1 /* SPDX-License-Identifier: LGPL-2.1+ */
3 This file is part of systemd.
5 Copyright 2014 Lennart Poettering
15 #include <sys/sendfile.h>
16 //#include <sys/stat.h>
17 #include <sys/xattr.h>
21 //#include "alloc-util.h"
22 //#include "btrfs-util.h"
23 //#include "chattr-util.h"
25 //#include "dirent-util.h"
26 //#include "fd-util.h"
28 //#include "fs-util.h"
32 //#include "string-util.h"
34 #include "time-util.h"
35 //#include "umask-util.h"
36 #include "user-util.h"
37 //#include "xattr-util.h"
39 #define COPY_BUFFER_SIZE (16*1024u)
41 static ssize_t try_copy_file_range(
42 int fd_in, loff_t *off_in,
43 int fd_out, loff_t *off_out,
53 r = copy_file_range(fd_in, off_in, fd_out, off_out, len, flags);
55 have = r >= 0 || errno != ENOSYS;
65 FD_IS_NONBLOCKING_PIPE,
68 static int fd_is_nonblock_pipe(int fd) {
72 /* Checks whether the specified file descriptor refers to a pipe, and if so if is has O_NONBLOCK set. */
74 if (fstat(fd, &st) < 0)
77 if (!S_ISFIFO(st.st_mode))
80 flags = fcntl(fd, F_GETFL);
84 return (flags & O_NONBLOCK) == O_NONBLOCK ? FD_IS_NONBLOCKING_PIPE : FD_IS_BLOCKING_PIPE;
92 size_t *ret_remains_size) {
94 bool try_cfr = true, try_sendfile = true, try_splice = true;
95 int r, nonblock_pipe = -1;
96 size_t m = SSIZE_MAX; /* that is the maximum that sendfile and c_f_r accept */
101 /* Tries to copy bytes from the file descriptor 'fdf' to 'fdt' in the smartest possible way. Copies a maximum
102 * of 'max_bytes', which may be specified as UINT64_MAX, in which no maximum is applied. Returns negative on
103 * error, zero if EOF is hit before the bytes limit is hit and positive otherwise. If the copy fails for some
104 * reason but we read but didn't yet write some data an ret_remains/ret_remains_size is not NULL, then it will
105 * be initialized with an allocated buffer containing this "remaining" data. Note that these two parameters are
106 * initialized with a valid buffer only on failure and only if there's actually data already read. Otherwise
107 * these parameters if non-NULL are set to NULL. */
111 if (ret_remains_size)
112 *ret_remains_size = 0;
114 #if 0 /// UNNEEDED by elogind
115 /* Try btrfs reflinks first. This only works on regular, seekable files, hence let's check the file offsets of
116 * source and destination first. */
117 if ((copy_flags & COPY_REFLINK)) {
120 foffset = lseek(fdf, 0, SEEK_CUR);
124 toffset = lseek(fdt, 0, SEEK_CUR);
127 if (foffset == 0 && toffset == 0 && max_bytes == UINT64_MAX)
128 r = btrfs_reflink(fdf, fdt); /* full file reflink */
130 r = btrfs_clone_range(fdf, foffset, fdt, toffset, max_bytes == UINT64_MAX ? 0 : max_bytes); /* partial reflink */
134 /* This worked, yay! Now — to be fully correct — let's adjust the file pointers */
135 if (max_bytes == UINT64_MAX) {
137 /* We cloned to the end of the source file, let's position the read
138 * pointer there, and query it at the same time. */
139 t = lseek(fdf, 0, SEEK_END);
145 /* Let's adjust the destination file write pointer by the same number
147 t = lseek(fdt, toffset + (t - foffset), SEEK_SET);
151 return 0; /* we copied the whole thing, hence hit EOF, return 0 */
153 t = lseek(fdf, foffset + max_bytes, SEEK_SET);
157 t = lseek(fdt, toffset + max_bytes, SEEK_SET);
161 return 1; /* we copied only some number of bytes, which worked, but this means we didn't hit EOF, return 1 */
165 log_debug_errno(r, "Reflinking didn't work, falling back to non-reflink copying: %m");
174 if (max_bytes != (uint64_t) -1) {
176 return 1; /* return > 0 if we hit the max_bytes limit */
182 /* First try copy_file_range(), unless we already tried */
184 n = try_copy_file_range(fdf, NULL, fdt, NULL, m, 0u);
186 if (!IN_SET(n, -EINVAL, -ENOSYS, -EXDEV, -EBADF))
190 /* use fallback below */
191 } else if (n == 0) /* EOF */
198 /* First try sendfile(), unless we already tried */
200 n = sendfile(fdt, fdf, NULL, m);
202 if (!IN_SET(errno, EINVAL, ENOSYS))
205 try_sendfile = false;
206 /* use fallback below */
207 } else if (n == 0) /* EOF */
214 /* Then try splice, unless we already tried. */
217 /* splice()'s asynchronous I/O support is a bit weird. When it encounters a pipe file
218 * descriptor, then it will ignore its O_NONBLOCK flag and instead only honour the
219 * SPLICE_F_NONBLOCK flag specified in its flag parameter. Let's hide this behaviour here, and
220 * check if either of the specified fds are a pipe, and if so, let's pass the flag
221 * automatically, depending on O_NONBLOCK being set.
