snapshot of https://github.com/seaweedfs/seaweedfs.git (master, 2026-09-16; blobs > 131072B omitted)

weed/storage/idx/check.go

@@ -0,0 +1,114 @@
+package idx
+
+import (
+	"fmt"
+	"io"
+	"sort"
+
+	"github.com/seaweedfs/seaweedfs/weed/storage/needle"
+	"github.com/seaweedfs/seaweedfs/weed/storage/types"
+)
+
+type indexEntry struct {
+	index  int
+	id     types.NeedleId
+	offset int64
+	size   types.Size
+}
+
+func (ie *indexEntry) Compare(other *indexEntry) int {
+	if ie.offset < other.offset {
+		return -1
+	}
+	if ie.offset > other.offset {
+		return 1
+	}
+	if ie.size < other.size {
+		return -1
+	}
+	if ie.size > other.size {
+		return 1
+	}
+	return 0
+}
+
+// CheckIndexFile verifies the integrity of a IDX/ECX index file. Returns a count of processed file entries, and slice of found errors.
+func CheckIndexFile(r io.ReaderAt, indexFileSize int64, version needle.Version) (int64, []error) {
+	errs := []error{}
+
+	entries := []*indexEntry{}
+	var i int
+	err := WalkIndexFile(r, 0, func(id types.NeedleId, offset types.Offset, size types.Size) error {
+		entries = append(entries, &indexEntry{
+			index:  i,
+			id:     id,
+			offset: offset.ToActualOffset(),
+			size:   size,
+		})
+		i++
+		return nil
+	})
+	if err != nil {
+		errs = append(errs, err)
+	}
+
+	sort.Slice(entries, func(i, j int) bool {
+		return entries[i].Compare(entries[j]) < 0
+	})
+
+	// Offset-0 logical tombstones (remote-tier deletes) occupy no physical .dat
+	// extent, so they cannot overlap anything — exclude them from the overlap
+	// check. They are still counted below for the index-size check.
+	physical := make([]*indexEntry, 0, len(entries))
+	for _, e := range entries {
+		if e.offset == 0 && e.size.IsDeleted() {
+			continue
+		}
+		physical = append(physical, e)
+	}
+
+	for i, e := range physical {
+		if i == 0 {
+			// nothing to check for the first entry
+			continue
+		}
+
+		start, end := e.offset, e.offset
+		if size := needle.GetActualSize(e.size, version); size != 0 {
+			end += size - 1
+		}
+
+		last := physical[i-1]
+		lastStart, lastEnd := last.offset, last.offset
+		if lastSize := needle.GetActualSize(last.size, version); lastSize != 0 {
+			lastEnd += lastSize - 1
+		}
+
+		// check if needles overlap
+		if start <= lastEnd {
+			errs = append(errs, fmt.Errorf(
+				"needle %d (#%d) at [%d-%d] overlaps needle %d at [%d-%d]",
+				e.id, e.index+1,
+				start, end,
+				last.id,
+				lastStart, lastEnd))
+		}
+
+		// The check below is intended to ensure all index entries are contiguous; unfortunately, Seaweed
+		// can delete index entries for files while keeping their data, so volumes with deleted files
+		// will fail this test :(
+		// See https://github.com/seaweedfs/seaweedfs/issues/8204 for details.
+		/*
+			if e.offset != lastEnd + 1 {
+				errs = append(errs, fmt.Errorf("offset %d for needle %d (#%d) doesn't match end of needle %d at %d", e.offset, e.id, e.index+1, last.id, lastEnd))
+			}
+		*/
+	}
+
+	count := int64(len(entries))
+	if got, want := count*types.NeedleMapEntrySize, indexFileSize; got != want {
+		errs = append(errs, fmt.Errorf("expected an index file of size %d, got %d", want, got))
+	}
+
+	return count, errs
+}

