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/benchmark_test.go8
-rw-r--r--pkg/sentry/fsimpl/tmpfs/device_file.go1
-rw-r--r--pkg/sentry/fsimpl/tmpfs/directory.go9
-rw-r--r--pkg/sentry/fsimpl/tmpfs/filesystem.go78
-rw-r--r--pkg/sentry/fsimpl/tmpfs/named_pipe.go5
-rw-r--r--pkg/sentry/fsimpl/tmpfs/pipe_test.go3
-rw-r--r--pkg/sentry/fsimpl/tmpfs/regular_file.go80
-rw-r--r--pkg/sentry/fsimpl/tmpfs/socket_file.go2
-rw-r--r--pkg/sentry/fsimpl/tmpfs/symlink.go1
-rw-r--r--pkg/sentry/fsimpl/tmpfs/tmpfs.go213
-rw-r--r--pkg/sentry/fsimpl/tmpfs/tmpfs_test.go3
11 files changed, 263 insertions, 140 deletions
diff --git a/pkg/sentry/fsimpl/tmpfs/benchmark_test.go b/pkg/sentry/fsimpl/tmpfs/benchmark_test.go
index d263147c2..3cc63e732 100644
--- a/pkg/sentry/fsimpl/tmpfs/benchmark_test.go
+++ b/pkg/sentry/fsimpl/tmpfs/benchmark_test.go
@@ -182,7 +182,7 @@ func BenchmarkVFS2TmpfsStat(b *testing.B) {
vfsObj.MustRegisterFilesystemType("tmpfs", tmpfs.FilesystemType{}, &vfs.RegisterFilesystemTypeOptions{
AllowUserMount: true,
})
- mntns, err := vfsObj.NewMountNamespace(ctx, creds, "", "tmpfs", &vfs.GetFilesystemOptions{})
+ mntns, err := vfsObj.NewMountNamespace(ctx, creds, "", "tmpfs", &vfs.MountOptions{})
if err != nil {
b.Fatalf("failed to create tmpfs root mount: %v", err)
}
@@ -193,6 +193,7 @@ func BenchmarkVFS2TmpfsStat(b *testing.B) {
// Create nested directories with given depth.
root := mntns.Root()
+ root.IncRef()
defer root.DecRef(ctx)
vd := root
vd.IncRef()
@@ -376,7 +377,7 @@ func BenchmarkVFS2TmpfsMountStat(b *testing.B) {
vfsObj.MustRegisterFilesystemType("tmpfs", tmpfs.FilesystemType{}, &vfs.RegisterFilesystemTypeOptions{
AllowUserMount: true,
})
- mntns, err := vfsObj.NewMountNamespace(ctx, creds, "", "tmpfs", &vfs.GetFilesystemOptions{})
+ mntns, err := vfsObj.NewMountNamespace(ctx, creds, "", "tmpfs", &vfs.MountOptions{})
if err != nil {
b.Fatalf("failed to create tmpfs root mount: %v", err)
}
@@ -387,6 +388,7 @@ func BenchmarkVFS2TmpfsMountStat(b *testing.B) {
// Create the mount point.
root := mntns.Root()
+ root.IncRef()
defer root.DecRef(ctx)
pop := vfs.PathOperation{
Root: root,
@@ -405,7 +407,7 @@ func BenchmarkVFS2TmpfsMountStat(b *testing.B) {
}
defer mountPoint.DecRef(ctx)
// Create and mount the submount.
- if err := vfsObj.MountAt(ctx, creds, "", &pop, "tmpfs", &vfs.MountOptions{}); err != nil {
+ if _, err := vfsObj.MountAt(ctx, creds, "", &pop, "tmpfs", &vfs.MountOptions{}); err != nil {
b.Fatalf("failed to mount tmpfs submount: %v", err)
}
filePathBuilder.WriteString(mountPointName)
diff --git a/pkg/sentry/fsimpl/tmpfs/device_file.go b/pkg/sentry/fsimpl/tmpfs/device_file.go
index ac54d420d..9129d35b7 100644
--- a/pkg/sentry/fsimpl/tmpfs/device_file.go
+++ b/pkg/sentry/fsimpl/tmpfs/device_file.go
@@ -22,6 +22,7 @@ import (
"gvisor.dev/gvisor/pkg/sentry/vfs"
)
+// +stateify savable
type deviceFile struct {
inode inode
kind vfs.DeviceKind
diff --git a/pkg/sentry/fsimpl/tmpfs/directory.go b/pkg/sentry/fsimpl/tmpfs/directory.go
index 78b4fc5be..e90669cf0 100644
--- a/pkg/sentry/fsimpl/tmpfs/directory.go
+++ b/pkg/sentry/fsimpl/tmpfs/directory.go
@@ -25,6 +25,7 @@ import (
"gvisor.dev/gvisor/pkg/syserror"
)
+// +stateify savable
type directory struct {
// Since directories can't be hard-linked, each directory can only be
// associated with a single dentry, which we can store in the directory
@@ -44,7 +45,7 @@ type directory struct {
// (with inode == nil) that represent the iteration position of
// directoryFDs. childList is used to support directoryFD.IterDirents()
// efficiently. childList is protected by iterMu.
