/
mount.go
1595 lines (1474 loc) · 47.8 KB
/
mount.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 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 vfs
import (
"bytes"
"fmt"
"math"
"sort"
"strings"
"gvisor.dev/gvisor/pkg/abi/linux"
"gvisor.dev/gvisor/pkg/atomicbitops"
"gvisor.dev/gvisor/pkg/cleanup"
"gvisor.dev/gvisor/pkg/context"
"gvisor.dev/gvisor/pkg/errors/linuxerr"
"gvisor.dev/gvisor/pkg/refs"
"gvisor.dev/gvisor/pkg/sentry/kernel/auth"
)
// MountMax is the maximum number of mounts allowed. In Linux this can be
// configured by the user at /proc/sys/fs/mount-max, but the default is
// 100,000. We set the gVisor limit to 10,000.
const (
MountMax = 10000
nsfsName = "nsfs"
cgroupFsName = "cgroup"
)
// A Mount is a replacement of a Dentry (Mount.key.point) from one Filesystem
// (Mount.key.parent.fs) with a Dentry (Mount.root) from another Filesystem
// (Mount.fs), which applies to path resolution in the context of a particular
// Mount (Mount.key.parent).
//
// Mounts are reference-counted. Unless otherwise specified, all Mount methods
// require that a reference is held.
//
// Mount and Filesystem are distinct types because it's possible for a single
// Filesystem to be mounted at multiple locations and/or in multiple mount
// namespaces.
//
// Mount is analogous to Linux's struct mount. (gVisor does not distinguish
// between struct mount and struct vfsmount.)
//
// +stateify savable
type Mount struct {
// vfs, fs, root are immutable. References are held on fs and root.
// Note that for a disconnected mount, root may be nil.
//
// Invariant: if not nil, root belongs to fs.
vfs *VirtualFilesystem
fs *Filesystem
root *Dentry
// ID is the immutable mount ID.
ID uint64
// Flags contains settings as specified for mount(2), e.g. MS_NOEXEC, except
// for MS_RDONLY which is tracked in "writers". flags is protected by
// VirtualFilesystem.mountMu.
flags MountFlags
// key is protected by VirtualFilesystem.mountMu and
// VirtualFilesystem.mounts.seq, and may be nil. References are held on
// key.parent and key.point if they are not nil.
//
// Invariant: key.parent != nil iff key.point != nil. key.point belongs to
// key.parent.fs.
key mountKey `state:".(VirtualDentry)"`
// ns is the namespace in which this Mount was mounted. ns is protected by
// VirtualFilesystem.mountMu.
ns *MountNamespace
// The lower 63 bits of refs are a reference count. The MSB of refs is set
// if the Mount has been eagerly umounted, as by umount(2) without the
// MNT_DETACH flag. refs is accessed using atomic memory operations.
refs atomicbitops.Int64
// children is the set of all Mounts for which Mount.key.parent is this
// Mount. children is protected by VirtualFilesystem.mountMu.
children map[*Mount]struct{}
// isShared indicates this mount has the MS_SHARED propagation type.
isShared bool
// sharedEntry is an entry in a circular list (ring) of mounts in a shared
// peer group.
sharedEntry mountEntry
// followerList is a list of mounts which has this mount as its leader.
followerList followerList
// followerEntry is an entry in a followerList.
followerEntry
// leader is the mount that this mount receives propagation events from.
leader *Mount
// groupID is the ID for this mount's shared peer group. If the mount is not
// in a peer group, this is 0.
groupID uint32
// umounted is true if VFS.umountRecursiveLocked() has been called on this
// Mount. VirtualFilesystem does not hold a reference on Mounts for which
// umounted is true. umounted is protected by VirtualFilesystem.mountMu.
umounted bool
// locked is true if the mount cannot be unmounted in the current mount
// namespace. It is analogous to MNT_LOCKED in Linux.
locked bool
// The lower 63 bits of writers is the number of calls to
// Mount.CheckBeginWrite() that have not yet been paired with a call to
// Mount.EndWrite(). The MSB of writers is set if MS_RDONLY is in effect.
