An independent adversarial review of the bucket metadata convergence
work found three defects it had introduced.
GetBucketInfo overwrote the physical creation probe with cached
metadata, which a bucket that never held a configuration legitimately
lacks. The new creation-time requirement then failed every policy, tag,
SSE, quota, versioning and Object Lock write on such a bucket, with no
operator recovery path, and initial synchronization skipped it silently.
Return the physical result unchanged when metadata is not requested, as
ListBuckets already does, recover the time during initial
synchronization, and pass it to MakeBucketHook so peers adopt the same
bucket generation.
Replication status compared parsed policies statement by statement while
heal compares the canonical key. An upgraded peer that stored an
equivalent statement order was therefore reported as mismatched forever,
and heal never had anything to write. Compare the key heal compares;
per-site presence counting is unchanged.
Heal diagnostics shared one log key across four conditions, so a real
peer RPC failure could be deduplicated away by an earlier message, and
they were logged at error level for the normal transient of a peer that
does not have the bucket yet. Give each reason its own key at warning
level, report only a field state that exists and still cannot be
ordered, and diagnose nothing when no site holds a state to propagate.
The recovery test now runs against the real ObjectLayer; the stub it
replaced returned the expected time and hid the defect. The policy
status test uses a statement order the canonical encoder reorders, and
adoption coverage is extended past a real field time.
Signed-off-by: Feng Ruohang <rh@vonng.com>
Return the committed object or part time on federation write responses and
capture it for CopyObject and UploadPartCopy without a follow-up read.
Keep successful writes compatible with targets that do not supply a time.
Reject authenticated raw SSE-C replica CopyObject across deployments before
forwarding. Extend the existing SSE fixtures to verify empty objects,
stored checksums, KMS contexts and multipart sources.
Refs: #169, #168, #171
Signed-off-by: Feng Ruohang <rh@vonng.com>
A cross-deployment CopyObject that requests
`x-amz-object-lock-legal-hold: ON` answered 200 while the destination
carried no hold. The resolved value reached the remote as ordinary user
metadata, `X-Amz-Meta-X-Amz-Object-Lock-Legal-Hold`, so nothing applied it.
Retention requested on the same copy survived, which is what made the loss
easy to miss.
The federation branch passes the resolved metadata map straight to
`Core.PutObject` as `PutObjectOptions.UserMetadata`. minio-go's `Header()`
writes the typed lock fields first, then prefixes every UserMetadata key it
does not recognise with `x-amz-meta-`; `supportedHeaders` covers
`x-amz-object-lock-mode` and `x-amz-object-lock-retain-until-date` but not
`x-amz-object-lock-legal-hold`, and `isAmzHeader` does not match it either.
Retention therefore arrives as real headers and the hold does not. The
high-level `validate()` that would have rejected the key never runs, because
`Core.PutObject` goes straight to the low-level PUT.
Carry the hold on the typed `LegalHold` option and forward a cloned map with
the raw key removed. The clone matters twice: typed fields are written before
the UserMetadata loop, so a leftover raw key would add a bogus `x-amz-meta-`
entry beside the correct header, and the proxy's own response and event
metadata are rebuilt from the resolved values rather than the forwarding map,
which no longer carries the hold.
Retention stays in the map deliberately. It already passes through as a
standard header, and moving it to the typed `RetainUntilDate` field would
format with `time.RFC3339` and truncate a retain-until date to whole seconds.
The new test asserts the wire: the remote must receive
`X-Amz-Object-Lock-Legal-Hold` and never the `x-amz-meta-` spelling, and the
destination version must actually store the hold. It fails without the change
with "legal hold forwarded as user metadata [ON]".
Fixes#166
Signed-off-by: Ayush Sharma <72848455+Aeirx@users.noreply.github.com>
Use ChangeMyPassword for the authenticated user and keep CreateUser for other users. Coordinate silo-pkg b3760f56ec23, mcli fa22b40b4eb7, Console 1b95b6cec652 and upstream minio-go 78bfa91607c2. Add legacy-policy and SDK streaming regressions plus upgrade guidance.
Signed-off-by: Feng Ruohang <rh@vonng.com>
A presigned or signed PUT authorized for a single object could be turned
into a server-side copy of any object the signing key can read by adding
an unsigned x-amz-copy-source header, executed as the signer. SigV4
verification only walked the signed-headers list, never the headers that
actually arrived; the meta-header check matched only X-Amz-Meta- and ran
only on the presigned path, so an unsigned x-amz-* header outside the
list was never seen while the router still dispatched the PUT to
CopyObjectHandler.
