Minor VFS2 interface changes.

- Remove the Filesystem argument from DentryImpl.*Ref(); in general DentryImpls
  that need the Filesystem for reference counting will probably also need it
  for other interface methods that don't plumb Filesystem, so it's easier to
  just store a pointer to the filesystem in the DentryImpl.

- Add a pointer to the VirtualFilesystem to Filesystem, which is needed by the
  gofer client to disown dentries for cache eviction triggered by dentry
  reference count changes.

- Rename FilesystemType.NewFilesystem to GetFilesystem; in some cases (e.g.
  sysfs, cgroupfs) it's much cleaner for there to be only one Filesystem that
  is used by all mounts, and in at least one case (devtmpfs) it's visibly
  incorrect not to do so, so NewFilesystem doesn't always actually create and
  return a *new* Filesystem.

- Require callers of FileDescription.Init() to increment Mount/Dentry
  references. This is because the gofer client may, in the OpenAt() path, take
  a reference on a dentry with 0 references, which is safe due to
  synchronization that is outside the scope of this CL, and it would be safer
  to still have its implementation of DentryImpl.IncRef() check for an
  increment for 0 references in other cases.

- Add FileDescription.TryIncRef. This is used by the gofer client to take
  references on "special file descriptions" (FDs for files such as pipes,
  sockets, and devices), which use per-FD handles (fids) instead of
  dentry-shared handles, for sync() and syncfs().

PiperOrigin-RevId: 282473364

10 files changed, 50 insertions, 38 deletions

diff --git a/pkg/sentry/fsimpl/ext/benchmark/benchmark_test.go b/pkg/sentry/fsimpl/ext/benchmark/benchmark_test.go
index 10a8083a0..94cd74095 100644
--- a/pkg/sentry/fsimpl/ext/benchmark/benchmark_test.go
+++ b/pkg/sentry/fsimpl/ext/benchmark/benchmark_test.go
@@ -50,7 +50,7 @@ func setUp(b *testing.B, imagePath string) (context.Context, *vfs.VirtualFilesys
 	// Create VFS.
 	vfsObj := vfs.New()
 	vfsObj.MustRegisterFilesystemType("extfs", ext.FilesystemType{})
-	mntns, err := vfsObj.NewMountNamespace(ctx, creds, imagePath, "extfs", &vfs.NewFilesystemOptions{InternalData: int(f.Fd())})
+	mntns, err := vfsObj.NewMountNamespace(ctx, creds, imagePath, "extfs", &vfs.GetFilesystemOptions{InternalData: int(f.Fd())})
 	if err != nil {
 		f.Close()
 		return nil, nil, nil, nil, err
@@ -81,7 +81,7 @@ func mount(b *testing.B, imagePath string, vfsfs *vfs.VirtualFilesystem, pop *vf
 	ctx := contexttest.Context(b)
 	creds := auth.CredentialsFromContext(ctx)
 
-	if err := vfsfs.NewMount(ctx, creds, imagePath, pop, "extfs", &vfs.NewFilesystemOptions{InternalData: int(f.Fd())}); err != nil {
+	if err := vfsfs.NewMount(ctx, creds, imagePath, pop, "extfs", &vfs.GetFilesystemOptions{InternalData: int(f.Fd())}); err != nil {
 		b.Fatalf("failed to mount tmpfs submount: %v", err)
 	}
 	return func() {
diff --git a/pkg/sentry/fsimpl/ext/block_map_file.go b/pkg/sentry/fsimpl/ext/block_map_file.go
index cea89bcd9..a2d8c3ad6 100644
--- a/pkg/sentry/fsimpl/ext/block_map_file.go
+++ b/pkg/sentry/fsimpl/ext/block_map_file.go
@@ -154,7 +154,7 @@ func (f *blockMapFile) read(curPhyBlk uint32, relFileOff uint64, height uint, ds
 			toRead = len(dst)
 		}
 
