summaryrefslogtreecommitdiffhomepage
path: root/pkg/sentry/fsimpl/gofer/filesystem.go
blob: 237d179212f6a019dea0450de9b0b6050e6a55f6 (plain)
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
1001
1002
1003
1004
1005
1006
1007
1008
1009
1010
1011
1012
1013
1014
1015
1016
1017
1018
1019
1020
1021
1022
1023
1024
1025
1026
1027
1028
1029
1030
1031
1032
1033
1034
1035
1036
1037
1038
1039
1040
1041
1042
1043
1044
1045
1046
1047
1048
1049
1050
1051
1052
1053
1054
1055
1056
1057
1058
1059
1060
1061
1062
1063
1064
1065
1066
1067
1068
1069
1070
1071
1072
1073
1074
1075
1076
1077
1078
1079
1080
1081
1082
1083
1084
1085
1086
1087
1088
1089
1090
1091
1092
1093
1094
1095
1096
1097
1098
1099
1100
1101
1102
1103
1104
1105
1106
1107
1108
1109
1110
1111
1112
1113
1114
1115
1116
1117
1118
1119
1120
1121
1122
1123
1124
1125
1126
1127
1128
1129
1130
1131
1132
1133
1134
1135
1136
1137
1138
1139
1140
1141
1142
1143
1144
1145
1146
1147
1148
1149
1150
1151
1152
1153
1154
1155
1156
1157
1158
1159
1160
1161
1162
1163
1164
1165
1166
1167
1168
1169
1170
1171
1172
1173
1174
1175
1176
1177
1178
1179
1180
1181
1182
1183
1184
1185
1186
1187
1188
1189
1190
1191
1192
1193
1194
1195
1196
1197
1198
1199
1200
1201
1202
1203
1204
1205
1206
1207
1208
1209
1210
1211
1212
1213
1214
1215
1216
1217
1218
1219
1220
1221
1222
1223
1224
1225
1226
1227
1228
1229
1230
1231
1232
1233
1234
1235
1236
1237
1238
1239
1240
1241
1242
1243
1244
1245
1246
1247
1248
1249
1250
1251
1252
1253
1254
1255
1256
1257
1258
1259
1260
1261
1262
1263
1264
1265
1266
1267
1268
1269
1270
1271
1272
1273
1274
1275
1276
1277
1278
1279
1280
1281
1282
1283
1284
1285
1286
1287
1288
1289
1290
1291
1292
1293
1294
1295
1296
1297
1298
1299
1300
1301
1302
1303
1304
1305
1306
1307
1308
1309
1310
1311
1312
1313
1314
1315
1316
1317
1318
1319
1320
1321
1322
1323
1324
1325
1326
1327
1328
1329
1330
1331
1332
1333
1334
1335
1336
1337
1338
1339
1340
1341
1342
1343
1344
1345
1346
1347
1348
1349
1350
1351
1352
1353
1354
1355
1356
1357
1358
1359
1360
1361
1362
1363
1364
1365
1366
1367
1368
1369
1370
1371
1372
1373
1374
1375
1376
1377
1378
1379
1380
1381
1382
1383
1384
1385
1386
1387
1388
1389
1390
1391
1392
1393
1394
1395
1396
1397
1398
1399
1400
1401
1402
1403
1404
1405
1406
1407
1408
1409
1410
1411
1412
1413
1414
1415
1416
1417
1418
1419
1420
1421
1422
1423
1424
1425
1426
1427
1428
1429
1430
1431
1432
1433
1434
1435
1436
1437
1438
1439
1440
1441
1442
1443
1444
1445
1446
1447
1448
1449
1450
1451
1452
1453
1454
1455
1456
1457
1458
1459
1460
1461
1462
1463
1464
1465
1466
1467
1468
1469
1470
1471
1472
1473
1474
1475
1476
1477
1478
1479
1480
1481
1482
1483
1484
1485
1486
1487
1488
1489
1490
1491
1492
1493
1494
1495
1496
1497
1498
1499
1500
1501
1502
1503
1504
1505
1506
1507
1508
1509
1510
1511
1512
1513
1514
1515
1516
1517
1518
1519
1520
1521
1522
1523
1524
1525
1526
1527
1528
1529
1530
1531
1532
1533
1534
1535
1536
1537
1538
1539
1540
1541
1542
1543
1544
1545
1546
1547
1548
1549
1550
1551
1552
1553
1554
1555
1556
1557
1558
1559
1560
1561
1562
1563
1564
1565
1566
1567
1568
1569
1570
1571
1572
1573
1574
1575
1576
1577
1578
1579
1580
1581
1582
1583
1584
1585
1586
1587
1588
1589
1590
1591
1592
1593
1594
1595
1596
1597
1598
1599
1600
1601
1602
1603
1604
1605
1606
1607
1608
1609
1610
1611
1612
1613
1614
1615
1616
1617
1618
1619
1620
1621
1622
1623
1624
1625
1626
1627
1628
1629
1630
1631
1632
1633
1634
1635
1636
// Copyright 2019 The gVisor Authors.
//
// Licensed under the Apache License, Version 2.0 (the "License");
// you may not use this file except in compliance with the License.
// You may obtain a copy of the License at
//
//     http://www.apache.org/licenses/LICENSE-2.0
//
// Unless required by applicable law or agreed to in writing, software
// distributed under the License is distributed on an "AS IS" BASIS,
// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
// See the License for the specific language governing permissions and
// limitations under the License.

package gofer

import (
	"fmt"
	"math"
	"strings"
	"sync"
	"sync/atomic"

	"gvisor.dev/gvisor/pkg/abi/linux"
	"gvisor.dev/gvisor/pkg/context"
	"gvisor.dev/gvisor/pkg/errors/linuxerr"
	"gvisor.dev/gvisor/pkg/fspath"
	"gvisor.dev/gvisor/pkg/p9"
	"gvisor.dev/gvisor/pkg/sentry/fsimpl/host"
	"gvisor.dev/gvisor/pkg/sentry/fsmetric"
	"gvisor.dev/gvisor/pkg/sentry/kernel"
	"gvisor.dev/gvisor/pkg/sentry/kernel/auth"
	"gvisor.dev/gvisor/pkg/sentry/kernel/pipe"
	"gvisor.dev/gvisor/pkg/sentry/socket/unix/transport"
	"gvisor.dev/gvisor/pkg/sentry/vfs"
	"gvisor.dev/gvisor/pkg/syserror"
)

// Sync implements vfs.FilesystemImpl.Sync.
func (fs *filesystem) Sync(ctx context.Context) error {
	// Snapshot current syncable dentries and special file FDs.
	fs.renameMu.RLock()
	fs.syncMu.Lock()
	ds := make([]*dentry, 0, len(fs.syncableDentries))
	for d := range fs.syncableDentries {
		// It's safe to use IncRef here even though fs.syncableDentries doesn't
		// hold references since we hold fs.renameMu. Note that we can't use
		// TryIncRef since cached dentries at zero references should still be
		// synced.
		d.IncRef()
		ds = append(ds, d)
	}
	fs.renameMu.RUnlock()
	sffds := make([]*specialFileFD, 0, len(fs.specialFileFDs))
	for sffd := range fs.specialFileFDs {
		// As above, fs.specialFileFDs doesn't hold references. However, unlike
		// dentries, an FD that has reached zero references can't be
		// resurrected, so we can use TryIncRef.
		if sffd.vfsfd.TryIncRef() {
			sffds = append(sffds, sffd)
		}
	}
	fs.syncMu.Unlock()

	// Return the first error we encounter, but sync everything we can
	// regardless.
	var retErr error

	// Sync syncable dentries.
	for _, d := range ds {
		err := d.syncCachedFile(ctx, true /* forFilesystemSync */)
		d.DecRef(ctx)
		if err != nil {
			ctx.Infof("gofer.filesystem.Sync: dentry.syncCachedFile failed: %v", err)
			if retErr == nil {
				retErr = err
			}
		}
	}

