-
Notifications
You must be signed in to change notification settings - Fork 363
/
storage_dest.go
1261 lines (1166 loc) · 55.5 KB
/
storage_dest.go
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848
849
850
851
852
853
854
855
856
857
858
859
860
861
862
863
864
865
866
867
868
869
870
871
872
873
874
875
876
877
878
879
880
881
882
883
884
885
886
887
888
889
890
891
892
893
894
895
896
897
898
899
900
901
902
903
904
905
906
907
908
909
910
911
912
913
914
915
916
917
918
919
920
921
922
923
924
925
926
927
928
929
930
931
932
933
934
935
936
937
938
939
940
941
942
943
944
945
946
947
948
949
950
951
952
953
954
955
956
957
958
959
960
961
962
963
964
965
966
967
968
969
970
971
972
973
974
975
976
977
978
979
980
981
982
983
984
985
986
987
988
989
990
991
992
993
994
995
996
997
998
999
1000
//go:build !containers_image_storage_stub
// +build !containers_image_storage_stub
package storage
import (
"bytes"
"context"
"encoding/json"
"errors"
"fmt"
"io"
"os"
"path/filepath"
"sync"
"sync/atomic"
"github.com/containers/image/v5/docker/reference"
"github.com/containers/image/v5/internal/imagedestination/impl"
"github.com/containers/image/v5/internal/imagedestination/stubs"
"github.com/containers/image/v5/internal/private"
"github.com/containers/image/v5/internal/putblobdigest"
"github.com/containers/image/v5/internal/set"
"github.com/containers/image/v5/internal/signature"
"github.com/containers/image/v5/internal/tmpdir"
"github.com/containers/image/v5/manifest"
"github.com/containers/image/v5/pkg/blobinfocache/none"
"github.com/containers/image/v5/types"
"github.com/containers/storage"
graphdriver "github.com/containers/storage/drivers"
"github.com/containers/storage/pkg/archive"
"github.com/containers/storage/pkg/chunked"
"github.com/containers/storage/pkg/ioutils"
digest "github.com/opencontainers/go-digest"
imgspecv1 "github.com/opencontainers/image-spec/specs-go/v1"
"github.com/sirupsen/logrus"
"golang.org/x/exp/slices"
)
var (
// ErrBlobDigestMismatch could potentially be returned when PutBlob() is given a blob
// with a digest-based name that doesn't match its contents.
// Deprecated: PutBlob() doesn't do this any more (it just accepts the caller’s value),
// and there is no known user of this error.
ErrBlobDigestMismatch = errors.New("blob digest mismatch")
// ErrBlobSizeMismatch is returned when PutBlob() is given a blob
// with an expected size that doesn't match the reader.
ErrBlobSizeMismatch = errors.New("blob size mismatch")
)
type storageImageDestination struct {
impl.Compat
impl.PropertyMethodsInitialize
stubs.ImplementsPutBlobPartial
stubs.AlwaysSupportsSignatures
imageRef storageReference
directory string // Temporary directory where we store blobs until Commit() time
nextTempFileID atomic.Int32 // A counter that we use for computing filenames to assign to blobs
manifest []byte // Manifest contents, temporary
manifestDigest digest.Digest // Valid if len(manifest) != 0
untrustedDiffIDValues []digest.Digest // From config’s RootFS.DiffIDs, valid if not nil
signatures []byte // Signature contents, temporary
signatureses map[digest.Digest][]byte // Instance signature contents, temporary
metadata storageImageMetadata // Metadata contents being built
// Mapping from layer (by index) to the associated ID in the storage.
// It's protected *implicitly* since `commitLayer()`, at any given
// time, can only be executed by *one* goroutine. Please refer to
// `queueOrCommit()` for further details on how the single-caller
// guarantee is implemented.
indexToStorageID map[int]string
// A storage destination may be used concurrently, due to HasThreadSafePutBlob.
lock sync.Mutex // Protects lockProtected
lockProtected storageImageDestinationLockProtected
}
// storageImageDestinationLockProtected contains storageImageDestination data which might be
// accessed concurrently, due to HasThreadSafePutBlob.
// _During the concurrent TryReusingBlob/PutBlob/* calls_ (but not necessarily during the final Commit)
// uses must hold storageImageDestination.lock.
type storageImageDestinationLockProtected struct {
currentIndex int // The index of the layer to be committed (i.e., lower indices have already been committed)
indexToAddedLayerInfo map[int]addedLayerInfo // Mapping from layer (by index) to blob to add to the image
// In general, a layer is identified either by (compressed) digest, or by TOC digest.
// When creating a layer, the c/storage layer metadata and image IDs must _only_ be based on trusted values
// we have computed ourselves. (Layer reuse can then look up against such trusted values, but it might not
// recompute those values for incoming layers — the point of the reuse is that we don’t need to consume the incoming layer.)
//
// Layer identification: For a layer, at least one of (indexToDiffID, indexToTOCDigest, blobDiffIDs) must be available
// before commitLayer is called.
// Layer is identified by the first of the three fields which exists, in that order (and the value must be trusted).
//
// Ideally we wouldn’t have blobDiffIDs, and just keep records by index, but the public API does not require the caller to provide layer
// indices; and configs don’t have layer indices. blobDiffIDs needs to exist for those cases.
indexToDiffID map[int]digest.Digest // Mapping from layer index to DiffID; exclusive with indexToTOCDigest
indexToTOCDigest map[int]digest.Digest // Mapping from layer index to a TOC Digest, IFF the layer was created/found/reused by TOC digest; exclusive with indexToDiffID
blobDiffIDs map[digest.Digest]digest.Digest // Mapping from layer blobsums to their corresponding DiffIDs
// Layer data: Before commitLayer is called, either at least one of (diffOutputs, blobAdditionalLayer, filenames)
// should be available; or indexToDiffID/indexToTOCDigest/blobDiffIDs should be enough to locate an existing c/storage layer.