// writers is accessed using atomic memory operations.
writers atomicbitops.Int64
}
func newMount(vfs *VirtualFilesystem, fs *Filesystem, root *Dentry, mntns *MountNamespace, opts *MountOptions) *Mount {
mnt := &Mount{
ID: vfs.lastMountID.Add(1),
flags: opts.Flags,
vfs: vfs,
fs: fs,
root: root,
ns: mntns,
locked: opts.Locked,
isShared: false,
refs: atomicbitops.FromInt64(1),
}
if opts.ReadOnly {
mnt.setReadOnlyLocked(true)
}
mnt.sharedEntry.Init(mnt)
refs.Register(mnt)
return mnt
}
// Options returns a copy of the MountOptions currently applicable to mnt.
func (mnt *Mount) Options() MountOptions {
mnt.vfs.lockMounts()
defer mnt.vfs.unlockMounts(context.Background())
return MountOptions{
Flags: mnt.flags,
ReadOnly: mnt.ReadOnlyLocked(),
}
}
// setMountOptions sets mnt's options to the given opts.
//
// Preconditions:
// - vfs.mountMu must be locked.
func (mnt *Mount) setMountOptions(opts *MountOptions) error {
if opts == nil {
return linuxerr.EINVAL
}
if err := mnt.setReadOnlyLocked(opts.ReadOnly); err != nil {
return err
}
mnt.flags = opts.Flags
return nil
}
// MountFlags returns a bit mask that indicates mount options.
func (mnt *Mount) MountFlags() uint64 {
mnt.vfs.lockMounts()
defer mnt.vfs.unlockMounts(context.Background())
var flags uint64
if mnt.flags.NoExec {
flags |= linux.ST_NOEXEC
}
if mnt.flags.NoATime {
flags |= linux.ST_NOATIME
}
if mnt.flags.NoDev {
flags |= linux.ST_NODEV
}
if mnt.flags.NoSUID {
flags |= linux.ST_NOSUID
}
if mnt.ReadOnlyLocked() {
flags |= linux.ST_RDONLY
}
return flags
}
func (mnt *Mount) isFollower() bool {
return mnt.leader != nil
}
func (mnt *Mount) neverConnected() bool {
return mnt.ns == nil
}
// coveringMount returns a mount that completely covers mnt if it exists and nil
// otherwise. A mount that covers another is one that is the only child of its
// parent and whose mountpoint is its parent's root.
func (mnt *Mount) coveringMount() *Mount {
if len(mnt.children) != 1 {
return nil
}
// Get the child from the children map.
var child *Mount
for child = range mnt.children {
break
}
if child.point() != mnt.root {
return nil
}
return child
}
// validInMountNS checks if the mount is valid in the current mount namespace. This includes
// checking if has previously been unmounted. It is analogous to fs/namespace.c:check_mnt() in
// Linux.
//
// +checklocks:vfs.mountMu
func (vfs *VirtualFilesystem) validInMountNS(ctx context.Context, mnt *Mount) bool {
if mntns := MountNamespaceFromContext(ctx); mntns != nil {
vfs.delayDecRef(mntns)
return mnt.ns == mntns && !mnt.umounted
}
return false
}
// NewFilesystem creates a new filesystem object not yet associated with any
// mounts. It can be installed into the filesystem tree with ConnectMountAt.
// Note that only the filesystem-specific mount options from opts are used by
// this function, mount flags are ignored. To set mount flags, pass them to a
// corresponding ConnectMountAt.
func (vfs *VirtualFilesystem) NewFilesystem(ctx context.Context, creds *auth.Credentials, source, fsTypeName string, opts *MountOptions) (*Filesystem, *Dentry, error) {
rft := vfs.getFilesystemType(fsTypeName)
if rft == nil {
return nil, nil, linuxerr.ENODEV
}
if !opts.GetFilesystemOptions.InternalMount && !rft.opts.AllowUserMount {
return nil, nil, linuxerr.ENODEV
}
return rft.fsType.GetFilesystem(ctx, vfs, creds, source, opts.GetFilesystemOptions)
}
// NewDisconnectedMount returns a Mount representing fs with the given root
// (which may be nil). The new Mount is not associated with any MountNamespace
// and is not connected to any other Mounts. References are taken on fs and
// root.
func (vfs *VirtualFilesystem) NewDisconnectedMount(fs *Filesystem, root *Dentry, opts *MountOptions) *Mount {
fs.IncRef()
if root != nil {
root.IncRef()
}
return newMount(vfs, fs, root, nil /* mntns */, opts)
}
// MountDisconnected creates a Filesystem configured by the given arguments,
// then returns a Mount representing it. The new Mount is not associated with
// any MountNamespace and is not connected to any other Mounts.