	// Sync special files, which may be writable but do not use dentry shared
	// handles (so they won't be synced by the above).
	for _, sffd := range sffds {
		err := sffd.sync(ctx, true /* forFilesystemSync */)
		sffd.vfsfd.DecRef(ctx)
		if err != nil {
			ctx.Infof("gofer.filesystem.Sync: specialFileFD.sync failed: %v", err)
			if retErr == nil {
				retErr = err
			}
		}
	}

	return retErr
}

// maxFilenameLen is the maximum length of a filename. This is dictated by 9P's
// encoding of strings, which uses 2 bytes for the length prefix.
const maxFilenameLen = (1 << 16) - 1

// dentrySlicePool is a pool of *[]*dentry used to store dentries for which
// dentry.checkCachingLocked() must be called. The pool holds pointers to
// slices because Go lacks generics, so sync.Pool operates on interface{}, so
// every call to (what should be) sync.Pool<[]*dentry>.Put() allocates a copy
// of the slice header on the heap.
var dentrySlicePool = sync.Pool{
	New: func() interface{} {
		ds := make([]*dentry, 0, 4) // arbitrary non-zero initial capacity
		return &ds
	},
}

func appendDentry(ds *[]*dentry, d *dentry) *[]*dentry {
	if ds == nil {
		ds = dentrySlicePool.Get().(*[]*dentry)
	}
	*ds = append(*ds, d)
	return ds
}

// Precondition: !parent.isSynthetic() && !child.isSynthetic().
func appendNewChildDentry(ds **[]*dentry, parent *dentry, child *dentry) {
	// The new child was added to parent and took a ref on the parent (hence
	// parent can be removed from cache). A new child has 0 refs for now. So
	// checkCachingLocked() should be called on both. Call it first on the parent
	// as it may create space in the cache for child to be inserted - hence
	// avoiding a cache eviction.
	*ds = appendDentry(*ds, parent)
	*ds = appendDentry(*ds, child)
}

// Preconditions: ds != nil.
func putDentrySlice(ds *[]*dentry) {
	// Allow dentries to be GC'd.
	for i := range *ds {
		(*ds)[i] = nil
	}
	*ds = (*ds)[:0]
	dentrySlicePool.Put(ds)
}

// renameMuRUnlockAndCheckCaching calls fs.renameMu.RUnlock(), then calls
// dentry.checkCachingLocked on all dentries in *dsp with fs.renameMu locked
// for writing.
//
// dsp is a pointer-to-pointer since defer evaluates its arguments immediately,
// but dentry slices are allocated lazily, and it's much easier to say "defer
// fs.renameMuRUnlockAndCheckCaching(&ds)" than "defer func() {
// fs.renameMuRUnlockAndCheckCaching(ds) }()" to work around this.
func (fs *filesystem) renameMuRUnlockAndCheckCaching(ctx context.Context, dsp **[]*dentry) {
	fs.renameMu.RUnlock()
	if *dsp == nil {
		return
	}
	ds := **dsp
	for _, d := range ds {
		d.checkCachingLocked(ctx, false /* renameMuWriteLocked */)
	}
	putDentrySlice(*dsp)
}

func (fs *filesystem) renameMuUnlockAndCheckCaching(ctx context.Context, ds **[]*dentry) {
	if *ds == nil {
		fs.renameMu.Unlock()
		return
	}
	for _, d := range **ds {
		d.checkCachingLocked(ctx, true /* renameMuWriteLocked */)
	}
	fs.renameMu.Unlock()
	putDentrySlice(*ds)
}

// stepLocked resolves rp.Component() to an existing file, starting from the
// given directory.
//
// Dentries which may become cached as a result of the traversal are appended
// to *ds.
//
// Preconditions:
// * fs.renameMu must be locked.
// * d.dirMu must be locked.
// * !rp.Done().
// * If !d.cachedMetadataAuthoritative(), then d and all children that are
//   part of rp must have been revalidated.
//
// Postconditions: The returned dentry's cached metadata is up to date.
func (fs *filesystem) stepLocked(ctx context.Context, rp *vfs.ResolvingPath, d *dentry, mayFollowSymlinks bool, ds **[]*dentry) (*dentry, bool, error) {
	if !d.isDir() {
		return nil, false, syserror.ENOTDIR
	}
	if err := d.checkPermissions(rp.Credentials(), vfs.MayExec); err != nil {
		return nil, false, err
	}
	followedSymlink := false
afterSymlink:
	name := rp.Component()
	if name == "." {
		rp.Advance()
		return d, followedSymlink, nil
	}
	if name == ".." {
		if isRoot, err := rp.CheckRoot(ctx, &d.vfsd); err != nil {
			return nil, false, err
		} else if isRoot || d.parent == nil {
			rp.Advance()
			return d, followedSymlink, nil
		}
		if err := rp.CheckMount(ctx, &d.parent.vfsd); err != nil {
			return nil, false, err
		}
		rp.Advance()
		return d.parent, followedSymlink, nil
	}
	child, err := fs.getChildLocked(ctx, d, name, ds)
	if err != nil {
		return nil, false, err
	}
	if err := rp.CheckMount(ctx, &child.vfsd); err != nil {
		return nil, false, err
	}
	if child.isSymlink() && mayFollowSymlinks && rp.ShouldFollowSymlink() {
		target, err := child.readlink(ctx, rp.Mount())
		if err != nil {
			return nil, false, err
		}
		if err := rp.HandleSymlink(target); err != nil {
			return nil, false, err
		}
		followedSymlink = true
		goto afterSymlink // don't check the current directory again
	}
	rp.Advance()
	return child, followedSymlink, nil
}

// getChildLocked returns a dentry representing the child of parent with the
// given name. Returns ENOENT if the child doesn't exist.
//
// Preconditions:
// * fs.renameMu must be locked.
// * parent.dirMu must be locked.
// * parent.isDir().
// * name is not "." or "..".
// * dentry at name has been revalidated
func (fs *filesystem) getChildLocked(ctx context.Context, parent *dentry, name string, ds **[]*dentry) (*dentry, error) {
	if len(name) > maxFilenameLen {
		return nil, linuxerr.ENAMETOOLONG
	}
	if child, ok := parent.children[name]; ok || parent.isSynthetic() {
		if child == nil {
			return nil, syserror.ENOENT
		}
		return child, nil
	}

	qid, file, attrMask, attr, err := parent.file.walkGetAttrOne(ctx, name)
	if err != nil {
		if linuxerr.Equals(linuxerr.ENOENT, err) {
			parent.cacheNegativeLookupLocked(name)
		}
		return nil, err
	}

	// Create a new dentry representing the file.
	child, err := fs.newDentry(ctx, file, qid, attrMask, &attr)
	if err != nil {
		file.close(ctx)
		delete(parent.children, name)
		return nil, err
	}
	parent.cacheNewChildLocked(child, name)
	appendNewChildDentry(ds, parent, child)
	return child, nil
}

// walkParentDirLocked resolves all but the last path component of rp to an
// existing directory, starting from the given directory (which is usually
// rp.Start().Impl().(*dentry)). It does not check that the returned directory
// is searchable by the provider of rp.
//
// Preconditions:
// * fs.renameMu must be locked.
// * !rp.Done().
// * If !d.cachedMetadataAuthoritative(), then d's cached metadata must be up
//   to date.
func (fs *filesystem) walkParentDirLocked(ctx context.Context, rp *vfs.ResolvingPath, d *dentry, ds **[]*dentry) (*dentry, error) {
	if err := fs.revalidateParentDir(ctx, rp, d, ds); err != nil {
		return nil, err
	}
	for !rp.Final() {
		d.dirMu.Lock()
		next, followedSymlink, err := fs.stepLocked(ctx, rp, d, true /* mayFollowSymlinks */, ds)
		d.dirMu.Unlock()
		if err != nil {
			return nil, err
		}
		d = next
		if followedSymlink {
			if err := fs.revalidateParentDir(ctx, rp, d, ds); err != nil {
				return nil, err
			}
		}
	}
	if !d.isDir() {
		return nil, syserror.ENOTDIR
	}
	return d, nil
}

// resolveLocked resolves rp to an existing file.
//
// Preconditions: fs.renameMu must be locked.
func (fs *filesystem) resolveLocked(ctx context.Context, rp *vfs.ResolvingPath, ds **[]*dentry) (*dentry, error) {
	d := rp.Start().Impl().(*dentry)
	if err := fs.revalidatePath(ctx, rp, d, ds); err != nil {
		return nil, err
	}
	for !rp.Done() {
		d.dirMu.Lock()
		next, followedSymlink, err := fs.stepLocked(ctx, rp, d, true /* mayFollowSymlinks */, ds)
		d.dirMu.Unlock()
		if err != nil {
			return nil, err
		}
		d = next
		if followedSymlink {
			if err := fs.revalidatePath(ctx, rp, d, ds); err != nil {
				return nil, err
			}
		}
	}
	if rp.MustBeDir() && !d.isDir() {
		return nil, syserror.ENOTDIR
	}
	return d, nil
}