// They are looked up in the order they are mentioned above.
diffOutputs map[int]*graphdriver.DriverWithDifferOutput // Mapping from layer index to a partially-pulled layer intermediate data
blobAdditionalLayer map[digest.Digest]storage.AdditionalLayer // Mapping from layer blobsums to their corresponding additional layer
// Mapping from layer blobsums to names of files we used to hold them. If set, fileSizes and blobDiffIDs must also be set.
filenames map[digest.Digest]string
// Mapping from layer blobsums to their sizes. If set, filenames and blobDiffIDs must also be set.
fileSizes map[digest.Digest]int64
}
// addedLayerInfo records data about a layer to use in this image.
type addedLayerInfo struct {
digest digest.Digest // Mandatory, the digest of the layer.
emptyLayer bool // The layer is an “empty”/“throwaway” one, and may or may not be physically represented in various transport / storage systems. false if the manifest type does not have the concept.
}
// newImageDestination sets us up to write a new image, caching blobs in a temporary directory until
// it's time to Commit() the image
func newImageDestination(sys *types.SystemContext, imageRef storageReference) (*storageImageDestination, error) {
directory, err := tmpdir.MkDirBigFileTemp(sys, "storage")
if err != nil {
return nil, fmt.Errorf("creating a temporary directory: %w", err)
}
dest := &storageImageDestination{
PropertyMethodsInitialize: impl.PropertyMethods(impl.Properties{
SupportedManifestMIMETypes: []string{
imgspecv1.MediaTypeImageManifest,
manifest.DockerV2Schema2MediaType,
manifest.DockerV2Schema1SignedMediaType,
manifest.DockerV2Schema1MediaType,
},
// We ultimately have to decompress layers to populate trees on disk
// and need to explicitly ask for it here, so that the layers' MIME
// types can be set accordingly.
DesiredLayerCompression: types.PreserveOriginal,
AcceptsForeignLayerURLs: false,
MustMatchRuntimeOS: true,
IgnoresEmbeddedDockerReference: true, // Yes, we want the unmodified manifest
HasThreadSafePutBlob: true,
}),
imageRef: imageRef,
directory: directory,
signatureses: make(map[digest.Digest][]byte),
metadata: storageImageMetadata{
SignatureSizes: []int{},
SignaturesSizes: make(map[digest.Digest][]int),
},
indexToStorageID: make(map[int]string),
lockProtected: storageImageDestinationLockProtected{
indexToAddedLayerInfo: make(map[int]addedLayerInfo),
indexToDiffID: make(map[int]digest.Digest),
indexToTOCDigest: make(map[int]digest.Digest),
blobDiffIDs: make(map[digest.Digest]digest.Digest),
diffOutputs: make(map[int]*graphdriver.DriverWithDifferOutput),
blobAdditionalLayer: make(map[digest.Digest]storage.AdditionalLayer),
filenames: make(map[digest.Digest]string),
fileSizes: make(map[digest.Digest]int64),
},
}
dest.Compat = impl.AddCompat(dest)
return dest, nil
}
// Reference returns the reference used to set up this destination. Note that this should directly correspond to user's intent,
// e.g. it should use the public hostname instead of the result of resolving CNAMEs or following redirects.
func (s *storageImageDestination) Reference() types.ImageReference {
return s.imageRef
}
// Close cleans up the temporary directory and additional layer store handlers.
func (s *storageImageDestination) Close() error {
// This is outside of the scope of HasThreadSafePutBlob, so we don’t need to hold s.lock.
for _, al := range s.lockProtected.blobAdditionalLayer {
al.Release()
}
for _, v := range s.lockProtected.diffOutputs {
if v.Target != "" {
_ = s.imageRef.transport.store.CleanupStagedLayer(v)
}
}
return os.RemoveAll(s.directory)
}
func (s *storageImageDestination) computeNextBlobCacheFile() string {
return filepath.Join(s.directory, fmt.Sprintf("%d", s.nextTempFileID.Add(1)))
}
// PutBlobWithOptions writes contents of stream and returns data representing the result.
// inputInfo.Digest can be optionally provided if known; if provided, and stream is read to the end without error, the digest MUST match the stream contents.
// inputInfo.Size is the expected length of stream, if known.
// inputInfo.MediaType describes the blob format, if known.
// WARNING: The contents of stream are being verified on the fly. Until stream.Read() returns io.EOF, the contents of the data SHOULD NOT be available
// to any other readers for download using the supplied digest.
// If stream.Read() at any time, ESPECIALLY at end of input, returns an error, PutBlob MUST 1) fail, and 2) delete any data stored so far.
func (s *storageImageDestination) PutBlobWithOptions(ctx context.Context, stream io.Reader, blobinfo types.BlobInfo, options private.PutBlobOptions) (private.UploadedBlob, error) {
info, err := s.putBlobToPendingFile(stream, blobinfo, &options)
if err != nil {
return info, err
}
if options.IsConfig || options.LayerIndex == nil {
return info, nil
}
return info, s.queueOrCommit(*options.LayerIndex, addedLayerInfo{
digest: info.Digest,
emptyLayer: options.EmptyLayer,
})
}
// putBlobToPendingFile implements ImageDestination.PutBlobWithOptions, storing stream into an on-disk file.