func (vfs *VirtualFilesystem) MountDisconnected(ctx context.Context, creds *auth.Credentials, source string, fsTypeName string, opts *MountOptions) (*Mount, error) {
fs, root, err := vfs.NewFilesystem(ctx, creds, source, fsTypeName, opts)
if err != nil {
return nil, err
}
return newMount(vfs, fs, root, nil /* mntns */, opts), nil
}
// attachTreeLocked attaches the mount tree at mnt to mp and propagates the mount to mp.mount's
// peers and followers. This method consumes the reference on mp. It is analogous to
// fs/namespace.c:attach_recursive_mnt() in Linux. The mount point mp must have its dentry locked
// before calling attachTreeLocked.
//
// +checklocks:vfs.mountMu
func (vfs *VirtualFilesystem) attachTreeLocked(ctx context.Context, mnt *Mount, mp VirtualDentry) error {
cleanup := cleanup.Make(func() {
vfs.cleanupGroupIDs(mnt.submountsLocked()) // +checklocksforce
mp.dentry.mu.Unlock()
vfs.delayDecRef(mp)
})
defer cleanup.Clean()
// This is equivalent to checking for SB_NOUSER in Linux, which is set on all
// anon mounts and sentry-internal filesystems like pipefs.
if mp.mount.neverConnected() {
return linuxerr.EINVAL
}
defer func() { mp.mount.ns.pending = 0 }()
if err := mp.mount.ns.checkMountCount(ctx, mnt); err != nil {
return err
}
var (
propMnts map[*Mount]struct{}
err error
)
if mp.mount.isShared {
if err := vfs.allocMountGroupIDs(mnt, true); err != nil {
return err
}
propMnts, err = vfs.doPropagation(ctx, mnt, mp)
if err != nil {
for pmnt := range propMnts {
if !pmnt.parent().neverConnected() {
pmnt.parent().ns.pending -= pmnt.countSubmountsLocked()
}
vfs.abortUncommitedMount(ctx, pmnt)
}
return err
}
}
cleanup.Release()
if mp.mount.isShared {
for _, m := range mnt.submountsLocked() {
m.isShared = true
}
}
vfs.mounts.seq.BeginWrite()
vfs.connectLocked(mnt, mp, mp.mount.ns)
vfs.mounts.seq.EndWrite()
mp.dentry.mu.Unlock()
vfs.commitChildren(ctx, mnt)
var owner *auth.UserNamespace
if mntns := MountNamespaceFromContext(ctx); mntns != nil {
owner = mntns.Owner
mntns.DecRef(ctx)
}
for pmnt := range propMnts {
vfs.commitMount(ctx, pmnt)
if pmnt.parent().ns.Owner != owner {
vfs.lockMountTree(pmnt)
}
pmnt.locked = false
}
return nil
}
// +checklocks:vfs.mountMu
func (vfs *VirtualFilesystem) lockMountTree(mnt *Mount) {
for _, m := range mnt.submountsLocked() {
// TODO(b/315839347): Add equivalents for MNT_LOCK_ATIME,
// MNT_LOCK_READONLY, etc.
m.locked = true
}
}
// +checklocks:vfs.mountMu
func (vfs *VirtualFilesystem) mountHasLockedChildren(mnt *Mount, vd VirtualDentry) bool {
for child := range mnt.children {
mp := child.getKey()
if !mp.mount.fs.Impl().IsDescendant(vd, mp) {
continue
}
if child.locked {
return true
}
}
return false
}
// ConnectMountAt connects mnt at the path represented by target.
//
// Preconditions: mnt must be disconnected.
func (vfs *VirtualFilesystem) ConnectMountAt(ctx context.Context, creds *auth.Credentials, mnt *Mount, target *PathOperation) error {
// We can't hold vfs.mountMu while calling FilesystemImpl methods due to
// lock ordering.
vd, err := vfs.GetDentryAt(ctx, creds, target, &GetDentryOptions{})
if err != nil {
return err
}
vfs.lockMounts()
defer vfs.unlockMounts(ctx)
mp, err := vfs.lockMountpoint(vd)
if err != nil {
return err
}
if mp.mount.neverConnected() || mp.mount.umounted {
mp.dentry.mu.Unlock()
vfs.delayDecRef(mp)
return linuxerr.EINVAL
}
return vfs.attachTreeLocked(ctx, mnt, mp)
}
// lockMountpoint returns VirtualDentry with a locked Dentry. If vd is a
// mountpoint, the method returns a VirtualDentry with a locked Dentry that is
// the top most mount stacked on that Dentry. This method consumes a reference
// on vd and returns a VirtualDentry with an extra reference. It is analogous to
// fs/namespace.c:do_lock_mount() in Linux.
