forked from keybase/client
-
Notifications
You must be signed in to change notification settings - Fork 0
/
folder_block_manager.go
1587 lines (1421 loc) · 47 KB
/
folder_block_manager.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
// Copyright 2016 Keybase Inc. All rights reserved.
// Use of this source code is governed by a BSD
// license that can be found in the LICENSE file.
package libkbfs
import (
"fmt"
"sync"
"time"
"github.com/keybase/backoff"
"github.com/keybase/client/go/kbfs/data"
"github.com/keybase/client/go/kbfs/env"
"github.com/keybase/client/go/kbfs/kbfsblock"
"github.com/keybase/client/go/kbfs/kbfsmd"
"github.com/keybase/client/go/kbfs/kbfssync"
"github.com/keybase/client/go/kbfs/tlf"
"github.com/keybase/client/go/kbfs/tlfhandle"
"github.com/keybase/client/go/logger"
"github.com/keybase/client/go/protocol/keybase1"
"github.com/pkg/errors"
"golang.org/x/net/context"
"golang.org/x/sync/errgroup"
)
type fbmHelper interface {
getMostRecentFullyMergedMD(ctx context.Context) (
ImmutableRootMetadata, error)
finalizeGCOp(ctx context.Context, gco *GCOp) error
getLatestMergedRevision(lState *kbfssync.LockState) kbfsmd.Revision
}
const (
// How many pointers to downgrade in a single Archive/Delete call.
numPointersToDowngradePerChunk = 20
// Once the number of pointers being deleted in a single gc op
// passes this threshold, we'll stop garbage collection at the
// current revision.
numPointersPerGCThresholdDefault = 100
// The most revisions to consider for each QR run.
numMaxRevisionsPerQR = 100
)
type blockDeleteType int
const (
// Delete the blocks only if the given MD failed to make it to the
// servers.
blockDeleteOnMDFail blockDeleteType = iota
// Always delete the blocks, without first checking if the given
// revision was successful. This is just an optimization to avoid
// fetching the MD from the server when we know for sure it had
// failed.
blockDeleteAlways
)
type blocksToDelete struct {
md ReadOnlyRootMetadata
blocks []data.BlockPointer
bdType blockDeleteType
backoff backoff.BackOff
}
// folderBlockManager is a helper class for managing the blocks in a
// particular TLF. It archives historical blocks and reclaims quota
// usage, all in the background.
type folderBlockManager struct {
appStateUpdater env.AppStateUpdater
config Config
log logger.Logger
shutdownChan chan struct{}
id tlf.ID
numPointersPerGCThreshold int
// A queue of MD updates for this folder that need to have their
// unref's blocks archived
archiveChan chan ReadOnlyRootMetadata
archivePauseChan chan (<-chan struct{})
// archiveGroup tracks the outstanding archives.
archiveGroup kbfssync.RepeatedWaitGroup
archiveCancelLock sync.Mutex
archiveCancel context.CancelFunc
// blocksToDeleteChan is a list of blocks, for a given
// metadata revision, that may have been Put as part of a failed
// MD write. These blocks should be deleted as soon as we know
// for sure that the MD write isn't visible to others.
// TODO: Persist these to disk?
blocksToDeleteChan chan blocksToDelete
blocksToDeletePauseChan chan (<-chan struct{})
blocksToDeleteWaitGroup kbfssync.RepeatedWaitGroup
blocksToDeleteCancelLock sync.Mutex
blocksToDeleteCancel context.CancelFunc
// forceReclamation forces the manager to start a reclamation
// process.
forceReclamationChan chan struct{}
// reclamationGroup tracks the outstanding quota reclamations.
reclamationGroup kbfssync.RepeatedWaitGroup
reclamationCancelLock sync.Mutex
reclamationCancel context.CancelFunc
// latestMergedChan signals when we learn about a newer latest
// merged revision for this TLF.
latestMergedChan chan struct{}
// cleanDiskCachesGroup tracks the outstanding disk-cache cleanings.
cleanDiskCachesGroup kbfssync.RepeatedWaitGroup
cleanDiskCacheCancelLock sync.Mutex
cleanDiskCacheCancel context.CancelFunc
helper fbmHelper
// Remembers what happened last time during quota reclamation.
