	}()
+	if err != nil {
+		return nil, err
+	}
+	// All FileInfo values above are owned decoded results. Encoding and merge
+	// allocation happen after RUnlock, as do all stream sends and cache RPCs.
+	encode := func(fi FileInfo) (*metaCacheEntry, error) {
+		xl := &xlMetaV2{}
+		if err := xl.AddVersion(fi); err != nil {
+			return nil, listingReadQuorumError(bucket, object)
+		}
+		entry := &metaCacheEntry{name: name, cached: xl, reusable: true}
+		var err error
+		entry.metadata, err = xl.AppendTo(metaDataPoolGet())
+		return entry, err
+	}
+	if currentOnly {
+		if current == nil {
+			return nil, nil
+		}
+		// Replace the whole group: no stale live copy or marker can override
+		// this current global decision.
+		return encode(*current)
+	}
+	versions := make([][]xlMetaV2ShallowVersion, 0, len(group))
+	for _, entry := range group {
+		if entry.listingSet != nil {
+			fi := reads[entry.listingSet].fi
+			xl := &xlMetaV2{}
+			if err := xl.AddVersion(fi); err != nil {
+				return nil, listingReadQuorumError(bucket, object)
+			}
+			versions = append(versions, xl.versions)
+			continue
+		}
+		xl, err := entry.xlmeta()
+		if err != nil {
+			return nil, listingReadQuorumError(bucket, object)
+		}
+		versions = append(versions, xl.versions)
+	}
+	// Preserve the existing cross-pool merge rule for recovered live metadata
+	// and delete markers when no source was strictly absent.
+	xl := &xlMetaV2{versions: mergeXLV2Versions(1, true, 0, versions...)}
+	// Do not retain shallow versions referencing reusable input buffers.
+	resolved := &metaCacheEntry{name: name, reusable: true}
+	resolved.metadata, err = xl.AppendTo(metaDataPoolGet())
+	return resolved, err
+}
+
 // listMerged will list across all sets and return a merged results stream.
 // The result channel is closed when no more results are expected.
 func (z *erasureServerPools) listMerged(ctx context.Context, o listPathOptions, results chan<- metaCacheEntry) error {
@@ -285,11 +455,14 @@ func (z *erasureServerPools) listMerged(ctx context.Context, o listPathOptions,
 				defer wg.Done()
 				err := set.listPath(listCtx, o, innerResults)
 				mu.Lock()
-				defer mu.Unlock()
 				if err == nil {
 					allAtEOF = false
 				}
 				errs[i] = err
+				mu.Unlock()
+				// The merger must observe the terminal error before treating
+				// this stream as exhausted and advancing to another pool.
+				xioutil.SafeClose(innerResults)
 			}(len(errs), set)
 			errs = append(errs, nil)
 		}
@@ -298,7 +471,18 @@ func (z *erasureServerPools) listMerged(ctx context.Context, o listPathOptions,
 
 	// Gather results to a single channel.
 	// Quorum is one since we are merging across sets.
-	err := mergeEntryChannels(ctx, inputs, results, 1)
+	err := mergeEntryChannelsWithResolver(ctx, inputs, results, 1, func(i int) error {
+		mu.Lock()
+		err := errs[i]
+		mu.Unlock()
+		// Keep the existing aggregate handling of absent paths and buckets.
+		if errors.Is(err, io.EOF) || isAllNotFound([]error{err}) {
+			return nil
+		}
+		return err
+	}, func(group []*metaCacheEntry) (*metaCacheEntry, error) {
+		return z.resolveListingGroup(ctx, o.Bucket, group)
+	})
 
 	cancelList()
 	wg.Wait()
@@ -415,13 +599,24 @@ func (z *erasureServerPools) listAndSave(ctx context.Context, o *listPathOptions
 		cancel()
 	}()
 
-	// Do listing...
+	// Publish the typed terminal result before either consumer observes EOF.
+	// mergeEntryChannels closes inCh before listMerged has inspected set errors.
+	listingDone := make(chan struct{})
+	var cacheErrorDone chan struct{}
+	var listingErr error
 	go func(o listPathOptions) {
-		err := z.listMerged(listCtx, o, inCh)
+		listingErr = z.listMerged(listCtx, o, inCh)
-		if err != nil {
+		failed := listingErr != nil && !errors.Is(listingErr, io.EOF)
-			meta.setErr(err.Error())
+		if failed {
+			cacheErrorDone = make(chan struct{})
 		}
-		o.debugln("listAndSave: listing", o.ID, "finished with ", err)
+		close(listingDone)
+		if failed {
+			meta.setErr(listingErr.Error())
+			cancel()
+			close(cacheErrorDone)
+		}
+		o.debugln("listAndSave: listing", o.ID, "finished with ", listingErr)
 	}(*o)
 
 	// Keep track of when we return since we no longer have to send entries to output.
@@ -451,16 +646,33 @@ func (z *erasureServerPools) listAndSave(ctx context.Context, o *listPathOptions
 			entry.reusable = returned
 			saveCh <- entry
 		}
+		// The typed scan result is enough to finish the HTTP response.
+		<-listingDone
 		if !returned {
 			xioutil.SafeClose(outCh)
 		}
+		// Closing saveCh writes an end-of-stream cache block. On failure,
+		// wait until the error is recorded and the writer is canceled.
+		if cacheErrorDone != nil {
+			<-cacheErrorDone
+		}
 		xioutil.SafeClose(saveCh)
 	}()
 
 	entries, err = filteredResults()
-	if err == nil {
+	if err == nil || errors.Is(err, io.EOF) {
-		// Check if listing recorded an error.
+		// Full pages may return before the scan ends. A final page has
-		err = meta.getErr()
+		// crossed the completion barrier above and must preserve scan errors.
+		select {
+		case <-listingDone:
+			if listingErr != nil && !errors.Is(listingErr, io.EOF) {
+				return entries, listingErr
+			}
+		default:
+		}
+		if cacheErr := meta.getErr(); cacheErr != nil {
+			return entries, cacheErr
+		}
 	}
 	return entries, err
 }

cmd/metacache-set.go

@@ -39,7 +39,6 @@ import (
 	"github.com/minio/minio/internal/bucket/versioning"
 	"github.com/minio/minio/internal/color"
 	"github.com/minio/minio/internal/hash"
-	xioutil "github.com/minio/minio/internal/ioutil"
 	"github.com/pgsty/silo-pkg/v3/console"
 )
 
@@ -423,63 +422,89 @@ func (er *erasureObjects) streamMetadataParts(ctx context.Context, o listPathOpt
 		retryDelay    = 50 * time.Millisecond
 		retryDelay250 = 250 * time.Millisecond
 	)
+	var lastStateCheck time.Time
+	retryWait := func() error {
+		if contextCanceled(ctx) {
+			return ctx.Err()
+		}
+		// A failed producer cannot publish another block. Check on the first
+		// failure, then at most once per retry interval while it is still running.
+		if time.Since(lastStateCheck) >= retryDelay250 {
+			lastStateCheck = time.Now()
+			if err := o.checkMetacacheState(ctx, rpc); err != nil {
+				return fmt.Errorf("remote listing canceled: %w", err)
+			}
+		}
+		retries++
+		delay := retryDelay250
+		if retries == 1 {
+			delay = retryDelay
+		}
+		timer := time.NewTimer(delay)
+		defer timer.Stop()
+		select {
+		case <-ctx.Done():
+			return ctx.Err()
+		case <-timer.C:
+			return nil
+		}
+	}
+	// Probe one online disk per retry, rotating so a missing replica cannot
+	// pin the reader to one disk. A miss only delays the next probe: it never
+	// establishes absence or replaces the quorum read after a probe succeeds.
+	// This bounds missing-block I/O without sleeping once for every disk.
+	nextProbeDisk := 0
+	probePart := func(part int) bool {
+		disks := er.getDisks()
+		for range disks {
+			if contextCanceled(ctx) {
+				return false
+			}
+			i := nextProbeDisk % len(disks)
+			nextProbeDisk = i + 1
+			disk := disks[i]
+			if disk == nil || !disk.IsOnline() {
+				continue
+			}
+			_, err := disk.ReadVersion(ctx, "", minioMetaBucket, o.objectPath(part), "", ReadOptions{})
+			return err == nil
+		}
+		return false
+	}
 
+	retryOtherError := false
 	for {
 		if contextCanceled(ctx) {
 			return entries, ctx.Err()
 		}
 
-		// If many failures, check the cache state.
-		if retries > 10 {
-			err := o.checkMetacacheState(ctx, rpc)
-			if err != nil {
-				return entries, fmt.Errorf("remote listing canceled: %w", err)
-			}
-			retries = 1
-		}
-
 		// All operations are performed without locks, so we must be careful and allow for failures.
-		// Read metadata associated with the object from a disk.
+		if retries > 0 && !retryOtherError && !probePart(0) {
-		if retries > 0 {
+			if err := retryWait(); err != nil {
-			for _, disk := range er.getDisks() {
+				return entries, err
-				if disk == nil {
-					continue
-				}
-				if !disk.IsOnline() {
-					continue
-				}
-				_, err := disk.ReadVersion(ctx, "", minioMetaBucket,
-					o.objectPath(0), "", ReadOptions{})
-				if err != nil {
-					time.Sleep(retryDelay250)
-					retries++
-					continue
-				}
-				break
-			}
-		}
-		retryWait := func() {
-			retries++
-			if retries == 1 {
-				time.Sleep(retryDelay)
-			} else {
-				time.Sleep(retryDelay250)
 			}
+			continue
 		}
 		// Load first part metadata...
 		// Read metadata associated with the object from all disks.
+		retryOtherError = false
 		fi, metaArr, onlineDisks, err := er.getObjectFileInfo(ctx, minioMetaBucket, o.objectPath(0), ObjectOptions{}, true)
 		if err != nil {
 			switch toObjectErr(err, minioMetaBucket, o.objectPath(0)).(type) {
 			case ObjectNotFound, InsufficientReadQuorum:
-				retryWait()
+				if err := retryWait(); err != nil {
+					return entries, err
+				}
 				continue
 			}
 			// Allow one fast retry for other errors.
 			if retries > 0 {
 				return entries, fmt.Errorf("reading first part metadata: %v", err)
 			}
-			retryWait()
+			retryOtherError = true
+			if err := retryWait(); err != nil {
+				return entries, err
+			}
 			continue
 		}
 
@@ -487,14 +512,9 @@ func (er *erasureObjects) streamMetadataParts(ctx context.Context, o listPathOpt
 		switch {
 		case err == nil:
 		case errors.Is(err, io.ErrUnexpectedEOF):
-			if retries == 10 {
+			if err := retryWait(); err != nil {
-				err := o.checkMetacacheState(ctx, rpc)
+				return entries, err
-				if err != nil {
-					return entries, fmt.Errorf("remote listing canceled: %w", err)
-				}
-				retries = -1
 			}
-			retryWait()
 			continue
 		case errors.Is(err, io.EOF):
 			return entries, io.EOF
@@ -508,39 +528,19 @@ func (er *erasureObjects) streamMetadataParts(ctx context.Context, o listPathOpt
 			}
 
 			if partN != loadedPart {
-				if retries > 10 {
+				if retries > 0 && !probePart(partN) {
-					err := o.checkMetacacheState(ctx, rpc)
+					if err := retryWait(); err != nil {
-					if err != nil {
+						return entries, err
-						return entries, fmt.Errorf("waiting for next part %d: %w", partN, err)
-					}
-					retries = 1
-				}
-
-				if retries > 0 {
-					// Load from one disk only
-					for _, disk := range er.getDisks() {
-						if disk == nil {
-							continue
-						}
-						if !disk.IsOnline() {
-							continue
-						}
-						_, err := disk.ReadVersion(ctx, "", minioMetaBucket,
-							o.objectPath(partN), "", ReadOptions{})
-						if err != nil {
-							time.Sleep(retryDelay250)
-							retries++
-							continue
-						}
-						break
 					}
+					continue
 				}
 