Reject any x-amz-* request header not covered by the signed headers, on
both the presigned (doesPresignedSignatureMatch) and Authorization-header
(doesSignatureMatch) paths, matching AWS S3. The check tests membership
in the signed set rather than value equality, so a header whose first
value is empty (e.g. {"", "/src/secret"}) cannot slip through.
X-Amz-Content-Sha256 is exempt (payload hash: read from the query for
presigned requests and bound into the string-to-sign for signed ones, so
it is self-protected) and X-Amz-Signature-Age is exempt (an internal
scratch header written after verification, so repeated verification of
the same request stays idempotent). The synthesized X-Amz-Tagging header
in PutObjectTagging is now injected after signature verification.
Tests that previously added x-amz-copy-source and friends after signing
(relying on the vulnerable behavior) now re-sign, mirroring real S3
clients. Adds checkUnsignedHeaders unit cases and TestPresignedVerifyIdempotent.
Reported by Oren Yomtov. Inherited unchanged from upstream minio/minio.
Tracked as SN-2026-011.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Signed-off-by: Feng Ruohang <rh@vonng.com>
Remove the eight explicit curve overrides so Go 1.27 honors tlsmlkem=0
across Server listeners, node links and outbound transports. Remove the
unused shared curve option and add wire-level regression coverage.
Document CA trust and TLS upgrade behavior, retain the investigation
artifacts, and exclude their synthetic routes from the rebrand guard.
The product compatibility baseline remains unchanged.
Validation: focused race tests, HTTP tests, lint, compatibility guard
positive/negative controls, and a fresh Linux build with three isolated
OIDC integration scenarios all pass.
Adversarial review: Claude Code Fable 5.1, max effort.
Final verdict: APPROVE FOR COMMIT.
Signed-off-by: Feng Ruohang <rh@vonng.com>
A multipart SSE-S3 source is encrypted per part, so its logical size is
the sum of the parts' decrypted sizes and the decrypting reader crosses
a part boundary. Copy such a source across the federation to a plain
and to an SSE-S3 destination and check the destination plaintext and
single-encryption size.
The test router registers routes in endpoint order and the plain
PutObject route has no query matcher, so the multipart endpoints are
listed first.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01FodsDpa6VkghaeRE6WjmEe
Signed-off-by: Feng Ruohang <rh@vonng.com>
putOptsFromReq handed the parsed kms.Context straight to
encrypt.NewSSEKMS. kms.Context implements encoding.TextMarshaler, so the
SDK serialized it as a JSON string, and a request without a context
still produced one because the nil Context is a typed nil inside the
interface value and marshals to "{}". The receiving ParseHTTP rejects
both forms, so every federated CopyObject to an SSE-KMS destination
failed with InvalidArgument once the forwarded stream was correct.
Pass a plain map, or nothing when no context was requested, and cover
SSE-KMS destinations with and without an explicit context.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01FodsDpa6VkghaeRE6WjmEe
Signed-off-by: Feng Ruohang <rh@vonng.com>
The legacy etcd bucket-federation branch of CopyObjectHandler reads its
source through getObjectNInfo, which yields the decrypted and
decompressed bytes, but it also ran the destination encryption locally
and then forwarded that stream to the remote PutObject with the source's
stored size and the destination SSE option. SSE to plain and plain to
SSE therefore failed on a Content-Length mismatch, while SSE to SSE
matched by coincidence: the remote encrypted the ciphertext a second
time and stored an unreadable object, and a destination GET returned
the inner ciphertext with HTTP 200.
The remote write owns the destination's storage transformations, so
hand it the logical bytes at their logical size and let it encrypt
exactly once. Compression was already excluded on this branch; apply
the same rule to encryption, size the forwarded reader by actualSize,
and declare that size on the forwarded PutObject.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01FodsDpa6VkghaeRE6WjmEe
Signed-off-by: Feng Ruohang <rh@vonng.com>
The #105 T3 change (PR #156) tried to keep the resident metadata cache
monotonic by guarding peer-reload publication on lastUpdate(). But
lastUpdate() is the max of per-config timestamps and cannot order whole
records: a node caching {policy@20, CORS@10} that receives a newer CORS@15
still has lastUpdate()==20, so the guard rejects the legitimately-newer
record and the periodic refresh (same comparator) cannot repair it. A
paused reload could also resurrect deleted resident state.
