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 O_NONBLOCK is set. */
74 if (fstat(fd, &st) < 0)
77 if (!S_ISFIFO(st.st_mode))
80 flags = fcntl(fd, F_GETFL);
84 return FLAGS_SET(flags, 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");
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;
312 if (max_bytes != (uint64_t) -1) {
313 assert(max_bytes >= (uint64_t) n);
316 /* sendfile accepts at most SSIZE_MAX-offset bytes to copy,
317 * so reduce our maximum by the amount we already copied,
318 * but don't go below our copy buffer size, unless we are
319 * close the limit of bytes we are allowed to copy. */
320 m = MAX(MIN(COPY_BUFFER_SIZE, max_bytes), m - n);
323 return 0; /* return 0 if we hit EOF earlier than the size limit */
326 #if 0 /// UNNEEDED by elogind
327 static int fd_copy_symlink(
330 const struct stat *st,
335 CopyFlags copy_flags) {
337 _cleanup_free_ char *target = NULL;
344 r = readlinkat_malloc(df, from, &target);
348 if (symlinkat(target, dt, to) < 0)
352 uid_is_valid(override_uid) ? override_uid : st->st_uid,
353 gid_is_valid(override_gid) ? override_gid : st->st_gid,
354 AT_SYMLINK_NOFOLLOW) < 0)
360 static int fd_copy_regular(
363 const struct stat *st,
368 CopyFlags copy_flags) {
370 _cleanup_close_ int fdf = -1, fdt = -1;
371 struct timespec ts[2];
378 fdf = openat(df, from, O_RDONLY|O_CLOEXEC|O_NOCTTY|O_NOFOLLOW);
382 fdt = openat(dt, to, O_WRONLY|O_CREAT|O_EXCL|O_CLOEXEC|O_NOCTTY|O_NOFOLLOW, st->st_mode & 07777);
386 r = copy_bytes(fdf, fdt, (uint64_t) -1, copy_flags);
388 (void) unlinkat(dt, to, 0);
393 uid_is_valid(override_uid) ? override_uid : st->st_uid,
394 gid_is_valid(override_gid) ? override_gid : st->st_gid) < 0)
397 if (fchmod(fdt, st->st_mode & 07777) < 0)
402 (void) futimens(fdt, ts);
403 (void) copy_xattr(fdf, fdt);
410 (void) unlinkat(dt, to, 0);
416 static int fd_copy_fifo(
419 const struct stat *st,
424 CopyFlags copy_flags) {
431 r = mkfifoat(dt, to, st->st_mode & 07777);
436 uid_is_valid(override_uid) ? override_uid : st->st_uid,
437 gid_is_valid(override_gid) ? override_gid : st->st_gid,
438 AT_SYMLINK_NOFOLLOW) < 0)
441 if (fchmodat(dt, to, st->st_mode & 07777, 0) < 0)
447 static int fd_copy_node(
450 const struct stat *st,
455 CopyFlags copy_flags) {
462 r = mknodat(dt, to, st->st_mode, st->st_rdev);
467 uid_is_valid(override_uid) ? override_uid : st->st_uid,
468 gid_is_valid(override_gid) ? override_gid : st->st_gid,
469 AT_SYMLINK_NOFOLLOW) < 0)
472 if (fchmodat(dt, to, st->st_mode & 07777, 0) < 0)
478 static int fd_copy_directory(
481 const struct stat *st,
484 dev_t original_device,
487 CopyFlags copy_flags) {
489 _cleanup_close_ int fdf = -1, fdt = -1;
490 _cleanup_closedir_ DIR *d = NULL;
499 fdf = openat(df, from, O_RDONLY|O_DIRECTORY|O_CLOEXEC|O_NOCTTY|O_NOFOLLOW);
501 fdf = fcntl(df, F_DUPFD_CLOEXEC, 3);
510 r = mkdirat(dt, to, st->st_mode & 07777);
513 else if (errno == EEXIST && (copy_flags & COPY_MERGE))
518 fdt = openat(dt, to, O_RDONLY|O_DIRECTORY|O_CLOEXEC|O_NOCTTY|O_NOFOLLOW);
524 FOREACH_DIRENT_ALL(de, d, return -errno) {
528 if (dot_or_dot_dot(de->d_name))
531 if (fstatat(dirfd(d), de->d_name, &buf, AT_SYMLINK_NOFOLLOW) < 0) {
536 if (buf.st_dev != original_device)
539 if (S_ISREG(buf.st_mode))
540 q = fd_copy_regular(dirfd(d), de->d_name, &buf, fdt, de->d_name, override_uid, override_gid, copy_flags);
541 else if (S_ISDIR(buf.st_mode))
542 q = fd_copy_directory(dirfd(d), de->d_name, &buf, fdt, de->d_name, original_device, override_uid, override_gid, copy_flags);
543 else if (S_ISLNK(buf.st_mode))
544 q = fd_copy_symlink(dirfd(d), de->d_name, &buf, fdt, de->d_name, override_uid, override_gid, copy_flags);
545 else if (S_ISFIFO(buf.st_mode))
