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);
113 if (IN_SET(errno, ENOSYS, EACCES, EPERM))
114 /* This kernel does not support name_to_handle_at() at all, or the syscall was blocked (maybe
115 * through seccomp, because we are running inside of a container?): fall back to simpler
117 goto fallback_fdinfo;
118 else if (errno == EOPNOTSUPP)
119 /* This kernel or file system does not support
120 * name_to_handle_at(), hence let's see if the
121 * upper fs supports it (in which case it is a
122 * mount point), otherwise fallback to the
123 * traditional stat() logic */
129 r = name_to_handle_at(fd, "", &h_parent.handle, &mount_id_parent, AT_EMPTY_PATH);
131 if (errno == EOPNOTSUPP) {
133 /* Neither parent nor child do name_to_handle_at()?
134 We have no choice but to fall back. */
135 goto fallback_fdinfo;
137 /* The parent can't do name_to_handle_at() but the
138 * directory we are interested in can?
139 * If so, it must be a mount point. */
145 /* The parent can do name_to_handle_at() but the
146 * directory we are interested in can't? If so, it
147 * must be a mount point. */
151 /* If the file handle for the directory we are
152 * interested in and its parent are identical, we
153 * assume this is the root directory, which is a mount
156 if (h.handle.handle_bytes == h_parent.handle.handle_bytes &&
157 h.handle.handle_type == h_parent.handle.handle_type &&
158 memcmp(h.handle.f_handle, h_parent.handle.f_handle, h.handle.handle_bytes) == 0)
161 return mount_id != mount_id_parent;
164 r = fd_fdinfo_mnt_id(fd, filename, flags, &mount_id);
165 if (IN_SET(r, -EOPNOTSUPP, -EACCES, -EPERM))
170 r = fd_fdinfo_mnt_id(fd, "", AT_EMPTY_PATH, &mount_id_parent);
174 if (mount_id != mount_id_parent)
177 /* Hmm, so, the mount ids are the same. This leaves one
178 * special case though for the root file system. For that,
179 * let's see if the parent directory has the same inode as we
180 * are interested in. Hence, let's also do fstat() checks now,
181 * too, but avoid the st_dev comparisons, since they aren't
182 * that useful on unionfs mounts. */
183 check_st_dev = false;
186 /* yay for fstatat() taking a different set of flags than the other
188 if (flags & AT_SYMLINK_FOLLOW)
189 flags &= ~AT_SYMLINK_FOLLOW;
191 flags |= AT_SYMLINK_NOFOLLOW;
192 if (fstatat(fd, filename, &a, flags) < 0)
195 if (fstatat(fd, "", &b, AT_EMPTY_PATH) < 0)
198 /* A directory with same device and inode as its parent? Must
199 * be the root directory */
200 if (a.st_dev == b.st_dev &&
201 a.st_ino == b.st_ino)
204 return check_st_dev && (a.st_dev != b.st_dev);
207 /* flags can be AT_SYMLINK_FOLLOW or 0 */
208 int path_is_mount_point(const char *t, const char *root, int flags) {
209 _cleanup_free_ char *canonical = NULL, *parent = NULL;
210 _cleanup_close_ int fd = -1;
215 if (path_equal(t, "/"))
218 /* we need to resolve symlinks manually, we can't just rely on
219 * fd_is_mount_point() to do that for us; if we have a structure like
220 * /bin -> /usr/bin/ and /usr is a mount point, then the parent that we
221 * look at needs to be /usr, not /. */
222 if (flags & AT_SYMLINK_FOLLOW) {
223 r = chase_symlinks(t, root, 0, &canonical);
230 parent = dirname_malloc(t);
234 fd = openat(AT_FDCWD, parent, O_DIRECTORY|O_CLOEXEC|O_PATH);
238 return fd_is_mount_point(fd, basename(t), flags);
241 #if 0 /// UNNEEDED by elogind
242 int umount_recursive(const char *prefix, int flags) {
246 /* Try to umount everything recursively below a
247 * directory. Also, take care of stacked mounts, and keep
248 * unmounting them until they are gone. */
251 _cleanup_fclose_ FILE *proc_self_mountinfo = NULL;
256 proc_self_mountinfo = fopen("/proc/self/mountinfo", "re");
257 if (!proc_self_mountinfo)
261 _cleanup_free_ char *path = NULL, *p = NULL;
264 k = fscanf(proc_self_mountinfo,
265 "%*s " /* (1) mount id */
266 "%*s " /* (2) parent id */
267 "%*s " /* (3) major:minor */
268 "%*s " /* (4) root */
269 "%ms " /* (5) mount point */
