diff options
Diffstat (limited to 'pkg/sentry/fsimpl/tmpfs')
-rw-r--r-- | pkg/sentry/fsimpl/tmpfs/BUILD | 1 | ||||
-rw-r--r-- | pkg/sentry/fsimpl/tmpfs/device_file.go | 4 | ||||
-rw-r--r-- | pkg/sentry/fsimpl/tmpfs/directory.go | 6 | ||||
-rw-r--r-- | pkg/sentry/fsimpl/tmpfs/filesystem.go | 70 | ||||
-rw-r--r-- | pkg/sentry/fsimpl/tmpfs/named_pipe.go | 4 | ||||
-rw-r--r-- | pkg/sentry/fsimpl/tmpfs/regular_file.go | 29 | ||||
-rw-r--r-- | pkg/sentry/fsimpl/tmpfs/socket_file.go | 4 | ||||
-rw-r--r-- | pkg/sentry/fsimpl/tmpfs/symlink.go | 4 | ||||
-rw-r--r-- | pkg/sentry/fsimpl/tmpfs/tmpfs.go | 191 |
9 files changed, 204 insertions, 109 deletions
diff --git a/pkg/sentry/fsimpl/tmpfs/BUILD b/pkg/sentry/fsimpl/tmpfs/BUILD index 007be1572..062321cbc 100644 --- a/pkg/sentry/fsimpl/tmpfs/BUILD +++ b/pkg/sentry/fsimpl/tmpfs/BUILD @@ -59,6 +59,7 @@ go_library( "//pkg/sentry/pgalloc", "//pkg/sentry/platform", "//pkg/sentry/socket/unix/transport", + "//pkg/sentry/uniqueid", "//pkg/sentry/usage", "//pkg/sentry/vfs", "//pkg/sentry/vfs/lock", diff --git a/pkg/sentry/fsimpl/tmpfs/device_file.go b/pkg/sentry/fsimpl/tmpfs/device_file.go index 83bf885ee..ac54d420d 100644 --- a/pkg/sentry/fsimpl/tmpfs/device_file.go +++ b/pkg/sentry/fsimpl/tmpfs/device_file.go @@ -29,7 +29,7 @@ type deviceFile struct { minor uint32 } -func (fs *filesystem) newDeviceFile(creds *auth.Credentials, mode linux.FileMode, kind vfs.DeviceKind, major, minor uint32) *inode { +func (fs *filesystem) newDeviceFile(kuid auth.KUID, kgid auth.KGID, mode linux.FileMode, kind vfs.DeviceKind, major, minor uint32) *inode { file := &deviceFile{ kind: kind, major: major, @@ -43,7 +43,7 @@ func (fs *filesystem) newDeviceFile(creds *auth.Credentials, mode linux.FileMode default: panic(fmt.Sprintf("invalid DeviceKind: %v", kind)) } - file.inode.init(file, fs, creds, mode) + file.inode.init(file, fs, kuid, kgid, mode) file.inode.nlink = 1 // from parent directory return &file.inode } diff --git a/pkg/sentry/fsimpl/tmpfs/directory.go b/pkg/sentry/fsimpl/tmpfs/directory.go index f2399981b..913b8a6c5 100644 --- a/pkg/sentry/fsimpl/tmpfs/directory.go +++ b/pkg/sentry/fsimpl/tmpfs/directory.go @@ -48,9 +48,9 @@ type directory struct { childList dentryList } -func (fs *filesystem) newDirectory(creds *auth.Credentials, mode linux.FileMode) *directory { +func (fs *filesystem) newDirectory(kuid auth.KUID, kgid auth.KGID, mode linux.FileMode) *directory { dir := &directory{} - dir.inode.init(dir, fs, creds, linux.S_IFDIR|mode) + dir.inode.init(dir, fs, kuid, kgid, linux.S_IFDIR|mode) dir.inode.nlink = 2 // from "." and parent directory or ".." for root dir.dentry.inode = &dir.inode dir.dentry.vfsd.Init(&dir.dentry) @@ -79,6 +79,7 @@ func (dir *directory) removeChildLocked(child *dentry) { dir.iterMu.Lock() dir.childList.Remove(child) dir.iterMu.Unlock() + child.unlinked = true } type directoryFD struct { @@ -112,6 +113,7 @@ func (fd *directoryFD) IterDirents(ctx context.Context, cb vfs.IterDirentsCallba 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 80fa7b29d..72399b321 100644 --- a/pkg/sentry/fsimpl/tmpfs/filesystem.go +++ b/pkg/sentry/fsimpl/tmpfs/filesystem.go @@ -177,6 +177,12 @@ func (fs *filesystem) doCreateAt(rp *vfs.ResolvingPath, dir bool, create func(pa if err := create(parentDir, name); err != nil { return err } + + ev := linux.IN_CREATE + if dir { + ev |= linux.IN_ISDIR + } + parentDir.inode.watches.Notify(name, uint32(ev), 0, vfs.InodeEvent) parentDir.inode.touchCMtime() return nil } @@ -241,6 +247,7 @@ func (fs *filesystem) LinkAt(ctx context.Context, rp *vfs.ResolvingPath, vd vfs. return syserror.EMLINK } d.inode.incLinksLocked() + d.inode.watches.Notify("", linux.IN_ATTRIB, 0, vfs.InodeEvent) parentDir.insertChildLocked(fs.newDentry(d.inode), name) return nil }) @@ -249,11 +256,12 @@ func (fs *filesystem) LinkAt(ctx context.Context, rp *vfs.ResolvingPath, vd vfs. // MkdirAt implements vfs.FilesystemImpl.MkdirAt. func (fs *filesystem) MkdirAt(ctx context.Context, rp *vfs.ResolvingPath, opts vfs.MkdirOptions) error { return fs.doCreateAt(rp, true /* dir */, func(parentDir *directory, name string) error { + creds := rp.Credentials() if parentDir.inode.nlink == maxLinks { return syserror.EMLINK } parentDir.inode.incLinksLocked() // from child's ".." - childDir := fs.newDirectory(rp.Credentials(), opts.Mode) + childDir := fs.newDirectory(creds.EffectiveKUID, creds.EffectiveKGID, opts.Mode) parentDir.insertChildLocked(&childDir.dentry, name) return nil }) @@ -262,18 +270,19 @@ func (fs *filesystem) MkdirAt(ctx context.Context, rp *vfs.ResolvingPath, opts v // MknodAt implements vfs.FilesystemImpl.MknodAt. func (fs *filesystem) MknodAt(ctx context.Context, rp *vfs.ResolvingPath, opts vfs.MknodOptions) error { return fs.doCreateAt(rp, false /* dir */, func(parentDir *directory, name string) error { + creds := rp.Credentials() var childInode *inode switch opts.Mode.FileType() { case 0, linux.S_IFREG: - childInode = fs.newRegularFile(rp.Credentials(), opts.Mode) + childInode = fs.newRegularFile(creds.EffectiveKUID, creds.EffectiveKGID, opts.Mode) case linux.S_IFIFO: - childInode = fs.newNamedPipe(rp.Credentials(), opts.Mode) + childInode = fs.newNamedPipe(creds.EffectiveKUID, creds.EffectiveKGID, opts.Mode) case linux.S_IFBLK: - childInode = fs.newDeviceFile(rp.Credentials(), opts.Mode, vfs.BlockDevice, opts.DevMajor, opts.DevMinor) + childInode = fs.newDeviceFile(creds.EffectiveKUID, creds.EffectiveKGID, opts.Mode, vfs.BlockDevice, opts.DevMajor, opts.DevMinor) case linux.S_IFCHR: - childInode = fs.newDeviceFile(rp.Credentials(), opts.Mode, vfs.CharDevice, opts.DevMajor, opts.DevMinor) + childInode = fs.newDeviceFile(creds.EffectiveKUID, creds.EffectiveKGID, opts.Mode, vfs.CharDevice, opts.DevMajor, opts.DevMinor) case linux.S_IFSOCK: - childInode = fs.newSocketFile(rp.Credentials(), opts.Mode, opts.Endpoint) + childInode = fs.newSocketFile(creds.EffectiveKUID, creds.EffectiveKGID, opts.Mode, opts.Endpoint) default: return syserror.EINVAL } @@ -348,12 +357,14 @@ afterTrailingSymlink: } defer rp.Mount().EndWrite() // Create and open the child. - child := fs.newDentry(fs.newRegularFile(rp.Credentials(), opts.Mode)) + creds := rp.Credentials() + child := fs.newDentry(fs.newRegularFile(creds.EffectiveKUID, creds.EffectiveKGID, opts.Mode)) parentDir.insertChildLocked(child, name) fd, err := child.open(ctx, rp, &opts, true) if err != nil { return nil, err } + parentDir.inode.watches.Notify(name, linux.IN_CREATE, 0, vfs.PathEvent) parentDir.inode.touchCMtime() return fd, nil } @@ -388,6 +399,7 @@ func (d *dentry) open(ctx context.Context, rp *vfs.ResolvingPath, opts *vfs.Open switch impl := d.inode.impl.