lastQRLock sync.Mutex
lastQRHeadRev kbfsmd.Revision
wasLastQRComplete bool
lastReclamationTime time.Time
lastReclaimedRev kbfsmd.Revision
}
func newFolderBlockManager(
appStateUpdater env.AppStateUpdater, config Config, fb data.FolderBranch,
bType branchType, helper fbmHelper) *folderBlockManager {
tlfStringFull := fb.Tlf.String()
log := config.MakeLogger(fmt.Sprintf("FBM %s", tlfStringFull[:8]))
var latestMergedChan chan struct{}
qrEnabled :=
fb.Branch == data.MasterBranch && config.Mode().QuotaReclamationEnabled()
if qrEnabled {
latestMergedChan = make(chan struct{}, 1)
}
fbm := &folderBlockManager{
appStateUpdater: appStateUpdater,
config: config,
log: log,
shutdownChan: make(chan struct{}),
id: fb.Tlf,
numPointersPerGCThreshold: numPointersPerGCThresholdDefault,
archiveChan: make(chan ReadOnlyRootMetadata, 500),
archivePauseChan: make(chan (<-chan struct{})),
blocksToDeleteChan: make(chan blocksToDelete, 25),
blocksToDeletePauseChan: make(chan (<-chan struct{})),
forceReclamationChan: make(chan struct{}, 1),
latestMergedChan: latestMergedChan,
helper: helper,
}
if bType != standard || !config.Mode().BlockManagementEnabled() {
return fbm
}
go fbm.archiveBlocksInBackground()
go fbm.deleteBlocksInBackground()
if qrEnabled {
go fbm.reclaimQuotaInBackground()
go fbm.cleanDiskCachesInBackground()
}
return fbm
}
func (fbm *folderBlockManager) setBlocksToDeleteCancel(cancel context.CancelFunc) {
fbm.blocksToDeleteCancelLock.Lock()
defer fbm.blocksToDeleteCancelLock.Unlock()
fbm.blocksToDeleteCancel = cancel
}
func (fbm *folderBlockManager) cancelBlocksToDelete() {
blocksToDeleteCancel := func() context.CancelFunc {
fbm.blocksToDeleteCancelLock.Lock()
defer fbm.blocksToDeleteCancelLock.Unlock()
blocksToDeleteCancel := fbm.blocksToDeleteCancel
fbm.blocksToDeleteCancel = nil
return blocksToDeleteCancel
}()
if blocksToDeleteCancel != nil {
blocksToDeleteCancel()
}
}
func (fbm *folderBlockManager) setArchiveCancel(cancel context.CancelFunc) {
fbm.archiveCancelLock.Lock()
defer fbm.archiveCancelLock.Unlock()
fbm.archiveCancel = cancel
}
func (fbm *folderBlockManager) cancelArchive() {
archiveCancel := func() context.CancelFunc {
fbm.archiveCancelLock.Lock()
defer fbm.archiveCancelLock.Unlock()
archiveCancel := fbm.archiveCancel
fbm.archiveCancel = nil
return archiveCancel
}()
if archiveCancel != nil {
archiveCancel()
}
}
func (fbm *folderBlockManager) setReclamationCancel(cancel context.CancelFunc) {
fbm.reclamationCancelLock.Lock()
defer fbm.reclamationCancelLock.Unlock()
fbm.reclamationCancel = cancel
}
func (fbm *folderBlockManager) cancelReclamation() {
reclamationCancel := func() context.CancelFunc {
fbm.reclamationCancelLock.Lock()
defer fbm.reclamationCancelLock.Unlock()
reclamationCancel := fbm.reclamationCancel
fbm.reclamationCancel = nil
return reclamationCancel
}()
if reclamationCancel != nil {
reclamationCancel()
}
}
func (fbm *folderBlockManager) setCleanDiskCacheCancel(
cancel context.CancelFunc) {
fbm.cleanDiskCacheCancelLock.Lock()
defer fbm.cleanDiskCacheCancelLock.Unlock()
fbm.cleanDiskCacheCancel = cancel
}
func (fbm *folderBlockManager) cancelCleanDiskCache() {
cleanDiskCacheCancel := func() context.CancelFunc {
fbm.cleanDiskCacheCancelLock.Lock()
defer fbm.cleanDiskCacheCancelLock.Unlock()
cleanDiskCacheCancel := fbm.cleanDiskCacheCancel
fbm.cleanDiskCacheCancel = nil
return cleanDiskCacheCancel
}()
if cleanDiskCacheCancel != nil {
cleanDiskCacheCancel()
}
}
func (fbm *folderBlockManager) shutdown() {
close(fbm.shutdownChan)
fbm.cancelArchive()
fbm.cancelBlocksToDelete()
fbm.cancelReclamation()
fbm.cancelCleanDiskCache()
}
// cleanUpBlockState cleans up any blocks that may have been orphaned
// by a failure during or after blocks have been sent to the
// server. This is usually used in a defer right before a call to
// fbo.doBlockPuts like so:
//
// defer func() {
// if err != nil {
// ...cleanUpBlockState(md.ReadOnly(), bps)
// }
// }()
//
// ... = ...doBlockPuts(ctx, md.ReadOnly(), *bps)
//
// The exception is for when blocks might get reused across multiple
// attempts at the same operation (like for a Sync). In that case,
// failed blocks should be built up in a separate data structure, and
// this should be called when the operation finally succeeds.