 				// Load partN metadata...
 				fi, metaArr, onlineDisks, err = er.getObjectFileInfo(ctx, minioMetaBucket, o.objectPath(partN), ObjectOptions{}, true)
 				if err != nil {
-					time.Sleep(retryDelay250)
+					if err := retryWait(); err != nil {
-					retries++
+						return entries, err
+					}
 					continue
 				}
 				loadedPart = partN
@@ -576,13 +576,15 @@ func (er *erasureObjects) streamMetadataParts(ctx context.Context, o listPathOpt
 			if err != nil && !errors.Is(err, io.EOF) {
 				switch toObjectErr(err, minioMetaBucket, o.objectPath(partN)).(type) {
 				case ObjectNotFound:
-					retries++
+					if err := retryWait(); err != nil {
-					time.Sleep(retryDelay250)
+						return entries, err
+					}
 					continue
 				case InsufficientReadQuorum:
-					retries++
 					loadedPart = -1
-					time.Sleep(retryDelay250)
+					if err := retryWait(); err != nil {
+						return entries, err
+					}
 					continue
 				default:
 					internalLogIf(ctx, err)
@@ -700,9 +702,46 @@ func getQuorumDisks(disks []StorageAPI, infos []DiskInfo, readQuorum int) (newDi
 	return newDisks
 }
 
+// resolveListEntry keeps the existing walk resolution and fallback eligibility.
+// Eligible latest-only disagreements are resolved at the pool merger, where a
+// current absence can replace the entire same-name group, including stale walks.
+func (er *erasureObjects) resolveListEntry(entries metaCacheEntries, errs []error, resolver *metadataResolutionParams) *metaCacheEntry {
+	entry, ok := entries.resolve(resolver)
+	if ok {
+		return entry
+	}
+	if resolver.requestedVersions != 1 {
+		return nil
+	}
+	valid := 0
+	for i := range entries {
+		if !entries[i].isObject() {
+			continue
+		}
+		if _, err := entries[i].xlmeta(); err != nil {
+			continue
+		}
+		valid++
+		if entry == nil {
+			entry = &entries[i]
+		}
+	}
+	for _, err := range errs {
+		if err != nil {
+			valid++
+		}
+	}
+	if entry == nil || valid < resolver.objQuorum {
+		return nil
+	}
+	// This entry exists only in the internal set-to-pool stream. Do not retain
+	// an arbitrary walk snapshot or encode the pending state as metadata.
+	return &metaCacheEntry{name: entry.name, listingSet: er}
+}
+
 // Will return io.EOF if continuing would not yield more results.
 func (er *erasureObjects) listPath(ctx context.Context, o listPathOptions, results chan<- metaCacheEntry) (err error) {
-	defer xioutil.SafeClose(results)
+	// The caller records the terminal error before closing results.
 	o.debugf(color.Green("listPath:")+" with options: %#v", o)
 
 	// get prioritized non-healing disks for listing
@@ -778,8 +817,8 @@ func (er *erasureObjects) listPath(ctx context.Context, o listPathOptions, resul
 		},
 		partial: func(entries metaCacheEntries, errs []error) {
 			// Results Disagree :-(
-			entry, ok := entries.resolve(&resolver)
+			entry := er.resolveListEntry(entries, errs, &resolver)
-			if ok {
+			if entry != nil {
 				select {
 				case <-ctxDone:
 				case results <- *entry:
@@ -809,7 +848,9 @@ func (m *metaCacheRPC) setErr(err string) {
 		// An error is already set.
 		return
 	}
-	meta, _ = m.o.updateMetacacheListing(meta, m.rpc)
+	if updated, err := m.o.updateMetacacheListing(meta, m.rpc); err == nil {
+		meta = updated
+	}
 	*m.meta = meta
 }
 

cmd/metacache-stream.go

@@ -126,6 +126,9 @@ func (w *metacacheWriter) write(objs ...metaCacheEntry) error {
 		}
 	}
 	for _, o := range objs {
+		if o.listingSet != nil {
+			return errors.New("metacacheWriter: unresolved listing entry")
+		}
 		if len(o.name) == 0 {
 			return errors.New("metacacheWriter: no name provided")
 		}
@@ -170,6 +173,9 @@ func (w *metacacheWriter) stream() (chan<- metaCacheEntry, error) {
 	go func() {
 		defer w.streamWg.Done()
 		for o := range objs {
+			if o.listingSet != nil && w.streamErr == nil {
+				w.streamErr = errors.New("metacacheWriter: unresolved listing entry")
+			}
 			if len(o.name) == 0 || w.streamErr != nil {
 				continue
 			}

cmd/object-api-errors.go

@@ -834,6 +834,9 @@ func (de DataMovementOverwriteErr) Error() string {
 	if de.VersionID != "" {
 		objInfoStr = fmt.Sprintf("%s version-id=%s", objInfoStr, de.VersionID)
 	}
+	if de.Err == errDataMovementSuperseded {
+		return fmt.Sprintf("data movement skipped, another pool holds a newer copy of %s", objInfoStr)
+	}
 	return fmt.Sprintf("invalid data movement operation, source and destination pool are the same for %s", objInfoStr)
 }
 
@@ -841,3 +844,11 @@ func isDataMovementOverWriteErr(err error) bool {
 	var de DataMovementOverwriteErr
 	return errors.As(err, &de)
 }
+
+// isDataMovementSuperseded reports a version that was not moved because
+// another pool already holds a newer copy, which reads return. The source copy
+// is a stale duplicate: the move is complete, and the source may be removed.
+func isDataMovementSuperseded(err error) bool {
+	var de DataMovementOverwriteErr
+	return errors.As(err, &de) && de.Err == errDataMovementSuperseded
+}

cmd/xl-storage-format-v2-cross-parity_test.go

@@ -0,0 +1,222 @@
+// Copyright (c) 2026 Feng Ruohang
+//
+// This file is part of Silo Object Storage stack
+//
+// This program is free software: you can redistribute it and/or modify
+// it under the terms of the GNU Affero General Public License as published by
+// the Free Software Foundation, either version 3 of the License, or
+// (at your option) any later version.
+//
+// This program is distributed in the hope that it will be useful
+// but WITHOUT ANY WARRANTY; without even the implied warranty of
+// MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+// GNU Affero General Public License for more details.
+//
+// You should have received a copy of the GNU Affero General Public License
+// along with this program.  If not, see <http://www.gnu.org/licenses/>.
+
+package cmd
+
+import (
+	"bytes"
+	"fmt"
+	"net/http"
+	"slices"
+	"testing"
+	"time"
+)
+
+func crossParityNullVersion(t *testing.T, generation int64, data, parity int) xlMetaV2ShallowVersion {
+	t.Helper()
+	fi := newFileInfo("bucket/object", data, parity)
+	fi.Erasure.Index = 1
+	fi.DataDir = "11111111-1111-1111-1111-111111111111"
+	fi.ModTime = time.Unix(generation, 0).UTC()
+	fi.Size = 1024
+	fi.Parts = []ObjectPartInfo{{Number: 1, Size: fi.Size, ActualSize: fi.Size}}
+	fi.Metadata = map[string]string{"etag": fmt.Sprintf("%032d", generation)}
+	var xl xlMetaV2
+	if err := xl.AddVersion(fi); err != nil {
+		t.Fatal(err)
+	}
+	return xl.versions[0]
+}
+
+// A restarted node can retain four 12+4 shards while a successful degraded
+// write leaves twelve newer 8+8 shards. Neither disk order nor the minority's
+// EC must hide the independently sufficient newer group.
+func TestMergeXLV2LatestNullCrossParity(t *testing.T) {
+	versions := []xlMetaV2ShallowVersion{
+		crossParityNullVersion(t, 100, 12, 4),
+		crossParityNullVersion(t, 200, 8, 8),
+		crossParityNullVersion(t, 300, 10, 6),
+	}
+	for _, tt := range []struct {
+		name   string
+		counts [3]int
+		empty  int
+		quorum int
+		want   int
+	}{
+		{"new-twelve-old-four", [3]int{4, 12}, 0, 8, 1},
+		{"old-twelve-new-four", [3]int{12, 4}, 0, 8, 0},
+		{"new-nine-old-seven", [3]int{7, 9}, 0, 8, 1},
+		{"eight-meets-eight", [3]int{4, 8}, 4, 8, 1},
+		{"eight-does-not-meet-nine", [3]int{8, 8}, 0, 9, -1},
+		{"three-subquorums", [3]int{7, 5, 4}, 0, 8, -1},
+	} {
+		t.Run(tt.name, func(t *testing.T) {
+			var input [][]xlMetaV2ShallowVersion
+			for group, count := range tt.counts {
+				for range count {
+					input = append(input, []xlMetaV2ShallowVersion{versions[group]})
+				}
+			}
+			input = append(input, make([][]xlMetaV2ShallowVersion, tt.empty)...)
+			for order, streams := range quorumVersionOrders(input) {
+				got := mergeXLV2Versions(tt.quorum, false, 1, streams...)
+				if tt.want < 0 {
+					if len(got) != 0 {
+						t.Fatalf("order %d: subquorum groups were combined: %v", order, got)
+					}
+					continue
+				}
+				if len(got) != 1 || got[0].header != versions[tt.want].header {
+					t.Fatalf("order %d: got %v, want group %d", order, got, tt.want)
+				}
+				xl := xlMetaV2{versions: got}
+				fi, err := xl.ToFileInfo("bucket", "object", "", false, true)
+				if err != nil || !fi.IsValid() || fi.Erasure.DataBlocks != int(versions[tt.want].header.EcM) {
+					t.Fatalf("order %d: selected metadata is invalid: %+v, %v", order, fi, err)
+				}
+			}
+		})
+	}
+}
+
+func TestMergeXLV2LatestNullCrossParityScope(t *testing.T) {
+	old := crossParityNullVersion(t, 100, 12, 4)
+	newer := crossParityNullVersion(t, 200, 8, 8)
+	for _, kind := range []string{"strict", "all-versions", "uuid", "delete", "free", "history", "legacy-modern-ec", "partial-ec"} {
+		t.Run(kind, func(t *testing.T) {
+			a, b := old, newer
+			strict, requested := false, 1
+			switch kind {
+			case "strict":
+				strict = true
+			case "all-versions":
+				requested = 0
+			case "uuid":
+				a.header.VersionID, b.header.VersionID = [16]byte{1}, [16]byte{1}
+			case "delete":
+				b.header.Type = DeleteType
+			case "free":
+				b.header.Flags |= xlFlagFreeVersion
+			case "legacy-modern-ec":
+				b.header.EcN, b.header.EcM = 0, 0
+			case "partial-ec":
+				b.header.EcN = 0
+			}
+			// These previously unresolved inputs remain outside the repair.
+			// Strict and legacy-header cases use an older majority followed by
+			// a newer minority; the other cases use the opposite grouping.
+			input := make([][]xlMetaV2ShallowVersion, 16)
+			for i := range input {
+				input[i] = []xlMetaV2ShallowVersion{b}
+				if i >= 12 {
+					input[i] = []xlMetaV2ShallowVersion{a}
+				}
+			}
+			if kind == "strict" || kind == "legacy-modern-ec" || kind == "partial-ec" {
+				for i := range input {
+					input[i] = []xlMetaV2ShallowVersion{a}
+					if i >= 12 {
+						input[i] = []xlMetaV2ShallowVersion{b}
+					}
+				}
+			}
+			if kind == "history" {
+				history := old
+				history.header.VersionID = [16]byte{1}
+				input[0] = append(input[0], history)
+			}
+			if got := mergeXLV2Versions(8, strict, requested, input...); len(got) != 0 {
+				t.Fatalf("excluded input changed: %v", got)
+			}
+		})
+	}
+}
+
+func TestCountXLV2MixedECGroups(t *testing.T) {
+	a := crossParityNullVersion(t, 100, 12, 4)
+	b := crossParityNullVersion(t, 100, 8, 8)
+	c := crossParityNullVersion(t, 200, 10, 6)
+	// Same-time groups remain separate, and equal counts have a deterministic
+	// final EC tiebreak even though sortsBefore does not compare EC itself.
+	input := [][]xlMetaV2ShallowVersion{{a}, {a}, {a}, {b}, {b}, {b}, {c}}
+	for order, streams := range quorumVersionOrders(input) {
+		tops := make([]xlMetaV2ShallowVersion, len(streams))
+		for i := range streams {
+			tops[i] = streams[i][0]
+		}
+		for range 20 {
+			got, count := countXLV2Versions(tops, c.header, c, false, true)
+			if count != 3 || got.header != a.header {
+				t.Fatalf("order %d: got %v, count %d; want distinct group of three", order, got.header, count)
+			}
+		}
+	}
+	differentSig := a
+	differentSig.header.Signature[0]++
+	if _, count := countXLV2Versions([]xlMetaV2ShallowVersion{a, a, differentSig}, a.header, a, false, true); count != 3 {
+		t.Fatalf("non-strict signature groups did not combine: count %d", count)
+	}
+	// Same EC with different time or flags must not accumulate votes either.
+	flags := a
+	flags.header.Flags ^= xlFlagInlineData
+	for _, other := range []xlMetaV2ShallowVersion{crossParityNullVersion(t, 200, 12, 4), flags} {
+		got, count := countXLV2Versions([]xlMetaV2ShallowVersion{a, a, other, other}, a.header, a, false, true)
+		if count != 2 {
+			t.Fatalf("incompatible header groups combined: %v, count %d", got.header, count)
+		}
+	}
+}
+
+// Exercise the regular full-metadata read and HTTP LIST with actual on-disk
+// object metadata, not merely the metadata selection helper.
+func TestListCrossParityReadableMajority(t *testing.T) {
+	for _, size := range []int{1024, 256 << 10} {
+		t.Run(fmt.Sprintf("bytes-%d", size), func(t *testing.T) {
+			z, bucket, router := nullQuorumBackend(t)
+			set := z.serverPools[0].sets[0]
+			disks := set.getDisks()
+			const object = "object"
+			mtime := time.Now().UTC().Add(-time.Hour)
+			oldBody := bytes.Repeat([]byte{'a'}, size)
+			newBody := bytes.Repeat([]byte{'b'}, size)
+			nullQuorumPut(t, z, bucket, object, oldBody, ObjectOptions{MTime: mtime})
+			old := make([][]byte, len(disks))
+			for i, disk := range disks {
+				old[i] = mustReadNullQuorumMeta(t, disk, bucket, object)
+			}
+			nullQuorumPut(t, z, bucket, object, newBody, ObjectOptions{MTime: mtime.Add(time.Minute), MaxParity: true})
+			for i := 12; i < len(disks); i++ {
+				if err := disks[i].WriteAll(t.Context(), bucket, object+"/"+xlStorageFormatFile, old[i]); err != nil {
+					t.Fatal(err)
+				}
+			}
+			for _, readData := range []bool{false, true} {
+				fi, _, _, err := set.getObjectFileInfo(t.Context(), bucket, object, ObjectOptions{}, readData)
+				if err != nil || !fi.ModTime.Equal(mtime.Add(time.Minute)) || fi.Erasure.DataBlocks != 8 {
+					t.Fatalf("readData=%v: full metadata read lost readable majority: %+v, %v", readData, fi, err)
+				}
+			}
+			if get := nullQuorumRequest(t, router, http.MethodGet, getGetObjectURL("", bucket, object)); get.Code != http.StatusOK || !bytes.Equal(get.Body.Bytes(), newBody) {
+				t.Fatalf("GET lost readable majority: status %d", get.Code)
+			}
+			if code, keys := nullQuorumListV2Keys(t, router, bucket); code != http.StatusOK || !slices.Equal(keys, []string{object}) {
+				t.Fatalf("LIST lost readable majority: %d %v", code, keys)
+			}
+		})
+	}
+}

