1 /* SPDX-License-Identifier: LGPL-2.1+ */
10 #include <sys/sendfile.h>
12 #include <sys/xattr.h>
16 #include "alloc-util.h"
17 #include "btrfs-util.h"
18 #include "chattr-util.h"
20 #include "dirent-util.h"
26 #include "mountpoint-util.h"
27 #include "stat-util.h"
28 #include "string-util.h"
30 #include "time-util.h"
31 #include "tmpfile-util.h"
32 #include "umask-util.h"
33 #include "user-util.h"
34 #include "xattr-util.h"
36 #define COPY_BUFFER_SIZE (16U*1024U)
38 /* A safety net for descending recursively into file system trees to copy. On Linux PATH_MAX is 4096, which means the
39 * deepest valid path one can build is around 2048, which we hence use as a safety net here, to not spin endlessly in
40 * case of bind mount cycles and suchlike. */
41 #define COPY_DEPTH_MAX 2048U
43 static ssize_t
try_copy_file_range(
44 int fd_in
, loff_t
*off_in
,
45 int fd_out
, loff_t
*off_out
,
55 r
= copy_file_range(fd_in
, off_in
, fd_out
, off_out
, len
, flags
);
57 have
= r
>= 0 || errno
!= ENOSYS
;
67 FD_IS_NONBLOCKING_PIPE
,
70 static int fd_is_nonblock_pipe(int fd
) {
74 /* Checks whether the specified file descriptor refers to a pipe, and if so if O_NONBLOCK is set. */
76 if (fstat(fd
, &st
) < 0)
79 if (!S_ISFIFO(st
.st_mode
))
82 flags
= fcntl(fd
, F_GETFL
);
86 return FLAGS_SET(flags
, O_NONBLOCK
) ? FD_IS_NONBLOCKING_PIPE
: FD_IS_BLOCKING_PIPE
;
94 size_t *ret_remains_size
,
95 copy_progress_bytes_t progress
,
98 bool try_cfr
= true, try_sendfile
= true, try_splice
= true;
99 int r
, nonblock_pipe
= -1;
100 size_t m
= SSIZE_MAX
; /* that is the maximum that sendfile and c_f_r accept */
105 /* Tries to copy bytes from the file descriptor 'fdf' to 'fdt' in the smartest possible way. Copies a maximum
106 * of 'max_bytes', which may be specified as UINT64_MAX, in which no maximum is applied. Returns negative on
107 * error, zero if EOF is hit before the bytes limit is hit and positive otherwise. If the copy fails for some
108 * reason but we read but didn't yet write some data an ret_remains/ret_remains_size is not NULL, then it will
109 * be initialized with an allocated buffer containing this "remaining" data. Note that these two parameters are
110 * initialized with a valid buffer only on failure and only if there's actually data already read. Otherwise
111 * these parameters if non-NULL are set to NULL. */
115 if (ret_remains_size
)
116 *ret_remains_size
= 0;
118 /* Try btrfs reflinks first. This only works on regular, seekable files, hence let's check the file offsets of
119 * source and destination first. */
120 if ((copy_flags
& COPY_REFLINK
)) {
123 foffset
= lseek(fdf
, 0, SEEK_CUR
);
127 toffset
= lseek(fdt
, 0, SEEK_CUR
);
130 if (foffset
== 0 && toffset
== 0 && max_bytes
== UINT64_MAX
)
131 r
= btrfs_reflink(fdf
, fdt
); /* full file reflink */
133 r
= btrfs_clone_range(fdf
, foffset
, fdt
, toffset
, max_bytes
== UINT64_MAX
