summaryrefslogtreecommitdiffhomepage
path: root/pkg/sentry/fsimpl/tmpfs
diff options
context:
space:
mode:
Diffstat (limited to 'pkg/sentry/fsimpl/tmpfs')
-rw-r--r--pkg/sentry/fsimpl/tmpfs/directory.go8
-rw-r--r--pkg/sentry/fsimpl/tmpfs/filesystem.go58
-rw-r--r--pkg/sentry/fsimpl/tmpfs/regular_file.go33
-rw-r--r--pkg/sentry/fsimpl/tmpfs/tmpfs.go23
4 files changed, 91 insertions, 31 deletions
diff --git a/pkg/sentry/fsimpl/tmpfs/directory.go b/pkg/sentry/fsimpl/tmpfs/directory.go
index 913b8a6c5..0a1ad4765 100644
--- a/pkg/sentry/fsimpl/tmpfs/directory.go
+++ b/pkg/sentry/fsimpl/tmpfs/directory.go
@@ -79,7 +79,10 @@ func (dir *directory) removeChildLocked(child *dentry) {
dir.iterMu.Lock()
dir.childList.Remove(child)
dir.iterMu.Unlock()
- child.unlinked = true
+}
+
+func (dir *directory) mayDelete(creds *auth.Credentials, child *dentry) error {
+ return vfs.CheckDeleteSticky(creds, linux.FileMode(atomic.LoadUint32(&dir.inode.mode)), auth.KUID(atomic.LoadUint32(&child.inode.uid)))
}
type directoryFD struct {
@@ -107,13 +110,14 @@ func (fd *directoryFD) IterDirents(ctx context.Context, cb vfs.IterDirentsCallba
fs := fd.filesystem()
dir := fd.inode().impl.(*directory)
+ defer fd.dentry().InotifyWithParent(linux.IN_ACCESS, 0, vfs.PathEvent)
+
// fs.mu is required to read d.parent and dentry.name.
fs.mu.RLock()
defer fs.mu.RUnlock()
dir.iterMu.Lock()
defer dir.iterMu.Unlock()
- fd.dentry().InotifyWithParent(linux.IN_ACCESS, 0, vfs.PathEvent)
fd.inode().touchAtime(fd.vfsfd.Mount())
if fd.off == 0 {
diff --git a/pkg/sentry/fsimpl/tmpfs/filesystem.go b/pkg/sentry/fsimpl/tmpfs/filesystem.go
index 72399b321..a0f20c2d4 100644
--- a/pkg/sentry/fsimpl/tmpfs/filesystem.go
+++ b/pkg/sentry/fsimpl/tmpfs/filesystem.go
@@ -79,7 +79,7 @@ afterSymlink:
}
if symlink, ok := child.inode.impl.(*symlink); ok && rp.ShouldFollowSymlink() {
// Symlink traversal updates access time.
- atomic.StoreInt64(&d.inode.atime, d.inode.fs.clock.Now().Nanoseconds())
+ child.inode.touchAtime(rp.Mount())
if err := rp.HandleSymlink(symlink.target); err != nil {
return nil, err
}
@@ -182,7 +182,7 @@ func (fs *filesystem) doCreateAt(rp *vfs.ResolvingPath, dir bool, create func(pa
if dir {
ev |= linux.IN_ISDIR
}
- parentDir.inode.watches.Notify(name, uint32(ev), 0, vfs.InodeEvent)
+ parentDir.inode.watches.Notify(name, uint32(ev), 0, vfs.InodeEvent, false /* unlinked */)
parentDir.inode.touchCMtime()
return nil
}
@@ -237,18 +237,22 @@ func (fs *filesystem) LinkAt(ctx context.Context, rp *vfs.ResolvingPath, vd vfs.