// The caller must arrange the blob to be eventually committed using s.commitLayer().
func (s *storageImageDestination) putBlobToPendingFile(stream io.Reader, blobinfo types.BlobInfo, options *private.PutBlobOptions) (private.UploadedBlob, error) {
// Stores a layer or data blob in our temporary directory, checking that any information
// in the blobinfo matches the incoming data.
if blobinfo.Digest != "" {
if err := blobinfo.Digest.Validate(); err != nil {
return private.UploadedBlob{}, fmt.Errorf("invalid digest %#v: %w", blobinfo.Digest.String(), err)
}
}
// Set up to digest the blob if necessary, and count its size while saving it to a file.
filename := s.computeNextBlobCacheFile()
file, err := os.OpenFile(filename, os.O_CREATE|os.O_TRUNC|os.O_WRONLY|os.O_EXCL, 0600)
if err != nil {
return private.UploadedBlob{}, fmt.Errorf("creating temporary file %q: %w", filename, err)
}
defer file.Close()
counter := ioutils.NewWriteCounter(file)
stream = io.TeeReader(stream, counter)
digester, stream := putblobdigest.DigestIfUnknown(stream, blobinfo)
decompressed, err := archive.DecompressStream(stream)
if err != nil {
return private.UploadedBlob{}, fmt.Errorf("setting up to decompress blob: %w", err)
}
diffID := digest.Canonical.Digester()
// Copy the data to the file.
// TODO: This can take quite some time, and should ideally be cancellable using context.Context.
_, err = io.Copy(diffID.Hash(), decompressed)
decompressed.Close()
if err != nil {
return private.UploadedBlob{}, fmt.Errorf("storing blob to file %q: %w", filename, err)
}
// Determine blob properties, and fail if information that we were given about the blob
// is known to be incorrect.
blobDigest := digester.Digest()
blobSize := blobinfo.Size
if blobSize < 0 {
blobSize = counter.Count
} else if blobinfo.Size != counter.Count {
return private.UploadedBlob{}, ErrBlobSizeMismatch
}
// Record information about the blob.
s.lock.Lock()
s.lockProtected.blobDiffIDs[blobDigest] = diffID.Digest()
s.lockProtected.fileSizes[blobDigest] = counter.Count
s.lockProtected.filenames[blobDigest] = filename
s.lock.Unlock()
// This is safe because we have just computed diffID, and blobDigest was either computed
// by us, or validated by the caller (usually copy.digestingReader).
options.Cache.RecordDigestUncompressedPair(blobDigest, diffID.Digest())
return private.UploadedBlob{
Digest: blobDigest,
Size: blobSize,
}, nil
}
type zstdFetcher struct {
chunkAccessor private.BlobChunkAccessor
ctx context.Context
blobInfo types.BlobInfo
}
// GetBlobAt converts from chunked.GetBlobAt to BlobChunkAccessor.GetBlobAt.
func (f *zstdFetcher) GetBlobAt(chunks []chunked.ImageSourceChunk) (chan io.ReadCloser, chan error, error) {
newChunks := make([]private.ImageSourceChunk, 0, len(chunks))
for _, v := range chunks {
i := private.ImageSourceChunk{
Offset: v.Offset,
Length: v.Length,
}
newChunks = append(newChunks, i)
}
rc, errs, err := f.chunkAccessor.GetBlobAt(f.ctx, f.blobInfo, newChunks)
if _, ok := err.(private.BadPartialRequestError); ok {
err = chunked.ErrBadRequest{}
}
return rc, errs, err
}
// PutBlobPartial attempts to create a blob using the data that is already present
// at the destination. chunkAccessor is accessed in a non-sequential way to retrieve the missing chunks.
// It is available only if SupportsPutBlobPartial().
// Even if SupportsPutBlobPartial() returns true, the call can fail, in which case the caller
// should fall back to PutBlobWithOptions.
func (s *storageImageDestination) PutBlobPartial(ctx context.Context, chunkAccessor private.BlobChunkAccessor, srcInfo types.BlobInfo, options private.PutBlobPartialOptions) (private.UploadedBlob, error) {
fetcher := zstdFetcher{
chunkAccessor: chunkAccessor,
ctx: ctx,
blobInfo: srcInfo,
}
differ, err := chunked.GetDiffer(ctx, s.imageRef.transport.store, srcInfo.Digest, srcInfo.Size, srcInfo.Annotations, &fetcher)
if err != nil {
return private.UploadedBlob{}, err
}
out, err := s.imageRef.transport.store.ApplyDiffWithDiffer("", nil, differ)
if err != nil {
return private.UploadedBlob{}, err
}
if out.TOCDigest == "" && out.UncompressedDigest == "" {
return private.UploadedBlob{}, errors.New("internal error: ApplyDiffWithDiffer succeeded with neither TOCDigest nor UncompressedDigest set")
}
blobDigest := srcInfo.Digest
s.lock.Lock()
if out.UncompressedDigest != "" {
s.lockProtected.indexToDiffID[options.LayerIndex] = out.UncompressedDigest
if out.TOCDigest != "" {
options.Cache.RecordTOCUncompressedPair(out.TOCDigest, out.UncompressedDigest)
}
// The computation of UncompressedDigest means the whole layer has been consumed; while doing that, chunked.GetDiffer is
// responsible for ensuring blobDigest has been validated.
// So, record also information about blobDigest, that might benefit reuse.
s.lockProtected.blobDiffIDs[blobDigest] = out.UncompressedDigest
options.Cache.RecordDigestUncompressedPair(blobDigest, out.UncompressedDigest)
} else {
// Don’t identify layers by TOC if UncompressedDigest is available.
// - Using UncompressedDigest allows image reuse with non-partially-pulled layers
// - If UncompressedDigest has been computed, that means the layer was read completely, and the TOC has been created from scratch.
// That TOC is quite unlikely to match with any other TOC value.
s.lockProtected.indexToTOCDigest[options.LayerIndex] = out.TOCDigest
}
s.lockProtected.diffOutputs[options.LayerIndex] = out
s.lock.Unlock()
return private.UploadedBlob{
Digest: blobDigest,
Size: srcInfo.Size,
}, nil
}
// TryReusingBlobWithOptions checks whether the transport already contains, or can efficiently reuse, a blob, and if so, applies it to the current destination
// (e.g. if the blob is a filesystem layer, this signifies that the changes it describes need to be applied again when composing a filesystem tree).
// info.Digest must not be empty.