func (fbm *folderBlockManager) cleanUpBlockState(
md ReadOnlyRootMetadata, bps blockPutState, bdType blockDeleteType) {
fbm.log.CDebugf(
context.TODO(), "Clean up md %d %s, bdType=%d", md.Revision(),
md.MergedStatus(), bdType)
expBackoff := backoff.NewExponentialBackOff()
// Never give up when trying to delete blocks; it might just take
// a long time to confirm with the server whether a revision
// succeeded or not.
expBackoff.MaxElapsedTime = 0
toDelete := blocksToDelete{
md: md,
bdType: bdType,
backoff: expBackoff,
}
toDelete.blocks = append(toDelete.blocks, bps.Ptrs()...)
fbm.enqueueBlocksToDelete(toDelete)
}
func (fbm *folderBlockManager) enqueueBlocksToDelete(toDelete blocksToDelete) {
fbm.blocksToDeleteWaitGroup.Add(1)
fbm.blocksToDeleteChan <- toDelete
}
func (fbm *folderBlockManager) enqueueBlocksToDeleteAfterShortDelay(
ctx context.Context, toDelete blocksToDelete) {
fbm.blocksToDeleteWaitGroup.Add(1)
duration := toDelete.backoff.NextBackOff()
if duration == backoff.Stop {
panic(fmt.Sprintf("Backoff stopped while checking whether we "+
"should delete revision %d", toDelete.md.Revision()))
}
time.AfterFunc(duration,
func() {
select {
case fbm.blocksToDeleteChan <- toDelete:
case <-fbm.shutdownChan:
fbm.blocksToDeleteWaitGroup.Done()
}
})
}
// enqueueBlocksToDeleteNoWait enqueues blocks to be deleted just like
// enqueueBlocksToDelete, except that when fbm.blocksToDeleteChan is full, it
// doesn't block, but instead spawns a goroutine to handle the sending.
//
// This is necessary to prevent a situation like following:
// 1. A delete fails when fbm.blocksToDeleteChan is full
// 2. The goroutine tries to put the failed toDelete back to
// fbm.blocksToDeleteChan
// 3. Step 2 becomes synchronous and is blocked because
// fbm.blocksToDeleteChan is already full
// 4. fbm.blocksToDeleteChan never gets drained because the goroutine that
// drains it is waiting for sending on the same channel.
// 5. Deadlock!
func (fbm *folderBlockManager) enqueueBlocksToDeleteNoWait(toDelete blocksToDelete) {
fbm.blocksToDeleteWaitGroup.Add(1)
select {
case fbm.blocksToDeleteChan <- toDelete:
return
default:
go func() { fbm.blocksToDeleteChan <- toDelete }()
}
}
func isArchivableOp(op op) bool {
switch op.(type) {
case *createOp:
return true
case *rmOp:
return true
case *renameOp:
return true
case *syncOp:
return true
case *setAttrOp:
return true
case *resolutionOp:
return true
default:
// rekey ops don't have anything to archive, and gc
// ops only have deleted blocks.
return false
}
}
func isArchivableMDOrError(md ReadOnlyRootMetadata) error {
if md.MergedStatus() != kbfsmd.Merged {
return fmt.Errorf("md rev=%d is not merged", md.Revision())
}
for _, op := range md.data.Changes.Ops {
if !isArchivableOp(op) {
return fmt.Errorf(
"md rev=%d has unarchivable op %s",
md.Revision(), op)
}
}
return nil
}
func (fbm *folderBlockManager) archiveUnrefBlocks(md ReadOnlyRootMetadata) {
// Don't archive for unmerged revisions, because conflict
// resolution might undo some of the unreferences.