(type) { 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 { return nil, err } @@ -403,15 +415,16 @@ func (d *dentry) open(ctx context.Context, rp *vfs.ResolvingPath, opts *vfs.Open return nil, syserror.EISDIR } 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 { return nil, err } return &fd.vfsfd, nil case *symlink: - // Can't open symlinks without O_PATH (which is unimplemented). + // TODO(gvisor.dev/issue/2782): Can't open symlinks without O_PATH. return nil, syserror.ELOOP case *namedPipe: - return impl.pipe.Open(ctx, rp.Mount(), &d.vfsd, opts.Flags) + return impl.pipe.Open(ctx, rp.Mount(), &d.vfsd, opts.Flags, &d.inode.locks) case *deviceFile: return rp.VirtualFilesystem().OpenDeviceSpecialFile(ctx, rp.Mount(), &d.vfsd, impl.kind, impl.major, impl.minor, opts) case *socketFile: @@ -559,6 +572,8 @@ func (fs *filesystem) RenameAt(ctx context.Context, rp *vfs.ResolvingPath, oldPa newParentDir.inode.touchCMtime() } renamed.inode.touchCtime() + + vfs.InotifyRename(ctx, &renamed.inode.watches, &oldParentDir.inode.watches, &newParentDir.inode.watches, oldName, newName, renamed.inode.isDir()) return nil } @@ -603,8 +618,11 @@ func (fs *filesystem) RmdirAt(ctx context.Context, rp *vfs.ResolvingPath) error return err } parentDir.removeChildLocked(child) - parentDir.inode.decLinksLocked() // from child's ".." + parentDir.inode.watches.Notify(name, linux.IN_DELETE|linux.IN_ISDIR, 0, vfs.InodeEvent) + // Remove links for child, child/., and child/.. child.inode.decLinksLocked() + child.inode.decLinksLocked() + parentDir.inode.decLinksLocked() vfsObj.CommitDeleteDentry(&child.vfsd) parentDir.inode.touchCMtime() return nil @@ -618,7 +636,14 @@ func (fs *filesystem) SetStatAt(ctx context.Context, rp *vfs.ResolvingPath, opts if err != nil { return err } - return d.inode.setStat(ctx, rp.Credentials(), &opts.Stat) + if err := d.inode.setStat(ctx, rp.Credentials(), &opts.Stat); err != nil { + return err + } + + if ev := vfs.InotifyEventFromStatMask(opts.Stat.Mask); ev != 0 { + d.InotifyWithParent(ev, 0, vfs.InodeEvent) + } + return nil } // StatAt implements vfs.FilesystemImpl.StatAt. @@ -656,7 +681,8 @@ func (fs *filesystem) StatFSAt(ctx context.Context, rp *vfs.ResolvingPath) (linu // SymlinkAt implements vfs.FilesystemImpl.SymlinkAt. func (fs *filesystem) SymlinkAt(ctx context.Context, rp *vfs.ResolvingPath, target string) error { return fs.doCreateAt(rp, false /* dir */, func(parentDir *directory, name string) error { - child := fs.newDentry(fs.newSymlink(rp.Credentials(), target)) + creds := rp.Credentials() + child := fs.newDentry(fs.newSymlink(creds.EffectiveKUID, creds.EffectiveKGID, 0777, target)) parentDir.insertChildLocked(child, name) return nil }) @@ -698,6 +724,12 @@ func (fs *filesystem) UnlinkAt(ctx context.Context, rp *vfs.ResolvingPath) error if err := vfsObj.PrepareDeleteDentry(mntns, &child.vfsd); err != nil { return err } + + // Generate inotify events. Note that this must take place before the link + // count of the child is decremented, or else the watches may be dropped + // before these events are