// If the blob has been successfully reused, returns (true, info, nil).
// If the transport can not reuse the requested blob, TryReusingBlob returns (false, {}, nil); it returns a non-nil error only on an unexpected failure.
func (s *storageImageDestination) TryReusingBlobWithOptions(ctx context.Context, blobinfo types.BlobInfo, options private.TryReusingBlobOptions) (bool, private.ReusedBlob, error) {
if !impl.OriginalCandidateMatchesTryReusingBlobOptions(options) {
return false, private.ReusedBlob{}, nil
}
reused, info, err := s.tryReusingBlobAsPending(blobinfo.Digest, blobinfo.Size, &options)
if err != nil || !reused || options.LayerIndex == nil {
return reused, info, err
}
return reused, info, s.queueOrCommit(*options.LayerIndex, addedLayerInfo{
digest: info.Digest,
emptyLayer: options.EmptyLayer,
})
}
// tryReusingBlobAsPending implements TryReusingBlobWithOptions for (blobDigest, size or -1), filling s.blobDiffIDs and other metadata.
// The caller must arrange the blob to be eventually committed using s.commitLayer().
func (s *storageImageDestination) tryReusingBlobAsPending(blobDigest digest.Digest, size int64, options *private.TryReusingBlobOptions) (bool, private.ReusedBlob, error) {
// lock the entire method as it executes fairly quickly
s.lock.Lock()
defer s.lock.Unlock()
if options.SrcRef != nil {
// Check if we have the layer in the underlying additional layer store.
aLayer, err := s.imageRef.transport.store.LookupAdditionalLayer(blobDigest, options.SrcRef.String())
if err != nil && !errors.Is(err, storage.ErrLayerUnknown) {
return false, private.ReusedBlob{}, fmt.Errorf(`looking for compressed layers with digest %q and labels: %w`, blobDigest, err)
} else if err == nil {
s.lockProtected.blobDiffIDs[blobDigest] = aLayer.UncompressedDigest()
s.lockProtected.blobAdditionalLayer[blobDigest] = aLayer
return true, private.ReusedBlob{
Digest: blobDigest,
Size: aLayer.CompressedSize(),
}, nil
}
}
if blobDigest == "" {
return false, private.ReusedBlob{}, errors.New(`Can not check for a blob with unknown digest`)
}
if err := blobDigest.Validate(); err != nil {
return false, private.ReusedBlob{}, fmt.Errorf("Can not check for a blob with invalid digest: %w", err)
}
if options.TOCDigest != "" {
if err := options.TOCDigest.Validate(); err != nil {
return false, private.ReusedBlob{}, fmt.Errorf("Can not check for a blob with invalid digest: %w", err)
}
}
// Check if we have a wasn't-compressed layer in storage that's based on that blob.
// Check if we've already cached it in a file.
if size, ok := s.lockProtected.fileSizes[blobDigest]; ok {
// s.lockProtected.blobDiffIDs is set either by putBlobToPendingFile or in createNewLayer when creating the
// filenames/fileSizes entry.
return true, private.ReusedBlob{
Digest: blobDigest,
Size: size,
}, nil
}
layers, err := s.imageRef.transport.store.LayersByUncompressedDigest(blobDigest)
if err != nil && !errors.Is(err, storage.ErrLayerUnknown) {
return false, private.ReusedBlob{}, fmt.Errorf(`looking for layers with digest %q: %w`, blobDigest, err)
}
if len(layers) > 0 {
s.lockProtected.blobDiffIDs[blobDigest] = blobDigest
return true, private.ReusedBlob{
Digest: blobDigest,
Size: layers[0].UncompressedSize,
}, nil
}
// Check if we have a was-compressed layer in storage that's based on that blob.
layers, err = s.imageRef.transport.store.LayersByCompressedDigest(blobDigest)
if err != nil && !errors.Is(err, storage.ErrLayerUnknown) {
return false, private.ReusedBlob{}, fmt.Errorf(`looking for compressed layers with digest %q: %w`, blobDigest, err)
}
if len(layers) > 0 {
// LayersByCompressedDigest only finds layers which were created from a full layer blob, and extracting that
// always sets UncompressedDigest.
diffID := layers[0].UncompressedDigest
if diffID == "" {
return false, private.ReusedBlob{}, fmt.Errorf("internal error: compressed layer %q (for compressed digest %q) does not have an uncompressed digest", layers[0].ID, blobDigest.String())
}
s.lockProtected.blobDiffIDs[blobDigest] = diffID
return true, private.ReusedBlob{
Digest: blobDigest,
Size: layers[0].CompressedSize,
}, nil
}
// Does the blob correspond to a known DiffID which we already have available?
// Because we must return the size, which is unknown for unavailable compressed blobs, the returned BlobInfo refers to the
// uncompressed layer, and that can happen only if options.CanSubstitute, or if the incoming manifest already specifies the size.
if options.CanSubstitute || size != -1 {
if uncompressedDigest := options.Cache.UncompressedDigest(blobDigest); uncompressedDigest != "" && uncompressedDigest != blobDigest {
layers, err := s.imageRef.transport.store.LayersByUncompressedDigest(uncompressedDigest)
if err != nil && !errors.Is(err, storage.ErrLayerUnknown) {
return false, private.ReusedBlob{}, fmt.Errorf(`looking for layers with digest %q: %w`, uncompressedDigest, err)
}
if found, reused := reusedBlobFromLayerLookupLocked(layers, blobDigest, size, options); found {
s.lockProtected.blobDiffIDs[blobDigest] = uncompressedDigest
return true, reused, nil
}
}
}
if options.TOCDigest != "" && options.LayerIndex != nil {
// Check if we know which which UncompressedDigest the TOC digest resolves to, and we have a match for that.
