filer: evaluate a write precondition in CreateEntry

Add an optional WriteCondition to CreateEntryRequest. When set, the filer evaluates it against the current entry while holding the per-path lock, so the check and the write are atomic on this filer, and returns PRECONDITION_FAILED when it does not hold. The condition is a single primitive (exists / not-exists / etag match / etag mismatch / [un]modified-since). The caller reduces request semantics such as RFC 7232 precedence to one primitive; the filer just compares. ETag comparison mirrors the S3 gateway's precedence (stored Seaweed ETag attribute, then the Md5/chunk fallback) so results match without coupling the filer to S3 handling.
2026-10-28 16:00:37 +01:00 · 2026-05-22 22:30:26 -07:00
1 parent 56c51e7f50
commit 4e421400e4
@@ -7,6 +7,7 @@ import (
 	"fmt"
 	"os"
 	"path/filepath"
+	"strings"
 	"time"

 	"github.com/seaweedfs/seaweedfs/weed/filer"
@@ -14,6 +15,7 @@ import (
 	"github.com/seaweedfs/seaweedfs/weed/operation"
 	"github.com/seaweedfs/seaweedfs/weed/pb/filer_pb"
 	"github.com/seaweedfs/seaweedfs/weed/pb/master_pb"
+	"github.com/seaweedfs/seaweedfs/weed/s3api/s3_constants"
 	"github.com/seaweedfs/seaweedfs/weed/storage/needle"
 	"github.com/seaweedfs/seaweedfs/weed/util"
 	"github.com/seaweedfs/seaweedfs/weed/wdclient"
@@ -191,6 +193,25 @@ func (fs *FilerServer) CreateEntry(ctx context.Context, req *filer_pb.CreateEntr
 	pathLock := fs.entryLockTable.AcquireLock("CreateEntry", fullpath, util.ExclusiveLock)
 	defer fs.entryLockTable.ReleaseLock(fullpath, pathLock)

+	// Evaluate the optional precondition against the current entry while the
+	// path lock is held, so the check and the write are atomic on this filer.
+	if req.Condition != nil && req.Condition.Kind != filer_pb.WriteCondition_NONE {
+		current, findErr := fs.filer.FindEntry(ctx, fullpath)
+		if findErr != nil && findErr != filer_pb.ErrNotFound {
+			return &filer_pb.CreateEntryResponse{}, fmt.Errorf("CreateEntry condition check %s: %w", fullpath, findErr)
+		}
+		if findErr == filer_pb.ErrNotFound {
+			current = nil
+		}
+		if !writeConditionSatisfied(req.Condition, current) {
+			glog.V(3).InfofCtx(ctx, "CreateEntry %s: precondition %v failed", fullpath, req.Condition.Kind)
+			return &filer_pb.CreateEntryResponse{
+				Error:     "precondition failed",
+				ErrorCode: filer_pb.FilerError_PRECONDITION_FAILED,
+			}, nil
+		}
+	}
+
 	ctx, eventSink := filer.WithMetadataEventSink(ctx)
 	createErr := fs.filer.CreateEntry(ctx, newEntry, req.OExcl, req.IsFromOtherCluster, req.Signatures, req.SkipCheckParentDirectory, so.MaxFileNameLength)

@@ -217,6 +238,44 @@ func (fs *FilerServer) CreateEntry(ctx context.Context, req *filer_pb.CreateEntr
 	return
 }