//
// +checklocks:vfs.mountMu
func (vfs *VirtualFilesystem) lockMountpoint(vd VirtualDentry) (VirtualDentry, error) {
vd.dentry.mu.Lock()
for {
if vd.mount.umounted || vd.dentry.dead {
vd.dentry.mu.Unlock()
vfs.delayDecRef(vd)
return VirtualDentry{}, linuxerr.ENOENT
}
// vd might have been mounted over between vfs.GetDentryAt() and
// vfs.mountMu.Lock().
if !vd.dentry.isMounted() {
break
}
nextmnt := vfs.mounts.Lookup(vd.mount, vd.dentry)
if nextmnt == nil {
break
}
// It's possible that nextmnt has been umounted but not disconnected,
// in which case vfs no longer holds a reference on it, and the last
// reference may be concurrently dropped even though we're holding
// vfs.mountMu.
if !nextmnt.tryIncMountedRef() {
break
}
// This can't fail since we're holding vfs.mountMu.
nextmnt.root.IncRef()
vd.dentry.mu.Unlock()
vfs.delayDecRef(vd)
vd = VirtualDentry{
mount: nextmnt,
dentry: nextmnt.root,
}
vd.dentry.mu.Lock()
}
return vd, nil
}
// CloneMountAt returns a new mount with the same fs, specified root and
// mount options. If mount options are nil, mnt's options are copied. The clone
// is added to mnt's peer group if mnt is shared. If not the clone is in a
// shared peer group by itself.
func (vfs *VirtualFilesystem) CloneMountAt(mnt *Mount, root *Dentry, mopts *MountOptions) (*Mount, error) {
vfs.lockMounts()
defer vfs.unlockMounts(context.Background())
return vfs.cloneMount(mnt, root, mopts, makeSharedClone)
}
// cloneMount returns a new mount with mnt.fs as the filesystem and root as the
// root, with a propagation type specified by cloneType. The returned mount has
// an extra reference. If mopts is nil, use the options found in mnt.
// This method is analogous to fs/namespace.c:clone_mnt() in Linux.
//
// +checklocks:vfs.mountMu
func (vfs *VirtualFilesystem) cloneMount(mnt *Mount, root *Dentry, mopts *MountOptions, cloneType int) (*Mount, error) {
opts := mopts
if opts == nil {
opts = &MountOptions{
Flags: mnt.flags,
ReadOnly: mnt.ReadOnlyLocked(),
}
}
clone := vfs.NewDisconnectedMount(mnt.fs, root, opts)
if cloneType&(makeFollowerClone|makePrivateClone|sharedToFollowerClone) != 0 {
clone.groupID = 0
} else {
clone.groupID = mnt.groupID
}
if cloneType&makeSharedClone != 0 && clone.groupID == 0 {
if err := vfs.allocateGroupID(clone); err != nil {
vfs.delayDecRef(clone)
return nil, err
}
}
clone.isShared = mnt.isShared
clone.locked = mnt.locked
if cloneType&makeFollowerClone != 0 || (cloneType&sharedToFollowerClone != 0 && mnt.isShared) {
mnt.followerList.PushFront(clone)
clone.leader = mnt
clone.isShared = false
} else if cloneType&makePrivateClone == 0 {
if cloneType&makeSharedClone != 0 || mnt.isShared {
mnt.sharedEntry.Add(&clone.sharedEntry)
}
if mnt.isFollower() {
mnt.leader.followerList.InsertAfter(mnt, clone)
}
clone.leader = mnt.leader
} else {
clone.isShared = false
}
if cloneType&makeSharedClone != 0 {
clone.isShared = true
}
return clone, nil
}
type cloneTreeNode struct {
prevMount *Mount
parentMount *Mount
}
// cloneMountTree creates a copy of mnt's tree with the specified root
// dentry at root. The new descendants are added to mnt's children list but are
// not connected with call to connectLocked.
// `cloneFunc` is a callback that is executed for each cloned mount.
// This method is analogous to fs/namespace.c:copy_tree() in Linux.
