Move VFS2 file description status flags to vfs.FileDescription.

PiperOrigin-RevId: 286616668

9 files changed, 107 insertions, 123 deletions

diff --git a/pkg/sentry/fsimpl/ext/file_description.go b/pkg/sentry/fsimpl/ext/file_description.go
index 5eca2b83f..841274daf 100644
--- a/pkg/sentry/fsimpl/ext/file_description.go
+++ b/pkg/sentry/fsimpl/ext/file_description.go
@@ -26,13 +26,6 @@ import (
 type fileDescription struct {
 	vfsfd vfs.FileDescription
 	vfs.FileDescriptionDefaultImpl
-
-	// flags is the same as vfs.OpenOptions.Flags which are passed to
-	// vfs.FilesystemImpl.OpenAt.
-	// TODO(b/134676337): syscalls like read(2), write(2), fchmod(2), fchown(2),
-	// fgetxattr(2), ioctl(2), mmap(2) should fail with EBADF if O_PATH is set.
-	// Only close(2), fstat(2), fstatfs(2) should work.
-	flags uint32
 }
 
 func (fd *fileDescription) filesystem() *filesystem {
@@ -43,18 +36,6 @@ func (fd *fileDescription) inode() *inode {
 	return fd.vfsfd.Dentry().Impl().(*dentry).inode
 }
 
-// StatusFlags implements vfs.FileDescriptionImpl.StatusFlags.
-func (fd *fileDescription) StatusFlags(ctx context.Context) (uint32, error) {
-	return fd.flags, nil
-}
-
-// SetStatusFlags implements vfs.FileDescriptionImpl.SetStatusFlags.
-func (fd *fileDescription) SetStatusFlags(ctx context.Context, flags uint32) error {
-	// None of the flags settable by fcntl(F_SETFL) are supported, so this is a
-	// no-op.
-	return nil
-}
-
 // Stat implements vfs.FileDescriptionImpl.Stat.
 func (fd *fileDescription) Stat(ctx context.Context, opts vfs.StatOptions) (linux.Statx, error) {
 	var stat linux.Statx
diff --git a/pkg/sentry/fsimpl/ext/inode.go b/pkg/sentry/fsimpl/ext/inode.go
index 24249525c..b2cc826c7 100644
--- a/pkg/sentry/fsimpl/ext/inode.go
+++ b/pkg/sentry/fsimpl/ext/inode.go
@@ -157,10 +157,9 @@ func (in *inode) open(rp *vfs.ResolvingPath, vfsd *vfs.Dentry, flags uint32) (*v
 	switch in.impl.(type) {
 	case *regularFile:
 		var fd regularFileFD
-		fd.flags = flags
 		mnt.IncRef()
 		vfsd.IncRef()
-		fd.vfsfd.Init(&fd, mnt, vfsd)
+		fd.vfsfd.Init(&fd, flags, mnt, vfsd, &vfs.FileDescriptionOptions{})
 		return &fd.vfsfd, nil
 	case *directory:
 		// Can't open directories writably. This check is not necessary for a read
@@ -169,10 +168,9 @@ func (in *inode) open(rp *vfs.ResolvingPath, vfsd *vfs.Dentry, flags uint32) (*v
 			return nil, syserror.EISDIR
 		}
 		var fd directoryFD
-		fd.flags = flags
 		mnt.IncRef()
 		vfsd.IncRef()
-		fd.vfsfd.Init(&fd, mnt, vfsd)
+		fd.vfsfd.Init(&fd, flags, mnt, vfsd, &vfs.FileDescriptionOptions{})
 		return &fd.vfsfd, nil
 	case *symlink:
 		if flags&linux.O_PATH == 0 {
@@ -180,10 +178,9 @@ func (in *inode) open(rp *vfs.ResolvingPath, vfsd *vfs.Dentry, flags uint32) (*v
 			return nil, syserror.ELOOP
 		}
 		var fd symlinkFD
-		fd.flags = flags
 		mnt.IncRef()
 		vfsd.IncRef()
-		fd.vfsfd.Init(&fd, mnt, vfsd)
+		fd.vfsfd.Init(&fd, flags, mnt, vfsd, &vfs.FileDescriptionOptions{})
 		return &fd.vfsfd, nil
 	default:
 		panic(fmt.Sprintf("unknown inode type: %T", in.impl))
diff --git a/pkg/sentry/fsimpl/kernfs/dynamic_bytes_file.go b/pkg/sentry/fsimpl/kernfs/dynamic_bytes_file.go
index 30c06baf0..51102ce48 100644
--- a/pkg/sentry/fsimpl/kernfs/dynamic_bytes_file.go
+++ b/pkg/sentry/fsimpl/kernfs/dynamic_bytes_file.go
@@ -65,17 +65,15 @@ type DynamicBytesFD struct {
 