// Prefer this over LayersByTOCDigest because we can identify the layer using UncompressedDigest, maximizing reuse.
if uncompressedDigest := options.Cache.UncompressedDigestForTOC(options.TOCDigest); uncompressedDigest != "" {
layers, err = s.imageRef.transport.store.LayersByUncompressedDigest(uncompressedDigest)
if err != nil && !errors.Is(err, storage.ErrLayerUnknown) {
return false, private.ReusedBlob{}, fmt.Errorf(`looking for layers with digest %q: %w`, uncompressedDigest, err)
}
if found, reused := reusedBlobFromLayerLookupLocked(layers, blobDigest, size, options); found {
s.lockProtected.indexToDiffID[*options.LayerIndex] = uncompressedDigest
reused.MatchedByTOCDigest = true
return true, reused, nil
}
}
// Check if we have a chunked layer in storage with the same TOC digest.
layers, err := s.imageRef.transport.store.LayersByTOCDigest(options.TOCDigest)
if err != nil && !errors.Is(err, storage.ErrLayerUnknown) {
return false, private.ReusedBlob{}, fmt.Errorf(`looking for layers with TOC digest %q: %w`, options.TOCDigest, err)
}
if found, reused := reusedBlobFromLayerLookupLocked(layers, blobDigest, size, options); found {
s.lockProtected.indexToTOCDigest[*options.LayerIndex] = options.TOCDigest
reused.MatchedByTOCDigest = true
return true, reused, nil
}
}
// Nope, we don't have it.
return false, private.ReusedBlob{}, nil
}
// reusedBlobFromLayerLookupLocked returns (true, ReusedBlob) if layers contain a usable match; or (false, ...) if not.
// The caller is still responsible for setting the layer identification fields, to allow the layer to be found again.
// The caller must hold s.lock.
func reusedBlobFromLayerLookupLocked(layers []storage.Layer, blobDigest digest.Digest, blobSize int64, options *private.TryReusingBlobOptions) (bool, private.ReusedBlob) {
if len(layers) > 0 {
if blobSize != -1 {
return true, private.ReusedBlob{
Digest: blobDigest,
Size: blobSize,
}
} else if options.CanSubstitute && layers[0].UncompressedDigest != "" {
return true, private.ReusedBlob{
Digest: layers[0].UncompressedDigest,
Size: layers[0].UncompressedSize,
}
}
}
return false, private.ReusedBlob{}
}
// computeID computes a recommended image ID based on information we have so far. If
// the manifest is not of a type that we recognize, we return an empty value, indicating
// that since we don't have a recommendation, a random ID should be used if one needs
// to be allocated.
func (s *storageImageDestination) computeID(m manifest.Manifest) string {
// This is outside of the scope of HasThreadSafePutBlob, so we don’t need to hold s.lock.
// Build the diffID list. We need the decompressed sums that we've been calculating to
// fill in the DiffIDs. It's expected (but not enforced by us) that the number of
// diffIDs corresponds to the number of non-EmptyLayer entries in the history.
var diffIDs []digest.Digest
switch m := m.(type) {
case *manifest.Schema1:
// Build a list of the diffIDs we've generated for the non-throwaway FS layers,
// in reverse of the order in which they were originally listed.
for i, compat := range m.ExtractedV1Compatibility {
if compat.ThrowAway {
continue
}
blobSum := m.FSLayers[i].BlobSum
diffID, ok := s.lockProtected.blobDiffIDs[blobSum]
if !ok {
// this can, in principle, legitimately happen when a layer is reused by TOC.
logrus.Infof("error looking up diffID for layer %q", blobSum.String())
return ""
}
diffIDs = append([]digest.Digest{diffID}, diffIDs...)
}
case *manifest.Schema2, *manifest.OCI1:
// We know the ID calculation doesn't actually use the diffIDs, so we don't need to populate
// the diffID list.
default:
return ""
}
// We want to use the same ID for “the same” images, but without risking unwanted sharing / malicious image corruption.
//
// Traditionally that means the same ~config digest, as computed by m.ImageID;
// but if we pull a layer by TOC, we verify the layer against neither the (compressed) blob digest in the manifest,
// nor against the config’s RootFS.DiffIDs. We don’t really want to do either, to allow partial layer pulls where we never see
// most of the data.
//
// So, if a layer is pulled by TOC (and we do validate against the TOC), the fact that we used the TOC, and the value of the TOC,
// must enter into the image ID computation.
// But for images where no TOC was used, continue to use IDs computed the traditional way, to maximize image reuse on upgrades,
// and to introduce the changed behavior only when partial pulls are used.
//
// Note that it’s not 100% guaranteed that an image pulled by TOC uses an OCI manifest; consider
// (skopeo copy --format v2s2 docker://…/zstd-chunked-image containers-storage:… ). So this is not happening only in the OCI case above.
ordinaryImageID, err := m.ImageID(diffIDs)
if err != nil {
return ""
}
tocIDInput := ""
hasLayerPulledByTOC := false
for i := range m.LayerInfos() {
layerValue := "" // An empty string is not a valid digest, so this is unambiguous with the TOC case.
tocDigest, ok := s.lockProtected.indexToTOCDigest[i] // "" if not a TOC, incl. the case where indexToDiffID is set
if ok {
hasLayerPulledByTOC = true
layerValue = tocDigest.String()
}
tocIDInput += layerValue + "|" // "|" can not be present in a TOC digest, so this is an unambiguous separator.
}
if !hasLayerPulledByTOC {
return ordinaryImageID
}
// ordinaryImageID is a digest of a config, which is a JSON value.