546 q = fd_copy_fifo(dirfd(d), de->d_name, &buf, fdt, de->d_name, override_uid, override_gid, copy_flags);
547 else if (S_ISBLK(buf.st_mode) || S_ISCHR(buf.st_mode) || S_ISSOCK(buf.st_mode))
548 q = fd_copy_node(dirfd(d), de->d_name, &buf, fdt, de->d_name, override_uid, override_gid, copy_flags);
552 if (q == -EEXIST && (copy_flags & COPY_MERGE))
560 struct timespec ut[2] = {
566 uid_is_valid(override_uid) ? override_uid : st->st_uid,
567 gid_is_valid(override_gid) ? override_gid : st->st_gid) < 0)
570 if (fchmod(fdt, st->st_mode & 07777) < 0)
573 (void) copy_xattr(dirfd(d), fdt);
574 (void) futimens(fdt, ut);
580 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) {
586 if (fstatat(fdf, from, &st, AT_SYMLINK_NOFOLLOW) < 0)
589 if (S_ISREG(st.st_mode))
590 return fd_copy_regular(fdf, from, &st, fdt, to, override_uid, override_gid, copy_flags);
591 else if (S_ISDIR(st.st_mode))
592 return fd_copy_directory(fdf, from, &st, fdt, to, st.st_dev, override_uid, override_gid, copy_flags);
593 else if (S_ISLNK(st.st_mode))
594 return fd_copy_symlink(fdf, from, &st, fdt, to, override_uid, override_gid, copy_flags);
595 else if (S_ISFIFO(st.st_mode))
596 return fd_copy_fifo(fdf, from, &st, fdt, to, override_uid, override_gid, copy_flags);
597 else if (S_ISBLK(st.st_mode) || S_ISCHR(st.st_mode) || S_ISSOCK(st.st_mode))
598 return fd_copy_node(fdf, from, &st, fdt, to, override_uid, override_gid, copy_flags);
603 int copy_tree(const char *from, const char *to, uid_t override_uid, gid_t override_gid, CopyFlags copy_flags) {
604 return copy_tree_at(AT_FDCWD, from, AT_FDCWD, to, override_uid, override_gid, copy_flags);
607 int copy_directory_fd(int dirfd, const char *to, CopyFlags copy_flags) {
613 if (fstat(dirfd, &st) < 0)
616 if (!S_ISDIR(st.st_mode))
619 return fd_copy_directory(dirfd, NULL, &st, AT_FDCWD, to, st.st_dev, UID_INVALID, GID_INVALID, copy_flags);
622 int copy_directory(const char *from, const char *to, CopyFlags copy_flags) {
628 if (lstat(from, &st) < 0)
631 if (!S_ISDIR(st.st_mode))
634 return fd_copy_directory(AT_FDCWD, from, &st, AT_FDCWD, to, st.st_dev, UID_INVALID, GID_INVALID, copy_flags);
637 int copy_file_fd(const char *from, int fdt, CopyFlags copy_flags) {
638 _cleanup_close_ int fdf = -1;
644 fdf = open(from, O_RDONLY|O_CLOEXEC|O_NOCTTY);
648 r = copy_bytes(fdf, fdt, (uint64_t) -1, copy_flags);
650 (void) copy_times(fdf, fdt);
651 (void) copy_xattr(fdf, fdt);
656 int copy_file(const char *from, const char *to, int flags, mode_t mode, unsigned chattr_flags, CopyFlags copy_flags) {
662 RUN_WITH_UMASK(0000) {
663 fdt = open(to, flags|O_WRONLY|O_CREAT|O_CLOEXEC|O_NOCTTY, mode);
668 if (chattr_flags != 0)
669 (void) chattr_fd(fdt, chattr_flags, (unsigned) -1);
671 r = copy_file_fd(from, fdt, copy_flags);
678 if (close(fdt) < 0) {
686 int copy_file_atomic(const char *from, const char *to, mode_t mode, unsigned chattr_flags, CopyFlags copy_flags) {
687 _cleanup_free_ char *t = NULL;
693 r = tempfn_random(to, NULL, &t);
697 r = copy_file(from, t, O_NOFOLLOW|O_EXCL, mode, chattr_flags, copy_flags);
701 if (copy_flags & COPY_REPLACE) {
702 r = renameat(AT_FDCWD, t, AT_FDCWD, to);
706 r = rename_noreplace(AT_FDCWD, t, AT_FDCWD, to);
715 int copy_times(int fdf, int fdt) {
716 struct timespec ut[2];
723 if (fstat(fdf, &st) < 0)
729 if (futimens(fdt, ut) < 0)
732 if (fd_getcrtime(fdf, &crtime) >= 0)
733 (void) fd_setcrtime(fdt, crtime);
738 int copy_xattr(int fdf, int fdt) {
739 _cleanup_free_ char *bufa = NULL, *bufb = NULL;
740 size_t sza = 100, szb = 100;
750 n = flistxattr(fdf, bufa, sza);
768 assert(l < (size_t) n);
770 if (startswith(p, "user.")) {
779 m = fgetxattr(fdf, p, bufb, szb);
781 if (errno == ERANGE) {
790 if (fsetxattr(fdt, p, bufb, m, 0) < 0)