-		n, _ := f.regFile.inode.dev.ReadAt(dst[:toRead], curPhyBlkOff+int64(relFileOff))
+		n, _ := f.regFile.inode.fs.dev.ReadAt(dst[:toRead], curPhyBlkOff+int64(relFileOff))
 		if n < toRead {
 			return n, syserror.EIO
 		}
@@ -174,7 +174,7 @@ func (f *blockMapFile) read(curPhyBlk uint32, relFileOff uint64, height uint, ds
 	curChildOff := relFileOff % childCov
 	for i := startIdx; i < endIdx; i++ {
 		var childPhyBlk uint32
-		err := readFromDisk(f.regFile.inode.dev, curPhyBlkOff+int64(i*4), &childPhyBlk)
+		err := readFromDisk(f.regFile.inode.fs.dev, curPhyBlkOff+int64(i*4), &childPhyBlk)
 		if err != nil {
 			return read, err
 		}
diff --git a/pkg/sentry/fsimpl/ext/block_map_test.go b/pkg/sentry/fsimpl/ext/block_map_test.go
index 213aa3919..181727ef7 100644
--- a/pkg/sentry/fsimpl/ext/block_map_test.go
+++ b/pkg/sentry/fsimpl/ext/block_map_test.go
@@ -87,12 +87,14 @@ func blockMapSetUp(t *testing.T) (*blockMapFile, []byte) {
 	mockDisk := make([]byte, mockBMDiskSize)
 	regFile := regularFile{
 		inode: inode{
+			fs: &filesystem{
+				dev: bytes.NewReader(mockDisk),
+			},
 			diskInode: &disklayout.InodeNew{
 				InodeOld: disklayout.InodeOld{
 					SizeLo: getMockBMFileFize(),
 				},
 			},
-			dev:     bytes.NewReader(mockDisk),
 			blkSize: uint64(mockBMBlkSize),
 		},
 	}
diff --git a/pkg/sentry/fsimpl/ext/dentry.go b/pkg/sentry/fsimpl/ext/dentry.go
index 054fb42b6..a080cb189 100644
--- a/pkg/sentry/fsimpl/ext/dentry.go
+++ b/pkg/sentry/fsimpl/ext/dentry.go
@@ -41,16 +41,18 @@ func newDentry(in *inode) *dentry {
 }
 
 // IncRef implements vfs.DentryImpl.IncRef.
-func (d *dentry) IncRef(vfsfs *vfs.Filesystem) {
+func (d *dentry) IncRef() {
 	d.inode.incRef()
 }
 
 // TryIncRef implements vfs.DentryImpl.TryIncRef.
-func (d *dentry) TryIncRef(vfsfs *vfs.Filesystem) bool {
+func (d *dentry) TryIncRef() bool {
 	return d.inode.tryIncRef()
 }
 
 // DecRef implements vfs.DentryImpl.DecRef.
-func (d *dentry) DecRef(vfsfs *vfs.Filesystem) {
-	d.inode.decRef(vfsfs.Impl().(*filesystem))
+func (d *dentry) DecRef() {
+	// FIXME(b/134676337): filesystem.mu may not be locked as required by
+	// inode.decRef().
+	d.inode.decRef()
 }
diff --git a/pkg/sentry/fsimpl/ext/ext.go b/pkg/sentry/fsimpl/ext/ext.go
index f10accafc..4b7d17dc6 100644
--- a/pkg/sentry/fsimpl/ext/ext.go
+++ b/pkg/sentry/fsimpl/ext/ext.go
@@ -40,14 +40,14 @@ var _ vfs.FilesystemType = (*FilesystemType)(nil)
 // Currently there are two ways of mounting an ext(2/3/4) fs:
 //   1. Specify a mount with our internal special MountType in the OCI spec.
 //   2. Expose the device to the container and mount it from application layer.
-func getDeviceFd(source string, opts vfs.NewFilesystemOptions) (io.ReaderAt, error) {
+func getDeviceFd(source string, opts vfs.GetFilesystemOptions) (io.ReaderAt, error) {
 	if opts.InternalData == nil {
 		// User mount call.
 		// TODO(b/134676337): Open the device specified by `source` and return that.
 		panic("unimplemented")
 	}
 
