fix(listing): snapshot asynchronous cache cleanup inputs

Capture the metacache peer before launching cleanup and pass listing
options by value. The parent can clear its cache ID without changing the
background RPC lookup or racing with its reads of shared request state.

Publish complete immutable disk snapshots through atomic.Value in the
listing fixtures instead of replacing getDisks after requests start.
Add a real peer RPC regression for cleanup using the original cache ID;
reverting the options copy reproduces both the race and an empty RPC ID.

Verified the final 18-test listing race suite, 30 repetitions of the real
peer regression, formatting checks and make lint. Prior runtime evidence
retains its original e90cb206 source identity.

Refs: https://github.com/pgsty/silo/issues/218
Refs: https://github.com/pgsty/silo/pull/239
Signed-off-by: Feng Ruohang <rh@vonng.com>

.github/workflows/go.yml

@@ -96,12 +96,20 @@ jobs:
       - name: Run multipart listing and cancellation tests under race detector
         run: go test -race ./cmd -run '^Test(MultipartListing|MultipartAbort|PaginateMultipartUploads|ListMultipartUploads)' -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|MergeXLV2LatestNullCrossParity|CountXLV2MixedECGroups|ListCrossParity)' -count=1 -timeout=5m
+
       - name: Run CPU metrics tests under race detector
         run: go test -race ./cmd -run '^TestLoadCPUMetrics' -count=1 -timeout=5m
 
       - name: Run tag replication tests under race detector
         run: go test -race ./cmd -run '^TestAPITagging' -count=1 -timeout=5m
 
+      - name: Run Object Lock authorization and overwrite tests under race detector
+        env:
+          MINIO_API_REQUESTS_MAX: "10000"
+        run: go test -race -tags kqueue,dev ./cmd -run '^Test(APIPutObjectRetention|APIVersionAddressedWrite|PutObjectRetention|DirectoryObjectLock|ObjectLockOverwrite|APICopyObjectPreservesVersionAfterSuspension|MultipartNullObjectLock|PoolsObjectLockOverwrite|MultipartVersioningTransitions|PlanReview|RecoveredMultipartUpload)' -count=1 -timeout=5m
+
   crosscompile:
     name: Cross Compile
     runs-on: ubuntu-latest

.github/workflows/release.yml

@@ -70,6 +70,12 @@ jobs:
           buildscripts/verify-rebrand.sh
           dockerscripts/docker-entrypoint_test.sh
 
+      # Keep this gate aligned with the Object Lock race regression in go.yml.
+      - name: Gate release on Object Lock authorization and overwrite tests under race detector
+        env:
+          MINIO_API_REQUESTS_MAX: "10000"
+        run: go test -race -tags kqueue,dev ./cmd -run '^Test(APIPutObjectRetention|APIVersionAddressedWrite|PutObjectRetention|DirectoryObjectLock|ObjectLockOverwrite|APICopyObjectPreservesVersionAfterSuspension|MultipartNullObjectLock|PoolsObjectLockOverwrite|MultipartVersioningTransitions|PlanReview|RecoveredMultipartUpload)' -count=1 -timeout=5m
+
       - name: Compute release variables
         env:
           # Passed through the environment, never interpolated into the script

Dockerfile.goreleaser

@@ -1,4 +1,4 @@
-FROM golang:1.27.1-alpine@sha256:cf6fca6641884b8433441b2b0652976f975e1d0fdd26d177eaaf8596087f3125 AS curl-build
+FROM golang:1.27.2-alpine@sha256:f92b6ef800e499660581efdabdf25d9d817a9d124eaf900924f0504e7e27e12d AS curl-build
 ARG TARGETARCH
 COPY dockerscripts/build-static-curl.sh /build/build-static-curl
 RUN /bin/sh /build/build-static-curl
@@ -9,7 +9,7 @@ COPY --from=curl-build /go/bin/curl /curl
 COPY --from=curl-build /etc/ssl/certs/ca-certificates.crt /etc/ssl/certs/ca-certificates.crt
 ENTRYPOINT ["/curl"]
 
-FROM golang:1.27.1-alpine@sha256:cf6fca6641884b8433441b2b0652976f975e1d0fdd26d177eaaf8596087f3125 AS build
+FROM golang:1.27.2-alpine@sha256:f92b6ef800e499660581efdabdf25d9d817a9d124eaf900924f0504e7e27e12d AS build
 
 ARG TARGETARCH
 

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 ?= v2.13.1
+GOLANGCI_VERSION ?= v2.14.0
 
 VERSION ?= $(shell git describe --tags)
 REPO ?= docker.io/pgsty

buildscripts/checkdeps.sh

@@ -7,7 +7,7 @@ _init() {
 
 	## Minimum required versions for build dependencies
 	GIT_VERSION="1.0"
-	GO_VERSION="1.27.1"
+	GO_VERSION="1.27.2"
 	OSX_VERSION="10.8"
 	KNAME=$(uname -s)
 	ARCH=$(uname -m)

cmd/api-errors-checksum_test.go

@@ -0,0 +1,45 @@
+// Copyright (c) 2026 PGSTY
+//
+// This file is part of SILO Object Storage.
+// SPDX-License-Identifier: AGPL-3.0-or-later
+//
+// 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"
+	"net/http"
+	"testing"
+
+	"github.com/minio/minio/internal/hash"
+)
+
+func TestChecksumMismatchAPIMessage(t *testing.T) {
+	for _, algorithm := range []string{"CRC32", "CRC32C", "SHA1", "SHA256", "CRC64NVME", ""} {
+		t.Run(algorithm, func(t *testing.T) {
+			mismatch := hash.ChecksumMismatch{Algorithm: algorithm, Want: "expected", Got: "actual"}
+			want := errorCodes.ToAPIErr(ErrContentChecksumMismatch).Description
+			if algorithm != "" {
+				want = fmt.Sprintf("The %s you specified did not match the calculated checksum.", algorithm)
+			}
+			for _, err := range []error{mismatch, fmt.Errorf("upload: %w", mismatch)} {
+				got := toAPIError(t.Context(), err)
+				if got.HTTPStatusCode != http.StatusBadRequest || got.Code != "XAmzContentChecksumMismatch" || got.Description != want {
+					t.Fatalf("toAPIError(%v) = %+v, want 400 XAmzContentChecksumMismatch: %s", err, got, want)
+				}
+			}
+		})
+	}
+}

cmd/api-errors.go

@@ -550,7 +550,7 @@ var errorCodes = errorCodeMap{
 	},
 	ErrBadDigest: {
 		Code:           "BadDigest",
-		Description:    "The Content-Md5 you specified did not match what we received.",
+		Description:    "The Content-MD5 you specified did not match what we received.",
 		HTTPStatusCode: http.StatusBadRequest,
 	},
 	ErrEntityTooSmall: {
@@ -2505,6 +2505,14 @@ func toAPIError(ctx context.Context, err error) APIError {
 	case errors.Is(err, errMissingPartChecksum):
 		apiErr.Description = strings.TrimPrefix(err.Error(), errMissingPartChecksum.Error()+": ")
 	}
+	// PutObject/UploadPart checksum mismatches name the algorithm in the
+	// message, matching S3 (e.g. "The CRC32 you specified did not match the
+	// calculated checksum."). Falls through to the generic table description
+	// when the algorithm is unknown.
+	var checksumMismatch hash.ChecksumMismatch
+	if errors.As(err, &checksumMismatch) && checksumMismatch.Algorithm != "" {
+		apiErr.Description = fmt.Sprintf("The %s you specified did not match the calculated checksum.", checksumMismatch.Algorithm)
+	}
 	switch apiErr.Code {
 	case "NotImplemented":
 		apiErr = APIError{

cmd/auth-handler.go

@@ -28,13 +28,10 @@ import (
 	"mime"
 	"net/http"
 	"net/url"
-	"strconv"
 	"strings"
 	"sync/atomic"
-	"time"
 
 	"github.com/minio/minio/internal/auth"
-	objectlock "github.com/minio/minio/internal/bucket/object/lock"
 	"github.com/minio/minio/internal/etag"
 	"github.com/minio/minio/internal/hash"
 	xhttp "github.com/minio/minio/internal/http"
@@ -733,48 +730,6 @@ func setAuthMiddleware(h http.Handler) http.Handler {
 	})
 }
 
-func isPutRetentionAllowed(bucketName, objectName string, retDays int, retDate time.Time, retMode objectlock.RetMode, byPassSet bool, r *http.Request, cred auth.Credentials, owner bool) (s3Err APIErrorCode) {
-	var retSet bool
-	if cred.AccessKey == "" {
-		return ErrAccessDenied
-	}
-
-	conditions := getConditionValues(r, "", cred)
-	conditions["object-lock-mode"] = []string{string(retMode)}
-	conditions["object-lock-retain-until-date"] = []string{retDate.UTC().Format(time.RFC3339)}
-	if retDays > 0 {
-		conditions["object-lock-remaining-retention-days"] = []string{strconv.Itoa(retDays)}
-	}
-	if retMode == objectlock.RetGovernance && byPassSet {
-		byPassSet = globalIAMSys.IsAllowed(policy.Args{
-			AccountName:     cred.AccessKey,
-			Groups:          cred.Groups,
-			Action:          policy.BypassGovernanceRetentionAction,
-			BucketName:      bucketName,
-			ObjectName:      objectName,
-			ConditionValues: conditions,
-			IsOwner:         owner,
-			Claims:          cred.Claims,
-		})
-	}
-	if globalIAMSys.IsAllowed(policy.Args{
-		AccountName:     cred.AccessKey,
-		Groups:          cred.Groups,
-		Action:          policy.PutObjectRetentionAction,
-		BucketName:      bucketName,
-		ConditionValues: conditions,
-		ObjectName:      objectName,
-		IsOwner:         owner,
-		Claims:          cred.Claims,
-	}) {
-		retSet = true
-	}
-	if byPassSet || retSet {
-		return ErrNone
-	}
-	return ErrAccessDenied
-}
-
 // isPutActionAllowed - check if PUT operation is allowed on the resource, this
 // call verifies bucket policies and IAM policies, supports multi user
 // checks etc.

cmd/batch-handlers.go

@@ -583,7 +583,11 @@ func toObjectInfo(bucket, object string, objInfo minio.ObjectInfo) ObjectInfo {
 func (r BatchJobReplicateV1) writeAsArchive(ctx context.Context, objAPI ObjectLayer, remoteClnt *minio.Client, entries []ObjectInfo, prefix string) error {
 	input := make(chan minio.SnowballObject, 1)
 	opts := minio.SnowballOptions{
-		Opts:     minio.PutObjectOptions{},
+		Opts: minio.PutObjectOptions{
+			Internal: minio.AdvancedPutOptions{
+				ReplicationRequest: true,
+			},
+		},
 		InMemory: *r.Source.Snowball.InMemory,
 		Compress: *r.Source.Snowball.Compress,
 		SkipErrs: *r.Source.Snowball.SkipErrs,

cmd/bucket-handlers.go

@@ -1273,6 +1273,12 @@ func (api objectAPIHandlers) PostPolicyBucketHandler(w http.ResponseWriter, r *h
 		writeErrorResponseHeadersOnly(w, toAPIError(ctx, err))
 		return
 	}
+	if isDirObject(object) {
+		if _, _, _, s3Err := checkPutObjectLockAllowed(ctx, r, bucket, object, objectAPI.GetObjectInfo, ErrNone, ErrNone, false); s3Err != ErrNone {
+			writeErrorResponse(ctx, w, errorCodes.ToAPIErr(s3Err), r.URL)
+			return
+		}
+	}
 	opts.WantChecksum = checksum
 
 	fanOutOpts := fanOutOptions{Checksum: checksum}

cmd/bucket-object-lock.go

@@ -20,8 +20,8 @@ package cmd
 import (
 	"context"
 	"errors"
-	"math"
 	"net/http"
+	"strconv"
 	"strings"
 	"time"
 
@@ -164,76 +164,129 @@ func enforceRetentionBypassForDelete(ctx context.Context, r *http.Request, bucke
 	return nil
 }
 
-// enforceRetentionBypassForPut enforces whether an existing object under governance can be overwritten
-// with governance bypass headers set in the request.
-// Objects under site wide WORM cannot be overwritten.
-// For objects in "Governance" mode, overwrite is allowed if a) object retention date is past OR
-// governance bypass headers are set and user has governance bypass permissions.
-// Objects in compliance mode can be overwritten only if retention date is being extended. No mode change is permitted.
+// isRetentionActionAllowed evaluates action for a PutObjectRetention request
+// with the object-lock condition keys derived from the requested retention.
+// The keys come from the request, not from the stored version, so the
+// handler can decide s3:PutObjectRetention before reading any object state.
+func isRetentionActionAllowed(action policy.Action, bucket, object string, objRetention *objectlock.ObjectRetention, r *http.Request, cred auth.Credentials, owner bool) bool {
+	if cred.AccessKey == "" {
+		return false
+	}
+	t, err := objectlock.UTCNowNTP()
+	if err != nil {
+		return false
+	}
+	conditions := getConditionValues(r, "", cred)
+	conditions["object-lock-mode"] = []string{string(objRetention.Mode)}
+	conditions["object-lock-retain-until-date"] = []string{objRetention.RetainUntilDate.UTC().Format(time.RFC3339)}
+	// Relative days from now, for "object-lock-remaining-retention-days".
+	// Clearing retention (<Retention/>) has a zero RetainUntilDate, which must
+	// evaluate as 0 remaining days rather than a large positive math.Abs delta
+	// from year 0001.
+	days := 0
+	if !objRetention.RetainUntilDate.IsZero() && objRetention.RetainUntilDate.After(t) {
+		// Time.Sub saturates after about 292 years. Count seconds instead so
+		// maximum-retention policies also apply to distant RFC3339 dates.
+		seconds := objRetention.RetainUntilDate.Unix() - t.Unix()
+		days = int(seconds / 86400)
+		if seconds%86400 != 0 || objRetention.RetainUntilDate.Nanosecond() > t.Nanosecond() {
+			days++
+		}
+	}
+	conditions["object-lock-remaining-retention-days"] = []string{strconv.Itoa(days)}
+	return globalIAMSys.IsAllowed(policy.Args{
+		AccountName:     cred.AccessKey,
+		Groups:          cred.Groups,
+		Action:          action,
+		BucketName:      bucket,
+		ConditionValues: conditions,
+		ObjectName:      object,
+		IsOwner:         owner,
+		Claims:          cred.Claims,
+	})
+}
+
+// enforceRetentionBypassForPut enforces the stored retention of oi against a
+// PutObjectRetention request. The caller must already hold
+// s3:PutObjectRetention; this only decides changes that depend on the stored
+// state. Expired or absent retention may be changed freely. Unexpired
+// GOVERNANCE retention may be shortened, removed or have its mode changed only
+// with the bypass header and s3:BypassGovernanceRetention. Unexpired
+// COMPLIANCE retention may only be extended.
 func enforceRetentionBypassForPut(ctx context.Context, r *http.Request, oi ObjectInfo, objRetention *objectlock.ObjectRetention, cred auth.Credentials, owner bool) error {
-	byPassSet := objectlock.IsObjectLockGovernanceBypassSet(r.Header)
+	ret := objectlock.GetObjectRetentionMeta(oi.UserDefined)
+	if !ret.Mode.Valid() {
+		return nil
+	}
 
 	t, err := objectlock.UTCNowNTP()
 	if err != nil {
 		internalLogIf(ctx, err, logger.WarningKind)
 		return ObjectLocked{Bucket: oi.Bucket, Object: oi.Name, VersionID: oi.VersionID}
 	}
+	if ret.RetainUntilDate.Before(t) {
+		return nil
+	}
 
-	// Pass in relative days from current time, to additionally
-	// to verify "object-lock-remaining-retention-days" policy if any.
-	days := int(math.Ceil(math.Abs(objRetention.RetainUntilDate.Sub(t).Hours()) / 24))
-
-	ret := objectlock.GetObjectRetentionMeta(oi.UserDefined)
-	if ret.Mode.Valid() {
-		// Retention has expired you may change whatever you like.
-		if ret.RetainUntilDate.Before(t) {
-			apiErr := isPutRetentionAllowed(oi.Bucket, oi.Name,
-				days, objRetention.RetainUntilDate.Time,
-				objRetention.Mode, byPassSet, r, cred,
-				owner)
-			if apiErr == ErrAccessDenied {
-				return errAuthentication
-			}
-			return nil
+	weakens := objRetention.Mode != ret.Mode || objRetention.RetainUntilDate.Before(ret.RetainUntilDate.Time)
+	if !weakens {
+		return nil
+	}
+	switch ret.Mode {
+	case objectlock.RetGovernance:
+		// https://docs.aws.amazon.com/AmazonS3/latest/userguide/object-lock.html#object-lock-bypass-governance
+		if !objectlock.IsObjectLockGovernanceBypassSet(r.Header) {
+			return ObjectLocked{Bucket: oi.Bucket, Object: oi.Name, VersionID: oi.VersionID}
 		}
-
-		switch ret.Mode {
-		case objectlock.RetGovernance:
-			govPerm := isPutRetentionAllowed(oi.Bucket, oi.Name, days,
-				objRetention.RetainUntilDate.Time, objRetention.Mode,
-				byPassSet, r, cred, owner)
-			// Governance mode retention period cannot be shortened, if x-amz-bypass-governance is not set.
-			if !byPassSet {
-				if objRetention.Mode != objectlock.RetGovernance || objRetention.RetainUntilDate.Before(ret.RetainUntilDate.Time) {
-					return ObjectLocked{Bucket: oi.Bucket, Object: oi.Name, VersionID: oi.VersionID}
-				}
-			}
-			if govPerm == ErrAccessDenied {
-				return errAuthentication
-			}
-			return nil
-		case objectlock.RetCompliance:
-			// Compliance retention mode cannot be changed or shortened.
-			// https://docs.aws.amazon.com/AmazonS3/latest/dev/object-lock-overview.html#object-lock-retention-modes
-			if objRetention.Mode != objectlock.RetCompliance || objRetention.RetainUntilDate.Before(ret.RetainUntilDate.Time) {
-				return ObjectLocked{Bucket: oi.Bucket, Object: oi.Name, VersionID: oi.VersionID}
-			}
-			apiErr := isPutRetentionAllowed(oi.Bucket, oi.Name,
-				days, objRetention.RetainUntilDate.Time, objRetention.Mode,
-				false, r, cred, owner)
-			if apiErr == ErrAccessDenied {
-				return errAuthentication
-			}
-			return nil
+		if !isRetentionActionAllowed(policy.BypassGovernanceRetentionAction, oi.Bucket, oi.Name, objRetention, r, cred, owner) {
+			return errAuthentication
 		}
 		return nil
-	} // No pre-existing retention metadata present.
+	default:
+		// Compliance retention mode cannot be changed or shortened.
+		// https://docs.aws.amazon.com/AmazonS3/latest/dev/object-lock-overview.html#object-lock-retention-modes
+		return ObjectLocked{Bucket: oi.Bucket, Object: oi.Name, VersionID: oi.VersionID}
+	}
+}
 