- iterMu sync.Mutex
+ iterMu sync.Mutex `state:"nosave"`
childList dentryList
}
@@ -57,8 +58,9 @@ func (fs *filesystem) newDirectory(kuid auth.KUID, kgid auth.KGID, mode linux.Fi
return dir
}
-// Preconditions: filesystem.mu must be locked for writing. dir must not
-// already contain a child with the given name.
+// Preconditions:
+// * filesystem.mu must be locked for writing.
+// * dir must not already contain a child with the given name.
func (dir *directory) insertChildLocked(child *dentry, name string) {
child.parent = &dir.dentry
child.name = name
@@ -85,6 +87,7 @@ 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)))
}
+// +stateify savable
type directoryFD struct {
fileDescription
vfs.DirectoryFileDescriptionDefaultImpl
diff --git a/pkg/sentry/fsimpl/tmpfs/filesystem.go b/pkg/sentry/fsimpl/tmpfs/filesystem.go
index b0ec177e6..e39cd305b 100644
--- a/pkg/sentry/fsimpl/tmpfs/filesystem.go
+++ b/pkg/sentry/fsimpl/tmpfs/filesystem.go
@@ -25,7 +25,6 @@ import (
"gvisor.dev/gvisor/pkg/sentry/socket/unix/transport"
"gvisor.dev/gvisor/pkg/sentry/vfs"
"gvisor.dev/gvisor/pkg/syserror"
- "gvisor.dev/gvisor/pkg/usermem"
)
// Sync implements vfs.FilesystemImpl.Sync.
@@ -39,7 +38,9 @@ func (fs *filesystem) Sync(ctx context.Context) error {
//
// stepLocked is loosely analogous to fs/namei.c:walk_component().
//
-// Preconditions: filesystem.mu must be locked. !rp.Done().
+// Preconditions:
+// * filesystem.mu must be locked.
+// * !rp.Done().
func stepLocked(ctx context.Context, rp *vfs.ResolvingPath, d *dentry) (*dentry, error) {
dir, ok := d.inode.impl.(*directory)
if !ok {
@@ -97,7 +98,9 @@ afterSymlink:
// walkParentDirLocked is loosely analogous to Linux's
// fs/namei.c:path_parentat().
//
-// Preconditions: filesystem.mu must be locked. !rp.Done().
+// Preconditions:
+// * filesystem.mu must be locked.
+// * !rp.Done().
func walkParentDirLocked(ctx context.Context, rp *vfs.ResolvingPath, d *dentry) (*directory, error) {
for !rp.Final() {
next, err := stepLocked(ctx, rp, d)
@@ -139,8 +142,9 @@ func resolveLocked(ctx context.Context, rp *vfs.ResolvingPath) (*dentry, error)
// doCreateAt is loosely analogous to a conjunction of Linux's
// fs/namei.c:filename_create() and done_path_create().
//
-// Preconditions: !rp.Done(). For the final path component in rp,
-// !rp.ShouldFollowSymlink().
+// Preconditions:
+// * !rp.Done().
+// * For the final path component in rp, !rp.ShouldFollowSymlink().
func (fs *filesystem) doCreateAt(ctx context.Context, rp *vfs.ResolvingPath, dir bool, create func(parentDir *directory, name string) error) error {
fs.mu.Lock()
defer fs.mu.Unlock()
@@ -669,11 +673,11 @@ func (fs *filesystem) SetStatAt(ctx context.Context, rp *vfs.ResolvingPath, opts
fs.mu.RUnlock()
return err
}
- if err := d.inode.setStat(ctx, rp.Credentials(), &opts); err != nil {
- fs.mu.RUnlock()
+ err = d.inode.setStat(ctx, rp.Credentials(), &opts)
+ fs.mu.RUnlock()
+ if err != nil {
return err
}
- fs.mu.RUnlock()
if ev := vfs.InotifyEventFromStatMask(opts.Stat.Mask); ev != 0 {
d.InotifyWithParent(ctx, ev, 0, vfs.InodeEvent)
@@ -701,16 +705,7 @@ func (fs *filesystem) StatFSAt(ctx context.Context, rp *vfs.ResolvingPath) (linu
if _, err := resolveLocked(ctx, rp); err != nil {
return linux.Statfs{}, err
}
- statfs := linux.Statfs{
- Type: linux.TMPFS_MAGIC,
- BlockSize: usermem.PageSize,
- FragmentSize: usermem.PageSize,
- NameLength: linux.NAME_MAX,
- // TODO(b/29637826): Allow configuring a tmpfs size and enforce it.
- Blocks: 0,
- BlocksFree: 0,
- }
- return statfs, nil
+ return globalStatfs, nil
}
// SymlinkAt implements vfs.FilesystemImpl.SymlinkAt.
@@ -775,7 +770,7 @@ func (fs *filesystem) UnlinkAt(ctx context.Context, rp *vfs.ResolvingPath) error
return nil
}
-// BoundEndpointAt implements FilesystemImpl.BoundEndpointAt.