// To avoid the risk of collisions, start the input with @ so that the input is not a valid JSON.
tocImageID := digest.FromString("@With TOC:" + tocIDInput).Hex()
logrus.Debugf("Ordinary storage image ID %s; a layer was looked up by TOC, so using image ID %s", ordinaryImageID, tocImageID)
return tocImageID
}
// getConfigBlob exists only to let us retrieve the configuration blob so that the manifest package can dig
// information out of it for Inspect().
func (s *storageImageDestination) getConfigBlob(info types.BlobInfo) ([]byte, error) {
if info.Digest == "" {
return nil, errors.New(`no digest supplied when reading blob`)
}
if err := info.Digest.Validate(); err != nil {
return nil, fmt.Errorf("invalid digest supplied when reading blob: %w", err)
}
// Assume it's a file, since we're only calling this from a place that expects to read files.
if filename, ok := s.lockProtected.filenames[info.Digest]; ok {
contents, err2 := os.ReadFile(filename)
if err2 != nil {
return nil, fmt.Errorf(`reading blob from file %q: %w`, filename, err2)
}
return contents, nil
}
// If it's not a file, it's a bug, because we're not expecting to be asked for a layer.
return nil, errors.New("blob not found")
}
// queueOrCommit queues the specified layer to be committed to the storage.
// If no other goroutine is already committing layers, the layer and all
// subsequent layers (if already queued) will be committed to the storage.
func (s *storageImageDestination) queueOrCommit(index int, info addedLayerInfo) error {
// NOTE: whenever the code below is touched, make sure that all code
// paths unlock the lock and to unlock it exactly once.
//
// Conceptually, the code is divided in two stages:
//
// 1) Queue in work by marking the layer as ready to be committed.
// If at least one previous/parent layer with a lower index has
// not yet been committed, return early.
//
// 2) Process the queued-in work by committing the "ready" layers
// in sequence. Make sure that more items can be queued-in
// during the comparatively I/O expensive task of committing a
// layer.
//
// The conceptual benefit of this design is that caller can continue
// pulling layers after an early return. At any given time, only one
// caller is the "worker" routine committing layers. All other routines
// can continue pulling and queuing in layers.
s.lock.Lock()
s.lockProtected.indexToAddedLayerInfo[index] = info
// We're still waiting for at least one previous/parent layer to be
// committed, so there's nothing to do.
if index != s.lockProtected.currentIndex {
s.lock.Unlock()
return nil
}
for {
info, ok := s.lockProtected.indexToAddedLayerInfo[index]
if !ok {
break
}
s.lock.Unlock()
// Note: commitLayer locks on-demand.
if stopQueue, err := s.commitLayer(index, info, -1); stopQueue || err != nil {
return err
}
s.lock.Lock()
index++
}
// Set the index at the very end to make sure that only one routine
// enters stage 2).
s.lockProtected.currentIndex = index
s.lock.Unlock()
return nil
}
// singleLayerIDComponent returns a single layer’s the input to computing a layer (chain) ID,
// and an indication whether the input already has the shape of a layer ID.
// It returns ("", false) if the layer is not found at all (which should never happen)
func (s *storageImageDestination) singleLayerIDComponent(layerIndex int, blobDigest digest.Digest) (string, bool) {
diffIDcomponent := func(d digest.Digest) string {
return d.Hex() // This looks like chain IDs, and it uses the traditional value.
}
s.lock.Lock()
defer s.lock.Unlock()
if d, found := s.lockProtected.indexToDiffID[layerIndex]; found {
return diffIDcomponent(d), true
}
if d, found := s.lockProtected.indexToTOCDigest[layerIndex]; found {
return "@TOC=" + d.Hex(), false // "@" is not a valid start of a digest.Digest, so this is unambiguous.
}
if d, found := s.lockProtected.blobDiffIDs[blobDigest]; found {
return diffIDcomponent(d), true
}
return "", false
}
// commitLayer commits the specified layer with the given index to the storage.
// size can usually be -1; it can be provided if the layer is not known to be already present in blobDiffIDs.
//
// If the layer cannot be committed yet, the function returns (true, nil).
//
// Note that the previous layer is expected to already be committed.
//
// Caution: this function must be called without holding `s.lock`. Callers
// must guarantee that, at any given time, at most one goroutine may execute
// `commitLayer()`.
func (s *storageImageDestination) commitLayer(index int, info addedLayerInfo, size int64) (bool, error) {
// Already committed? Return early.
if _, alreadyCommitted := s.indexToStorageID[index]; alreadyCommitted {
return false, nil
}
// Start with an empty string or the previous layer ID. Note that
// `s.indexToStorageID` can only be accessed by *one* goroutine at any
// given time. Hence, we don't need to lock accesses.
var parentLayer string
if index != 0 {
prev, ok := s.indexToStorageID[index-1]
if !ok {
return false, fmt.Errorf("Internal error: commitLayer called with previous layer %d not committed yet", index-1)
}
parentLayer = prev
}
// Carry over the previous ID for empty non-base layers.
if info.emptyLayer {
s.indexToStorageID[index] = parentLayer
return false, nil
}
// Check if there's already a layer with the ID that we'd give to the result of applying
// this layer blob to its parent, if it has one, or the blob's hex value otherwise.
// The layerID refers either to the DiffID or the digest of the TOC.
layerIDComponent, layerIDComponentStandalone := s.singleLayerIDComponent(index, info.digest)
if layerIDComponent == "" {
// Check if it's elsewhere and the caller just forgot to pass it to us in a PutBlob() / TryReusingBlob() / …
//
// Use none.NoCache to avoid a repeated DiffID lookup in the BlobInfoCache: a caller
// that relies on using a blob digest that has never been seen by the store had better call
// TryReusingBlob; not calling PutBlob already violates the documented API, so there’s only
// so far we are going to accommodate that (if we should be doing that at all).
//
// We are also ignoring lookups by TOC, and other non-trivial situations.