//
// +checklocks:vfs.mountMu
func (vfs *VirtualFilesystem) cloneMountTree(ctx context.Context, mnt *Mount, root *Dentry, cloneType int, cloneFunc func(ctx context.Context, oldmnt, newMnt *Mount)) (*Mount, error) {
clone, err := vfs.cloneMount(mnt, root, nil, cloneType)
if err != nil {
return nil, err
}
if cloneFunc != nil {
cloneFunc(ctx, mnt, clone)
}
queue := []cloneTreeNode{{mnt, clone}}
for len(queue) != 0 {
p := queue[len(queue)-1]
queue = queue[:len(queue)-1]
for c := range p.prevMount.children {
if mp := c.getKey(); p.prevMount == mnt && !mp.mount.fs.Impl().IsDescendant(VirtualDentry{mnt, root}, mp) {
continue
}
m, err := vfs.cloneMount(c, c.root, nil, cloneType)
if err != nil {
vfs.abortUncommitedMount(ctx, clone)
return nil, err
}
mp := VirtualDentry{
mount: p.parentMount,
dentry: c.point(),
}
mp.IncRef()
m.setKey(mp)
if p.parentMount.children == nil {
p.parentMount.children = make(map[*Mount]struct{})
}
p.parentMount.children[m] = struct{}{}
if len(c.children) != 0 {
queue = append(queue, cloneTreeNode{c, m})
}
if cloneFunc != nil {
cloneFunc(ctx, c, m)
}
}
}
return clone, nil
}
// BindAt creates a clone of the source path's parent mount and mounts it at
// the target path. The new mount's root dentry is one pointed to by the source
// path.
func (vfs *VirtualFilesystem) BindAt(ctx context.Context, creds *auth.Credentials, source, target *PathOperation, recursive bool) error {
sourceVd, err := vfs.GetDentryAt(ctx, creds, source, &GetDentryOptions{})
if err != nil {
return err
}
defer sourceVd.DecRef(ctx)
targetVd, err := vfs.GetDentryAt(ctx, creds, target, &GetDentryOptions{})
if err != nil {
return err
}
vfs.lockMounts()
defer vfs.unlockMounts(ctx)
mp, err := vfs.lockMountpoint(targetVd)
if err != nil {
return err
}
cleanup := cleanup.Make(func() {
mp.dentry.mu.Unlock()
vfs.delayDecRef(mp) // +checklocksforce
})
defer cleanup.Clean()
// Namespace mounts can be binded to other mount points.
fsName := sourceVd.mount.Filesystem().FilesystemType().Name()
if !vfs.validInMountNS(ctx, sourceVd.mount) && fsName != nsfsName && fsName != cgroupFsName {
return linuxerr.EINVAL
}
if !vfs.validInMountNS(ctx, mp.mount) {
return linuxerr.EINVAL
}
var clone *Mount
if recursive {
clone, err = vfs.cloneMountTree(ctx, sourceVd.mount, sourceVd.dentry, 0, nil)
} else {
if vfs.mountHasLockedChildren(sourceVd.mount, sourceVd) {
return linuxerr.EINVAL
}
clone, err = vfs.cloneMount(sourceVd.mount, sourceVd.dentry, nil, 0)
}
if err != nil {
return err
}
cleanup.Release()
vfs.delayDecRef(clone)
clone.locked = false
if err := vfs.attachTreeLocked(ctx, clone, mp); err != nil {
vfs.abortUncomittedChildren(ctx, clone)
return err
}
return nil
}
// RemountAt changes the mountflags and data of an existing mount without having to unmount and remount the filesystem.
func (vfs *VirtualFilesystem) RemountAt(ctx context.Context, creds *auth.Credentials, pop *PathOperation, opts *MountOptions) error {
vd, err := vfs.getMountpoint(ctx, creds, pop)
if err != nil {
return err
}
defer vd.DecRef(ctx)
vfs.lockMounts()
defer vfs.unlockMounts(ctx)
mnt := vd.Mount()
if !vfs.validInMountNS(ctx, mnt) {
return linuxerr.EINVAL
}
return mnt.setMountOptions(opts)
}
// MountAt creates and mounts a Filesystem configured by the given arguments.
// The VirtualFilesystem will hold a reference to the Mount until it is
// unmounted.