// doCreateAt checks that creating a file at rp is permitted, then invokes
// createInRemoteDir (if the parent directory is a real remote directory) or
// createInSyntheticDir (if the parent directory is synthetic) to do so.
//
// Preconditions:
// * !rp.Done().
// * For the final path component in rp, !rp.ShouldFollowSymlink().
func (fs *filesystem) doCreateAt(ctx context.Context, rp *vfs.ResolvingPath, dir bool, createInRemoteDir func(parent *dentry, name string, ds **[]*dentry) error, createInSyntheticDir func(parent *dentry, name string) error) error {
	var ds *[]*dentry
	fs.renameMu.RLock()
	defer fs.renameMuRUnlockAndCheckCaching(ctx, &ds)
	start := rp.Start().Impl().(*dentry)
	parent, err := fs.walkParentDirLocked(ctx, rp, start, &ds)
	if err != nil {
		return err
	}

	// Order of checks is important. First check if parent directory can be
	// executed, then check for existence, and lastly check if mount is writable.
	if err := parent.checkPermissions(rp.Credentials(), vfs.MayExec); err != nil {
		return err
	}
	name := rp.Component()
	if name == "." || name == ".." {
		return syserror.EEXIST
	}
	if parent.isDeleted() {
		return syserror.ENOENT
	}
	if err := fs.revalidateOne(ctx, rp.VirtualFilesystem(), parent, name, &ds); err != nil {
		return err
	}

	parent.dirMu.Lock()
	defer parent.dirMu.Unlock()

	if len(name) > maxFilenameLen {
		return linuxerr.ENAMETOOLONG
	}
	// Check for existence only if caching information is available. Otherwise,
	// don't check for existence just yet. We will check for existence if the
	// checks for writability fail below. Existence check is done by the creation
	// RPCs themselves.
	if child, ok := parent.children[name]; ok && child != nil {
		return syserror.EEXIST
	}
	checkExistence := func() error {
		if child, err := fs.getChildLocked(ctx, parent, name, &ds); err != nil && !linuxerr.Equals(linuxerr.ENOENT, err) {
			return err
		} else if child != nil {
			return syserror.EEXIST
		}
		return nil
	}

	mnt := rp.Mount()
	if err := mnt.CheckBeginWrite(); err != nil {
		// Existence check takes precedence.
		if existenceErr := checkExistence(); existenceErr != nil {
			return existenceErr
		}
		return err
	}
	defer mnt.EndWrite()

	if err := parent.checkPermissions(rp.Credentials(), vfs.MayWrite); err != nil {
		// Existence check takes precedence.
		if existenceErr := checkExistence(); existenceErr != nil {
			return existenceErr
		}
		return err
	}
	if !dir && rp.MustBeDir() {
		return syserror.ENOENT
	}
	if parent.isSynthetic() {
		if createInSyntheticDir == nil {
			return linuxerr.EPERM
		}
		if err := createInSyntheticDir(parent, name); err != nil {
			return err
		}
		parent.touchCMtime()
		parent.dirents = nil
		ev := linux.IN_CREATE
		if dir {
			ev |= linux.IN_ISDIR
		}
		parent.watches.Notify(ctx, name, uint32(ev), 0, vfs.InodeEvent, false /* unlinked */)
		return nil
	}
	// No cached dentry exists; however, in InteropModeShared there might still be
	// an existing file at name. Just attempt the file creation RPC anyways. If a
	// file does exist, the RPC will fail with EEXIST like we would have.
	if err := createInRemoteDir(parent, name, &ds); err != nil {
		return err
	}
	if fs.opts.interop != InteropModeShared {
		if child, ok := parent.children[name]; ok && child == nil {
			// Delete the now-stale negative dentry.
			delete(parent.children, name)
		}
		parent.touchCMtime()
		parent.dirents = nil
	}
	ev := linux.IN_CREATE
	if dir {
		ev |= linux.IN_ISDIR
	}
	parent.watches.Notify(ctx, name, uint32(ev), 0, vfs.InodeEvent, false /* unlinked */)
	return nil
}

// Preconditions: !rp.Done().
func (fs *filesystem) unlinkAt(ctx context.Context, rp *vfs.ResolvingPath, dir bool) error {
	var ds *[]*dentry
	fs.renameMu.RLock()
	defer fs.renameMuRUnlockAndCheckCaching(ctx, &ds)
	start := rp.Start().Impl().(*dentry)
	parent, err := fs.walkParentDirLocked(ctx, rp, start, &ds)
	if err != nil {
		return err
	}
	if err := parent.checkPermissions(rp.Credentials(), vfs.MayWrite|vfs.MayExec); err != nil {
		return err
	}
	if err := rp.Mount().CheckBeginWrite(); err != nil {
		return err
	}
	defer rp.Mount().EndWrite()

	name := rp.Component()
	if dir {
		if name == "." {
			return linuxerr.EINVAL
		}
		if name == ".." {
			return linuxerr.ENOTEMPTY
		}
	} else {
		if name == "." || name == ".." {
			return syserror.EISDIR
		}
	}

	vfsObj := rp.VirtualFilesystem()
	if err := fs.revalidateOne(ctx, vfsObj, parent, rp.Component(), &ds); err != nil {
		return err
	}

	mntns := vfs.MountNamespaceFromContext(ctx)
	defer mntns.DecRef(ctx)

	parent.dirMu.Lock()
	defer parent.dirMu.Unlock()

	// Load child if sticky bit is set because we need to determine whether
	// deletion is allowed.
	var child *dentry
	if atomic.LoadUint32(&parent.mode)&linux.ModeSticky == 0 {
		var ok bool
		child, ok = parent.children[name]
		if ok && child == nil {
			// Hit a negative cached entry, child doesn't exist.
			return syserror.ENOENT
		}
	} else {
		child, _, err = fs.stepLocked(ctx, rp, parent, false /* mayFollowSymlinks */, &ds)
		if err != nil {
			return err
		}
		if err := parent.mayDelete(rp.Credentials(), child); err != nil {
			return err
		}
	}

	// If a child dentry exists, prepare to delete it. This should fail if it is
	// a mount point. We detect mount points by speculatively calling
	// PrepareDeleteDentry, which fails if child is a mount point.
	//
	// Also note that if child is nil, then it can't be a mount point.
	if child != nil {
		// Hold child.dirMu so we can check child.children and
		// child.syntheticChildren. We don't access these fields until a bit later,
		// but locking child.dirMu after calling vfs.PrepareDeleteDentry() would
		// create an inconsistent lock ordering between dentry.dirMu and
		// vfs.Dentry.mu (in the VFS lock order, it would make dentry.dirMu both "a
		// FilesystemImpl lock" and "a lock acquired by a FilesystemImpl between
		// PrepareDeleteDentry and CommitDeleteDentry). To avoid this, lock
		// child.dirMu before calling PrepareDeleteDentry.
		child.dirMu.Lock()
		defer child.dirMu.Unlock()
		if err := vfsObj.PrepareDeleteDentry(mntns, &child.vfsd); err != nil {
			return err
		}
	}
	flags := uint32(0)
	// If a dentry exists, use it for best-effort checks on its deletability.
	if dir {
		if child != nil {
			// child must be an empty directory.
			if child.syntheticChildren != 0 {
				// This is definitely not an empty directory, irrespective of
				// fs.opts.interop.
				vfsObj.AbortDeleteDentry(&child.vfsd)
				return linuxerr.ENOTEMPTY
			}
			// If InteropModeShared is in effect and the first call to
			// PrepareDeleteDentry above succeeded, then child wasn't
			// revalidated (so we can't expect its file type to be correct) and
			// individually revalidating its children (to confirm that they
			// still exist) would be a waste of time.
			if child.cachedMetadataAuthoritative() {
				if !child.isDir() {
					vfsObj.AbortDeleteDentry(&child.vfsd)
					return syserror.ENOTDIR
				}
				for _, grandchild := range child.children {
					if grandchild != nil {
						vfsObj.AbortDeleteDentry(&child.vfsd)
						return linuxerr.ENOTEMPTY
					}
				}
			}
		}
		flags = linux.AT_REMOVEDIR
	} else {
		// child must be a non-directory file.
		if child != nil && child.isDir() {
			vfsObj.AbortDeleteDentry(&child.vfsd)
			return syserror.EISDIR
		}
		if rp.MustBeDir() {
			if child != nil {
				vfsObj.AbortDeleteDentry(&child.vfsd)
			}
			return syserror.ENOTDIR
		}
	}
	if parent.isSynthetic() {
		if child == nil {
			return syserror.ENOENT
		}
	} else if child == nil || !child.isSynthetic() {
		err = parent.file.unlinkAt(ctx, name, flags)
		if err != nil {
			if child != nil {
				vfsObj.AbortDeleteDentry(&child.vfsd)
			}
			return err
		}
	}