cmd/xl-storage-format-v2.go

@@ -1944,7 +1944,7 @@ func mergeXLV2Versions(quorum int, strict bool, requestedVersions int, versions
 	// Keep the original stream shapes: pruning must not make a versioned object
 	// eligible for the single null-version recount below.
 	originalVersions := versions
-	var checkedSingleNull, singleNull bool
+	var checkedSingleNull, singleNull, mixedECNull bool
 	// Shallow copy input
 	versions = append(make([][]xlMetaV2ShallowVersion, 0, len(versions)), versions...)
 
@@ -2019,19 +2019,24 @@ func mergeXLV2Versions(quorum int, strict bool, requestedVersions int, versions
 					// Version IDs match, but otherwise unable to resolve.
 					// We are either strict, or don't have enough information to match.
 					// Switch to a pure counting algo.
-					latest, latestCount = countXLV2Versions(tops, ver.header, latest, strict)
+					latest, latestCount = countXLV2Versions(tops, ver.header, latest, strict, false)
 					break
 				}
 			}
 			if latestCount < quorum {
 				if !checkedSingleNull {
-					singleNull = singleNullVersionStreams(originalVersions)
+					singleNull = singleNullVersionStreams(originalVersions, false)
+					// Availability-optimized writes can change EC while a drive
+					// is offline. Only latest-only non-strict reads may recount
+					// these ordinary null versions across separate modern EC groups.
+					mixedECNull = !singleNull && !strict && requestedVersions == 1 &&
+						singleNullVersionStreams(originalVersions, true)
 					checkedSingleNull = true
 				}
-				if singleNull {
+				if singleNull || mixedECNull {
 					// A newer minority at the end can hide an older quorum from
 					// the selection loop. Recount before discarding the null ID.
-					if candidate, count := countXLV2Versions(tops, latest.header, latest, strict); count >= quorum {
+					if candidate, count := countXLV2Versions(tops, latest.header, latest, strict, mixedECNull); count >= quorum {
 						latest, latestCount = candidate, count
 					}
 				}
@@ -2098,7 +2103,8 @@ func mergeXLV2Versions(quorum int, strict bool, requestedVersions int, versions
 
 // singleNullVersionStreams excludes histories and other version types from the
 // additional recount. Check the original inputs, before any stream is pruned.
-func singleNullVersionStreams(versions [][]xlMetaV2ShallowVersion) bool {
+// Mixed EC groups require known modern EC headers on every non-empty stream.
+func singleNullVersionStreams(versions [][]xlMetaV2ShallowVersion, allowMixedEC bool) bool {
 	var ec xlMetaV2VersionHeader
 	var haveEC bool
 	for _, stream := range versions {
@@ -2112,7 +2118,10 @@ func singleNullVersionStreams(versions [][]xlMetaV2ShallowVersion) bool {
 		if h.VersionID != [16]byte{} || h.Type != ObjectType || h.FreeVersion() {
 			return false
 		}
-		if haveEC && (h.EcN != ec.EcN || h.EcM != ec.EcM) {
+		if allowMixedEC && !h.hasEC() {
+			return false
+		}
+		if !allowMixedEC && haveEC && (h.EcN != ec.EcN || h.EcM != ec.EcM) {
 			return false
 		}
 		ec, haveEC = h, true
@@ -2121,16 +2130,18 @@ func singleNullVersionStreams(versions [][]xlMetaV2ShallowVersion) bool {
 }
 
 // countXLV2Versions selects the most frequent compatible header for reference's
-// VersionID, retaining the existing sort tiebreak and last matching entry.
+// VersionID and the last matching entry. The mixed-EC null-version recount adds
-func countXLV2Versions(tops []xlMetaV2ShallowVersion, reference xlMetaV2VersionHeader, latest xlMetaV2ShallowVersion, strict bool) (xlMetaV2ShallowVersion, int) {
+// a stable EC tiebreak without changing the ordinary counting path.
+func countXLV2Versions(tops []xlMetaV2ShallowVersion, reference xlMetaV2VersionHeader, latest xlMetaV2ShallowVersion, strict, allowMixedEC bool) (xlMetaV2ShallowVersion, int) {
 	x := make(map[xlMetaV2VersionHeader]int, len(tops))
 	for _, a := range tops {
 		if a.header.VersionID != reference.VersionID {
 			continue
 		}
 		if !strict {
-			// we must match EC, when we are not strict.
+			// The narrow null-version recount can consider other EC groups,
-			if !a.header.matchesEC(reference) {
+			// but EcN/EcM stay in the map key: their votes never combine.
+			if !allowMixedEC && !a.header.matchesEC(reference) {
 				continue
 			}
 			a.header.Signature = [4]byte{}
@@ -2142,9 +2153,19 @@ func countXLV2Versions(tops []xlMetaV2ShallowVersion, reference xlMetaV2VersionH
 		if v < latestCount {
 			continue
 		}
-		if v == latestCount && latest.header.sortsBefore(k) {
+		if v == latestCount {
-			// Tiebreak, use sort.
+			if allowMixedEC {
-			continue
+				current := latest.header
+				current.Signature = [4]byte{}
+				// Signatures are not part of these groups. Use EC counts as
+				// the final tiebreak when the existing sort fields are equal.
+				if current.sortsBefore(k) || !k.sortsBefore(current) &&
+					(current.EcM > k.EcM || current.EcM == k.EcM && current.EcN > k.EcN) {
+					continue
+				}
+			} else if latest.header.sortsBefore(k) {
+				continue
+			}
 		}
 		for _, a := range tops {
 			hdr := a.header

docs/debugging/reorder-disks/go.mod

@@ -2,4 +2,4 @@ module github.com/minio/minio/docs/debugging/reorder-disks
 
 go 1.26.0
 
-require github.com/pgsty/silo-pkg/v3 v3.13.2
+require github.com/pgsty/silo-pkg/v3 v3.14.3

docs/debugging/reorder-disks/go.sum

@@ -1,2 +1,2 @@
-github.com/pgsty/silo-pkg/v3 v3.13.2 h1:Clw11c/J54Tx6pijNCWtXiC7e0fwP/f5Tgeb6fsXg2w=
+github.com/pgsty/silo-pkg/v3 v3.14.3 h1:hMDxOo36F1whGRkYScrUGGwn98FvsmJ8hVM6pE4i8Bs=
-github.com/pgsty/silo-pkg/v3 v3.13.2/go.mod h1:0GmaDA0ArQ8bkAI/obiSNTBQzdgBu6W0e0olDCbyXXo=
+github.com/pgsty/silo-pkg/v3 v3.14.3/go.mod h1:zqIm3FMETmwPHRKuUExCDGfYkCNXt+vpfBnPM821RdI=

go.mod

@@ -4,9 +4,9 @@ go 1.27.2
 
 // Console and MC retain their historical module paths for best-effort upstream
 // compatibility. Pin the maintained PGSTY implementations used by SILO.
-replace github.com/minio/console => github.com/pgsty/silo-console v0.0.0-20261015132217-e5b165e4c0ee
+replace github.com/minio/console => github.com/pgsty/silo-console v0.0.0-20261017112417-ef2dd2c41d21
 