Per the maintainer decision, revert the reload publication to its original
unconditional (acceptable-until-refresh) behavior:
- remove setReloaded and restore the plain Set plus notification/target
registry updates in LoadBucketMetadataHandler;
- restore refreshBucketsMetadataLoop's own lastUpdate() staleness check and
globalEventNotifier.set / globalBucketTargetSys.set publication;
- restore the unconditional GetConfig cache-miss publication;
- document the known freshness limitation at the reload site (the periodic
refresh is best-effort and cannot repair an equal-maximum-timestamp
divergence).
The T1 lifecycle merge-under-lock (UpdateExpiryLCConfig) and both T2 fixes
(DeleteBucket takes metadata.lock before deleting; saveMetadata and
loadBucketMetadataParseUnderLock recheck physical bucket existence) are
kept fully intact.
Tests:
- drop the T3 reproductions (overlapping-reload resident-cache test and the
peer-reload-preserves-current-targets publication test);
- add lockBucketMetadataAcquireHook, a nil-in-production atomic test hook in
the shared metadata.lock path, so tests can deterministically observe a
caller (notably DeleteBucket, whose lock is taken through its
erasureServerPools receiver and is invisible to an injected object layer)
reaching the lock;
- rewrite the T2 delete-race ghost test to hold metadata.lock MID-SAVE (past
saveMetadata's existence recheck) and synchronize on the delete's actual
lock attempt via the hook, so it isolates the lock-before-delete fix:
removing only DeleteBucket's metadata.lock (recheck kept) now fails it;
- rewrite the cancellation test to observe the delete's actual lock attempt,
then cancel and await its error while still holding the lock, so a
scheduling-delayed delete stopped by the canceled context can no longer
pass on a broken tree.
Refs #105. Follows #156.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Signed-off-by: Feng Ruohang <rh@vonng.com>
Three residual checksum defects in the legacy etcd federation branch of
CopyObjectHandler, found by post-merge review of #157.
1. Empty-source 500 regression. A checksum-less object gains the S3 default
CRC-64NVME (WantServerSideChecksumType is set), but minio-go streams no
trailing checksum for a 0-byte body (contentLength == 0), so the remote
computed none, federatedChecksumValue was empty, hash.NewChecksumWithType
returned nil, and the handler returned 500 -- so every empty-object
federated copy failed. For a 0-byte source, forward the empty-content digest
as an ordinary checksum request header instead, so the remote validates,
persists and returns it, matching the local path (e.g. CRC32 "AAAAAA==").
2. Inherited full-object checksum dropped. When the source already carries a
full-object checksum, the local path sets dstOpts.WantChecksum, not
WantServerSideChecksumType (only multipart-composite sources are promoted).
The federated branch inspected only WantServerSideChecksumType, so a
checksum-bearing source's checksum was silently discarded on a federated
copy that requested no algorithm. Forward WantChecksum.Encoded (always a
plain digest) as a checksum header so the remote validates and persists it,
and bind the returned value, matching local persistence.
3. Multipart-suffixed remote value accepted. The bind accepted a value like
"NSRBwg==-0": NewChecksumWithType parses the "-N" as ChecksumMultipart with
WantParts 0 and the length-only validator passes, so the destination was
returned as COMPOSITE. A single forwarded PutObject must yield a full-object
digest, so reject a multipart-marked parsed value in addition to the
existing nil (missing/malformed) rejection.
A forwarded checksum request header is stripped from objInfo.UserDefined so it
is not mistaken for object metadata.
Out of scope: the SSE federated-copy corruption (srcInfo.Reader/Size mismatch
for encrypted sources) predates this work and is filed separately.
New federated regressions cover empty source with requested and default
checksum (200 + correct value + persisted), an inherited full-object checksum
preserved without a requested algorithm, and a multipart-suffixed remote value
rejected. Red/green verified for each against the merged code.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Signed-off-by: Feng Ruohang <rh@vonng.com>
Audit of the three deferred #105 follow-ups. Each reproduces with a
deterministic red test in cmd/bucket-metadata-race_test.go, and each fix is
the minimal change that turns its test green while preserving the
<bucket>.lck -> metadata.lock -> .metadata.bin lock order established by #103.
