Implement statx.

We don't have the plumbing for btime yet, so that field is left off. The
returned mask indicates that btime is absent.

Fixes #343

PiperOrigin-RevId: 254575752

pkg/abi/linux/file.go

@@ -181,6 +181,57 @@ type Stat struct {
 // SizeOfStat is the size of a Stat struct.
 var SizeOfStat = binary.Size(Stat{})
 
+// Flags for statx.
+const (
+	AT_STATX_SYNC_TYPE    = 0x6000
+	AT_STATX_SYNC_AS_STAT = 0x0000
+	AT_STATX_FORCE_SYNC   = 0x2000
+	AT_STATX_DONT_SYNC    = 0x4000
+)
+
+// Mask values for statx.
+const (
+	STATX_TYPE        = 0x00000001
+	STATX_MODE        = 0x00000002
+	STATX_NLINK       = 0x00000004
+	STATX_UID         = 0x00000008
+	STATX_GID         = 0x00000010
+	STATX_ATIME       = 0x00000020
+	STATX_MTIME       = 0x00000040
+	STATX_CTIME       = 0x00000080
+	STATX_INO         = 0x00000100
+	STATX_SIZE        = 0x00000200
+	STATX_BLOCKS      = 0x00000400
+	STATX_BASIC_STATS = 0x000007ff
+	STATX_BTIME       = 0x00000800
+	STATX_ALL         = 0x00000fff
+	STATX__RESERVED   = 0x80000000
+)
+
+// Statx represents struct statx.
+type Statx struct {
+	Mask           uint32
+	Blksize        uint32
+	Attributes     uint64
+	Nlink          uint32
+	UID            uint32
+	GID            uint32
+	Mode           uint16
+	_              uint16
+	Ino            uint64
+	Size           uint64
+	Blocks         uint64
+	AttributesMask uint64
+	Atime          StatxTimestamp
+	Btime          StatxTimestamp
+	Ctime          StatxTimestamp
+	Mtime          StatxTimestamp
+	RdevMajor      uint32
+	RdevMinor      uint32
+	DevMajor       uint32
+	DevMinor       uint32
+}
+
 // FileMode represents a mode_t.
 type FileMode uint
 

pkg/abi/linux/time.go

@@ -226,3 +226,18 @@ type Tms struct {
 // TimerID represents type timer_t, which identifies a POSIX per-process
 // interval timer.
 type TimerID int32
+
+// StatxTimestamp represents struct statx_timestamp.
+type StatxTimestamp struct {
+	Sec  int64
+	Nsec uint32
+	_    int32
+}
+
+// NsecToStatxTimestamp translates nanoseconds to StatxTimestamp.
+func NsecToStatxTimestamp(nsec int64) (ts StatxTimestamp) {
+	return StatxTimestamp{
+		Sec:  nsec / 1e9,
+		Nsec: uint32(nsec % 1e9),
+	}
+}

pkg/sentry/fs/attr.go

@@ -89,6 +89,28 @@ func (n InodeType) String() string {
 	}
 }
 
+// LinuxType returns the linux file type for this inode type.
+func (n InodeType) LinuxType() uint32 {
+	switch n {
+	case RegularFile, SpecialFile:
+		return linux.ModeRegular
+	case Directory, SpecialDirectory:
+		return linux.ModeDirectory
+	case Symlink:
+		return linux.ModeSymlink
+	case Pipe:
+		return linux.ModeNamedPipe
+	case CharacterDevice:
+		return linux.ModeCharacterDevice
+	case BlockDevice:
+		return linux.ModeBlockDevice
+	case Socket:
+		return linux.ModeSocket
+	default:
+		return 0
+	}
+}
+
 // StableAttr contains Inode attributes that will be stable throughout the
 // lifetime of the Inode.
 //

pkg/sentry/kernel/time/time.go

@@ -142,6 +142,11 @@ func (t Time) Timeval() linux.Timeval {
 	return linux.NsecToTimeval(t.Nanoseconds())
 }
 
