2 This file is part of systemd.
4 Copyright 2010 Lennart Poettering
6 systemd is free software; you can redistribute it and/or modify it
7 under the terms of the GNU Lesser General Public License as published by
8 the Free Software Foundation; either version 2.1 of the License, or
9 (at your option) any later version.
11 systemd is distributed in the hope that it will be useful, but
12 WITHOUT ANY WARRANTY; without even the implied warranty of
13 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
14 Lesser General Public License for more details.
16 You should have received a copy of the GNU Lesser General Public License
17 along with systemd; If not, see <http://www.gnu.org/licenses/>.
23 #include <sys/mount.h>
25 #include <sys/statvfs.h>
28 #include "alloc-util.h"
33 #include "mount-util.h"
34 #include "parse-util.h"
35 #include "path-util.h"
37 #include "stdio-util.h"
38 #include "string-util.h"
40 static int fd_fdinfo_mnt_id(int fd, const char *filename, int flags, int *mnt_id) {
41 char path[strlen("/proc/self/fdinfo/") + DECIMAL_STR_MAX(int)];
42 _cleanup_free_ char *fdinfo = NULL;
43 _cleanup_close_ int subfd = -1;
47 if ((flags & AT_EMPTY_PATH) && isempty(filename))
48 xsprintf(path, "/proc/self/fdinfo/%i", fd);
50 subfd = openat(fd, filename, O_CLOEXEC|O_PATH);
54 xsprintf(path, "/proc/self/fdinfo/%i", subfd);
57 r = read_full_file(path, &fdinfo, NULL);
58 if (r == -ENOENT) /* The fdinfo directory is a relatively new addition */
63 p = startswith(fdinfo, "mnt_id:");
65 p = strstr(fdinfo, "\nmnt_id:");
66 if (!p) /* The mnt_id field is a relatively new addition */
72 p += strspn(p, WHITESPACE);
73 p[strcspn(p, WHITESPACE)] = 0;
75 return safe_atoi(p, mnt_id);
78 int fd_is_mount_point(int fd, const char *filename, int flags) {
79 union file_handle_union h = FILE_HANDLE_INIT, h_parent = FILE_HANDLE_INIT;
80 int mount_id = -1, mount_id_parent = -1;
81 bool nosupp = false, check_st_dev = true;
88 /* First we will try the name_to_handle_at() syscall, which
89 * tells us the mount id and an opaque file "handle". It is
90 * not supported everywhere though (kernel compile-time
91 * option, not all file systems are hooked up). If it works
92 * the mount id is usually good enough to tell us whether
93 * something is a mount point.
95 * If that didn't work we will try to read the mount id from
96 * /proc/self/fdinfo/<fd>. This is almost as good as
97 * name_to_handle_at(), however, does not return the
98 * opaque file handle. The opaque file handle is pretty useful
99 * to detect the root directory, which we should always
100 * consider a mount point. Hence we use this only as
101 * fallback. Exporting the mnt_id in fdinfo is a pretty recent
104 * As last fallback we do traditional fstat() based st_dev
105 * comparisons. This is how things were traditionally done,
106 * but unionfs breaks this since it exposes file
107 * systems with a variety of st_dev reported. Also, btrfs
108 * subvolumes have different st_dev, even though they aren't
109 * real mounts of their own. */
111 r = name_to_handle_at(fd, filename, &h.handle, &mount_id, flags);
114 /* This kernel does not support name_to_handle_at()
115 * fall back to simpler logic. */
116 goto fallback_fdinfo;
117 else if (errno == EOPNOTSUPP)
118 /* This kernel or file system does not support
119 * name_to_handle_at(), hence let's see if the
120 * upper fs supports it (in which case it is a
121 * mount point), otherwise fallback to the
122 * traditional stat() logic */
128 r = name_to_handle_at(fd, "", &h_parent.handle, &mount_id_parent, AT_EMPTY_PATH);
130 if (errno == EOPNOTSUPP) {
132 /* Neither parent nor child do name_to_handle_at()?
133 We have no choice but to fall back. */
134 goto fallback_fdinfo;
136 /* The parent can't do name_to_handle_at() but the
137 * directory we are interested in can?