270 "%*s" /* (6) mount options */
271 "%*[^-]" /* (7) optional fields */
272 "- " /* (8) separator */
273 "%*s " /* (9) file system type */
274 "%*s" /* (10) mount source */
275 "%*s" /* (11) mount options 2 */
276 "%*[^\n]", /* some rubbish at the end */
285 r = cunescape(path, UNESCAPE_RELAX, &p);
289 if (!path_startswith(p, prefix))
292 if (umount2(p, flags) < 0) {
293 r = log_debug_errno(errno, "Failed to umount %s: %m", p);
297 log_debug("Successfully unmounted %s", p);
310 static int get_mount_flags(const char *path, unsigned long *flags) {
313 if (statvfs(path, &buf) < 0)
319 int bind_remount_recursive(const char *prefix, bool ro, char **blacklist) {
320 _cleanup_set_free_free_ Set *done = NULL;
321 _cleanup_free_ char *cleaned = NULL;
324 /* Recursively remount a directory (and all its submounts) read-only or read-write. If the directory is already
325 * mounted, we reuse the mount and simply mark it MS_BIND|MS_RDONLY (or remove the MS_RDONLY for read-write
326 * operation). If it isn't we first make it one. Afterwards we apply MS_BIND|MS_RDONLY (or remove MS_RDONLY) to
327 * all submounts we can access, too. When mounts are stacked on the same mount point we only care for each
328 * individual "top-level" mount on each point, as we cannot influence/access the underlying mounts anyway. We
329 * do not have any effect on future submounts that might get propagated, they migt be writable. This includes
330 * future submounts that have been triggered via autofs.
332 * If the "blacklist" parameter is specified it may contain a list of subtrees to exclude from the
333 * remount operation. Note that we'll ignore the blacklist for the top-level path. */
335 cleaned = strdup(prefix);
339 path_kill_slashes(cleaned);
341 done = set_new(&string_hash_ops);
346 _cleanup_fclose_ FILE *proc_self_mountinfo = NULL;
347 _cleanup_set_free_free_ Set *todo = NULL;
348 bool top_autofs = false;
350 unsigned long orig_flags;
352 todo = set_new(&string_hash_ops);
356 proc_self_mountinfo = fopen("/proc/self/mountinfo", "re");
357 if (!proc_self_mountinfo)
361 _cleanup_free_ char *path = NULL, *p = NULL, *type = NULL;
364 k = fscanf(proc_self_mountinfo,
365 "%*s " /* (1) mount id */
366 "%*s " /* (2) parent id */
367 "%*s " /* (3) major:minor */
368 "%*s " /* (4) root */
369 "%ms " /* (5) mount point */
370 "%*s" /* (6) mount options (superblock) */
371 "%*[^-]" /* (7) optional fields */
372 "- " /* (8) separator */
373 "%ms " /* (9) file system type */
374 "%*s" /* (10) mount source */
375 "%*s" /* (11) mount options (bind mount) */
376 "%*[^\n]", /* some rubbish at the end */
386 r = cunescape(path, UNESCAPE_RELAX, &p);
390 if (!path_startswith(p, cleaned))
393 /* Ignore this mount if it is blacklisted, but only if it isn't the top-level mount we shall
395 if (!path_equal(cleaned, p)) {
396 bool blacklisted = false;
399 STRV_FOREACH(i, blacklist) {
401 if (path_equal(*i, cleaned))
404 if (!path_startswith(*i, cleaned))
407 if (path_startswith(p, *i)) {
409 log_debug("Not remounting %s, because blacklisted by %s, called for %s", p, *i, cleaned);
417 /* Let's ignore autofs mounts. If they aren't
418 * triggered yet, we want to avoid triggering
419 * them, as we don't make any guarantees for
420 * future submounts anyway. If they are
421 * already triggered, then we will find
422 * another entry for this. */
423 if (streq(type, "autofs")) {
424 top_autofs = top_autofs || path_equal(cleaned, p);
428 if (!set_contains(done, p)) {
429 r = set_consume(todo, p);
438 /* If we have no submounts to process anymore and if
439 * the root is either already done, or an autofs, we
441 if (set_isempty(todo) &&
442 (top_autofs || set_contains(done, cleaned)))
445 if (!set_contains(done, cleaned) &&
446 !set_contains(todo, cleaned)) {
447 /* The prefix directory itself is not yet a mount, make it one. */
448 if (mount(cleaned, cleaned, NULL, MS_BIND|MS_REC, NULL) < 0)
452 (void) get_mount_flags(cleaned, &orig_flags);