//
// This method returns the mounted Mount without a reference, for convenience
// during VFS setup when there is no chance of racing with unmount.
func (vfs *VirtualFilesystem) MountAt(ctx context.Context, creds *auth.Credentials, source string, target *PathOperation, fsTypeName string, opts *MountOptions) (*Mount, error) {
mnt, err := vfs.MountDisconnected(ctx, creds, source, fsTypeName, opts)
if err != nil {
return nil, err
}
defer mnt.DecRef(ctx)
if err := vfs.ConnectMountAt(ctx, creds, mnt, target); err != nil {
return nil, err
}
return mnt, nil
}
// UmountAt removes the Mount at the given path.
func (vfs *VirtualFilesystem) UmountAt(ctx context.Context, creds *auth.Credentials, pop *PathOperation, opts *UmountOptions) error {
if opts.Flags&^(linux.MNT_FORCE|linux.MNT_DETACH) != 0 {
return linuxerr.EINVAL
}
// MNT_FORCE is currently unimplemented except for the permission check.
// Force unmounting specifically requires CAP_SYS_ADMIN in the root user
// namespace, and not in the owner user namespace for the target mount. See
// fs/namespace.c:SYSCALL_DEFINE2(umount, ...)
if opts.Flags&linux.MNT_FORCE != 0 && creds.HasCapabilityIn(linux.CAP_SYS_ADMIN, creds.UserNamespace.Root()) {
return linuxerr.EPERM
}
vd, err := vfs.getMountpoint(ctx, creds, pop)
if err != nil {
return err
}
defer vd.DecRef(ctx)
vfs.lockMounts()
defer vfs.unlockMounts(ctx)
if vd.mount.locked {
return linuxerr.EINVAL
}
if !vfs.validInMountNS(ctx, vd.mount) {
return linuxerr.EINVAL
}
if vd.mount == vd.mount.ns.root {
return linuxerr.EINVAL
}
if opts.Flags&linux.MNT_DETACH == 0 && vfs.arePropMountsBusy(vd.mount) {
return linuxerr.EBUSY
}
// TODO(gvisor.dev/issue/1035): Linux special-cases umount of the caller's
// root, which we don't implement yet (we'll just fail it since the caller
// holds a reference on it).
vfs.umountTreeLocked(vd.mount, &umountRecursiveOptions{
eager: opts.Flags&linux.MNT_DETACH == 0,
disconnectHierarchy: true,
propagate: true,
})
return nil
}
// mountHasExpectedRefs checks that mnt has the correct number of references
// before a umount. It is analogous to fs/pnode.c:do_refcount_check().
//
// +checklocks:vfs.mountMu
func (vfs *VirtualFilesystem) mountHasExpectedRefs(mnt *Mount) bool {
expectedRefs := int64(1)
if !mnt.umounted {
expectedRefs++
}
if mnt.coveringMount() != nil {
expectedRefs++
}
return mnt.refs.Load()&^math.MinInt64 == expectedRefs // mask out MSB
}
// +stateify savable
type umountRecursiveOptions struct {
// If eager is true, ensure that future calls to Mount.tryIncMountedRef()
// on umounted mounts fail.
//
// eager is analogous to Linux's UMOUNT_SYNC.
eager bool
// If disconnectHierarchy is true, Mounts that are umounted hierarchically
// should be disconnected from their parents. (Mounts whose parents are not
// umounted, which in most cases means the Mount passed to the initial call
// to umountRecursiveLocked, are unconditionally disconnected for
// consistency with Linux.)
//
// disconnectHierarchy is analogous to Linux's !UMOUNT_CONNECTED.
disconnectHierarchy bool
// If propagate is true, mounts located at the same point on the mount's
// parent's peers and follows will also be umounted if they do not have any
// children.
//
// propagate is analogous to Linux's UMOUNT_PROPAGATE.
propagate bool
}
// shouldUmount returns if this mount should be disconnected from its parent.
// It is analogous to fs/namespace.c:disconnect_mount() in Linux.
//
// +checklocks:vfs.mountMu
func (vfs *VirtualFilesystem) shouldUmount(mnt *Mount, opts *umountRecursiveOptions) bool {
// Always disconnect when it's not a lazy unmount.
if opts.eager {
return true
}
// If a mount does not have a parent, it won't be disconnected but will be
// DecRef-ed.
if mnt.parent() == nil {
return true
}
// Always unmount if the parent is not marked as unmounted.
if !mnt.parent().umounted {
return true
}
// If the parent is marked as unmounted, we can only unmount is
// UMOUNT_CONNECTED is false.
if !opts.disconnectHierarchy {
return false
}
if mnt.locked {
return false
}
return true
}
// umountTreeLocked marks mnt and its descendants as umounted.