added. + vfs.InotifyRemoveChild(&child.inode.watches, &parentDir.inode.watches, name) + parentDir.removeChildLocked(child) child.inode.decLinksLocked() vfsObj.CommitDeleteDentry(&child.vfsd) @@ -754,7 +786,12 @@ func (fs *filesystem) SetxattrAt(ctx context.Context, rp *vfs.ResolvingPath, opt if err != nil { return err } - return d.inode.setxattr(rp.Credentials(), &opts) + if err := d.inode.setxattr(rp.Credentials(), &opts); err != nil { + return err + } + + d.InotifyWithParent(linux.IN_ATTRIB, 0, vfs.InodeEvent) + return nil } // RemovexattrAt implements vfs.FilesystemImpl.RemovexattrAt. @@ -765,7 +802,12 @@ func (fs *filesystem) RemovexattrAt(ctx context.Context, rp *vfs.ResolvingPath, if err != nil { return err } - return d.inode.removexattr(rp.Credentials(), name) + if err := d.inode.removexattr(rp.Credentials(), name); err != nil { + return err + } + + d.InotifyWithParent(linux.IN_ATTRIB, 0, vfs.InodeEvent) + return nil } // PrependPath implements vfs.FilesystemImpl.PrependPath. diff --git a/pkg/sentry/fsimpl/tmpfs/named_pipe.go b/pkg/sentry/fsimpl/tmpfs/named_pipe.go index 8d77b3fa8..739350cf0 100644 --- a/pkg/sentry/fsimpl/tmpfs/named_pipe.go +++ b/pkg/sentry/fsimpl/tmpfs/named_pipe.go @@ -30,9 +30,9 @@ type namedPipe struct { // Preconditions: // * fs.mu must be locked. // * rp.Mount().CheckBeginWrite() has been called successfully. -func (fs *filesystem) newNamedPipe(creds *auth.Credentials, mode linux.FileMode) *inode { +func (fs *filesystem) newNamedPipe(kuid auth.KUID, kgid auth.KGID, mode linux.FileMode) *inode { file := &namedPipe{pipe: pipe.NewVFSPipe(true /* isNamed */, pipe.DefaultPipeSize, usermem.PageSize)} - file.inode.init(file, fs, creds, linux.S_IFIFO|mode) + file.inode.init(file, fs, kuid, kgid, linux.S_IFIFO|mode) file.inode.nlink = 1 // Only the parent has a link. return &file.inode } diff --git a/pkg/sentry/fsimpl/tmpfs/regular_file.go b/pkg/sentry/fsimpl/tmpfs/regular_file.go index 3f433d666..77447b32c 100644 --- a/pkg/sentry/fsimpl/tmpfs/regular_file.go +++ b/pkg/sentry/fsimpl/tmpfs/regular_file.go @@ -25,7 +25,6 @@ import ( "gvisor.dev/gvisor/pkg/safemem" "gvisor.dev/gvisor/pkg/sentry/fs" "gvisor.dev/gvisor/pkg/sentry/fs/fsutil" - "gvisor.dev/gvisor/pkg/sentry/fs/lock" "gvisor.dev/gvisor/pkg/sentry/kernel/auth" "gvisor.dev/gvisor/pkg/sentry/memmap" "gvisor.dev/gvisor/pkg/sentry/pgalloc" @@ -85,12 +84,12 @@ type regularFile struct { size uint64 } -func (fs *filesystem) newRegularFile(creds *auth.Credentials, mode linux.FileMode) *inode { +func (fs *filesystem) newRegularFile(kuid auth.KUID, kgid auth.KGID, mode linux.FileMode) *inode { file := ®ularFile{ memFile: fs.memFile, seals: linux.F_SEAL_SEAL, } - file.inode.init(file, fs, creds, linux.S_IFREG|mode) + file.inode.init(file, fs, kuid, kgid, linux.S_IFREG|mode) file.inode.nlink = 1 // from parent directory return &file.inode } @@ -312,7 +311,7 @@ func (fd *regularFileFD) PWrite(ctx context.Context, src usermem.IOSequence, off f := fd.inode().impl.