return syserror.EXDEV
}
d := vd.Dentry().Impl().(*dentry)
- if d.inode.isDir() {
+ i := d.inode
+ if i.isDir() {
return syserror.EPERM
}
- if d.inode.nlink == 0 {
+ if err := vfs.MayLink(auth.CredentialsFromContext(ctx), linux.FileMode(atomic.LoadUint32(&i.mode)), auth.KUID(atomic.LoadUint32(&i.uid)), auth.KGID(atomic.LoadUint32(&i.gid))); err != nil {
+ return err
+ }
+ if i.nlink == 0 {
return syserror.ENOENT
}
- if d.inode.nlink == maxLinks {
+ if i.nlink == maxLinks {
return syserror.EMLINK
}
- d.inode.incLinksLocked()
- d.inode.watches.Notify("", linux.IN_ATTRIB, 0, vfs.InodeEvent)
- parentDir.insertChildLocked(fs.newDentry(d.inode), name)
+ i.incLinksLocked()
+ i.watches.Notify("", linux.IN_ATTRIB, 0, vfs.InodeEvent, false /* unlinked */)
+ parentDir.insertChildLocked(fs.newDentry(i), name)
return nil
})
}
@@ -273,7 +277,7 @@ func (fs *filesystem) MknodAt(ctx context.Context, rp *vfs.ResolvingPath, opts v
creds := rp.Credentials()
var childInode *inode
switch opts.Mode.FileType() {
- case 0, linux.S_IFREG:
+ case linux.S_IFREG:
childInode = fs.newRegularFile(creds.EffectiveKUID, creds.EffectiveKGID, opts.Mode)
case linux.S_IFIFO:
childInode = fs.newNamedPipe(creds.EffectiveKUID, creds.EffectiveKGID, opts.Mode)
@@ -364,10 +368,13 @@ afterTrailingSymlink:
if err != nil {
return nil, err
}
- parentDir.inode.watches.Notify(name, linux.IN_CREATE, 0, vfs.PathEvent)
+ parentDir.inode.watches.Notify(name, linux.IN_CREATE, 0, vfs.PathEvent, false /* unlinked */)
parentDir.inode.touchCMtime()
return fd, nil
}
+ if mustCreate {
+ return nil, syserror.EEXIST
+ }
// Is the file mounted over?
if err := rp.CheckMount(&child.vfsd); err != nil {
return nil, err
@@ -375,7 +382,7 @@ afterTrailingSymlink:
// Do we need to resolve a trailing symlink?
if symlink, ok := child.inode.impl.(*symlink); ok && rp.ShouldFollowSymlink() {
// Symlink traversal updates access time.
- atomic.StoreInt64(&child.inode.atime, child.inode.fs.clock.Now().Nanoseconds())
+ child.inode.touchAtime(rp.Mount())
if err := rp.HandleSymlink(symlink.target); err != nil {
return nil, err
}
@@ -400,10 +407,10 @@ func (d *dentry) open(ctx context.Context, rp *vfs.ResolvingPath, opts *vfs.Open
case *regularFile:
var fd regularFileFD
fd.LockFD.Init(&d.inode.locks)
- if err := fd.vfsfd.Init(&fd, opts.Flags, rp.Mount(), &d.vfsd, &vfs.FileDescriptionOptions{}); err != nil {
+ if err := fd.vfsfd.Init(&fd, opts.Flags, rp.Mount(), &d.vfsd, &vfs.FileDescriptionOptions{AllowDirectIO: true}); err != nil {
return nil, err
}
- if opts.Flags&linux.O_TRUNC != 0 {
+ if !afterCreate && opts.Flags&linux.O_TRUNC != 0 {
if _, err := impl.truncate(0); err != nil {
return nil, err
}
@@ -416,7 +423,7 @@ func (d *dentry) open(ctx context.Context, rp *vfs.ResolvingPath, opts *vfs.Open
}
var fd directoryFD
fd.LockFD.Init(&d.inode.locks)
- if err := fd.vfsfd.Init(&fd, opts.Flags, rp.Mount(), &d.vfsd, &vfs.FileDescriptionOptions{}); err != nil {
+ if err := fd.vfsfd.Init(&fd, opts.Flags, rp.Mount(), &d.vfsd, &vfs.FileDescriptionOptions{AllowDirectIO: true}); err != nil {
return nil, err
}
return &fd.vfsfd, nil
@@ -485,6 +492,9 @@ func (fs *filesystem) RenameAt(ctx context.Context, rp *vfs.ResolvingPath, oldPa
if !ok {
return syserror.ENOENT
}
+ if err := oldParentDir.mayDelete(rp.Credentials(), renamed); err != nil {
+ return err
+ }
// Note that we don't need to call rp.CheckMount(), since if renamed is a
// mount point then we want to rename the mount point, not anything in the
// mounted filesystem.