138 * If so, it must be a mount point. */
144 /* The parent can do name_to_handle_at() but the
145 * directory we are interested in can't? If so, it
146 * must be a mount point. */
150 /* If the file handle for the directory we are
151 * interested in and its parent are identical, we
152 * assume this is the root directory, which is a mount
155 if (h.handle.handle_bytes == h_parent.handle.handle_bytes &&
156 h.handle.handle_type == h_parent.handle.handle_type &&
157 memcmp(h.handle.f_handle, h_parent.handle.f_handle, h.handle.handle_bytes) == 0)
160 return mount_id != mount_id_parent;
163 r = fd_fdinfo_mnt_id(fd, filename, flags, &mount_id);
164 if (r == -EOPNOTSUPP)
169 r = fd_fdinfo_mnt_id(fd, "", AT_EMPTY_PATH, &mount_id_parent);
173 if (mount_id != mount_id_parent)
176 /* Hmm, so, the mount ids are the same. This leaves one
177 * special case though for the root file system. For that,
178 * let's see if the parent directory has the same inode as we
179 * are interested in. Hence, let's also do fstat() checks now,
180 * too, but avoid the st_dev comparisons, since they aren't
181 * that useful on unionfs mounts. */
182 check_st_dev = false;
185 /* yay for fstatat() taking a different set of flags than the other
187 if (flags & AT_SYMLINK_FOLLOW)
188 flags &= ~AT_SYMLINK_FOLLOW;
190 flags |= AT_SYMLINK_NOFOLLOW;
191 if (fstatat(fd, filename, &a, flags) < 0)
194 if (fstatat(fd, "", &b, AT_EMPTY_PATH) < 0)
197 /* A directory with same device and inode as its parent? Must
198 * be the root directory */
199 if (a.st_dev == b.st_dev &&
200 a.st_ino == b.st_ino)
203 return check_st_dev && (a.st_dev != b.st_dev);
206 /* flags can be AT_SYMLINK_FOLLOW or 0 */
207 int path_is_mount_point(const char *t, int flags) {
208 _cleanup_close_ int fd = -1;
209 _cleanup_free_ char *canonical = NULL, *parent = NULL;
213 if (path_equal(t, "/"))
216 /* we need to resolve symlinks manually, we can't just rely on
217 * fd_is_mount_point() to do that for us; if we have a structure like
218 * /bin -> /usr/bin/ and /usr is a mount point, then the parent that we
219 * look at needs to be /usr, not /. */
220 if (flags & AT_SYMLINK_FOLLOW) {
221 canonical = canonicalize_file_name(t);
228 parent = dirname_malloc(t);
232 fd = openat(AT_FDCWD, parent, O_DIRECTORY|O_CLOEXEC|O_PATH);
236 return fd_is_mount_point(fd, basename(t), flags);
239 #if 0 /// UNNEEDED by elogind
240 int umount_recursive(const char *prefix, int flags) {
244 /* Try to umount everything recursively below a
245 * directory. Also, take care of stacked mounts, and keep
246 * unmounting them until they are gone. */
249 _cleanup_fclose_ FILE *proc_self_mountinfo = NULL;
254 proc_self_mountinfo = fopen("/proc/self/mountinfo", "re");
255 if (!proc_self_mountinfo)
259 _cleanup_free_ char *path = NULL, *p = NULL;
262 k = fscanf(proc_self_mountinfo,
263 "%*s " /* (1) mount id */
264 "%*s " /* (2) parent id */
265 "%*s " /* (3) major:minor */
266 "%*s " /* (4) root */
267 "%ms " /* (5) mount point */
268 "%*s" /* (6) mount options */
269 "%*[^-]" /* (7) optional fields */
270 "- " /* (8) separator */
271 "%*s " /* (9) file system type */
272 "%*s" /* (10) mount source */
273 "%*s" /* (11) mount options 2 */
274 "%*[^\n]", /* some rubbish at the end */
283 r = cunescape(path, UNESCAPE_RELAX, &p);
287 if (!path_startswith(p, prefix))
290 if (umount2(p, flags) < 0) {
291 r = log_debug_errno(errno, "Failed to umount %s: %m", p);
295 log_debug("Successfully unmounted %s", p);
308 static int get_mount_flags(const char *path, unsigned long *flags) {
311 if (statvfs(path, &buf) < 0)
317 int bind_remount_recursive(const char *prefix, bool ro, char **blacklist) {
318 _cleanup_set_free_free_ Set *done = NULL;
319 _cleanup_free_ char *cleaned = NULL;
322 /* Recursively remount a directory (and all its submounts) read-only or read-write. If the directory is already
323 * mounted, we reuse the mount and simply mark it MS_BIND|MS_RDONLY (or remove the MS_RDONLY for read-write
324 * operation). If it isn't we first make it one. Afterwards we apply MS_BIND|MS_RDONLY (or remove MS_RDONLY) to
325 * all submounts we can access, too. When mounts are stacked on the same mount point we only care for each
326 * individual "top-level" mount on each point, as we cannot influence/access the underlying mounts anyway. We
327 * do not have any effect on future submounts that might get propagated, they migt be writable. This includes
328 * future submounts that have been triggered via autofs.