-	apiErr := isPutRetentionAllowed(oi.Bucket, oi.Name,
-		days, objRetention.RetainUntilDate.Time,
-		objRetention.Mode, byPassSet, r, cred, owner)
-	if apiErr == ErrAccessDenied {
-		return errAuthentication
+// checkObjectLockedForOverwrite re-checks under the object write lock whether an
+// in-place write to the null version (such as a directory object) would
+// overwrite a locked version in an Object Lock bucket.
+func checkObjectLockedForOverwrite(ctx context.Context, bucket, object string, getObjectInfoFn GetObjectInfoFn) error {
+	if isMinioMetaBucketName(bucket) {
+		return nil
+	}
+	retentionCfg, err := globalBucketObjectLockSys.Get(bucket)
+	if err != nil {
+		return err
+	}
+	if !retentionCfg.LockEnabled {
+		return nil
+	}
+	objInfo, err := getObjectInfoFn(ctx, bucket, object, ObjectOptions{
+		VersionID:  nullVersionID,
+		NoLock:     true,
+		NoAuditLog: true,
+	})
+	if err != nil {
+		if isErrObjectNotFound(err) || isErrVersionNotFound(err) || errors.As(err, &MethodNotAllowed{}) {
+			return nil
+		}
+		return err
+	}
+	if objectlock.GetObjectLegalHoldMeta(objInfo.UserDefined).Status == objectlock.LegalHoldOn {
+		return ObjectLocked{Bucket: bucket, Object: decodeDirObject(object), VersionID: nullVersionID}
+	}
+	r := objectlock.GetObjectRetentionMeta(objInfo.UserDefined)
+	if r.Mode == objectlock.RetCompliance || r.Mode == objectlock.RetGovernance {
+		t, err := objectlock.UTCNowNTP()
+		if err != nil {
+			internalLogIf(ctx, err, logger.WarningKind)
+			return ObjectLocked{Bucket: bucket, Object: decodeDirObject(object), VersionID: nullVersionID}
+		}
+		if r.RetainUntilDate.After(t) {
+			return ObjectLocked{Bucket: bucket, Object: decodeDirObject(object), VersionID: nullVersionID}
+		}
 	}
 	return nil
 }
@@ -274,16 +327,24 @@ func checkPutObjectLockAllowed(ctx context.Context, rq *http.Request, bucket, ob
 	if err != nil {
 		return mode, retainDate, legalHold, toAPIErrorCode(ctx, err)
 	}
+	if isDirObject(object) && opts.VersionID == "" {
+		opts.VersionID = nullVersionID
+	}
 
 	if opts.VersionID != "" && !replicaTrusted {
-		if objInfo, err := getObjectInfoFn(ctx, bucket, object, opts); err == nil {
+		objInfo, err := getObjectInfoFn(ctx, bucket, object, opts)
+		if err != nil {
+			if !isErrObjectNotFound(err) && !isErrVersionNotFound(err) && !errors.As(err, &MethodNotAllowed{}) {
+				return mode, retainDate, legalHold, toAPIErrorCode(ctx, err)
+			}
+		} else {
 			r := objectlock.GetObjectRetentionMeta(objInfo.UserDefined)
 			t, err := objectlock.UTCNowNTP()
 			if err != nil {
 				internalLogIf(ctx, err, logger.WarningKind)
 				return mode, retainDate, legalHold, ErrObjectLocked
 			}
-			if r.Mode == objectlock.RetCompliance && r.RetainUntilDate.After(t) {
+			if (r.Mode == objectlock.RetCompliance || r.Mode == objectlock.RetGovernance) && r.RetainUntilDate.After(t) {
 				return mode, retainDate, legalHold, ErrObjectLocked
 			}
 			mode = r.Mode

cmd/data-scanner.go

@@ -744,10 +744,10 @@ func (f *folderScanner) scanFolder(ctx context.Context, folder cachedFolder, int
 					}
 				},
 				// Some disks have data for this.
-				partial: func(entries metaCacheEntries, errs []error) {
+				partial: func(entries metaCacheEntries, errs []error) error {
 					if !f.shouldHeal() {
 						cancel()
-						return
+						return nil
 					}
 					entry, ok := entries.resolve(&resolver)
 					if !ok {
@@ -772,7 +772,7 @@ func (f *folderScanner) scanFolder(ctx context.Context, folder cachedFolder, int
 					}
 
 					if entry.isDir() {
-						return
+						return nil
 					}
 
 					// We got an entry which we should be able to heal.
@@ -787,7 +787,7 @@ func (f *folderScanner) scanFolder(ctx context.Context, folder cachedFolder, int
 							scannerLogIf(ctx, err)
 						}
 						foundObjs = foundObjs || err == nil
-						return
+						return nil
 					}
 
 					custom["versions"] = fmt.Sprint(len(fiv.Versions))
@@ -814,6 +814,7 @@ func (f *folderScanner) scanFolder(ctx context.Context, folder cachedFolder, int
 					}
 					custom["success_versions"] = fmt.Sprint(successVersions)
 					custom["failed_versions"] = fmt.Sprint(failVersions)
+					return nil
 				},
 				// Too many disks failed.
 				finished: func(errs []error) {

cmd/erasure-healing.go

@@ -86,7 +86,7 @@ func (er erasureObjects) listAndHeal(ctx context.Context, bucket, prefix string,
 				cancel()
 			}
 		},
-		partial: func(entries metaCacheEntries, _ []error) {
+		partial: func(entries metaCacheEntries, _ []error) error {
 			entry, ok := entries.resolve(&resolver)
 			if !ok {
 				// check if we can get one entry at least
@@ -94,12 +94,12 @@ func (er erasureObjects) listAndHeal(ctx context.Context, bucket, prefix string,
 				entry, _ = entries.firstFound()
 			}
 			if !recursive && prefix != entry.name {
-				return
+				return nil
 			}
 			if err := healEntry(bucket, *entry, scanMode); err != nil {
 				cancel()
-				return
 			}
+			return nil
 		},
 		finished: nil,
 	}

cmd/erasure-metadata-utils_test.go

@@ -20,6 +20,7 @@ package cmd
 import (
 	"context"
 	"encoding/hex"
+	"errors"
 	"fmt"
 	"math/rand"
 	"reflect"
@@ -110,6 +111,96 @@ func TestReduceErrs(t *testing.T) {
 	}
 }
 
+// TestReduceWriteQuorumErrsCommitMetaMissing verifies that an in-quorum
+// errCommitMetaMissing reduces to the sentinel, which renameData() converts
+// to errErasureWriteQuorum, while below-quorum reduces directly to
+// errErasureWriteQuorum.
+func TestReduceWriteQuorumErrsCommitMetaMissing(t *testing.T) {
+	errs := make([]error, 16)
+	for i := range 9 {
+		errs[i] = errCommitMetaMissing
+	}
+	if err := reduceWriteQuorumErrs(t.Context(), errs, objectOpIgnoredErrs, 9); !errors.Is(err, errCommitMetaMissing) {
+		t.Errorf("expected %s, got %s", errCommitMetaMissing, err)
+	}
+	// The sentinel must be distinguishable from errFileNotFound so that
+	// commit failures are never mapped to ObjectNotFound (404).
+	if errors.Is(errCommitMetaMissing, errFileNotFound) {
+		t.Error("errCommitMetaMissing must not match errFileNotFound")
+	}
+
+	// Below quorum the reduction must yield a write-quorum failure (503);
+	// eight sentinels against eight committed drives is not a quorum.
+	belowQuorum := make([]error, 16)
+	for i := range 8 {
+		belowQuorum[i] = errCommitMetaMissing
+	}
+	if err := reduceWriteQuorumErrs(t.Context(), belowQuorum, objectOpIgnoredErrs, 9); !errors.Is(err, errErasureWriteQuorum) {
+		t.Errorf("expected %s, got %s", errErasureWriteQuorum, err)
+	}
+}
+
+// renameDataFailingDisk is a StorageAPI stub returning a fixed error
+// from RenameData, enough to drive renameData() in unit tests.
+type renameDataFailingDisk struct {
+	StorageAPI
+	err       error
+	undoWrite bool
+}
+
+func (d *renameDataFailingDisk) RenameData(context.Context, string, string, FileInfo, string, string, RenameOptions) (RenameDataResp, error) {
+	return RenameDataResp{}, d.err
+}
+
+func (d *renameDataFailingDisk) DeleteVersion(_ context.Context, _, _ string, _ FileInfo, _ bool, opts DeleteOptions) error {
+	d.undoWrite = opts.UndoWrite
+	return nil
+}
+
+// TestRenameDataCommitMetaMissing verifies that renameData() converts an
+// in-quorum errCommitMetaMissing into errErasureWriteQuorum (HTTP 503),
+// while an in-quorum errFileNotFound keeps its meaning so that
+// PutObject() still reports IncompleteBody for client disconnects.
+func TestRenameDataCommitMetaMissing(t *testing.T) {
+	for _, tc := range []struct {
+		name     string
+		failures int
+		diskErr  error
+		wantErr  error
+	}{
+		{"all_commit_metadata_missing", 16, errCommitMetaMissing, errErasureWriteQuorum},
+		{"quorum_commit_metadata_missing", 9, errCommitMetaMissing, errErasureWriteQuorum},
+		{"neither_side_has_quorum", 8, errCommitMetaMissing, errErasureWriteQuorum},
+		{"successful_write_quorum", 7, errCommitMetaMissing, nil},
+		{"staging_failure", 16, errFileNotFound, errFileNotFound},
+	} {
+		t.Run(tc.name, func(t *testing.T) {
+			metadata := make([]FileInfo, 16)
+			disks := make([]StorageAPI, 16)
+			for i := range disks {
+				metadata[i] = FileInfo{Deleted: true} // valid without erasure info
+				disk := &renameDataFailingDisk{}
+				if i < tc.failures {
+					disk.err = tc.diskErr
+				}
+				disks[i] = disk
+			}
+			_, _, _, err := renameData(t.Context(), disks, minioMetaTmpBucket, "tmp", metadata, "bucket", "object", 9)
+			if !errors.Is(err, tc.wantErr) {
+				t.Fatalf("expected %v, got %v", tc.wantErr, err)
+			}
+			for i, disk := range disks {
+				// Failed writes undo only the drives that committed. A
+				// successful quorum must never be rolled back.
+				wantUndo := tc.wantErr != nil && i >= tc.failures
+				if got := disk.(*renameDataFailingDisk).undoWrite; got != wantUndo {
+					t.Errorf("disk %d: UndoWrite = %v, want %v", i, got, wantUndo)
+				}
+			}
+		})
+	}
+}
+
 // TestHashOrder - test order of ints in array
 func TestHashOrder(t *testing.T) {
 	testCases := []struct {

cmd/erasure-multipart-cleanup_test.go

@@ -0,0 +1,177 @@
+// Copyright (c) 2026 clown15
+//
+// 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"
+	"encoding/base64"
+	"fmt"
+	"net/url"
+	"os"
+	"path/filepath"
+	"testing"
+	"time"
+)
+
+func TestExtractModTimeFromUploadIDDir(t *testing.T) {
+	now := time.Unix(1792056747, 123456789).UTC()
+	nano := now.UnixNano()
+	uuid := "d16ad42a-6fe5-4a3b-8092-69983142238f"
+
+	testCases := []struct {
+		name        string
+		dirName     string
+		expectedMod time.Time
+		wantZero    bool
+	}{
+		{
+			name:        "Modern Plaintext Format (<UUID>x<UnixNano>)",
+			dirName:     fmt.Sprintf("%sx%d", uuid, nano),
+			expectedMod: time.Unix(0, nano),
+			wantZero:    false,
+		},
+		{
+			// This valid UUID decodes to bytes containing 'x'. Checking
+			// for that byte before parsing plaintext loses the timestamp.
+			name:        "Plaintext Also Decodes To Base64 Containing x",
+			dirName:     fmt.Sprintf("4eacd477-e733-4851-b821-6fd42785ff87x%d", nano),
+			expectedMod: now,
+		},
+		{
+			name:        "Legacy Format Without Timestamp (<UUID>)",
+			dirName:     uuid,
+			expectedMod: time.Time{},
+			wantZero:    true,
+		},
+		{
+			name:        "Base64 Encoded Format",
+			dirName:     base64.RawURLEncoding.EncodeToString([]byte(fmt.Sprintf("%sx%d", uuid, nano))),
+			expectedMod: time.Unix(0, nano),
+			wantZero:    false,
+		},
+		{
+			name:        "Encoded Format With Deployment Prefix",
+			dirName:     base64.RawURLEncoding.EncodeToString([]byte(fmt.Sprintf("%s.%sx%d", uuid, uuid, nano))),
+			expectedMod: now,
+		},
+		{
+			name:        "Malformed Suffix (<UUID>xinvalid)",
+			dirName:     fmt.Sprintf("%sxnotanumber", uuid),
+			expectedMod: time.Time{},
+			wantZero:    true,
+		},
+		{
+			name:     "Timestamp Overflow",
+			dirName:  uuid + "x9223372036854775808",
+			wantZero: true,
+		},
+		{
+			name:     "Malformed Plaintext UUID",
+			dirName:  fmt.Sprintf("not-a-uuidx%d", nano),
+			wantZero: true,
+		},
+		{
+			name:        "Empty Directory Name",
+			dirName:     "",
+			expectedMod: time.Time{},
+			wantZero:    true,
+		},
+	}
+
+	for _, tc := range testCases {
+		t.Run(tc.name, func(t *testing.T) {
+			got := extractModTimeFromUploadIDDir(tc.dirName)
+			if tc.wantZero {
+				if !got.IsZero() {
+					t.Fatalf("expected zero time for %q, got %v", tc.dirName, got)
+				}
+				return
+			}
+			if !got.Equal(tc.expectedMod) {
+				t.Fatalf("extractModTimeFromUploadIDDir(%q) = %v; want %v", tc.dirName, got, tc.expectedMod)
+			}
+		})
+	}
+}
+
+type multipartCleanupDisk struct {
+	StorageAPI
+	root    string
+	modTime time.Time
+	readErr error
+	reads   int
+}
+
+func (d *multipartCleanupDisk) Endpoint() Endpoint {
+	return Endpoint{URL: &url.URL{Path: d.root}}
+}
+
+func (d *multipartCleanupDisk) ReadVersion(context.Context, string, string, string, string, ReadOptions) (FileInfo, error) {
+	d.reads++
+	return FileInfo{ModTime: d.modTime}, d.readErr
+}
+
+func (d *multipartCleanupDisk) StatVol(context.Context, string) (VolInfo, error) {
+	return VolInfo{Created: time.Now()}, nil
+}
+
+func TestCleanupStaleUploadsOnDisk(t *testing.T) {
+	const uuid = "d16ad42a-6fe5-4a3b-8092-69983142238f"
+	expiry := globalAPIConfig.getStaleUploadsExpiry()
+	for _, tc := range []struct {
+		name, prefix string
+		age          time.Duration
+		legacy       bool
+		readErr      error
+		wantReads    int
+		wantDeleted  bool
+	}{
+		{name: "recent", prefix: uuid, age: expiry / 2, readErr: errFileNotFound},
+		{name: "plaintext_base64_collision", prefix: "4eacd477-e733-4851-b821-6fd42785ff87", age: expiry / 2, readErr: errFileNotFound},
+		{name: "expired", prefix: uuid, age: 2 * expiry, readErr: errFileNotFound, wantDeleted: true},
+		{name: "legacy_recent", prefix: uuid, age: expiry / 2, legacy: true, wantReads: 1},
+		{name: "legacy_expired", prefix: uuid, age: 2 * expiry, legacy: true, wantReads: 1, wantDeleted: true},
+		{name: "legacy_metadata_missing", prefix: uuid, age: 2 * expiry, legacy: true, readErr: errFileNotFound, wantReads: 1},
+		{name: "malformed_metadata_missing", prefix: "not-a-uuid", age: 2 * expiry, readErr: errFileNotFound, wantReads: 1},
+	} {
+		t.Run(tc.name, func(t *testing.T) {
+			modTime := time.Now().Add(-tc.age)
+			id := tc.prefix
+			if !tc.legacy {
+				id = fmt.Sprintf("%sx%d", tc.prefix, modTime.UnixNano())
+			}
+			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 {
+				t.Fatal(err)
+			}
+			erasureObjects{}.cleanupStaleUploadsOnDisk(t.Context(), disk)
+			if disk.reads != tc.wantReads {
+				t.Errorf("ReadVersion calls = %d, want %d", disk.reads, tc.wantReads)
+			}
+			_, err := os.Stat(uploadDir)
+			if tc.wantDeleted {
+				if !os.IsNotExist(err) {
+					t.Errorf("expired upload must be removed: %v", err)
+				}
+			} else if err != nil {
+				t.Errorf("upload must be preserved: %v", err)
+			}
+		})
+	}
+}

cmd/erasure-multipart-fullobject_test.go

@@ -253,7 +253,7 @@ func testAPICompleteMultipartFullObjectChecksumMismatch(obj ObjectLayer, instanc
 	if apiErr.Code != "BadDigest" {
 		t.Fatalf("%s: expected BadDigest, got %q", instanceType, apiErr.Code)
 	}
-	if want := "The CRC32 checksum you specified did not match the calculated checksum."; apiErr.Message != want {
+	if want := "The CRC32 you specified did not match the calculated checksum."; apiErr.Message != want {
 		t.Fatalf("%s: expected message %q, got %q", instanceType, want, apiErr.Message)
 	}
 
@@ -361,7 +361,7 @@ func testAPICompleteMultipartCompositeChecksumMismatch(obj ObjectLayer, instance
 	if apiErr.Code != "BadDigest" {
 		t.Fatalf("%s: expected BadDigest, got %q", instanceType, apiErr.Code)
 	}
-	if want := "The CRC32 checksum you specified did not match the calculated checksum."; apiErr.Message != want {
+	if want := "The CRC32 you specified did not match the calculated checksum."; apiErr.Message != want {
 		t.Fatalf("%s: expected message %q, got %q", instanceType, want, apiErr.Message)
 	}
 	if _, err := obj.GetObjectInfo(t.Context(), bucketName, objectName, ObjectOptions{}); err == nil {

cmd/erasure-multipart.go

@@ -174,6 +174,24 @@ func (er erasureObjects) deleteAll(ctx context.Context, bucket, prefix string) {
 	wg.Wait()
 }
 
+// extractModTimeFromUploadIDDir extracts creation time from the upload ID directory on disk.
+// On-disk directory names are of the form <UUID>x<UnixNano>.
+// Parse the native format first: plaintext UUIDs can also be valid Base64.
+// Preserve the previous encoded-format parser for compatibility.
+func extractModTimeFromUploadIDDir(uploadIDDir string) time.Time {
+	if modTime, ok := multipartUploadTime(uploadIDDir); ok {
+		return modTime
+	}
+	if b64, err := base64.RawURLEncoding.DecodeString(uploadIDDir); err == nil {
+		if split := strings.Split(string(b64), "x"); len(split) == 2 {
+			if t, err := strconv.ParseInt(split[1], 10, 64); err == nil {
+				return time.Unix(0, t)
+			}
+		}
+	}
+	return time.Time{}
+}
+
 // Remove the old multipart uploads on the given disk.
 func (er erasureObjects) cleanupStaleUploadsOnDisk(ctx context.Context, disk StorageAPI) {
 	drivePath := disk.Endpoint().Path
@@ -181,16 +199,7 @@ func (er erasureObjects) cleanupStaleUploadsOnDisk(ctx context.Context, disk Sto
 	readDirFn(pathJoin(drivePath, minioMetaMultipartBucket), func(shaDir string, typ os.FileMode) error {
 		readDirFn(pathJoin(drivePath, minioMetaMultipartBucket, shaDir), func(uploadIDDir string, typ os.FileMode) error {
 			uploadIDPath := pathJoin(shaDir, uploadIDDir)
-			var modTime time.Time
-			// Upload IDs are of the form base64_url(<UUID>x<UnixNano>), we can extract the time from the UUID.
-			if b64, err := base64.RawURLEncoding.DecodeString(uploadIDDir); err == nil {
-				if split := strings.Split(string(b64), "x"); len(split) == 2 {
-					t, err := strconv.ParseInt(split[1], 10, 64)
-					if err == nil {
-						modTime = time.Unix(0, t)
-					}
-				}
-			}
+			modTime := extractModTimeFromUploadIDDir(uploadIDDir)
 			// Fallback for older uploads without time in the ID.
 			if modTime.IsZero() {
 				wait := deleteMultipartCleanupSleeper.Timer(ctx)
@@ -1696,6 +1705,32 @@ func (er erasureObjects) CompleteMultipartUpload(ctx context.Context, bucket str
 		defer lk.Unlock(lkctx)
 	}
 
+	if !opts.ReplicationRequest && !opts.ReplicaLockReconcile && !opts.DataMovement {
+		getObjectInfo := er.getObjectInfo
+		if opts.replicaObjectInfo != nil {
+			getObjectInfo = opts.replicaObjectInfo
+		}
+		if fi.VersionID == "" || fi.VersionID == nullVersionID {
+			if err := checkObjectLockedForOverwrite(ctx, bucket, object, getObjectInfo); err != nil {
+				return oi, toObjectErr(err, bucket, object)
+			}
+		} else {
+			_, gerr := getObjectInfo(ctx, bucket, object, ObjectOptions{
+				VersionID:        fi.VersionID,
+				Versioned:        opts.Versioned,
+				VersionSuspended: opts.VersionSuspended,
+				NoLock:           true,
+			})
+			switch {
+			case gerr == nil, errors.As(gerr, &MethodNotAllowed{}):
+				return oi, toObjectErr(errAuthentication, bucket, object)
+			case isErrVersionNotFound(gerr) || isErrObjectNotFound(gerr):
+			default:
+				return oi, toObjectErr(gerr, bucket, object)
+			}
+		}
+	}
+
 	er.cleanupMultipartPath(ctx, paths...) // cleanup all part.N.meta, and skipped part.N's before final rename().
 
 	defer func() {

cmd/erasure-object.go

@@ -150,6 +150,17 @@ func (er erasureObjects) CopyObject(ctx context.Context, srcBucket, srcObject, d
 		}
 	}
 
+	if !dstOpts.ReplicationRequest && !dstOpts.ReplicaLockReconcile && !dstOpts.DataMovement &&
+		(versionID == "" || versionID == nullVersionID) {
+		getObjectInfo := er.getObjectInfo
+		if dstOpts.replicaObjectInfo != nil {
+			getObjectInfo = dstOpts.replicaObjectInfo
+		}
+		if err := checkObjectLockedForOverwrite(ctx, dstBucket, dstObject, getObjectInfo); err != nil {
+			return oi, toObjectErr(err, dstBucket, dstObject)
+		}
+	}
+
 	modTime = UTCNow() // We only preserve modTime if dstOpts.MTime is true.
 	// in all other cases mtime is latest.
 