weed/storage/idx/check_test.go

@@ -0,0 +1,128 @@
+package idx
+
+import (
+	"bytes"
+	"fmt"
+	"os"
+	"reflect"
+	"testing"
+
+	"github.com/seaweedfs/seaweedfs/weed/storage/needle"
+	"github.com/seaweedfs/seaweedfs/weed/storage/types"
+)
+
+func idxEntryBytes(id types.NeedleId, offset types.Offset, size types.Size) []byte {
+	b := make([]byte, types.NeedleIdSize+types.OffsetSize+types.SizeSize)
+	types.NeedleIdToBytes(b[0:types.NeedleIdSize], id)
+	types.OffsetToBytes(b[types.NeedleIdSize:types.NeedleIdSize+types.OffsetSize], offset)
+	types.SizeToBytes(b[types.NeedleIdSize+types.OffsetSize:], size)
+	return b
+}
+
+// An offset-0 logical tombstone (remote-tier delete) occupies no physical .dat
+// extent, so it must not be flagged as overlapping a real needle. The tombstone
+// row is still counted for the index-size check.
+func TestCheckIndexFile_IgnoresOffset0Tombstone(t *testing.T) {
+	var buf []byte
+	buf = append(buf, idxEntryBytes(types.NeedleId(1), types.ToOffset(8), types.Size(100))...)
+	buf = append(buf, idxEntryBytes(types.NeedleId(2), types.ToOffset(0), types.TombstoneFileSize)...)
+
+	count, errs := CheckIndexFile(bytes.NewReader(buf), int64(len(buf)), needle.Version3)
+	if count != 2 {
+		t.Errorf("tombstone row must still count: got %d", count)
+	}
+	if len(errs) != 0 {
+		t.Errorf("offset-0 tombstone must not overlap: %v", errs)
+	}
+}
+
+func TestCheckIndexFile(t *testing.T) {
+	testCases := []struct {
+		name      string
+		indexPath string
+		version   needle.Version
+		want      int64
+		wantErrs  []error
+	}{
+		{
+			name:      "healthy index",
+			indexPath: "./test_files/simple_index.idx",
+			version:   needle.Version3,
+			want:      161,
+			wantErrs:  []error{},
+		},
+		{
+			name:      "healthy index with deleted files",
+			indexPath: "./test_files/deleted_files.idx",
+			version:   needle.Version3,
+			want:      230,
+			wantErrs:  []error{},
+		},
+		{
+			name:      "damaged index (bitrot)",
+			indexPath: "./test_files/simple_index_bitrot.idx",
+			version:   needle.Version3,
+			want:      161,
+			wantErrs: []error{
+				fmt.Errorf("needle 3544668469065756977 (#2) at [6602459528-7427766999] overlaps needle 49 at [6602459528-7427766999]"),
+				fmt.Errorf("expected an index file of size 2577, got 2576"),
+			},
+		},
+		{
+			name:      "damaged index (truncated)",
+			indexPath: "./test_files/simple_index_truncated.idx",
+			version:   needle.Version3,
+			want:      158,
+			wantErrs: []error{
+				fmt.Errorf("expected an index file of size 2540, got 2528"),
+			},
+		},
+		{
+			name:      "healthy EC index with deleted files",
+			indexPath: "./test_files/deleted_files.ecx",
+			version:   needle.Version3,
+			want:      116,
+			wantErrs:  []error{},
+		},
+		{
+			name:      "damaged EC index (bitrot)",
+			indexPath: "./test_files/deleted_files_bitrot.ecx",
+			version:   needle.Version3,
+			want:      116,
+			wantErrs: []error{
+				fmt.Errorf("needle 3223857 (#110) at [6602459528-7427767055] overlaps needle 12593 at [6601933184-7407907279]"),
+				fmt.Errorf("needle 3544668469065757234 (#43) at [6737203600-7579354079] overlaps needle 3223857 at [6602459528-7427767055]"),
+				fmt.Errorf("needle 3421236 (#112) at [7006693800-7899362591] overlaps needle 3544668469065757234 at [6737203600-7579354079]"),
+				fmt.Errorf("needle 310 (#113) at [7276179888-8185702583] overlaps needle 3421236 at [7006693800-7899362591]"),
+				fmt.Errorf("needle 7089336938131513954 (#52) at [13204919056-13205053935] overlaps needle 27410143614427489 at [13070174984-14703946887]"),
+				fmt.Errorf("needle 25186 (#50) at [13204919056-14855533967] overlaps needle 7089336938131513954 at [13204919056-13205053935]"),
+				fmt.Errorf("needle 7089336938131513954 (#51) at [13204919056-14855533967] overlaps needle 25186 at [13204919056-14855533967]"),
+				fmt.Errorf("expected an index file of size 1857, got 1856"),
+			},
+		},
+	}
+
+	for _, tc := range testCases {
+		t.Run(tc.name, func(t *testing.T) {
+			idx, err := os.OpenFile(tc.indexPath, os.O_RDONLY, 0)
+			if err != nil {
+				t.Fatalf("failed to open index file: %v", err)
+			}
+			defer idx.Close()
+
+			idxStat, err := idx.Stat()
+			if err != nil {
+				t.Fatalf("failed to stat index file: %v", err)
+			}
+
+			got, gotErrs := CheckIndexFile(idx, idxStat.Size(), tc.version)
+
+			if got != tc.want {
+				t.Errorf("expected %d files processed, got %d", tc.want, got)
+			}
+			if !reflect.DeepEqual(gotErrs, tc.wantErrs) {
+				t.Errorf("expected errors %v, got %v", tc.wantErrs, gotErrs)
+			}
+		})
+	}
+}