1. Lifecycle expiry merge lost update (persistent). PeerBucketLCConfigHandler
and healBucketILMExpiry read the current lifecycle with an unlocked
GetConfigFromDisk, merged the replicated expiry rules with the local
transition rules, then wrote the pre-computed blob via Update. Any lifecycle
transition change committed between the merge read and the merge write was
silently lost. New BucketMetadataSys.UpdateExpiryLCConfig performs the read,
merge, and save under one metadata.lock; mergeExpiryWithLCConfig now takes
the locked snapshot and validates object-lock retention from it instead of
re-reading (avoids a re-entrant metadata load under the lock).
2. DeleteBucket ghost .metadata.bin (persistent). DeleteBucket took only
<bucket>.lck while config writers take only metadata.lock, so a writer that
was mid-save could re-create .metadata.bin after the prefix purge. The purge
now runs under metadata.lock, with a best-effort unlocked fallback so a
delete is never blocked from completing.
3. Overlapping peer reloads publishing a stale resident cache (freshness only;
the persisted record stays correct). LoadBucketMetadataHandler and the
GetConfig cache-miss path published with an unconditional Set, so a reload
that read an older revision could overwrite a newer resident record until the
next refresh. New BucketMetadataSys.setReloaded (and a matching GetConfig
guard) refuses to regress a newer resident record, mirroring
refreshBucketsMetadataLoop.
Verification: go build -tags kqueue,dev ./...; go vet ./cmd; gofmt clean;
rebrand-guard baseline unchanged; go test -tags kqueue,dev ./cmd (207s) green;
new tests plus the #103 metadata suite green under -race.
Refs #105. Parent #102. Foundation #103.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01L7qJqWwy8oFA6aCXWRzXQe
Signed-off-by: Feng Ruohang <rh@vonng.com>
The legacy etcd federation branch of CopyObjectHandler forwards the copied
bytes with minio-go Core.PutObject but never asked the remote for a checksum
and discarded any it returned, so a cross-deployment whole-object copy that
requested a checksum returned 200 with an empty checksum, and a checksum-less
source did not gain the S3 default CRC-64NVME that the local path assigns. The
request was neither honored nor rejected. This is the whole-object counterpart
of #72, which repaired the same class of defect for federated UploadPartCopy.
When a server-side checksum is wanted -- explicitly requested, inherited from a
multipart source, or the CRC-64NVME default for a checksum-less object, all
already captured in dstOpts.WantServerSideChecksumType -- the forwarded
PutObject now streams a trailing checksum of that type, so the remote computes
and persists it and echoes it in the response. The value the remote reports for
that exact write is bound into objInfo.Checksum, matching how the local
CopyObject path carries checksums into the CopyObjectResult. Reading the value
from the same UploadInfo that produced the ETag keeps the pair bound to one
write.
Only the requested algorithm is returned; a malformed or absent remote value
leaves objInfo.Checksum unset, so an ordinary copy that wanted no checksum
still returns none. Two small mapping helpers convert between the server's
hash.ChecksumType and the minio-go request type and response field.
New end-to-end tests drive the real federation branch through
getRemoteInstanceClient and minio-go into a second in-process deployment and
assert that CRC32/CRC32C/SHA256/CRC64NVME and the no-algorithm default are all
returned in the CopyObjectResult and persisted on the destination, and that a
requested algorithm never leaks other algorithms into the response.
Fixes#99
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01L7qJqWwy8oFA6aCXWRzXQe
Signed-off-by: Feng Ruohang <rh@vonng.com>
The embedded Console reaches the S3/STS API at https://127.0.0.1:<port>
(minioConfigToConsoleFeatures), a loopback endpoint whose TLS certificate is
not expected to carry a 127.0.0.1 SAN. silo-console v2.3.x began verifying
every outbound TLS peer, so the Console's STS AssumeRole handshake to that
loopback endpoint now fails certificate validation and BOTH local and LDAP
logins fail with a generic "invalid login". The failure happens in the Console
HTTP client before any request reaches a server auth/STS/LDAP handler, so no
server-side auth error is logged, matching the report.