+// writeConditionSatisfied reports whether the precondition holds against the
+// current entry (nil if absent). The caller reduces request semantics to one
+// primitive; each kind is a single comparison evaluated under the path lock.
+func writeConditionSatisfied(cond *filer_pb.WriteCondition, current *filer.Entry) bool {
+	exists := current != nil
+	switch cond.Kind {
+	case filer_pb.WriteCondition_IF_NOT_EXISTS:
+		return !exists
+	case filer_pb.WriteCondition_IF_EXISTS:
+		return exists
+	case filer_pb.WriteCondition_IF_ETAG_MATCH:
+		return exists && storedEntryETag(current) == normalizeETag(cond.Etag)
+	case filer_pb.WriteCondition_IF_ETAG_NOT_MATCH:
+		return !exists || storedEntryETag(current) != normalizeETag(cond.Etag)
+	case filer_pb.WriteCondition_IF_UNMODIFIED_SINCE:
+		return !exists || current.Attr.Mtime.Unix() <= cond.UnixTime
+	case filer_pb.WriteCondition_IF_MODIFIED_SINCE:
+		return !exists || current.Attr.Mtime.Unix() > cond.UnixTime
+	default:
+		return true
+	}
+}
+
+// storedEntryETag mirrors the S3 gateway's ETag precedence (the stored
+// Seaweed ETag extended attribute, then the chunk/Md5 fallback) so conditional
+// comparisons match what the gateway computes, without coupling the filer to
+// S3 request handling.
+func storedEntryETag(entry *filer.Entry) string {
+	if v, ok := entry.Extended[s3_constants.ExtETagKey]; ok && len(v) > 0 {
+		return normalizeETag(string(v))
+	}
+	return normalizeETag(filer.ETagEntry(entry))
+}
+
+func normalizeETag(etag string) string {
+	return strings.Trim(etag, `"`)
+}
+
 func (fs *FilerServer) UpdateEntry(ctx context.Context, req *filer_pb.UpdateEntryRequest) (*filer_pb.UpdateEntryResponse, error) {