 	vfsfd vfs.FileDescription
 	inode Inode
-	flags uint32
 }
 
 // Init initializes a DynamicBytesFD.
 func (fd *DynamicBytesFD) Init(m *vfs.Mount, d *vfs.Dentry, data vfs.DynamicBytesSource, flags uint32) {
 	m.IncRef() // DecRef in vfs.FileDescription.vd.DecRef on final ref.
 	d.IncRef() // DecRef in vfs.FileDescription.vd.DecRef on final ref.
-	fd.flags = flags
 	fd.inode = d.Impl().(*Dentry).inode
 	fd.SetDataSource(data)
-	fd.vfsfd.Init(fd, m, d)
+	fd.vfsfd.Init(fd, flags, m, d, &vfs.FileDescriptionOptions{})
 }
 
 // Seek implements vfs.FileDescriptionImpl.Seek.
@@ -117,15 +115,3 @@ func (fd *DynamicBytesFD) SetStat(context.Context, vfs.SetStatOptions) error {
 	// DynamicBytesFiles are immutable.
 	return syserror.EPERM
 }
-
-// StatusFlags implements vfs.FileDescriptionImpl.StatusFlags.
-func (fd *DynamicBytesFD) StatusFlags(ctx context.Context) (uint32, error) {
-	return fd.flags, nil
-}
-
-// SetStatusFlags implements vfs.FileDescriptionImpl.SetStatusFlags.
-func (fd *DynamicBytesFD) SetStatusFlags(ctx context.Context, flags uint32) error {
-	// None of the flags settable by fcntl(F_SETFL) are supported, so this is a
-	// no-op.
-	return nil
-}
diff --git a/pkg/sentry/fsimpl/kernfs/fd_impl_util.go b/pkg/sentry/fsimpl/kernfs/fd_impl_util.go
index d6c18937a..bd402330f 100644
--- a/pkg/sentry/fsimpl/kernfs/fd_impl_util.go
+++ b/pkg/sentry/fsimpl/kernfs/fd_impl_util.go
@@ -39,7 +39,6 @@ type GenericDirectoryFD struct {
 
 	vfsfd    vfs.FileDescription
 	children *OrderedChildren
-	flags    uint32
 	off      int64
 }
 
@@ -48,8 +47,7 @@ func (fd *GenericDirectoryFD) Init(m *vfs.Mount, d *vfs.Dentry, children *Ordere
 	m.IncRef() // DecRef in vfs.FileDescription.vd.DecRef on final ref.
 	d.IncRef() // DecRef in vfs.FileDescription.vd.DecRef on final ref.
 	fd.children = children
-	fd.flags = flags
-	fd.vfsfd.Init(fd, m, d)
+	fd.vfsfd.Init(fd, flags, m, d, &vfs.FileDescriptionOptions{})
 }
 
 // VFSFileDescription returns a pointer to the vfs.FileDescription representing
@@ -180,18 +178,6 @@ func (fd *GenericDirectoryFD) Seek(ctx context.Context, offset int64, whence int
 	return offset, nil
 }
 