(*regularFile) if end := offset + srclen; end < offset { // Overflow. - return 0, syserror.EFBIG + return 0, syserror.EINVAL } var err error @@ -366,28 +365,6 @@ func (fd *regularFileFD) Sync(ctx context.Context) error { return nil } -// LockBSD implements vfs.FileDescriptionImpl.LockBSD. -func (fd *regularFileFD) LockBSD(ctx context.Context, uid lock.UniqueID, t lock.LockType, block lock.Blocker) error { - return fd.inode().lockBSD(uid, t, block) -} - -// UnlockBSD implements vfs.FileDescriptionImpl.UnlockBSD. -func (fd *regularFileFD) UnlockBSD(ctx context.Context, uid lock.UniqueID) error { - fd.inode().unlockBSD(uid) - return nil -} - -// LockPOSIX implements vfs.FileDescriptionImpl.LockPOSIX. -func (fd *regularFileFD) LockPOSIX(ctx context.Context, uid lock.UniqueID, t lock.LockType, rng lock.LockRange, block lock.Blocker) error { - return fd.inode().lockPOSIX(uid, t, rng, block) -} - -// UnlockPOSIX implements vfs.FileDescriptionImpl.UnlockPOSIX. -func (fd *regularFileFD) UnlockPOSIX(ctx context.Context, uid lock.UniqueID, rng lock.LockRange) error { - fd.inode().unlockPOSIX(uid, rng) - return nil -} - // ConfigureMMap implements vfs.FileDescriptionImpl.ConfigureMMap. func (fd *regularFileFD) ConfigureMMap(ctx context.Context, opts *memmap.MMapOpts) error { file := fd.inode().impl.(*regularFile) diff --git a/pkg/sentry/fsimpl/tmpfs/socket_file.go b/pkg/sentry/fsimpl/tmpfs/socket_file.go index 25c2321af..3ed650474 100644 --- a/pkg/sentry/fsimpl/tmpfs/socket_file.go +++ b/pkg/sentry/fsimpl/tmpfs/socket_file.go @@ -26,9 +26,9 @@ type socketFile struct { ep transport.BoundEndpoint } -func (fs *filesystem) newSocketFile(creds *auth.Credentials, mode linux.FileMode, ep transport.BoundEndpoint) *inode { +func (fs *filesystem) newSocketFile(kuid auth.KUID, kgid auth.KGID, mode linux.FileMode, ep transport.BoundEndpoint) *inode { file := &socketFile{ep: ep} - file.inode.init(file, fs, creds, mode) + file.inode.init(file, fs, kuid, kgid, mode) file.inode.nlink = 1 // from parent directory return &file.inode } diff --git a/pkg/sentry/fsimpl/tmpfs/symlink.go b/pkg/sentry/fsimpl/tmpfs/symlink.go index 47e075ed4..b0de5fabe 100644 --- a/pkg/sentry/fsimpl/tmpfs/symlink.go +++ b/pkg/sentry/fsimpl/tmpfs/symlink.go @@ -24,11 +24,11 @@ type symlink struct { target string // immutable } -func (fs *filesystem) newSymlink(creds *auth.Credentials, target string) *inode { +func (fs *filesystem) newSymlink(kuid auth.KUID, kgid auth.KGID, mode linux.FileMode, target string) *inode { link := &symlink{ target: target, } - link.inode.init(link, fs, creds, linux.S_IFLNK|0777) + link.inode.init(link, fs, kuid, kgid, linux.S_IFLNK|mode) link.inode.nlink = 1 // from parent directory return &link.inode } diff --git a/pkg/sentry/fsimpl/tmpfs/tmpfs.go b/pkg/sentry/fsimpl/tmpfs/tmpfs.go index 1e781aecd..71a7522af 100644 --- a/pkg/sentry/fsimpl/tmpfs/tmpfs.go +++ b/pkg/sentry/fsimpl/tmpfs/tmpfs.go @@ -30,12 +30,12 @@ package tmpfs import ( "fmt" "math" + "strconv" "strings" "sync/atomic" "gvisor.dev/gvisor/pkg/abi/linux" "gvisor.dev/gvisor/pkg/context" - fslock "gvisor.dev/gvisor/pkg/sentry/fs/lock" "gvisor.dev/gvisor/pkg/sentry/kernel/auth" "gvisor.dev/gvisor/pkg/sentry/kernel/time" "gvisor.dev/gvisor/pkg/sentry/pgalloc" @@ -112,6 +112,58 @@ func (fstype FilesystemType) GetFilesystem(ctx context.Context, vfsObj *vfs.Virt } } + mopts := vfs.GenericParseMountOptions(opts.Data) + rootMode := linux.FileMode(0777) + if rootFileType == linux.S_IFDIR { + rootMode = 01777 + } + modeStr, ok := mopts["mode"] + if ok { + delete(mopts, "mode") + mode, err := strconv.ParseUint(modeStr, 8, 32) + if err != nil { + ctx.Warningf("tmpfs.FilesystemType.GetFilesystem: invalid mode: %q", modeStr) + return nil, nil, syserror.EINVAL + } + rootMode = linux.FileMode(mode & 07777) + } + rootKUID := creds.EffectiveKUID + uidStr, ok := mopts["uid"] + if ok { + delete(mopts, "uid") + uid, err := strconv.ParseUint(uidStr, 10, 32) + if err != nil { + ctx.Warningf("tmpfs.FilesystemType.GetFilesystem: invalid uid: %q", uidStr) + return nil, nil, syserror.EINVAL + } + kuid := creds.UserNamespace.MapToKUID(auth.UID(uid)) + if !kuid.Ok() { + ctx.Warningf("tmpfs.FilesystemType.GetFilesystem: unmapped uid: %d", uid) + return nil, nil, syserror.EINVAL + } + rootKUID = kuid + } + rootKGID := creds.EffectiveKGID + gidStr, ok := mopts["gid"] + if ok { + delete(mopts, "gid") + gid, err := strconv.ParseUint(gidStr, 10, 32) + if err != nil { + ctx.Warningf("tmpfs.FilesystemType.GetFilesystem: invalid gid: %q", gidStr) + return nil, nil, syserror.EINVAL + } + kgid := creds.UserNamespace.MapToKGID(auth.GID(gid)) + if !kgid.Ok() { + ctx.Warningf("tmpfs.FilesystemType.GetFilesystem: unmapped gid: %d", gid) + return nil, nil, syserror.EINVAL + } + rootKGID = kgid + } + if len(mopts) != 0 { + ctx.Warningf("tmpfs.FilesystemType.GetFilesystem: unknown options: %v", mopts) + return nil, nil, syserror.EINVAL + } + devMinor, err := vfsObj.GetAnonBlockDevMinor() if err != nil { return nil, nil, err @@ -127,11 +179,11 @@ func (fstype FilesystemType) GetFilesystem(ctx context.Context, vfsObj *vfs.Virt var root *dentry switch rootFileType { case linux.S_IFREG: - root = fs.newDentry(fs.newRegularFile(creds, 0777)) + root = fs.newDentry(fs.newRegularFile(rootKUID, rootKGID, rootMode)) case linux.S_IFLNK: - root = fs.newDentry(fs.newSymlink(creds, tmpfsOpts.RootSymlinkTarget)) + root = fs.newDentry(fs.newSymlink(rootKUID, rootKGID, rootMode, tmpfsOpts.RootSymlinkTarget)) case linux.S_IFDIR: - root = &fs.newDirectory(creds, 01777).dentry + root = &fs.newDirectory(rootKUID, rootKGID, rootMode).dentry default: fs.vfsfs.DecRef() return nil, nil, fmt.Errorf("invalid tmpfs root file type: %#o", rootFileType) @@ -163,6 +215,11 @@ 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 @@ -201,6 +258,26 @@ func (d *dentry) DecRef() { d.inode.decRef() } +// InotifyWithParent implements vfs.DentryImpl.InotifyWithParent. +func (d *dentry) InotifyWithParent(events uint32, cookie uint32, et vfs.EventType) { + if d.inode.isDir() { + events |= linux.IN_ISDIR + } + + // 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.inode.watches.Notify("", events, cookie, et) +} + +// Watches implements vfs.DentryImpl.Watches. +func (d *dentry) Watches() *vfs.Watches { + return &d.inode.watches +} + // inode represents a filesystem object. type inode struct { // fs is the owning filesystem. fs is immutable. @@ -209,11 +286,9 @@ type inode struct { // refs is a reference count. refs is accessed using atomic memory // operations. // - // A reference is held on all inodes that are reachable in the filesystem - // tree. For non-directories (which may have multiple hard links), this - // means that a reference is dropped when nlink reaches 0. For directories, - // nlink never reaches 0 due to the "." entry; instead, - // filesystem.RmdirAt() drops the reference. + // A reference is held on all inodes as long as they are reachable in the + // filesystem tree, i.e. nlink is nonzero. This reference is dropped when + // nlink reaches 0. refs int64 // xattrs implements extended attributes. @@ -235,23 +310,25 @@ type inode struct { ctime int64 // nanoseconds mtime int64 // nanoseconds - // Advisory file locks, which lock at the inode level. locks lock.FileLocks + // Inotify watches for this inode. + watches vfs.Watches + impl interface{} // immutable } const maxLinks = math.MaxUint32 -func (i *inode) init(impl interface{}, fs *filesystem, creds *auth.Credentials, mode linux.FileMode) { +func (i *inode) init(impl interface{}, fs *filesystem, kuid auth.KUID, kgid auth.KGID, mode linux.FileMode) { if mode.FileType() == 0 { panic("file type is required in FileMode") } i.fs = fs i.refs = 1 i.mode = uint32(mode) - i.uid = uint32(creds.EffectiveKUID) - i.gid = uint32(creds.EffectiveKGID) + i.uid = uint32(kuid) + i.gid = uint32(kgid) i.ino = atomic.AddUint64(&fs.nextInoMinusOne, 1) // Tmpfs creation sets atime, ctime, and mtime to current time. now := fs.clock.Now().Nanoseconds() @@ -259,6 +336,7 @@ func (i *inode) init(impl interface{}, fs *filesystem, creds *auth.Credentials, i.ctime = now i.mtime = now // i.nlink initialized by caller + i.watches = vfs.Watches{} i.impl = impl } @@ -276,14 +354,17 @@ func (i *inode) incLinksLocked() { atomic.AddUint32(&i.nlink, 1) } -// decLinksLocked decrements i's link count. +// decLinksLocked decrements i's link count. If the link count reaches 0, we +// remove a reference on i as well. // // Preconditions: filesystem.mu must be locked for writing. i.nlink != 0. func (i *inode) decLinksLocked() { if i.nlink == 0 { panic("tmpfs.inode.decLinksLocked() called with no existing links") } - atomic.AddUint32(&i.nlink, ^uint32(0)) + if atomic.AddUint32(&i.nlink, ^uint32(0)) == 0 { + i.decRef() + } } func (i *inode) incRef() { @@ -306,6 +387,7 @@ func (i *inode) tryIncRef() bool { func (i *inode) decRef() { if refs := atomic.AddInt64(&i.refs, -1); refs == 0 { + i.watches.HandleDeletion() if regFile, ok := i.impl.(*regularFile); ok { // Release memory used by regFile to store data. Since regFile is // no longer usable, we don't need to grab any locks or update any @@ -455,44 +537,6 @@ func (i *inode) setStat(ctx context.Context, creds *auth.Credentials, stat *linu return nil } -// TODO(gvisor.dev/issue/1480): support file locking for file types other than regular. -func (i *inode) lockBSD(uid fslock.UniqueID, t fslock.LockType, block fslock.Blocker) error { - switch i.impl.(type) { - case *regularFile: - return i.locks.LockBSD(uid, t, block) - } - return syserror.EBADF -} - -// TODO(gvisor.dev/issue/1480): support file locking for file types other than regular. -func (i *inode) unlockBSD(uid fslock.UniqueID) error { - switch i.impl.(type) { - case *regularFile: - i.locks.UnlockBSD(uid) - return nil - } - return syserror.EBADF -} - -// TODO(gvisor.dev/issue/1480): support file locking for file types other than regular. -func (i *inode) lockPOSIX(uid fslock.UniqueID, t fslock.LockType, rng fslock.LockRange, block fslock.Blocker) error { - switch i.impl.(type) { - case *regularFile: - return i.locks.LockPOSIX(uid, t, rng, block) - } - return syserror.EBADF -} - -// TODO(gvisor.dev/issue/1480): support file locking for file types other than regular. -func (i *inode) unlockPOSIX(uid fslock.UniqueID, rng fslock.LockRange) error { - switch i.impl.