if md.MergedStatus() != kbfsmd.Merged {
return
}
if err := isArchivableMDOrError(md); err != nil {
panic(err)
}
fbm.archiveGroup.Add(1)
fbm.archiveChan <- md
}
// archiveUnrefBlocksNoWait enqueues the MD for archiving without
// blocking. By the time it returns, the archive group has been
// incremented so future waits will block on this archive. This
// method is for internal use within folderBlockManager only.
func (fbm *folderBlockManager) archiveUnrefBlocksNoWait(md ReadOnlyRootMetadata) {
// Don't archive for unmerged revisions, because conflict
// resolution might undo some of the unreferences.
if md.MergedStatus() != kbfsmd.Merged {
return
}
if err := isArchivableMDOrError(md); err != nil {
panic(err)
}
fbm.archiveGroup.Add(1)
// Don't block if the channel is full; instead do the send in a
// background goroutine. We've already done the Add above, so the
// wait calls should all work just fine.
select {
case fbm.archiveChan <- md:
return
default:
go func() { fbm.archiveChan <- md }()
}
}
func (fbm *folderBlockManager) waitForArchives(ctx context.Context) error {
return fbm.archiveGroup.Wait(ctx)
}
func (fbm *folderBlockManager) waitForDeletingBlocks(ctx context.Context) error {
return fbm.blocksToDeleteWaitGroup.Wait(ctx)
}
func (fbm *folderBlockManager) waitForQuotaReclamations(
ctx context.Context) error {
return fbm.reclamationGroup.Wait(ctx)
}
func (fbm *folderBlockManager) waitForDiskCacheCleans(
ctx context.Context) error {
return fbm.cleanDiskCachesGroup.Wait(ctx)
}
func (fbm *folderBlockManager) forceQuotaReclamation() {
fbm.reclamationGroup.Add(1)
select {
case fbm.forceReclamationChan <- struct{}{}:
default:
fbm.reclamationGroup.Done()
}
}
// doChunkedDowngrades sends batched archive or delete messages to the
// block server for the given block pointers. For deletes, it returns
// a list of block IDs that no longer have any references.
func (fbm *folderBlockManager) doChunkedDowngrades(ctx context.Context,
tlfID tlf.ID, ptrs []data.BlockPointer, archive bool) (
[]kbfsblock.ID, error) {
fbm.log.CDebugf(ctx, "Downgrading %d pointers (archive=%t)",
len(ptrs), archive)
bops := fbm.config.BlockOps()
// Round up to find the number of chunks.
numChunks := (len(ptrs) + numPointersToDowngradePerChunk - 1) /
numPointersToDowngradePerChunk
numWorkers := numChunks
if numWorkers > maxParallelBlockPuts {
numWorkers = maxParallelBlockPuts
}
chunks := make(chan []data.BlockPointer, numChunks)
var wg sync.WaitGroup
defer wg.Wait()
ctx, cancel := context.WithCancel(ctx)
defer cancel()
type workerResult struct {
zeroRefCounts []kbfsblock.ID
err error
}
chunkResults := make(chan workerResult, numChunks)
worker := func() {
defer wg.Done()
for chunk := range chunks {
var res workerResult
fbm.log.CDebugf(ctx, "Downgrading chunk of %d pointers", len(chunk))
if archive {
res.err = bops.Archive(ctx, tlfID, chunk)
} else {
var liveCounts map[kbfsblock.ID]int
liveCounts, res.err = bops.Delete(ctx, tlfID, chunk)
if res.err == nil {
for id, count := range liveCounts {
if count == 0 {
res.zeroRefCounts = append(res.zeroRefCounts, id)
}
}
}
}
chunkResults <- res
select {
// return early if the context has been canceled
case <-ctx.Done():
return
default:
}
}
}
for i := 0; i < numWorkers; i++ {
wg.Add(1)
go worker()
}
for start := 0; start < len(ptrs); start += numPointersToDowngradePerChunk {
end := start + numPointersToDowngradePerChunk
if end > len(ptrs) {
end = len(ptrs)
}
chunks <- ptrs[start:end]
}
close(chunks)
var zeroRefCounts []kbfsblock.ID
for i := 0; i < numChunks; i++ {
result := <-chunkResults
if result.err != nil {
// deferred cancel will stop the other workers.
return nil, result.err
}
zeroRefCounts = append(zeroRefCounts, result.zeroRefCounts...)