-// StatusFlags implements vfs.FileDescriptionImpl.StatusFlags.
-func (fd *GenericDirectoryFD) StatusFlags(ctx context.Context) (uint32, error) {
-	return fd.flags, nil
-}
-
-// SetStatusFlags implements vfs.FileDescriptionImpl.SetStatusFlags.
-func (fd *GenericDirectoryFD) SetStatusFlags(ctx context.Context, flags uint32) error {
-	// None of the flags settable by fcntl(F_SETFL) are supported, so this is a
-	// no-op.
-	return nil
-}
-
 // Stat implements vfs.FileDescriptionImpl.Stat.
 func (fd *GenericDirectoryFD) Stat(ctx context.Context, opts vfs.StatOptions) (linux.Statx, error) {
 	fs := fd.filesystem()
diff --git a/pkg/sentry/fsimpl/memfs/filesystem.go b/pkg/sentry/fsimpl/memfs/filesystem.go
index 22f1e811f..af4389459 100644
--- a/pkg/sentry/fsimpl/memfs/filesystem.go
+++ b/pkg/sentry/fsimpl/memfs/filesystem.go
@@ -282,9 +282,8 @@ func (fs *filesystem) MknodAt(ctx context.Context, rp *vfs.ResolvingPath, opts v
 func (fs *filesystem) OpenAt(ctx context.Context, rp *vfs.ResolvingPath, opts vfs.OpenOptions) (*vfs.FileDescription, error) {
 	// Filter out flags that are not supported by memfs. O_DIRECTORY and
 	// O_NOFOLLOW have no effect here (they're handled by VFS by setting
-	// appropriate bits in rp), but are returned by
-	// FileDescriptionImpl.StatusFlags(). O_NONBLOCK is supported only by
-	// pipes.
+	// appropriate bits in rp), but are visible in FD status flags. O_NONBLOCK
+	// is supported only by pipes.
 	opts.Flags &= linux.O_ACCMODE | linux.O_CREAT | linux.O_EXCL | linux.O_TRUNC | linux.O_DIRECTORY | linux.O_NOFOLLOW | linux.O_NONBLOCK
 
 	if opts.Flags&linux.O_CREAT == 0 {
@@ -384,7 +383,6 @@ func (i *inode) open(ctx context.Context, rp *vfs.ResolvingPath, vfsd *vfs.Dentr
 	switch impl := i.impl.(type) {
 	case *regularFile:
 		var fd regularFileFD
-		fd.flags = flags
 		fd.readable = vfs.MayReadFileWithOpenFlags(flags)
 		fd.writable = vfs.MayWriteFileWithOpenFlags(flags)
 		if fd.writable {
@@ -395,7 +393,7 @@ func (i *inode) open(ctx context.Context, rp *vfs.ResolvingPath, vfsd *vfs.Dentr
 		}
 		mnt.IncRef()
 		vfsd.IncRef()
-		fd.vfsfd.Init(&fd, mnt, vfsd)
+		fd.vfsfd.Init(&fd, flags, mnt, vfsd, &vfs.FileDescriptionOptions{})
 		if flags&linux.O_TRUNC != 0 {
 			impl.mu.Lock()
 			impl.data = impl.data[:0]
@@ -411,8 +409,7 @@ func (i *inode) open(ctx context.Context, rp *vfs.ResolvingPath, vfsd *vfs.Dentr
 		var fd directoryFD
 		mnt.IncRef()
 		vfsd.IncRef()
-		fd.vfsfd.Init(&fd, mnt, vfsd)
-		fd.flags = flags
+		fd.vfsfd.Init(&fd, flags, mnt, vfsd, &vfs.FileDescriptionOptions{})
 		return &fd.vfsfd, nil
 	case *symlink:
 		// Can't open symlinks without O_PATH (which is unimplemented).
diff --git a/pkg/sentry/fsimpl/memfs/memfs.go b/pkg/sentry/fsimpl/memfs/memfs.go
index 4cb2a4e0f..9d509f6e4 100644
--- a/pkg/sentry/fsimpl/memfs/memfs.go
+++ b/pkg/sentry/fsimpl/memfs/memfs.go
@@ -261,8 +261,6 @@ func (i *inode) direntType() uint8 {
 type fileDescription struct {
 	vfsfd vfs.FileDescription
 	vfs.FileDescriptionDefaultImpl
-
-	flags uint32 // status flags; immutable
 }
 