//
// umountTreeLocked is analogous to Linux's fs/namespace.c:umount_tree().
// +checklocks:vfs.mountMu
func (vfs *VirtualFilesystem) umountTreeLocked(mnt *Mount, opts *umountRecursiveOptions) {
if opts.propagate {
vfs.unlockPropagationMounts(mnt)
}
umountMnts := mnt.submountsLocked()
for _, mnt := range umountMnts {
vfs.umount(mnt)
}
if opts.propagate {
umountMnts = append(umountMnts, vfs.propagateUmount(umountMnts)...)
}
vfs.mounts.seq.BeginWrite()
for _, mnt := range umountMnts {
if opts.eager {
for {
refs := mnt.refs.Load()
if refs < 0 {
break
}
if mnt.refs.CompareAndSwap(refs, refs|math.MinInt64) {
break
}
}
}
if mnt.parent() != nil {
vfs.delayDecRef(mnt.getKey())
if vfs.shouldUmount(mnt, opts) {
vfs.disconnectLocked(mnt)
} else {
// Restore mnt in it's parent children list with a reference, but leave
// it marked as unmounted. These partly unmounted mounts are cleaned up
// in vfs.forgetDeadMountpoints and Mount.destroy. We keep the extra
// reference on the mount but remove a reference on the mount point so
// that mount.Destroy is called when there are no other references on
// the parent.
mnt.IncRef()
mnt.parent().children[mnt] = struct{}{}
}
}
vfs.setPropagation(mnt, linux.MS_PRIVATE)
}
vfs.mounts.seq.EndWrite()
}
// +checklocks:vfs.mountMu
func (vfs *VirtualFilesystem) umount(mnt *Mount) {
if !mnt.umounted {
mnt.umounted = true
vfs.delayDecRef(mnt)
}
if parent := mnt.parent(); parent != nil {
delete(parent.children, mnt)
}
}
// changeMountpoint disconnects mnt from its current mount point and connects
// it to mp. It must be called from a vfs.mounts.seq writer critical section.
//
// +checklocks:vfs.mountMu
func (vfs *VirtualFilesystem) changeMountpoint(mnt *Mount, mp VirtualDentry) {
mp.dentry.mu.Lock()
vfs.delayDecRef(vfs.disconnectLocked(mnt))
vfs.delayDecRef(mnt)
mp.IncRef()
vfs.connectLocked(mnt, mp, mp.mount.ns)
mp.dentry.mu.Unlock()
}
// connectLocked makes vd the mount parent/point for mnt. It consumes
// references held by vd.
//
// Preconditions:
// - vfs.mountMu must be locked.
// - vfs.mounts.seq must be in a writer critical section.
// - d.mu must be locked.
// - mnt.parent() == nil or mnt.parent().children doesn't contain mnt.
// i.e. mnt must not already be connected.
func (vfs *VirtualFilesystem) connectLocked(mnt *Mount, vd VirtualDentry, mntns *MountNamespace) {
if checkInvariants {
if mnt.parent() != nil && mnt.parent().children != nil {
if _, ok := mnt.parent().children[mnt]; ok {
panic("VFS.connectLocked called on connected mount")
}
}
}
mnt.IncRef() // dropped by vfs.umount().
mnt.setKey(vd)
if vd.mount.children == nil {
vd.mount.children = make(map[*Mount]struct{})
}
vd.mount.children[mnt] = struct{}{}
vd.dentry.mounts.Add(1)
mnt.ns = mntns
mntns.mountpoints[vd.dentry]++
mntns.mounts++
vfs.mounts.insertSeqed(mnt)
vfsmpmounts, ok := vfs.mountpoints[vd.dentry]
if !ok {
vfsmpmounts = make(map[*Mount]struct{})
vfs.mountpoints[vd.dentry] = vfsmpmounts
}
vfsmpmounts[mnt] = struct{}{}
vfs.maybeResolveMountPromise(vd)
}
// disconnectLocked makes vd have no mount parent/point and returns its old
// mount parent/point with a reference held.
//
// Preconditions:
// - vfs.mountMu must be locked.
// - vfs.mounts.seq must be in a writer critical section.