@@ -599,6 +609,9 @@ func (fs *filesystem) RmdirAt(ctx context.Context, rp *vfs.ResolvingPath) error
if !ok {
return syserror.ENOENT
}
+ if err := parentDir.mayDelete(rp.Credentials(), child); err != nil {
+ return err
+ }
childDir, ok := child.inode.impl.(*directory)
if !ok {
return syserror.ENOTDIR
@@ -618,7 +631,7 @@ func (fs *filesystem) RmdirAt(ctx context.Context, rp *vfs.ResolvingPath) error
return err
}
parentDir.removeChildLocked(child)
- parentDir.inode.watches.Notify(name, linux.IN_DELETE|linux.IN_ISDIR, 0, vfs.InodeEvent)
+ parentDir.inode.watches.Notify(name, linux.IN_DELETE|linux.IN_ISDIR, 0, vfs.InodeEvent, true /* unlinked */)
// Remove links for child, child/., and child/..
child.inode.decLinksLocked()
child.inode.decLinksLocked()
@@ -631,14 +644,16 @@ func (fs *filesystem) RmdirAt(ctx context.Context, rp *vfs.ResolvingPath) error
// SetStatAt implements vfs.FilesystemImpl.SetStatAt.
func (fs *filesystem) SetStatAt(ctx context.Context, rp *vfs.ResolvingPath, opts vfs.SetStatOptions) error {
fs.mu.RLock()
- defer fs.mu.RUnlock()
d, err := resolveLocked(rp)
if err != nil {
+ fs.mu.RUnlock()
return err
}
if err := d.inode.setStat(ctx, rp.Credentials(), &opts.Stat); err != nil {
+ fs.mu.RUnlock()
return err
}
+ fs.mu.RUnlock()
if ev := vfs.InotifyEventFromStatMask(opts.Stat.Mask); ev != 0 {
d.InotifyWithParent(ev, 0, vfs.InodeEvent)
@@ -707,6 +722,9 @@ func (fs *filesystem) UnlinkAt(ctx context.Context, rp *vfs.ResolvingPath) error
if !ok {
return syserror.ENOENT
}
+ if err := parentDir.mayDelete(rp.Credentials(), child); err != nil {
+ return err
+ }
if child.inode.isDir() {
return syserror.EISDIR
}
@@ -781,14 +799,16 @@ func (fs *filesystem) GetxattrAt(ctx context.Context, rp *vfs.ResolvingPath, opt
// SetxattrAt implements vfs.FilesystemImpl.SetxattrAt.
func (fs *filesystem) SetxattrAt(ctx context.Context, rp *vfs.ResolvingPath, opts vfs.SetxattrOptions) error {
fs.mu.RLock()
- defer fs.mu.RUnlock()
d, err := resolveLocked(rp)
if err != nil {
+ fs.mu.RUnlock()
return err
}
if err := d.inode.setxattr(rp.Credentials(), &opts); err != nil {
+ fs.mu.RUnlock()
return err
}
+ fs.mu.RUnlock()
d.InotifyWithParent(linux.IN_ATTRIB, 0, vfs.InodeEvent)
return nil
@@ -797,14 +817,16 @@ func (fs *filesystem) SetxattrAt(ctx context.Context, rp *vfs.ResolvingPath, opt
// RemovexattrAt implements vfs.FilesystemImpl.RemovexattrAt.
func (fs *filesystem) RemovexattrAt(ctx context.Context, rp *vfs.ResolvingPath, name string) error {
fs.mu.RLock()
- defer fs.mu.RUnlock()
d, err := resolveLocked(rp)
if err != nil {
+ fs.mu.RUnlock()
return err
}
if err := d.inode.removexattr(rp.Credentials(), name); err != nil {
+ fs.mu.RUnlock()
return err
}
+ fs.mu.RUnlock()
d.InotifyWithParent(linux.IN_ATTRIB, 0, vfs.InodeEvent)
return nil
diff --git a/pkg/sentry/fsimpl/tmpfs/regular_file.go b/pkg/sentry/fsimpl/tmpfs/regular_file.go
index bfd9c5995..1cdb46e6f 100644
--- a/pkg/sentry/fsimpl/tmpfs/regular_file.go
+++ b/pkg/sentry/fsimpl/tmpfs/regular_file.go
@@ -274,11 +274,35 @@ func (fd *regularFileFD) Release() {
// noop
}
+// Allocate implements vfs.FileDescriptionImpl.Allocate.