-	// NewFilesystem call originated from within the sentry.
+	// GetFilesystem call originated from within the sentry.
 	devFd, ok := opts.InternalData.(int)
 	if !ok {
 		return nil, errors.New("internal data for ext fs must be an int containing the file descriptor to device")
@@ -91,8 +91,8 @@ func isCompatible(sb disklayout.SuperBlock) bool {
 	return true
 }
 
-// NewFilesystem implements vfs.FilesystemType.NewFilesystem.
-func (FilesystemType) NewFilesystem(ctx context.Context, creds *auth.Credentials, source string, opts vfs.NewFilesystemOptions) (*vfs.Filesystem, *vfs.Dentry, error) {
+// GetFilesystem implements vfs.FilesystemType.GetFilesystem.
+func (FilesystemType) GetFilesystem(ctx context.Context, vfsObj *vfs.VirtualFilesystem, creds *auth.Credentials, source string, opts vfs.GetFilesystemOptions) (*vfs.Filesystem, *vfs.Dentry, error) {
 	// TODO(b/134676337): Ensure that the user is mounting readonly. If not,
 	// EACCESS should be returned according to mount(2). Filesystem independent
 	// flags (like readonly) are currently not available in pkg/sentry/vfs.
@@ -103,7 +103,7 @@ func (FilesystemType) NewFilesystem(ctx context.Context, creds *auth.Credentials
 	}
 
 	fs := filesystem{dev: dev, inodeCache: make(map[uint32]*inode)}
-	fs.vfsfs.Init(&fs)
+	fs.vfsfs.Init(vfsObj, &fs)
 	fs.sb, err = readSuperBlock(dev)
 	if err != nil {
 		return nil, nil, err
diff --git a/pkg/sentry/fsimpl/ext/ext_test.go b/pkg/sentry/fsimpl/ext/ext_test.go
index 1aa2bd6a4..307e4d68c 100644
--- a/pkg/sentry/fsimpl/ext/ext_test.go
+++ b/pkg/sentry/fsimpl/ext/ext_test.go
@@ -66,7 +66,7 @@ func setUp(t *testing.T, imagePath string) (context.Context, *vfs.VirtualFilesys
 	// Create VFS.
 	vfsObj := vfs.New()
 	vfsObj.MustRegisterFilesystemType("extfs", FilesystemType{})
-	mntns, err := vfsObj.NewMountNamespace(ctx, creds, localImagePath, "extfs", &vfs.NewFilesystemOptions{InternalData: int(f.Fd())})
+	mntns, err := vfsObj.NewMountNamespace(ctx, creds, localImagePath, "extfs", &vfs.GetFilesystemOptions{InternalData: int(f.Fd())})
 	if err != nil {
 		f.Close()
 		return nil, nil, nil, nil, err
@@ -509,27 +509,27 @@ func TestIterDirents(t *testing.T) {
 	}
 