-replace github.com/minio/mc => github.com/pgsty/mc v0.0.0-20260916070421-e952aa78f10a
+replace github.com/minio/mc => github.com/pgsty/mc v0.0.0-20261017101407-fa3b8bf72458
 
 // v22.7.0 does not compile on NetBSD because its unix implementation uses
 // CLOCK_MONOTONIC, which is unavailable there. Keep the last portable release
@@ -41,7 +41,7 @@ require (
 	github.com/fatih/color v1.19.0
 	github.com/felixge/fgprof v0.9.5
 	github.com/fraugster/parquet-go v0.12.0
-	github.com/go-ldap/ldap/v3 v3.4.14
+	github.com/go-ldap/ldap/v3 v3.4.15
 	github.com/go-openapi/loads v0.25.2
 	github.com/go-sql-driver/mysql v1.9.3
 	github.com/gobwas/ws v1.4.0
@@ -50,8 +50,8 @@ require (
 	github.com/google/uuid v1.6.0
 	github.com/inconshreveable/mousetrap v1.1.0
 	github.com/json-iterator/go v1.1.12
-	github.com/klauspost/compress v1.20.0
+	github.com/klauspost/compress v1.20.3
-	github.com/klauspost/cpuid/v2 v2.4.0
+	github.com/klauspost/cpuid/v2 v2.4.1
 	github.com/klauspost/filepathx v1.1.1
 	github.com/klauspost/pgzip v1.2.6
 	github.com/klauspost/readahead v1.4.0
@@ -68,7 +68,7 @@ require (
 	github.com/minio/kms-go/kes v0.3.1
 	github.com/minio/kms-go/kms v0.6.0
 	github.com/minio/madmin-go/v3 v3.0.110
-	github.com/minio/minio-go/v7 v7.3.1-0.20260915093545-32e1f32cb176
+	github.com/minio/minio-go/v7 v7.4.1
 	github.com/minio/mux v1.10.1
 	github.com/minio/selfupdate v0.6.0
 	github.com/minio/simdjson-go v0.4.5
@@ -81,8 +81,8 @@ require (
 	github.com/nats-io/stan.go v0.10.4
 	github.com/ncw/directio v1.0.5
 	github.com/nsqio/go-nsq v1.1.0
-	github.com/pgsty/silo-pkg/v3 v3.14.1
+	github.com/pgsty/silo-pkg/v3 v3.14.3
-	github.com/philhofer/fwd v1.2.0
+	github.com/philhofer/fwd v1.2.1
 	github.com/pierrec/lz4/v4 v4.1.30
 	github.com/pkg/errors v0.9.1
 	github.com/pkg/sftp v1.13.11
@@ -97,21 +97,21 @@ require (
 	github.com/rs/cors v1.11.1
 	github.com/secure-io/sio-go v0.3.1
 	github.com/shirou/gopsutil/v3 v3.24.5
-	github.com/tinylib/msgp v1.6.4
+	github.com/tinylib/msgp v1.6.5
 	github.com/valyala/bytebufferpool v1.0.0
 	github.com/xdg/scram v1.0.5
 	github.com/zeebo/xxh3 v1.1.0
-	go.etcd.io/etcd/api/v3 v3.7.1
+	go.etcd.io/etcd/api/v3 v3.7.2
-	go.etcd.io/etcd/client/v3 v3.7.1
+	go.etcd.io/etcd/client/v3 v3.7.2
 	go.uber.org/atomic v1.11.0
 	go.uber.org/zap v1.28.0
 	go.yaml.in/yaml/v3 v3.0.5
 	goftp.io/server/v2 v2.0.3
-	golang.org/x/crypto v0.57.0
+	golang.org/x/crypto v0.58.0
 	golang.org/x/oauth2 v0.37.0
-	golang.org/x/sync v0.23.0
+	golang.org/x/sync v0.24.0
-	golang.org/x/sys v0.48.0
+	golang.org/x/sys v0.49.0
-	golang.org/x/term v0.46.0
+	golang.org/x/term v0.47.0
 	golang.org/x/time v0.16.0
 	google.golang.org/api v0.290.0
 	gopkg.in/yaml.v2 v2.4.0
@@ -164,7 +164,7 @@ require (
 	github.com/fatih/structs v1.1.0 // indirect
 	github.com/felixge/httpsnoop v1.1.0 // indirect
 	github.com/go-asn1-ber/asn1-ber v1.5.8 // indirect
-	github.com/go-jose/go-jose/v4 v4.1.4 // indirect
+	github.com/go-jose/go-jose/v4 v4.1.6 // indirect
 	github.com/go-logr/logr v1.4.4 // indirect
 	github.com/go-logr/stdr v1.2.2 // indirect
 	github.com/go-ole/go-ole v1.3.0 // indirect
@@ -282,7 +282,7 @@ require (
 	github.com/xdg/stringprep v1.0.3 // indirect
 	github.com/xo/terminfo v1.0.0 // indirect
 	github.com/yusufpapurcu/wmi v1.2.4 // indirect
-	go.etcd.io/etcd/client/pkg/v3 v3.7.1 // indirect
+	go.etcd.io/etcd/client/pkg/v3 v3.7.2 // indirect
 	go.opentelemetry.io/auto/sdk v1.2.1 // indirect
 	go.opentelemetry.io/contrib/detectors/gcp v1.44.0 // indirect
 	go.opentelemetry.io/contrib/instrumentation/google.golang.org/grpc/otelgrpc v0.70.0 // indirect
@@ -293,10 +293,10 @@ require (
 	go.opentelemetry.io/otel/sdk/metric v1.45.0 // indirect
 	go.opentelemetry.io/otel/trace v1.45.0 // indirect
 	go.uber.org/multierr v1.11.0 // indirect
-	golang.org/x/mod v0.41.0 // indirect
+	golang.org/x/mod v0.42.0 // indirect
-	golang.org/x/net v0.60.0 // indirect
+	golang.org/x/net v0.61.0 // indirect
-	golang.org/x/text v0.42.0 // indirect
+	golang.org/x/text v0.43.0 // indirect
-	golang.org/x/tools v0.50.0 // indirect
+	golang.org/x/tools v0.52.0 // indirect
 	google.golang.org/genproto v0.0.0-20260319201613-d00831a3d3e7 // indirect
 	google.golang.org/genproto/googleapis/api v0.0.0-20260831171406-18b4a7587f8a // indirect
 	google.golang.org/genproto/googleapis/rpc v0.0.0-20260831171406-18b4a7587f8a // indirect