@@ -1072,6 +1083,12 @@ func renameData(ctx context.Context, disks []StorageAPI, srcBucket, srcEntry str
 	errs := g.Wait()
 
 	err := reduceWriteQuorumErrs(ctx, errs, objectOpIgnoredErrs, writeQuorum)
+	if errors.Is(err, errCommitMetaMissing) {
+		// Commit metadata went missing on a quorum of drives; the write
+		// did not commit. Return a 503 class error so clients retry
+		// instead of treating the object as absent (404).
+		err = errErasureWriteQuorum
+	}
 	if err != nil {
 		dg := errgroup.WithNErrs(len(disks))
 		for index, nerr := range errs {
@@ -1594,6 +1611,17 @@ func (er erasureObjects) putObject(ctx context.Context, bucket string, object st
 		defer lk.Unlock(lkctx)
 	}
 
+	if !opts.ReplicationRequest && !opts.ReplicaLockReconcile && !opts.DataMovement &&
+		(fi.VersionID == "" || fi.VersionID == nullVersionID) {
+		getObjectInfo := er.getObjectInfo
+		if opts.replicaObjectInfo != nil {
+			getObjectInfo = opts.replicaObjectInfo
+		}
+		if err := checkObjectLockedForOverwrite(ctx, bucket, object, getObjectInfo); err != nil {
+			return ObjectInfo{}, toObjectErr(err, bucket, object)
+		}
+	}
+
 	// Rename the successfully written temporary object to final location.
 	onlineDisks, versions, oldDataDir, err := renameData(ctx, onlineDisks, minioMetaTmpBucket, tempObj, partsMetadata, bucket, object, writeQuorum)
 	if err != nil {

cmd/erasure-server-pool-decom.go

@@ -730,11 +730,12 @@ func (set *erasureObjects) listObjectsToDecommission(ctx context.Context, bi dec
 		minDisks:       listingQuorum,
 		reportNotFound: false,
 		agreed:         fn,
-		partial: func(entries metaCacheEntries, _ []error) {
+		partial: func(entries metaCacheEntries, _ []error) error {
 			entry, ok := entries.resolve(&resolver)
 			if ok {
 				fn(*entry)
 			}
+			return nil
 		},
 		finished: nil,
 	})

cmd/erasure-server-pool-multipart-http_test.go

@@ -265,6 +265,7 @@ func TestPoolsMultipartConditionalLatestVersionAndCallbackOnce(t *testing.T) {
 			if explicit {
 				current = putConsistencyObject(t, z, bucket, object, 1, "latest-other-version", ObjectOptions{Versioned: true, MTime: UTCNow().Add(-time.Minute)})
 			}
+			opts.VersionID = ""
 			id, parts := multipartConditionUpload(t, z, bucket, object, 1, opts)
 			called := 0
 			opts.CheckPrecondFn = func(oi ObjectInfo) bool { called++; return oi.ETag != current.ETag }

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

@@ -633,6 +633,8 @@ func TestPoolsConditionalPutReplicaAvailability(t *testing.T) {
 	}
 }
 
+// A public client cannot address a destination version, with or without a
+// precondition; only trusted replication may (SN-2026-016).
 func TestPoolsConditionalPutDestinationVersionHTTP(t *testing.T) {
 	z, _ := consistencyPools(t)
 	bucket, router := conditionalPutBucket(t, z, "versioned")
@@ -641,21 +643,14 @@ func TestPoolsConditionalPutDestinationVersionHTTP(t *testing.T) {
 	current := putConsistencyObject(t, z, bucket, object, 1, "current", ObjectOptions{Versioned: true, MTime: UTCNow().Add(-time.Minute)})
 	defer conditionalPutPool(t, z, object, 0)()
 	url := getPutObjectURL("", bucket, object) + "?versionId=" + old.VersionID
-	for _, stale := range []bool{true, false} {
-		tag, status := current.ETag, http.StatusOK
-		if stale {
-			tag, status = old.ETag, http.StatusPreconditionFailed
-		}
+	for _, tag := range []string{old.ETag, current.ETag} {
 		put := multipartConditionRequest(t, router, http.MethodPut, url, "replacement", map[string]string{xhttp.IfMatch: fmt.Sprintf("%q", tag)})
-		if put.Code != status {
-			t.Fatalf("version-addressed public PUT: %d want %d: %s", put.Code, status, put.Body.String())
-		}
-		if got := put.Header()[xhttp.AmzVersionID]; !stale && (len(got) != 1 || got[0] != old.VersionID) {
-			t.Fatalf("write version changed: %v", put.Header())
+		if put.Code != http.StatusForbidden {
+			t.Fatalf("version-addressed public PUT: %d want 403: %s", put.Code, put.Body.String())
 		}
 	}
 	get := multipartConditionRequest(t, router, http.MethodGet, url, "", nil)
-	if get.Code != http.StatusOK || get.Body.String() != "replacement" {
+	if get.Code != http.StatusOK || get.Body.String() != "old" {
 		t.Fatalf("addressed GET: %d %q", get.Code, get.Body.String())
 	}
 }

cmd/erasure-server-pool-rebalance.go

@@ -525,11 +525,12 @@ func (set *erasureObjects) listObjectsToRebalance(ctx context.Context, bucketNam
 		minDisks:       listingQuorum,
 		reportNotFound: false,
 		agreed:         fn,
-		partial: func(entries metaCacheEntries, _ []error) {
+		partial: func(entries metaCacheEntries, _ []error) error {
 			entry, ok := entries.resolve(&resolver)
 			if ok {
 				fn(*entry)
 			}
+			return nil
 		},
 		finished: nil,
 	})

cmd/erasure-server-pool.go

@@ -1199,6 +1199,10 @@ func (z *erasureServerPools) PutObject(ctx context.Context, bucket string, objec
 
 	if opts.ReplicaLockReconcile || opts.DataMovement {
 		opts.ReplicaLockReconcile = true
+	}
+	// 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 {
 		opts.replicaObjectInfo = z.replicaObjectInfo
 	}
 	oi, err := z.serverPools[idx].PutObject(ctx, bucket, object, data, opts)
@@ -1510,9 +1514,9 @@ func (z *erasureServerPools) CopyObject(ctx context.Context, srcBucket, srcObjec
 		return objInfo, err
 	}
 
-	if !z.SinglePool() && dstOpts.ReplicaLockReconcile {
+	if !z.SinglePool() && (dstOpts.ReplicaLockReconcile || !dstOpts.ReplicationRequest) {
 		dstOpts.replicaObjectInfo = z.replicaObjectInfo
-		if !dstOpts.DataMovement {
+		if dstOpts.ReplicaLockReconcile && !dstOpts.DataMovement {
 			defer func() {
 				if err == nil {
 					err = z.retireReplicaCopies(ctx, dstBucket, dstObject, poolIdx, objInfo)
@@ -1549,13 +1553,16 @@ func (z *erasureServerPools) CopyObject(ctx context.Context, srcBucket, srcObjec
 		ServerSideEncryption:       dstOpts.ServerSideEncryption,
 		UserDefined:                srcInfo.UserDefined,
 		Versioned:                  dstOpts.Versioned,
+		VersionSuspended:           dstOpts.VersionSuspended,
 		VersionID:                  dstOpts.VersionID,
 		MTime:                      dstOpts.MTime,
 		NoLock:                     true,
 		EncryptFn:                  dstOpts.EncryptFn,
 		WantChecksum:               dstOpts.WantChecksum,
 		WantServerSideChecksumType: dstOpts.WantServerSideChecksumType,
+		ReplicationRequest:         dstOpts.ReplicationRequest,
 		ReplicaLockReconcile:       dstOpts.ReplicaLockReconcile,
+		DataMovement:               dstOpts.DataMovement,
 		replicaObjectInfo:          dstOpts.replicaObjectInfo,
 	}
 
@@ -2305,6 +2312,8 @@ func (z *erasureServerPools) CompleteMultipartUpload(ctx context.Context, bucket
 
 	if opts.ReplicaLockReconcile || opts.DataMovement {
 		opts.ReplicaLockReconcile = true
+	}
+	if opts.ReplicaLockReconcile || !opts.ReplicationRequest {
 		opts.replicaObjectInfo = z.replicaObjectInfo
 	}
 
@@ -2642,11 +2651,15 @@ func (z *erasureServerPools) Walk(ctx context.Context, bucket, prefix string, re
 					minDisks:       listingQuorum,
 					reportNotFound: false,
 					agreed:         send,
-					partial: func(entries metaCacheEntries, _ []error) {
-						entry, ok := entries.resolve(&resolver)
-						if ok {
+					partial: func(entries metaCacheEntries, errs []error) error {
+						entry, err := set.resolveListEntry(ctx, bucket, entries, errs, &resolver, z.listingNSLock(set))
+						if err != nil {
+							return err
+						}
+						if entry != nil {
 							send(*entry)
 						}
+						return nil
 					},
 					finished: nil,
 				}

cmd/erasure-sets.go

@@ -869,11 +869,17 @@ func (s *erasureSets) CopyObject(ctx context.Context, srcBucket, srcObject, dstB
 		ServerSideEncryption:       dstOpts.ServerSideEncryption,
 		UserDefined:                srcInfo.UserDefined,
 		Versioned:                  dstOpts.Versioned,
+		VersionSuspended:           dstOpts.VersionSuspended,
 		VersionID:                  dstOpts.VersionID,
 		MTime:                      dstOpts.MTime,
+		NoLock:                     dstOpts.NoLock,
 		EncryptFn:                  dstOpts.EncryptFn,
 		WantChecksum:               dstOpts.WantChecksum,
 		WantServerSideChecksumType: dstOpts.WantServerSideChecksumType,
+		ReplicationRequest:         dstOpts.ReplicationRequest,
+		ReplicaLockReconcile:       dstOpts.ReplicaLockReconcile,
+		DataMovement:               dstOpts.DataMovement,
+		replicaObjectInfo:          dstOpts.replicaObjectInfo,
 	}
 
 	return dstSet.putObject(ctx, dstBucket, dstObject, srcInfo.PutObjReader, putOpts)

cmd/global-heal.go

@@ -529,7 +529,7 @@ func (er *erasureObjects) healErasureSet(ctx context.Context, buckets []string,
 				jt.Take()
 				go healEntry(bucket, entry)
 			},
-			partial: func(entries metaCacheEntries, _ []error) {
+			partial: func(entries metaCacheEntries, _ []error) error {
 				entry, ok := entries.resolve(&resolver)
 				if !ok {
 					// check if we can get one entry at least
@@ -538,6 +538,7 @@ func (er *erasureObjects) healErasureSet(ctx context.Context, buckets []string,
 				}
 				jt.Take()
 				go healEntry(bucket, *entry)
+				return nil
 			},
 			finished: func(errs []error) {
 				success := countErrs(errs, nil)

cmd/metacache-entries.go

@@ -670,6 +670,12 @@ 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 {
 	defer xioutil.SafeClose(out)
 	top := make([]*metaCacheEntry, len(in))
 	nDone := 0
@@ -683,6 +689,9 @@ func mergeEntryChannels(ctx context.Context, in []chan metaCacheEntry, out chan<
 				return ctx.Err()
 			case v, ok := <-in[0]:
 				if !ok {
+					if terminalErr != nil {
+						return terminalErr(0)
+					}
 					return nil
 				}
 				select {
@@ -700,6 +709,11 @@ 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 {

cmd/metacache-listing-cache_test.go

@@ -0,0 +1,326 @@
+// 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"
+	"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)
+	}
+}

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 := set.resolveListEntry(t.Context(), bucket, entries, nil, &resolver, z.listingNSLock(set))
+	if entry != nil || !isErrReadQuorum(err) {
+		t.Fatalf("split live metadata became absence: entry=%v err=%v", entry, err)
+	}
+}

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 +1,84 @@
+// 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},
+	} {
+		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" }}
+			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 {
+				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)
+				}
+			}()
+			reader := z.listingNSLock(target)("bucket", "object")
+			ctx, cancel := context.WithTimeout(t.Context(), 80*time.Millisecond)
+			defer cancel()
+			rctx, err := reader.GetRLock(ctx, globalOperationTimeout)
+			if err == nil {
+				reader.RUnlock(rctx)
+				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()
+			rctx, err = reader.GetRLock(ctx, globalOperationTimeout)
+			if err != nil {
+				t.Fatalf("fallback read lock failed after writer unlock: %v", err)
+			}
+			reader.RUnlock(rctx)
+		})
+	}
+}

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 = set.resolveListEntry(ctx, bucket, entries, errs, &resolver, z.listingNSLock(set))
+	}()
+	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-null-quorum_test.go

@@ -52,6 +52,37 @@ 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 +150,30 @@ 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 +282,357 @@ func TestListObjectsSingleNullQuorumHTTP(t *testing.T) {
 	}
 }
 
+func TestListObjectsFallsBackToReadableMetadata(t *testing.T) {
+	z, bucket, router := nullQuorumBackend(t)
+	set := z.serverPools[0].sets[0]
+	disks := set.getDisks()
+	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}
+		}
+	}
+	set.getDisks = func() []StorageAPI { return 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())
+	}
+}
+
+// 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 TestListFallbackDoesNotTrustNotFound(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 := set.resolveListEntry(t.Context(), bucket, entries, errs, &resolver, z.listingNSLock(set))
+	if entry != nil || !isErrReadQuorum(err) {
+		t.Fatalf("ambiguous not-found: got entry=%v err=%v; want read quorum 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-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 +264,16 @@ func (z *erasureServerPools) listPathShortcutEOF(ctx context.Context, bucket str
 	return io.EOF
 }
 
+// listingNSLock chooses the namespace used by regular object writes.
+// Multiple pools serialize at the pools layer; single-pool PUT and GET use
+// the destination set. This keeps the existing single-pool lock routing.
+func (z *erasureServerPools) listingNSLock(er *erasureObjects) func(string, ...string) RWLocker {
+	if z.SinglePool() {
+		return er.NewNSLock
+	}
+	return z.NewNSLock
+}
+
 // 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 {
@@ -283,13 +293,16 @@ func (z *erasureServerPools) listMerged(ctx context.Context, o listPathOptions,
 			inputs = append(inputs, innerResults)
 			go func(i int, set *erasureObjects) {
 				defer wg.Done()
-				err := set.listPath(listCtx, o, innerResults)
+				err := set.listPath(listCtx, o, innerResults, z.listingNSLock(set))
 				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 +311,16 @@ 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 := mergeEntryChannelsWithErrors(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
+	})
 
 	cancelList()
 	wg.Wait()
@@ -415,13 +437,18 @@ 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 listingErr error
 	go func(o listPathOptions) {
-		err := z.listMerged(listCtx, o, inCh)
-		if err != nil {
-			meta.setErr(err.Error())
+		listingErr = z.listMerged(listCtx, o, inCh)
+		if listingErr != nil && !errors.Is(listingErr, io.EOF) {
+			meta.setErr(listingErr.Error())
+			cancel()
 		}
-		o.debugln("listAndSave: listing", o.ID, "finished with ", err)
+		close(listingDone)
+		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.
@@ -451,6 +478,9 @@ func (z *erasureServerPools) listAndSave(ctx context.Context, o *listPathOptions
 			entry.reusable = returned
 			saveCh <- entry
 		}
+		// Do not publish an end-of-stream cache block or a final response
+		// before the scan has recorded its terminal error.
+		<-listingDone
 		if !returned {
 			xioutil.SafeClose(outCh)
 		}
@@ -458,9 +488,19 @@ func (z *erasureServerPools) listAndSave(ctx context.Context, o *listPathOptions
 	}()
 
 	entries, err = filteredResults()
-	if err == nil {
-		// Check if listing recorded an error.
-		err = meta.getErr()
+	if err == nil || errors.Is(err, io.EOF) {
+		// Full pages may return before the scan ends. A final page has
+		// 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"
 )
 
@@ -700,9 +699,88 @@ func getQuorumDisks(disks []StorageAPI, infos []DiskInfo, readQuorum int) (newDi
 	return newDisks
 }
 
+// resolveListEntry resolves walk snapshots and, for latest-only listings,
+// falls back to the regular object read path when those snapshots disagree.
+//
+// Only unresolved entries pay for the additional namespace read lock and
+// metadata read. The caller supplies the lock domain used by object writes.
+func (er *erasureObjects) resolveListEntry(ctx context.Context, bucket string, entries metaCacheEntries, errs []error, resolver *metadataResolutionParams, newNSLock func(string, ...string) RWLocker) (*metaCacheEntry, error) {
+	entry, ok := entries.resolve(resolver)
+	if ok {
+		return entry, nil
+	}
+	if resolver.requestedVersions != 1 {
+		return nil, 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, nil
+	}
+	readQuorumErr := func() error {
+		return InsufficientReadQuorum{
+			Bucket: bucket,
+			Object: entry.name,
+			Err:    errErasureReadQuorum,
+			Type:   RQInconsistentMeta,
+		}
+	}
+	// Read under the object read lock, exactly as the GET path does. A
+	// lock-free read of an object being overwritten stably reports a live
+	// object as absent, so an unlocked fallback only trades one silent
+	// omission for another.
+	object := encodeDirObject(entry.name)
+	lock := newNSLock(bucket, object)
+	lockCtx, cancel := context.WithTimeout(ctx, 5*time.Second)
+	defer cancel()
+	lkctx, err := lock.GetRLock(lockCtx, globalOperationTimeout)
+	if err != nil {
+		return nil, err
+	}
+	defer lock.RUnlock(lkctx)
+
+	fi, _, _, err := er.getObjectFileInfo(lkctx.Context(), bucket, object, ObjectOptions{}, false)
+	if err != nil {
+		// getObjectFileInfo maps inconsistent metadata to not-found, so the
+		// walk evidence means not-found is still undecidable here.
+		if isErrObjectNotFound(err) || isErrVersionNotFound(err) {
+			return nil, readQuorumErr()
+		}
+		return nil, err
+	}
+	// Preserve delete markers until the cross-pool merge has discarded any
+	// older live copy. The normal listing filter removes the merged marker.
+	xl := &xlMetaV2{}
+	if err = xl.AddVersion(fi); err != nil {
+		return nil, readQuorumErr()
+	}
+	resolved := &metaCacheEntry{name: entry.name, cached: xl, reusable: true}
+	resolved.metadata, err = xl.AppendTo(metaDataPoolGet())
+	if err != nil {
+		return nil, err
+	}
+	return resolved, nil
+}
+
 // 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)
+func (er *erasureObjects) listPath(ctx context.Context, o listPathOptions, results chan<- metaCacheEntry, newNSLock func(string, ...string) RWLocker) (err error) {
+	// 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
@@ -776,15 +854,19 @@ func (er *erasureObjects) listPath(ctx context.Context, o listPathOptions, resul
 			case results <- entry:
 			}
 		},
-		partial: func(entries metaCacheEntries, errs []error) {
+		partial: func(entries metaCacheEntries, errs []error) error {
 			// Results Disagree :-(
-			entry, ok := entries.resolve(&resolver)
-			if ok {
+			entry, err := er.resolveListEntry(ctx, o.Bucket, entries, errs, &resolver, newNSLock)
+			if err != nil {
+				return err
+			}
+			if entry != nil {
 				select {
 				case <-ctxDone:
 				case results <- *entry:
 				}
 			}
+			return nil
 		},
 	})
 }
@@ -980,7 +1062,7 @@ type listPathRawOptions struct {
 	// partial will be called when there is disagreement between disks.
 	// if disk did not return any result, but also haven't errored
 	// the entry will be empty and errs will
-	partial func(entries metaCacheEntries, errs []error)
+	partial func(entries metaCacheEntries, errs []error) error
 
 	// finished will be called when all streams have finished and
 	// more than one disk returned an error.
@@ -1191,7 +1273,9 @@ func listPathRaw(ctx context.Context, opts listPathRawOptions) (err error) {
 			continue
 		}
 		if opts.partial != nil {
-			opts.partial(topEntries, errs)
+			if err := opts.partial(topEntries, errs); err != nil {
+				return err
+			}
 		}
 		// Skip the inputs we used.
 		for i, r := range readers {

cmd/object-api-errors.go

@@ -162,7 +162,7 @@ func toObjectErr(oerr error, params ...string) error {
 			apiErr.Type = v.Type
 		}
 		return apiErr
-	case errErasureWriteQuorum.Error():
+	case errErasureWriteQuorum.Error(), errCommitMetaMissing.Error():
 		apiErr := InsufficientWriteQuorum{}
 		if len(params) >= 1 {
 			apiErr.Bucket = params[0]
@@ -668,7 +668,7 @@ var (
 func completeMultipartChecksumMismatch(algorithm string) error {
 	description := "The checksum you specified did not match the calculated checksum."
 	if algorithm != "" {
-		description = fmt.Sprintf("The %s checksum you specified did not match the calculated checksum.", algorithm)
+		description = fmt.Sprintf("The %s you specified did not match the calculated checksum.", algorithm)
 	}
 	return fmt.Errorf("%w: %s", errCompleteMultipartChecksumMismatch, description)
 }

cmd/object-api-options.go

@@ -347,6 +347,11 @@ func putOpts(ctx context.Context, bucket, object, vid string, hdrs http.Header,
 	versionSuspended := globalBucketVersioningSys.PrefixSuspended(bucket, object)
 
 	vid = strings.TrimSpace(vid)
+	// A caller-chosen version ID writes that version in place. Only trusted
+	// replication may recreate a source version under its original ID.
+	if vid != "" && !trustedReplication {
+		return opts, errAuthentication
+	}
 	if vid != "" && vid != nullVersionID {
 		_, err := uuid.Parse(vid)
 		if err != nil {

cmd/object-handlers-chunked-checksum_test.go

@@ -62,12 +62,13 @@ func testAPIPutObjectChunkedChecksum(obj ObjectLayer, instanceType, bucketName s
 	// A well-formed 4-byte value that does not match the content.
 	wrongCRC := base64.StdEncoding.EncodeToString([]byte{0x01, 0x02, 0x03, 0x04})
 
-	apiCode := func(rec *httptest.ResponseRecorder) string {
+	apiErrOf := func(rec *httptest.ResponseRecorder) APIErrorResponse {
 		var apiErr APIErrorResponse
 		b, _ := io.ReadAll(rec.Body)
 		_ = xml.Unmarshal(b, &apiErr)
-		return apiErr.Code
+		return apiErr
 	}
+	apiCode := func(rec *httptest.ResponseRecorder) string { return apiErrOf(rec).Code }
 