	// Generate inotify events for rmdir or unlink.
	if dir {
		parent.watches.Notify(ctx, name, linux.IN_DELETE|linux.IN_ISDIR, 0, vfs.InodeEvent, true /* unlinked */)
	} else {
		var cw *vfs.Watches
		if child != nil {
			cw = &child.watches
		}
		vfs.InotifyRemoveChild(ctx, cw, &parent.watches, name)
	}

	if child != nil {
		vfsObj.CommitDeleteDentry(ctx, &child.vfsd)
		child.setDeleted()
		if child.isSynthetic() {
			parent.syntheticChildren--
			child.decRefNoCaching()
		}
		ds = appendDentry(ds, child)
	}
	parent.cacheNegativeLookupLocked(name)
	if parent.cachedMetadataAuthoritative() {
		parent.dirents = nil
		parent.touchCMtime()
		if dir {
			parent.decLinks()
		}
	}
	return nil
}

// AccessAt implements vfs.Filesystem.Impl.AccessAt.
func (fs *filesystem) AccessAt(ctx context.Context, rp *vfs.ResolvingPath, creds *auth.Credentials, ats vfs.AccessTypes) error {
	var ds *[]*dentry
	fs.renameMu.RLock()
	defer fs.renameMuRUnlockAndCheckCaching(ctx, &ds)
	d, err := fs.resolveLocked(ctx, rp, &ds)
	if err != nil {
		return err
	}
	return d.checkPermissions(creds, ats)
}

// GetDentryAt implements vfs.FilesystemImpl.GetDentryAt.
func (fs *filesystem) GetDentryAt(ctx context.Context, rp *vfs.ResolvingPath, opts vfs.GetDentryOptions) (*vfs.Dentry, error) {
	var ds *[]*dentry
	fs.renameMu.RLock()
	defer fs.renameMuRUnlockAndCheckCaching(ctx, &ds)
	d, err := fs.resolveLocked(ctx, rp, &ds)
	if err != nil {
		return nil, err
	}
	if opts.CheckSearchable {
		if !d.isDir() {
			return nil, syserror.ENOTDIR
		}
		if err := d.checkPermissions(rp.Credentials(), vfs.MayExec); err != nil {
			return nil, err
		}
	}
	d.IncRef()
	// Call d.checkCachingLocked() so it can be removed from the cache if needed.
	ds = appendDentry(ds, d)
	return &d.vfsd, nil
}

// GetParentDentryAt implements vfs.FilesystemImpl.GetParentDentryAt.
func (fs *filesystem) GetParentDentryAt(ctx context.Context, rp *vfs.ResolvingPath) (*vfs.Dentry, error) {
	var ds *[]*dentry
	fs.renameMu.RLock()
	defer fs.renameMuRUnlockAndCheckCaching(ctx, &ds)
	start := rp.Start().Impl().(*dentry)
	d, err := fs.walkParentDirLocked(ctx, rp, start, &ds)
	if err != nil {
		return nil, err
	}
	d.IncRef()
	// Call d.checkCachingLocked() so it can be removed from the cache if needed.
	ds = appendDentry(ds, d)
	return &d.vfsd, nil
}

// LinkAt implements vfs.FilesystemImpl.LinkAt.
func (fs *filesystem) LinkAt(ctx context.Context, rp *vfs.ResolvingPath, vd vfs.VirtualDentry) error {
	return fs.doCreateAt(ctx, rp, false /* dir */, func(parent *dentry, childName string, _ **[]*dentry) error {
		if rp.Mount() != vd.Mount() {
			return linuxerr.EXDEV
		}
		d := vd.Dentry().Impl().(*dentry)
		if d.isDir() {
			return linuxerr.EPERM
		}
		gid := auth.KGID(atomic.LoadUint32(&d.gid))
		uid := auth.KUID(atomic.LoadUint32(&d.uid))
		mode := linux.FileMode(atomic.LoadUint32(&d.mode))
		if err := vfs.MayLink(rp.Credentials(), mode, uid, gid); err != nil {
			return err
		}
		if d.nlink == 0 {
			return syserror.ENOENT
		}
		if d.nlink == math.MaxUint32 {
			return linuxerr.EMLINK
		}
		if err := parent.file.link(ctx, d.file, childName); err != nil {
			return err
		}

		// Success!
		atomic.AddUint32(&d.nlink, 1)
		return nil
	}, nil)
}

// MkdirAt implements vfs.FilesystemImpl.MkdirAt.
func (fs *filesystem) MkdirAt(ctx context.Context, rp *vfs.ResolvingPath, opts vfs.MkdirOptions) error {
	creds := rp.Credentials()
	return fs.doCreateAt(ctx, rp, true /* dir */, func(parent *dentry, name string, ds **[]*dentry) error {
		// If the parent is a setgid directory, use the parent's GID
		// rather than the caller's and enable setgid.
		kgid := creds.EffectiveKGID
		mode := opts.Mode
		if atomic.LoadUint32(&parent.mode)&linux.S_ISGID != 0 {
			kgid = auth.KGID(atomic.LoadUint32(&parent.gid))
			mode |= linux.S_ISGID
		}
		if _, err := parent.file.mkdir(ctx, name, p9.FileMode(mode), (p9.UID)(creds.EffectiveKUID), p9.GID(kgid)); err != nil {
			if !opts.ForSyntheticMountpoint || linuxerr.Equals(linuxerr.EEXIST, err) {
				return err
			}
			ctx.Infof("Failed to create remote directory %q: %v; falling back to synthetic directory", name, err)
			parent.createSyntheticChildLocked(&createSyntheticOpts{
				name: name,
				mode: linux.S_IFDIR | opts.Mode,
				kuid: creds.EffectiveKUID,
				kgid: creds.EffectiveKGID,
			})
			*ds = appendDentry(*ds, parent)
		}
		if fs.opts.interop != InteropModeShared {
			parent.incLinks()
		}
		return nil
	}, func(parent *dentry, name string) error {
		if !opts.ForSyntheticMountpoint {
			// Can't create non-synthetic files in synthetic directories.
			return linuxerr.EPERM
		}
		parent.createSyntheticChildLocked(&createSyntheticOpts{
			name: name,
			mode: linux.S_IFDIR | opts.Mode,
			kuid: creds.EffectiveKUID,
			kgid: creds.EffectiveKGID,
		})
		parent.incLinks()
		return nil
	})
}

// MknodAt implements vfs.FilesystemImpl.MknodAt.
func (fs *filesystem) MknodAt(ctx context.Context, rp *vfs.ResolvingPath, opts vfs.MknodOptions) error {
	return fs.doCreateAt(ctx, rp, false /* dir */, func(parent *dentry, name string, ds **[]*dentry) error {
		creds := rp.Credentials()
		_, err := parent.file.mknod(ctx, name, (p9.FileMode)(opts.Mode), opts.DevMajor, opts.DevMinor, (p9.UID)(creds.EffectiveKUID), (p9.GID)(creds.EffectiveKGID))
		if !linuxerr.Equals(linuxerr.EPERM, err) {
			return err
		}

		// EPERM means that gofer does not allow creating a socket or pipe. Fallback
		// to creating a synthetic one, i.e. one that is kept entirely in memory.