+// BoundEndpointAt implements vfs.FilesystemImpl.BoundEndpointAt.
func (fs *filesystem) BoundEndpointAt(ctx context.Context, rp *vfs.ResolvingPath, opts vfs.BoundEndpointOptions) (transport.BoundEndpoint, error) {
fs.mu.RLock()
defer fs.mu.RUnlock()
@@ -788,65 +783,68 @@ func (fs *filesystem) BoundEndpointAt(ctx context.Context, rp *vfs.ResolvingPath
}
switch impl := d.inode.impl.(type) {
case *socketFile:
+ if impl.ep == nil {
+ return nil, syserror.ECONNREFUSED
+ }
return impl.ep, nil
default:
return nil, syserror.ECONNREFUSED
}
}
-// ListxattrAt implements vfs.FilesystemImpl.ListxattrAt.
-func (fs *filesystem) ListxattrAt(ctx context.Context, rp *vfs.ResolvingPath, size uint64) ([]string, error) {
+// ListXattrAt implements vfs.FilesystemImpl.ListXattrAt.
+func (fs *filesystem) ListXattrAt(ctx context.Context, rp *vfs.ResolvingPath, size uint64) ([]string, error) {
fs.mu.RLock()
defer fs.mu.RUnlock()
d, err := resolveLocked(ctx, rp)
if err != nil {
return nil, err
}
- return d.inode.listxattr(size)
+ return d.inode.listXattr(size)
}
-// GetxattrAt implements vfs.FilesystemImpl.GetxattrAt.
-func (fs *filesystem) GetxattrAt(ctx context.Context, rp *vfs.ResolvingPath, opts vfs.GetxattrOptions) (string, error) {
+// GetXattrAt implements vfs.FilesystemImpl.GetXattrAt.
+func (fs *filesystem) GetXattrAt(ctx context.Context, rp *vfs.ResolvingPath, opts vfs.GetXattrOptions) (string, error) {
fs.mu.RLock()
defer fs.mu.RUnlock()
d, err := resolveLocked(ctx, rp)
if err != nil {
return "", err
}
- return d.inode.getxattr(rp.Credentials(), &opts)
+ return d.inode.getXattr(rp.Credentials(), &opts)
}
-// SetxattrAt implements vfs.FilesystemImpl.SetxattrAt.
-func (fs *filesystem) SetxattrAt(ctx context.Context, rp *vfs.ResolvingPath, opts vfs.SetxattrOptions) error {
+// SetXattrAt implements vfs.FilesystemImpl.SetXattrAt.
+func (fs *filesystem) SetXattrAt(ctx context.Context, rp *vfs.ResolvingPath, opts vfs.SetXattrOptions) error {
fs.mu.RLock()
d, err := resolveLocked(ctx, rp)
if err != nil {
fs.mu.RUnlock()
return err
}
- if err := d.inode.setxattr(rp.Credentials(), &opts); err != nil {
- fs.mu.RUnlock()
+ err = d.inode.setXattr(rp.Credentials(), &opts)
+ fs.mu.RUnlock()
+ if err != nil {
return err
}
- fs.mu.RUnlock()
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 {
+// RemoveXattrAt implements vfs.FilesystemImpl.RemoveXattrAt.
+func (fs *filesystem) RemoveXattrAt(ctx context.Context, rp *vfs.ResolvingPath, name string) error {
fs.mu.RLock()
d, err := resolveLocked(ctx, rp)
if err != nil {
fs.mu.RUnlock()
return err
}
- if err := d.inode.removexattr(rp.Credentials(), name); err != nil {
- fs.mu.RUnlock()
+ err = d.inode.removeXattr(rp.Credentials(), name)
+ fs.mu.RUnlock()
+ if err != nil {
return err
}
- fs.mu.RUnlock()
d.InotifyWithParent(ctx, linux.IN_ATTRIB, 0, vfs.InodeEvent)
return nil
@@ -867,8 +865,16 @@ func (fs *filesystem) PrependPath(ctx context.Context, vfsroot, vd vfs.VirtualDe
}
if d.parent == nil {
if d.name != "" {
- // This must be an anonymous memfd file.
+ // This file must have been created by
+ // newUnlinkedRegularFileDescription(). In Linux,
+ // mm/shmem.c:__shmem_file_setup() =>
+ // fs/file_table.c:alloc_file_pseudo() sets the created
+ // dentry's dentry_operations to anon_ops, for which d_dname ==
+ // simple_dname. fs/d_path.c:simple_dname() defines the
+ // dentry's pathname to be its name, prefixed with "/" and
+ // suffixed with " (deleted)".
b.PrependComponent("/" + d.name)
+ b.AppendString(" (deleted)")
return vfs.PrependPathSyntheticError{}
}
return vfs.PrependPathAtNonMountRootError{}
diff --git a/pkg/sentry/fsimpl/tmpfs/named_pipe.go b/pkg/sentry/fsimpl/tmpfs/named_pipe.go
index 739350cf0..d772db9e9 100644
--- a/pkg/sentry/fsimpl/tmpfs/named_pipe.go
+++ b/pkg/sentry/fsimpl/tmpfs/named_pipe.go
@@ -21,6 +21,7 @@ import (
"gvisor.dev/gvisor/pkg/usermem"
)
+// +stateify savable
type namedPipe struct {
inode inode
@@ -28,8 +29,8 @@ type namedPipe struct {
}
// Preconditions:
-// * fs.mu must be locked.