? 0 : max_bytes
); /* partial reflink */
137 /* This worked, yay! Now — to be fully correct — let's adjust the file pointers */
138 if (max_bytes
== UINT64_MAX
) {
140 /* We cloned to the end of the source file, let's position the read
141 * pointer there, and query it at the same time. */
142 t
= lseek(fdf
, 0, SEEK_END
);
148 /* Let's adjust the destination file write pointer by the same number
150 t
= lseek(fdt
, toffset
+ (t
- foffset
), SEEK_SET
);
154 return 0; /* we copied the whole thing, hence hit EOF, return 0 */
156 t
= lseek(fdf
, foffset
+ max_bytes
, SEEK_SET
);
160 t
= lseek(fdt
, toffset
+ max_bytes
, SEEK_SET
);
164 return 1; /* we copied only some number of bytes, which worked, but this means we didn't hit EOF, return 1 */
175 return 1; /* return > 0 if we hit the max_bytes limit */
177 if (max_bytes
!= UINT64_MAX
&& m
> max_bytes
)
180 /* First try copy_file_range(), unless we already tried */
182 n
= try_copy_file_range(fdf
, NULL
, fdt
, NULL
, m
, 0u);
184 if (!IN_SET(n
, -EINVAL
, -ENOSYS
, -EXDEV
, -EBADF
))
188 /* use fallback below */
189 } else if (n
== 0) /* EOF */
196 /* First try sendfile(), unless we already tried */
198 n
= sendfile(fdt
, fdf
, NULL
, m
);
200 if (!IN_SET(errno
, EINVAL
, ENOSYS
))
203 try_sendfile
= false;
204 /* use fallback below */
205 } else if (n
== 0) /* EOF */
212 /* Then try splice, unless we already tried. */
215 /* splice()'s asynchronous I/O support is a bit weird. When it encounters a pipe file
216 * descriptor, then it will ignore its O_NONBLOCK flag and instead only honour the
217 * SPLICE_F_NONBLOCK flag specified in its flag parameter. Let's hide this behaviour here, and
218 * check if either of the specified fds are a pipe, and if so, let's pass the flag
219 * automatically, depending on O_NONBLOCK being set.
221 * Here's a twist though: when we use it to move data between two pipes of which one has
222 * O_NONBLOCK set and the other has not, then we have no individual control over O_NONBLOCK
223 * behaviour. Hence in that case we can't use splice() and still guarantee systematic
224 * O_NONBLOCK behaviour, hence don't. */
226 if (nonblock_pipe
< 0) {
229 /* Check if either of these fds is a pipe, and if so non-blocking or not */
230 a
= fd_is_nonblock_pipe(fdf
);
234 b
= fd_is_nonblock_pipe(fdt
);
238 if ((a
== FD_IS_NO_PIPE
&& b
== FD_IS_NO_PIPE
) ||
239 (a
== FD_IS_BLOCKING_PIPE
&& b
== FD_IS_NONBLOCKING_PIPE
) ||
240 (a
== FD_IS_NONBLOCKING_PIPE
&& b
== FD_IS_BLOCKING_PIPE
))
242 /* splice() only works if one of the fds is a pipe. If neither is, let's skip
243 * this step right-away. As mentioned above, if one of the two fds refers to a
244 * blocking pipe and the other to a non-blocking pipe, we can't use splice()
245 * either, hence don't try either. This hence means we can only use splice() if
246 * either only one of the two fds is a pipe, or if both are pipes with the same