330 * If the "blacklist" parameter is specified it may contain a list of subtrees to exclude from the
331 * remount operation. Note that we'll ignore the blacklist for the top-level path. */
333 cleaned = strdup(prefix);
337 path_kill_slashes(cleaned);
339 done = set_new(&string_hash_ops);
344 _cleanup_fclose_ FILE *proc_self_mountinfo = NULL;
345 _cleanup_set_free_free_ Set *todo = NULL;
346 bool top_autofs = false;
348 unsigned long orig_flags;
350 todo = set_new(&string_hash_ops);
354 proc_self_mountinfo = fopen("/proc/self/mountinfo", "re");
355 if (!proc_self_mountinfo)
359 _cleanup_free_ char *path = NULL, *p = NULL, *type = NULL;
362 k = fscanf(proc_self_mountinfo,
363 "%*s " /* (1) mount id */
364 "%*s " /* (2) parent id */
365 "%*s " /* (3) major:minor */
366 "%*s " /* (4) root */
367 "%ms " /* (5) mount point */
368 "%*s" /* (6) mount options (superblock) */
369 "%*[^-]" /* (7) optional fields */
370 "- " /* (8) separator */
371 "%ms " /* (9) file system type */
372 "%*s" /* (10) mount source */
373 "%*s" /* (11) mount options (bind mount) */
374 "%*[^\n]", /* some rubbish at the end */
384 r = cunescape(path, UNESCAPE_RELAX, &p);
388 if (!path_startswith(p, cleaned))
391 /* Ignore this mount if it is blacklisted, but only if it isn't the top-level mount we shall
393 if (!path_equal(cleaned, p)) {
394 bool blacklisted = false;
397 STRV_FOREACH(i, blacklist) {
399 if (path_equal(*i, cleaned))
402 if (!path_startswith(*i, cleaned))
405 if (path_startswith(p, *i)) {
407 log_debug("Not remounting %s, because blacklisted by %s, called for %s", p, *i, cleaned);
415 /* Let's ignore autofs mounts. If they aren't
416 * triggered yet, we want to avoid triggering
417 * them, as we don't make any guarantees for
418 * future submounts anyway. If they are
419 * already triggered, then we will find
420 * another entry for this. */
421 if (streq(type, "autofs")) {
422 top_autofs = top_autofs || path_equal(cleaned, p);
426 if (!set_contains(done, p)) {
427 r = set_consume(todo, p);
436 /* If we have no submounts to process anymore and if
437 * the root is either already done, or an autofs, we
439 if (set_isempty(todo) &&
440 (top_autofs || set_contains(done, cleaned)))
443 if (!set_contains(done, cleaned) &&
444 !set_contains(todo, cleaned)) {
445 /* The prefix directory itself is not yet a mount, make it one. */
446 if (mount(cleaned, cleaned, NULL, MS_BIND|MS_REC, NULL) < 0)
450 (void) get_mount_flags(cleaned, &orig_flags);
451 orig_flags &= ~MS_RDONLY;
453 if (mount(NULL, prefix, NULL, orig_flags|MS_BIND|MS_REMOUNT|(ro ? MS_RDONLY : 0), NULL) < 0)
456 log_debug("Made top-level directory %s a mount point.", prefix);
462 r = set_consume(done, x);
467 while ((x = set_steal_first(todo))) {
469 r = set_consume(done, x);
470 if (r == -EEXIST || r == 0)
475 /* Deal with mount points that are obstructed by a later mount */
476 r = path_is_mount_point(x, 0);
477 if (r == -ENOENT || r == 0)
482 /* Try to reuse the original flag set */