(type) { - case *regularFile: - i.locks.UnlockPOSIX(uid, rng) - return nil - } - return syserror.EBADF -} - // allocatedBlocksForSize returns the number of 512B blocks needed to // accommodate the given size in bytes, as appropriate for struct // stat::st_blocks and struct statx::stx_blocks. (Note that this 512B block @@ -531,6 +575,9 @@ func (i *inode) isDir() bool { } func (i *inode) touchAtime(mnt *vfs.Mount) { + if mnt.Flags.NoATime { + return + } if err := mnt.CheckBeginWrite(); err != nil { return } @@ -621,14 +668,19 @@ func (i *inode) userXattrSupported() bool { type fileDescription struct { vfsfd vfs.FileDescription vfs.FileDescriptionDefaultImpl + vfs.LockFD } func (fd *fileDescription) filesystem() *filesystem { return fd.vfsfd.Mount().Filesystem().Impl().(*filesystem) } +func (fd *fileDescription) dentry() *dentry { + return fd.vfsfd.Dentry().Impl().(*dentry) +} + func (fd *fileDescription) inode() *inode { - return fd.vfsfd.Dentry().Impl().(*dentry).inode + return fd.dentry().inode } // Stat implements vfs.FileDescriptionImpl.Stat. @@ -641,7 +693,15 @@ func (fd *fileDescription) Stat(ctx context.Context, opts vfs.StatOptions) (linu // SetStat implements vfs.FileDescriptionImpl.SetStat. func (fd *fileDescription) SetStat(ctx context.Context, opts vfs.SetStatOptions) error { creds := auth.CredentialsFromContext(ctx) - return fd.inode().setStat(ctx, creds, &opts.Stat) + d := fd.dentry() + if err := d.inode.setStat(ctx, creds, &opts.Stat); err != nil { + return err + } + + if ev := vfs.InotifyEventFromStatMask(opts.Stat.Mask); ev != 0 { + d.InotifyWithParent(ev, 0, vfs.InodeEvent) + } + return nil } // Listxattr implements vfs.FileDescriptionImpl.Listxattr. @@ -656,12 +716,26 @@ func (fd *fileDescription) Getxattr(ctx context.Context, opts vfs.GetxattrOption // Setxattr implements vfs.FileDescriptionImpl.Setxattr. func (fd *fileDescription) Setxattr(ctx context.Context, opts vfs.SetxattrOptions) error { - return fd.inode().setxattr(auth.CredentialsFromContext(ctx), &opts) + d := fd.dentry() + if err := d.inode.setxattr(auth.CredentialsFromContext(ctx), &opts); err != nil { + return err + } + + // Generate inotify events. + d.InotifyWithParent(linux.IN_ATTRIB, 0, vfs.InodeEvent) + return nil } // Removexattr implements vfs.FileDescriptionImpl.Removexattr. func (fd *fileDescription) Removexattr(ctx context.Context, name string) error { - return fd.inode().removexattr(auth.CredentialsFromContext(ctx), name) + d := fd.dentry() + if err := d.inode.removexattr(auth.CredentialsFromContext(ctx), name); err != nil { + return err + } + + // Generate inotify events. + d.InotifyWithParent(linux.IN_ATTRIB, 0, vfs.InodeEvent) + return nil } // NewMemfd creates a new tmpfs regular file and file description that can back @@ -674,8 +748,7 @@ func NewMemfd(mount *vfs.Mount, creds *auth.Credentials, allowSeals bool, name s // Per Linux, mm/shmem.c:__shmem_file_setup(), memfd inodes are set up with // S_IRWXUGO. - mode := linux.FileMode(0777) - inode := fs.newRegularFile(creds, mode) + inode := fs.newRegularFile(creds.EffectiveKUID, creds.EffectiveKGID, 0777) rf := inode.impl.(*regularFile) if allowSeals { rf.seals = 0 |