247 * nonblocking flag setting. */
251 nonblock_pipe
= a
== FD_IS_NONBLOCKING_PIPE
|| b
== FD_IS_NONBLOCKING_PIPE
;
256 n
= splice(fdf
, NULL
, fdt
, NULL
, m
, nonblock_pipe
? SPLICE_F_NONBLOCK
: 0);
258 if (!IN_SET(errno
, EINVAL
, ENOSYS
))
262 /* use fallback below */
263 } else if (n
== 0) /* EOF */
270 /* As a fallback just copy bits by hand */
272 uint8_t buf
[MIN(m
, COPY_BUFFER_SIZE
)], *p
= buf
;
275 n
= read(fdf
, buf
, sizeof buf
);
278 if (n
== 0) /* EOF */
285 k
= write(fdt
, p
, z
);
299 if (ret_remains_size
)
300 *ret_remains_size
= z
;
313 r
= progress(n
, userdata
);
318 if (max_bytes
!= (uint64_t) -1) {
319 assert(max_bytes
>= (uint64_t) n
);
323 /* sendfile accepts at most SSIZE_MAX-offset bytes to copy,
324 * so reduce our maximum by the amount we already copied,
325 * but don't go below our copy buffer size, unless we are
326 * close the limit of bytes we are allowed to copy. */
327 m
= MAX(MIN(COPY_BUFFER_SIZE
, max_bytes
), m
- n
);
330 return 0; /* return 0 if we hit EOF earlier than the size limit */
333 static int fd_copy_symlink(
336 const struct stat
*st
,
341 CopyFlags copy_flags
) {
343 _cleanup_free_
char *target
= NULL
;
350 r
= readlinkat_malloc(df
, from
, &target
);
354 if (symlinkat(target
, dt
, to
) < 0)
358 uid_is_valid(override_uid
) ? override_uid
: st
->st_uid
,
359 gid_is_valid(override_gid
) ? override_gid
: st
->st_gid
,
360 AT_SYMLINK_NOFOLLOW
) < 0)
366 static int fd_copy_regular(
369 const struct stat
*st
,
374 CopyFlags copy_flags
,
375 copy_progress_bytes_t progress
,
378 _cleanup_close_
int fdf
= -1, fdt
= -1;
379 struct timespec ts
[2];
386 fdf
= openat(df
, from
, O_RDONLY
|O_CLOEXEC
|O_NOCTTY
|O_NOFOLLOW
);
390 fdt
= openat(dt
, to
, O_WRONLY
|O_CREAT
|O_EXCL
|O_CLOEXEC
|O_NOCTTY
|O_NOFOLLOW
, st
->st_mode
& 07777);
394 r
= copy_bytes_full(fdf
, fdt
, (uint64_t) -1, copy_flags
, NULL
, NULL
, progress
, userdata
);
396 (void) unlinkat(dt
, to
, 0);
401 uid_is_valid(override_uid
) ? override_uid
: st
->st_uid
,
402 gid_is_valid(override_gid
) ? override_gid
: st
->st_gid
) < 0)
405 if (fchmod(fdt
, st
->st_mode
& 07777) < 0)
410 (void) futimens(fdt
, ts
);
411 (void) copy_xattr(fdf
, fdt
);
418 (void) unlinkat(dt
, to
, 0);
424 static int fd_copy_fifo(
427 const struct stat
*st
,
432 CopyFlags copy_flags
) {
439 r
= mkfifoat(dt
, to
, st
->st_mode
& 07777);
444 uid_is_valid(override_uid
) ? override_uid
: st
->st_uid
,
445 gid_is_valid(override_gid
) ? override_gid
: st
->st_gid
,
446 AT_SYMLINK_NOFOLLOW
) < 0)
449 if (fchmodat(dt
, to
, st
->st_mode
& 07777, 0) < 0)
455 static int fd_copy_node(
458 const struct stat
*st
,
463 CopyFlags copy_flags
) {
470 r
= mknodat(dt
, to
, st
->st_mode
, st
->st_rdev
);
475 uid_is_valid(override_uid
) ? override_uid
: st
->st_uid
,
476 gid_is_valid(override_gid
) ? override_gid
: st
->st_gid
,
477 AT_SYMLINK_NOFOLLOW
) < 0)
480 if (fchmodat(dt
