Added stub FUSE filesystem

Allow FUSE filesystems to be mounted using libfuse.
The appropriate flags and mount options are parsed and
understood by fusefs.

8 files changed, 297 insertions, 2 deletions

diff --git a/pkg/sentry/fsimpl/fuse/BUILD b/pkg/sentry/fsimpl/fuse/BUILD
index 3e00c2abb..1fb9c438b 100644
--- a/pkg/sentry/fsimpl/fuse/BUILD
+++ b/pkg/sentry/fsimpl/fuse/BUILD
@@ -6,15 +6,20 @@ go_library(
     name = "fuse",
     srcs = [
         "dev.go",
+        "fusefs.go",
     ],
     visibility = ["//pkg/sentry:internal"],
     deps = [
         "//pkg/abi/linux",
         "//pkg/context",
+        "//pkg/log",
         "//pkg/sentry/fsimpl/devtmpfs",
+        "//pkg/sentry/fsimpl/kernfs",
         "//pkg/sentry/kernel",
+        "//pkg/sentry/kernel/auth",
         "//pkg/sentry/vfs",
         "//pkg/syserror",
         "//pkg/usermem",
+        "@org_golang_x_sys//unix:go_default_library",
     ],
 )
diff --git a/pkg/sentry/fsimpl/fuse/dev.go b/pkg/sentry/fsimpl/fuse/dev.go
index dc33268af..c9e12a94f 100644
--- a/pkg/sentry/fsimpl/fuse/dev.go
+++ b/pkg/sentry/fsimpl/fuse/dev.go
@@ -51,6 +51,9 @@ type DeviceFD struct {
 	vfs.DentryMetadataFileDescriptionImpl
 	vfs.NoLockFD
 
+	// mounted specifies whether a FUSE filesystem was mounted using the DeviceFD.
+	mounted bool
+
 	// TODO(gvisor.dev/issue/2987): Add all the data structures needed to enqueue
 	// and deque requests, control synchronization and establish communication
 	// between the FUSE kernel module and the /dev/fuse character device.
@@ -61,26 +64,51 @@ func (fd *DeviceFD) Release() {}
 
 // PRead implements vfs.FileDescriptionImpl.PRead.
 func (fd *DeviceFD) PRead(ctx context.Context, dst usermem.IOSequence, offset int64, opts vfs.ReadOptions) (int64, error) {
+	// Operations on /dev/fuse don't make sense until a FUSE filesystem is mounted.
+	if !fd.mounted {
+		return 0, syserror.EPERM
+	}
+
 	return 0, syserror.ENOSYS
 }
 
 // Read implements vfs.FileDescriptionImpl.Read.
 func (fd *DeviceFD) Read(ctx context.Context, dst usermem.IOSequence, opts vfs.ReadOptions) (int64, error) {
+	// Operations on /dev/fuse don't make sense until a FUSE filesystem is mounted.
+	if !fd.mounted {
+		return 0, syserror.EPERM
+	}
+
 	return 0, syserror.ENOSYS
 }
 
 // PWrite implements vfs.FileDescriptionImpl.PWrite.
 func (fd *DeviceFD) PWrite(ctx context.Context, src usermem.IOSequence, offset int64, opts vfs.WriteOptions) (int64, error) {
+	// Operations on /dev/fuse don't make sense until a FUSE filesystem is mounted.
+	if !fd.mounted {
+		return 0, syserror.EPERM
+	}
+
 	return 0, syserror.ENOSYS
 }
 
 // Write implements vfs.FileDescriptionImpl.Write.
 func (fd *DeviceFD) Write(ctx context.Context, src usermem.IOSequence, opts vfs.WriteOptions) (int64, error) {
+	// Operations on /dev/fuse don't make sense until a FUSE filesystem is mounted.
+	if !fd.mounted {
+		return 0, syserror.EPERM
+	}
+
 	return 0, syserror.ENOSYS
 }
 
 // Seek implements vfs.FileDescriptionImpl.Seek.
 func (fd *DeviceFD) Seek(ctx context.Context, offset int64, whence int32) (int64, error) {
+	// Operations on /dev/fuse don't make sense until a FUSE filesystem is mounted.
+	if !fd.mounted {
+		return 0, syserror.EPERM
+	}
+
 	return 0, syserror.ENOSYS
 }
 