Restore the documented loopback bypass by opting the embedded Console into its
endpoint-scoped CONSOLE_MINIO_SERVER_TLS_SKIP_VERIFY switch whenever the server
falls back to the 127.0.0.1 endpoint under TLS. The exemption is scoped to that
single loopback origin inside Console; every other HTTPS peer (IdP, Prometheus,
webhooks) stays verified, preserving the v2.3.x hardening. An explicitly
configured endpoint is reached under its own verified name and is never
exempted. initConsoleServer unsets CONSOLE_* before re-deriving them, so the
switch cannot be supplied by the operator on the embedded path; the server must
assert it.
Fixes#108
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01L7qJqWwy8oFA6aCXWRzXQe
Signed-off-by: Feng Ruohang <rh@vonng.com>
The legacy etcd federation branch of CopyObjectHandler forwards the copied
source metadata to the remote deployment with minio-go Core.PutObject, after
removing only two reserved keys (compression and actual-size). Every small
object is stored inline, so its stored metadata also carries
X-Minio-Internal-inline-data; the remote's setRequestLimitMiddleware rejects
any request bearing a reserved-prefix header (containsReservedMetadata), so
the forwarded write failed with 400 InvalidArgument "Your metadata headers
are not supported." for the default COPY metadata directive.
A plain federated PutObject must not carry any internal storage metadata, so
strip the whole reserved-prefix class before forwarding instead of an
enumerated subset. Enumerating a third key would only defer the next leak:
besides inline-data, replication bookkeeping (replica/replication status and
timestamps) is added to the same map earlier in the handler and would be
rejected just the same. None of these keys is required by the remote for a
correct plain PutObject; they are internal storage details the remote sets
for itself. The stripping uses stringsHasPrefixFold, matching the remote's
own case-insensitive detection. Ordinary user metadata (x-amz-meta-*) is
untouched and still copied.
The pre-existing UUID ETag on the federated write (no Content-MD5 is sent) is
out of scope and left unchanged, as recorded in the issue.
A new end-to-end test drives the real federation branch through
getRemoteInstanceClient and minio-go into a second in-process deployment,
copying an inline source with the default COPY directive. It asserts the copy
now succeeds, that no reserved-prefix header reaches the remote on any
forwarded request, and that copied user metadata survives.
Fixes#100
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01L7qJqWwy8oFA6aCXWRzXQe
Signed-off-by: Feng Ruohang <rh@vonng.com>
PutObjectHandler's precondition callback ran DecryptObjectInfo on the
stored object before checkPreconditionsPUT, so an authenticated raw
SSE-C replica overwrite was rejected when the stored version could not
decrypt. A replica a pre-fix destination (issue #109) left as
compress(ciphertext) or a re-encrypted body has an invalid decrypted
length, so DecryptObjectInfo returned errObjectTampered and the
retransmission that repairs it never ran -- the version stayed damaged
through resync. #134's raw-replica exemption only covered the
version/ETag duplicate check inside checkPreconditionsPUT, one step too
late.
Skip the stored object's decryption precondition only for a PURE raw
SSE-C replica overwrite (a trusted SSE-C replica write with no public
precondition), keyed on the incoming request's restored SSE-C metadata,
the same predicate checkPreconditionsPUT uses. Such a write fully
replaces the object, so requiring the damaged stored version to decrypt
is both wrong and unnecessary. A conditional request keeps the check:
DecryptObjectInfo also normalizes the stored sealed ETag to the
client-visible one, and If-Match/If-None-Match must compare against
that, not the sealed ETag -- skipping it for every replica inverted both
conditions. Ordinary writes and non-SSE-C replicas are unchanged.
Adds red/green regressions: a raw retransmit over a version staged as an
undecodable body returns 500 XMinioObjectTampered before this change and
200 with full customer-key recovery after; and a conditional replica PUT
(If-Match / If-None-Match) on the client-visible ETag is honoured rather
than inverted. Fixes the single-PUT compression-damage recovery gap in
Signed-off-by: Feng Ruohang <rh@vonng.com>
#120.
Rebased onto current main. #119 made PutObjectPart derive an encrypted
part's plaintext length and reject a part that cannot be a valid sio
stream, so TestReplicaLockReconcileNullVersion's completeNullMPU fixture
(a 4-byte plaintext part under SSE-C metadata) no longer stores; build it
with sio.Encrypt like the other encrypted-part fixtures. Also regenerate
the rebrand-guard baseline for the replication SSE header the retransmit
path reintroduces (headers 84 -> 85). Mechanical integration only.
Signed-off-by: Feng Ruohang <rh@vonng.com>
Feng Ruohang