, to
, st
->st_mode
& 07777, 0) < 0)
486 static int fd_copy_directory(
489 const struct stat
*st
,
492 dev_t original_device
,
496 CopyFlags copy_flags
,
497 const char *display_path
,
498 copy_progress_path_t progress_path
,
499 copy_progress_bytes_t progress_bytes
,
502 _cleanup_close_
int fdf
= -1, fdt
= -1;
503 _cleanup_closedir_
DIR *d
= NULL
;
505 bool exists
, created
;
512 return -ENAMETOOLONG
;
515 fdf
= openat(df
, from
, O_RDONLY
|O_DIRECTORY
|O_CLOEXEC
|O_NOCTTY
|O_NOFOLLOW
);
517 fdf
= fcntl(df
, F_DUPFD_CLOEXEC
, 3);
527 if (copy_flags
& COPY_MERGE_EMPTY
) {
528 r
= dir_is_empty_at(dt
, to
);
529 if (r
< 0 && r
!= -ENOENT
)
538 r
= mkdirat(dt
, to
, st
->st_mode
& 07777);
541 else if (errno
== EEXIST
&& (copy_flags
& COPY_MERGE
))
547 fdt
= openat(dt
, to
, O_RDONLY
|O_DIRECTORY
|O_CLOEXEC
|O_NOCTTY
|O_NOFOLLOW
);
553 FOREACH_DIRENT_ALL(de
, d
, return -errno
) {
554 const char *child_display_path
= NULL
;
555 _cleanup_free_
char *dp
= NULL
;
559 if (dot_or_dot_dot(de
->d_name
))
562 if (fstatat(dirfd(d
), de
->d_name
, &buf
, AT_SYMLINK_NOFOLLOW
) < 0) {
569 child_display_path
= dp
= strjoin(display_path
, "/", de
->d_name
);
571 child_display_path
= de
->d_name
;
573 r
= progress_path(child_display_path
, &buf
, userdata
);
578 if (S_ISDIR(buf
.st_mode
)) {
580 * Don't descend into directories on other file systems, if this is requested. We do a simple
581 * .st_dev check here, which basically comes for free. Note that we do this check only on
582 * directories, not other kind of file system objects, for two reason:
584 * • The kernel's overlayfs pseudo file system that overlays multiple real file systems
585 * propagates the .st_dev field of the file system a file originates from all the way up
586 * through the stack to stat(). It doesn't do that for directories however. This means that
587 * comparing .st_dev on non-directories suggests that they all are mount points. To avoid
588 * confusion we hence avoid relying on this check for regular files.
590 * • The main reason we do this check at all is to protect ourselves from bind mount cycles,
591 * where we really want to avoid descending down in all eternity. However the .st_dev check
592 * is usually not sufficient for this protection anyway, as bind mount cycles from the same
593 * file system onto itself can't be detected that way. (Note we also do a recursion depth
594 * check, which is probably the better protection in this regard, which is why
595 * COPY_SAME_MOUNT is optional).
598 if (FLAGS_SET(copy_flags
, COPY_SAME_MOUNT
)) {
599 if (buf
.st_dev
!= original_device
)
602 r
= fd_is_mount_point(dirfd(d
), de
->d_name
, 0);
609 q
= fd_copy_directory(dirfd(d
), de
->d_name
, &buf
, fdt
, de
->d_name
, original_device
, depth_left
-1, override_uid
, override_gid
, copy_flags
, child_display_path
, progress_path
, progress_bytes
, userdata
);
610 } else if (S_ISREG(buf