+func (fd *regularFileFD) Allocate(ctx context.Context, mode, offset, length uint64) error {
+ f := fd.inode().impl.(*regularFile)
+
+ f.inode.mu.Lock()
+ defer f.inode.mu.Unlock()
+ oldSize := f.size
+ size := offset + length
+ if oldSize >= size {
+ return nil
+ }
+ _, err := f.truncateLocked(size)
+ return err
+}
+
// PRead implements vfs.FileDescriptionImpl.PRead.
func (fd *regularFileFD) PRead(ctx context.Context, dst usermem.IOSequence, offset int64, opts vfs.ReadOptions) (int64, error) {
if offset < 0 {
return 0, syserror.EINVAL
}
+
+ // Check that flags are supported. RWF_DSYNC/RWF_SYNC can be ignored since
+ // all state is in-memory.
+ //
+ // TODO(gvisor.dev/issue/2601): Support select preadv2 flags.
+ if opts.Flags&^(linux.RWF_HIPRI|linux.RWF_DSYNC|linux.RWF_SYNC) != 0 {
+ return 0, syserror.EOPNOTSUPP
+ }
+
if dst.NumBytes() == 0 {
return 0, nil
}
@@ -304,6 +328,15 @@ func (fd *regularFileFD) PWrite(ctx context.Context, src usermem.IOSequence, off
if offset < 0 {
return 0, syserror.EINVAL
}
+
+ // Check that flags are supported. RWF_DSYNC/RWF_SYNC can be ignored since
+ // all state is in-memory.
+ //
+ // TODO(gvisor.dev/issue/2601): Support select preadv2 flags.
+ if opts.Flags&^(linux.RWF_HIPRI|linux.RWF_DSYNC|linux.RWF_SYNC) != 0 {
+ return 0, syserror.EOPNOTSUPP
+ }
+
srclen := src.NumBytes()
if srclen == 0 {
return 0, nil
diff --git a/pkg/sentry/fsimpl/tmpfs/tmpfs.go b/pkg/sentry/fsimpl/tmpfs/tmpfs.go
index a94333ee0..d7f4f0779 100644
--- a/pkg/sentry/fsimpl/tmpfs/tmpfs.go
+++ b/pkg/sentry/fsimpl/tmpfs/tmpfs.go
@@ -215,11 +215,6 @@ type dentry struct {
// filesystem.mu.
name string
- // unlinked indicates whether this dentry has been unlinked from its parent.
- // It is only set to true on an unlink operation, and never set from true to
- // false. unlinked is protected by filesystem.mu.
- unlinked bool
-
// dentryEntry (ugh) links dentries into their parent directory.childList.
dentryEntry
@@ -259,18 +254,22 @@ func (d *dentry) DecRef() {
}
// InotifyWithParent implements vfs.DentryImpl.InotifyWithParent.
-func (d *dentry) InotifyWithParent(events uint32, cookie uint32, et vfs.EventType) {
+func (d *dentry) InotifyWithParent(events, cookie uint32, et vfs.EventType) {
if d.inode.isDir() {
events |= linux.IN_ISDIR
}
+ // tmpfs never calls VFS.InvalidateDentry(), so d.vfsd.IsDead() indicates
+ // that d was deleted.
+ deleted := d.vfsd.IsDead()
+
+ d.inode.fs.mu.RLock()
// The ordering below is important, Linux always notifies the parent first.
if d.parent != nil {
- // Note that d.parent or d.name may be stale if there is a concurrent
- // rename operation. Inotify does not provide consistency guarantees.
- d.parent.inode.watches.NotifyWithExclusions(d.name, events, cookie, et, d.unlinked)
+ d.parent.inode.watches.Notify(d.name, events, cookie, et, deleted)
}
- d.inode.watches.Notify("", events, cookie, et)
+ d.inode.watches.Notify("", events, cookie, et, deleted)
+ d.inode.fs.mu.RUnlock()
}
// Watches implements vfs.DentryImpl.Watches.
@@ -278,6 +277,9 @@ func (d *dentry) Watches() *vfs.Watches {
return &d.inode.watches
}
+// OnZeroWatches implements vfs.Dentry.OnZeroWatches.
+func (d *dentry) OnZeroWatches() {}
+
// inode represents a filesystem object.
type inode struct {
// fs is the owning filesystem. fs is immutable.
@@ -336,7 +338,6 @@ func (i *inode) init(impl interface{}, fs *filesystem, kuid auth.KUID, kgid auth
i.ctime = now
i.mtime = now
// i.nlink initialized by caller
- i.watches = vfs.Watches{}
i.impl = impl
}
generated by cgit v1.2.3 (git 2.39.1) at 2025-04-08 12:17:35 +0000