-// * rp.Mount().CheckBeginWrite() has been called successfully.
+// * fs.mu must be locked.
+// * rp.Mount().CheckBeginWrite() has been called successfully.
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, kuid, kgid, linux.S_IFIFO|mode)
diff --git a/pkg/sentry/fsimpl/tmpfs/pipe_test.go b/pkg/sentry/fsimpl/tmpfs/pipe_test.go
index ec2701d8b..2f856ce36 100644
--- a/pkg/sentry/fsimpl/tmpfs/pipe_test.go
+++ b/pkg/sentry/fsimpl/tmpfs/pipe_test.go
@@ -158,13 +158,14 @@ func setup(t *testing.T) (context.Context, *auth.Credentials, *vfs.VirtualFilesy
vfsObj.MustRegisterFilesystemType("tmpfs", FilesystemType{}, &vfs.RegisterFilesystemTypeOptions{
AllowUserMount: true,
})
- mntns, err := vfsObj.NewMountNamespace(ctx, creds, "", "tmpfs", &vfs.GetFilesystemOptions{})
+ mntns, err := vfsObj.NewMountNamespace(ctx, creds, "", "tmpfs", &vfs.MountOptions{})
if err != nil {
t.Fatalf("failed to create tmpfs root mount: %v", err)
}
// Create the pipe.
root := mntns.Root()
+ root.IncRef()
pop := vfs.PathOperation{
Root: root,
Start: root,
diff --git a/pkg/sentry/fsimpl/tmpfs/regular_file.go b/pkg/sentry/fsimpl/tmpfs/regular_file.go
index 0710b65db..ce4e3eda7 100644
--- a/pkg/sentry/fsimpl/tmpfs/regular_file.go
+++ b/pkg/sentry/fsimpl/tmpfs/regular_file.go
@@ -36,12 +36,18 @@ import (
)
// regularFile is a regular (=S_IFREG) tmpfs file.
+//
+// +stateify savable
type regularFile struct {
inode inode
// memFile is a platform.File used to allocate pages to this regularFile.
memFile *pgalloc.MemoryFile
+ // memoryUsageKind is the memory accounting category under which pages backing
+ // this regularFile's contents are accounted.
+ memoryUsageKind usage.MemoryKind
+
// mapsMu protects mappings.
mapsMu sync.Mutex `state:"nosave"`
@@ -62,7 +68,7 @@ type regularFile struct {
writableMappingPages uint64
// dataMu protects the fields below.
- dataMu sync.RWMutex
+ dataMu sync.RWMutex `state:"nosave"`
// data maps offsets into the file to offsets into memFile that store
// the file's data.
@@ -86,14 +92,75 @@ type regularFile struct {
func (fs *filesystem) newRegularFile(kuid auth.KUID, kgid auth.KGID, mode linux.FileMode) *inode {
file := &regularFile{
- memFile: fs.memFile,
- seals: linux.F_SEAL_SEAL,
+ memFile: fs.memFile,
+ memoryUsageKind: usage.Tmpfs,
+ seals: linux.F_SEAL_SEAL,
}
file.inode.init(file, fs, kuid, kgid, linux.S_IFREG|mode)
file.inode.nlink = 1 // from parent directory
return &file.inode
}
+// newUnlinkedRegularFileDescription creates a regular file on the tmpfs
+// filesystem represented by mount and returns an FD representing that file.
+// The new file is not reachable by path traversal from any other file.
+//
+// newUnlinkedRegularFileDescription is analogous to Linux's
+// mm/shmem.c:__shmem_file_setup().
+//
+// Preconditions: mount must be a tmpfs mount.
+func newUnlinkedRegularFileDescription(ctx context.Context, creds *auth.Credentials, mount *vfs.Mount, name string) (*regularFileFD, error) {
+ fs, ok := mount.Filesystem().Impl().(*filesystem)
+ if !ok {
+ panic("tmpfs.newUnlinkedRegularFileDescription() called with non-tmpfs mount")
+ }
+
+ inode := fs.newRegularFile(creds.EffectiveKUID, creds.EffectiveKGID, 0777)
+ d := fs.newDentry(inode)
+ defer d.DecRef(ctx)
+ d.name = name
+
+ fd := &regularFileFD{}
+ fd.Init(&inode.locks)
+ flags := uint32(linux.O_RDWR)
+ if err := fd.vfsfd.Init(fd, flags, mount, &d.vfsd, &vfs.FileDescriptionOptions{}); err != nil {
+ return nil, err
+ }
+ return fd, nil
+}
+
+// NewZeroFile creates a new regular file and file description as for
+// mmap(MAP_SHARED | MAP_ANONYMOUS). The file has the given size and is
+// initially (implicitly) filled with zeroes.
+//
+// Preconditions: mount must be a tmpfs mount.
+func NewZeroFile(ctx context.Context, creds *auth.Credentials, mount *vfs.Mount, size uint64) (*vfs.FileDescription, error) {
+ // Compare mm/shmem.c:shmem_zero_setup().
+ fd, err := newUnlinkedRegularFileDescription(ctx, creds, mount, "dev/zero")
+ if err != nil {
+ return nil, err
+ }
+ rf := fd.inode().impl.(*regularFile)
+ rf.memoryUsageKind = usage.Anonymous
+ rf.size = size
+ return &fd.vfsfd, err
+}
+
+// NewMemfd creates a new regular file and file description as for
+// memfd_create.