.st_mode
))
611 q
= fd_copy_regular(dirfd(d
), de
->d_name
, &buf
, fdt
, de
->d_name
, override_uid
, override_gid
, copy_flags
, progress_bytes
, userdata
);
612 else if (S_ISLNK(buf
.st_mode
))
613 q
= fd_copy_symlink(dirfd(d
), de
->d_name
, &buf
, fdt
, de
->d_name
, override_uid
, override_gid
, copy_flags
);
614 else if (S_ISFIFO(buf
.st_mode
))
615 q
= fd_copy_fifo(dirfd(d
), de
->d_name
, &buf
, fdt
, de
->d_name
, override_uid
, override_gid
, copy_flags
);
616 else if (S_ISBLK(buf
.st_mode
) || S_ISCHR(buf
.st_mode
) || S_ISSOCK(buf
.st_mode
))
617 q
= fd_copy_node(dirfd(d
), de
->d_name
, &buf
, fdt
, de
->d_name
, override_uid
, override_gid
, copy_flags
);
621 if (q
== -EEXIST
&& (copy_flags
& COPY_MERGE
))
629 struct timespec ut
[2] = {
635 uid_is_valid(override_uid
) ? override_uid
: st
->st_uid
,
636 gid_is_valid(override_gid
) ? override_gid
: st
->st_gid
) < 0)
639 if (fchmod(fdt
, st
->st_mode
& 07777) < 0)
642 (void) copy_xattr(dirfd(d
), fdt
);
643 (void) futimens(fdt
, ut
);
649 int copy_tree_at_full(
656 CopyFlags copy_flags
,
657 copy_progress_path_t progress_path
,
658 copy_progress_bytes_t progress_bytes
,
666 if (fstatat(fdf
, from
, &st
, AT_SYMLINK_NOFOLLOW
) < 0)
669 if (S_ISREG(st
.st_mode
))
670 return fd_copy_regular(fdf
, from
, &st
, fdt
, to
, override_uid
, override_gid
, copy_flags
, progress_bytes
, userdata
);
671 else if (S_ISDIR(st
.st_mode
))
672 return fd_copy_directory(fdf
, from
, &st
, fdt
, to
, st
.st_dev
, COPY_DEPTH_MAX
, override_uid
, override_gid
, copy_flags
, NULL
, progress_path
, progress_bytes
, userdata
);
673 else if (S_ISLNK(st
.st_mode
))
674 return fd_copy_symlink(fdf
, from
, &st
, fdt
, to
, override_uid
, override_gid
, copy_flags
);
675 else if (S_ISFIFO(st
.st_mode
))
676 return fd_copy_fifo(fdf
, from
, &st
, fdt
, to
, override_uid
, override_gid
, copy_flags
);
677 else if (S_ISBLK(st
.st_mode
) || S_ISCHR(st
.st_mode
) || S_ISSOCK(st
.st_mode
))
678 return fd_copy_node(fdf
, from
, &st
, fdt
, to
, override_uid
, override_gid
, copy_flags
);
683 int copy_directory_fd_full(
686 CopyFlags copy_flags
,
687 copy_progress_path_t progress_path
,
688 copy_progress_bytes_t progress_bytes
,
696 if (fstat(dirfd
, &st
) < 0)
699 if (!S_ISDIR(st
.st_mode
))
702 return fd_copy_directory(dirfd
, NULL
, &st
, AT_FDCWD
, to
, st
.st_dev
, COPY_DEPTH_MAX
, UID_INVALID
, GID_INVALID
, copy_flags
, NULL
, progress_path
, progress_bytes
, userdata
);
705 int copy_directory_full(
708 CopyFlags copy_flags
,
709 copy_progress_path_t progress_path
,
710 copy_progress_bytes_t progress_bytes
,
718 if (lstat(from
, &st
) < 0)
721 if (!S_ISDIR(st
.st_mode
))
724 return fd_copy_directory(AT_FDCWD
, from
, &st
, AT_FDCWD
, to
, st
.st_dev
, COPY_DEPTH_MAX
, UID_INVALID
, GID_INVALID
, copy_flags
, NULL
, progress_path
, progress_bytes
, userdata
);
727 int copy_file_fd_full(
730 CopyFlags copy_flags