453 orig_flags &= ~MS_RDONLY;
455 if (mount(NULL, prefix, NULL, orig_flags|MS_BIND|MS_REMOUNT|(ro ? MS_RDONLY : 0), NULL) < 0)
458 log_debug("Made top-level directory %s a mount point.", prefix);
464 r = set_consume(done, x);
469 while ((x = set_steal_first(todo))) {
471 r = set_consume(done, x);
472 if (r == -EEXIST || r == 0)
477 /* Deal with mount points that are obstructed by a later mount */
478 r = path_is_mount_point(x, NULL, 0);
479 if (r == -ENOENT || r == 0)
484 /* Try to reuse the original flag set */
486 (void) get_mount_flags(x, &orig_flags);
487 orig_flags &= ~MS_RDONLY;
489 if (mount(NULL, x, NULL, orig_flags|MS_BIND|MS_REMOUNT|(ro ? MS_RDONLY : 0), NULL) < 0)
492 log_debug("Remounted %s read-only.", x);
497 int mount_move_root(const char *path) {
503 if (mount(path, "/", NULL, MS_MOVE, NULL) < 0)
515 bool fstype_is_network(const char *fstype) {
516 static const char table[] =
528 "pvfs2\0" /* OrangeFS */
535 x = startswith(fstype, "fuse.");
539 return nulstr_contains(table, fstype);
542 int repeat_unmount(const char *path, int flags) {
547 /* If there are multiple mounts on a mount point, this
548 * removes them all */
551 if (umount2(path, flags) < 0) {
564 const char* mode_to_inaccessible_node(mode_t mode) {
565 /* This function maps a node type to the correspondent inaccessible node type.
566 * Character and block inaccessible devices may not be created (because major=0 and minor=0),
567 * in such case we map character and block devices to the inaccessible node type socket. */
568 switch(mode & S_IFMT) {
570 return "/run/systemd/inaccessible/reg";
572 return "/run/systemd/inaccessible/dir";
574 if (access("/run/systemd/inaccessible/chr", F_OK) == 0)
575 return "/run/systemd/inaccessible/chr";
576 return "/run/systemd/inaccessible/sock";
578 if (access("/run/systemd/inaccessible/blk", F_OK) == 0)
579 return "/run/systemd/inaccessible/blk";
580 return "/run/systemd/inaccessible/sock";
582 return "/run/systemd/inaccessible/fifo";
584 return "/run/systemd/inaccessible/sock";
589 #if 0 /// UNNEEDED by elogind
590 #define FLAG(name) (flags & name ? STRINGIFY(name) "|" : "")
591 static char* mount_flags_to_string(long unsigned flags) {
593 _cleanup_free_ char *y = NULL;
594 long unsigned overflow;
596 overflow = flags & ~(MS_RDONLY |
621 if (flags == 0 || overflow != 0)
622 if (asprintf(&y, "%lx", overflow) < 0)
625 x = strjoin(FLAG(MS_RDONLY),
629 FLAG(MS_SYNCHRONOUS),
647 FLAG(MS_STRICTATIME),
653 x[strlen(x) - 1] = '\0'; /* truncate the last | */
663 const char *options) {
665 _cleanup_free_ char *fl = NULL;
667 fl = mount_flags_to_string(flags);
669 if ((flags & MS_REMOUNT) && !what && !type)
670 log_debug("Remounting %s (%s \"%s\")...",
671 where, strnull(fl), strempty(options));
672 else if (!what && !type)
673 log_debug("Mounting %s (%s \"%s\")...",
674 where, strnull(fl), strempty(options));
675 else if ((flags & MS_BIND) && !type)
676 log_debug("Bind-mounting %s on %s (%s \"%s\")...",
677 what, where, strnull(fl), strempty(options));
678 else if (flags & MS_MOVE)
679 log_debug("Moving mount %s → %s (%s \"%s\")...",
680 what, where, strnull(fl), strempty(options));
682 log_debug("Mounting %s on %s (%s \"%s\")...",
683 strna(type), where, strnull(fl), strempty(options));
684 if (mount(what, where, type, flags, options) < 0)
685 return log_full_errno(error_log_level, errno,
686 "Failed to mount %s on %s (%s \"%s\"): %m",
687 strna(type), where, strnull(fl), strempty(options));
691 int umount_verbose(const char *what) {
692 log_debug("Umounting %s...", what);
693 if (umount(what) < 0)
694 return log_error_errno(errno, "Failed to unmount %s: %m", what);
699 const char *mount_propagation_flags_to_string(unsigned long flags) {
701 switch (flags & (MS_SHARED|MS_SLAVE|MS_PRIVATE)) {
716 int mount_propagation_flags_from_string(const char *name, unsigned long *ret) {
720 else if (streq(name, "shared"))
722 else if (streq(name, "slave"))
724 else if (streq(name, "private"))