go.sum

@@ -186,10 +186,10 @@ github.com/fraugster/parquet-go v0.12.0/go.mod h1:dGzUxdNqXsAijatByVgbAWVPlFirnh
 github.com/fsnotify/fsnotify v1.4.7/go.mod h1:jwhsz4b93w/PPRr/qN1Yymfu8t87LnFCMoQvtojpjFo=
 github.com/go-asn1-ber/asn1-ber v1.5.8 h1:H9AZkK22UOmfX8J84ubyaZxKJZ3FMHVwn8swoMML7iQ=
 github.com/go-asn1-ber/asn1-ber v1.5.8/go.mod h1:hEBeB/ic+5LoWskz+yKT7vGhhPYkProFKoKdwZRWMe0=
-github.com/go-jose/go-jose/v4 v4.1.4 h1:moDMcTHmvE6Groj34emNPLs/qtYXRVcd6S7NHbHz3kA=
+github.com/go-jose/go-jose/v4 v4.1.6 h1:sNRkftxFJxwsILiQukqp5z83QBSmntVwmgRY7AqBDO4=
-github.com/go-jose/go-jose/v4 v4.1.4/go.mod h1:x4oUasVrzR7071A4TnHLGSPpNOm2a21K9Kf04k1rs08=
+github.com/go-jose/go-jose/v4 v4.1.6/go.mod h1:x4oUasVrzR7071A4TnHLGSPpNOm2a21K9Kf04k1rs08=
-github.com/go-ldap/ldap/v3 v3.4.14 h1:D6PYdEgsaVzsXyr6w/yDC06Ria4uUhWm+Rb+er8lfAs=
+github.com/go-ldap/ldap/v3 v3.4.15 h1:eRB4f4Ecn1POL1mO86+uUDlecmyWsdiJwh60bAUS4UY=
-github.com/go-ldap/ldap/v3 v3.4.14/go.mod h1:S4eJUMUNjDkE0ZJtIZdybwyb03sGGLW6gxXT1Hs8VKA=
+github.com/go-ldap/ldap/v3 v3.4.15/go.mod h1:jeD0c98EdZQCKKKKe4kTWs5aCf6y+j8KuYsYLznZSQk=
 github.com/go-logr/logr v1.2.2/go.mod h1:jdQByPbusPIv2/zmleS9BjJVeZ6kBagPoEUsqbVz/1A=
 github.com/go-logr/logr v1.4.4 h1:tG4xh9yMsRCAiodLVTxyrkzSZ9+o0L1Kg/+cPVcbP/8=
 github.com/go-logr/logr v1.4.4/go.mod h1:9T104GzyrTigFIr8wt5mBrctHMim0Nb2HLGrmQ40KvY=
@@ -374,11 +374,11 @@ github.com/keybase/go-keychain v0.0.1/go.mod h1:PdEILRW3i9D8JcdM+FmY6RwkHGnhHxXw
 github.com/kisielk/errcheck v1.5.0/go.mod h1:pFxgyoBC7bSaBwPgfKdkLd5X25qrDl4LWUI2bnpBCr8=
 github.com/kisielk/gotool v1.0.0/go.mod h1:XhKaO+MFFWcvkIS/tQcRk01m1F5IRFswLeQ+oQHNcck=
 github.com/klauspost/compress v1.14.4/go.mod h1:/3/Vjq9QcHkK5uEr5lBEmyoZ1iFhe47etQ6QUkpK6sk=
-github.com/klauspost/compress v1.20.0 h1:a3C1ke2ohxFymNlb2HWAHjDeKCI90scRskErZkR0ezA=
+github.com/klauspost/compress v1.20.3 h1:zWNoFqq/lSuYQRim3dUXlXeA0IqwuZGicdG5ZTfY5j4=
-github.com/klauspost/compress v1.20.0/go.mod h1:LUdAzn7YLVvxLpc7y3V1m40wESHTgc1422pwwBSKYuI=
+github.com/klauspost/compress v1.20.3/go.mod h1:LUdAzn7YLVvxLpc7y3V1m40wESHTgc1422pwwBSKYuI=
 github.com/klauspost/cpuid/v2 v2.0.1/go.mod h1:FInQzS24/EEf25PyTYn52gqo7WaD8xa0213Md/qVLRg=
-github.com/klauspost/cpuid/v2 v2.4.0 h1:S6Hrbc7+ywsr0r+RLapfGBHfyefhCTwEh3A0tV913Dw=
+github.com/klauspost/cpuid/v2 v2.4.1 h1:I9KxadAkvo+QjtcEE2f2tyfEvmIla/T0m+14PrZNl0U=
-github.com/klauspost/cpuid/v2 v2.4.0/go.mod h1:19jmZ9mjzoF//ddRSUsv0zfBTJWh3QJh9FNxZTMrGxU=
+github.com/klauspost/cpuid/v2 v2.4.1/go.mod h1:19jmZ9mjzoF//ddRSUsv0zfBTJWh3QJh9FNxZTMrGxU=
 github.com/klauspost/crc32 v1.3.0 h1:sSmTt3gUt81RP655XGZPElI0PelVTZ6YwCRnPSupoFM=
 github.com/klauspost/crc32 v1.3.0/go.mod h1:D7kQaZhnkX/Y0tstFGf8VUzv2UofNGqCjnC3zdHB0Hw=
 github.com/klauspost/filepathx v1.1.1 h1:201zvAsL1PhZvmXTP+QLer3AavWrO3U1NILWpniHK4w=
@@ -474,8 +474,8 @@ github.com/minio/madmin-go/v3 v3.0.110 h1:FIYekj7YPc430ffpXFWiUtyut3qBt/unIAcDzJ
 github.com/minio/madmin-go/v3 v3.0.110/go.mod h1:WOe2kYmYl1OIlY2DSRHVQ8j1v4OItARQ6jGyQqcCud8=
 github.com/minio/md5-simd v1.1.2 h1:Gdi1DZK69+ZVMoNHRXJyNcxrMA4dSxoYHZSQbirFg34=
 github.com/minio/md5-simd v1.1.2/go.mod h1:MzdKDxYpY2BT9XQFocsiZf/NKVtR7nkE4RoEpN+20RM=
-github.com/minio/minio-go/v7 v7.3.1-0.20260915093545-32e1f32cb176 h1:TIrbAkJoMN/kbtV4LAitf/eidrbvvYcujXPbSe+4j/0=
+github.com/minio/minio-go/v7 v7.4.1 h1:PtU/zIOfQRWZ+QydXAZysc+xZYHdG+BBR4Z6TUvy73w=
-github.com/minio/minio-go/v7 v7.3.1-0.20260915093545-32e1f32cb176/go.mod h1:KUPWdecEO1LWyUz+sTGXAuf2jZHrPh5fCsRH86QbPfk=
+github.com/minio/minio-go/v7 v7.4.1/go.mod h1:nQnQi7zTfXR4lREtYXZQWqo5aw0Rqv+ZqCGrTORygRk=
 github.com/minio/mux v1.10.1 h1:grrK8SwRKbkNFE6qG7WAvFGH09bB46d5teOOtKfQ14s=
 github.com/minio/mux v1.10.1/go.mod h1:INYT4sMSTJy0QWUEA/E2DZNxJ5sAxIwbnyZjkzNFRfE=
 github.com/minio/pkg/v3 v3.6.1 h1:gaNT80BS/iuIany5ylTkVmfN4s6UYY30OtImFv4GQA8=
@@ -545,14 +545,14 @@ github.com/orisano/pixelmatch v0.0.0-20220722002657-fb0b55479cde/go.mod h1:nZgzb
 github.com/pascaldekloe/goe v0.1.0/go.mod h1:lzWF7FIEvWOWxwDKqyGYQf6ZUaNfKdP144TG7ZOy1lc=
 github.com/pborman/getopt v0.0.0-20170112200414-7148bc3a4c30/go.mod h1:85jBQOZwpVEaDAr341tbn15RS4fCAsIst0qp7i8ex1o=
 github.com/pelletier/go-toml v1.2.0/go.mod h1:5z9KED0ma1S8pY6P1sdut58dfprrGBbd/94hg7ilaic=
-github.com/pgsty/mc v0.0.0-20260916070421-e952aa78f10a h1:FY1vcTQT67HGBe23ozlhOiQnWCzUCGmeSN5x6DT/kFs=
+github.com/pgsty/mc v0.0.0-20261017101407-fa3b8bf72458 h1:f3kbkjtjj6D7PXmsIeKrv8phXncq1OVrxKM2e4I6rVk=
-github.com/pgsty/mc v0.0.0-20260916070421-e952aa78f10a/go.mod h1:6z9lX8kOvaWz2vLlP7BjjF3JWHnIH86XDA2HwcmpZH8=
+github.com/pgsty/mc v0.0.0-20261017101407-fa3b8bf72458/go.mod h1:n8NDi0J3hPUGCIM9tOxlhYXkYS9xiTVIxlxa3Rwx1cw=
-github.com/pgsty/silo-console v0.0.0-20261015132217-e5b165e4c0ee h1:e51AnrGe0Zq+nHTYaGfv/GkehQMO9Mekuw/ew6zYwOM=
+github.com/pgsty/silo-console v0.0.0-20261017112417-ef2dd2c41d21 h1:PHxPswGiEwU9MVisu0/Ab7AcNX4IveiX1thQzUXOsM4=
-github.com/pgsty/silo-console v0.0.0-20261015132217-e5b165e4c0ee/go.mod h1:jpt39bx4wjhXoFB+BTOa/bx/pmU3pXzp2cT7PnT21Lo=
+github.com/pgsty/silo-console v0.0.0-20261017112417-ef2dd2c41d21/go.mod h1:HFNaafu5iCO9bWS86PIyTbgF+yPD4phFt+gxML/1jJU=
-github.com/pgsty/silo-pkg/v3 v3.14.1 h1:8MONT3Hky9EOnsuPk29590pEBE6MILaOnJtaejPPi7M=
+github.com/pgsty/silo-pkg/v3 v3.14.3 h1:hMDxOo36F1whGRkYScrUGGwn98FvsmJ8hVM6pE4i8Bs=
-github.com/pgsty/silo-pkg/v3 v3.14.1/go.mod h1:jxKWxi52jSsDaDhXXPqjOyPDYdxx2iqdBXyLfM/xSIw=
+github.com/pgsty/silo-pkg/v3 v3.14.3/go.mod h1:zqIm3FMETmwPHRKuUExCDGfYkCNXt+vpfBnPM821RdI=
-github.com/philhofer/fwd v1.2.0 h1:e6DnBTl7vGY+Gz322/ASL4Gyp1FspeMvx1RNDoToZuM=
+github.com/philhofer/fwd v1.2.1 h1:UtpS73q8bFgy/yH3gCq8LSNvj5gv/FowNOiWZa7uKwQ=
-github.com/philhofer/fwd v1.2.0/go.mod h1:RqIHx9QI14HlwKwm98g9Re5prTQ6LdeRQn+gXJFxsJM=
+github.com/philhofer/fwd v1.2.1/go.mod h1:RqIHx9QI14HlwKwm98g9Re5prTQ6LdeRQn+gXJFxsJM=
 github.com/pierrec/lz4/v4 v4.1.30 h1:cchX8N2DVP668WkElI9QMwVyoNabLkq1LofDHFeIrdg=
 github.com/pierrec/lz4/v4 v4.1.30/go.mod h1:EoQMVJgeeEOMsCqCzqFm2O0cJvljX2nGZjcRIPL34O4=
 github.com/pkg/browser v0.0.0-20240102092130-5ac0b6a4141c h1:+mdjkGKdHQG3305AYmdv1U2eRNDiU2ErMBj1gwrq8eQ=
@@ -650,8 +650,8 @@ github.com/tidwall/match v1.2.0 h1:0pt8FlkOwjN2fPt4bIl4BoNxb98gGHN2ObFEDkrfZnM=
 github.com/tidwall/match v1.2.0/go.mod h1:eRSPERbgtNPcGhD8UCthc6PmLEQXEWd3PRB5JTxsfmM=
 github.com/tidwall/pretty v1.2.1 h1:qjsOFOWWQl+N3RsoF5/ssm1pHmJJwhjlSbZ51I6wMl4=
 github.com/tidwall/pretty v1.2.1/go.mod h1:ITEVvHYasfjBbM0u2Pg8T2nJnzm8xPwvNhhsoaGGjNU=
-github.com/tinylib/msgp v1.6.4 h1:mOwYbyYDLPj35mkA2BjjYejgJk9BuHxDdvRnb6v2ZcQ=
+github.com/tinylib/msgp v1.6.5 h1:iAH6XTP7BpzErjBZlujJQXxA+GmRi2YWs6M4KdLtNbE=
-github.com/tinylib/msgp v1.6.4/go.mod h1:RSp0LW9oSxFut3KzESt5Voq4GVWyS+PSulT77roAqEA=
+github.com/tinylib/msgp v1.6.5/go.mod h1:RSp0LW9oSxFut3KzESt5Voq4GVWyS+PSulT77roAqEA=
 github.com/tklauser/go-sysconf v0.4.0 h1:7H0uAN+7RkwWRaxhYXDLqa5V3LPrJeV8wmD9dRUgPQU=
 github.com/tklauser/go-sysconf v0.4.0/go.mod h1:8mTNWyog7H+MpKijp4VmKJAd2bbYQ2zuUwkYRbUArPI=
 github.com/tklauser/numcpus v0.12.0 h1:NR85qdvHA9pFse3x3weVZ0r0ST8R6l5RHbZrlRaqob4=
@@ -686,12 +686,12 @@ github.com/zeebo/xxh3 v1.1.0 h1:s7DLGDK45Dyfg7++yxI0khrfwq9661w9EN78eP/UZVs=
 github.com/zeebo/xxh3 v1.1.0/go.mod h1:IisAie1LELR4xhVinxWS5+zf1lA4p0MW4T+w+W07F5s=
 go.etcd.io/bbolt v1.3.6 h1:/ecaJf0sk1l4l6V4awd65v2C3ILy7MSj+s/x1ADCIMU=
 go.etcd.io/bbolt v1.3.6/go.mod h1:qXsaaIqmgQH0T+OPdb99Bf+PKfBBQVAdyD6TY9G8XM4=
-go.etcd.io/etcd/api/v3 v3.7.1 h1:KJG0/DcWGfe3Y1otDf/fsBf0TSSgpxZ5RO/L8SFt73E=
+go.etcd.io/etcd/api/v3 v3.7.2 h1:xgt/6el1LsPWWYNLkhMAK4tZm6dF+1sCqDecpE5gdbk=
-go.etcd.io/etcd/api/v3 v3.7.1/go.mod h1:8bXIpCMeV7E3/XL0Ix123ATn3dB+0V7d9zklHbB0m78=
+go.etcd.io/etcd/api/v3 v3.7.2/go.mod h1:RoRCBRt9BfBff1pIGZLUVMiz7wu3bY+b2qLysGu1HY4=
-go.etcd.io/etcd/client/pkg/v3 v3.7.1 h1:rKYsj3pRkR0eK3yjT3XOgrhqfmIfj9pzNgxjh7mfFv4=
+go.etcd.io/etcd/client/pkg/v3 v3.7.2 h1:SVtlR7tiSVAYOQ4nWPIyFXb4RMgEcnzeAG9RQ8MoNDU=
-go.etcd.io/etcd/client/pkg/v3 v3.7.1/go.mod h1:cnzZGIUzSfjEwLC6UBVsSXlEK1eepS/JUD7wE6PLRT0=
+go.etcd.io/etcd/client/pkg/v3 v3.7.2/go.mod h1:HsSux/B3ahgyw/D5+d4YbZqicOi0mEbuxm6lIUdjAoI=
-go.etcd.io/etcd/client/v3 v3.7.1 h1:0PEMMC0KuZmVIN+RAbdqfkZ45pYTgKVtmBEbRCvZFUg=