,
731 copy_progress_bytes_t progress_bytes
,
734 _cleanup_close_
int fdf
= -1;
740 fdf
= open(from
, O_RDONLY
|O_CLOEXEC
|O_NOCTTY
);
744 r
= copy_bytes_full(fdf
, fdt
, (uint64_t) -1, copy_flags
, NULL
, NULL
, progress_bytes
, userdata
);
746 (void) copy_times(fdf
, fdt
, copy_flags
);
747 (void) copy_xattr(fdf
, fdt
);
757 unsigned chattr_flags
,
758 unsigned chattr_mask
,
759 CopyFlags copy_flags
,
760 copy_progress_bytes_t progress_bytes
,
768 RUN_WITH_UMASK(0000) {
769 fdt
= open(to
, flags
|O_WRONLY
|O_CREAT
|O_CLOEXEC
|O_NOCTTY
, mode
);
774 if (chattr_mask
!= 0)
775 (void) chattr_fd(fdt
, chattr_flags
, chattr_mask
& CHATTR_EARLY_FL
, NULL
);
777 r
= copy_file_fd_full(from
, fdt
, copy_flags
, progress_bytes
, userdata
);
784 if (chattr_mask
!= 0)
785 (void) chattr_fd(fdt
, chattr_flags
, chattr_mask
& ~CHATTR_EARLY_FL
, NULL
);
787 if (close(fdt
) < 0) {
795 int copy_file_atomic_full(
799 unsigned chattr_flags
,
800 unsigned chattr_mask
,
801 CopyFlags copy_flags
,
802 copy_progress_bytes_t progress_bytes
,
805 _cleanup_(unlink_and_freep
) char *t
= NULL
;
806 _cleanup_close_
int fdt
= -1;
812 /* We try to use O_TMPFILE here to create the file if we can. Note that that only works if COPY_REPLACE is not
813 * set though as we need to use linkat() for linking the O_TMPFILE file into the file system but that system
814 * call can't replace existing files. Hence, if COPY_REPLACE is set we create a temporary name in the file
815 * system right-away and unconditionally which we then can renameat() to the right name after we completed
818 if (copy_flags
& COPY_REPLACE
) {
819 r
= tempfn_random(to
, NULL
, &t
);
823 fdt
= open(t
, O_CREAT
|O_EXCL
|O_NOFOLLOW
|O_NOCTTY
|O_WRONLY
|O_CLOEXEC
, 0600);
829 fdt
= open_tmpfile_linkable(to
, O_WRONLY
|O_CLOEXEC
, &t
);
834 if (chattr_mask
!= 0)
835 (void) chattr_fd(fdt
, chattr_flags
, chattr_mask
& CHATTR_EARLY_FL
, NULL
);
837 r
= copy_file_fd_full(from
, fdt
, copy_flags
, progress_bytes
, userdata
);
841 if (fchmod(fdt
, mode
) < 0)
844 if (copy_flags
& COPY_REPLACE
) {
845 if (renameat(AT_FDCWD
, t
, AT_FDCWD
, to
) < 0)
848 r
= link_tmpfile(fdt
, t
, to
);
853 if (chattr_mask
!= 0)
854 (void) chattr_fd(fdt
, chattr_flags
, chattr_mask
& ~CHATTR_EARLY_FL
, NULL
);
860 int copy_times(int fdf
, int fdt
, CopyFlags flags
) {
861 struct timespec ut
[2];
867 if (fstat(fdf
, &st
) < 0)
873 if (futimens(fdt
, ut
) < 0)
876 if (FLAGS_SET(flags
, COPY_CRTIME
)) {
879 if (fd_getcrtime(fdf
, &crtime
) >= 0)
880 (void) fd_setcrtime(fdt
, crtime
);
886 int copy_xattr(int fdf
, int fdt
) {
887 _cleanup_free_
char *bufa
= NULL
, *bufb
= NULL
;
888 size_t sza
= 100, szb
= 100;
898 n
= flistxattr(fdf
, bufa
, sza
);
916 assert(l
< (size_t) n
);
918 if (startswith(p
, "user.")) {
927 m
= fgetxattr(fdf
, p
, bufb
, szb
);
929 if (errno
== ERANGE
) {
938 if (fsetxattr(fdt
, p
, bufb
, m
, 0) < 0)