 func (fd *fileDescription) filesystem() *filesystem {
@@ -273,18 +271,6 @@ func (fd *fileDescription) inode() *inode {
 	return fd.vfsfd.Dentry().Impl().(*dentry).inode
 }
 
-// StatusFlags implements vfs.FileDescriptionImpl.StatusFlags.
-func (fd *fileDescription) StatusFlags(ctx context.Context) (uint32, error) {
-	return fd.flags, nil
-}
-
-// SetStatusFlags implements vfs.FileDescriptionImpl.SetStatusFlags.
-func (fd *fileDescription) SetStatusFlags(ctx context.Context, flags uint32) error {
-	// None of the flags settable by fcntl(F_SETFL) are supported, so this is a
-	// no-op.
-	return nil
-}
-
 // Stat implements vfs.FileDescriptionImpl.Stat.
 func (fd *fileDescription) Stat(ctx context.Context, opts vfs.StatOptions) (linux.Statx, error) {
 	var stat linux.Statx
diff --git a/pkg/sentry/fsimpl/memfs/named_pipe.go b/pkg/sentry/fsimpl/memfs/named_pipe.go
index 91cb4b1fc..d5060850e 100644
--- a/pkg/sentry/fsimpl/memfs/named_pipe.go
+++ b/pkg/sentry/fsimpl/memfs/named_pipe.go
@@ -57,6 +57,6 @@ func newNamedPipeFD(ctx context.Context, np *namedPipe, rp *vfs.ResolvingPath, v
 	mnt := rp.Mount()
 	mnt.IncRef()
 	vfsd.IncRef()
-	fd.vfsfd.Init(&fd, mnt, vfsd)
+	fd.vfsfd.Init(&fd, flags, mnt, vfsd, &vfs.FileDescriptionOptions{})
 	return &fd.vfsfd, nil
 }
diff --git a/pkg/sentry/vfs/file_description.go b/pkg/sentry/vfs/file_description.go
index c5a9adca3..df03886c3 100644
--- a/pkg/sentry/vfs/file_description.go
+++ b/pkg/sentry/vfs/file_description.go
@@ -20,6 +20,7 @@ import (
 	"gvisor.dev/gvisor/pkg/abi/linux"
 	"gvisor.dev/gvisor/pkg/sentry/arch"
 	"gvisor.dev/gvisor/pkg/sentry/context"
+	"gvisor.dev/gvisor/pkg/sentry/kernel/auth"
 	"gvisor.dev/gvisor/pkg/sentry/memmap"
 	"gvisor.dev/gvisor/pkg/sentry/usermem"
 	"gvisor.dev/gvisor/pkg/syserror"
@@ -39,49 +40,43 @@ type FileDescription struct {
 	// operations.
 	refs int64
 
+	// statusFlags contains status flags, "initialized by open(2) and possibly
+	// modified by fcntl()" - fcntl(2). statusFlags is accessed using atomic
+	// memory operations.
+	statusFlags uint32
+
 	// vd is the filesystem location at which this FileDescription was opened.
 	// A reference is held on vd. vd is immutable.
 	vd VirtualDentry
 
+	opts FileDescriptionOptions
+
 	// impl is the FileDescriptionImpl associated with this Filesystem. impl is
 	// immutable. This should be the last field in FileDescription.
 	impl FileDescriptionImpl
 }
 