+//
+// Preconditions: mount must be a tmpfs mount.
+func NewMemfd(ctx context.Context, creds *auth.Credentials, mount *vfs.Mount, allowSeals bool, name string) (*vfs.FileDescription, error) {
+ fd, err := newUnlinkedRegularFileDescription(ctx, creds, mount, name)
+ if err != nil {
+ return nil, err
+ }
+ if allowSeals {
+ fd.inode().impl.(*regularFile).seals = 0
+ }
+ return &fd.vfsfd, nil
+}
+
// truncate grows or shrinks the file to the given size. It returns true if the
// file size was updated.
func (rf *regularFile) truncate(newSize uint64) (bool, error) {
@@ -226,7 +293,7 @@ func (rf *regularFile) Translate(ctx context.Context, required, optional memmap.
optional.End = pgend
}
- cerr := rf.data.Fill(ctx, required, optional, rf.memFile, usage.Tmpfs, func(_ context.Context, dsts safemem.BlockSeq, _ uint64) (uint64, error) {
+ cerr := rf.data.Fill(ctx, required, optional, rf.size, rf.memFile, rf.memoryUsageKind, func(_ context.Context, dsts safemem.BlockSeq, _ uint64) (uint64, error) {
// Newly-allocated pages are zeroed, so we don't need to do anything.
return dsts.NumBytes(), nil
})
@@ -260,13 +327,14 @@ func (*regularFile) InvalidateUnsavable(context.Context) error {
return nil
}
+// +stateify savable
type regularFileFD struct {
fileDescription
// off is the file offset. off is accessed using atomic memory operations.
// offMu serializes operations that may mutate off.
off int64
- offMu sync.Mutex
+ offMu sync.Mutex `state:"nosave"`
}
// Release implements vfs.FileDescriptionImpl.Release.
@@ -575,7 +643,7 @@ func (rw *regularFileReadWriter) WriteFromBlocks(srcs safemem.BlockSeq) (uint64,
case gap.Ok():
// Allocate memory for the write.
gapMR := gap.Range().Intersect(pgMR)
- fr, err := rw.file.memFile.Allocate(gapMR.Length(), usage.Tmpfs)
+ fr, err := rw.file.memFile.Allocate(gapMR.Length(), rw.file.memoryUsageKind)
if err != nil {
retErr = err
goto exitLoop
diff --git a/pkg/sentry/fsimpl/tmpfs/socket_file.go b/pkg/sentry/fsimpl/tmpfs/socket_file.go
index 3ed650474..5699d5975 100644
--- a/pkg/sentry/fsimpl/tmpfs/socket_file.go
+++ b/pkg/sentry/fsimpl/tmpfs/socket_file.go
@@ -21,6 +21,8 @@ import (
)
// socketFile is a socket (=S_IFSOCK) tmpfs file.
+//
+// +stateify savable
type socketFile struct {
inode inode
ep transport.BoundEndpoint
diff --git a/pkg/sentry/fsimpl/tmpfs/symlink.go b/pkg/sentry/fsimpl/tmpfs/symlink.go
index b0de5fabe..a102a2ee2 100644
--- a/pkg/sentry/fsimpl/tmpfs/symlink.go
+++ b/pkg/sentry/fsimpl/tmpfs/symlink.go
@@ -19,6 +19,7 @@ import (
"gvisor.dev/gvisor/pkg/sentry/kernel/auth"
)
+// +stateify savable
type symlink struct {
inode inode
target string // immutable
diff --git a/pkg/sentry/fsimpl/tmpfs/tmpfs.go b/pkg/sentry/fsimpl/tmpfs/tmpfs.go
index de2af6d01..e2a0aac69 100644
--- a/pkg/sentry/fsimpl/tmpfs/tmpfs.go
+++ b/pkg/sentry/fsimpl/tmpfs/tmpfs.go
@@ -51,9 +51,13 @@ import (
const Name = "tmpfs"
// FilesystemType implements vfs.FilesystemType.
+//
+// +stateify savable
type FilesystemType struct{}
// filesystem implements vfs.FilesystemImpl.
+//
+// +stateify savable
type filesystem struct {
vfsfs vfs.Filesystem
@@ -67,9 +71,11 @@ type filesystem struct {
devMinor uint32
// mu serializes changes to the Dentry tree.
- mu sync.RWMutex
+ mu sync.RWMutex `state:"nosave"`
nextInoMinusOne uint64 // accessed using atomic memory operations
+
+ root *dentry
}
// Name implements vfs.FilesystemType.Name.
@@ -77,7 +83,12 @@ func (FilesystemType) Name() string {
return Name
}
+// Release implements vfs.FilesystemType.Release.
+func (FilesystemType) Release(ctx context.Context) {}
+
// FilesystemOpts is used to pass configuration data to tmpfs.
+//
+// +stateify savable
type FilesystemOpts struct {
// RootFileType is the FileType of the filesystem root. Valid values
// are: S_IFDIR, S_IFREG, and S_IFLNK. Defaults to S_IFDIR.