 	// newChunkedJavaForm builds a non-trailer signed chunked PutObject request that
 	// mirrors the AWS Java SDK v2 wire form: the checksum value sits in the header
@@ -131,8 +132,12 @@ func testAPIPutObjectChunkedChecksum(obj ObjectLayer, instanceType, bucketName s
 		if rec.Code != http.StatusBadRequest {
 			t.Fatalf("%s: chunked+wrong CRC32: expected 400, got %d", instanceType, rec.Code)
 		}
-		if code := apiCode(rec); code != "XAmzContentChecksumMismatch" {
-			t.Fatalf("%s: chunked+wrong CRC32: want XAmzContentChecksumMismatch, got %q", instanceType, code)
+		gotErr := apiErrOf(rec)
+		if gotErr.Code != "XAmzContentChecksumMismatch" {
+			t.Fatalf("%s: chunked+wrong CRC32: want XAmzContentChecksumMismatch, got %q", instanceType, gotErr.Code)
+		}
+		if gotErr.Message != "The CRC32 you specified did not match the calculated checksum." {
+			t.Fatalf("%s: chunked+wrong CRC32: message = %q, want S3 algorithm-specific wording", instanceType, gotErr.Message)
 		}
 	}
 }

cmd/object-handlers.go

@@ -3435,27 +3435,16 @@ func (api objectAPIHandlers) PutObjectRetentionHandler(w http.ResponseWriter, r
 		return
 	}
 
-	// Check permissions to perform this object retention operation
 	if s3Error := authenticateRequest(ctx, r, policy.PutObjectRetentionAction); s3Error != ErrNone {
 		writeErrorResponse(ctx, w, errorCodes.ToAPIErr(s3Error), r.URL)
 		return
 	}
 
-	if _, err := objectAPI.GetBucketInfo(ctx, bucket, BucketOptions{}); err != nil {
-		writeErrorResponse(ctx, w, toAPIError(ctx, err), r.URL)
-		return
-	}
-
 	if !validateLengthAndChecksum(r) {
 		writeErrorResponse(ctx, w, errorCodes.ToAPIErr(ErrMissingContentMD5), r.URL)
 		return
 	}
 
-	if rcfg, _ := globalBucketObjectLockSys.Get(bucket); !rcfg.LockEnabled {
-		writeErrorResponse(ctx, w, errorCodes.ToAPIErr(ErrInvalidBucketObjectLockConfiguration), r.URL)
-		return
-	}
-
 	objRetention, err := objectlock.ParseObjectRetention(r.Body)
 	if err != nil {
 		apiErr := errorCodes.ToAPIErr(ErrMalformedXML)
@@ -3467,6 +3456,24 @@ func (api objectAPIHandlers) PutObjectRetentionHandler(w http.ResponseWriter, r
 	reqInfo := logger.GetReqInfo(ctx)
 	reqInfo.SetTags("retention", objRetention.String())
 
+	// Check permissions to perform this object retention operation. The
+	// object-lock condition keys come from the request body, so this is
+	// decided before any bucket or object state is read.
+	if !isRetentionActionAllowed(policy.PutObjectRetentionAction, bucket, object, objRetention, r, reqInfo.Cred, reqInfo.Owner) {
+		writeErrorResponse(ctx, w, errorCodes.ToAPIErr(ErrAccessDenied), r.URL)
+		return
+	}
+
+	if _, err := objectAPI.GetBucketInfo(ctx, bucket, BucketOptions{}); err != nil {
+		writeErrorResponse(ctx, w, toAPIError(ctx, err), r.URL)
+		return
+	}
+
+	if rcfg, _ := globalBucketObjectLockSys.Get(bucket); !rcfg.LockEnabled {
+		writeErrorResponse(ctx, w, errorCodes.ToAPIErr(ErrInvalidBucketObjectLockConfiguration), r.URL)
+		return
+	}
+
 	opts, err := getOpts(ctx, r, bucket, object)
 	if err != nil {
 		writeErrorResponse(ctx, w, toAPIError(ctx, err), r.URL)

cmd/object-multipart-handlers.go

@@ -1166,7 +1166,7 @@ func (api objectAPIHandlers) CompleteMultipartUploadHandler(w http.ResponseWrite
 		return
 	}
 
-	if _, _, _, s3Err := checkPutObjectLockAllowed(ctx, r, bucket, object, objectAPI.GetObjectInfo, ErrNone, ErrNone, false); s3Err != ErrNone {
+	if _, _, _, s3Err := checkPutObjectLockAllowed(ctx, r, bucket, object, objectAPI.GetObjectInfo, ErrNone, ErrNone, trustedReplication); s3Err != ErrNone {
 		writeErrorResponse(ctx, w, errorCodes.ToAPIErr(s3Err), r.URL)
 		return
 	}