}
return zeroRefCounts, nil
}
// deleteBlockRefs sends batched delete messages to the block server
// for the given block pointers. It returns a list of block IDs that
// no longer have any references.
func (fbm *folderBlockManager) deleteBlockRefs(ctx context.Context,
tlfID tlf.ID, ptrs []data.BlockPointer) ([]kbfsblock.ID, error) {
return fbm.doChunkedDowngrades(ctx, tlfID, ptrs, false)
}
func (fbm *folderBlockManager) processBlocksToDelete(ctx context.Context, toDelete blocksToDelete) error {
// also attempt to delete any error references
defer fbm.blocksToDeleteWaitGroup.Done()
// Make sure all blocks in the journal (if journaling is enabled)
// are flushed before attempting to delete any of them.
if jManager, err := GetJournalManager(fbm.config); err == nil {
fbm.log.CDebugf(ctx, "Waiting for journal to flush")
if err := jManager.WaitForCompleteFlush(ctx, fbm.id); err != nil {
return err
}
}
fbm.log.CDebugf(ctx, "Checking deleted blocks for revision %d",
toDelete.md.Revision())
// Make sure that the MD didn't actually become part of the folder
// history. (This could happen if the Sync was canceled while the
// MD put was outstanding.) If the private MD is not set, there's
// no way the revision made it to the server, so we are free to
// clean it up without checking with the server.
if toDelete.bdType == blockDeleteOnMDFail &&
toDelete.md.bareMd.GetSerializedPrivateMetadata() != nil {
// Don't use `getSingleMD` here, since it returns an error if
// the revision isn't found, and that's useful information for
// us here.
rmds, err := getMDRange(
ctx, fbm.config, fbm.id, toDelete.md.BID(), toDelete.md.Revision(),
toDelete.md.Revision(), toDelete.md.MergedStatus(), nil)
if err != nil {
fbm.log.CDebugf(ctx,
"Error trying to get MD %d; retrying after a delay",
toDelete.md.Revision())
// We don't know whether or not the revision made it to
// the server, so try again. But don't re-enqueue
// immediately to avoid fast infinite loops.
fbm.enqueueBlocksToDeleteAfterShortDelay(ctx, toDelete)
return nil
}
var rmd ImmutableRootMetadata
if len(rmds) == 0 {
// The rmd.mdID check below will fail intentionally since
// rmd is empty. Note that this assumes that the MD
// servers don't cache negative lookups, or if they do,
// they use synchronous cache invalidations for that case.
// If we ever allow MD servers to cache negative lookups,
// we'll have to retry here for at least the amount of the
// maximum allowable cache timeout.
fbm.log.CDebugf(ctx, "No revision %d found on MD server, so we "+
"can safely archive", toDelete.md.Revision())
} else {
rmd = rmds[0]
}
mdID, err := kbfsmd.MakeID(fbm.config.Codec(), toDelete.md.bareMd)
if err != nil {
fbm.log.CErrorf(ctx, "Error when comparing dirs: %v", err)
} else if mdID == rmd.mdID {
if err := isArchivableMDOrError(rmd.ReadOnly()); err != nil {
fbm.log.CDebugf(ctx, "Skipping archiving for non-deleted, "+
"unarchivable revision %d: %v", rmd.Revision(), err)
return nil
}
// This md is part of the history of the folder, so we
// shouldn't delete the blocks. But, since this MD put
// seems to have succeeded, we should archive it.
fbm.log.CDebugf(ctx, "Not deleting blocks from revision %d; "+
"archiving it", rmd.Revision())
// Don't block on archiving the MD, because that could
// lead to deadlock.
fbm.archiveUnrefBlocksNoWait(rmd.ReadOnly())
return nil
}
// Otherwise something else has been written over
// this MD, so get rid of the blocks.