+// FileDescriptionOptions contains options to FileDescription.Init().
+type FileDescriptionOptions struct {
+	// If AllowDirectIO is true, allow O_DIRECT to be set on the file. This is
+	// usually only the case if O_DIRECT would actually have an effect.
+	AllowDirectIO bool
+}
+
 // Init must be called before first use of fd. It takes ownership of references
-// on mnt and d held by the caller.
-func (fd *FileDescription) Init(impl FileDescriptionImpl, mnt *Mount, d *Dentry) {
+// on mnt and d held by the caller. statusFlags is the initial file description
+// status flags, which is usually the full set of flags passed to open(2).
+func (fd *FileDescription) Init(impl FileDescriptionImpl, statusFlags uint32, mnt *Mount, d *Dentry, opts *FileDescriptionOptions) {
 	fd.refs = 1
+	fd.statusFlags = statusFlags | linux.O_LARGEFILE
 	fd.vd = VirtualDentry{
 		mount:  mnt,
 		dentry: d,
 	}
+	fd.opts = *opts
 	fd.impl = impl
 }
 
-// Impl returns the FileDescriptionImpl associated with fd.
-func (fd *FileDescription) Impl() FileDescriptionImpl {
-	return fd.impl
-}
-
-// Mount returns the mount on which fd was opened. It does not take a reference
-// on the returned Mount.
-func (fd *FileDescription) Mount() *Mount {
-	return fd.vd.mount
-}
-
-// Dentry returns the dentry at which fd was opened. It does not take a
-// reference on the returned Dentry.
-func (fd *FileDescription) Dentry() *Dentry {
-	return fd.vd.dentry
-}
-
-// VirtualDentry returns the location at which fd was opened. It does not take
-// a reference on the returned VirtualDentry.
-func (fd *FileDescription) VirtualDentry() VirtualDentry {
-	return fd.vd
-}
-
 // IncRef increments fd's reference count.
 func (fd *FileDescription) IncRef() {
 	atomic.AddInt64(&fd.refs, 1)
@@ -113,6 +108,82 @@ func (fd *FileDescription) DecRef() {
 	}
 }
 
+// Mount returns the mount on which fd was opened. It does not take a reference
+// on the returned Mount.
+func (fd *FileDescription) Mount() *Mount {
+	return fd.vd.mount
+}
+
+// Dentry returns the dentry at which fd was opened. It does not take a
+// reference on the returned Dentry.
+func (fd *FileDescription) Dentry() *Dentry {
+	return fd.vd.dentry
+}
+
+// VirtualDentry returns the location at which fd was opened. It does not take
+// a reference on the returned VirtualDentry.
+func (fd *FileDescription) VirtualDentry() VirtualDentry {
+	return fd.vd
+}
+
+// StatusFlags returns file description status flags, as for fcntl(F_GETFL).
+func (fd *FileDescription) StatusFlags() uint32 {
+	return atomic.LoadUint32(&fd.statusFlags)
+}
+
+// SetStatusFlags sets file description status flags, as for fcntl(F_SETFL).
+func (fd *FileDescription) SetStatusFlags(ctx context.Context, creds *auth.Credentials, flags uint32) error {
+	// Compare Linux's fs/fcntl.c:setfl().
+	oldFlags := fd.StatusFlags()
+	// Linux documents this check as "O_APPEND cannot be cleared if the file is
+	// marked as append-only and the file is open for write", which would make
+	// sense. However, the check as actually implemented seems to be "O_APPEND
+	// cannot be changed if the file is marked as append-only".
+	if (flags^oldFlags)&linux.O_APPEND != 0 {
+		stat, err := fd.impl.Stat(ctx, StatOptions{
+			// There is no mask bit for stx_attributes.
+			Mask: 0,
+			// Linux just reads inode::i_flags directly.
+			Sync: linux.AT_STATX_DONT_SYNC,
+		})
+		if err != nil {
+			return err
+		}
+		if (stat.AttributesMask&linux.STATX_ATTR_APPEND != 0) && (stat.Attributes&linux.STATX_ATTR_APPEND != 0) {
+			return syserror.EPERM
+		}
+	}
+	if (flags&linux.O_NOATIME != 0) && (oldFlags&linux.O_NOATIME == 0) {
+		stat, err := fd.impl.Stat(ctx, StatOptions{
+			Mask: linux.STATX_UID,
+			// Linux's inode_owner_or_capable() just reads inode::i_uid
+			// directly.
+			Sync: linux.AT_STATX_DONT_SYNC,
+		})
+		if err != nil {
+			return err
+		}
+		if stat.Mask&linux.STATX_UID == 0 {
+			return syserror.EPERM
+		}
+		if !CanActAsOwner(creds, auth.KUID(stat.UID)) {
+			return syserror.EPERM
+		}
+	}
+	if flags&linux.O_DIRECT != 0 && !fd.opts.AllowDirectIO {
+		return syserror.EINVAL
+	}
+	// TODO(jamieliu): FileDescriptionImpl.SetOAsync()?
+	const settableFlags = linux.O_APPEND | linux.O_ASYNC | linux.O_DIRECT | linux.O_NOATIME | linux.O_NONBLOCK
+	atomic.StoreUint32(&fd.statusFlags, (oldFlags&^settableFlags)|(flags&settableFlags))
+	return nil
+}
+
+// Impl returns the FileDescriptionImpl associated with fd.
+func (fd *FileDescription) Impl() FileDescriptionImpl {
+	return fd.impl
+}
+
 // FileDescriptionImpl contains implementation details for an FileDescription.
 // Implementations of FileDescriptionImpl should contain their associated
 // FileDescription by value as their first field.
@@ -132,14 +203,6 @@ type FileDescriptionImpl interface {
 	// prevent the file descriptor from being closed.
 	OnClose(ctx context.Context) error
 