diff --git a/pkg/sentry/fsimpl/fuse/fusefs.go b/pkg/sentry/fsimpl/fuse/fusefs.go
new file mode 100644
index 000000000..6d40e2d87
--- /dev/null
+++ b/pkg/sentry/fsimpl/fuse/fusefs.go
@@ -0,0 +1,201 @@
+// Copyright 2019 The gVisor Authors.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+// Package fuse implements fusefs.
+package fuse
+
+import (
+	"strconv"
+
+	"gvisor.dev/gvisor/pkg/abi/linux"
+	"gvisor.dev/gvisor/pkg/context"
+	"gvisor.dev/gvisor/pkg/log"
+	"gvisor.dev/gvisor/pkg/sentry/fsimpl/kernfs"
+	"gvisor.dev/gvisor/pkg/sentry/kernel"
+	"gvisor.dev/gvisor/pkg/sentry/kernel/auth"
+	"gvisor.dev/gvisor/pkg/sentry/vfs"
+	"gvisor.dev/gvisor/pkg/syserror"
+)
+
+// Name is the default filesystem name.
+const Name = "fuse"
+
+// FilesystemType implements vfs.FilesystemType.
+type FilesystemType struct{}
+
+type filesystemOptions struct {
+	// userID specifies the numeric uid of the mount owner.
+	// This option should not be specified by the filesystem owner.
+	// It is set by libfuse (or, if libfuse is not used, must be set
+	// by the filesystem itself). For more information, see man page
+	// for fuse(8)
+	userID uint32
+
+	// groupID specifies the numeric gid of the mount owner.
+	// This option should not be specified by the filesystem owner.
+	// It is set by libfuse (or, if libfuse is not used, must be set
+	// by the filesystem itself). For more information, see man page
+	// for fuse(8)
+	groupID uint32
+
+	// rootMode specifies the the file mode of the filesystem's root.
+	rootMode linux.FileMode
+}
+
+// filesystem implements vfs.FilesystemImpl.
+type filesystem struct {
+	kernfs.Filesystem
+	devMinor uint32
+
+	// fuseFD is the FD returned when opening /dev/fuse. It is used for communication
+	// between the FUSE server daemon and the sentry fusefs.
+	fuseFD *DeviceFD
+
+	// opts is the options the fusefs is initialized with.
+	opts filesystemOptions
+}
+
+// Name implements vfs.FilesystemType.Name.
+func (FilesystemType) Name() string {
+	return Name
+}
+
+// GetFilesystem implements vfs.FilesystemType.GetFilesystem.
+func (fsType FilesystemType) GetFilesystem(ctx context.Context, vfsObj *vfs.VirtualFilesystem, creds *auth.Credentials, source string, opts vfs.GetFilesystemOptions) (*vfs.Filesystem, *vfs.Dentry, error) {
+	devMinor, err := vfsObj.GetAnonBlockDevMinor()
+	if err != nil {
+		return nil, nil, err
+	}
+
+	var fsopts filesystemOptions
+	mopts := vfs.GenericParseMountOptions(opts.Data)
+	deviceDescriptorStr, ok := mopts["fd"]
+	if !ok {
+		log.Warningf("%s.GetFilesystem: communication file descriptor N (obtained by opening /dev/fuse) must be specified as 'fd=N'", fsType.Name())
+		return nil, nil, syserror.EINVAL
+	}
+	delete(mopts, "fd")
+
+	deviceDescriptor, err := strconv.ParseInt(deviceDescriptorStr, 10 /* base */, 32 /* bitSize */)
+	if err != nil {
+		return nil, nil, err
+	}
+
+	kernelTask := kernel.TaskFromContext(ctx)
+	if kernelTask == nil {
+		log.Warningf("%s.GetFilesystem: couldn't get kernel task from context", fsType.Name())
+		return nil, nil, syserror.EINVAL
+	}
+	fuseFd := kernelTask.GetFileVFS2(int32(deviceDescriptor))
+
+	// Parse and set all the other supported FUSE mount options.
+	// TODO: Expand the supported mount options.
+	if userIDStr, ok := mopts["user_id"]; ok {
+		delete(mopts, "user_id")
+		userID, err := strconv.ParseUint(userIDStr, 10, 32)
+		if err != nil {
+			log.Warningf("%s.GetFilesystem: invalid user_id: user_id=%s", fsType.Name(), userIDStr)
+			return nil, nil, syserror.EINVAL
+		}
+		fsopts.userID = uint32(userID)
+	}
+
+	if groupIDStr, ok := mopts["group_id"]; ok {
+		delete(mopts, "group_id")
+		groupID, err := strconv.ParseUint(groupIDStr, 10, 32)
+		if err != nil {
+			log.Warningf("%s.GetFilesystem: invalid group_id: group_id=%s", fsType.Name(), groupIDStr)
+			return nil, nil, syserror.EINVAL
+		}
+		fsopts.groupID = uint32(groupID)
+	}
+
+	rootMode := linux.FileMode(0777)
+	modeStr, ok := mopts["rootmode"]
+	if ok {
+		delete(mopts, "rootmode")
+		mode, err := strconv.ParseUint(modeStr, 8, 32)
+		if err != nil {
+			log.Warningf("%s.GetFilesystem: invalid mode: %q", fsType.Name(), modeStr)
+			return nil, nil, syserror.EINVAL
+		}
+		rootMode = linux.FileMode(mode & 07777)
+	}
+	fsopts.rootMode = rootMode
+
+	// Check for unparsed options.
+	if len(mopts) != 0 {
+		log.Warningf("%s.GetFilesystem: unknown options: %v", fsType.Name(), mopts)
+		return nil, nil, syserror.EINVAL
+	}
+
+	// Mark the device as ready so it can be used. /dev/fuse can only be used if the FD was used to
+	// mount a FUSE filesystem.
+	fuseFD := fuseFd.Impl().(*DeviceFD)
+	fuseFD.mounted = true
+
+	fs := &filesystem{
+		devMinor: devMinor,
+		fuseFD:   fuseFD,
+		opts:     fsopts,
+	}
+
+	fs.VFSFilesystem().Init(vfsObj, &fsType, fs)
+
+	// TODO: dispatch a FUSE_INIT request to the FUSE daemon server before
+	//  returning. Mount will not block on this dispatched request.
+
+	// root is the fusefs root directory.
+	defaultFusefsDirMode := linux.FileMode(0755)
+	root := fs.newInode(creds, defaultFusefsDirMode)
+
+	return fs.VFSFilesystem(), root.VFSDentry(), nil
+}
+
+// Release implements vfs.FilesystemImpl.Release.
+func (fs *filesystem) Release() {
+	fs.Filesystem.VFSFilesystem().VirtualFilesystem().PutAnonBlockDevMinor(fs.devMinor)
+	fs.Filesystem.Release()
+}
+
+// Inode implements kernfs.Inode.
+type Inode struct {
+	kernfs.InodeAttrs
+	kernfs.InodeNoDynamicLookup
+	kernfs.InodeNotSymlink
+	kernfs.InodeDirectoryNoNewChildren
+	kernfs.OrderedChildren
+
+	locks vfs.FileLocks
+
+	dentry kernfs.Dentry
+}
+
+func (fs *filesystem) newInode(creds *auth.Credentials, mode linux.FileMode) *kernfs.Dentry {
+	i := &Inode{}
+	i.InodeAttrs.Init(creds, linux.UNNAMED_MAJOR, fs.devMinor, fs.NextIno(), linux.ModeDirectory|0755)
+	i.OrderedChildren.Init(kernfs.OrderedChildrenOptions{})
+	i.dentry.Init(i)
+
+	return &i.dentry
+}
+
+// Open implements kernfs.Inode.Open.
+func (i *Inode) Open(ctx context.Context, rp *vfs.ResolvingPath, vfsd *vfs.Dentry, opts vfs.OpenOptions) (*vfs.FileDescription, error) {
+	fd, err := kernfs.NewGenericDirectoryFD(rp.Mount(), vfsd, &i.OrderedChildren, &i.locks, &opts)
+	if err != nil {
+		return nil, err
+	}
+	return fd.VFSFileDescription(), nil
+}
diff --git a/pkg/sentry/syscalls/linux/vfs2/mount.go b/pkg/sentry/syscalls/linux/vfs2/mount.go
index adeaa39cc..ea337de7c 100644
--- a/pkg/sentry/syscalls/linux/vfs2/mount.go
+++ b/pkg/sentry/syscalls/linux/vfs2/mount.go
@@ -77,8 +77,7 @@ func Mount(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscall
 