fbm.log.CDebugf(ctx, "Cleaning up blocks for failed revision %d",
toDelete.md.Revision())
} else {
fbm.log.CDebugf(ctx, "Cleaning up blocks for revision %d",
toDelete.md.Revision())
}
_, err := fbm.deleteBlockRefs(ctx, toDelete.md.TlfID(), toDelete.blocks)
// Ignore permanent errors
_, isPermErr := err.(kbfsblock.ServerError)
_, isNonceNonExistentErr := err.(kbfsblock.ServerErrorNonceNonExistent)
_, isBadRequestErr := err.(kbfsblock.ServerErrorBadRequest)
if err != nil {
fbm.log.CWarningf(ctx, "Couldn't delete some ref in batch %v: %v",
toDelete.blocks, err)
if !isPermErr && !isNonceNonExistentErr && !isBadRequestErr {
fbm.enqueueBlocksToDeleteNoWait(toDelete)
return nil
}
}
return nil
}
// CtxFBMTagKey is the type used for unique context tags within
// folderBlockManager
type CtxFBMTagKey int
const (
// CtxFBMIDKey is the type of the tag for unique operation IDs
// within folderBlockManager.
CtxFBMIDKey CtxFBMTagKey = iota
)
// CtxFBMOpID is the display name for the unique operation
// folderBlockManager ID tag.
const CtxFBMOpID = "FBMID"
func (fbm *folderBlockManager) ctxWithFBMID(
ctx context.Context) context.Context {
return CtxWithRandomIDReplayable(ctx, CtxFBMIDKey, CtxFBMOpID, fbm.log)
}
// Run the passed function with a context that's canceled on shutdown.
func (fbm *folderBlockManager) runUnlessShutdownWithCtx(
ctx context.Context, fn func(ctx context.Context) error) error {
ctx, cancelFunc := context.WithCancel(ctx)
defer cancelFunc()
errChan := make(chan error, 1)
go func() {
errChan <- fn(ctx)
}()
select {
case err := <-errChan:
return err
case <-fbm.shutdownChan:
return errors.New("shutdown received")
}
}
// Run the passed function with a context that's canceled on shutdown.
func (fbm *folderBlockManager) runUnlessShutdown(
fn func(ctx context.Context) error) error {
ctx := fbm.ctxWithFBMID(context.Background())
return fbm.runUnlessShutdownWithCtx(ctx, fn)
}
func (fbm *folderBlockManager) archiveBlockRefs(ctx context.Context,
tlfID tlf.ID, ptrs []data.BlockPointer) error {
_, err := fbm.doChunkedDowngrades(ctx, tlfID, ptrs, true)
return err
}
type unrefIterator struct {
nextPtr int
}
// getUnrefPointersFromMD returns a slice of BlockPointers that were
// unreferenced by the given `rmd`. If there are too many pointers to
// process, given the current mode, then it will return a partial
// list, plus a non-nil `iter` parameter that can be passed into a
// subsequent call to get the next set of unreferenced BlockPointers
// from the same MD. If a nil `iter` is given, pointers are returned
// from the beginning of the list.
func (fbm *folderBlockManager) getUnrefPointersFromMD(
rmd ReadOnlyRootMetadata, includeGC bool, iter *unrefIterator) (
ptrs []data.BlockPointer, nextIter *unrefIterator) {
currPtr := 0
complete := true
nextPtr := 0
if iter != nil {
nextPtr = iter.nextPtr
}
ptrMap := make(map[data.BlockPointer]bool)
max := fbm.config.Mode().MaxBlockPtrsToManageAtOnce()
opLoop:
for _, op := range rmd.data.Changes.Ops {
if _, ok := op.(*GCOp); !includeGC && ok {
continue
}
for _, ptr := range op.Unrefs() {
currPtr++
// Skip past any ptrs we've already processed.
if currPtr <= nextPtr {
continue
}
// Can be zeroPtr in weird failed sync scenarios.
// See syncInfo.replaceRemovedBlock for an example
// of how this can happen.
if ptr != data.ZeroPtr && !ptrMap[ptr] {
ptrMap[ptr] = true
}
nextPtr++
if max >= 0 && len(ptrMap) >= max {
complete = false
break opLoop
}
}
for _, update := range op.allUpdates() {
currPtr++
// Skip past any ptrs we've already processed.
if currPtr <= nextPtr {
continue
}
// It's legal for there to be an "update" between
// two identical pointers (usually because of
// conflict resolution), so ignore that for quota
// reclamation purposes.
if update.Ref != update.Unref && !ptrMap[update.Unref] {
ptrMap[update.Unref] = true
}
nextPtr++
if max >= 0 && len(ptrMap) >= max {
complete = false
break opLoop
}
}
}
ptrs = make([]data.BlockPointer, 0, len(ptrMap))
for ptr := range ptrMap {
ptrs = append(ptrs, ptr)
}
if !complete {
nextIter = &unrefIterator{nextPtr}
}
return ptrs, nextIter
}
func (fbm *folderBlockManager) archiveAllBlocksInMD(md ReadOnlyRootMetadata) {
// This func doesn't take any locks, though it can
// block md writes due to the buffered channel.