		// Check that we're not overriding an existing file with a synthetic one.
		_, _, err = fs.stepLocked(ctx, rp, parent, true, ds)
		switch {
		case err == nil:
			// Step succeeded, another file exists.
			return syserror.EEXIST
		case !linuxerr.Equals(linuxerr.ENOENT, err):
			// Unexpected error.
			return err
		}

		switch opts.Mode.FileType() {
		case linux.S_IFSOCK:
			parent.createSyntheticChildLocked(&createSyntheticOpts{
				name:     name,
				mode:     opts.Mode,
				kuid:     creds.EffectiveKUID,
				kgid:     creds.EffectiveKGID,
				endpoint: opts.Endpoint,
			})
			*ds = appendDentry(*ds, parent)
			return nil
		case linux.S_IFIFO:
			parent.createSyntheticChildLocked(&createSyntheticOpts{
				name: name,
				mode: opts.Mode,
				kuid: creds.EffectiveKUID,
				kgid: creds.EffectiveKGID,
				pipe: pipe.NewVFSPipe(true /* isNamed */, pipe.DefaultPipeSize),
			})
			*ds = appendDentry(*ds, parent)
			return nil
		}
		// Retain error from gofer if synthetic file cannot be created internally.
		return linuxerr.EPERM
	}, nil)
}

// OpenAt implements vfs.FilesystemImpl.OpenAt.
func (fs *filesystem) OpenAt(ctx context.Context, rp *vfs.ResolvingPath, opts vfs.OpenOptions) (*vfs.FileDescription, error) {
	// Reject O_TMPFILE, which is not supported; supporting it correctly in the
	// presence of other remote filesystem users requires remote filesystem
	// support, and it isn't clear that there's any way to implement this in
	// 9P.
	if opts.Flags&linux.O_TMPFILE != 0 {
		return nil, syserror.EOPNOTSUPP
	}
	mayCreate := opts.Flags&linux.O_CREAT != 0
	mustCreate := opts.Flags&(linux.O_CREAT|linux.O_EXCL) == (linux.O_CREAT | linux.O_EXCL)

	var ds *[]*dentry
	fs.renameMu.RLock()
	unlocked := false
	unlock := func() {
		if !unlocked {
			fs.renameMuRUnlockAndCheckCaching(ctx, &ds)
			unlocked = true
		}
	}
	defer unlock()

	start := rp.Start().Impl().(*dentry)
	if rp.Done() {
		// Reject attempts to open mount root directory with O_CREAT.
		if mayCreate && rp.MustBeDir() {
			return nil, syserror.EISDIR
		}
		if mustCreate {
			return nil, syserror.EEXIST
		}
		if !start.cachedMetadataAuthoritative() {
			// Refresh dentry's attributes before opening.
			if err := start.updateFromGetattr(ctx); err != nil {
				return nil, err
			}
		}
		start.IncRef()
		defer start.DecRef(ctx)
		unlock()
		// start is intentionally not added to ds (which would remove it from the
		// cache) because doing so regresses performance in practice.
		return start.open(ctx, rp, &opts)
	}

afterTrailingSymlink:
	parent, err := fs.walkParentDirLocked(ctx, rp, start, &ds)
	if err != nil {
		return nil, err
	}
	// Check for search permission in the parent directory.
	if err := parent.checkPermissions(rp.Credentials(), vfs.MayExec); err != nil {
		return nil, err
	}
	// Reject attempts to open directories with O_CREAT.
	if mayCreate && rp.MustBeDir() {
		return nil, syserror.EISDIR
	}
	if err := fs.revalidateOne(ctx, rp.VirtualFilesystem(), parent, rp.Component(), &ds); err != nil {
		return nil, err
	}
	// Determine whether or not we need to create a file.
	parent.dirMu.Lock()
	child, _, err := fs.stepLocked(ctx, rp, parent, false /* mayFollowSymlinks */, &ds)
	if linuxerr.Equals(linuxerr.ENOENT, err) && mayCreate {
		if parent.isSynthetic() {
			parent.dirMu.Unlock()
			return nil, linuxerr.EPERM
		}
		fd, err := parent.createAndOpenChildLocked(ctx, rp, &opts, &ds)
		parent.dirMu.Unlock()
		return fd, err
	}
	parent.dirMu.Unlock()
	if err != nil {
		return nil, err
	}
	if mustCreate {
		return nil, syserror.EEXIST
	}
	// Open existing child or follow symlink.
	if child.isSymlink() && rp.ShouldFollowSymlink() {
		target, err := child.readlink(ctx, rp.Mount())
		if err != nil {
			return nil, err
		}
		if err := rp.HandleSymlink(target); err != nil {
			return nil, err
		}
		start = parent
		goto afterTrailingSymlink
	}
	if rp.MustBeDir() && !child.isDir() {
		return nil, syserror.ENOTDIR
	}
	child.IncRef()
	defer child.DecRef(ctx)
	unlock()
	// child is intentionally not added to ds (which would remove it from the
	// cache) because doing so regresses performance in practice.
	return child.open(ctx, rp, &opts)
}

// Preconditions: The caller must hold no locks (since opening pipes may block
// indefinitely).
func (d *dentry) open(ctx context.Context, rp *vfs.ResolvingPath, opts *vfs.OpenOptions) (*vfs.FileDescription, error) {
	ats := vfs.AccessTypesForOpenFlags(opts)
	if err := d.checkPermissions(rp.Credentials(), ats); err != nil {
		return nil, err
	}

	trunc := opts.Flags&linux.O_TRUNC != 0 && d.fileType() == linux.S_IFREG
	if trunc {
		// Lock metadataMu *while* we open a regular file with O_TRUNC because
		// open(2) will change the file size on server.
		d.metadataMu.Lock()
		defer d.metadataMu.Unlock()
	}

	var vfd *vfs.FileDescription
	var err error
	mnt := rp.Mount()
	switch d.fileType() {
	case linux.S_IFREG:
		if !d.fs.opts.regularFilesUseSpecialFileFD {
			if err := d.ensureSharedHandle(ctx, ats.MayRead(), ats.MayWrite(), trunc); err != nil {
				return nil, err
			}
			fd, err := newRegularFileFD(mnt, d, opts.Flags)
			if err != nil {
				return nil, err
			}
			vfd = &fd.vfsfd
		}
	case linux.S_IFDIR:
		// Can't open directories with O_CREAT.
		if opts.Flags&linux.O_CREAT != 0 {
			return nil, syserror.EISDIR
		}
		// Can't open directories writably.
		if ats&vfs.MayWrite != 0 {
			return nil, syserror.EISDIR
		}
		if opts.Flags&linux.O_DIRECT != 0 {
			return nil, linuxerr.EINVAL
		}
		if !d.isSynthetic() {
			if err := d.ensureSharedHandle(ctx, ats&vfs.MayRead != 0, false /* write */, false /* trunc */); err != nil {
				return nil, err
			}
		}
		fd := &directoryFD{}
		fd.LockFD.Init(&d.locks)
		if err := fd.vfsfd.Init(fd, opts.Flags, mnt, &d.vfsd, &vfs.FileDescriptionOptions{}); err != nil {
			return nil, err
		}
		if atomic.LoadInt32(&d.readFD) >= 0 {
			fsmetric.GoferOpensHost.Increment()
		} else {
			fsmetric.GoferOpens9P.Increment()
		}
		return &fd.vfsfd, nil
	case linux.S_IFLNK:
		// Can't open symlinks without O_PATH, which is handled at the VFS layer.
		return nil, linuxerr.ELOOP
	case linux.S_IFSOCK:
		if d.isSynthetic() {
			return nil, linuxerr.ENXIO
		}
		if d.fs.iopts.OpenSocketsByConnecting {
			return d.openSocketByConnecting(ctx, opts)
		}
	case linux.S_IFIFO:
		if d.isSynthetic() {
			return d.pipe.Open(ctx, mnt, &d.vfsd, opts.Flags, &d.locks)
		}
	}

	if vfd == nil {
		if vfd, err = d.openSpecialFile(ctx, mnt, opts); err != nil {
			return nil, err
		}
	}

	if trunc {
		// If no errors occured so far then update file size in memory. This
		// step is required even if !d.cachedMetadataAuthoritative() because
		// d.mappings has to be updated.
		// d.metadataMu has already been acquired if trunc == true.
		d.updateSizeLocked(0)