cmd/object-retention-authz_test.go

@@ -0,0 +1,679 @@
+// Copyright (c) 2015-2026 MinIO, Inc.
+// Copyright (c) 2026 PGSTY
+//
+// This file is part of MinIO 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.
+
+package cmd
+
+import (
+	"archive/tar"
+	"bytes"
+	"context"
+	"crypto/md5"
+	"encoding/base64"
+	"encoding/xml"
+	"fmt"
+	"io"
+	"net/http"
+	"net/http/httptest"
+	"net/url"
+	"strings"
+	"testing"
+	"time"
+
+	"github.com/minio/madmin-go/v3"
+	"github.com/minio/minio/internal/amztime"
+	"github.com/minio/minio/internal/auth"
+	objectlock "github.com/minio/minio/internal/bucket/object/lock"
+	xhttp "github.com/minio/minio/internal/http"
+	"github.com/pgsty/silo-pkg/v3/policy"
+)
+
+func retentionAuthzRequest(t *testing.T, router http.Handler, creds auth.Credentials, method, target string, body []byte, headers map[string]string) *httptest.ResponseRecorder {
+	t.Helper()
+	var rd io.ReadSeeker
+	if body != nil {
+		rd = bytes.NewReader(body)
+	}
+	req, err := newTestSignedRequestV4(method, target, int64(len(body)), rd, creds.AccessKey, creds.SecretKey, headers)
+	if err != nil {
+		t.Fatal(err)
+	}
+	rec := httptest.NewRecorder()
+	router.ServeHTTP(rec, req)
+	return rec
+}
+
+func retentionBody(mode string, until time.Time) []byte {
+	if mode == "" {
+		return []byte(`<Retention xmlns="http://s3.amazonaws.com/doc/2006-03-01/"></Retention>`)
+	}
+	return fmt.Appendf(nil, `<Retention xmlns="http://s3.amazonaws.com/doc/2006-03-01/"><Mode>%s</Mode><RetainUntilDate>%s</RetainUntilDate></Retention>`,
+		mode, until.UTC().Format(time.RFC3339))
+}
+
+func putRetention(t *testing.T, router http.Handler, cred auth.Credentials, bucket, key, vid, mode string, until time.Time, bypass bool) int {
+	t.Helper()
+	body := retentionBody(mode, until)
+	sum := md5.Sum(body)
+	headers := map[string]string{xhttp.ContentMD5: base64.StdEncoding.EncodeToString(sum[:])}
+	if bypass {
+		headers[objectlock.AmzObjectLockBypassRetGovernance] = "true"
+	}
+	rec := retentionAuthzRequest(t, router, cred, http.MethodPut,
+		makeTestTargetURL("", bucket, key, url.Values{"retention": {""}, "versionId": {vid}}), body, headers)
+	return rec.Code
+}
+
+func newRetentionPolicyUser(t *testing.T, instanceType, policyJSON string) auth.Credentials {
+	t.Helper()
+	ctx := t.Context()
+	accessKey, secretKey, err := auth.GenerateCredentials()
+	if err != nil {
+		t.Fatalf("%s: generate credentials: %v", instanceType, err)
+	}
+	creds := auth.Credentials{AccessKey: accessKey, SecretKey: secretKey}
+	if _, err = globalIAMSys.CreateUser(ctx, creds.AccessKey, madmin.AddOrUpdateUserReq{
+		SecretKey: creds.SecretKey,
+		Status:    madmin.AccountEnabled,
+	}); err != nil {
+		t.Fatalf("%s: create user: %v", instanceType, err)
+	}
+	parsed, err := policy.ParseConfig(strings.NewReader(policyJSON))
+	if err != nil {
+		t.Fatalf("%s: parse policy: %v", instanceType, err)
+	}
+	policyName := "retention-authz-" + mustGetUUID()
+	if _, err = globalIAMSys.SetPolicy(ctx, policyName, *parsed); err != nil {
+		t.Fatalf("%s: install policy: %v", instanceType, err)
+	}
+	if _, err = globalIAMSys.PolicyDBSet(ctx, creds.AccessKey, policyName, regUser, false); err != nil {
+		t.Fatalf("%s: attach policy: %v", instanceType, err)
+	}
+	return creds
+}
+
+// TestAPIPutObjectRetentionAuthorization covers SN-2026-015: the governance
+// bypass header is intent only; s3:PutObjectRetention is always required and
+// weakening unexpired GOVERNANCE also requires s3:BypassGovernanceRetention.
+func TestAPIPutObjectRetentionAuthorization(t *testing.T) {
+	defer DetectTestLeak(t)()
+	ExecObjectLayerAPITest(ExecObjectLayerAPITestArgs{
+		t:          t,
+		objAPITest: testAPIPutObjectRetentionAuthorization,
+	})
+}
+
+func testAPIPutObjectRetentionAuthorization(obj ObjectLayer, instanceType, bucket string, router http.Handler, root auth.Credentials, t *testing.T) {
+	enableBucketObjectLock(t, bucket)
+
+	noRet := newObjectAttributesAuthzUser(t, instanceType, bucket, `"s3:GetObject"`)
+	retOnly := newObjectAttributesAuthzUser(t, instanceType, bucket, `"s3:PutObjectRetention"`)
+	retBypass := newObjectAttributesAuthzUser(t, instanceType, bucket, `"s3:PutObjectRetention","s3:BypassGovernanceRetention"`)
+	bypassOnly := newObjectAttributesAuthzUser(t, instanceType, bucket, `"s3:BypassGovernanceRetention"`)
+	condDenyShort := newRetentionPolicyUser(t, instanceType, `{
+		"Version": "2012-10-17",
+		"Statement": [
+			{
+				"Effect": "Allow",
+				"Action": ["s3:PutObjectRetention", "s3:BypassGovernanceRetention"],
+				"Resource": ["arn:aws:s3:::`+bucket+`/*"]
+			},
+			{
+				"Effect": "Deny",
+				"Action": ["s3:PutObjectRetention"],
+				"Resource": ["arn:aws:s3:::`+bucket+`/*"],
+				"Condition": {"NumericLessThan": {"s3:object-lock-remaining-retention-days": "60"}}
+			}
+		]
+	}`)
+	explicitDenyBypass := newRetentionPolicyUser(t, instanceType, `{
+		"Version": "2012-10-17",
+		"Statement": [
+			{
+				"Effect": "Allow",
+				"Action": ["s3:*"],
+				"Resource": ["arn:aws:s3:::`+bucket+`/*"]
+			},
+			{
+				"Effect": "Deny",
+				"Action": ["s3:BypassGovernanceRetention"],
+				"Resource": ["arn:aws:s3:::`+bucket+`/*"]
+			}
+		]
+	}`)
+
+	far := UTCNow().Add(365 * 24 * time.Hour).Truncate(time.Second)
+	nearer := UTCNow().Add(30 * 24 * time.Hour).Truncate(time.Second)
+	farther := far.Add(24 * time.Hour)
+
+	put := func(t *testing.T, object, mode string) string {
+		t.Helper()
+		payload := []byte("retention authorization")
+		opts := ObjectOptions{Versioned: true, UserDefined: map[string]string{}}
+		if mode != "" {
+			opts.UserDefined[strings.ToLower(xhttp.AmzObjectLockMode)] = mode
+			opts.UserDefined[strings.ToLower(xhttp.AmzObjectLockRetainUntilDate)] = amztime.ISO8601Format(far)
+		}
+		info, err := obj.PutObject(t.Context(), bucket, object,
+			mustGetPutObjReader(t, bytes.NewReader(payload), int64(len(payload)), "", ""), opts)
+		if err != nil {
+			t.Fatal(err)
+		}
+		return info.VersionID
+	}
+	stored := func(t *testing.T, object, vid string) objectlock.ObjectRetention {
+		t.Helper()
+		info, err := obj.GetObjectInfo(t.Context(), bucket, object, ObjectOptions{VersionID: vid})
+		if err != nil {
+			t.Fatal(err)
+		}
+		return objectlock.GetObjectRetentionMeta(info.UserDefined)
+	}
+
+	for _, test := range []struct {
+		name     string
+		creds    auth.Credentials
+		existing string // stored retention mode on the version, until far
+		mode     string // requested mode, "" removes retention
+		until    time.Time
+		bypass   bool
+		want     int
+		changed  bool
+	}{
+		{name: "no-permission removal", creds: noRet, existing: "GOVERNANCE", bypass: true, want: http.StatusForbidden},
+		{name: "no-permission mode change", creds: noRet, existing: "GOVERNANCE", mode: "COMPLIANCE", until: nearer, bypass: true, want: http.StatusForbidden},
+		{name: "no-permission new compliance", creds: noRet, mode: "COMPLIANCE", until: nearer, bypass: true, want: http.StatusForbidden},
+		{name: "retention-only shorten with header", creds: retOnly, existing: "GOVERNANCE", mode: "GOVERNANCE", until: nearer, bypass: true, want: http.StatusForbidden},
+		{name: "retention-only removal with header", creds: retOnly, existing: "GOVERNANCE", bypass: true, want: http.StatusForbidden},
+		{name: "retention-only mode change with header", creds: retOnly, existing: "GOVERNANCE", mode: "COMPLIANCE", until: nearer, bypass: true, want: http.StatusForbidden},
+		{name: "retention-only shorten without header", creds: retOnly, existing: "GOVERNANCE", mode: "GOVERNANCE", until: nearer, want: http.StatusBadRequest},
+		{name: "retention-only extend with header", creds: retOnly, existing: "GOVERNANCE", mode: "GOVERNANCE", until: farther, bypass: true, want: http.StatusOK, changed: true},
+		{name: "retention-only new retention", creds: retOnly, mode: "GOVERNANCE", until: nearer, want: http.StatusOK, changed: true},
+		{name: "bypass shorten with header", creds: retBypass, existing: "GOVERNANCE", mode: "GOVERNANCE", until: nearer, bypass: true, want: http.StatusOK, changed: true},
+		{name: "bypass removal with header", creds: retBypass, existing: "GOVERNANCE", bypass: true, want: http.StatusOK, changed: true},
+		{name: "bypass shorten without header", creds: retBypass, existing: "GOVERNANCE", mode: "GOVERNANCE", until: nearer, want: http.StatusBadRequest},
+		{name: "bypass shorten compliance", creds: retBypass, existing: "COMPLIANCE", mode: "COMPLIANCE", until: nearer, bypass: true, want: http.StatusBadRequest},
+		{name: "bypass-only extend", creds: bypassOnly, existing: "GOVERNANCE", mode: "GOVERNANCE", until: farther, bypass: true, want: http.StatusForbidden},
+		{name: "conditional-deny-put-retention with bypass header", creds: condDenyShort, existing: "GOVERNANCE", mode: "GOVERNANCE", until: nearer, bypass: true, want: http.StatusForbidden},
+		{name: "conditional-allow-put-retention satisfied", creds: condDenyShort, existing: "GOVERNANCE", mode: "GOVERNANCE", until: farther, bypass: true, want: http.StatusOK, changed: true},
+		{name: "explicit-deny-bypass shorten with header", creds: explicitDenyBypass, existing: "GOVERNANCE", mode: "GOVERNANCE", until: nearer, bypass: true, want: http.StatusForbidden},
+		{name: "owner shorten with header", creds: root, existing: "GOVERNANCE", mode: "GOVERNANCE", until: nearer, bypass: true, want: http.StatusOK, changed: true},
+	} {
+		t.Run(instanceType+"/"+test.name, func(t *testing.T) {
+			object := "retention-authz/" + strings.ReplaceAll(test.name, " ", "-")
+			vid := put(t, object, test.existing)
+			before := stored(t, object, vid)
+
+			body := retentionBody(test.mode, test.until)
+			sum := md5.Sum(body)
+			headers := map[string]string{xhttp.ContentMD5: base64.StdEncoding.EncodeToString(sum[:])}
+			if test.bypass {
+				headers[objectlock.AmzObjectLockBypassRetGovernance] = "true"
+			}
+			target := makeTestTargetURL("", bucket, object, url.Values{"retention": {""}, "versionId": {vid}})
+			rec := retentionAuthzRequest(t, router, test.creds, http.MethodPut, target, body, headers)
+			if rec.Code != test.want {
+				t.Fatalf("status = %d, want %d: %s", rec.Code, test.want, rec.Body.String())
+			}
+			after := stored(t, object, vid)
+			if changed := after.Mode != before.Mode || !after.RetainUntilDate.Equal(before.RetainUntilDate.Time); changed != test.changed {
+				t.Fatalf("retention changed = %v, want %v (before %s, after %s)", changed, test.changed, before.String(), after.String())
+			}
+		})
+	}
+
+	t.Run(instanceType+"/no-permission missing bucket", func(t *testing.T) {
+		body := retentionBody("", time.Time{})
+		sum := md5.Sum(body)
+		target := makeTestTargetURL("", "retention-authz-missing", "object", url.Values{"retention": {""}})
+		rec := retentionAuthzRequest(t, router, noRet, http.MethodPut, target, body,
+			map[string]string{xhttp.ContentMD5: base64.StdEncoding.EncodeToString(sum[:])})
+		if rec.Code != http.StatusForbidden {
+			t.Fatalf("status = %d, want 403 before bucket lookup: %s", rec.Code, rec.Body.String())
+		}
+	})
+}
+
+// TestAPIVersionAddressedWriteRequiresReplicationTrust covers SN-2026-016: a
+// caller-chosen destination versionId is honored only for trusted replication,
+// and write paths cannot overwrite locked versions via directory keys or
+// version-addressed multipart completions.
+func TestAPIVersionAddressedWriteRequiresReplicationTrust(t *testing.T) {
+	defer DetectTestLeak(t)()
+	ExecObjectLayerAPITest(ExecObjectLayerAPITestArgs{
+		t:                 t,
+		objAPITest:        testAPIVersionAddressedWriteRequiresReplicationTrust,
+		makeBucketOptions: MakeBucketOptions{VersioningEnabled: true, LockEnabled: true},
+	})
+}
+
+func testAPIVersionAddressedWriteRequiresReplicationTrust(obj ObjectLayer, instanceType, bucket string, router http.Handler, root auth.Credentials, t *testing.T) {
+	writer := newObjectAttributesAuthzUser(t, instanceType, bucket, `"s3:GetObject","s3:PutObject"`)
+	original := []byte("original version content")
+	info, err := obj.PutObject(t.Context(), bucket, "addressed",
+		mustGetPutObjReader(t, bytes.NewReader(original), int64(len(original)), "", ""), ObjectOptions{Versioned: true})
+	if err != nil {
+		t.Fatal(err)
+	}
+	vid := info.VersionID
+
+	unchanged := func(t *testing.T) {
+		t.Helper()
+		got, err := obj.GetObjectInfo(t.Context(), bucket, "addressed", ObjectOptions{VersionID: vid})
+		if err != nil {
+			t.Fatal(err)
+		}
+		if got.Size != int64(len(original)) || got.ETag != info.ETag {
+			t.Fatalf("version %s was rewritten: size %d etag %s", vid, got.Size, got.ETag)
+		}
+	}
+
+	for _, creds := range []struct {
+		name  string
+		creds auth.Credentials
+	}{{"writer", writer}, {"owner", root}} {
+		t.Run(instanceType+"/"+creds.name+"/put", func(t *testing.T) {
+			body := []byte("replacement")
+			target := makeTestTargetURL("", bucket, "addressed", url.Values{"versionId": {vid}})
+			rec := retentionAuthzRequest(t, router, creds.creds, http.MethodPut, target, body, nil)
+			if rec.Code != http.StatusForbidden {
+				t.Fatalf("status = %d, want 403: %s", rec.Code, rec.Body.String())
+			}
+			unchanged(t)
+		})
+		t.Run(instanceType+"/"+creds.name+"/null-version", func(t *testing.T) {
+			body := []byte("replacement")
+			target := makeTestTargetURL("", bucket, "addressed", url.Values{"versionId": {nullVersionID}})
+			rec := retentionAuthzRequest(t, router, creds.creds, http.MethodPut, target, body, nil)
+			if rec.Code != http.StatusForbidden {
+				t.Fatalf("status = %d, want 403: %s", rec.Code, rec.Body.String())
+			}
+		})
+		t.Run(instanceType+"/"+creds.name+"/copy", func(t *testing.T) {
+			target := makeTestTargetURL("", bucket, "addressed", url.Values{"versionId": {vid}})
+			req, err := newTestSignedRequestV4(http.MethodPut, target, 0, nil, creds.creds.AccessKey, creds.creds.SecretKey,
+				map[string]string{xhttp.AmzCopySource: "/" + bucket + "/addressed?versionId=" + vid, xhttp.AmzMetadataDirective: "REPLACE"})
+			if err != nil {
+				t.Fatal(err)
+			}
+			rec := httptest.NewRecorder()
+			router.ServeHTTP(rec, req)
+			if rec.Code != http.StatusForbidden {
+				t.Fatalf("status = %d, want 403: %s", rec.Code, rec.Body.String())
+			}
+			unchanged(t)
+		})
+		t.Run(instanceType+"/"+creds.name+"/multipart", func(t *testing.T) {
+			target := makeTestTargetURL("", bucket, "addressed", url.Values{"uploads": {""}, "versionId": {vid}})
+			rec := retentionAuthzRequest(t, router, creds.creds, http.MethodPost, target, nil, nil)
+			if rec.Code != http.StatusForbidden {
+				t.Fatalf("status = %d, want 403: %s", rec.Code, rec.Body.String())
+			}
+		})
+		t.Run(instanceType+"/"+creds.name+"/complete-multipart-existing-version", func(t *testing.T) {
+			// Simulate a pending upload persisted with an existing version ID
+			// (e.g. initiated pre-upgrade or by a replication peer).
+			up, err := obj.NewMultipartUpload(t.Context(), bucket, "addressed", ObjectOptions{
+				Versioned: true,
+				VersionID: vid,
+			})
+			if err != nil {
+				t.Fatal(err)
+			}
+			defer obj.AbortMultipartUpload(t.Context(), bucket, "addressed", up.UploadID, ObjectOptions{})
+
+			partPayload := []byte("replacement-part")
+			pi, err := obj.PutObjectPart(t.Context(), bucket, "addressed", up.UploadID, 1,
+				mustGetPutObjReader(t, bytes.NewReader(partPayload), int64(len(partPayload)), "", ""), ObjectOptions{})
+			if err != nil {
+				t.Fatal(err)
+			}
+			completeBody := fmt.Appendf(nil, `<CompleteMultipartUpload><Part><PartNumber>1</PartNumber><ETag>"%s"</ETag></Part></CompleteMultipartUpload>`, pi.ETag)
+			target := makeTestTargetURL("", bucket, "addressed", url.Values{"uploadId": {up.UploadID}})
+			rec := retentionAuthzRequest(t, router, creds.creds, http.MethodPost, target, completeBody, nil)
+			if rec.Code != http.StatusForbidden {
+				t.Fatalf("status = %d, want 403: %s", rec.Code, rec.Body.String())
+			}
+			unchanged(t)
+		})
+	}
+
+	t.Run(instanceType+"/locked-directory-object", func(t *testing.T) {
+		far := UTCNow().Add(365 * 24 * time.Hour).Truncate(time.Second)
+		for _, lockCase := range []struct {
+			name string
+			meta map[string]string
+		}{
+			{
+				name: "governance",
+				meta: map[string]string{
+					strings.ToLower(xhttp.AmzObjectLockMode):            "GOVERNANCE",
+					strings.ToLower(xhttp.AmzObjectLockRetainUntilDate): amztime.ISO8601Format(far),
+				},
+			},
+			{
+				name: "compliance",
+				meta: map[string]string{
+					strings.ToLower(xhttp.AmzObjectLockMode):            "COMPLIANCE",
+					strings.ToLower(xhttp.AmzObjectLockRetainUntilDate): amztime.ISO8601Format(far),
+				},
+			},
+			{
+				name: "legal-hold",
+				meta: map[string]string{
+					strings.ToLower(xhttp.AmzObjectLockLegalHold): "ON",
+				},
+			},
+		} {
+			t.Run(lockCase.name, func(t *testing.T) {
+				dirKey := "locked-dir-" + lockCase.name + "/"
+				if _, err := obj.PutObject(t.Context(), bucket, dirKey,
+					mustGetPutObjReader(t, bytes.NewReader(nil), 0, "", ""),
+					ObjectOptions{Versioned: true, VersionID: nullVersionID, UserDefined: lockCase.meta}); err != nil {
+					t.Fatal(err)
+				}
+				target := makeTestTargetURL("", bucket, dirKey, nil)
+				rec := retentionAuthzRequest(t, router, writer, http.MethodPut, target, nil, nil)
+				if rec.Code != http.StatusBadRequest {
+					t.Fatalf("PUT locked dir status = %d, want 400: %s", rec.Code, rec.Body.String())
+				}
+				got, err := obj.GetObjectInfo(t.Context(), bucket, dirKey, ObjectOptions{VersionID: nullVersionID})
+				if err != nil {
+					t.Fatal(err)
+				}
+				for k, wantVal := range lockCase.meta {
+					if got.UserDefined[k] != wantVal {
+						t.Fatalf("metadata %s = %q, want %q", k, got.UserDefined[k], wantVal)
+					}
+				}
+			})
+		}
+	})
+
+	t.Run(instanceType+"/snowball-archive-version-id", func(t *testing.T) {
+		var tarBuf bytes.Buffer
+		tw := tar.NewWriter(&tarBuf)
+		entryData := []byte("snowball-overwrite")
+		if err := tw.WriteHeader(&tar.Header{
+			Typeflag:   tar.TypeReg,
+			Name:       "addressed",
+			Size:       int64(len(entryData)),
+			Format:     tar.FormatPAX,
+			PAXRecords: map[string]string{"minio.versionId": vid},
+		}); err != nil {
+			t.Fatal(err)
+		}
+		if _, err := tw.Write(entryData); err != nil {
+			t.Fatal(err)
+		}
+		if err := tw.Close(); err != nil {
+			t.Fatal(err)
+		}
+
+		// Untrusted writer cannot overwrite an existing version via PAX minio.versionId.
+		target := makeTestTargetURL("", bucket, "archive.tar", nil)
+		rec := retentionAuthzRequest(t, router, writer, http.MethodPut, target, tarBuf.Bytes(),
+			map[string]string{"x-amz-meta-snowball-auto-extract": "true"})
+		if rec.Code != http.StatusForbidden {
+			t.Fatalf("untrusted snowball status = %d, want 403: %s", rec.Code, rec.Body.String())
+		}
+		unchanged(t)
+
+		// Trusted batch replication (with outer MinIOSourceReplicationRequest) preserves PAX minio.versionId.
+		replicaVID := mustGetUUID()
+		var repBuf bytes.Buffer
+		repTW := tar.NewWriter(&repBuf)
+		repData := []byte("replicated-via-snowball")
+		if err := repTW.WriteHeader(&tar.Header{
+			Typeflag:   tar.TypeReg,
+			Name:       "snowball-replicated",
+			Size:       int64(len(repData)),
+			Format:     tar.FormatPAX,
+			PAXRecords: map[string]string{"minio.versionId": replicaVID},
+		}); err != nil {
+			t.Fatal(err)
+		}
+		if _, err := repTW.Write(repData); err != nil {
+			t.Fatal(err)
+		}
+		if err := repTW.Close(); err != nil {
+			t.Fatal(err)
+		}
+		repRec := retentionAuthzRequest(t, router, root, http.MethodPut, target, repBuf.Bytes(),
+			map[string]string{
+				"x-amz-meta-snowball-auto-extract":  "true",
+				xhttp.MinIOSourceReplicationRequest: "true",
+			})
+		if repRec.Code != http.StatusOK {
+			t.Fatalf("trusted snowball status = %d, want 200: %s", repRec.Code, repRec.Body.String())
+		}
+		if _, err := obj.GetObjectInfo(t.Context(), bucket, "snowball-replicated", ObjectOptions{VersionID: replicaVID}); err != nil {
+			t.Fatalf("trusted snowball replication did not preserve version ID: %v", err)
+		}
+	})
+
+	t.Run(instanceType+"/trusted replication", func(t *testing.T) {
+		replicaVID := mustGetUUID()
+		body := []byte("replicated version")
+		target := makeTestTargetURL("", bucket, "replicated", url.Values{"versionId": {replicaVID}})
+		rec := retentionAuthzRequest(t, router, root, http.MethodPut, target, body,
+			map[string]string{xhttp.MinIOSourceReplicationRequest: "true"})
+		if rec.Code != http.StatusOK {
+			t.Fatalf("status = %d, want 200: %s", rec.Code, rec.Body.String())
+		}
+		if _, err := obj.GetObjectInfo(t.Context(), bucket, "replicated", ObjectOptions{VersionID: replicaVID}); err != nil {
+			t.Fatalf("trusted replication did not keep its version ID: %v", err)
+		}
+	})
+}
+
+func TestPutObjectRetentionClearMinimumDaysPolicy(t *testing.T) {
+	ExecObjectLayerAPITest(ExecObjectLayerAPITestArgs{
+		t:                 t,
+		makeBucketOptions: MakeBucketOptions{VersioningEnabled: true, LockEnabled: true},
+		objAPITest: func(obj ObjectLayer, instance, bucket string, router http.Handler, root auth.Credentials, t *testing.T) {
+			for _, conditionalAllow := range []bool{false, true} {
+				t.Run(fmt.Sprintf("%s/conditional-allow-%v", instance, conditionalAllow), func(t *testing.T) {
+					statement := `{"Effect":"Allow","Action":["s3:PutObjectRetention","s3:BypassGovernanceRetention"],"Resource":["arn:aws:s3:::` + bucket + `/*"]}, {"Effect":"Deny","Action":["s3:PutObjectRetention"],"Resource":["arn:aws:s3:::` + bucket + `/*"],"Condition":{"NumericLessThan":{"s3:object-lock-remaining-retention-days":"60"}}}`
+					if conditionalAllow {
+						statement = `{"Effect":"Allow","Action":["s3:BypassGovernanceRetention"],"Resource":["arn:aws:s3:::` + bucket + `/*"]}, {"Effect":"Allow","Action":["s3:PutObjectRetention"],"Resource":["arn:aws:s3:::` + bucket + `/*"],"Condition":{"NumericGreaterThanEquals":{"s3:object-lock-remaining-retention-days":"60"}}}`
+					}
+					cred := newRetentionPolicyUser(t, instance, `{"Version":"2012-10-17","Statement":[`+statement+`]}`)
+					key := fmt.Sprintf("review-minimum-%v", conditionalAllow)
+					far := UTCNow().Add(365 * 24 * time.Hour).Truncate(time.Second)
+					info, err := obj.PutObject(t.Context(), bucket, key, mustGetPutObjReader(t, bytes.NewReader(nil), 0, "", ""), ObjectOptions{
+						Versioned: true,
+						UserDefined: map[string]string{
+							strings.ToLower(xhttp.AmzObjectLockMode):            "GOVERNANCE",
+							strings.ToLower(xhttp.AmzObjectLockRetainUntilDate): amztime.ISO8601Format(far),
+						},
+					})
+					if err != nil {
+						t.Fatal(err)
+					}
+					if status := putRetention(t, router, cred, bucket, key, info.VersionID, "GOVERNANCE", UTCNow().Add(30*24*time.Hour), true); status != http.StatusForbidden {
+						t.Fatalf("shortening to 30 days status = %d, want 403", status)
+					}
+					status := putRetention(t, router, cred, bucket, key, info.VersionID, "", time.Time{}, true)
+					after, err := obj.GetObjectInfo(t.Context(), bucket, key, ObjectOptions{VersionID: info.VersionID})
+					if err != nil {
+						t.Fatal(err)
+					}
+					ret := objectlock.GetObjectRetentionMeta(after.UserDefined)
+					if status != http.StatusForbidden || ret.Mode != objectlock.RetGovernance {
+						t.Fatalf("clearing retention status = %d, stored retention = %s; want 403 and original GOVERNANCE", status, ret.String())
+					}
+				})
+			}
+		},
+	})
+}
+
+type hookBeforeWrite struct {
+	ObjectLayer
+	key    string
+	before func()
+}
+
+func (p *hookBeforeWrite) fire(key string) {
+	if key == p.key && p.before != nil {
+		f := p.before
+		p.before = nil
+		f()
+	}
+}
+
+func (p *hookBeforeWrite) PutObject(ctx context.Context, bucket, key string, r *PutObjReader, opts ObjectOptions) (ObjectInfo, error) {
+	p.fire(key)
+	return p.ObjectLayer.PutObject(ctx, bucket, key, r, opts)
+}
+
+func (p *hookBeforeWrite) CopyObject(ctx context.Context, srcBucket, srcObject, dstBucket, dstObject string, srcInfo ObjectInfo, srcOpts, dstOpts ObjectOptions) (ObjectInfo, error) {
+	p.fire(dstObject)
+	return p.ObjectLayer.CopyObject(ctx, srcBucket, srcObject, dstBucket, dstObject, srcInfo, srcOpts, dstOpts)
+}
+
+func (p *hookBeforeWrite) CompleteMultipartUpload(ctx context.Context, bucket, key, uploadID string, parts []CompletePart, opts ObjectOptions) (ObjectInfo, error) {
+	p.fire(key)
+	return p.ObjectLayer.CompleteMultipartUpload(ctx, bucket, key, uploadID, parts, opts)
+}
+
+func TestDirectoryObjectLockAfterPreflight(t *testing.T) {
+	ExecObjectLayerAPITest(ExecObjectLayerAPITestArgs{
+		t:                 t,
+		makeBucketOptions: MakeBucketOptions{VersioningEnabled: true, LockEnabled: true},
+		objAPITest: func(obj ObjectLayer, instance, bucket string, router http.Handler, root auth.Credentials, t *testing.T) {
+			writer := newObjectAttributesAuthzUser(t, instance, bucket, `"s3:GetObject","s3:PutObject"`)
+			srcDir := "other-dir/"
+			if rec := retentionAuthzRequest(t, router, writer, http.MethodPut, makeTestTargetURL("", bucket, srcDir, nil), nil, nil); rec.Code != http.StatusOK {
+				t.Fatalf("srcDir PUT: %d %s", rec.Code, rec.Body.String())
+			}
+
+			for _, mode := range []string{"GOVERNANCE", "COMPLIANCE", "LEGAL-HOLD"} {
+				for _, op := range []string{"PutObject", "CopyObject", "SelfCopy"} {
+					t.Run(instance+"/"+mode+"/"+op, func(t *testing.T) {
+						key := fmt.Sprintf("review-race-%s-%s/", mode, op)
+						if rec := retentionAuthzRequest(t, router, writer, http.MethodPut, makeTestTargetURL("", bucket, key, nil), nil, nil); rec.Code != http.StatusOK {
+							t.Fatalf("initial PUT: %d %s", rec.Code, rec.Body.String())
+						}
+
+						var locked ObjectInfo
+						hook := &hookBeforeWrite{ObjectLayer: obj, key: key}
+						hook.before = func() {
+							if mode == "LEGAL-HOLD" {
+								body := []byte(`<LegalHold><Status>ON</Status></LegalHold>`)
+								sum := md5.Sum(body)
+								rec := retentionAuthzRequest(t, router, root, http.MethodPut, makeTestTargetURL("", bucket, key, url.Values{"legal-hold": {""}, "versionId": {nullVersionID}}), body, map[string]string{xhttp.ContentMD5: base64.StdEncoding.EncodeToString(sum[:])})
+								if rec.Code != http.StatusOK {
+									t.Fatalf("set legal hold: %d %s", rec.Code, rec.Body.String())
+								}
+							} else if status := putRetention(t, router, root, bucket, key, nullVersionID, mode, UTCNow().Add(365*24*time.Hour), false); status != http.StatusOK {
+								t.Fatalf("set retention: %d", status)
+							}
+							var err error
+							locked, err = obj.GetObjectInfo(t.Context(), bucket, key, ObjectOptions{VersionID: nullVersionID})
+							if err != nil {
+								t.Fatal(err)
+							}
+						}
+						setObjectLayer(hook)
+						defer setObjectLayer(obj)
+
+						var rec *httptest.ResponseRecorder
+						switch op {
+						case "PutObject":
+							rec = retentionAuthzRequest(t, router, writer, http.MethodPut, makeTestTargetURL("", bucket, key, nil), nil, nil)
+						case "CopyObject":
+							rec = retentionAuthzRequest(t, router, writer, http.MethodPut, makeTestTargetURL("", bucket, key, nil), nil, map[string]string{
+								xhttp.AmzCopySource: pathJoin(bucket, srcDir),
+							})
+						case "SelfCopy":
+							rec = retentionAuthzRequest(t, router, writer, http.MethodPut, makeTestTargetURL("", bucket, key, nil), nil, map[string]string{
+								xhttp.AmzCopySource: pathJoin(bucket, key), xhttp.AmzMetadataDirective: "REPLACE", "x-amz-meta-updated": "true",
+							})
+						}
+						if hook.before != nil {
+							t.Fatalf("%s never reached backend: %d %s", op, rec.Code, rec.Body.String())
+						}
+						after, err := obj.GetObjectInfo(t.Context(), bucket, key, ObjectOptions{VersionID: nullVersionID})
+						if err != nil {
+							t.Fatal(err)
+						}
+						if rec.Code != http.StatusBadRequest {
+							t.Fatalf("locked null write via %s: %d %s", op, rec.Code, rec.Body.String())
+						}
+						multipartConditionError(t, rec, "InvalidRequest")
+						retentionOverwriteUnchanged(t, locked, after)
+					})
+				}
+			}
+		},
+	})
+}
+
+func TestMultipartVersioningTransitions(t *testing.T) {
+	ExecObjectLayerAPITest(ExecObjectLayerAPITestArgs{
+		t: t,
+		objAPITest: func(obj ObjectLayer, instance, bucket string, router http.Handler, root auth.Credentials, t *testing.T) {
+			t.Run(instance+"/unversioned-to-enabled", func(t *testing.T) {
+				key := "review-upload-before-versioning"
+				initRec := retentionAuthzRequest(t, router, root, http.MethodPost, makeTestTargetURL("", bucket, key, url.Values{"uploads": {""}}), nil, nil)
+				if initRec.Code != http.StatusOK {
+					t.Fatalf("init: %d %s", initRec.Code, initRec.Body.String())
+				}
+				var mp InitiateMultipartUploadResponse
+				if err := xml.Unmarshal(initRec.Body.Bytes(), &mp); err != nil {
+					t.Fatal(err)
+				}
+				part := retentionAuthzRequest(t, router, root, http.MethodPut, getPutObjectPartURL("", bucket, key, mp.UploadID, "1"), []byte("ordinary pending upload"), nil)
+				if part.Code != http.StatusOK {
+					t.Fatalf("part: %d %s", part.Code, part.Body.String())
+				}
+				if _, err := globalBucketMetadataSys.Update(t.Context(), bucket, bucketVersioningConfig, []byte(`<VersioningConfiguration><Status>Enabled</Status></VersioningConfiguration>`)); err != nil {
+					t.Fatal(err)
+				}
+				body := fmt.Appendf(nil, `<CompleteMultipartUpload><Part><PartNumber>1</PartNumber><ETag>%s</ETag></Part></CompleteMultipartUpload>`, multipartConditionResponseETag(part))
+				rec := retentionAuthzRequest(t, router, root, http.MethodPost, makeTestTargetURL("", bucket, key, url.Values{"uploadId": {mp.UploadID}}), body, nil)
+				if rec.Code != http.StatusOK {
+					t.Fatalf("ordinary upload initiated before versioning fails completion: %d %s", rec.Code, rec.Body.String())
+				}
+			})
+			t.Run(instance+"/addressed-upload-after-suspension", func(t *testing.T) {
+				key := "review-addressed-before-suspension"
+				original := []byte("original")
+				info, err := obj.PutObject(t.Context(), bucket, key, mustGetPutObjReader(t, bytes.NewReader(original), int64(len(original)), "", ""), ObjectOptions{Versioned: true})
+				if err != nil {
+					t.Fatal(err)
+				}
+				up, err := obj.NewMultipartUpload(t.Context(), bucket, key, ObjectOptions{Versioned: true, VersionID: info.VersionID})
+				if err != nil {
+					t.Fatal(err)
+				}
+				part := []byte("replaced existing version")
+				pi, err := obj.PutObjectPart(t.Context(), bucket, key, up.UploadID, 1, mustGetPutObjReader(t, bytes.NewReader(part), int64(len(part)), "", ""), ObjectOptions{})
+				if err != nil {
+					t.Fatal(err)
+				}
+				if _, err := globalBucketMetadataSys.Update(t.Context(), bucket, bucketVersioningConfig, []byte(`<VersioningConfiguration><Status>Suspended</Status></VersioningConfiguration>`)); err != nil {
+					t.Fatal(err)
+				}
+				writer := newObjectAttributesAuthzUser(t, instance, bucket, `"s3:PutObject"`)
+				body := fmt.Appendf(nil, `<CompleteMultipartUpload><Part><PartNumber>1</PartNumber><ETag>"%s"</ETag></Part></CompleteMultipartUpload>`, pi.ETag)
+				rec := retentionAuthzRequest(t, router, writer, http.MethodPost, makeTestTargetURL("", bucket, key, url.Values{"uploadId": {up.UploadID}}), body, nil)
+				after, err := obj.GetObjectInfo(t.Context(), bucket, key, ObjectOptions{VersionID: info.VersionID})
+				if err != nil {
+					t.Fatal(err)
+				}
+				if rec.Code != http.StatusForbidden || after.ETag != info.ETag {
+					t.Fatalf("suspended-bucket existing UUID overwritten: status=%d old-etag=%s new-etag=%s body=%s", rec.Code, info.ETag, after.ETag, rec.Body.String())
+				}
+			})
+		},
+	})
+}