// Those can only happen using the c/image/internal/private API,
// so those internal callers should be fixed to follow the API instead of expanding this fallback.
logrus.Debugf("looking for diffID for blob=%+v", info.digest)
// Use tryReusingBlobAsPending, not the top-level TryReusingBlobWithOptions, to prevent recursion via queueOrCommit.
has, _, err := s.tryReusingBlobAsPending(info.digest, size, &private.TryReusingBlobOptions{
Cache: none.NoCache,
CanSubstitute: false,
})
if err != nil {
return false, fmt.Errorf("checking for a layer based on blob %q: %w", info.digest.String(), err)
}
if !has {
return false, fmt.Errorf("error determining uncompressed digest for blob %q", info.digest.String())
}
layerIDComponent, layerIDComponentStandalone = s.singleLayerIDComponent(index, info.digest)
if layerIDComponent == "" {
return false, fmt.Errorf("we have blob %q, but don't know its layer ID", info.digest.String())
}
}
id := layerIDComponent
if !layerIDComponentStandalone || parentLayer != "" {
id = digest.Canonical.FromString(parentLayer + "+" + layerIDComponent).Hex()
}
if layer, err2 := s.imageRef.transport.store.Layer(id); layer != nil && err2 == nil {
// There's already a layer that should have the right contents, just reuse it.
s.indexToStorageID[index] = layer.ID
return false, nil
}
layer, err := s.createNewLayer(index, info.digest, parentLayer, id)
if err != nil {
return false, err
}
if layer == nil {
return true, nil
}
s.indexToStorageID[index] = layer.ID
return false, nil
}
// createNewLayer creates a new layer newLayerID for (index, layerDigest) on top of parentLayer (which may be "").
// If the layer cannot be committed yet, the function returns (nil, nil).
func (s *storageImageDestination) createNewLayer(index int, layerDigest digest.Digest, parentLayer, newLayerID string) (*storage.Layer, error) {
s.lock.Lock()
diffOutput, ok := s.lockProtected.diffOutputs[index]
s.lock.Unlock()
if ok {
var untrustedUncompressedDigest digest.Digest
if diffOutput.UncompressedDigest == "" {
d, err := s.untrustedLayerDiffID(index)
if err != nil {
return nil, err
}
if d == "" {
logrus.Debugf("Skipping commit for layer %q, manifest not yet available", newLayerID)
return nil, nil
}
untrustedUncompressedDigest = d
}
flags := make(map[string]interface{})
if untrustedUncompressedDigest != "" {
flags[expectedLayerDiffIDFlag] = untrustedUncompressedDigest
logrus.Debugf("Setting uncompressed digest to %q for layer %q", untrustedUncompressedDigest, newLayerID)
}
args := storage.ApplyStagedLayerOptions{
ID: newLayerID,
ParentLayer: parentLayer,
DiffOutput: diffOutput,
DiffOptions: &graphdriver.ApplyDiffWithDifferOpts{
Flags: flags,
},
}
layer, err := s.imageRef.transport.store.ApplyStagedLayer(args)
if err != nil && !errors.Is(err, storage.ErrDuplicateID) {
return nil, fmt.Errorf("failed to put layer using a partial pull: %w", err)
}
return layer, nil
}
s.lock.Lock()
al, ok := s.lockProtected.blobAdditionalLayer[layerDigest]
s.lock.Unlock()
if ok {
layer, err := al.PutAs(newLayerID, parentLayer, nil)
if err != nil && !errors.Is(err, storage.ErrDuplicateID) {
return nil, fmt.Errorf("failed to put layer from digest and labels: %w", err)
}
return layer, nil
}
// Check if we previously cached a file with that blob's contents. If we didn't,
// then we need to read the desired contents from a layer.
var trustedUncompressedDigest, trustedOriginalDigest digest.Digest // For storage.LayerOptions
s.lock.Lock()
tocDigest := s.lockProtected.indexToTOCDigest[index] // "" if not set
// We look up optionalDiffID in blobDiffIDs first, contrary to singleLayerIDComponent, to bias towards
// setting layerDigestIsTrusted to true.
optionalDiffID, layerDigestIsTrusted := s.lockProtected.blobDiffIDs[layerDigest]
if !layerDigestIsTrusted {
optionalDiffID = s.lockProtected.indexToDiffID[index] // "" if not set
}
filename, gotFilename := s.lockProtected.filenames[layerDigest]
s.lock.Unlock()
if gotFilename && tocDigest == "" {
// If tocDigest != "", if we now happen to find a layerDigest match, the newLayerID has already been computed as TOC-based,
// and we don't know the relationship of the layerDigest and TOC digest.
// We could recompute newLayerID to be DiffID-based and use the file, but such a within-image layer
// reuse is expected to be pretty rare; instead, ignore the unexpected file match and proceed to the
// originally-planned TOC match.
// Because tocDigest == "", optionaldiffID must have been set; and even if it weren’t, PutLayer will recompute the digest from the stream.
trustedUncompressedDigest = optionalDiffID
trustedOriginalDigest = layerDigest // The code setting .filenames[layerDigest] is responsible for the contents matching.
} else {
// Try to find the layer with contents matching the data we use.
var layer *storage.Layer // = nil
if optionalDiffID != "" {
if layers, err2 := s.imageRef.transport.store.LayersByUncompressedDigest(optionalDiffID); err2 == nil && len(layers) > 0 {
layer = &layers[0]
}
}
if layer == nil {
if tocDigest != "" {
if layers, err2 := s.imageRef.transport.store.LayersByTOCDigest(tocDigest); err2 == nil && len(layers) > 0 {
layer = &layers[0]
} else {
return nil, fmt.Errorf("locating layer for TOC digest %q (diffID %q): %w", tocDigest, optionalDiffID, err2)
}
} else {
// Because tocDigest == "", optionaldiffID must have been set.
if !layerDigestIsTrusted {
// We can get here if we got optionalDiffID from a TOC lookup in BlobInfoCache, and we originally found a layer
// … and that layer is no longer available.
return nil, fmt.Errorf("layer for diffID %q is no longer available", optionalDiffID)
}
if layers, err2 := s.imageRef.transport.store.LayersByCompressedDigest(layerDigest); err2 == nil && len(layers) > 0 {
layer = &layers[0]
} else {
return nil, fmt.Errorf("locating layer for blob %q (diffID %q): %w", layerDigest, optionalDiffID, err2)
}
}
}
// Read the layer's contents.