// - mnt.parent() != nil.
func (vfs *VirtualFilesystem) disconnectLocked(mnt *Mount) VirtualDentry {
vd := mnt.getKey()
if checkInvariants {
if vd.mount == nil {
panic("VFS.disconnectLocked called on disconnected mount")
}
if mnt.ns.mountpoints[vd.dentry] == 0 {
panic("VFS.disconnectLocked called on dentry with zero mountpoints.")
}
if mnt.ns.mounts == 0 {
panic("VFS.disconnectLocked called on namespace with zero mounts.")
}
}
delete(vd.mount.children, mnt)
vd.dentry.mounts.Add(math.MaxUint32) // -1
mnt.ns.mountpoints[vd.dentry]--
mnt.ns.mounts--
if mnt.ns.mountpoints[vd.dentry] == 0 {
delete(mnt.ns.mountpoints, vd.dentry)
}
vfs.mounts.removeSeqed(mnt)
mnt.setKey(VirtualDentry{}) // Clear mnt.key.
vfsmpmounts := vfs.mountpoints[vd.dentry]
delete(vfsmpmounts, mnt)
if len(vfsmpmounts) == 0 {
delete(vfs.mountpoints, vd.dentry)
}
return vd
}
// tryIncMountedRef increments mnt's reference count and returns true. If mnt's
// reference count is already zero, or has been eagerly umounted,
// tryIncMountedRef does nothing and returns false.
//
// tryIncMountedRef does not require that a reference is held on mnt.
func (mnt *Mount) tryIncMountedRef() bool {
for {
r := mnt.refs.Load()
if r <= 0 { // r < 0 => MSB set => eagerly unmounted
return false
}
if mnt.refs.CompareAndSwap(r, r+1) {
if mnt.LogRefs() {
refs.LogTryIncRef(mnt, r+1)
}
return true
}
}
}
// IncRef increments mnt's reference count.
func (mnt *Mount) IncRef() {
// In general, negative values for mnt.refs are valid because the MSB is
// the eager-unmount bit.
r := mnt.refs.Add(1)
if mnt.LogRefs() {
refs.LogIncRef(mnt, r)
}
}
// DecRef decrements mnt's reference count.
func (mnt *Mount) DecRef(ctx context.Context) {
r := mnt.refs.Add(-1)
if mnt.LogRefs() {
refs.LogDecRef(mnt, r)
}
if r&^math.MinInt64 == 0 { // mask out MSB
refs.Unregister(mnt)
mnt.destroy(ctx)
}
}
func (mnt *Mount) destroy(ctx context.Context) {
mnt.vfs.lockMounts()
defer mnt.vfs.unlockMounts(ctx)
if mnt.parent() != nil {
mnt.vfs.mounts.seq.BeginWrite()
vd := mnt.vfs.disconnectLocked(mnt)
if vd.Ok() {
mnt.vfs.delayDecRef(vd)
}
mnt.vfs.mounts.seq.EndWrite()
}
// Cleanup any leftover children. The mount point has already been decref'd in
// umount so we just need to clean up the actual mounts.
if len(mnt.children) != 0 {
mnt.vfs.mounts.seq.BeginWrite()
for child := range mnt.children {
if checkInvariants {
if !child.umounted {
panic("children of a mount that has no references should already be marked as unmounted.")
}
}
mnt.vfs.disconnectLocked(child)
mnt.vfs.delayDecRef(child)
}
mnt.vfs.mounts.seq.EndWrite()
}
if mnt.root != nil {
mnt.vfs.delayDecRef(mnt.root)
}
mnt.vfs.delayDecRef(mnt.fs)
}
// RefType implements refs.CheckedObject.Type.
func (mnt *Mount) RefType() string {
return "vfs.Mount"
}
// LeakMessage implements refs.CheckedObject.LeakMessage.
func (mnt *Mount) LeakMessage() string {
return fmt.Sprintf("[vfs.Mount %p] reference count of %d instead of 0", mnt, mnt.refs.Load())
}
// LogRefs implements refs.CheckedObject.LogRefs.
//
// This should only be set to true for debugging purposes, as it can generate an
// extremely large amount of output and drastically degrade performance.
func (mnt *Mount) LogRefs() bool {
return false
}
// getMountAt returns the last Mount in the stack mounted at (mnt, d). It takes
// a reference on the returned Mount. If (mnt, d) is not a mount point,