cmd/object-retention-overwrite_test.go

@@ -0,0 +1,546 @@
+// Copyright (c) 2026 Feng Ruohang
+// SPDX-License-Identifier: AGPL-3.0-or-later
+//
+// 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"
+	"crypto/md5"
+	"encoding/base64"
+	"errors"
+	"fmt"
+	"io"
+	"net/http"
+	"net/http/httptest"
+	"net/url"
+	"strings"
+	"sync"
+	"testing"
+	"time"
+
+	"github.com/minio/minio/internal/amztime"
+	"github.com/minio/minio/internal/auth"
+	objectlock "github.com/minio/minio/internal/bucket/object/lock"
+	xhttp "github.com/minio/minio/internal/http"
+)
+
+func retentionOverwriteUnchanged(t *testing.T, before, after ObjectInfo) {
+	t.Helper()
+	if after.ETag != before.ETag || after.Size != before.Size || after.VersionID != before.VersionID || !after.ModTime.Equal(before.ModTime) {
+		t.Error("rejected write changed the stored object")
+	}
+	if objectlock.GetObjectRetentionMeta(after.UserDefined).String() != objectlock.GetObjectRetentionMeta(before.UserDefined).String() || objectlock.GetObjectLegalHoldMeta(after.UserDefined).Status != objectlock.GetObjectLegalHoldMeta(before.UserDefined).Status {
+		t.Error("rejected write changed the stored lock")
+	}
+}
+
+func TestObjectLockOverwriteConfigUnavailable(t *testing.T) {
+	previous := newObjectLayerFn()
+	setObjectLayer(nil)
+	defer setObjectLayer(previous)
+	err := checkObjectLockedForOverwrite(t.Context(), "lock-config-unavailable", "object", func(context.Context, string, string, ObjectOptions) (ObjectInfo, error) {
+		t.Fatal("object must not be read without a bucket configuration")
+		return ObjectInfo{}, nil
+	})
+	if !errors.Is(err, errServerNotInitialized) {
+		t.Fatalf("unavailable lock configuration must fail closed, got %v", err)
+	}
+}
+
+func TestPutObjectRetentionMaximumDaysPolicy(t *testing.T) {
+	ExecObjectLayerAPITest(ExecObjectLayerAPITestArgs{
+		t:                 t,
+		makeBucketOptions: MakeBucketOptions{VersioningEnabled: true, LockEnabled: true},
+		objAPITest: func(obj ObjectLayer, instance, bucket string, router http.Handler, _ auth.Credentials, t *testing.T) {
+			for _, conditionalAllow := range []bool{false, true} {
+				t.Run(fmt.Sprintf("%s/conditional-allow=%t", instance, conditionalAllow), func(t *testing.T) {
+					statement := fmt.Sprintf(`{"Effect":"Allow","Action":["s3:PutObjectRetention"],"Resource":["arn:aws:s3:::%s/*"],"Condition":{"NumericLessThanEquals":{"s3:object-lock-remaining-retention-days":"200000"}}}`, bucket)
+					if !conditionalAllow {
+						statement = fmt.Sprintf(`{"Effect":"Allow","Action":["s3:PutObjectRetention"],"Resource":["arn:aws:s3:::%s/*"]},{"Effect":"Deny","Action":["s3:PutObjectRetention"],"Resource":["arn:aws:s3:::%s/*"],"Condition":{"NumericGreaterThan":{"s3:object-lock-remaining-retention-days":"200000"}}}`, bucket, bucket)
+					}
+					cred := newRetentionPolicyUser(t, instance, `{"Version":"2012-10-17","Statement":[`+statement+`]}`)
+					key := fmt.Sprintf("maximum-retention-%t", conditionalAllow)
+					initial, err := obj.PutObject(t.Context(), bucket, key, mustGetPutObjReader(t, bytes.NewReader(nil), 0, "", ""), ObjectOptions{Versioned: true})
+					if err != nil {
+						t.Fatal(err)
+					}
+					// Both dates exceed time.Duration's range (~292 years). A saturated
+					// Time.Sub would report 106752 days and incorrectly allow 1000 years.
+					if status := putRetention(t, router, cred, bucket, key, initial.VersionID, "GOVERNANCE", UTCNow().AddDate(1000, 0, 0), false); status != http.StatusForbidden {
+						t.Fatalf("1000-year retention: want 403, got %d", status)
+					}
+					after, err := obj.GetObjectInfo(t.Context(), bucket, key, ObjectOptions{VersionID: initial.VersionID})
+					if err != nil {
+						t.Fatal(err)
+					}
+					retentionOverwriteUnchanged(t, initial, after)
+					if status := putRetention(t, router, cred, bucket, key, initial.VersionID, "GOVERNANCE", UTCNow().AddDate(500, 0, 0), false); status != http.StatusOK {
+						t.Fatalf("500-year retention: want 200, got %d", status)
+					}
+				})
+			}
+		},
+	})
+}
+
+func TestAPICopyObjectPreservesVersionAfterSuspension(t *testing.T) {
+	ExecObjectLayerAPITest(ExecObjectLayerAPITestArgs{
+		t:                 t,
+		makeBucketOptions: MakeBucketOptions{VersioningEnabled: true},
+		objAPITest: func(obj ObjectLayer, instance, bucket string, router http.Handler, root auth.Credentials, t *testing.T) {
+			t.Run(instance, func(t *testing.T) {
+				key := "self-copy-after-suspension"
+				original, err := obj.PutObject(t.Context(), bucket, key, mustGetPutObjReader(t, bytes.NewBufferString("payload"), 7, "", ""), ObjectOptions{
+					Versioned: true, UserDefined: map[string]string{"x-amz-meta-state": "original"},
+				})
+				if err != nil {
+					t.Fatal(err)
+				}
+				if _, err := globalBucketMetadataSys.Update(t.Context(), bucket, bucketVersioningConfig,
+					[]byte(`<VersioningConfiguration><Status>Suspended</Status></VersioningConfiguration>`)); err != nil {
+					t.Fatal(err)
+				}
+				rec := retentionAuthzRequest(t, router, root, http.MethodPut, makeTestTargetURL("", bucket, key, nil), nil, map[string]string{
+					xhttp.AmzCopySource: "/" + bucket + "/" + key, xhttp.AmzMetadataDirective: "REPLACE", "x-amz-meta-state": "replacement",
+				})
+				if rec.Code != http.StatusOK {
+					t.Fatalf("copy: %d %s", rec.Code, rec.Body.String())
+				}
+				old, err := obj.GetObjectInfo(t.Context(), bucket, key, ObjectOptions{VersionID: original.VersionID})
+				if err != nil {
+					t.Fatal(err)
+				}
+				if old.UserDefined["x-amz-meta-state"] != "original" {
+					t.Errorf("copy changed metadata of existing UUID %s: %v", original.VersionID, old.UserDefined)
+				}
+				current, err := obj.GetObjectInfo(t.Context(), bucket, key, ObjectOptions{VersionID: nullVersionID})
+				if err != nil {
+					t.Fatalf("copy must create the null version: %v", err)
+				}
+				if current.UserDefined["X-Amz-Meta-State"] != "replacement" {
+					t.Errorf("new null version metadata: %v", current.UserDefined)
+				}
+			})
+		},
+	})
+}
+
+// A persisted null upload can survive an upgrade or a versioning transition.
+// The completion must see locks committed after the handler's preflight.
+func TestMultipartNullObjectLockAfterPreflight(t *testing.T) {
+	ExecObjectLayerAPITest(ExecObjectLayerAPITestArgs{
+		t:                 t,
+		makeBucketOptions: MakeBucketOptions{VersioningEnabled: true, LockEnabled: true},
+		objAPITest: func(obj ObjectLayer, instance, bucket string, router http.Handler, root auth.Credentials, t *testing.T) {
+			writer := newObjectAttributesAuthzUser(t, instance, bucket, `"s3:PutObject"`)
+			for _, mode := range []string{"GOVERNANCE", "COMPLIANCE", "LEGAL-HOLD"} {
+				t.Run(instance+"/"+mode, func(t *testing.T) {
+					ctx := t.Context()
+					key := "multipart-null-" + mode
+					opts := ObjectOptions{Versioned: true, VersionID: nullVersionID, ReplicationRequest: true}
+					if _, err := obj.PutObject(ctx, bucket, key, mustGetPutObjReader(t, bytes.NewBufferString("original"), 8, "", ""), opts); err != nil {
+						t.Fatal(err)
+					}
+					mp, err := obj.NewMultipartUpload(ctx, bucket, key, opts)
+					if err != nil {
+						t.Fatal(err)
+					}
+					part, err := obj.PutObjectPart(ctx, bucket, key, mp.UploadID, 1, mustGetPutObjReader(t, bytes.NewBufferString("replacement"), 11, "", ""), ObjectOptions{})
+					if err != nil {
+						t.Fatal(err)
+					}
+					var locked ObjectInfo
+					hook := &hookBeforeWrite{ObjectLayer: obj, key: key, before: func() {
+						if mode == "LEGAL-HOLD" {
+							body := []byte(`<LegalHold><Status>ON</Status></LegalHold>`)
+							sum := md5.Sum(body)
+							rec := retentionAuthzRequest(t, router, root, http.MethodPut, makeTestTargetURL("", bucket, key, url.Values{"legal-hold": {""}, "versionId": {nullVersionID}}), body, map[string]string{xhttp.ContentMD5: base64.StdEncoding.EncodeToString(sum[:])})
+							if rec.Code != http.StatusOK {
+								t.Fatalf("set legal hold: %d %s", rec.Code, rec.Body.String())
+							}
+						} else if status := putRetention(t, router, root, bucket, key, nullVersionID, mode, UTCNow().Add(365*24*time.Hour), false); status != http.StatusOK {
+							t.Fatalf("set retention: %d", status)
+						}
+						var err error
+						locked, err = obj.GetObjectInfo(ctx, bucket, key, ObjectOptions{VersionID: nullVersionID})
+						if err != nil {
+							t.Fatal(err)
+						}
+					}}
+					setObjectLayer(hook)
+					defer setObjectLayer(obj)
+					body := []byte(multipartConditionCompleteBody([]CompletePart{{PartNumber: 1, ETag: part.ETag}}))
+					rec := retentionAuthzRequest(t, router, writer, http.MethodPost, makeTestTargetURL("", bucket, key, url.Values{"uploadId": {mp.UploadID}}), body, nil)
+					if hook.before != nil || rec.Code != http.StatusBadRequest {
+						t.Fatalf("completion after locking: %d %s", rec.Code, rec.Body.String())
+					}
+					multipartConditionError(t, rec, "InvalidRequest")
+					after, err := obj.GetObjectInfo(ctx, bucket, key, ObjectOptions{VersionID: nullVersionID})
+					if err != nil {
+						t.Fatal(err)
+					}
+					retentionOverwriteUnchanged(t, locked, after)
+					parts, err := obj.ListObjectParts(ctx, bucket, key, mp.UploadID, 0, 100, ObjectOptions{})
+					if err != nil || len(parts.Parts) != 1 || parts.Parts[0].ETag != part.ETag {
+						t.Fatalf("rejected completion removed upload parts: %v %+v", err, parts.Parts)
+					}
+				})
+			}
+		},
+	})
+}
+
+// The newer, unlocked copy is in the write pool; the lock exists only in the
+// other pool. A set-local check would miss it even while holding a write lock.
+func TestPoolsObjectLockOverwrite(t *testing.T) {
+	z, _ := consistencyPools(t)
+	bucket, _, err := initAPIHandlerTest(t.Context(), z, nil, MakeBucketOptions{VersioningEnabled: true, LockEnabled: true})
+	if err != nil {
+		t.Fatal(err)
+	}
+	for _, state := range []string{"GOVERNANCE", "COMPLIANCE", "LEGAL-HOLD", "expired", "unlocked", "unreadable"} {
+		for _, op := range []string{"put", "copy", "self-copy", "multipart"} {
+			t.Run(state+"/"+op, func(t *testing.T) {
+				ctx := t.Context()
+				key := encodeDirObject("null-" + state + "-" + op + "/")
+				meta := map[string]string{}
+				switch state {
+				case "GOVERNANCE", "COMPLIANCE":
+					meta[strings.ToLower(xhttp.AmzObjectLockMode)] = state
+					meta[strings.ToLower(xhttp.AmzObjectLockRetainUntilDate)] = amztime.ISO8601Format(UTCNow().Add(365 * 24 * time.Hour))
+				case "LEGAL-HOLD":
+					meta[strings.ToLower(xhttp.AmzObjectLockLegalHold)] = "ON"
+				case "expired":
+					meta[strings.ToLower(xhttp.AmzObjectLockMode)] = "GOVERNANCE"
+					meta[strings.ToLower(xhttp.AmzObjectLockRetainUntilDate)] = amztime.ISO8601Format(UTCNow().Add(-24 * time.Hour))
+				}
+				seed := ObjectOptions{Versioned: true, VersionID: nullVersionID, ReplicationRequest: true, MTime: UTCNow().Add(-time.Hour), UserDefined: meta}
+				putConsistencyObject(t, z, bucket, key, 1, "original", seed)
+				seed.UserDefined = map[string]string{}
+				seed.MTime = UTCNow()
+				putConsistencyObject(t, z, bucket, key, 0, "original", seed)
+				before := make([]ObjectInfo, len(z.serverPools))
+				for i, pool := range z.serverPools {
+					var err error
+					before[i], err = pool.GetObjectInfo(ctx, bucket, key, ObjectOptions{VersionID: nullVersionID})
+					if err != nil {
+						t.Fatal(err)
+					}
+				}
+				defer conditionalPutPool(t, z, key, 0)()
+				var restore func()
+				if state == "unreadable" {
+					restore = conditionalPutSwapDisks(z.serverPools[1], key, func(d StorageAPI) StorageAPI {
+						return consistencyReadFaultDisk{StorageAPI: d, bucket: bucket, object: key}
+					})
+					defer restore()
+				}
+				opts := ObjectOptions{Versioned: true, VersionID: nullVersionID, UserDefined: map[string]string{"x-amz-meta-updated": "true"}}
+				var writeErr error
+				switch op {
+				case "put":
+					_, writeErr = z.PutObject(ctx, bucket, key, mustGetPutObjReader(t, bytes.NewBufferString("replacement"), 11, "", ""), opts)
+				case "copy", "self-copy":
+					source := key
+					if op == "copy" {
+						source = "source-" + state
+						putConsistencyObject(t, z, bucket, source, 0, "replacement", ObjectOptions{Versioned: true})
+					}
+					gr, gerr := z.GetObjectNInfo(ctx, bucket, source, nil, nil, ObjectOptions{})
+					if gerr != nil {
+						t.Fatal(gerr)
+					}
+					defer gr.Close()
+					info := gr.ObjInfo
+					info.UserDefined = opts.UserDefined
+					info.metadataOnly = op == "self-copy"
+					info.PutObjReader = mustGetPutObjReader(t, gr, info.Size, "", "")
+					info.Reader = info.PutObjReader.Reader
+					_, writeErr = z.CopyObject(ctx, bucket, source, bucket, key, info, ObjectOptions{VersionID: nullVersionID}, opts)
+				case "multipart":
+					id, parts := multipartConditionUpload(t, z, bucket, key, 0, opts)
+					_, writeErr = z.CompleteMultipartUpload(ctx, bucket, key, id, parts, ObjectOptions{Versioned: true})
+				}
+				allowed := state == "expired" || state == "unlocked"
+				if allowed {
+					if writeErr != nil {
+						t.Fatalf("unlocked write: %v", writeErr)
+					}
+					return
+				}
+				if state == "unreadable" {
+					if !isErrReadQuorum(writeErr) {
+						t.Fatalf("unreadable pool: want quorum error, got %v", writeErr)
+					}
+					restore()
+				} else if !errors.As(writeErr, &ObjectLocked{}) {
+					t.Fatalf("lock in other pool: want ObjectLocked, got %v", writeErr)
+				}
+				for i, original := range before {
+					after, gerr := z.serverPools[i].GetObjectInfo(ctx, bucket, key, ObjectOptions{VersionID: nullVersionID})
+					if gerr != nil {
+						t.Fatal(gerr)
+					}
+					retentionOverwriteUnchanged(t, original, after)
+				}
+			})
+		}
+	}
+}
+
+type planReviewGatedBody struct {
+	io.ReadCloser
+	first   sync.Once
+	ready   chan struct{}
+	release chan struct{}
+}
+
+func (g *planReviewGatedBody) Read(p []byte) (int, error) {
+	g.first.Do(func() { close(g.ready); <-g.release })
+	return g.ReadCloser.Read(p)
+}
+
+func TestPlanReviewSlowDirectoryPut(t *testing.T) {
+	ExecObjectLayerAPITest(ExecObjectLayerAPITestArgs{
+		t:                 t,
+		makeBucketOptions: MakeBucketOptions{VersioningEnabled: true, LockEnabled: true},
+		objAPITest: func(obj ObjectLayer, instance, bucket string, router http.Handler, root auth.Credentials, t *testing.T) {
+			writer := newObjectAttributesAuthzUser(t, instance, bucket, `"s3:PutObject"`)
+			t.Run(instance, func(t *testing.T) {
+				key := "slow-body-directory/"
+				before, err := obj.PutObject(t.Context(), bucket, key, mustGetPutObjReader(t, bytes.NewBufferString("original"), 8, "", ""), ObjectOptions{Versioned: true, VersionID: nullVersionID})
+				if err != nil {
+					t.Fatal(err)
+				}
+				payload := []byte("replacement-body")
+				req, err := newTestSignedRequestV4(http.MethodPut, makeTestTargetURL("", bucket, key, nil), int64(len(payload)), bytes.NewReader(payload), writer.AccessKey, writer.SecretKey, nil)
+				if err != nil {
+					t.Fatal(err)
+				}
+				gate := &planReviewGatedBody{ReadCloser: req.Body, ready: make(chan struct{}), release: make(chan struct{})}
+				req.Body = gate
+				var once sync.Once
+				release := func() { once.Do(func() { close(gate.release) }) }
+				defer release()
+				rec := httptest.NewRecorder()
+				done := make(chan struct{})
+				go func() { router.ServeHTTP(rec, req); close(done) }()
+				select {
+				case <-gate.ready:
+				case <-done:
+					t.Fatalf("PUT finished before body: %d %s", rec.Code, rec.Body.String())
+				case <-time.After(5 * time.Second):
+					t.Fatal("PUT did not reach body read")
+				}
+				locked := make(chan int, 1)
+				go func() {
+					locked <- putRetention(t, router, root, bucket, key, nullVersionID, "COMPLIANCE", UTCNow().Add(365*24*time.Hour), false)
+				}()
+				select {
+				case status := <-locked:
+					if status != http.StatusOK {
+						t.Fatalf("set lock while body paused: %d", status)
+					}
+				case <-time.After(5 * time.Second):
+					t.Fatal("retention update blocked behind paused body")
+				}
+				release()
+				select {
+				case <-done:
+				case <-time.After(5 * time.Second):
+					t.Fatal("PUT did not finish")
+				}
+				after, err := obj.GetObjectInfo(t.Context(), bucket, key, ObjectOptions{VersionID: nullVersionID})
+				if err != nil {
+					t.Fatal(err)
+				}
+				if rec.Code != http.StatusBadRequest || after.ETag != before.ETag || objectlock.GetObjectRetentionMeta(after.UserDefined).Mode != objectlock.RetCompliance {
+					t.Fatal("paused ordinary PUT overwrote the COMPLIANCE null version after the lock was acknowledged")
+				}
+			})
+		},
+	})
+}
+
+type planReviewReadFault struct {
+	ObjectLayer
+	key       string
+	fired     bool
+	putCalled bool
+}
+
+func (p *planReviewReadFault) GetObjectInfo(ctx context.Context, bucket, key string, opts ObjectOptions) (ObjectInfo, error) {
+	if key == p.key && opts.VersionID == nullVersionID && !p.fired {
+		p.fired = true
+		return ObjectInfo{}, InsufficientReadQuorum{}
+	}
+	return p.ObjectLayer.GetObjectInfo(ctx, bucket, key, opts)
+}
+
+func (p *planReviewReadFault) PutObject(ctx context.Context, bucket, key string, data *PutObjReader, opts ObjectOptions) (ObjectInfo, error) {
+	if key == p.key {
+		p.putCalled = true
+	}
+	return p.ObjectLayer.PutObject(ctx, bucket, key, data, opts)
+}
+
+func TestPlanReviewPreflightReadFailure(t *testing.T) {
+	ExecObjectLayerAPITest(ExecObjectLayerAPITestArgs{
+		t:                 t,
+		makeBucketOptions: MakeBucketOptions{VersioningEnabled: true, LockEnabled: true},
+		objAPITest: func(obj ObjectLayer, instance, bucket string, router http.Handler, _ auth.Credentials, t *testing.T) {
+			writer := newObjectAttributesAuthzUser(t, instance, bucket, `"s3:PutObject"`)
+			t.Run(instance, func(t *testing.T) {
+				key := "locked-directory-read-fault/"
+				_, err := obj.PutObject(t.Context(), bucket, key, mustGetPutObjReader(t, bytes.NewBufferString("original"), 8, "", ""), ObjectOptions{
+					Versioned: true, VersionID: nullVersionID, ReplicationRequest: true,
+					UserDefined: map[string]string{
+						strings.ToLower(xhttp.AmzObjectLockMode):            "COMPLIANCE",
+						strings.ToLower(xhttp.AmzObjectLockRetainUntilDate): amztime.ISO8601Format(UTCNow().Add(365 * 24 * time.Hour)),
+					},
+				})
+				if err != nil {
+					t.Fatal(err)
+				}
+				initial, err := obj.GetObjectInfo(t.Context(), bucket, key, ObjectOptions{VersionID: nullVersionID})
+				if err != nil {
+					t.Fatal(err)
+				}
+				fault := &planReviewReadFault{ObjectLayer: obj, key: key}
+				setObjectLayer(fault)
+				defer setObjectLayer(obj)
+				rec := retentionAuthzRequest(t, router, writer, http.MethodPut, makeTestTargetURL("", bucket, key, nil), []byte("replacement"), nil)
+				if !fault.fired {
+					t.Fatal("preflight fault was not injected")
+				}
+				if fault.putCalled {
+					t.Fatal("preflight read error allowed PutObject to proceed to storage layer")
+				}
+				after, err := obj.GetObjectInfo(t.Context(), bucket, key, ObjectOptions{VersionID: nullVersionID})
+				if err != nil {
+					t.Fatal(err)
+				}
+				if rec.Code < 400 {
+					t.Errorf("metadata read quorum error was treated as absence: status=%d", rec.Code)
+				}
+				retentionOverwriteUnchanged(t, initial, after)
+			})
+		},
+	})
+}
+
+func TestPlanReviewLegacyMPUCompliance(t *testing.T) {
+	ExecObjectLayerAPITest(ExecObjectLayerAPITestArgs{
+		t:                 t,
+		makeBucketOptions: MakeBucketOptions{VersioningEnabled: true, LockEnabled: true},
+		objAPITest: func(obj ObjectLayer, instance, bucket string, router http.Handler, root auth.Credentials, t *testing.T) {
+			writer := newObjectAttributesAuthzUser(t, instance, bucket, `"s3:PutObject"`)
+			t.Run(instance, func(t *testing.T) {
+				key := "legacy-mpu-compliance"
+				initial, err := obj.PutObject(t.Context(), bucket, key, mustGetPutObjReader(t, bytes.NewBufferString("original"), 8, "", ""), ObjectOptions{Versioned: true})
+				if err != nil {
+					t.Fatal(err)
+				}
+				up, err := obj.NewMultipartUpload(t.Context(), bucket, key, ObjectOptions{Versioned: true, VersionID: initial.VersionID})
+				if err != nil {
+					t.Fatal(err)
+				}
+				payload := []byte("replacement-part")
+				part, err := obj.PutObjectPart(t.Context(), bucket, key, up.UploadID, 1, mustGetPutObjReader(t, bytes.NewReader(payload), int64(len(payload)), "", ""), ObjectOptions{})
+				if err != nil {
+					t.Fatal(err)
+				}
+				if status := putRetention(t, router, root, bucket, key, initial.VersionID, "COMPLIANCE", UTCNow().Add(365*24*time.Hour), false); status != http.StatusOK {
+					t.Fatalf("set compliance: %d", status)
+				}
+				before, err := obj.GetObjectInfo(t.Context(), bucket, key, ObjectOptions{VersionID: initial.VersionID})
+				if err != nil {
+					t.Fatal(err)
+				}
+				body := fmt.Appendf(nil, `<CompleteMultipartUpload><Part><PartNumber>1</PartNumber><ETag>"%s"</ETag></Part></CompleteMultipartUpload>`, part.ETag)
+				rec := retentionAuthzRequest(t, router, writer, http.MethodPost, makeTestTargetURL("", bucket, key, url.Values{"uploadId": {up.UploadID}}), body, nil)
+				after, err := obj.GetObjectInfo(t.Context(), bucket, key, ObjectOptions{VersionID: initial.VersionID})
+				if err != nil {
+					t.Fatal(err)
+				}
+				if rec.Code != http.StatusForbidden {
+					t.Errorf("legacy upload replaced locked UUID: status=%d", rec.Code)
+				}
+				retentionOverwriteUnchanged(t, before, after)
+			})
+		},
+	})
+}
+
+// A recovered upload retains an authentic server-generated version ID while a
+// surviving replica reintroduces its already-completed version and lock.
+// Only the fixture uses replication privileges; completion uses PutObject only.
+func TestRecoveredMultipartUploadAfterReplica(t *testing.T) {
+	ExecObjectLayerAPITest(ExecObjectLayerAPITestArgs{
+		t:                 t,
+		makeBucketOptions: MakeBucketOptions{VersioningEnabled: true, LockEnabled: true},
+		objAPITest: func(obj ObjectLayer, instance, bucket string, router http.Handler, root auth.Credentials, t *testing.T) {
+			writer := newObjectAttributesAuthzUser(t, instance, bucket, `"s3:PutObject"`)
+			t.Run(instance, func(t *testing.T) {
+				z := obj.(*erasureServerPools)
+				key := "recovered-generated-upload"
+				up, err := obj.NewMultipartUpload(t.Context(), bucket, key, ObjectOptions{Versioned: true})
+				if err != nil {
+					t.Fatal(err)
+				}
+				er := z.serverPools[0].getHashedSet(key)
+				fi, _, err := er.checkUploadIDExists(t.Context(), bucket, key, up.UploadID, true)
+				if err != nil {
+					t.Fatal(err)
+				}
+				_, err = obj.PutObject(t.Context(), bucket, key,
+					mustGetPutObjReader(t, bytes.NewBufferString("protected-original"), 18, "", ""),
+					ObjectOptions{Versioned: true, VersionID: fi.VersionID, ReplicationRequest: true})
+				if err != nil {
+					t.Fatal(err)
+				}
+				if status := putRetention(t, router, root, bucket, key, fi.VersionID, "COMPLIANCE", UTCNow().Add(365*24*time.Hour), false); status != http.StatusOK {
+					t.Fatalf("setting compliance lock: %d", status)
+				}
+				before, err := obj.GetObjectInfo(t.Context(), bucket, key, ObjectOptions{VersionID: fi.VersionID})
+				if err != nil {
+					t.Fatal(err)
+				}
+				partRec := retentionAuthzRequest(t, router, writer, http.MethodPut,
+					makeTestTargetURL("", bucket, key, url.Values{"uploadId": {up.UploadID}, "partNumber": {"1"}}),
+					[]byte("attacker-replacement"), nil)
+				if partRec.Code != http.StatusOK {
+					t.Fatalf("upload part: %d %s", partRec.Code, partRec.Body.String())
+				}
+				etag := multipartConditionResponseETag(partRec)
+				body := []byte(multipartConditionCompleteBody([]CompletePart{{PartNumber: 1, ETag: etag}}))
+				rec := retentionAuthzRequest(t, router, writer, http.MethodPost,
+					makeTestTargetURL("", bucket, key, url.Values{"uploadId": {up.UploadID}}), body, nil)
+				after, err := obj.GetObjectInfo(t.Context(), bucket, key, ObjectOptions{VersionID: fi.VersionID})
+				if err != nil {
+					t.Fatal(err)
+				}
+				if rec.Code != http.StatusForbidden {
+					t.Errorf("recovered upload bypassed existing UUID guard: status=%d", rec.Code)
+				}
+				retentionOverwriteUnchanged(t, before, after)
+			})
+		},
+	})
+}