@@ -188,6 +199,7 @@ func (fstype FilesystemType) GetFilesystem(ctx context.Context, vfsObj *vfs.Virt
fs.vfsfs.DecRef(ctx)
return nil, nil, fmt.Errorf("invalid tmpfs root file type: %#o", rootFileType)
}
+ fs.root = root
return &fs.vfsfs, &root.vfsd, nil
}
@@ -199,9 +211,61 @@ func NewFilesystem(ctx context.Context, vfsObj *vfs.VirtualFilesystem, creds *au
// Release implements vfs.FilesystemImpl.Release.
func (fs *filesystem) Release(ctx context.Context) {
fs.vfsfs.VirtualFilesystem().PutAnonBlockDevMinor(fs.devMinor)
+ fs.mu.Lock()
+ if fs.root.inode.isDir() {
+ fs.root.releaseChildrenLocked(ctx)
+ }
+ fs.mu.Unlock()
+}
+
+// releaseChildrenLocked is called on the mount point by filesystem.Release() to
+// destroy all objects in the mount. It performs a depth-first walk of the
+// filesystem and "unlinks" everything by decrementing link counts
+// appropriately. There should be no open file descriptors when this is called,
+// so each inode should only have one outstanding reference that is removed once
+// its link count hits zero.
+//
+// Note that we do not update filesystem state precisely while tearing down (for
+// instance, the child maps are ignored)--we only care to remove all remaining
+// references so that every filesystem object gets destroyed. Also note that we
+// do not need to trigger DecRef on the mount point itself or any child mount;
+// these are taken care of by the destructor of the enclosing MountNamespace.
+//
+// Precondition: filesystem.mu is held.
+func (d *dentry) releaseChildrenLocked(ctx context.Context) {
+ dir := d.inode.impl.(*directory)
+ for _, child := range dir.childMap {
+ if child.inode.isDir() {
+ child.releaseChildrenLocked(ctx)
+ child.inode.decLinksLocked(ctx) // link for child/.
+ dir.inode.decLinksLocked(ctx) // link for child/..
+ }
+ child.inode.decLinksLocked(ctx) // link for child
+ }
+}
+
+// immutable
+var globalStatfs = linux.Statfs{
+ Type: linux.TMPFS_MAGIC,
+ BlockSize: usermem.PageSize,
+ FragmentSize: usermem.PageSize,
+ NameLength: linux.NAME_MAX,
+
+ // tmpfs currently does not support configurable size limits. In Linux,
+ // such a tmpfs mount will return f_blocks == f_bfree == f_bavail == 0 from
+ // statfs(2). However, many applications treat this as having a size limit
+ // of 0. To work around this, claim to have a very large but non-zero size,
+ // chosen to ensure that BlockSize * Blocks does not overflow int64 (which
+ // applications may also handle incorrectly).
+ // TODO(b/29637826): allow configuring a tmpfs size and enforce it.
+ Blocks: math.MaxInt64 / usermem.PageSize,
+ BlocksFree: math.MaxInt64 / usermem.PageSize,
+ BlocksAvailable: math.MaxInt64 / usermem.PageSize,
}
// dentry implements vfs.DentryImpl.
+//
+// +stateify savable
type dentry struct {
vfsd vfs.Dentry
@@ -281,6 +345,8 @@ func (d *dentry) Watches() *vfs.Watches {
func (d *dentry) OnZeroWatches(context.Context) {}
// inode represents a filesystem object.
+//
+// +stateify savable
type inode struct {
// fs is the owning filesystem. fs is immutable.
fs *filesystem
@@ -297,12 +363,12 @@ type inode struct {
// Inode metadata. Writing multiple fields atomically requires holding
// mu, othewise atomic operations can be used.
- mu sync.Mutex
- mode uint32 // file type and mode
- nlink uint32 // protected by filesystem.mu instead of inode.mu
- uid uint32 // auth.KUID, but stored as raw uint32 for sync/atomic
- gid uint32 // auth.KGID, but ...
- ino uint64 // immutable
+ mu sync.Mutex `state:"nosave"`
+ mode uint32 // file type and mode
+ nlink uint32 // protected by filesystem.mu instead of inode.mu
+ uid uint32 // auth.KUID, but stored as raw uint32 for sync/atomic
+ gid uint32 // auth.KGID, but ...
+ ino uint64 // immutable
// Linux's tmpfs has no concept of btime.
atime int64 // nanoseconds
@@ -340,8 +406,10 @@ func (i *inode) init(impl interface{}, fs *filesystem, kuid auth.KUID, kgid auth
// incLinksLocked increments i's link count.
//
-// Preconditions: filesystem.mu must be locked for writing. i.nlink != 0.
-// i.nlink < maxLinks.
+// Preconditions:
+// * filesystem.mu must be locked for writing.
+// * i.nlink != 0.
+// * i.nlink < maxLinks.
func (i *inode) incLinksLocked() {
if i.nlink == 0 {
panic("tmpfs.inode.incLinksLocked() called with no existing links")
@@ -355,7 +423,9 @@ func (i *inode) incLinksLocked() {
// 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.
+// Preconditions:
+// * filesystem.mu must be locked for writing.