		if d.cachedMetadataAuthoritative() {
			d.touchCMtimeLocked()
		}
	}
	return vfd, err
}

func (d *dentry) openSocketByConnecting(ctx context.Context, opts *vfs.OpenOptions) (*vfs.FileDescription, error) {
	if opts.Flags&linux.O_DIRECT != 0 {
		return nil, linuxerr.EINVAL
	}
	fdObj, err := d.file.connect(ctx, p9.AnonymousSocket)
	if err != nil {
		return nil, err
	}
	fd, err := host.NewFD(ctx, kernel.KernelFromContext(ctx).HostMount(), fdObj.FD(), &host.NewFDOptions{
		HaveFlags: true,
		Flags:     opts.Flags,
	})
	if err != nil {
		fdObj.Close()
		return nil, err
	}
	fdObj.Release()
	return fd, nil
}

func (d *dentry) openSpecialFile(ctx context.Context, mnt *vfs.Mount, opts *vfs.OpenOptions) (*vfs.FileDescription, error) {
	ats := vfs.AccessTypesForOpenFlags(opts)
	if opts.Flags&linux.O_DIRECT != 0 {
		return nil, linuxerr.EINVAL
	}
	// We assume that the server silently inserts O_NONBLOCK in the open flags
	// for all named pipes (because all existing gofers do this).
	//
	// NOTE(b/133875563): This makes named pipe opens racy, because the
	// mechanisms for translating nonblocking to blocking opens can only detect
	// the instantaneous presence of a peer holding the other end of the pipe
	// open, not whether the pipe was *previously* opened by a peer that has
	// since closed its end.
	isBlockingOpenOfNamedPipe := d.fileType() == linux.S_IFIFO && opts.Flags&linux.O_NONBLOCK == 0
retry:
	h, err := openHandle(ctx, d.file, ats.MayRead(), ats.MayWrite(), opts.Flags&linux.O_TRUNC != 0)
	if err != nil {
		if isBlockingOpenOfNamedPipe && ats == vfs.MayWrite && linuxerr.Equals(linuxerr.ENXIO, err) {
			// An attempt to open a named pipe with O_WRONLY|O_NONBLOCK fails
			// with ENXIO if opening the same named pipe with O_WRONLY would
			// block because there are no readers of the pipe.
			if err := sleepBetweenNamedPipeOpenChecks(ctx); err != nil {
				return nil, err
			}
			goto retry
		}
		return nil, err
	}
	if isBlockingOpenOfNamedPipe && ats == vfs.MayRead && h.fd >= 0 {
		if err := blockUntilNonblockingPipeHasWriter(ctx, h.fd); err != nil {
			h.close(ctx)
			return nil, err
		}
	}
	fd, err := newSpecialFileFD(h, mnt, d, opts.Flags)
	if err != nil {
		h.close(ctx)
		return nil, err
	}
	return &fd.vfsfd, nil
}

// Preconditions:
// * d.fs.renameMu must be locked.
// * d.dirMu must be locked.
// * !d.isSynthetic().
func (d *dentry) createAndOpenChildLocked(ctx context.Context, rp *vfs.ResolvingPath, opts *vfs.OpenOptions, ds **[]*dentry) (*vfs.FileDescription, error) {
	if err := d.checkPermissions(rp.Credentials(), vfs.MayWrite); err != nil {
		return nil, err
	}
	if d.isDeleted() {
		return nil, syserror.ENOENT
	}
	mnt := rp.Mount()
	if err := mnt.CheckBeginWrite(); err != nil {
		return nil, err
	}
	defer mnt.EndWrite()

	// 9P2000.L's lcreate takes a fid representing the parent directory, and
	// converts it into an open fid representing the created file, so we need
	// to duplicate the directory fid first.
	_, dirfile, err := d.file.walk(ctx, nil)
	if err != nil {
		return nil, err
	}
	creds := rp.Credentials()
	name := rp.Component()
	// We only want the access mode for creating the file.
	createFlags := p9.OpenFlags(opts.Flags) & p9.OpenFlagsModeMask

	// If the parent is a setgid directory, use the parent's GID rather
	// than the caller's.
	kgid := creds.EffectiveKGID
	if atomic.LoadUint32(&d.mode)&linux.S_ISGID != 0 {
		kgid = auth.KGID(atomic.LoadUint32(&d.gid))
	}

	fdobj, openFile, createQID, _, err := dirfile.create(ctx, name, createFlags, p9.FileMode(opts.Mode), (p9.UID)(creds.EffectiveKUID), p9.GID(kgid))
	if err != nil {
		dirfile.close(ctx)
		return nil, err
	}
	// Then we need to walk to the file we just created to get a non-open fid
	// representing it, and to get its metadata. This must use d.file since, as
	// explained above, dirfile was invalidated by dirfile.Create().
	_, nonOpenFile, attrMask, attr, err := d.file.walkGetAttrOne(ctx, name)
	if err != nil {
		openFile.close(ctx)
		if fdobj != nil {
			fdobj.Close()
		}
		return nil, err
	}

	// Construct the new dentry.
	child, err := d.fs.newDentry(ctx, nonOpenFile, createQID, attrMask, &attr)
	if err != nil {
		nonOpenFile.close(ctx)
		openFile.close(ctx)
		if fdobj != nil {
			fdobj.Close()
		}
		return nil, err
	}
	// Incorporate the fid that was opened by lcreate.
	useRegularFileFD := child.fileType() == linux.S_IFREG && !d.fs.opts.regularFilesUseSpecialFileFD
	if useRegularFileFD {
		openFD := int32(-1)
		if fdobj != nil {
			openFD = int32(fdobj.Release())
		}
		child.handleMu.Lock()
		if vfs.MayReadFileWithOpenFlags(opts.Flags) {
			child.readFile = openFile
			if fdobj != nil {
				child.readFD = openFD
				child.mmapFD = openFD
			}
		}
		if vfs.MayWriteFileWithOpenFlags(opts.Flags) {
			child.writeFile = openFile
			child.writeFD = openFD
		}
		child.handleMu.Unlock()
	}
	// Insert the dentry into the tree.
	d.cacheNewChildLocked(child, name)
	appendNewChildDentry(ds, d, child)
	if d.cachedMetadataAuthoritative() {
		d.touchCMtime()
		d.dirents = nil
	}

	// Finally, construct a file description representing the created file.
	var childVFSFD *vfs.FileDescription
	if useRegularFileFD {
		fd, err := newRegularFileFD(mnt, child, opts.Flags)
		if err != nil {
			return nil, err
		}
		childVFSFD = &fd.vfsfd
	} else {
		h := handle{
			file: openFile,
			fd:   -1,
		}
		if fdobj != nil {
			h.fd = int32(fdobj.Release())
		}
		fd, err := newSpecialFileFD(h, mnt, child, opts.Flags)
		if err != nil {
			h.close(ctx)
			return nil, err
		}
		childVFSFD = &fd.vfsfd
	}
	d.watches.Notify(ctx, name, linux.IN_CREATE, 0, vfs.PathEvent, false /* unlinked */)
	return childVFSFD, nil
}

// ReadlinkAt implements vfs.FilesystemImpl.ReadlinkAt.
func (fs *filesystem) ReadlinkAt(ctx context.Context, rp *vfs.ResolvingPath) (string, error) {
	var ds *[]*dentry
	fs.renameMu.RLock()
	defer fs.renameMuRUnlockAndCheckCaching(ctx, &ds)
	d, err := fs.resolveLocked(ctx, rp, &ds)
	if err != nil {
		return "", err
	}
	if !d.isSymlink() {
		return "", linuxerr.EINVAL
	}
	return d.readlink(ctx, rp.Mount())
}