 	// Silently allow MS_NOSUID, since we don't implement set-id bits
 	// anyway.
-	const unsupportedFlags = linux.MS_NODEV |
-		linux.MS_NODIRATIME | linux.MS_STRICTATIME
+	const unsupportedFlags = linux.MS_NODIRATIME | linux.MS_STRICTATIME
 
 	// Linux just allows passing any flags to mount(2) - it won't fail when
 	// unknown or unsupported flags are passed. Since we don't implement
@@ -94,6 +93,12 @@ func Mount(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscall
 	if flags&linux.MS_NOEXEC == linux.MS_NOEXEC {
 		opts.Flags.NoExec = true
 	}
+	if flags&linux.MS_NODEV == linux.MS_NODEV {
+		opts.Flags.NoDev = true
+	}
+	if flags&linux.MS_NOSUID == linux.MS_NOSUID {
+		opts.Flags.NoSUID = true
+	}
 	if flags&linux.MS_RDONLY == linux.MS_RDONLY {
 		opts.ReadOnly = true
 	}
diff --git a/pkg/sentry/vfs/options.go b/pkg/sentry/vfs/options.go
index f223aeda8..d37208a1f 100644
--- a/pkg/sentry/vfs/options.go
+++ b/pkg/sentry/vfs/options.go
@@ -79,6 +79,17 @@ type MountFlags struct {
 	// NoATime is equivalent to MS_NOATIME and indicates that the
 	// filesystem should not update access time in-place.
 	NoATime bool
+
+	// NoDev is equivalent to MS_NODEV and indicates that the
+	// filesystem should not allow access to devices (special files).
+	// TODO(gVisor.dev/issue/3186): respect this flag in non FUSE
+	// filesystems.
+	NoDev bool
+
+	// NoSUID is equivalent to MS_NOSUID and indicates that the
+	// filesystem should not honor set-user-ID and set-group-ID bits or
+	// file capabilities when executing programs.
+	NoSUID bool
 }
 