+// * i.nlink != 0.
func (i *inode) decLinksLocked(ctx context.Context) {
if i.nlink == 0 {
panic("tmpfs.inode.decLinksLocked() called with no existing links")
@@ -594,66 +664,59 @@ func (i *inode) touchCMtime() {
i.mu.Unlock()
}
-// Preconditions: The caller has called vfs.Mount.CheckBeginWrite() and holds
-// inode.mu.
+// Preconditions:
+// * The caller has called vfs.Mount.CheckBeginWrite().
+// * inode.mu must be locked.
func (i *inode) touchCMtimeLocked() {
now := i.fs.clock.Now().Nanoseconds()
atomic.StoreInt64(&i.mtime, now)
atomic.StoreInt64(&i.ctime, now)
}
-func (i *inode) listxattr(size uint64) ([]string, error) {
- return i.xattrs.Listxattr(size)
+func (i *inode) listXattr(size uint64) ([]string, error) {
+ return i.xattrs.ListXattr(size)
}
-func (i *inode) getxattr(creds *auth.Credentials, opts *vfs.GetxattrOptions) (string, error) {
- if err := i.checkPermissions(creds, vfs.MayRead); err != nil {
+func (i *inode) getXattr(creds *auth.Credentials, opts *vfs.GetXattrOptions) (string, error) {
+ if err := i.checkXattrPermissions(creds, opts.Name, vfs.MayRead); err != nil {
return "", err
}
- if !strings.HasPrefix(opts.Name, linux.XATTR_USER_PREFIX) {
- return "", syserror.EOPNOTSUPP
- }
- if !i.userXattrSupported() {
- return "", syserror.ENODATA
- }
- return i.xattrs.Getxattr(opts)
+ return i.xattrs.GetXattr(opts)
}
-func (i *inode) setxattr(creds *auth.Credentials, opts *vfs.SetxattrOptions) error {
- if err := i.checkPermissions(creds, vfs.MayWrite); err != nil {
+func (i *inode) setXattr(creds *auth.Credentials, opts *vfs.SetXattrOptions) error {
+ if err := i.checkXattrPermissions(creds, opts.Name, vfs.MayWrite); err != nil {
return err
}
- if !strings.HasPrefix(opts.Name, linux.XATTR_USER_PREFIX) {
- return syserror.EOPNOTSUPP
- }
- if !i.userXattrSupported() {
- return syserror.EPERM
- }
- return i.xattrs.Setxattr(opts)
+ return i.xattrs.SetXattr(opts)
}
-func (i *inode) removexattr(creds *auth.Credentials, name string) error {
- if err := i.checkPermissions(creds, vfs.MayWrite); err != nil {
+func (i *inode) removeXattr(creds *auth.Credentials, name string) error {
+ if err := i.checkXattrPermissions(creds, name, vfs.MayWrite); err != nil {
return err
}
- if !strings.HasPrefix(name, linux.XATTR_USER_PREFIX) {
+ return i.xattrs.RemoveXattr(name)
+}
+
+func (i *inode) checkXattrPermissions(creds *auth.Credentials, name string, ats vfs.AccessTypes) error {
+ // We currently only support extended attributes in the user.* and
+ // trusted.* namespaces. See b/148380782.
+ if !strings.HasPrefix(name, linux.XATTR_USER_PREFIX) && !strings.HasPrefix(name, linux.XATTR_TRUSTED_PREFIX) {
return syserror.EOPNOTSUPP
}
- if !i.userXattrSupported() {
- return syserror.EPERM
+ mode := linux.FileMode(atomic.LoadUint32(&i.mode))
+ kuid := auth.KUID(atomic.LoadUint32(&i.uid))
+ kgid := auth.KGID(atomic.LoadUint32(&i.gid))
+ if err := vfs.GenericCheckPermissions(creds, ats, mode, kuid, kgid); err != nil {
+ return err
}
- return i.xattrs.Removexattr(name)
-}
-
-// Extended attributes in the user.* namespace are only supported for regular
-// files and directories.
-func (i *inode) userXattrSupported() bool {
- filetype := linux.S_IFMT & atomic.LoadUint32(&i.mode)
- return filetype == linux.S_IFREG || filetype == linux.S_IFDIR
+ return vfs.CheckXattrPermissions(creds, ats, mode, kuid, name)
}
// fileDescription is embedded by tmpfs implementations of
// vfs.FileDescriptionImpl.
+//
+// +stateify savable
type fileDescription struct {
vfsfd vfs.FileDescription
vfs.FileDescriptionDefaultImpl
@@ -693,20 +756,25 @@ func (fd *fileDescription) SetStat(ctx context.Context, opts vfs.SetStatOptions)
return nil
}
-// Listxattr implements vfs.FileDescriptionImpl.Listxattr.
-func (fd *fileDescription) Listxattr(ctx context.Context, size uint64) ([]string, error) {
- return fd.inode().listxattr(size)
+// StatFS implements vfs.FileDescriptionImpl.StatFS.
+func (fd *fileDescription) StatFS(ctx context.Context) (linux.Statfs, error) {
+ return globalStatfs, nil
}
-// Getxattr implements vfs.FileDescriptionImpl.Getxattr.