 	wantDirents := []vfs.Dirent{
-		vfs.Dirent{
+		{
 			Name: ".",
 			Type: linux.DT_DIR,
 		},
-		vfs.Dirent{
+		{
 			Name: "..",
 			Type: linux.DT_DIR,
 		},
-		vfs.Dirent{
+		{
 			Name: "lost+found",
 			Type: linux.DT_DIR,
 		},
-		vfs.Dirent{
+		{
 			Name: "file.txt",
 			Type: linux.DT_REG,
 		},
-		vfs.Dirent{
+		{
 			Name: "bigfile.txt",
 			Type: linux.DT_REG,
 		},
-		vfs.Dirent{
+		{
 			Name: "symlink.txt",
 			Type: linux.DT_LNK,
 		},
diff --git a/pkg/sentry/fsimpl/ext/extent_file.go b/pkg/sentry/fsimpl/ext/extent_file.go
index 38b68a2d3..3d3ebaca6 100644
--- a/pkg/sentry/fsimpl/ext/extent_file.go
+++ b/pkg/sentry/fsimpl/ext/extent_file.go
@@ -99,7 +99,7 @@ func (f *extentFile) buildExtTree() error {
 func (f *extentFile) buildExtTreeFromDisk(entry disklayout.ExtentEntry) (*disklayout.ExtentNode, error) {
 	var header disklayout.ExtentHeader
 	off := entry.PhysicalBlock() * f.regFile.inode.blkSize
-	err := readFromDisk(f.regFile.inode.dev, int64(off), &header)
+	err := readFromDisk(f.regFile.inode.fs.dev, int64(off), &header)
 	if err != nil {
 		return nil, err
 	}
@@ -115,7 +115,7 @@ func (f *extentFile) buildExtTreeFromDisk(entry disklayout.ExtentEntry) (*diskla
 			curEntry = &disklayout.ExtentIdx{}
 		}
 
-		err := readFromDisk(f.regFile.inode.dev, int64(off), curEntry)
+		err := readFromDisk(f.regFile.inode.fs.dev, int64(off), curEntry)
 		if err != nil {
 			return nil, err
 		}
@@ -229,7 +229,7 @@ func (f *extentFile) readFromExtent(ex *disklayout.Extent, off uint64, dst []byt
 		toRead = len(dst)
 	}
 
-	n, _ := f.regFile.inode.dev.ReadAt(dst[:toRead], int64(readStart))
+	n, _ := f.regFile.inode.fs.dev.ReadAt(dst[:toRead], int64(readStart))
 	if n < toRead {
 		return n, syserror.EIO
 	}
diff --git a/pkg/sentry/fsimpl/ext/extent_test.go b/pkg/sentry/fsimpl/ext/extent_test.go
index 42d0a484b..a2382daa3 100644
--- a/pkg/sentry/fsimpl/ext/extent_test.go
+++ b/pkg/sentry/fsimpl/ext/extent_test.go
@@ -180,13 +180,15 @@ func extentTreeSetUp(t *testing.T, root *disklayout.ExtentNode) (*extentFile, []
 	mockExtentFile := &extentFile{
 		regFile: regularFile{
 			inode: inode{
+				fs: &filesystem{
+					dev: bytes.NewReader(mockDisk),
+				},
 				diskInode: &disklayout.InodeNew{
 					InodeOld: disklayout.InodeOld{
 						SizeLo: uint32(mockExtentBlkSize) * getNumPhyBlks(root),
 					},
 				},
 				blkSize: mockExtentBlkSize,
-				dev:     bytes.NewReader(mockDisk),
 			},
 		},
 	}
diff --git a/pkg/sentry/fsimpl/ext/file_description.go b/pkg/sentry/fsimpl/ext/file_description.go
index 4d18b28cb..5eca2b83f 100644
--- a/pkg/sentry/fsimpl/ext/file_description.go
+++ b/pkg/sentry/fsimpl/ext/file_description.go
@@ -36,11 +36,11 @@ type fileDescription struct {
 }
 
 func (fd *fileDescription) filesystem() *filesystem {
-	return fd.vfsfd.VirtualDentry().Mount().Filesystem().Impl().(*filesystem)
+	return fd.vfsfd.Mount().Filesystem().Impl().(*filesystem)
 }
 
 func (fd *fileDescription) inode() *inode {
-	return fd.vfsfd.VirtualDentry().Dentry().Impl().(*dentry).inode
+	return fd.vfsfd.Dentry().Impl().(*dentry).inode
 }
 