// RenameAt implements vfs.FilesystemImpl.RenameAt.
func (fs *filesystem) RenameAt(ctx context.Context, rp *vfs.ResolvingPath, oldParentVD vfs.VirtualDentry, oldName string, opts vfs.RenameOptions) error {
	// Resolve newParent first to verify that it's on this Mount.
	var ds *[]*dentry
	fs.renameMu.Lock()
	defer fs.renameMuUnlockAndCheckCaching(ctx, &ds)
	newParent, err := fs.walkParentDirLocked(ctx, rp, rp.Start().Impl().(*dentry), &ds)
	if err != nil {
		return err
	}

	if opts.Flags&^linux.RENAME_NOREPLACE != 0 {
		return linuxerr.EINVAL
	}
	if fs.opts.interop == InteropModeShared && opts.Flags&linux.RENAME_NOREPLACE != 0 {
		// Requires 9P support to synchronize with other remote filesystem
		// users.
		return linuxerr.EINVAL
	}

	newName := rp.Component()
	if newName == "." || newName == ".." {
		if opts.Flags&linux.RENAME_NOREPLACE != 0 {
			return syserror.EEXIST
		}
		return linuxerr.EBUSY
	}
	mnt := rp.Mount()
	if mnt != oldParentVD.Mount() {
		return linuxerr.EXDEV
	}
	if err := mnt.CheckBeginWrite(); err != nil {
		return err
	}
	defer mnt.EndWrite()

	oldParent := oldParentVD.Dentry().Impl().(*dentry)
	if !oldParent.cachedMetadataAuthoritative() {
		if err := oldParent.updateFromGetattr(ctx); err != nil {
			return err
		}
	}
	creds := rp.Credentials()
	if err := oldParent.checkPermissions(creds, vfs.MayWrite|vfs.MayExec); err != nil {
		return err
	}

	vfsObj := rp.VirtualFilesystem()
	if err := fs.revalidateOne(ctx, vfsObj, newParent, newName, &ds); err != nil {
		return err
	}
	if err := fs.revalidateOne(ctx, vfsObj, oldParent, oldName, &ds); err != nil {
		return err
	}

	// We need a dentry representing the renamed file since, if it's a
	// directory, we need to check for write permission on it.
	oldParent.dirMu.Lock()
	defer oldParent.dirMu.Unlock()
	renamed, err := fs.getChildLocked(ctx, oldParent, oldName, &ds)
	if err != nil {
		return err
	}
	if err := oldParent.mayDelete(creds, renamed); err != nil {
		return err
	}
	if renamed.isDir() {
		if renamed == newParent || genericIsAncestorDentry(renamed, newParent) {
			return linuxerr.EINVAL
		}
		if oldParent != newParent {
			if err := renamed.checkPermissions(creds, vfs.MayWrite); err != nil {
				return err
			}
		}
	} else {
		if opts.MustBeDir || rp.MustBeDir() {
			return syserror.ENOTDIR
		}
	}

	if oldParent != newParent {
		if err := newParent.checkPermissions(creds, vfs.MayWrite|vfs.MayExec); err != nil {
			return err
		}
		newParent.dirMu.Lock()
		defer newParent.dirMu.Unlock()
	}
	if newParent.isDeleted() {
		return syserror.ENOENT
	}
	replaced, err := fs.getChildLocked(ctx, newParent, newName, &ds)
	if err != nil && !linuxerr.Equals(linuxerr.ENOENT, err) {
		return err
	}
	var replacedVFSD *vfs.Dentry
	if replaced != nil {
		if opts.Flags&linux.RENAME_NOREPLACE != 0 {
			return syserror.EEXIST
		}
		replacedVFSD = &replaced.vfsd
		if replaced.isDir() {
			if !renamed.isDir() {
				return syserror.EISDIR
			}
			if genericIsAncestorDentry(replaced, renamed) {
				return linuxerr.ENOTEMPTY
			}
		} else {
			if rp.MustBeDir() || renamed.isDir() {
				return syserror.ENOTDIR
			}
		}
	}

	if oldParent == newParent && oldName == newName {
		return nil
	}
	mntns := vfs.MountNamespaceFromContext(ctx)
	defer mntns.DecRef(ctx)
	if err := vfsObj.PrepareRenameDentry(mntns, &renamed.vfsd, replacedVFSD); err != nil {
		return err
	}

	// Update the remote filesystem.
	if !renamed.isSynthetic() {
		if err := renamed.file.rename(ctx, newParent.file, newName); err != nil {
			vfsObj.AbortRenameDentry(&renamed.vfsd, replacedVFSD)
			return err
		}
	} else if replaced != nil && !replaced.isSynthetic() {
		// We are replacing an existing real file with a synthetic one, so we
		// need to unlink the former.
		flags := uint32(0)
		if replaced.isDir() {
			flags = linux.AT_REMOVEDIR
		}
		if err := newParent.file.unlinkAt(ctx, newName, flags); err != nil {
			vfsObj.AbortRenameDentry(&renamed.vfsd, replacedVFSD)
			return err
		}
	}

	// Update the dentry tree.
	vfsObj.CommitRenameReplaceDentry(ctx, &renamed.vfsd, replacedVFSD)
	if replaced != nil {
		replaced.setDeleted()
		if replaced.isSynthetic() {
			newParent.syntheticChildren--
			replaced.decRefNoCaching()
		}
		ds = appendDentry(ds, replaced)
	}
	oldParent.cacheNegativeLookupLocked(oldName)
	// We don't use newParent.cacheNewChildLocked() since we don't want to mess
	// with reference counts and queue oldParent for checkCachingLocked if the
	// parent isn't actually changing.
	if oldParent != newParent {
		oldParent.decRefNoCaching()
		newParent.IncRef()
		ds = appendDentry(ds, newParent)
		ds = appendDentry(ds, oldParent)
		if renamed.isSynthetic() {
			oldParent.syntheticChildren--
			newParent.syntheticChildren++
		}
		renamed.parent = newParent
	}
	renamed.name = newName
	if newParent.children == nil {
		newParent.children = make(map[string]*dentry)
	}
	newParent.children[newName] = renamed

	// Update metadata.
	if renamed.cachedMetadataAuthoritative() {
		renamed.touchCtime()
	}
	if oldParent.cachedMetadataAuthoritative() {
		oldParent.dirents = nil
		oldParent.touchCMtime()
		if renamed.isDir() {
			oldParent.decLinks()
		}
	}
	if newParent.cachedMetadataAuthoritative() {
		newParent.dirents = nil
		newParent.touchCMtime()
		if renamed.isDir() && (replaced == nil || !replaced.isDir()) {
			// Increase the link count if we did not replace another directory.
			newParent.incLinks()
		}
	}
	vfs.InotifyRename(ctx, &renamed.watches, &oldParent.watches, &newParent.watches, oldName, newName, renamed.isDir())
	return nil
}

// RmdirAt implements vfs.FilesystemImpl.RmdirAt.
func (fs *filesystem) RmdirAt(ctx context.Context, rp *vfs.ResolvingPath) error {
	return fs.unlinkAt(ctx, rp, true /* dir */)
}

// SetStatAt implements vfs.FilesystemImpl.SetStatAt.
func (fs *filesystem) SetStatAt(ctx context.Context, rp *vfs.ResolvingPath, opts vfs.SetStatOptions) error {
	var ds *[]*dentry
	fs.renameMu.RLock()
	d, err := fs.resolveLocked(ctx, rp, &ds)
	if err != nil {
		fs.renameMuRUnlockAndCheckCaching(ctx, &ds)
		return err
	}
	err = d.setStat(ctx, rp.Credentials(), &opts, rp.Mount())
	fs.renameMuRUnlockAndCheckCaching(ctx, &ds)
	if err != nil {
		return err
	}

	if ev := vfs.InotifyEventFromStatMask(opts.Stat.Mask); ev != 0 {
		d.InotifyWithParent(ctx, ev, 0, vfs.InodeEvent)
	}
	return nil
}

// StatAt implements vfs.FilesystemImpl.StatAt.
func (fs *filesystem) StatAt(ctx context.Context, rp *vfs.ResolvingPath, opts vfs.StatOptions) (linux.Statx, error) {
	var ds *[]*dentry
	fs.renameMu.RLock()
	defer fs.renameMuRUnlockAndCheckCaching(ctx, &ds)
	d, err := fs.resolveLocked(ctx, rp, &ds)
	if err != nil {
		return linux.Statx{}, err
	}
	// Since walking updates metadata for all traversed dentries under
	// InteropModeShared, including the returned one, we can return cached
	// metadata here regardless of fs.opts.interop.
	var stat linux.Statx
	d.statTo(&stat)
	return stat, nil
}