noCompression := archive.Uncompressed
diffOptions := &storage.DiffOptions{
Compression: &noCompression,
}
diff, err2 := s.imageRef.transport.store.Diff("", layer.ID, diffOptions)
if err2 != nil {
return nil, fmt.Errorf("reading layer %q for blob %q/%q/%q: %w", layer.ID, layerDigest, tocDigest, optionalDiffID, err2)
}
// Copy the layer diff to a file. Diff() takes a lock that it holds
// until the ReadCloser that it returns is closed, and PutLayer() wants
// the same lock, so the diff can't just be directly streamed from one
// to the other.
filename = s.computeNextBlobCacheFile()
file, err := os.OpenFile(filename, os.O_CREATE|os.O_TRUNC|os.O_WRONLY|os.O_EXCL, 0o600)
if err != nil {
diff.Close()
return nil, fmt.Errorf("creating temporary file %q: %w", filename, err)
}
// Copy the data to the file.
// TODO: This can take quite some time, and should ideally be cancellable using
// ctx.Done().
fileSize, err := io.Copy(file, diff)
diff.Close()
file.Close()
if err != nil {
return nil, fmt.Errorf("storing blob to file %q: %w", filename, err)
}
if optionalDiffID == "" && layer.UncompressedDigest != "" {
optionalDiffID = layer.UncompressedDigest
}
// The stream we have is uncompressed, this matches contents of the stream.
// If tocDigest != "", trustedUncompressedDigest might still be ""; in that case PutLayer will compute the value from the stream.
trustedUncompressedDigest = optionalDiffID
// FIXME? trustedOriginalDigest could be set to layerDigest IF layerDigestIsTrusted.
// But for c/storage to reasonably use it (as a CompressedDigest value), we should also ensure the CompressedSize of the created
// layer is correct, and the API does not currently make it possible (.CompressedSize is set from the input stream).
//
// We can legitimately set storage.LayerOptions.OriginalDigest to "",
// but that would just result in PutLayer computing the digest of the input stream == optionalDiffID.
// So, instead, set .OriginalDigest to the value we know already, to avoid that digest computation.
trustedOriginalDigest = optionalDiffID
// Allow using the already-collected layer contents without extracting the layer again.
//
// This only matches against the uncompressed digest.
// We don’t have the original compressed data here to trivially set filenames[layerDigest].
// In particular we can’t achieve the correct Layer.CompressedSize value with the current c/storage API.
// Within-image layer reuse is probably very rare, for now we prefer to avoid that complexity.
if trustedUncompressedDigest != "" {
s.lock.Lock()
s.lockProtected.blobDiffIDs[trustedUncompressedDigest] = trustedUncompressedDigest
s.lockProtected.filenames[trustedUncompressedDigest] = filename
s.lockProtected.fileSizes[trustedUncompressedDigest] = fileSize
s.lock.Unlock()
}
}
// Read the cached blob and use it as a diff.
file, err := os.Open(filename)
if err != nil {
return nil, fmt.Errorf("opening file %q: %w", filename, err)
}
defer file.Close()
// Build the new layer using the diff, regardless of where it came from.
// TODO: This can take quite some time, and should ideally be cancellable using ctx.Done().
layer, _, err := s.imageRef.transport.store.PutLayer(newLayerID, parentLayer, nil, "", false, &storage.LayerOptions{
OriginalDigest: trustedOriginalDigest,
UncompressedDigest: trustedUncompressedDigest,
}, file)
if err != nil && !errors.Is(err, storage.ErrDuplicateID) {
return nil, fmt.Errorf("adding layer with blob %q/%q/%q: %w", layerDigest, tocDigest, optionalDiffID, err)
}
return layer, nil
}
// untrustedLayerDiffID returns a DiffID value for layerIndex from the image’s config.
// If the value is not yet available (but it can be available after s.manifets is set), it returns ("", nil).
// WARNING: We don’t validate the DiffID value against the layer contents; it must not be used for any deduplication.
func (s *storageImageDestination) untrustedLayerDiffID(layerIndex int) (digest.Digest, error) {
// At this point, we are either inside the multi-threaded scope of HasThreadSafePutBlob, and
// nothing is writing to s.manifest yet, or PutManifest has been called and s.manifest != nil.
// Either way this function does not need the protection of s.lock.
if s.manifest == nil {
logrus.Debugf("Skipping commit for layer %d, manifest not yet available", layerIndex)
return "", nil
}
if s.untrustedDiffIDValues == nil {
mt := manifest.GuessMIMEType(s.manifest)
if mt != imgspecv1.MediaTypeImageManifest {
// We could, in principle, build an ImageSource, support arbitrary image formats using image.FromUnparsedImage,
// and then use types.Image.OCIConfig so that we can parse the image.
//
// In practice, this should, right now, only matter for pulls of OCI images (this code path implies that a layer has annotation),
// while converting to a non-OCI formats, using a manual (skopeo copy) or something similar, not (podman pull).
// So it is not implemented yet.
return "", fmt.Errorf("determining DiffID for manifest type %q is not yet supported", mt)
}
man, err := manifest.FromBlob(s.manifest, mt)
if err != nil {
return "", fmt.Errorf("parsing manifest: %w", err)
}
cb, err := s.getConfigBlob(man.ConfigInfo())
if err != nil {
return "", err
}
// retrieve the expected uncompressed digest from the config blob.
configOCI := &imgspecv1.Image{}
if err := json.Unmarshal(cb, configOCI); err != nil {
return "", err
}
s.untrustedDiffIDValues = slices.Clone(configOCI.RootFS.DiffIDs)
if s.untrustedDiffIDValues == nil { // Unlikely but possible in theory…