-	// StatusFlags returns file description status flags, as for
-	// fcntl(F_GETFL).
-	StatusFlags(ctx context.Context) (uint32, error)
-
-	// SetStatusFlags sets file description status flags, as for
-	// fcntl(F_SETFL).
-	SetStatusFlags(ctx context.Context, flags uint32) error
-
 	// Stat returns metadata for the file represented by the FileDescription.
 	Stat(ctx context.Context, opts StatOptions) (linux.Statx, error)
 
@@ -264,18 +327,6 @@ func (fd *FileDescription) OnClose(ctx context.Context) error {
 	return fd.impl.OnClose(ctx)
 }
 
-// StatusFlags returns file description status flags, as for fcntl(F_GETFL).
-func (fd *FileDescription) StatusFlags(ctx context.Context) (uint32, error) {
-	flags, err := fd.impl.StatusFlags(ctx)
-	flags |= linux.O_LARGEFILE
-	return flags, err
-}
-
-// SetStatusFlags sets file description status flags, as for fcntl(F_SETFL).
-func (fd *FileDescription) SetStatusFlags(ctx context.Context, flags uint32) error {
-	return fd.impl.SetStatusFlags(ctx, flags)
-}
-
 // Stat returns metadata for the file represented by fd.
 func (fd *FileDescription) Stat(ctx context.Context, opts StatOptions) (linux.Statx, error) {
 	return fd.impl.Stat(ctx, opts)
diff --git a/pkg/sentry/vfs/file_description_impl_util_test.go b/pkg/sentry/vfs/file_description_impl_util_test.go
index ac7799296..678be07fe 100644
--- a/pkg/sentry/vfs/file_description_impl_util_test.go
+++ b/pkg/sentry/vfs/file_description_impl_util_test.go
@@ -48,7 +48,7 @@ type genCountFD struct {
 
 func newGenCountFD(mnt *Mount, vfsd *Dentry) *FileDescription {
 	var fd genCountFD
-	fd.vfsfd.Init(&fd, mnt, vfsd)
+	fd.vfsfd.Init(&fd, 0 /* statusFlags */, mnt, vfsd, &FileDescriptionOptions{})
 	fd.DynamicBytesFileDescriptionImpl.SetDataSource(&fd)
 	return &fd.vfsfd
 }