// So use the long timeout to make sure things get
// unblocked eventually, but no need for a short
// timeout.
ctx := fbm.ctxWithFBMID(context.Background())
ctx, cancel := context.WithTimeout(ctx, data.BackgroundTaskTimeout)
fbm.setArchiveCancel(cancel)
defer fbm.cancelArchive()
iter := &unrefIterator{0}
defer fbm.archiveGroup.Done()
for iter != nil {
var ptrs []data.BlockPointer
ptrs, iter = fbm.getUnrefPointersFromMD(md, true, iter)
_ = fbm.runUnlessShutdownWithCtx(
ctx, func(ctx context.Context) (err error) {
fbm.log.CDebugf(
ctx, "Archiving %d block pointers as a result "+
"of revision %d", len(ptrs), md.Revision())
err = fbm.archiveBlockRefs(ctx, md.TlfID(), ptrs)
if err != nil {
fbm.log.CWarningf(
ctx, "Couldn't archive blocks: %v", err)
return err
}
return nil
})
if iter != nil {
fbm.log.CDebugf(
ctx, "Archived %d pointers for revision %d, "+
"now looking for more", len(ptrs), md.Revision())
}
}
}
func (fbm *folderBlockManager) archiveBlocksInBackground() {
for {
select {
case md := <-fbm.archiveChan:
fbm.archiveAllBlocksInMD(md)
case unpause := <-fbm.archivePauseChan:
_ = fbm.runUnlessShutdown(func(ctx context.Context) (err error) {
fbm.log.CInfof(ctx, "Archives paused")
// wait to be unpaused
select {
case <-unpause:
fbm.log.CInfof(ctx, "Archives unpaused")
case <-ctx.Done():
return ctx.Err()
}
return nil
})
case <-fbm.shutdownChan:
return
}
}
}
func (fbm *folderBlockManager) deleteBlocksInBackground() {
for {
select {
case toDelete := <-fbm.blocksToDeleteChan:
_ = fbm.runUnlessShutdown(func(ctx context.Context) (err error) {
ctx, cancel := context.WithTimeout(
ctx, data.BackgroundTaskTimeout)
fbm.setBlocksToDeleteCancel(cancel)
defer fbm.cancelBlocksToDelete()
if err := fbm.processBlocksToDelete(ctx, toDelete); err != nil {
fbm.log.CDebugf(ctx, "Error deleting blocks: %v", err)
return err
}
return nil
})
case unpause := <-fbm.blocksToDeletePauseChan:
_ = fbm.runUnlessShutdown(func(ctx context.Context) (err error) {
fbm.log.CInfof(ctx, "deleteBlocks paused")
select {
case <-unpause:
fbm.log.CInfof(ctx, "deleteBlocks unpaused")
case <-ctx.Done():
return ctx.Err()
}
return nil
})
case <-fbm.shutdownChan:
return
}
}
}
func (fbm *folderBlockManager) isOldEnough(rmd ImmutableRootMetadata) bool {
// Trust the server's timestamp on this MD.
mtime := rmd.localTimestamp
unrefAge := fbm.config.Mode().QuotaReclamationMinUnrefAge()
return mtime.Add(unrefAge).Before(fbm.config.Clock().Now())
}
// getMostRecentGCRevision returns the latest revision that was
// scrubbed by the previous gc op.
func (fbm *folderBlockManager) getMostRecentGCRevision(
ctx context.Context, head ReadOnlyRootMetadata) (
lastGCRev kbfsmd.Revision, err error) {
if head.data.LastGCRevision >= kbfsmd.RevisionInitial {
fbm.log.CDebugf(ctx, "Found last gc revision %d in "+
"head MD revision %d", head.data.LastGCRevision,
head.Revision())
return head.data.LastGCRevision, nil
}
// Very old TLFs might not have a filled-in `LastGCRevision`, so
// we need to walk backwards to find the latest gcOp.