+// StatxTimestamp converts Time to a Linux statx_timestamp.
+func (t Time) StatxTimestamp() linux.StatxTimestamp {
+	return linux.NsecToStatxTimestamp(t.Nanoseconds())
+}
+
 // Add adds the duration of d to t.
 func (t Time) Add(d time.Duration) Time {
 	if t.ns > 0 && d.Nanoseconds() > math.MaxInt64-int64(t.ns) {

pkg/sentry/syscalls/linux/linux64.go

@@ -379,6 +379,7 @@ var AMD64 = &kernel.SyscallTable{
 		326: syscalls.ErrorWithEvent("copy_file_range", syscall.ENOSYS, "", nil),
 		327: syscalls.Undocumented("preadv2", Preadv2),
 		328: syscalls.Undocumented("pwritev2", Pwritev2),
+		397: syscalls.Undocumented("statx", Statx),
 	},
 
 	Emulate: map[usermem.Addr]uintptr{

pkg/sentry/syscalls/linux/sys_stat.go

@@ -132,24 +132,6 @@ func fstat(t *kernel.Task, f *fs.File, statAddr usermem.Addr) error {
 // t.CopyObjectOut has noticeable performance impact due to its many slice
 // allocations and use of reflection.
 func copyOutStat(t *kernel.Task, dst usermem.Addr, sattr fs.StableAttr, uattr fs.UnstableAttr) error {
-	var mode uint32
-	switch sattr.Type {
-	case fs.RegularFile, fs.SpecialFile:
-		mode |= linux.ModeRegular
-	case fs.Symlink:
-		mode |= linux.ModeSymlink
-	case fs.Directory, fs.SpecialDirectory:
-		mode |= linux.ModeDirectory
-	case fs.Pipe:
-		mode |= linux.ModeNamedPipe
-	case fs.CharacterDevice:
-		mode |= linux.ModeCharacterDevice
-	case fs.BlockDevice:
-		mode |= linux.ModeBlockDevice
-	case fs.Socket:
-		mode |= linux.ModeSocket
-	}
-
 	b := t.CopyScratchBuffer(int(linux.SizeOfStat))[:0]
 
 	// Dev (uint64)
@@ -159,7 +141,7 @@ func copyOutStat(t *kernel.Task, dst usermem.Addr, sattr fs.StableAttr, uattr fs
 	// Nlink (uint64)
 	b = binary.AppendUint64(b, usermem.ByteOrder, uattr.Links)
 	// Mode (uint32)
-	b = binary.AppendUint32(b, usermem.ByteOrder, mode|uint32(uattr.Perms.LinuxMode()))
+	b = binary.AppendUint32(b, usermem.ByteOrder, sattr.Type.LinuxType()|uint32(uattr.Perms.LinuxMode()))
 	// UID (uint32)
 	b = binary.AppendUint32(b, usermem.ByteOrder, uint32(uattr.Owner.UID.In(t.UserNamespace()).OrOverflow()))
 	// GID (uint32)
@@ -194,6 +176,98 @@ func copyOutStat(t *kernel.Task, dst usermem.Addr, sattr fs.StableAttr, uattr fs
 	return err
 }
 