+go.etcd.io/etcd/client/v3 v3.7.2 h1:Z66GqDQDI7zPDfVSsIBqGSK4mJYLtv8ESwXa4mPf+wY=
-go.etcd.io/etcd/client/v3 v3.7.1/go.mod h1:ffNqALa8tRCYhYo1F9oR489y23K39Gz+BSR3ApAGYq0=
+go.etcd.io/etcd/client/v3 v3.7.2/go.mod h1:x03t1qMs4tGZirCDJlMuzPBJdQffXJImIyEjLhNBCsY=
 go.opentelemetry.io/auto/sdk v1.2.1 h1:jXsnJ4Lmnqd11kwkBV2LgLoFMZKizbCi5fNZ/ipaZ64=
 go.opentelemetry.io/auto/sdk v1.2.1/go.mod h1:KRTj+aOaElaLi+wW1kO/DZRXwkF4C5xPbEe3ZiIhN7Y=
 go.opentelemetry.io/contrib/detectors/gcp v1.44.0 h1:NmLfL734pJhM0JKaYd2Y28+nY9dPRWYAAbxhRCrKXPw=
@@ -740,15 +740,15 @@ golang.org/x/crypto v0.0.0-20211209193657-4570a0811e8b/go.mod h1:IxCIyHEi3zRg3s0
 golang.org/x/crypto v0.0.0-20220315160706-3147a52a75dd/go.mod h1:IxCIyHEi3zRg3s0A5j5BB6A9Jmi73HwBIUl50j+osU4=
 golang.org/x/crypto v0.5.0/go.mod h1:NK/OQwhpMQP3MwtdjgLlYHnH9ebylxKWv3e0fK+mkQU=
 golang.org/x/crypto v0.6.0/go.mod h1:OFC/31mSvZgRz0V1QTNCzfAI1aIRzbiufJtkMIlEp58=
-golang.org/x/crypto v0.57.0 h1:3ZVCjf8Ggz7zneR/EHRVx68Ctf+2pmIMP2UFhh9cC6M=
+golang.org/x/crypto v0.58.0 h1:COkYLr4k7nDI4r6QtMML2AfzeIswWHAJthLS6K/M/54=
-golang.org/x/crypto v0.57.0/go.mod h1:Fdz0i5U6CoizGwLda9DttjSk6qlZo25zYNtR+ycvuZA=
+golang.org/x/crypto v0.58.0/go.mod h1:Xh+kl5A+M0gMiWDytH2Hhr5Z7WbNCpv/eB/EFgfSk+w=
-golang.org/x/exp v0.0.0-20260908205506-85c1c2202aba h1:Ck8QetSgk912qxWLMCKxd0in+aiyBQyDSMae6e/xmpU=
+golang.org/x/exp v0.0.0-20261014143641-f0251eb4aa3e h1:HgI6W1OolKK6tcV74is+rl9SMT0njsHlp0TbQiGXI/A=
-golang.org/x/exp v0.0.0-20260908205506-85c1c2202aba/go.mod h1:50RgIsmK7OwqzTTeqcSXQW8SswW0o8fRcDxmqGluJ8E=
+golang.org/x/exp v0.0.0-20261014143641-f0251eb4aa3e/go.mod h1:DBoxwWhjaW8DJJw/YeM4zpeOKMC1FMXSblsYy+1mFSw=
 golang.org/x/mod v0.2.0/go.mod h1:s0Qsj1ACt9ePp/hMypM3fl4fZqREWJwdYDEqhRiZZUA=
 golang.org/x/mod v0.3.0/go.mod h1:s0Qsj1ACt9ePp/hMypM3fl4fZqREWJwdYDEqhRiZZUA=
 golang.org/x/mod v0.6.0-dev.0.20220419223038-86c51ed26bb4/go.mod h1:jJ57K6gSWd91VN4djpZkiMVwK6gcyfeH4XE8wZrZaV4=
-golang.org/x/mod v0.41.0 h1:qJmnOUb4YB+FsEuM3HcWucdZASCPGhsX6uljO6pog0c=
+golang.org/x/mod v0.42.0 h1:ICeL3t2L30ljSPHHxDnR4fqVcBt/zR2zJAb9juavmQg=
-golang.org/x/mod v0.41.0/go.mod h1:Ek9pY8RKWXwsWvd3rQiHYtMqkjSUV+s1Rj7j4H5Ur6o=
+golang.org/x/mod v0.42.0/go.mod h1:u5fzIrRInkL8MaO1cbRkUHT64v303YIT5KEqpNr7jRs=
 golang.org/x/net v0.0.0-20181201002055-351d144fa1fc/go.mod h1:mL1N/T3taQHkDXs73rZJwtUhF3w3ftmwwsq0BUmARs4=
 golang.org/x/net v0.0.0-20190311183353-d8887717615a/go.mod h1:t9HGtf8HONx5eT2rtn7q6eTqICYqUVnKs3thJo3Qplg=
 golang.org/x/net v0.0.0-20190404232315-eb5bcb51f2a3/go.mod h1:t9HGtf8HONx5eT2rtn7q6eTqICYqUVnKs3thJo3Qplg=
@@ -762,8 +762,8 @@ golang.org/x/net v0.0.0-20220722155237-a158d28d115b/go.mod h1:XRhObCWvk6IyKnWLug
 golang.org/x/net v0.5.0/go.mod h1:DivGGAXEgPSlEBzxGzZI+ZLohi+xUj054jfeKui00ws=
 golang.org/x/net v0.6.0/go.mod h1:2Tu9+aMcznHK/AK1HMvgo6xiTLG5rD5rZLDS+rp2Bjs=
 golang.org/x/net v0.7.0/go.mod h1:2Tu9+aMcznHK/AK1HMvgo6xiTLG5rD5rZLDS+rp2Bjs=
-golang.org/x/net v0.60.0 h1:79p50tfZlm0J9YfoDsSi639qSXNGVwEzOPLCxM2FsYU=
+golang.org/x/net v0.61.0 h1:RR+6j/BTBCrggOLugxa+66R8FvoN6MHlzDlZIbN9YFw=
-golang.org/x/net v0.60.0/go.mod h1:2DA/G1UfVbCpQPeWTmMPGY7Cs2PkBkwu743bVX5PIVg=
+golang.org/x/net v0.61.0/go.mod h1:WgDCOTH8iwtB66T4/GfI94AHo/hhdb7Hah9NDseQfiI=
 golang.org/x/oauth2 v0.37.0 h1:JUlcxA8oAtauLfiH8FX2/FkAWHAdi0QtGCGc+hofE98=
 golang.org/x/oauth2 v0.37.0/go.mod h1:IxwZNxUULJmpBFf9K/9NTMSIfZZuvuTy1gGxhigP/58=
 golang.org/x/sync v0.0.0-20181108010431-42b317875d0f/go.mod h1:RxMgew5VJxzue5/jJTE5uejpjVlOe/izrB70Jof72aM=
@@ -773,8 +773,8 @@ golang.org/x/sync v0.0.0-20190911185100-cd5d95a43a6e/go.mod h1:RxMgew5VJxzue5/jJ
 golang.org/x/sync v0.0.0-20201020160332-67f06af15bc9/go.mod h1:RxMgew5VJxzue5/jJTE5uejpjVlOe/izrB70Jof72aM=
 golang.org/x/sync v0.0.0-20201207232520-09787c993a3a/go.mod h1:RxMgew5VJxzue5/jJTE5uejpjVlOe/izrB70Jof72aM=
 golang.org/x/sync v0.0.0-20220722155255-886fb9371eb4/go.mod h1:RxMgew5VJxzue5/jJTE5uejpjVlOe/izrB70Jof72aM=
-golang.org/x/sync v0.23.0 h1:KameEIfc1IkluZyXWLn39Wd4tURc6GbCiISGiZm2bQk=
+golang.org/x/sync v0.24.0 h1:iqiDrGoDgyuvAKrhLQthUEzU/R378N78gXbBjxed7oA=
-golang.org/x/sync v0.23.0/go.mod h1:sUUOizhqBxiL6pEWpqNLUiaJn1ShEbZ6BBqskPbjZm0=
+golang.org/x/sync v0.24.0/go.mod h1:sUUOizhqBxiL6pEWpqNLUiaJn1ShEbZ6BBqskPbjZm0=
 golang.org/x/sys v0.0.0-20180926160741-c2ed4eda69e7/go.mod h1:STP8DvDyc/dI5b8T5hshtkjS+E42TnysNCUPdjciGhY=
 golang.org/x/sys v0.0.0-20181205085412-a5c9d58dba9a/go.mod h1:STP8DvDyc/dI5b8T5hshtkjS+E42TnysNCUPdjciGhY=
 golang.org/x/sys v0.0.0-20190130150945-aca44879d564/go.mod h1:STP8DvDyc/dI5b8T5hshtkjS+E42TnysNCUPdjciGhY=
@@ -805,23 +805,23 @@ golang.org/x/sys v0.4.0/go.mod h1:oPkhp1MJrh7nUepCBck5+mAzfO9JrbApNNgaTdGDITg=
 golang.org/x/sys v0.5.0/go.mod h1:oPkhp1MJrh7nUepCBck5+mAzfO9JrbApNNgaTdGDITg=
 golang.org/x/sys v0.6.0/go.mod h1:oPkhp1MJrh7nUepCBck5+mAzfO9JrbApNNgaTdGDITg=
 golang.org/x/sys v0.21.0/go.mod h1:/VUhepiaJMQUp4+oa/7Zr1D23ma6VTLIYjOOTFZPUcA=
-golang.org/x/sys v0.48.0 h1:bbX/i/6MgT9BVLM9RT1thmxL04yeTAhbEz4SyadbXoo=
+golang.org/x/sys v0.49.0 h1:XbzkgYJHdqh/8m2Uu0W/dQv8nktxx4BFHp1M0gROTXA=
-golang.org/x/sys v0.48.0/go.mod h1:hNLxWAXmnKAxqDtdwIYC4bM9oQPEecfsnNMuSxOs3og=
+golang.org/x/sys v0.49.0/go.mod h1:hNLxWAXmnKAxqDtdwIYC4bM9oQPEecfsnNMuSxOs3og=
 golang.org/x/term v0.0.0-20201117132131-f5c789dd3221/go.mod h1:Nr5EML6q2oocZ2LXRh80K7BxOlk5/8JxuGnuhpl+muw=
 golang.org/x/term v0.0.0-20201126162022-7de9c90e9dd1/go.mod h1:bj7SfCRtBDWHUb9snDiAeCFNEtKQo2Wmx5Cou7ajbmo=
 golang.org/x/term v0.0.0-20210927222741-03fcf44c2211/go.mod h1:jbD1KX2456YbFQfuXm/mYQcufACuNUgVhRMnK/tPxf8=
 golang.org/x/term v0.4.0/go.mod h1:9P2UbLfCdcvo3p/nzKvsmas4TnlujnuoV9hGgYzW1lQ=
 golang.org/x/term v0.5.0/go.mod h1:jMB1sMXY+tzblOD4FWmEbocvup2/aLOaQEp7JmGp78k=
-golang.org/x/term v0.46.0 h1:3+OXuTbaKDgwk8jTi3aSLHRlmWqHEUDUtxnbFigO4YE=
+golang.org/x/term v0.47.0 h1:cpyOwQkr2Nzxe2p6KnrTG7Ei2wtzjsuSjhpXXepZ/BM=
-golang.org/x/term v0.46.0/go.mod h1:+K02xbkittuwc0Am4abfA3Fc+XRGXkvBXNO88NCXPoc=
+golang.org/x/term v0.47.0/go.mod h1:efAfAuG6RKABvg1ArGebRkvX6LXul14SM67o2e4mUCY=
 golang.org/x/text v0.3.0/go.mod h1:NqM8EUOU14njkJ3fqMW+pc6Ldnwhi/IjpwHt7yyuwOQ=
 golang.org/x/text v0.3.3/go.mod h1:5Zoc/QRtKVWzQhOtBMvqHzDpF6irO9z98xDceosuGiQ=
 golang.org/x/text v0.3.6/go.mod h1:5Zoc/QRtKVWzQhOtBMvqHzDpF6irO9z98xDceosuGiQ=
 golang.org/x/text v0.3.7/go.mod h1:u+2+/6zg+i71rQMx5EYifcz6MCKuco9NR6JIITiCfzQ=
 golang.org/x/text v0.6.0/go.mod h1:mrYo+phRRbMaCq/xk9113O4dZlRixOauAjOtrjsXDZ8=
 golang.org/x/text v0.7.0/go.mod h1:mrYo+phRRbMaCq/xk9113O4dZlRixOauAjOtrjsXDZ8=
-golang.org/x/text v0.42.0 h1:JbOZXgfeCPU9gacVtYliJqOhD+zhrEqK4LfdpmlUZqI=
+golang.org/x/text v0.43.0 h1:1QivrlhwAsnMOcqgOdl9my7s9OyZl8O5/UGUTJVXoWc=
-golang.org/x/text v0.42.0/go.mod h1:ojzP1Z+2QtioaF8DTtO8K5q7JWVVYwZKenzujK0Zd0E=
+golang.org/x/text v0.43.0/go.mod h1:hbSIYA/amXcRXTFSZqh8tWHEoqRvvbKPJ2uz/KU+a+s=
 golang.org/x/time v0.0.0-20211116232009-f0f3c7e86c11/go.mod h1:tRJNPiyCQ0inRvYxbN9jk5I+vvW/OXSQhTDSoE431IQ=
 golang.org/x/time v0.16.0 h1:vMb6ptszcQMkcwiRTAuNNU50gom6++Q/6gY2hDM6VDE=
 golang.org/x/time v0.16.0/go.mod h1:rVKOqvZeKvrDKTQiAHJ7wmwP0RzleSphoEA9RcdLA0s=
@@ -831,8 +831,8 @@ golang.org/x/tools v0.0.0-20191119224855-298f0cb1881e/go.mod h1:b+2E5dAYhXwXZwtn
 golang.org/x/tools v0.0.0-20200619180055-7c47624df98f/go.mod h1:EkVYQZoAsY45+roYkvgYkIh4xh/qjgUK9TdY2XT94GE=
 golang.org/x/tools v0.0.0-20210106214847-113979e3529a/go.mod h1:emZCQorbCU4vsT4fOWvOPXz4eW1wZW4PmDk9uLelYpA=
 golang.org/x/tools v0.1.12/go.mod h1:hNGJHUnrk76NpqgfD5Aqm5Crs+Hm0VOH/i9J2+nxYbc=
-golang.org/x/tools v0.50.0 h1:c2ifzfcuY7L90lZ2aKd8S4K2NpASF08SZx9ZuJkHmSU=
+golang.org/x/tools v0.52.0 h1:ryxg4U63oaUYk7uOLTF37n43f4qgtGvqV4Ioit1BaEs=
-golang.org/x/tools v0.50.0/go.mod h1:7ulVMw3831Mwi5EZD6RomGyffr4VFjuNYXf2BbCEAV0=
+golang.org/x/tools v0.52.0/go.mod h1:ItcP5UNaVGG+Yc4eIYmaFLNJ4soEopkx5Y2d4lcEurs=
 golang.org/x/xerrors v0.0.0-20190717185122-a985d3407aa7/go.mod h1:I/5z698sn9Ka8TeJc9MKroUUfqBBauWjQqLJ2OPfmY0=
 golang.org/x/xerrors v0.0.0-20191011141410-1b5146add898/go.mod h1:I/5z698sn9Ka8TeJc9MKroUUfqBBauWjQqLJ2OPfmY0=
 golang.org/x/xerrors v0.0.0-20191204190536-9bdfabe68543/go.mod h1:I/5z698sn9Ka8TeJc9MKroUUfqBBauWjQqLJ2OPfmY0=