 	glog.V(4).InfofCtx(ctx, "UpdateEntry %v", req)
@@ -0,0 +1,108 @@
+package weed_server
+
+import (
+	"context"
+	"testing"
+	"time"
+
+	"github.com/seaweedfs/seaweedfs/weed/filer"
+	"github.com/seaweedfs/seaweedfs/weed/pb/filer_pb"
+	"github.com/seaweedfs/seaweedfs/weed/s3api/s3_constants"
+	"github.com/seaweedfs/seaweedfs/weed/util"
+)
+
+func entryWithETag(etag string, mtime time.Time) *filer.Entry {
+	return &filer.Entry{
+		FullPath: "/test/obj",
+		Attr:     filer.Attr{Mtime: mtime},
+		Extended: map[string][]byte{s3_constants.ExtETagKey: []byte(etag)},
+	}
+}
+
+func TestWriteConditionSatisfied(t *testing.T) {
+	base := time.Unix(1700000000, 0)
+	present := entryWithETag("abc", base)
+
+	cases := []struct {
+		name string
+		cond *filer_pb.WriteCondition
+		cur  *filer.Entry
+		want bool
+	}{
+		{"none-absent", &filer_pb.WriteCondition{Kind: filer_pb.WriteCondition_NONE}, nil, true},
+		{"ifnotexists-absent", &filer_pb.WriteCondition{Kind: filer_pb.WriteCondition_IF_NOT_EXISTS}, nil, true},
+		{"ifnotexists-present", &filer_pb.WriteCondition{Kind: filer_pb.WriteCondition_IF_NOT_EXISTS}, present, false},
+		{"ifexists-absent", &filer_pb.WriteCondition{Kind: filer_pb.WriteCondition_IF_EXISTS}, nil, false},
+		{"ifexists-present", &filer_pb.WriteCondition{Kind: filer_pb.WriteCondition_IF_EXISTS}, present, true},
+		{"etagmatch-hit", &filer_pb.WriteCondition{Kind: filer_pb.WriteCondition_IF_ETAG_MATCH, Etag: `"abc"`}, present, true},
+		{"etagmatch-miss", &filer_pb.WriteCondition{Kind: filer_pb.WriteCondition_IF_ETAG_MATCH, Etag: `"zzz"`}, present, false},
+		{"etagmatch-absent", &filer_pb.WriteCondition{Kind: filer_pb.WriteCondition_IF_ETAG_MATCH, Etag: `"abc"`}, nil, false},
+		{"etagnotmatch-hit", &filer_pb.WriteCondition{Kind: filer_pb.WriteCondition_IF_ETAG_NOT_MATCH, Etag: `"abc"`}, present, false},
+		{"etagnotmatch-miss", &filer_pb.WriteCondition{Kind: filer_pb.WriteCondition_IF_ETAG_NOT_MATCH, Etag: `"zzz"`}, present, true},
+		{"etagnotmatch-absent", &filer_pb.WriteCondition{Kind: filer_pb.WriteCondition_IF_ETAG_NOT_MATCH, Etag: `"abc"`}, nil, true},
+		{"unmodsince-ok", &filer_pb.WriteCondition{Kind: filer_pb.WriteCondition_IF_UNMODIFIED_SINCE, UnixTime: base.Unix()}, present, true},
+		{"unmodsince-fail", &filer_pb.WriteCondition{Kind: filer_pb.WriteCondition_IF_UNMODIFIED_SINCE, UnixTime: base.Unix() - 1}, present, false},
+		{"modsince-ok", &filer_pb.WriteCondition{Kind: filer_pb.WriteCondition_IF_MODIFIED_SINCE, UnixTime: base.Unix() - 1}, present, true},
+		{"modsince-fail", &filer_pb.WriteCondition{Kind: filer_pb.WriteCondition_IF_MODIFIED_SINCE, UnixTime: base.Unix()}, present, false},
+	}
+	for _, tc := range cases {
+		if got := writeConditionSatisfied(tc.cond, tc.cur); got != tc.want {
+			t.Errorf("%s: got %v want %v", tc.name, got, tc.want)
+		}
+	}
+}
+
+// storedEntryETag prefers the stored Seaweed ETag attribute and falls back to
+// the Md5-derived ETag, matching the S3 gateway.
+func TestStoredEntryETag(t *testing.T) {
+	withExt := entryWithETag("explicit", time.Unix(0, 0))
+	if got := storedEntryETag(withExt); got != "explicit" {
+		t.Errorf("extended etag: got %q", got)
+	}
+	md5Only := &filer.Entry{Attr: filer.Attr{Md5: []byte{0xab, 0xcd}}}
+	if got := storedEntryETag(md5Only); got != "abcd" {
+		t.Errorf("md5 fallback: got %q", got)
+	}
+}
+
+// The CreateEntry handler enforces the precondition atomically: a matching
+// If-Match overwrites, a non-matching one returns PRECONDITION_FAILED.
+func TestCreateEntryConditionEnforced(t *testing.T) {
+	store := newRenameTestStore()
+	store.entries["/test/obj"] = &filer.Entry{
+		FullPath: "/test/obj",
+		Attr:     filer.Attr{Inode: 1, Mtime: time.Unix(1700000000, 0)},
+		Extended: map[string][]byte{s3_constants.ExtETagKey: []byte("abc")},
+	}
+	f := newRenameTestFiler(store)
+	f.DirBucketsPath = "/buckets"
+	fs := &FilerServer{filer: f, option: &FilerOption{}, entryLockTable: util.NewLockTable[util.FullPath]()}
+
+	req := func(etag string) *filer_pb.CreateEntryRequest {
+		return &filer_pb.CreateEntryRequest{
+			Directory:                "/test",
+			SkipCheckParentDirectory: true,
+			Entry: &filer_pb.Entry{
+				Name:       "obj",
+				Attributes: &filer_pb.FuseAttributes{Mtime: 1700000001, FileMode: 0644, Inode: 2},
+			},
+			Condition: &filer_pb.WriteCondition{Kind: filer_pb.WriteCondition_IF_ETAG_MATCH, Etag: etag},
+		}
+	}
+
+	resp, err := fs.CreateEntry(context.Background(), req(`"zzz"`))
+	if err != nil {
+		t.Fatalf("unexpected err: %v", err)
+	}
+	if resp.ErrorCode != filer_pb.FilerError_PRECONDITION_FAILED {
+		t.Fatalf("mismatched etag: want PRECONDITION_FAILED, got %v (%q)", resp.ErrorCode, resp.Error)
+	}
+
+	resp, err = fs.CreateEntry(context.Background(), req(`"abc"`))
+	if err != nil {
+		t.Fatalf("unexpected err: %v", err)
+	}
+	if resp.Error != "" {
+		t.Fatalf("matching etag should overwrite, got error %q", resp.Error)
+	}
+}