 // StatusFlags implements vfs.FileDescriptionImpl.StatusFlags.
diff --git a/pkg/sentry/fsimpl/ext/inode.go b/pkg/sentry/fsimpl/ext/inode.go
index e6c847a71..24249525c 100644
--- a/pkg/sentry/fsimpl/ext/inode.go
+++ b/pkg/sentry/fsimpl/ext/inode.go
@@ -16,7 +16,6 @@ package ext
 
 import (
 	"fmt"
-	"io"
 	"sync/atomic"
 
 	"gvisor.dev/gvisor/pkg/abi/linux"
@@ -42,13 +41,13 @@ type inode struct {
 	// refs is a reference count. refs is accessed using atomic memory operations.
 	refs int64
 
+	// fs is the containing filesystem.
+	fs *filesystem
+
 	// inodeNum is the inode number of this inode on disk. This is used to
 	// identify inodes within the ext filesystem.
 	inodeNum uint32
 
-	// dev represents the underlying device. Same as filesystem.dev.
-	dev io.ReaderAt
-
 	// blkSize is the fs data block size. Same as filesystem.sb.BlockSize().
 	blkSize uint64
 
@@ -81,10 +80,10 @@ func (in *inode) tryIncRef() bool {
 // decRef decrements the inode ref count and releases the inode resources if
 // the ref count hits 0.
 //
-// Precondition: Must have locked fs.mu.
-func (in *inode) decRef(fs *filesystem) {
+// Precondition: Must have locked filesystem.mu.
+func (in *inode) decRef() {
 	if refs := atomic.AddInt64(&in.refs, -1); refs == 0 {
-		delete(fs.inodeCache, in.inodeNum)
+		delete(in.fs.inodeCache, in.inodeNum)
 	} else if refs < 0 {
 		panic("ext.inode.decRef() called without holding a reference")
 	}
@@ -117,8 +116,8 @@ func newInode(fs *filesystem, inodeNum uint32) (*inode, error) {
 
 	// Build the inode based on its type.
 	inode := inode{
+		fs:        fs,
 		inodeNum:  inodeNum,
-		dev:       fs.dev,
 		blkSize:   blkSize,
 		diskInode: diskInode,
 	}
@@ -154,11 +153,14 @@ func (in *inode) open(rp *vfs.ResolvingPath, vfsd *vfs.Dentry, flags uint32) (*v
 	if err := in.checkPermissions(rp.Credentials(), ats); err != nil {
 		return nil, err
 	}
+	mnt := rp.Mount()
 	switch in.impl.(type) {
 	case *regularFile:
 		var fd regularFileFD
 		fd.flags = flags
-		fd.vfsfd.Init(&fd, rp.Mount(), vfsd)
+		mnt.IncRef()
+		vfsd.IncRef()
+		fd.vfsfd.Init(&fd, mnt, vfsd)
 		return &fd.vfsfd, nil
 	case *directory:
 		// Can't open directories writably. This check is not necessary for a read
@@ -167,8 +169,10 @@ func (in *inode) open(rp *vfs.ResolvingPath, vfsd *vfs.Dentry, flags uint32) (*v
 			return nil, syserror.EISDIR
 		}
 		var fd directoryFD
-		fd.vfsfd.Init(&fd, rp.Mount(), vfsd)
 		fd.flags = flags
+		mnt.IncRef()
+		vfsd.IncRef()
+		fd.vfsfd.Init(&fd, mnt, vfsd)
 		return &fd.vfsfd, nil
 	case *symlink:
 		if flags&linux.O_PATH == 0 {
@@ -177,7 +181,9 @@ func (in *inode) open(rp *vfs.ResolvingPath, vfsd *vfs.Dentry, flags uint32) (*v
 		}
 		var fd symlinkFD
 		fd.flags = flags
-		fd.vfsfd.Init(&fd, rp.Mount(), vfsd)
+		mnt.IncRef()
+		vfsd.IncRef()
+		fd.vfsfd.Init(&fd, mnt, vfsd)
 		return &fd.vfsfd, nil
 	default:
 		panic(fmt.Sprintf("unknown inode type: %T", in.impl))