helm/silo/values.yaml

@@ -25,7 +25,7 @@ imagePullSecrets: []
 ##
 mcImage:
   repository: pgsty/mc
-  tag: RELEASE.2026-09-16T00-00-00Z
+  tag: RELEASE.2026-10-18T00-00-00Z
   pullPolicy: IfNotPresent
 
 ## Silo mode, i.e. standalone or distributed.

internal/config/api/api.go

@@ -68,7 +68,7 @@ const (
 
 	EnvAPIStaleUploadsCleanupInterval = "MINIO_API_STALE_UPLOADS_CLEANUP_INTERVAL"
 	EnvAPIStaleUploadsExpiry          = "MINIO_API_STALE_UPLOADS_EXPIRY"
-	EnvAPIMultipartListing            = "MINIO_API_MULTIPART_LISTING"
+	EnvAPIMultipartListing            = "MINIO_API_MULTIPART_LISTING" // Retired; used only for a startup warning.
 	EnvAPIDeleteCleanupInterval       = "MINIO_API_DELETE_CLEANUP_INTERVAL"
 	EnvDeleteCleanupInterval          = "MINIO_DELETE_CLEANUP_INTERVAL"
 	EnvAPIODirect                     = "MINIO_API_ODIRECT"
@@ -180,7 +180,6 @@ type Config struct {
 	TransitionWorkers           int           `json:"transition_workers"`
 	StaleUploadsCleanupInterval time.Duration `json:"stale_uploads_cleanup_interval"`
 	StaleUploadsExpiry          time.Duration `json:"stale_uploads_expiry"`
-	MultipartListing            string        `json:"multipart_listing"`
 	DeleteCleanupInterval       time.Duration `json:"delete_cleanup_interval"`
 	EnableODirect               bool          `json:"enable_odirect"`
 	GzipObjects                 bool          `json:"gzip_objects"`
@@ -209,7 +208,7 @@ func LookupConfig(kvs config.KVS) (cfg Config, err error) {
 		apiReplicationWorkers,
 		apiReplicationFailedWorkers,
 		"expiry_workers",
-		apiMultipartListing, // Ignore the retired shared-config key during migration.
+		apiMultipartListing, // Tolerate the retired key without changing other API settings.
 	}
 
 	disableODirect := env.Get(EnvAPIDisableODirect, kvs.Get(apiDisableODirect)) == config.EnableOn
@@ -344,11 +343,5 @@ func LookupConfig(kvs config.KVS) (cfg Config, err error) {
 		cfg.ObjectMaxVersions = math.MaxInt64
 	}
 
-	cfg.MultipartListing = env.Get(EnvAPIMultipartListing, "legacy")
-	if cfg.MultipartListing != "strict" && cfg.MultipartListing != "legacy" {
-		cfg.MultipartListing = "legacy"
-		return cfg, fmt.Errorf("%s must be strict or legacy; using legacy", EnvAPIMultipartListing)
-	}
-
 	return cfg, nil
 }

internal/config/api/api_test.go

@@ -11,22 +11,21 @@ import (
 	"github.com/minio/minio/internal/config"
 )
 
-func TestMultipartListingMigrationMode(t *testing.T) {
+func TestMultipartListingRetiredConfiguration(t *testing.T) {
 	t.Setenv(EnvAPIRequestsMax, "")
 	t.Setenv(EnvAPISyncEvents, "")
 	t.Setenv(EnvAPIObjectMaxVersions, "")
 	for _, tc := range []struct {
-		name, stored, override, want string
+		name, stored, override string
-		invalid                      bool
 	}{
-		{name: "default", want: "legacy"},
+		{name: "default"},
-		{name: "explicit-legacy", override: "legacy", want: "legacy"},
+		{name: "explicit-legacy", override: "legacy"},
-		{name: "explicit-strict", override: "strict", want: "strict"},
+		{name: "explicit-strict", override: "strict"},
-		{name: "retired-stored-strict", stored: "strict", want: "legacy"},
+		{name: "retired-stored-strict", stored: "strict"},
-		{name: "retired-stored-legacy", stored: "legacy", want: "legacy"},
+		{name: "retired-stored-legacy", stored: "legacy"},
-		{name: "retired-stored-invalid", stored: "automatic", want: "legacy"},
+		{name: "retired-stored-invalid", stored: "automatic"},
-		{name: "environment-override", stored: "legacy", override: "strict", want: "strict"},
+		{name: "environment-override", stored: "legacy", override: "strict"},
-		{name: "invalid-environment", stored: "strict", override: "automatic", want: "legacy", invalid: true},
+		{name: "invalid-environment", stored: "strict", override: "automatic"},
 	} {
 		t.Run(tc.name, func(t *testing.T) {
 			t.Setenv(EnvAPIMultipartListing, tc.override)
@@ -38,11 +37,11 @@ func TestMultipartListingMigrationMode(t *testing.T) {
 				kvs.Set(apiMultipartListing, tc.stored)
 			}
 			cfg, err := LookupConfig(kvs)
-			if (err != nil) != tc.invalid || cfg.MultipartListing != tc.want {
+			if err != nil {
-				t.Fatalf("mode=%q err=%v, want %q invalid=%v", cfg.MultipartListing, err, tc.want, tc.invalid)
+				t.Fatalf("retired setting must be ignored: %v", err)
 			}
 			if cfg.RequestsMax != 17 || !cfg.SyncEvents || cfg.ObjectMaxVersions != 71 {
-				t.Fatalf("migration setting discarded unrelated config: %+v", cfg)
+				t.Fatalf("retired setting discarded unrelated config: %+v", cfg)
 			}
 		})
 	}
@@ -58,7 +57,7 @@ func TestMultipartListingConfigPersistence(t *testing.T) {
 	t.Cleanup(func() { config.DefaultKVS = previous })
 	for _, help := range Help {
 		if help.Key == apiMultipartListing {
-			t.Fatal("process-only setting advertised as shared config")
+			t.Fatal("retired setting advertised as shared config")
 		}
 	}
 	for _, input := range []string{
@@ -82,11 +81,11 @@ func TestMultipartListingConfigPersistence(t *testing.T) {
 			}
 			kvs := loaded.Merge()[config.APISubSys][config.Default]
 			if _, present := kvs.Lookup(apiMultipartListing); present {
-				t.Fatal("process-only setting persisted in shared config")
+				t.Fatal("retired setting persisted in shared config")
 			}
 			cfg, err := LookupConfig(kvs)
-			if err != nil || cfg.MultipartListing != "strict" {
+			if err != nil {
-				t.Fatalf("environment mode lost: %+v %v", cfg, err)
+				t.Fatalf("retired environment setting caused failure: %+v %v", cfg, err)
 			}
 			if input != "" && cfg.RequestsMax != 17 {
 				t.Fatalf("unrelated setting lost: %+v", cfg)