-func (fd *fileDescription) Getxattr(ctx context.Context, opts vfs.GetxattrOptions) (string, error) {
- return fd.inode().getxattr(auth.CredentialsFromContext(ctx), &opts)
+// ListXattr implements vfs.FileDescriptionImpl.ListXattr.
+func (fd *fileDescription) ListXattr(ctx context.Context, size uint64) ([]string, error) {
+ return fd.inode().listXattr(size)
}
-// Setxattr implements vfs.FileDescriptionImpl.Setxattr.
-func (fd *fileDescription) Setxattr(ctx context.Context, opts vfs.SetxattrOptions) error {
+// GetXattr implements vfs.FileDescriptionImpl.GetXattr.
+func (fd *fileDescription) GetXattr(ctx context.Context, opts vfs.GetXattrOptions) (string, error) {
+ return fd.inode().getXattr(auth.CredentialsFromContext(ctx), &opts)
+}
+
+// SetXattr implements vfs.FileDescriptionImpl.SetXattr.
+func (fd *fileDescription) SetXattr(ctx context.Context, opts vfs.SetXattrOptions) error {
d := fd.dentry()
- if err := d.inode.setxattr(auth.CredentialsFromContext(ctx), &opts); err != nil {
+ if err := d.inode.setXattr(auth.CredentialsFromContext(ctx), &opts); err != nil {
return err
}
@@ -715,10 +783,10 @@ func (fd *fileDescription) Setxattr(ctx context.Context, opts vfs.SetxattrOption
return nil
}
-// Removexattr implements vfs.FileDescriptionImpl.Removexattr.
-func (fd *fileDescription) Removexattr(ctx context.Context, name string) error {
+// RemoveXattr implements vfs.FileDescriptionImpl.RemoveXattr.
+func (fd *fileDescription) RemoveXattr(ctx context.Context, name string) error {
d := fd.dentry()
- if err := d.inode.removexattr(auth.CredentialsFromContext(ctx), name); err != nil {
+ if err := d.inode.removeXattr(auth.CredentialsFromContext(ctx), name); err != nil {
return err
}
@@ -727,37 +795,6 @@ func (fd *fileDescription) Removexattr(ctx context.Context, name string) error {
return nil
}
-// NewMemfd creates a new tmpfs regular file and file description that can back
-// an anonymous fd created by memfd_create.
-func NewMemfd(ctx context.Context, creds *auth.Credentials, mount *vfs.Mount, allowSeals bool, name string) (*vfs.FileDescription, error) {
- fs, ok := mount.Filesystem().Impl().(*filesystem)
- if !ok {
- panic("NewMemfd() called with non-tmpfs mount")
- }
-
- // Per Linux, mm/shmem.c:__shmem_file_setup(), memfd inodes are set up with
- // S_IRWXUGO.
- inode := fs.newRegularFile(creds.EffectiveKUID, creds.EffectiveKGID, 0777)
- rf := inode.impl.(*regularFile)
- if allowSeals {
- rf.seals = 0
- }
-
- d := fs.newDentry(inode)
- defer d.DecRef(ctx)
- d.name = name
-
- // Per Linux, mm/shmem.c:__shmem_file_setup(), memfd files are set up with
- // FMODE_READ | FMODE_WRITE.
- var fd regularFileFD
- fd.Init(&inode.locks)
- flags := uint32(linux.O_RDWR)
- if err := fd.vfsfd.Init(&fd, flags, mount, &d.vfsd, &vfs.FileDescriptionOptions{}); err != nil {
- return nil, err
- }
- return &fd.vfsfd, nil
-}
-
// LockPOSIX implements vfs.FileDescriptionImpl.LockPOSIX.
func (fd *fileDescription) LockPOSIX(ctx context.Context, uid fslock.UniqueID, t fslock.LockType, start, length uint64, whence int16, block fslock.Blocker) error {
return fd.Locks().LockPOSIX(ctx, &fd.vfsfd, uid, t, start, length, whence, block)
diff --git a/pkg/sentry/fsimpl/tmpfs/tmpfs_test.go b/pkg/sentry/fsimpl/tmpfs/tmpfs_test.go
index 6f3e3ae6f..fc5323abc 100644
--- a/pkg/sentry/fsimpl/tmpfs/tmpfs_test.go
+++ b/pkg/sentry/fsimpl/tmpfs/tmpfs_test.go
@@ -41,11 +41,12 @@ func newTmpfsRoot(ctx context.Context) (*vfs.VirtualFilesystem, vfs.VirtualDentr
vfsObj.MustRegisterFilesystemType("tmpfs", FilesystemType{}, &vfs.RegisterFilesystemTypeOptions{
AllowUserMount: true,
})
- mntns, err := vfsObj.NewMountNamespace(ctx, creds, "", "tmpfs", &vfs.GetFilesystemOptions{})
+ mntns, err := vfsObj.NewMountNamespace(ctx, creds, "", "tmpfs", &vfs.MountOptions{})
if err != nil {
return nil, vfs.VirtualDentry{}, nil, fmt.Errorf("failed to create tmpfs root mount: %v", err)
}
root := mntns.Root()
+ root.IncRef()
return vfsObj, root, func() {
root.DecRef(ctx)
mntns.DecRef(ctx)
generated by cgit v1.2.3 (git 2.39.1) at 2025-04-22 14:14:18 +0000