223 * Here's a twist though: when we use it to move data between two pipes of which one has
224 * O_NONBLOCK set and the other has not, then we have no individual control over O_NONBLOCK
225 * behaviour. Hence in that case we can't use splice() and still guarantee systematic
226 * O_NONBLOCK behaviour, hence don't. */
228 if (nonblock_pipe < 0) {
231 /* Check if either of these fds is a pipe, and if so non-blocking or not */
232 a = fd_is_nonblock_pipe(fdf);
236 b = fd_is_nonblock_pipe(fdt);
240 if ((a == FD_IS_NO_PIPE && b == FD_IS_NO_PIPE) ||
241 (a == FD_IS_BLOCKING_PIPE && b == FD_IS_NONBLOCKING_PIPE) ||
242 (a == FD_IS_NONBLOCKING_PIPE && b == FD_IS_BLOCKING_PIPE))
244 /* splice() only works if one of the fds is a pipe. If neither is, let's skip
245 * this step right-away. As mentioned above, if one of the two fds refers to a
246 * blocking pipe and the other to a non-blocking pipe, we can't use splice()
247 * either, hence don't try either. This hence means we can only use splice() if
248 * either only one of the two fds is a pipe, or if both are pipes with the same
249 * nonblocking flag setting. */
253 nonblock_pipe = a == FD_IS_NONBLOCKING_PIPE || b == FD_IS_NONBLOCKING_PIPE;
258 n = splice(fdf, NULL, fdt, NULL, m, nonblock_pipe ? SPLICE_F_NONBLOCK : 0);
260 if (!IN_SET(errno, EINVAL, ENOSYS))
264 /* use fallback below */
265 } else if (n == 0) /* EOF */
272 /* As a fallback just copy bits by hand */
274 uint8_t buf[MIN(m, COPY_BUFFER_SIZE)], *p = buf;
277 n = read(fdf, buf, sizeof buf);
280 if (n == 0) /* EOF */
287 k = write(fdt, p, z);
301 if (ret_remains_size)
302 *ret_remains_size = z;
314 if (max_bytes != (uint64_t) -1) {
315 assert(max_bytes >= (uint64_t) n);
318 /* sendfile accepts at most SSIZE_MAX-offset bytes to copy,
319 * so reduce our maximum by the amount we already copied,
320 * but don't go below our copy buffer size, unless we are
321 * close the limit of bytes we are allowed to copy. */
322 m = MAX(MIN(COPY_BUFFER_SIZE, max_bytes), m - n);
325 return 0; /* return 0 if we hit EOF earlier than the size limit */
328 #if 0 /// UNNEEDED by elogind
329 static int fd_copy_symlink(
332 const struct stat *st,
337 CopyFlags copy_flags) {
339 _cleanup_free_ char *target = NULL;
346 r = readlinkat_malloc(df, from, &target);
350 if (symlinkat(target, dt, to) < 0)
354 uid_is_valid(override_uid) ? override_uid : st->st_uid,
355 gid_is_valid(override_gid) ? override_gid : st->st_gid,
356 AT_SYMLINK_NOFOLLOW) < 0)
362 static int fd_copy_regular(
365 const struct stat *st,
370 CopyFlags copy_flags) {
372 _cleanup_close_ int fdf = -1, fdt = -1;
373 struct timespec ts[2];
380 fdf = openat(df, from, O_RDONLY|O_CLOEXEC|O_NOCTTY|O_NOFOLLOW);
384 fdt = openat(dt, to, O_WRONLY|O_CREAT|O_EXCL|O_CLOEXEC|O_NOCTTY|O_NOFOLLOW, st->st_mode & 07777);
388 r = copy_bytes(fdf, fdt, (uint64_t) -1, copy_flags);
390 (void) unlinkat(dt, to, 0);
395 uid_is_valid(override_uid) ? override_uid : st->st_uid,