+// Statx implements linux syscall statx(2).
+func Statx(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) {
+	fd := kdefs.FD(args[0].Int())
+	pathAddr := args[1].Pointer()
+	flags := args[2].Int()
+	mask := args[3].Uint()
+	statxAddr := args[4].Pointer()
+
+	if mask&linux.STATX__RESERVED > 0 {
+		return 0, nil, syserror.EINVAL
+	}
+	if flags&linux.AT_STATX_SYNC_TYPE == linux.AT_STATX_SYNC_TYPE {
+		return 0, nil, syserror.EINVAL
+	}
+
+	path, dirPath, err := copyInPath(t, pathAddr, flags&linux.AT_EMPTY_PATH != 0)
+	if err != nil {
+		return 0, nil, err
+	}
+
+	if path == "" {
+		file := t.FDMap().GetFile(fd)
+		if file == nil {
+			return 0, nil, syserror.EBADF
+		}
+		defer file.DecRef()
+		uattr, err := file.UnstableAttr(t)
+		if err != nil {
+			return 0, nil, err
+		}
+		return 0, nil, statx(t, file.Dirent.Inode.StableAttr, uattr, statxAddr)
+	}
+
+	resolve := dirPath || flags&linux.AT_SYMLINK_NOFOLLOW == 0
+
+	return 0, nil, fileOpOn(t, fd, path, resolve, func(root *fs.Dirent, d *fs.Dirent, _ uint) error {
+		if dirPath && !fs.IsDir(d.Inode.StableAttr) {
+			return syserror.ENOTDIR
+		}
+		uattr, err := d.Inode.UnstableAttr(t)
+		if err != nil {
+			return err
+		}
+		return statx(t, d.Inode.StableAttr, uattr, statxAddr)
+	})
+}
+
+func statx(t *kernel.Task, sattr fs.StableAttr, uattr fs.UnstableAttr, statxAddr usermem.Addr) error {
+	// "[T]he kernel may return fields that weren't requested and may fail to
+	// return fields that were requested, depending on what the backing
+	// filesystem supports.
+	// [...]
+	// A filesystem may also fill in fields that the caller didn't ask for
+	// if it has values for them available and the information is available
+	// at no extra cost. If this happens, the corresponding bits will be
+	// set in stx_mask." -- statx(2)
+	//
+	// We fill in all the values we have (which currently does not include
+	// btime, see b/135608823), regardless of what the user asked for. The
+	// STATX_BASIC_STATS mask indicates that all fields are present except
+	// for btime.
+
+	devMajor, devMinor := linux.DecodeDeviceID(uint32(sattr.DeviceID))
+	s := linux.Statx{
+		// TODO(b/135608823): Support btime, and then change this to
+		// STATX_ALL to indicate presence of btime.
+		Mask: linux.STATX_BASIC_STATS,
+
+		// No attributes, and none supported.
+		Attributes:     0,
+		AttributesMask: 0,
+
+		Blksize:   uint32(sattr.BlockSize),
+		Nlink:     uint32(uattr.Links),
+		UID:       uint32(uattr.Owner.UID.In(t.UserNamespace()).OrOverflow()),
+		GID:       uint32(uattr.Owner.GID.In(t.UserNamespace()).OrOverflow()),
+		Mode:      uint16(sattr.Type.LinuxType()) | uint16(uattr.Perms.LinuxMode()),
+		Ino:       sattr.InodeID,
+		Size:      uint64(uattr.Size),
+		Blocks:    uint64(uattr.Usage) / 512,
+		Atime:     uattr.AccessTime.StatxTimestamp(),
+		Ctime:     uattr.StatusChangeTime.StatxTimestamp(),
+		Mtime:     uattr.ModificationTime.StatxTimestamp(),
+		RdevMajor: uint32(sattr.DeviceFileMajor),
+		RdevMinor: sattr.DeviceFileMinor,
+		DevMajor:  uint32(devMajor),
+		DevMinor:  devMinor,
+	}
+	_, err := t.CopyOut(statxAddr, &s)
+	return err
+}
+
 // Statfs implements linux syscall statfs(2).
 func Statfs(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) {
 	addr := args[0].Pointer()
@@ -252,8 +326,6 @@ func statfsImpl(t *kernel.Task, d *fs.Dirent, addr usermem.Addr) error {
 		FragmentSize: d.Inode.StableAttr.BlockSize,
 		// Leave other fields 0 like simple_statfs does.
 	}
-	if _, err := t.CopyOut(addr, &statfs); err != nil {
+	_, err = t.CopyOut(addr, &statfs)
-		return err
+	return err
-	}
-	return nil
 }

test/syscalls/linux/stat.cc

@@ -16,7 +16,9 @@
 #include <fcntl.h>
 #include <sys/stat.h>
 #include <sys/statfs.h>
+#include <sys/types.h>
 #include <unistd.h>
+
 #include <string>
 #include <vector>
 
@@ -554,6 +556,99 @@ TEST(SimpleStatTest, AnonDeviceAllocatesUniqueInodesAcrossSaveRestore) {
   EXPECT_EQ(st2_after.st_ino, st2.st_ino);
 }
 
