Merge release-20200522.0-89-g21b6bc72 (automated)

2 files changed, 147 insertions, 2 deletions

diff --git a/pkg/sentry/syscalls/linux/vfs2/mount.go b/pkg/sentry/syscalls/linux/vfs2/mount.go
new file mode 100644
index 000000000..adeaa39cc
--- /dev/null
+++ b/pkg/sentry/syscalls/linux/vfs2/mount.go
@@ -0,0 +1,145 @@
+// Copyright 2020 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 vfs2
+
+import (
+	"gvisor.dev/gvisor/pkg/abi/linux"
+	"gvisor.dev/gvisor/pkg/sentry/arch"
+	"gvisor.dev/gvisor/pkg/sentry/kernel"
+	"gvisor.dev/gvisor/pkg/sentry/vfs"
+	"gvisor.dev/gvisor/pkg/syserror"
+	"gvisor.dev/gvisor/pkg/usermem"
+)
+
+// Mount implements Linux syscall mount(2).
+func Mount(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) {
+	sourceAddr := args[0].Pointer()
+	targetAddr := args[1].Pointer()
+	typeAddr := args[2].Pointer()
+	flags := args[3].Uint64()
+	dataAddr := args[4].Pointer()
+
+	// For null-terminated strings related to mount(2), Linux copies in at most
+	// a page worth of data. See fs/namespace.c:copy_mount_string().
+	fsType, err := t.CopyInString(typeAddr, usermem.PageSize)
+	if err != nil {
+		return 0, nil, err
+	}
+	source, err := t.CopyInString(sourceAddr, usermem.PageSize)
+	if err != nil {
+		return 0, nil, err
+	}
+
+	targetPath, err := copyInPath(t, targetAddr)
+	if err != nil {
+		return 0, nil, err
+	}
+
+	data := ""
+	if dataAddr != 0 {
+		// In Linux, a full page is always copied in regardless of null
+		// character placement, and the address is passed to each file system.
+		// Most file systems always treat this data as a string, though, and so
+		// do all of the ones we implement.
+		data, err = t.CopyInString(dataAddr, usermem.PageSize)
+		if err != nil {
+			return 0, nil, err
+		}
+	}
+
+	// Ignore magic value that was required before Linux 2.4.
+	if flags&linux.MS_MGC_MSK == linux.MS_MGC_VAL {
+		flags = flags &^ linux.MS_MGC_MSK
+	}
+
+	// Must have CAP_SYS_ADMIN in the current mount namespace's associated user
+	// namespace.
+	creds := t.Credentials()
+	if !creds.HasCapabilityIn(linux.CAP_SYS_ADMIN, t.MountNamespaceVFS2().Owner) {
+		return 0, nil, syserror.EPERM
+	}
+
+	const unsupportedOps = linux.MS_REMOUNT | linux.MS_BIND |
+		linux.MS_SHARED | linux.MS_PRIVATE | linux.MS_SLAVE |
+		linux.MS_UNBINDABLE | linux.MS_MOVE
+
+	// Silently allow MS_NOSUID, since we don't implement set-id bits
+	// anyway.
+	const unsupportedFlags = linux.MS_NODEV |
+		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
+	// everything, we fail explicitly on flags that are unimplemented.
+	if flags&(unsupportedOps|unsupportedFlags) != 0 {
+		return 0, nil, syserror.EINVAL
+	}
+
+	var opts vfs.MountOptions
+	if flags&linux.MS_NOATIME == linux.MS_NOATIME {
+		opts.Flags.NoATime = true
+	}
+	if flags&linux.MS_NOEXEC == linux.MS_NOEXEC {
+		opts.Flags.NoExec = true
+	}
+	if flags&linux.MS_RDONLY == linux.MS_RDONLY {
+		opts.ReadOnly = true
+	}
+	opts.GetFilesystemOptions.Data = data
+
+	target, err := getTaskPathOperation(t, linux.AT_FDCWD, targetPath, disallowEmptyPath, nofollowFinalSymlink)
+	if err != nil {
+		return 0, nil, err
+	}
+	defer target.Release()
+
+	return 0, nil, t.Kernel().VFS().MountAt(t, creds, source, &target.pop, fsType, &opts)
+}
+
+// Umount2 implements Linux syscall umount2(2).
+func Umount2(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) {
+	addr := args[0].Pointer()
+	flags := args[1].Int()
+
+	// Must have CAP_SYS_ADMIN in the mount namespace's associated user
+	// namespace.
+	//
+	// Currently, this is always the init task's user namespace.
+	creds := t.Credentials()
+	if !creds.HasCapabilityIn(linux.CAP_SYS_ADMIN, t.MountNamespaceVFS2().Owner) {
+		return 0, nil, syserror.EPERM
+	}
+
+	const unsupported = linux.MNT_FORCE | linux.MNT_EXPIRE
+	if flags&unsupported != 0 {
+		return 0, nil, syserror.EINVAL
+	}
+
+	path, err := copyInPath(t, addr)
+	if err != nil {
+		return 0, nil, err
+	}
+	tpop, err := getTaskPathOperation(t, linux.AT_FDCWD, path, disallowEmptyPath, nofollowFinalSymlink)
+	if err != nil {
+		return 0, nil, err
+	}
+	defer tpop.Release()
+
+	opts := vfs.UmountOptions{
+		Flags: uint32(flags),
+	}
+
+	return 0, nil, t.Kernel().VFS().UmountAt(t, creds, &tpop.pop, &opts)
+}
diff --git a/pkg/sentry/syscalls/linux/vfs2/vfs2.go b/pkg/sentry/syscalls/linux/vfs2/vfs2.go
index ef8358b8a..7b6e7571a 100644
--- a/pkg/sentry/syscalls/linux/vfs2/vfs2.go
+++ b/pkg/sentry/syscalls/linux/vfs2/vfs2.go
@@ -90,8 +90,8 @@ func Override() {
 	s.Table[138] = syscalls.Supported("fstatfs", Fstatfs)
 	s.Table[161] = syscalls.Supported("chroot", Chroot)
 	s.Table[162] = syscalls.Supported("sync", Sync)
-	delete(s.Table, 165) // mount
-	delete(s.Table, 166) // umount2
+	s.Table[165] = syscalls.Supported("mount", Mount)
+	s.Table[166] = syscalls.Supported("umount2", Umount2)
 	delete(s.Table, 187) // readahead
 	s.Table[188] = syscalls.Supported("setxattr", Setxattr)
 	s.Table[189] = syscalls.Supported("lsetxattr", Lsetxattr)