396 gid_is_valid(override_gid) ? override_gid : st->st_gid) < 0)
399 if (fchmod(fdt, st->st_mode & 07777) < 0)
404 (void) futimens(fdt, ts);
405 (void) copy_xattr(fdf, fdt);
412 (void) unlinkat(dt, to, 0);
418 static int fd_copy_fifo(
421 const struct stat *st,
426 CopyFlags copy_flags) {
433 r = mkfifoat(dt, to, st->st_mode & 07777);
438 uid_is_valid(override_uid) ? override_uid : st->st_uid,
439 gid_is_valid(override_gid) ? override_gid : st->st_gid,
440 AT_SYMLINK_NOFOLLOW) < 0)
443 if (fchmodat(dt, to, st->st_mode & 07777, 0) < 0)
449 static int fd_copy_node(
452 const struct stat *st,
457 CopyFlags copy_flags) {
464 r = mknodat(dt, to, st->st_mode, st->st_rdev);
469 uid_is_valid(override_uid) ? override_uid : st->st_uid,
470 gid_is_valid(override_gid) ? override_gid : st->st_gid,
471 AT_SYMLINK_NOFOLLOW) < 0)
474 if (fchmodat(dt, to, st->st_mode & 07777, 0) < 0)
480 static int fd_copy_directory(
483 const struct stat *st,
486 dev_t original_device,
489 CopyFlags copy_flags) {
491 _cleanup_close_ int fdf = -1, fdt = -1;
492 _cleanup_closedir_ DIR *d = NULL;
501 fdf = openat(df, from, O_RDONLY|O_DIRECTORY|O_CLOEXEC|O_NOCTTY|O_NOFOLLOW);
503 fdf = fcntl(df, F_DUPFD_CLOEXEC, 3);
512 r = mkdirat(dt, to, st->st_mode & 07777);
515 else if (errno == EEXIST && (copy_flags & COPY_MERGE))
520 fdt = openat(dt, to, O_RDONLY|O_DIRECTORY|O_CLOEXEC|O_NOCTTY|O_NOFOLLOW);
526 FOREACH_DIRENT_ALL(de, d, return -errno) {
530 if (dot_or_dot_dot(de->d_name))
533 if (fstatat(dirfd(d), de->d_name, &buf, AT_SYMLINK_NOFOLLOW) < 0) {
538 if (buf.st_dev != original_device)
541 if (S_ISREG(buf.st_mode))
542 q = fd_copy_regular(dirfd(d), de->d_name, &buf, fdt, de->d_name, override_uid, override_gid, copy_flags);
543 else if (S_ISDIR(buf.st_mode))
544 q = fd_copy_directory(dirfd(d), de->d_name, &buf, fdt, de->d_name, original_device, override_uid, override_gid, copy_flags);
545 else if (S_ISLNK(buf.st_mode))
546 q = fd_copy_symlink(dirfd(d), de->d_name, &buf, fdt, de->d_name, override_uid, override_gid, copy_flags);
547 else if (S_ISFIFO(buf.st_mode))
548 q = fd_copy_fifo(dirfd(d), de->d_name, &buf, fdt, de->d_name, override_uid, override_gid, copy_flags);
549 else if (S_ISBLK(buf.st_mode) || S_ISCHR(buf.st_mode) || S_ISSOCK(buf.st_mode))
550 q = fd_copy_node(dirfd(d), de->d_name, &buf, fdt, de->d_name, override_uid, override_gid, copy_flags);
554 if (q == -EEXIST && (copy_flags & COPY_MERGE))
562 struct timespec ut[2] = {
568 uid_is_valid(override_uid) ? override_uid : st->st_uid,
569 gid_is_valid(override_gid) ? override_gid : st->st_gid) < 0)
572 if (fchmod(fdt, st->st_mode & 07777) < 0)
575 (void) copy_xattr(dirfd(d), fdt);
576 (void) futimens(fdt, ut);
582 int copy_tree_at(int fdf, const char *from, int fdt, const char *to, uid_t override_uid, gid_t override_gid, CopyFlags copy_flags) {
588 if (fstatat(fdf, from, &st, AT_SYMLINK_NOFOLLOW) < 0)
591 if (S_ISREG(st.st_mode))