cmd/replication-tagging-order_test.go

@@ -238,7 +238,7 @@ func r5APIOrdering(t *testing.T, encrypted bool) {
 						if err != nil {
 							t.Fatal(err)
 						}
-						original, err = obj.CompleteMultipartUpload(t.Context(), bucket, name, mp.UploadID, []CompletePart{{PartNumber: 1, ETag: part.ETag}}, ObjectOptions{Versioned: true})
+						original, err = obj.CompleteMultipartUpload(t.Context(), bucket, name, mp.UploadID, []CompletePart{{PartNumber: 1, ETag: part.ETag}}, ObjectOptions{Versioned: true, ReplicationRequest: true})
 						if err != nil {
 							t.Fatal(err)
 						}

cmd/storage-errors.go

@@ -70,6 +70,11 @@ var errDiskAccessDenied = StorageErr("drive access denied")
 // errFileNotFound - cannot find the file.
 var errFileNotFound = StorageErr("file not found")
 
+// errCommitMetaMissing - the staged xl.meta disappeared between the
+// data rename and the meta-file commit; a write-quorum failure, not
+// a missing object.
+var errCommitMetaMissing = StorageErr("commit metadata is missing")
+
 // errFileNotFound - cannot find requested file version.
 var errFileVersionNotFound = StorageErr("file version not found")
 

cmd/storage-rest-client.go

@@ -106,6 +106,8 @@ func toStorageErr(err error) error {
 		return errVolumeExists
 	case errFileNotFound.Error():
 		return errFileNotFound
+	case errCommitMetaMissing.Error():
+		return errCommitMetaMissing
 	case errFileVersionNotFound.Error():
 		return errFileVersionNotFound
 	case errFileNameTooLong.Error():

cmd/storage-rest-server.go

@@ -290,6 +290,9 @@ func (s *storageRESTServer) NSScannerHandler(ctx context.Context, params *nsScan
 		for update := range updates {
 			resp := storageNSScannerRPC.NewResponse()
 			resp.Update = &update
+			// Pooled responses may carry fields from a previous
+			// scan, always clear the fields not being set.
+			resp.Final = nil
 			out <- resp
 		}
 	}()
@@ -301,6 +304,7 @@ func (s *storageRESTServer) NSScannerHandler(ctx context.Context, params *nsScan
 	// Send final response.
 	resp := storageNSScannerRPC.NewResponse()
 	resp.Final = &ui
+	resp.Update = nil
 	out <- resp
 	return nil
 }

cmd/storage-rest-server_test.go

@@ -0,0 +1,94 @@
+// Copyright (c) 2026 Julien Laurenceau and Feng Ruohang
+//
+// This file is part of Silo Object Storage.
+//
+// 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"
+
+	"github.com/minio/madmin-go/v3"
+	"github.com/minio/minio/internal/grid"
+)
+
+type scannerResponseTestDisk struct {
+	StorageAPI
+}
+
+func (scannerResponseTestDisk) NSScanner(_ context.Context, cache dataUsageCache, updates chan<- dataUsageEntry, _ madmin.HealScanMode, _ func() bool) (dataUsageCache, error) {
+	defer close(updates)
+	updates <- dataUsageEntry{Size: int64(cache.Info.NextCycle)}
+	cache.Info.NextCycle++
+	return cache, nil
+}
+
+func TestNSScannerHandlerClearsStaleResponses(t *testing.T) {
+	endpoint, err := NewEndpoint(t.TempDir())
+	if err != nil {
+		t.Fatal(err)
+	}
+	globalLocalDrivesMu.Lock()
+	oldDrives, oldSets := globalLocalDrivesMap, globalLocalSetDrives
+	globalLocalDrivesMap = map[string]StorageAPI{endpoint.String(): scannerResponseTestDisk{}}
+	globalLocalSetDrives = nil
+	globalLocalDrivesMu.Unlock()
+	t.Cleanup(func() {
+		globalLocalDrivesMu.Lock()
+		globalLocalDrivesMap, globalLocalSetDrives = oldDrives, oldSets
+		globalLocalDrivesMu.Unlock()
+	})
+
+	oldRPC := storageNSScannerRPC
+	t.Cleanup(func() { storageNSScannerRPC = oldRPC })
+	// Every new allocation is dirty too: the check must not depend on
+	// sync.Pool returning a previously used object.
+	storageNSScannerRPC = grid.NewStream[*nsScannerOptions, grid.NoPayload, *nsScannerResp](grid.HandlerNSScanner,
+		func() *nsScannerOptions { return &nsScannerOptions{} }, nil,
+		func() *nsScannerResp {
+			return &nsScannerResp{Update: &dataUsageEntry{Size: -1}, Final: &dataUsageCache{}}
+		})
+	server := storageRESTServer{endpoint: endpoint}
+	for cycle := uint32(1); cycle <= 3; cycle++ {
+		out := make(chan *nsScannerResp, 2)
+		params := &nsScannerOptions{Cache: &dataUsageCache{Info: dataUsageCacheInfo{Name: "bucket", NextCycle: cycle}}}
+		if err := server.NSScannerHandler(t.Context(), params, out); err != nil {
+			t.Fatal(err)
+		}
+		if len(out) != 2 {
+			t.Fatalf("cycle %d: got %d responses, want an update and a final cache", cycle, len(out))
+		}
+		for i := 0; i < 2; i++ {
+			resp := <-out
+			wire, err := resp.MarshalMsg(nil)
+			storageNSScannerRPC.PutResponse(resp)
+			if err != nil {
+				t.Fatal(err)
+			}
+			var decoded nsScannerResp
+			if _, err := decoded.UnmarshalMsg(wire); err != nil {
+				t.Fatal(err)
+			}
+			if i == 0 {
+				if decoded.Update == nil || decoded.Final != nil || decoded.Update.Size != int64(cycle) {
+					t.Errorf("cycle %d: invalid update response: %+v", cycle, decoded)
+				}
+			} else if decoded.Update != nil || decoded.Final == nil || decoded.Final.Info.NextCycle != cycle+1 {
+				t.Errorf("cycle %d: invalid final response: %+v", cycle, decoded)
+			}
+		}
+	}
+}

cmd/xl-storage-errors_test.go

@@ -18,10 +18,16 @@
 package cmd
 
 import (
+	"context"
+	"errors"
+	"net/http"
 	"os"
+	"path/filepath"
 	"runtime"
+	"sync"
 	"syscall"
 	"testing"
+	"time"
 )
 
 func TestSysErrors(t *testing.T) {
@@ -56,3 +62,120 @@ func TestSysErrors(t *testing.T) {
 		}
 	}
 }
+
+// TestCommitMetaMissingMapping verifies the RPC and object-layer mapping
+// of errCommitMetaMissing, returned when the staged xl.meta disappears
+// between the data rename and the meta-file commit in RenameData().
+func TestCommitMetaMissingMapping(t *testing.T) {
+	// The sentinel must survive a storage REST/grid round-trip; errors
+	// cross the wire as strings and are remapped by toStorageErr().
+	if got := toStorageErr(errors.New(errCommitMetaMissing.Error())); !errors.Is(got, errCommitMetaMissing) {
+		t.Fatalf("expected %s, got %s", errCommitMetaMissing, got)
+	}
+	if got := toStorageErr(errCommitMetaMissing); got != errCommitMetaMissing {
+		t.Fatalf("expected %s, got %s", errCommitMetaMissing, got)
+	}
+
+	// toObjectErr() must map the sentinel to InsufficientWriteQuorum
+	// (503), never to ObjectNotFound (404).
+	err := toObjectErr(errCommitMetaMissing, "bucket", "object")
+	qerr, ok := err.(InsufficientWriteQuorum)
+	if !ok {
+		t.Fatalf("expected InsufficientWriteQuorum, got %s (%T)", err, err)
+	}
+	if qerr.Bucket != "bucket" || qerr.Object != "object" {
+		t.Fatalf("unexpected bucket/object %s/%s", qerr.Bucket, qerr.Object)
+	}
+	apiErr := toAPIError(t.Context(), err)
+	if apiErr.Code != "SlowDownWrite" || apiErr.HTTPStatusCode != http.StatusServiceUnavailable {
+		t.Fatalf("expected SlowDownWrite/503, got %+v", apiErr)
+	}
+}
+
+// missingCommitMetaContext removes staged metadata at an existing cancellation
+// check, after WriteAll and (for non-inline objects) the data rename succeeded.
+// This deterministically models concurrent cleanup without changing production
+// code or relying on a filesystem race. The context itself remains live.
+type missingCommitMetaContext struct {
+	context.Context
+	mu             sync.Mutex
+	meta, dataDest string
+	removed        bool
+	removeErr      error
+}
+
+func (c *missingCommitMetaContext) Done() <-chan struct{} {
+	c.mu.Lock()
+	defer c.mu.Unlock()
+	if !c.removed {
+		if _, err := os.Stat(c.meta); err == nil {
+			dataMoved := c.dataDest == ""
+			if !dataMoved {
+				_, err = os.Stat(c.dataDest)
+				dataMoved = err == nil
+			}
+			if dataMoved {
+				c.removeErr = os.Remove(c.meta)
+				c.removed = c.removeErr == nil
+			}
+		}
+	}
+	return c.Context.Done()
+}
+
+func TestXLStorageRenameDataCommitMetaMissing(t *testing.T) {
+	for _, tc := range []struct {
+		name     string
+		dataFile bool
+	}{
+		{"empty_object", false},
+		{"data_directory", true},
+	} {
+		t.Run(tc.name, func(t *testing.T) {
+			disk, root, err := newXLStorageTestSetup(t)
+			if err != nil {
+				t.Fatal(err)
+			}
+			defer disk.Close()
+			s := disk.storage
+			if err := s.MakeVol(t.Context(), "bucket"); err != nil {
+				t.Fatal(err)
+			}
+			fi := newFileInfo("object", 2, 2)
+			fi.DataDir, fi.VersionID = mustGetUUID(), mustGetUUID()
+			fi.Erasure.Index = 1
+			fi.ModTime = time.Now().UTC()
+			fi.Fresh = true
+			ctx := &missingCommitMetaContext{
+				Context: t.Context(),
+				meta:    filepath.Join(root, minioMetaTmpBucket, "staged", xlStorageFormatFile),
+			}
+			if tc.dataFile {
+				fi.Size = 1
+				fi.AddObjectPart(1, "", 1, 1, fi.ModTime, nil, nil)
+				part := filepath.Join(root, minioMetaTmpBucket, "staged", fi.DataDir, "part.1")
+				if err := os.MkdirAll(filepath.Dir(part), 0o755); err != nil {
+					t.Fatal(err)
+				}
+				if err := os.WriteFile(part, []byte("a"), 0o644); err != nil {
+					t.Fatal(err)
+				}
+				ctx.dataDest = filepath.Join(root, "bucket", "object", fi.DataDir)
+			}
+			_, err = s.RenameData(ctx, minioMetaTmpBucket, "staged", fi, "bucket", "object", RenameOptions{})
+			if !ctx.removed {
+				t.Fatalf("metadata removal did not execute: removal error = %v, RenameData error = %v", ctx.removeErr, err)
+			}
+			if !errors.Is(err, errCommitMetaMissing) {
+				t.Fatalf("expected commit metadata missing, got %v", err)
+			}
+			apiErr := toAPIError(t.Context(), toObjectErr(err, "bucket", "object"))
+			if apiErr.Code != "SlowDownWrite" || apiErr.HTTPStatusCode != http.StatusServiceUnavailable {
+				t.Fatalf("expected SlowDownWrite/503, got %+v", apiErr)
+			}
+			if _, err := os.Stat(filepath.Join(root, "bucket", "object", xlStorageFormatFile)); !os.IsNotExist(err) {
+				t.Fatalf("failed commit must not produce destination metadata: %v", err)
+			}
+		})
+	}
+}

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-legacy.go

@@ -40,6 +40,10 @@ func (x *xlMetaV2VersionHeader) unmarshalV(v uint8, bts []byte) (o []byte, err e
 
 // unmarshalV1 decodes version 1, never released.
 func (x *xlMetaV2VersionHeader) unmarshalV1(bts []byte) (o []byte, err error) {
+	// Fields absent from the V1 wire format must be reset since the
+	// header may be reused across decodes.
+	x.Signature = [4]byte{}
+	x.EcN, x.EcM = 0, 0
 	var zb0001 uint32
 	zb0001, bts, err = msgp.ReadArrayHeaderBytes(bts)
 	if err != nil {
@@ -134,7 +138,7 @@ type xlMetaV2VersionHeaderV2 struct {
 
 // UnmarshalMsg implements msgp.Unmarshaler
 func (z *xlMetaV2VersionHeaderV2) UnmarshalMsg(bts []byte) (o []byte, err error) {
-	z.EcN, z.EcN = 0, 0
+	z.EcN, z.EcM = 0, 0
 	var zb0001 uint32
 	zb0001, bts, err = msgp.ReadArrayHeaderBytes(bts)
 	if err != nil {
@@ -184,7 +188,7 @@ func (z *xlMetaV2VersionHeaderV2) UnmarshalMsg(bts []byte) (o []byte, err error)
 
 // DecodeMsg implements msgp.Decodable
 func (z *xlMetaV2VersionHeaderV2) DecodeMsg(dc *msgp.Reader) (err error) {
-	z.EcN, z.EcN = 0, 0
+	z.EcN, z.EcM = 0, 0
 	var zb0001 uint32
 	zb0001, err = dc.ReadArrayHeader()
 	if err != nil {