+#ifndef SYS_statx
+#if defined(__x86_64__)
+#define SYS_statx 397
+#else
+#error "Unknown architecture"
+#endif
+#endif  // SYS_statx
+
+#ifndef STATX_ALL
+#define STATX_ALL 0x00000fffU
+#endif  // STATX_ALL
+
+// struct kernel_statx_timestamp is a Linux statx_timestamp struct.
+struct kernel_statx_timestamp {
+  int64_t tv_sec;
+  uint32_t tv_nsec;
+  int32_t __reserved;
+};
+
+// struct kernel_statx is a Linux statx struct. Old versions of glibc do not
+// expose it. See include/uapi/linux/stat.h
+struct kernel_statx {
+  uint32_t stx_mask;
+  uint32_t stx_blksize;
+  uint64_t stx_attributes;
+  uint32_t stx_nlink;
+  uint32_t stx_uid;
+  uint32_t stx_gid;
+  uint16_t stx_mode;
+  uint16_t __spare0[1];
+  uint64_t stx_ino;
+  uint64_t stx_size;
+  uint64_t stx_blocks;
+  uint64_t stx_attributes_mask;
+  struct kernel_statx_timestamp stx_atime;
+  struct kernel_statx_timestamp stx_btime;
+  struct kernel_statx_timestamp stx_ctime;
+  struct kernel_statx_timestamp stx_mtime;
+  uint32_t stx_rdev_major;
+  uint32_t stx_rdev_minor;
+  uint32_t stx_dev_major;
+  uint32_t stx_dev_minor;
+  uint64_t __spare2[14];
+};
+
+int statx(int dirfd, const char *pathname, int flags, unsigned int mask,
+          struct kernel_statx *statxbuf) {
+  return syscall(SYS_statx, dirfd, pathname, flags, mask, statxbuf);
+}
+
+TEST_F(StatTest, StatxAbsPath) {
+  SKIP_IF(statx(-1, nullptr, 0, 0, 0) < 0 && errno == ENOSYS);
+
+  struct kernel_statx stx;
+  EXPECT_THAT(statx(-1, test_file_name_.c_str(), 0, STATX_ALL, &stx),
+              SyscallSucceeds());
+  EXPECT_TRUE(S_ISREG(stx.stx_mode));
+}
+
+TEST_F(StatTest, StatxRelPathDirFD) {
+  SKIP_IF(statx(-1, nullptr, 0, 0, 0) < 0 && errno == ENOSYS);
+
+  struct kernel_statx stx;
+  auto const dirfd =
+      ASSERT_NO_ERRNO_AND_VALUE(Open(GetAbsoluteTestTmpdir(), O_RDONLY));
+  auto filename = std::string(Basename(test_file_name_));
+
+  EXPECT_THAT(statx(dirfd.get(), filename.c_str(), 0, STATX_ALL, &stx),
+              SyscallSucceeds());
+  EXPECT_TRUE(S_ISREG(stx.stx_mode));
+}
+
+TEST_F(StatTest, StatxRelPathCwd) {
+  SKIP_IF(statx(-1, nullptr, 0, 0, 0) < 0 && errno == ENOSYS);
+
+  ASSERT_THAT(chdir(GetAbsoluteTestTmpdir().c_str()), SyscallSucceeds());
+  auto filename = std::string(Basename(test_file_name_));
+  struct kernel_statx stx;
+  EXPECT_THAT(statx(AT_FDCWD, filename.c_str(), 0, STATX_ALL, &stx),
+              SyscallSucceeds());
+  EXPECT_TRUE(S_ISREG(stx.stx_mode));
+}
+
+TEST_F(StatTest, StatxEmptyPath) {
+  SKIP_IF(statx(-1, nullptr, 0, 0, 0) < 0 && errno == ENOSYS);
+
+  const auto fd = ASSERT_NO_ERRNO_AND_VALUE(Open(test_file_name_, O_RDONLY));
+  struct kernel_statx stx;
+  EXPECT_THAT(statx(fd.get(), "", AT_EMPTY_PATH, STATX_ALL, &stx),
+              SyscallSucceeds());
+  EXPECT_TRUE(S_ISREG(stx.stx_mode));
+}
+
 }  // namespace
 
 }  // namespace testing