weed/storage/idx/walk.go

@@ -0,0 +1,54 @@
+package idx
+
+import (
+	"io"
+
+	"github.com/seaweedfs/seaweedfs/weed/glog"
+	"github.com/seaweedfs/seaweedfs/weed/storage/types"
+)
+
+// walks through the index file, calls fn function with each key, offset, size
+// stops with the error returned by the fn function
+func WalkIndexFile(r io.ReaderAt, startFrom uint64, fn func(key types.NeedleId, offset types.Offset, size types.Size) error) error {
+	readerOffset := int64(startFrom * types.NeedleMapEntrySize)
+	bytes := make([]byte, types.NeedleMapEntrySize*RowsToRead)
+	count, e := r.ReadAt(bytes, readerOffset)
+	if count == 0 && e == io.EOF {
+		return nil
+	}
+	glog.V(3).Infof("readerOffset %d count %d err: %v", readerOffset, count, e)
+	readerOffset += int64(count)
+	var (
+		key    types.NeedleId
+		offset types.Offset
+		size   types.Size
+		i      int
+	)
+
+	for count > 0 && e == nil || e == io.EOF {
+		for i = 0; i+types.NeedleMapEntrySize <= count; i += types.NeedleMapEntrySize {
+			key, offset, size = IdxFileEntry(bytes[i : i+types.NeedleMapEntrySize])
+			if e = fn(key, offset, size); e != nil {
+				return e
+			}
+		}
+		if e == io.EOF {
+			return nil
+		}
+		count, e = r.ReadAt(bytes, readerOffset)
+		glog.V(3).Infof("readerOffset %d count %d err: %v", readerOffset, count, e)
+		readerOffset += int64(count)
+	}
+	return e
+}
+
+func IdxFileEntry(bytes []byte) (key types.NeedleId, offset types.Offset, size types.Size) {
+	key = types.BytesToNeedleId(bytes[:types.NeedleIdSize])
+	offset = types.BytesToOffset(bytes[types.NeedleIdSize : types.NeedleIdSize+types.OffsetSize])
+	size = types.BytesToSize(bytes[types.NeedleIdSize+types.OffsetSize : types.NeedleIdSize+types.OffsetSize+types.SizeSize])
+	return
+}
+
+const (
+	RowsToRead = 1024
+)