// StatFSAt implements vfs.FilesystemImpl.StatFSAt.
func (fs *filesystem) StatFSAt(ctx context.Context, rp *vfs.ResolvingPath) (linux.Statfs, error) {
	var ds *[]*dentry
	fs.renameMu.RLock()
	defer fs.renameMuRUnlockAndCheckCaching(ctx, &ds)
	d, err := fs.resolveLocked(ctx, rp, &ds)
	if err != nil {
		return linux.Statfs{}, err
	}
	// If d is synthetic, invoke statfs on the first ancestor of d that isn't.
	for d.isSynthetic() {
		d = d.parent
	}
	fsstat, err := d.file.statFS(ctx)
	if err != nil {
		return linux.Statfs{}, err
	}
	nameLen := uint64(fsstat.NameLength)
	if nameLen > maxFilenameLen {
		nameLen = maxFilenameLen
	}
	return linux.Statfs{
		// This is primarily for distinguishing a gofer file system in
		// tests. Testing is important, so instead of defining
		// something completely random, use a standard value.
		Type:            linux.V9FS_MAGIC,
		BlockSize:       int64(fsstat.BlockSize),
		Blocks:          fsstat.Blocks,
		BlocksFree:      fsstat.BlocksFree,
		BlocksAvailable: fsstat.BlocksAvailable,
		Files:           fsstat.Files,
		FilesFree:       fsstat.FilesFree,
		NameLength:      nameLen,
	}, nil
}

// SymlinkAt implements vfs.FilesystemImpl.SymlinkAt.
func (fs *filesystem) SymlinkAt(ctx context.Context, rp *vfs.ResolvingPath, target string) error {
	return fs.doCreateAt(ctx, rp, false /* dir */, func(parent *dentry, name string, _ **[]*dentry) error {
		creds := rp.Credentials()
		_, err := parent.file.symlink(ctx, target, name, (p9.UID)(creds.EffectiveKUID), (p9.GID)(creds.EffectiveKGID))
		return err
	}, nil)
}

// UnlinkAt implements vfs.FilesystemImpl.UnlinkAt.
func (fs *filesystem) UnlinkAt(ctx context.Context, rp *vfs.ResolvingPath) error {
	return fs.unlinkAt(ctx, rp, false /* dir */)
}

// BoundEndpointAt implements vfs.FilesystemImpl.BoundEndpointAt.
func (fs *filesystem) BoundEndpointAt(ctx context.Context, rp *vfs.ResolvingPath, opts vfs.BoundEndpointOptions) (transport.BoundEndpoint, error) {
	var ds *[]*dentry
	fs.renameMu.RLock()
	defer fs.renameMuRUnlockAndCheckCaching(ctx, &ds)
	d, err := fs.resolveLocked(ctx, rp, &ds)
	if err != nil {
		return nil, err
	}
	if err := d.checkPermissions(rp.Credentials(), vfs.MayWrite); err != nil {
		return nil, err
	}
	if d.isSocket() {
		if !d.isSynthetic() {
			d.IncRef()
			ds = appendDentry(ds, d)
			return &endpoint{
				dentry: d,
				path:   opts.Addr,
			}, nil
		}
		if d.endpoint != nil {
			return d.endpoint, nil
		}
	}
	return nil, linuxerr.ECONNREFUSED
}

// ListXattrAt implements vfs.FilesystemImpl.ListXattrAt.
func (fs *filesystem) ListXattrAt(ctx context.Context, rp *vfs.ResolvingPath, size uint64) ([]string, error) {
	var ds *[]*dentry
	fs.renameMu.RLock()
	defer fs.renameMuRUnlockAndCheckCaching(ctx, &ds)
	d, err := fs.resolveLocked(ctx, rp, &ds)
	if err != nil {
		return nil, err
	}
	return d.listXattr(ctx, rp.Credentials(), size)
}

// GetXattrAt implements vfs.FilesystemImpl.GetXattrAt.
func (fs *filesystem) GetXattrAt(ctx context.Context, rp *vfs.ResolvingPath, opts vfs.GetXattrOptions) (string, error) {
	var ds *[]*dentry
	fs.renameMu.RLock()
	defer fs.renameMuRUnlockAndCheckCaching(ctx, &ds)
	d, err := fs.resolveLocked(ctx, rp, &ds)
	if err != nil {
		return "", err
	}
	return d.getXattr(ctx, rp.Credentials(), &opts)
}

// SetXattrAt implements vfs.FilesystemImpl.SetXattrAt.
func (fs *filesystem) SetXattrAt(ctx context.Context, rp *vfs.ResolvingPath, opts vfs.SetXattrOptions) error {
	var ds *[]*dentry
	fs.renameMu.RLock()
	d, err := fs.resolveLocked(ctx, rp, &ds)
	if err != nil {
		fs.renameMuRUnlockAndCheckCaching(ctx, &ds)
		return err
	}
	err = d.setXattr(ctx, rp.Credentials(), &opts)
	fs.renameMuRUnlockAndCheckCaching(ctx, &ds)
	if err != nil {
		return err
	}

	d.InotifyWithParent(ctx, linux.IN_ATTRIB, 0, vfs.InodeEvent)
	return nil
}

// RemoveXattrAt implements vfs.FilesystemImpl.RemoveXattrAt.
func (fs *filesystem) RemoveXattrAt(ctx context.Context, rp *vfs.ResolvingPath, name string) error {
	var ds *[]*dentry
	fs.renameMu.RLock()
	d, err := fs.resolveLocked(ctx, rp, &ds)
	if err != nil {
		fs.renameMuRUnlockAndCheckCaching(ctx, &ds)
		return err
	}
	err = d.removeXattr(ctx, rp.Credentials(), name)
	fs.renameMuRUnlockAndCheckCaching(ctx, &ds)
	if err != nil {
		return err
	}

	d.InotifyWithParent(ctx, linux.IN_ATTRIB, 0, vfs.InodeEvent)
	return nil
}

// PrependPath implements vfs.FilesystemImpl.PrependPath.
func (fs *filesystem) PrependPath(ctx context.Context, vfsroot, vd vfs.VirtualDentry, b *fspath.Builder) error {
	fs.renameMu.RLock()
	defer fs.renameMu.RUnlock()
	return genericPrependPath(vfsroot, vd.Mount(), vd.Dentry().Impl().(*dentry), b)
}

type mopt struct {
	key   string
	value interface{}
}

func (m mopt) String() string {
	if m.value == nil {
		return fmt.Sprintf("%s", m.key)
	}
	return fmt.Sprintf("%s=%v", m.key, m.value)
}

// MountOptions implements vfs.FilesystemImpl.MountOptions.
func (fs *filesystem) MountOptions() string {
	optsKV := []mopt{
		{moptTransport, transportModeFD}, // Only valid value, currently.
		{moptReadFD, fs.opts.fd},         // Currently, read and write FD are the same.
		{moptWriteFD, fs.opts.fd},        // Currently, read and write FD are the same.
		{moptAname, fs.opts.aname},
		{moptDfltUID, fs.opts.dfltuid},
		{moptDfltGID, fs.opts.dfltgid},
		{moptMsize, fs.opts.msize},
		{moptVersion, fs.opts.version},
		{moptDentryCacheLimit, fs.opts.maxCachedDentries},
	}

	switch fs.opts.interop {
	case InteropModeExclusive:
		optsKV = append(optsKV, mopt{moptCache, cacheFSCache})
	case InteropModeWritethrough:
		optsKV = append(optsKV, mopt{moptCache, cacheFSCacheWritethrough})
	case InteropModeShared:
		if fs.opts.regularFilesUseSpecialFileFD {
			optsKV = append(optsKV, mopt{moptCache, cacheNone})
		} else {
			optsKV = append(optsKV, mopt{moptCache, cacheRemoteRevalidating})
		}
	}
	if fs.opts.forcePageCache {
		optsKV = append(optsKV, mopt{moptForcePageCache, nil})
	}
	if fs.opts.limitHostFDTranslation {
		optsKV = append(optsKV, mopt{moptLimitHostFDTranslation, nil})
	}
	if fs.opts.overlayfsStaleRead {
		optsKV = append(optsKV, mopt{moptOverlayfsStaleRead, nil})
	}

	opts := make([]string, 0, len(optsKV))
	for _, opt := range optsKV {
		opts = append(opts, opt.String())
	}
	return strings.Join(opts, ",")
}