endRev := head.Revision()
for {
startRev := endRev - maxMDsAtATime + 1 // (kbfsmd.Revision is signed)
if startRev < kbfsmd.RevisionInitial {
startRev = kbfsmd.RevisionInitial
}
rmds, err := getMDRange(
ctx, fbm.config, fbm.id, kbfsmd.NullBranchID, startRev,
endRev, kbfsmd.Merged, nil)
if err != nil {
return kbfsmd.RevisionUninitialized, err
}
numNew := len(rmds)
for i := len(rmds) - 1; i >= 0; i-- {
rmd := rmds[i]
if rmd.data.LastGCRevision >= kbfsmd.RevisionInitial {
fbm.log.CDebugf(ctx, "Found last gc revision %d in "+
"MD revision %d", rmd.data.LastGCRevision,
rmd.Revision())
return rmd.data.LastGCRevision, nil
}
for j := len(rmd.data.Changes.Ops) - 1; j >= 0; j-- {
GCOp, ok := rmd.data.Changes.Ops[j].(*GCOp)
if !ok || GCOp.LatestRev == kbfsmd.RevisionUninitialized {
continue
}
fbm.log.CDebugf(ctx, "Found last gc op: %s", GCOp)
return GCOp.LatestRev, nil
}
}
if numNew > 0 {
endRev = rmds[0].Revision() - 1
}
if numNew < maxMDsAtATime || endRev < kbfsmd.RevisionInitial {
// Never been GC'd.
return kbfsmd.RevisionUninitialized, nil
}
}
}
// getUnrefBlocks returns a slice containing all the block pointers
// that were unreferenced after the earliestRev, up to and including
// those in latestRev. If the number of pointers is too large, it
// will shorten the range of the revisions being reclaimed, and return
// the latest revision represented in the returned slice of pointers.
func (fbm *folderBlockManager) getUnreferencedBlocks(
ctx context.Context, earliestRev, mostRecentRev kbfsmd.Revision) (
ptrs []data.BlockPointer, lastRev kbfsmd.Revision,
complete bool, err error) {
fbm.log.CDebugf(ctx, "Getting unreferenced blocks between revisions "+
"%d and %d", earliestRev, mostRecentRev)
defer func() {
if err == nil {
fbm.log.CDebugf(ctx, "Found %d pointers to clean between "+
"revisions %d and %d", len(ptrs), earliestRev, lastRev)
}
}()
// Walk forward, starting from just after earliestRev, until we
// get enough pointers or until we reach the head or a revision
// that's not old enough, gathering pointers to GC.
startRev := earliestRev + 1
outer:
for {
endRev := startRev + maxMDsAtATime
if endRev > mostRecentRev {
endRev = mostRecentRev
}
rmds, err := getMDRange(
ctx, fbm.config, fbm.id, kbfsmd.NullBranchID, startRev,
endRev, kbfsmd.Merged, nil)
if err != nil {
return nil, kbfsmd.RevisionUninitialized, false, err
}
numNew := len(rmds)
for _, rmd := range rmds {
if !fbm.isOldEnough(rmd) {
fbm.log.CDebugf(ctx, "Revision %d is too recent; stopping QR",
rmd.Revision())
complete = true
break outer
}
lastRev = rmd.Revision()
// A garbage-collection op *must* contain all pointers in
// its respective op. If this device can't handle it,
// error the process and let another device take care of
// it.
newPtrs, iter := fbm.getUnrefPointersFromMD(
rmd.ReadOnlyRootMetadata, false, &unrefIterator{0})
if iter != nil {
return nil, kbfsmd.RevisionUninitialized, false, errors.New(
fmt.Sprintf(
"Can't handle the unref'd pointers of revision %d",
lastRev))
}
ptrs = append(ptrs, newPtrs...)
// TODO: when can we clean up the MD's unembedded block
// changes pointer? It's not safe until we know for sure
// that all existing clients have received the latest
// update (and also that there are no outstanding staged
// branches). Let's do that as part of the bigger issue
// KBFS-793 -- for now we have to leak those blocks.
if len(ptrs) > fbm.numPointersPerGCThreshold {
fbm.log.CDebugf(ctx, "Shortening GC range to [%d:%d]",
earliestRev, rmd.Revision())
break outer
}
}
if numNew > 0 {