 // MountOptions contains options to VirtualFilesystem.MountAt().
diff --git a/runsc/boot/vfs.go b/runsc/boot/vfs.go
index 56f4ba15d..9a1ed8e9e 100644
--- a/runsc/boot/vfs.go
+++ b/runsc/boot/vfs.go
@@ -77,6 +77,10 @@ func registerFilesystems(k *kernel.Kernel) error {
 		AllowUserMount: true,
 		AllowUserList:  true,
 	})
+	vfsObj.MustRegisterFilesystemType(fuse.Name, &fuse.FilesystemType{}, &vfs.RegisterFilesystemTypeOptions{
+		AllowUserMount: true,
+		AllowUserList:  true,
+	})
 
 	// Setup files in devtmpfs.
 	if err := memdev.Register(vfsObj); err != nil {
@@ -119,6 +123,7 @@ func registerFilesystems(k *kernel.Kernel) error {
 			return fmt.Errorf("creating fusedev devtmpfs files: %w", err)
 		}
 	}
+
 	return nil
 }
 
diff --git a/test/syscalls/linux/dev.cc b/test/syscalls/linux/dev.cc
index 6fa16208e..69e8b76d3 100644
--- a/test/syscalls/linux/dev.cc
+++ b/test/syscalls/linux/dev.cc
@@ -161,6 +161,18 @@ TEST(DevTest, OpenDevFuse) {
   ASSERT_NO_ERRNO_AND_VALUE(Open("/dev/fuse", O_RDONLY));
 }
 
+TEST(DevTest, ReadDevFuseWithoutMount) {
+  // Note(gvisor.dev/issue/3076) This won't work in the sentry until the new
+  // device registration is complete.
+  SKIP_IF(IsRunningWithVFS1() || IsRunningOnGvisor());
+
+  const FileDescriptor fd =
+    ASSERT_NO_ERRNO_AND_VALUE(Open("/dev/fuse", O_RDONLY));
+
+  std::vector<char> buf(1);
+  EXPECT_THAT(ReadFd(fd.get(), buf.data(), sizeof(buf)), SyscallFailsWithErrno(EPERM));
+}
+
 }  // namespace
 }  // namespace testing
 
diff --git a/test/syscalls/linux/mount.cc b/test/syscalls/linux/mount.cc
index a3e9745cf..7664fa73d 100644
--- a/test/syscalls/linux/mount.cc
+++ b/test/syscalls/linux/mount.cc
@@ -321,6 +321,34 @@ TEST(MountTest, RenameRemoveMountPoint) {
   ASSERT_THAT(rmdir(dir.path().c_str()), SyscallFailsWithErrno(EBUSY));
 }
 
+TEST(MountTest, MountFuseFilesystemNoDevice) {
+  SKIP_IF(!ASSERT_NO_ERRNO_AND_VALUE(HaveCapability(CAP_SYS_ADMIN)));
+
+  // Note(gvisor.dev/issue/3076) This won't work in the sentry until the new
+  // device registration is complete.
+  SKIP_IF(IsRunningWithVFS1() || IsRunningOnGvisor());
+
+  auto const dir = ASSERT_NO_ERRNO_AND_VALUE(TempPath::CreateDir());
+  EXPECT_THAT(mount("", dir.path().c_str(), "fuse", 0, ""),
+              SyscallFailsWithErrno(EINVAL));
+}
+
+TEST(MountTest, MountFuseFilesystem) {
+  SKIP_IF(!ASSERT_NO_ERRNO_AND_VALUE(HaveCapability(CAP_SYS_ADMIN)));
+
+  // Note(gvisor.dev/issue/3076) This won't work in the sentry until the new
+  // device registration is complete.
+  SKIP_IF(IsRunningWithVFS1() || IsRunningOnGvisor());
+
+  const FileDescriptor fd =
+      ASSERT_NO_ERRNO_AND_VALUE(Open("/dev/fuse", O_WRONLY));
+  std::string mopts = "fd=" + std::to_string(fd.get());
+
+  auto const dir = ASSERT_NO_ERRNO_AND_VALUE(TempPath::CreateDir());
+  auto const mount =
+      ASSERT_NO_ERRNO_AND_VALUE(Mount("", dir.path(), "fuse", 0, mopts, 0));
+}
+
 }  // namespace
 
 }  // namespace testing