internal/dsync/drwmutex-retry_test.go

@@ -0,0 +1,218 @@
+// Copyright (c) 2026 Feng Ruohang
+//
+// This file is part of Silo Object Storage stack
+//
+// This program is free software: you can redistribute it and/or modify
+// it under the terms of the GNU Affero General Public License as published by
+// the Free Software Foundation, either version 3 of the License, or
+// (at your option) any later version.
+//
+// This program is distributed in the hope that it will be useful
+// but WITHOUT ANY WARRANTY; without even the implied warranty of
+// MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+// GNU Affero General Public License for more details.
+//
+// You should have received a copy of the GNU Affero General Public License
+// along with this program.  If not, see <http://www.gnu.org/licenses/>.
+
+package dsync
+
+import (
+	"context"
+	"math/rand"
+	"sync/atomic"
+	"testing"
+	"testing/synctest"
+	"time"
+)
+
+// retryTestLocker deliberately leaves unexpected methods unimplemented.
+// These tests exercise the real DRWMutex retry and cleanup code without HTTP.
+type retryTestLocker struct {
+	NetLocker
+	acquire func(context.Context, LockArgs) (bool, error)
+	release func(context.Context, LockArgs) (bool, error)
+}
+
+func (l retryTestLocker) Lock(ctx context.Context, args LockArgs) (bool, error) {
+	return l.acquire(ctx, args)
+}
+
+func (l retryTestLocker) RLock(ctx context.Context, args LockArgs) (bool, error) {
+	return l.acquire(ctx, args)
+}
+
+func (l retryTestLocker) Unlock(ctx context.Context, args LockArgs) (bool, error) {
+	return l.release(ctx, args)
+}
+
+func (l retryTestLocker) RUnlock(ctx context.Context, args LockArgs) (bool, error) {
+	return l.release(ctx, args)
+}
+
+func (l retryTestLocker) String() string { return "retry-test-locker" }
+
+func newRetryTestMutex(lockers []NetLocker) *DRWMutex {
+	dm := NewDRWMutex(&Dsync{
+		GetLockers: func() ([]NetLocker, string) { return lockers, "retry-test-owner" },
+		Timeouts:   DefaultTimeouts,
+	}, "retry-test-resource")
+	dm.rng = rand.New(&lockedRandSource{src: rand.NewSource(1)})
+	return dm
+}
+
+func retryTestAcquire(ctx context.Context, dm *DRWMutex, read bool, opts Options) bool {
+	if read {
+		return dm.GetRLock(ctx, nil, "retry-test-uid", "retry-test", opts)
+	}
+	return dm.GetLock(ctx, nil, "retry-test-uid", "retry-test", opts)
+}
+
+func TestDRWMutexRetryCancellation(t *testing.T) {
+	for _, read := range []bool{false, true} {
+		for _, opts := range []Options{
+			{Timeout: time.Hour},
+			{Timeout: time.Hour, RetryInterval: time.Hour},
+		} {
+			synctest.Test(t, func(t *testing.T) {
+				var calls atomic.Int64
+				lockers := make([]NetLocker, 4)
+				for i := range lockers {
+					lockers[i] = retryTestLocker{acquire: func(context.Context, LockArgs) (bool, error) {
+						calls.Add(1)
+						return false, nil
+					}}
+				}
+				dm := newRetryTestMutex(lockers)
+				ctx, cancel := context.WithCancel(t.Context())
+				defer cancel()
+				done := make(chan bool, 1)
+				go func() {
+					done <- retryTestAcquire(ctx, dm, read, opts)
+				}()
+				synctest.Wait()
+				if got := calls.Load(); got != 4 {
+					t.Fatalf("first attempt made %d RPCs, want 4", got)
+				}
+				cancel()
+				synctest.Wait()
+				select {
+				case locked := <-done:
+					if locked {
+						t.Fatal("canceled waiter acquired lock")
+					}
+				default:
+					t.Errorf("read=%v opts=%+v: cancellation did not interrupt retry wait", read, opts)
+					<-done
+				}
+				time.Sleep(time.Second)
+				synctest.Wait()
+				if got := calls.Load(); got != 4 {
+					t.Fatalf("canceled waiter issued a new RPC attempt: %d calls", got)
+				}
+			})
+		}
+	}
+}
+
+func TestDRWMutexRetryDeadline(t *testing.T) {
+	for _, interval := range []time.Duration{0, time.Hour} {
+		synctest.Test(t, func(t *testing.T) {
+			locker := retryTestLocker{acquire: func(context.Context, LockArgs) (bool, error) {
+				return false, nil
+			}}
+			dm := newRetryTestMutex([]NetLocker{locker})
+			start := time.Now()
+			if retryTestAcquire(t.Context(), dm, false, Options{Timeout: 10 * time.Millisecond, RetryInterval: interval}) {
+				t.Fatal("unavailable lock was acquired")
+			}
+			if elapsed := time.Since(start); elapsed != 10*time.Millisecond {
+				t.Fatalf("retry wait overran deadline: %v", elapsed)
+			}
+		})
+	}
+}
+
+func TestDRWMutexCanceledPartialAndLateGrants(t *testing.T) {
+	for _, read := range []bool{false, true} {
+		for _, slowRelease := range []bool{false, true} {
+			synctest.Test(t, func(t *testing.T) {
+				lateGrant := make(chan struct{})
+				releaseGate := make(chan struct{})
+				if !slowRelease {
+					close(releaseGate)
+				}
+				var calls, releases atomic.Int64
+				var held [4]atomic.Bool
+				lockers := make([]NetLocker, len(held))
+				for i := range lockers {
+					lockers[i] = retryTestLocker{
+						acquire: func(ctx context.Context, _ LockArgs) (bool, error) {
+							calls.Add(1)
+							if i > 0 {
+								<-lateGrant
+							}
+							if ctx.Err() != nil {
+								t.Error("NetLocker acquisition unexpectedly inherits caller cancellation")
+							}
+							held[i].Store(true)
+							return true, nil
+						},
+						release: func(context.Context, LockArgs) (bool, error) {
+							if i == 0 {
+								<-releaseGate
+							}
+							held[i].Store(false)
+							releases.Add(1)
+							return true, nil
+						},
+					}
+				}
+				dm := newRetryTestMutex(lockers)
+				ctx, cancel := context.WithCancel(t.Context())
+				defer cancel()
+				done := make(chan bool, 1)
+				go func() { done <- retryTestAcquire(ctx, dm, read, Options{Timeout: time.Hour}) }()
+				synctest.Wait()
+				if !held[0].Load() || calls.Load() != 4 {
+					t.Fatal("partial acquisition was not established")
+				}
+				cancel()
+				synctest.Wait()
+				if slowRelease {
+					// Cancellation of retry sleep is not a whole-call bound:
+					// synchronous cleanup of a partial grant can still block.
+					select {
+					case <-done:
+						t.Fatal("acquisition returned before partial-grant cleanup completed")
+					default:
+					}
+					close(releaseGate)
+					synctest.Wait()
+				}
+				select {
+				case locked := <-done:
+					if locked {
+						t.Fatal("canceled partial acquisition returned success")
+					}
+				default:
+					t.Error("acquisition did not return after partial-grant cleanup")
+					<-done
+				}
+				if held[0].Load() || releases.Load() != 1 {
+					t.Fatal("partial grant was not released")
+				}
+				close(lateGrant)
+				synctest.Wait()
+				for i := range held {
+					if held[i].Load() {
+						t.Errorf("late grant on node %d was not released", i)
+					}
+				}
+				if calls.Load() != 4 || releases.Load() != 4 {
+					t.Fatalf("calls=%d releases=%d, want 4/4", calls.Load(), releases.Load())
+				}
+			})
+		}
+	}
+}

internal/dsync/drwmutex.go

@@ -260,14 +260,22 @@ func (dm *DRWMutex) lockBlocking(ctx context.Context, lockLossCallback func(), i
 				return locked
 			}
 
+			var retryInterval time.Duration
 			switch {
 			case opts.RetryInterval < 0:
 				return false
 			case opts.RetryInterval > 0:
-				time.Sleep(opts.RetryInterval)
+				retryInterval = opts.RetryInterval
 			default:
 				attempt++
-				time.Sleep(lockRetryBackOff(dm.rng, attempt))
+				retryInterval = lockRetryBackOff(dm.rng, attempt)
+			}
+			timer := time.NewTimer(retryInterval)
+			select {
+			case <-ctx.Done():
+				timer.Stop()
+				return false
+			case <-timer.C:
 			}
 		}
 	}

internal/lsync/lrwmutex-retry_test.go

@@ -0,0 +1,72 @@
+// Copyright (c) 2026 Feng Ruohang
+//
+// This file is part of Silo Object Storage stack
+//
+// This program is free software: you can redistribute it and/or modify
+// it under the terms of the GNU Affero General Public License as published by
+// the Free Software Foundation, either version 3 of the License, or
+// (at your option) any later version.
+//
+// This program is distributed in the hope that it will be useful
+// but WITHOUT ANY WARRANTY; without even the implied warranty of
+// MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+// GNU Affero General Public License for more details.
+//
+// You should have received a copy of the GNU Affero General Public License
+// along with this program.  If not, see <http://www.gnu.org/licenses/>.
+
+package lsync
+
+import (
+	"context"
+	"testing"
+	"testing/synctest"
+	"time"
+)
+
+func TestLRWMutexRetryCancellation(t *testing.T) {
+	for _, read := range []bool{false, true} {
+		synctest.Test(t, func(t *testing.T) {
+			lm := NewLRWMutex()
+			lm.Lock()
+			defer lm.Unlock()
+			ctx, cancel := context.WithCancel(t.Context())
+			defer cancel()
+			done := make(chan bool, 1)
+			go func() {
+				if read {
+					done <- lm.GetRLock(ctx, "waiter", "retry-test", time.Hour)
+				} else {
+					done <- lm.GetLock(ctx, "waiter", "retry-test", time.Hour)
+				}
+			}()
+			synctest.Wait()
+			cancel()
+			synctest.Wait()
+			select {
+			case locked := <-done:
+				if locked {
+					t.Fatal("canceled waiter acquired lock")
+				}
+			default:
+				t.Error("cancellation did not interrupt local retry wait")
+				<-done
+			}
+		})
+	}
+}
+
+func TestLRWMutexRetryDeadline(t *testing.T) {
+	synctest.Test(t, func(t *testing.T) {
+		lm := NewLRWMutex()
+		lm.Lock()
+		defer lm.Unlock()
+		start := time.Now()
+		if lm.GetLock(t.Context(), "waiter", "retry-test", time.Millisecond) {
+			t.Fatal("unavailable lock was acquired")
+		}
+		if elapsed := time.Since(start); elapsed != time.Millisecond {
+			t.Fatalf("local retry wait overran deadline: %v", elapsed)
+		}
+	})
+}

internal/lsync/lrwmutex.go

@@ -115,7 +115,13 @@ func (lm *LRWMutex) lockLoop(ctx context.Context, id, source string, timeout tim
 			if lm.lock(id, source, isWriteLock) {
 				return true
 			}
-			time.Sleep(time.Duration(r.Float64() * float64(lockRetryInterval)))
+			timer := time.NewTimer(time.Duration(r.Float64() * float64(lockRetryInterval)))
+			select {
+			case <-retryCtx.Done():
+				timer.Stop()
+				return false
+			case <-timer.C:
+			}
 		}
 	}
 }