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.
-func countXLV2Versions(tops []xlMetaV2ShallowVersion, reference xlMetaV2VersionHeader, latest xlMetaV2ShallowVersion, strict bool) (xlMetaV2ShallowVersion, int) {
+// VersionID and the last matching entry. The mixed-EC null-version recount adds
+// 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.
-			if !a.header.matchesEC(reference) {
+			// The narrow null-version recount can consider other EC groups,
+			// 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) {
-			// Tiebreak, use sort.
-			continue
+		if v == latestCount {
+			if allowMixedEC {
+				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

cmd/xl-storage-format-v2_test.go

@@ -38,6 +38,7 @@ import (
 	"github.com/klauspost/compress/zstd"
 	"github.com/minio/minio/internal/bucket/lifecycle"
 	xhttp "github.com/minio/minio/internal/http"
+	"github.com/tinylib/msgp/msgp"
 )
 
 func TestReadXLMetaNoData(t *testing.T) {
@@ -1190,3 +1191,92 @@ func BenchmarkToFileInfoNoParts(b *testing.B) {
 func BenchmarkToFileInfoWithParts(b *testing.B) {
 	benchmarkManyPartsOptionally(b, true)
 }
+
+// TestXLMetaV2HeaderLegacyUnmarshalResetsEc checks that decoding legacy
+// version headers, which carry no erasure metadata fields, cannot leak
+// stale EcN/EcM values from a previously decoded object into the same
+// structure.
+func TestXLMetaV2HeaderLegacyUnmarshalResetsEc(t *testing.T) {
+	base := xlMetaV2VersionHeader{
+		VersionID: [16]byte{1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16},
+		ModTime:   time.Now().Unix(),
+		Signature: [4]byte{1, 2, 3, 4},
+		Type:      ObjectType,
+		Flags:     xlFlagInlineData,
+	}
+
+	// Legacy V2 header wire format: 5 fields, no EcN/EcM.
+	var legacy []byte
+	legacy = msgp.AppendArrayHeader(legacy, 5)
+	legacy = msgp.AppendBytes(legacy, base.VersionID[:])
+	legacy = msgp.AppendInt64(legacy, base.ModTime)
+	legacy = msgp.AppendBytes(legacy, base.Signature[:])
+	legacy = msgp.AppendUint8(legacy, uint8(base.Type))
+	legacy = msgp.AppendUint8(legacy, uint8(base.Flags))
+
+	for _, decoder := range []string{"UnmarshalMsg", "DecodeMsg", "loadIndexed"} {
+		t.Run(decoder, func(t *testing.T) {
+			header := base
+			header.EcN, header.EcM = 4, 8
+			x2 := xlMetaV2VersionHeaderV2{xlMetaV2VersionHeader: &header}
+			var err error
+			switch decoder {
+			case "UnmarshalMsg":
+				_, err = x2.UnmarshalMsg(legacy)
+			case "DecodeMsg":
+				err = x2.DecodeMsg(msgp.NewReader(bytes.NewReader(legacy)))
+			case "loadIndexed":
+				indexed := msgp.AppendUint(nil, 2)
+				indexed = msgp.AppendUint(indexed, xlMetaVersion)
+				indexed = msgp.AppendInt(indexed, 1)
+				indexed = msgp.AppendBytes(indexed, legacy)
+				indexed = msgp.AppendBytes(indexed, nil)
+				xl := xlMetaV2{versions: []xlMetaV2ShallowVersion{{header: header}}}
+				err = xl.loadIndexed(indexed, nil)
+				header = xl.versions[0].header
+			}
+			if err != nil {
+				t.Fatal(err)
+			}
+			if header != base {
+				t.Fatalf("legacy header: got %+v, want %+v", header, base)
+			}
+		})
+	}
+
+	// V1 headers (4 fields, never released) must zero the fields too.
+	var v1 []byte
+	v1 = msgp.AppendArrayHeader(v1, 4)
+	v1 = msgp.AppendBytes(v1, base.VersionID[:])
+	v1 = msgp.AppendInt64(v1, base.ModTime)
+	v1 = msgp.AppendUint8(v1, uint8(base.Type))
+	v1 = msgp.AppendUint8(v1, uint8(base.Flags))
+
+	header := base
+	header.EcN, header.EcM = 4, 8
+	header.Signature = [4]byte{9, 9, 9, 9}
+	if _, err := header.unmarshalV1(v1); err != nil {
+		t.Fatalf("legacy V1 header unmarshal failed: %v", err)
+	}
+	if header.EcN != 0 || header.EcM != 0 {
+		t.Fatalf("stale erasure metadata leaked on V1 decode: EcN %d, EcM %d", header.EcN, header.EcM)
+	}
+	if header.Signature != ([4]byte{}) {
+		t.Fatalf("stale signature leaked on V1 decode: %v", header.Signature)
+	}
+
+	// Current headers must keep their erasure metadata.
+	current := base
+	current.EcN, current.EcM = 4, 8
+	wire, err := current.MarshalMsg(nil)
+	if err != nil {
+		t.Fatalf("header marshal failed: %v", err)
+	}
+	decoded := base
+	if _, err := decoded.UnmarshalMsg(wire); err != nil {
+		t.Fatalf("header unmarshal failed: %v", err)
+	}
+	if decoded.EcN != 4 || decoded.EcM != 8 {
+		t.Fatalf("erasure metadata lost on current decode: EcN %d, EcM %d", decoded.EcN, decoded.EcM)
+	}
+}

cmd/xl-storage.go

@@ -2925,7 +2925,15 @@ func (s *xlStorage) RenameData(ctx context.Context, srcVolume, srcPath string, f
 		// if its a partial rename() do not attempt to delete recursively.
 		// this can be healed since all parts are available.
 		s.deleteFile(dstVolumeDir, dstDataPath, false, false)
-		return res, osErrToFileErr(err)
+		err = osErrToFileErr(err)
+		if errors.Is(err, errFileNotFound) {
+			// The staged xl.meta disappeared between staging and commit.
+			// This is a write-quorum failure, not a missing object; use a
+			// dedicated error so the object layer returns 503 instead of
+			// mapping it to ObjectNotFound (404).
+			err = errCommitMetaMissing
+		}
+		return res, err
 	}
 
 	if srcVolume != minioMetaMultipartBucket {

docs/distributed/README.md

@@ -10,13 +10,13 @@ Silo in distributed mode can help you setup a highly-available storage system wi
 
 Distributed Silo provides protection against multiple node/drive failures and [bit rot](https://github.com/pgsty/silo/blob/main/docs/erasure/README.md#what-is-bit-rot-protection) using [erasure code](https://silo.pgsty.com/operations/concepts/erasure-coding/). As the minimum drives required for distributed Silo is 2 (same as minimum drives required for erasure coding), erasure code automatically kicks in as you launch distributed Silo.
 
-If one or more drives are offline at the start of a PutObject or NewMultipartUpload operation the object will have additional data protection bits added automatically to provide additional safety for these objects.
+With the default `availability` storage-class optimization, Silo can increase parity for new objects when drives are offline, up to half the drives in the erasure set. Writes must still meet the quorum for the resulting shard layout. This does not guarantee unchanged availability or keep a two-drive `EC:1` set writable after one drive fails. See the [storage-class reference](https://silo.pgsty.com/reference/minio-server/settings/storage-class/) for details.
 
 ### High availability
 
-A stand-alone Silo server would go down if the server hosting the drives goes offline. In contrast, a distributed Silo setup with _m_ servers and _n_ drives will have your data safe as long as _m/2_ servers or _m*n_/2 or more drives are online.
+A stand-alone Silo server becomes unavailable if its host goes offline. In a distributed deployment, determine node-failure tolerance from how many drives each failed node removes from every erasure set and from the parity recorded for the affected objects. Aggregate online node or drive counts alone do not establish read or write quorum.
 
-For example, an 16-server distributed setup with 200 drives per node would continue serving files, up to 4 servers can be offline in default configuration i.e around 800 drives down Silo would continue to read and write objects.
+For example, if a 16-node pool has 16-drive erasure sets with one drive per node in each set, a fixed `EC:4` layout has a read and write quorum of 12. Losing four nodes leaves 12 drives in every set. This result depends on that drive placement and parity; it cannot be generalized to arbitrary failures of the same total number of drives. In a two-node, two-drive `EC:1` set, losing one node leaves only one drive, so existing intact objects can remain readable but writes cannot continue.
 
 Refer to sizing guide for more understanding on default values chosen depending on your erasure stripe size [here](https://github.com/pgsty/silo/blob/main/docs/distributed/SIZING.md). Parity settings can be changed using [storage classes](https://github.com/pgsty/silo/tree/main/docs/erasure/storage-class).
 

docs/erasure/README.md

@@ -1,18 +1,18 @@
 # Silo Erasure Code Quickstart Guide
 
-Silo protects data against hardware failures and silent data corruption using erasure code and checksums. With the highest level of redundancy, you may lose up to half (N/2) of the total drives and still be able to recover the data.
+Silo protects data against hardware failures and silent data corruption using erasure coding and checksums. Recovery depends on the healthy shards and metadata remaining in each object's erasure set. With maximum parity, an object can tolerate the loss of up to `floor(N/2)` shards in its `N`-drive set; the deployment's total online drive count is not sufficient to determine recoverability.
 
 ## What is Erasure Code?
 
-Erasure code is a mathematical algorithm to reconstruct missing or corrupted data. Silo uses Reed-Solomon code to shard objects into variable data and parity blocks. For example, in a 12 drive setup, an object can be sharded to a variable number of data and parity blocks across all the drives - ranging from six data and six parity blocks to ten data and two parity blocks.
+Erasure coding uses mathematical algorithms to reconstruct missing or corrupted data. Silo uses Reed-Solomon codes to split objects into data and parity shards. For a 12-drive erasure set, parity can range from `EC:0` (12 data shards and no parity) to `EC:6` (six data shards and six parity shards).
 
-By default, Silo shards the objects across N/2 data and N/2 parity drives. Though, you can use [storage classes](https://github.com/pgsty/silo/tree/main/docs/erasure/storage-class) to use a custom configuration. We recommend N/2 data and parity blocks, as it ensures the best protection from drive failures.
+The default parity depends on the erasure set size: `EC:0` for 1 drive, `EC:1` for 2–3 drives, `EC:2` for 4–5 drives, `EC:3` for 6–7 drives, and `EC:4` for 8–16 drives. See the [Silo storage-class reference](https://silo.pgsty.com/reference/minio-server/settings/storage-class/) for configuration details.
 
-In 12 drive example above, with Silo server running in the default configuration, you can lose any of the six drives and still reconstruct the data reliably from the remaining drives.
+In the 12-drive example, the default `EC:4` layout uses eight data shards and four parity shards. An existing object can tolerate four unavailable shards if the remaining shards and metadata are intact. Explicit `EC:6` instead uses six data and six parity shards, with a read quorum of six and a write quorum of seven.
 
 ## Why is Erasure Code useful?
 
-Erasure code protects data from multiple drives failure, unlike RAID or replication. For example, RAID6 can protect against two drive failure whereas in Silo erasure code you can lose as many as half of drives and still the data remains safe. Further, Silo's erasure code is at the object level and can heal one object at a time. For RAID, healing can be done only at the volume level which translates into high downtime. As Silo encodes each object individually, it can heal objects incrementally. Storage servers once deployed should not require drive replacement or healing for the lifetime of the server. Silo's erasure coded backend is designed for operational efficiency and takes full advantage of hardware acceleration whenever available.
+Erasure coding protects objects against drive failures within their erasure set, up to the parity recorded for each object. Silo encodes objects individually and can heal them incrementally when enough healthy shards and consistent metadata remain. Failed drives still need repair or replacement to restore redundancy; erasure coding does not eliminate that maintenance.
 
 ![Erasure](https://github.com/pgsty/silo/blob/main/docs/screenshots/erasure-code.jpg?raw=true)
 
@@ -64,4 +64,4 @@ podman run \
 
 ### 3. Test your setup
 
-You may unplug drives randomly and continue to perform I/O on the system.
+In a test deployment, take drives offline and verify reads and writes against the quorum required by each erasure set and object layout. Restore the drives after each test; continued I/O depends on the remaining healthy shards and metadata.

docs/erasure/storage-class/README.md

@@ -38,31 +38,23 @@ You can calculate _approximate_ storage usage ratio using the formula - total dr
 
 ### Allowed values for STANDARD storage class
 
-`STANDARD` storage class implies more parity than `REDUCED_REDUNDANCY` class. So, `STANDARD` parity drives should be
+`STANDARD` supports `EC:0` without erasure-code redundancy and nonzero parity values up to `floor(N/2)`, where `N` is the number of drives in the erasure set. When both `STANDARD` and `REDUCED_REDUNDANCY` parity are nonzero, `STANDARD` must be greater than or equal to `REDUCED_REDUNDANCY`; equal parity is allowed.
 
-- Greater than or equal to 2, if `REDUCED_REDUNDANCY` parity is not set.
-- Greater than `REDUCED_REDUNDANCY` parity, if it is set.
+The default `STANDARD` parity is:
 
-Parity blocks can not be higher than data blocks, so `STANDARD` storage class parity can not be higher than N/2. (N being total number of drives)
-
-The default value for the `STANDARD` storage class depends on the number of volumes in the erasure set:
-
-| Erasure Set Size | Default Parity (EC:N) |
-|------------------|-----------------------|
-| 5 or fewer       |                 EC:2  |
-| 6-7              |                 EC:3  |
-| 8 or more        |                 EC:4  |
+| Erasure Set Size | Default Parity (EC:M) |
+| --- | --- |
+| 1 | EC:0 |
+| 2–3 | EC:1 |
+| 4–5 | EC:2 |
+| 6–7 | EC:3 |
+| 8–16 | EC:4 |
 
 For more complete documentation on Erasure Set sizing, see the [Silo Documentation on Erasure Sets](https://silo.pgsty.com/operations/concepts/erasure-coding/#minio-ec-erasure-set).
 
 ### Allowed values for REDUCED_REDUNDANCY storage class
 
-`REDUCED_REDUNDANCY` implies lesser parity than `STANDARD` class. So,`REDUCED_REDUNDANCY` parity drives should be
-
-- Less than N/2, if `STANDARD` parity is not set.
-- Less than `STANDARD` Parity, if it is set.
-
-Default value for `REDUCED_REDUNDANCY` storage class is `1`.
+`REDUCED_REDUNDANCY` parity can be zero or up to `floor(N/2)`. When both storage classes have nonzero parity, its parity must be less than or equal to `STANDARD` parity. The default is `EC:1` for multi-drive sets and `EC:0` for single-drive deployments. See the [Silo storage-class reference](https://silo.pgsty.com/reference/minio-server/settings/storage-class/) for the maintained configuration documentation.
 
 ## Get started with Storage Class
 

go.mod

@@ -1,10 +1,10 @@
 module github.com/minio/minio
 
-go 1.27.1
+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 v0.0.0-20260916075814-1360e26d976d
+replace github.com/minio/console => github.com/pgsty/silo-console v0.0.0-20261015132217-e5b165e4c0ee
 
 replace github.com/minio/mc => github.com/pgsty/mc v0.0.0-20260916070421-e952aa78f10a
 
@@ -294,7 +294,7 @@ require (
 	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/net v0.59.0 // indirect
+	golang.org/x/net v0.60.0 // indirect
 	golang.org/x/text v0.42.0 // indirect
 	golang.org/x/tools v0.50.0 // indirect
 	google.golang.org/genproto v0.0.0-20260319201613-d00831a3d3e7 // indirect

go.sum

@@ -547,8 +547,8 @@ github.com/pborman/getopt v0.0.0-20170112200414-7148bc3a4c30/go.mod h1:85jBQOZwp
 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 v0.0.0-20260916070421-e952aa78f10a/go.mod h1:6z9lX8kOvaWz2vLlP7BjjF3JWHnIH86XDA2HwcmpZH8=
-github.com/pgsty/silo-console v0.0.0-20260916075814-1360e26d976d h1:A9ou1GkUckGZd3BOgEXeZ393nbYST967SKcI4Lgr/Vs=
-github.com/pgsty/silo-console v0.0.0-20260916075814-1360e26d976d/go.mod h1:JS165CF2yoTBIx1X8k1IQ6vunJcZFG3pRZlWHJ8UT8o=
+github.com/pgsty/silo-console v0.0.0-20261015132217-e5b165e4c0ee h1:e51AnrGe0Zq+nHTYaGfv/GkehQMO9Mekuw/ew6zYwOM=
+github.com/pgsty/silo-console v0.0.0-20261015132217-e5b165e4c0ee/go.mod h1:jpt39bx4wjhXoFB+BTOa/bx/pmU3pXzp2cT7PnT21Lo=
 github.com/pgsty/silo-pkg/v3 v3.14.1 h1:8MONT3Hky9EOnsuPk29590pEBE6MILaOnJtaejPPi7M=
 github.com/pgsty/silo-pkg/v3 v3.14.1/go.mod h1:jxKWxi52jSsDaDhXXPqjOyPDYdxx2iqdBXyLfM/xSIw=
 github.com/philhofer/fwd v1.2.0 h1:e6DnBTl7vGY+Gz322/ASL4Gyp1FspeMvx1RNDoToZuM=
@@ -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 v0.59.0 h1:5zfYln+w5XCxwrnMMJPufRgNoXEaGxl0wo5GqPXyues=
-golang.org/x/net v0.59.0/go.mod h1:2DA/G1UfVbCpQPeWTmMPGY7Cs2PkBkwu743bVX5PIVg=
+golang.org/x/net v0.60.0 h1:79p50tfZlm0J9YfoDsSi639qSXNGVwEzOPLCxM2FsYU=
+golang.org/x/net v0.60.0/go.mod h1:2DA/G1UfVbCpQPeWTmMPGY7Cs2PkBkwu743bVX5PIVg=
 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=

internal/config/lambda/event/targetidset.go

@@ -24,7 +24,7 @@ type TargetIDSet map[TargetID]struct{}
 
 // IsEmpty returns true if the set is empty.
 func (set TargetIDSet) IsEmpty() bool {
-	return len(set) != 0
+	return len(set) == 0
 }
 
 // Clone - returns copy of this set.

internal/config/lambda/event/targetidset_test.go

@@ -22,6 +22,22 @@ import (
 	"testing"
 )
 
+func TestTargetIDSetIsEmpty(t *testing.T) {
+	testCases := []struct {
+		set      TargetIDSet
+		expected bool
+	}{
+		{NewTargetIDSet(), true},
+		{NewTargetIDSet(TargetID{"1", "webhook"}), false},
+	}
+
+	for i, testCase := range testCases {
+		if got := testCase.set.IsEmpty(); got != testCase.expected {
+			t.Fatalf("test %v: expected: %v, got: %v", i+1, testCase.expected, got)
+		}
+	}
+}
+
 func TestTargetIDSetClone(t *testing.T) {
 	testCases := []struct {
 		set           TargetIDSet

internal/hash/checksum-errors_test.go

@@ -0,0 +1,73 @@
+// Copyright (c) 2026 PGSTY
+//
+// This file is part of SILO Object Storage.
+// SPDX-License-Identifier: AGPL-3.0-or-later
+//
+// 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 hash
+
+import (
+	"bytes"
+	"errors"
+	"io"
+	"net/http"
+	"net/http/httptest"
+	"testing"
+)
+
+func TestChecksumMismatchAlgorithm(t *testing.T) {
+	content := []byte("actual content")
+	for _, typ := range []ChecksumType{ChecksumCRC32, ChecksumCRC32C, ChecksumSHA1, ChecksumSHA256, ChecksumCRC64NVME} {
+		t.Run(typ.String(), func(t *testing.T) {
+			wrong := NewChecksumFromData(typ, []byte("wrong content"))
+			check := func(t *testing.T, err error) {
+				t.Helper()
+				var mismatch ChecksumMismatch
+				if !errors.As(err, &mismatch) || mismatch.Algorithm != typ.String() {
+					t.Fatalf("error=%+v, want %s checksum mismatch", err, typ)
+				}
+			}
+			check(t, wrong.Matches(content, 0))
+			parts := NewChecksumFromData(typ, content)
+			parts.WantParts = 2
+			check(t, parts.Matches(content, 1))
+			for _, mode := range []string{"header", "trailer", "malformed-trailer"} {
+				t.Run(mode, func(t *testing.T) {
+					reader, err := NewReader(t.Context(), bytes.NewReader(content), int64(len(content)), "", "", int64(len(content)))
+					if err != nil {
+						t.Fatal(err)
+					}
+					req := httptest.NewRequest(http.MethodPut, "http://localhost/bucket/object", nil)
+					if mode == "header" {
+						req.Header.Set(typ.Key(), wrong.Encoded)
+					} else {
+						req.Header.Set("x-amz-trailer", typ.Key())
+						req.Trailer = make(http.Header)
+						value := wrong.Encoded
+						if mode == "malformed-trailer" {
+							value = "!invalid-base64!"
+						}
+						req.Trailer.Set(typ.Key(), value)
+					}
+					if err = reader.AddChecksum(req, false); err != nil {
+						t.Fatal(err)
+					}
+					_, err = io.ReadAll(reader)
+					check(t, err)
+				})
+			}
+		})
+	}
+}

internal/hash/checksum.go

@@ -584,15 +584,17 @@ func (c Checksum) Matches(content []byte, parts int) error {
 	sum := hasher.Sum(nil)
 	if c.WantParts > 0 && c.WantParts != parts {
 		return ChecksumMismatch{
-			Want: fmt.Sprintf("%s-%d", c.Encoded, c.WantParts),
-			Got:  fmt.Sprintf("%s-%d", base64.StdEncoding.EncodeToString(sum), parts),
+			Algorithm: c.Type.String(),
+			Want:      fmt.Sprintf("%s-%d", c.Encoded, c.WantParts),
+			Got:       fmt.Sprintf("%s-%d", base64.StdEncoding.EncodeToString(sum), parts),
 		}
 	}
 
 	if !bytes.Equal(sum, c.Raw) {
 		return ChecksumMismatch{
-			Want: c.Encoded,
-			Got:  base64.StdEncoding.EncodeToString(sum),
+			Algorithm: c.Type.String(),
+			Want:      c.Encoded,
+			Got:       base64.StdEncoding.EncodeToString(sum),
 		}
 	}
 	return nil

internal/hash/errors.go

@@ -74,8 +74,11 @@ func (e SizeMismatch) Error() string {
 
 // ChecksumMismatch - when content checksum does not match with what was sent from client.
 type ChecksumMismatch struct {
-	Want string
-	Got  string
+	// Algorithm carries the requested checksum type (e.g. "CRC32") so API
+	// error rendering can name it, matching S3's wording.
+	Algorithm string
+	Want      string
+	Got       string
 }
 
 func (e ChecksumMismatch) Error() string {

internal/hash/reader.go

@@ -302,13 +302,14 @@ func (r *Reader) Read(p []byte) (int, error) {
 				r.contentHash.Encoded = r.trailer.Get(r.contentHash.Type.Key())
 				r.contentHash.Raw, err = base64.StdEncoding.DecodeString(r.contentHash.Encoded)
 				if err != nil || len(r.contentHash.Raw) == 0 {
-					return 0, ChecksumMismatch{Got: r.contentHash.Encoded}
+					return 0, ChecksumMismatch{Algorithm: r.contentHash.Type.String(), Got: r.contentHash.Encoded}
 				}
 			}
 			if sum := r.contentHasher.Sum(nil); !bytes.Equal(r.contentHash.Raw, sum) {
 				err := ChecksumMismatch{
-					Want: r.contentHash.Encoded,
-					Got:  base64.StdEncoding.EncodeToString(sum),
+					Algorithm: r.contentHash.Type.String(),
+					Want:      r.contentHash.Encoded,
+					Got:       base64.StdEncoding.EncodeToString(sum),
 				}
 				return n, err
 			}