484 (void) get_mount_flags(x, &orig_flags);
485 orig_flags &= ~MS_RDONLY;
487 if (mount(NULL, x, NULL, orig_flags|MS_BIND|MS_REMOUNT|(ro ? MS_RDONLY : 0), NULL) < 0)
490 log_debug("Remounted %s read-only.", x);
495 int mount_move_root(const char *path) {
501 if (mount(path, "/", NULL, MS_MOVE, NULL) < 0)
513 bool fstype_is_network(const char *fstype) {
514 static const char table[] =
526 "pvfs2\0" /* OrangeFS */
532 x = startswith(fstype, "fuse.");
536 return nulstr_contains(table, fstype);
539 int repeat_unmount(const char *path, int flags) {
544 /* If there are multiple mounts on a mount point, this
545 * removes them all */
548 if (umount2(path, flags) < 0) {
561 const char* mode_to_inaccessible_node(mode_t mode) {
562 /* This function maps a node type to the correspondent inaccessible node type.
563 * Character and block inaccessible devices may not be created (because major=0 and minor=0),
564 * in such case we map character and block devices to the inaccessible node type socket. */
565 switch(mode & S_IFMT) {
567 return "/run/systemd/inaccessible/reg";
569 return "/run/systemd/inaccessible/dir";
571 if (access("/run/systemd/inaccessible/chr", F_OK) == 0)
572 return "/run/systemd/inaccessible/chr";
573 return "/run/systemd/inaccessible/sock";
575 if (access("/run/systemd/inaccessible/blk", F_OK) == 0)
576 return "/run/systemd/inaccessible/blk";
577 return "/run/systemd/inaccessible/sock";
579 return "/run/systemd/inaccessible/fifo";
581 return "/run/systemd/inaccessible/sock";
586 #if 0 /// UNNEEDED by elogind
587 #define FLAG(name) (flags & name ? STRINGIFY(name) "|" : "")
588 static char* mount_flags_to_string(long unsigned flags) {
590 _cleanup_free_ char *y = NULL;
591 long unsigned overflow;
593 overflow = flags & ~(MS_RDONLY |
618 if (flags == 0 || overflow != 0)
619 if (asprintf(&y, "%lx", overflow) < 0)
622 x = strjoin(FLAG(MS_RDONLY),
626 FLAG(MS_SYNCHRONOUS),
644 FLAG(MS_STRICTATIME),
650 x[strlen(x) - 1] = '\0'; /* truncate the last | */
660 const char *options) {
662 _cleanup_free_ char *fl = NULL;
664 fl = mount_flags_to_string(flags);
666 if ((flags & MS_REMOUNT) && !what && !type)
667 log_debug("Remounting %s (%s \"%s\")...",
668 where, strnull(fl), strempty(options));
669 else if (!what && !type)
670 log_debug("Mounting %s (%s \"%s\")...",
671 where, strnull(fl), strempty(options));
672 else if ((flags & MS_BIND) && !type)
673 log_debug("Bind-mounting %s on %s (%s \"%s\")...",
674 what, where, strnull(fl), strempty(options));
676 log_debug("Mounting %s on %s (%s \"%s\")...",
677 strna(type), where, strnull(fl), strempty(options));
678 if (mount(what, where, type, flags, options) < 0)
679 return log_full_errno(error_log_level, errno,
680 "Failed to mount %s on %s (%s \"%s\"): %m",
681 strna(type), where, strnull(fl), strempty(options));
685 int umount_verbose(const char *what) {
686 log_debug("Umounting %s...", what);
687 if (umount(what) < 0)
688 return log_error_errno(errno, "Failed to unmount %s: %m", what);