592 return fd_copy_regular(fdf, from, &st, fdt, to, override_uid, override_gid, copy_flags);
593 else if (S_ISDIR(st.st_mode))
594 return fd_copy_directory(fdf, from, &st, fdt, to, st.st_dev, override_uid, override_gid, copy_flags);
595 else if (S_ISLNK(st.st_mode))
596 return fd_copy_symlink(fdf, from, &st, fdt, to, override_uid, override_gid, copy_flags);
597 else if (S_ISFIFO(st.st_mode))
598 return fd_copy_fifo(fdf, from, &st, fdt, to, override_uid, override_gid, copy_flags);
599 else if (S_ISBLK(st.st_mode) || S_ISCHR(st.st_mode) || S_ISSOCK(st.st_mode))
600 return fd_copy_node(fdf, from, &st, fdt, to, override_uid, override_gid, copy_flags);
605 int copy_tree(const char *from, const char *to, uid_t override_uid, gid_t override_gid, CopyFlags copy_flags) {
606 return copy_tree_at(AT_FDCWD, from, AT_FDCWD, to, override_uid, override_gid, copy_flags);
609 int copy_directory_fd(int dirfd, const char *to, CopyFlags copy_flags) {
615 if (fstat(dirfd, &st) < 0)
618 if (!S_ISDIR(st.st_mode))
621 return fd_copy_directory(dirfd, NULL, &st, AT_FDCWD, to, st.st_dev, UID_INVALID, GID_INVALID, copy_flags);
624 int copy_directory(const char *from, const char *to, CopyFlags copy_flags) {
630 if (lstat(from, &st) < 0)
633 if (!S_ISDIR(st.st_mode))
636 return fd_copy_directory(AT_FDCWD, from, &st, AT_FDCWD, to, st.st_dev, UID_INVALID, GID_INVALID, copy_flags);
639 int copy_file_fd(const char *from, int fdt, CopyFlags copy_flags) {
640 _cleanup_close_ int fdf = -1;
646 fdf = open(from, O_RDONLY|O_CLOEXEC|O_NOCTTY);
650 r = copy_bytes(fdf, fdt, (uint64_t) -1, copy_flags);
652 (void) copy_times(fdf, fdt);
653 (void) copy_xattr(fdf, fdt);
658 int copy_file(const char *from, const char *to, int flags, mode_t mode, unsigned chattr_flags, CopyFlags copy_flags) {
664 RUN_WITH_UMASK(0000) {
665 fdt = open(to, flags|O_WRONLY|O_CREAT|O_CLOEXEC|O_NOCTTY, mode);
670 if (chattr_flags != 0)
671 (void) chattr_fd(fdt, chattr_flags, (unsigned) -1);
673 r = copy_file_fd(from, fdt, copy_flags);
680 if (close(fdt) < 0) {
688 int copy_file_atomic(const char *from, const char *to, mode_t mode, unsigned chattr_flags, CopyFlags copy_flags) {
689 _cleanup_free_ char *t = NULL;
695 r = tempfn_random(to, NULL, &t);
699 r = copy_file(from, t, O_NOFOLLOW|O_EXCL, mode, chattr_flags, copy_flags);
703 if (copy_flags & COPY_REPLACE) {
704 r = renameat(AT_FDCWD, t, AT_FDCWD, to);
708 r = rename_noreplace(AT_FDCWD, t, AT_FDCWD, to);
717 int copy_times(int fdf, int fdt) {
718 struct timespec ut[2];
725 if (fstat(fdf, &st) < 0)
731 if (futimens(fdt, ut) < 0)
734 if (fd_getcrtime(fdf, &crtime) >= 0)
735 (void) fd_setcrtime(fdt, crtime);
740 int copy_xattr(int fdf, int fdt) {
741 _cleanup_free_ char *bufa = NULL, *bufb = NULL;
742 size_t sza = 100, szb = 100;
752 n = flistxattr(fdf, bufa, sza);
770 assert(l < (size_t) n);
772 if (startswith(p, "user.")) {
781 m = fgetxattr(fdf, p, bufb, szb);
783 if (errno == ERANGE) {
792 if (fsetxattr(fdt, p, bufb, m, 0) < 0)