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"
34 #include "mount-util.h"
35 #include "parse-util.h"
36 #include "path-util.h"
38 #include "stdio-util.h"
39 #include "string-util.h"
42 static int fd_fdinfo_mnt_id(int fd, const char *filename, int flags, int *mnt_id) {
43 char path[strlen("/proc/self/fdinfo/") + DECIMAL_STR_MAX(int)];
44 _cleanup_free_ char *fdinfo = NULL;
45 _cleanup_close_ int subfd = -1;
49 if ((flags & AT_EMPTY_PATH) && isempty(filename))
50 xsprintf(path, "/proc/self/fdinfo/%i", fd);
52 subfd = openat(fd, filename, O_CLOEXEC|O_PATH);
56 xsprintf(path, "/proc/self/fdinfo/%i", subfd);
59 r = read_full_file(path, &fdinfo, NULL);
60 if (r == -ENOENT) /* The fdinfo directory is a relatively new addition */
65 p = startswith(fdinfo, "mnt_id:");
67 p = strstr(fdinfo, "\nmnt_id:");
68 if (!p) /* The mnt_id field is a relatively new addition */
74 p += strspn(p, WHITESPACE);
75 p[strcspn(p, WHITESPACE)] = 0;
77 return safe_atoi(p, mnt_id);
80 int fd_is_mount_point(int fd, const char *filename, int flags) {
81 union file_handle_union h = FILE_HANDLE_INIT, h_parent = FILE_HANDLE_INIT;
82 int mount_id = -1, mount_id_parent = -1;
83 bool nosupp = false, check_st_dev = true;
90 /* First we will try the name_to_handle_at() syscall, which
91 * tells us the mount id and an opaque file "handle". It is
92 * not supported everywhere though (kernel compile-time
93 * option, not all file systems are hooked up). If it works
94 * the mount id is usually good enough to tell us whether
95 * something is a mount point.
97 * If that didn't work we will try to read the mount id from
98 * /proc/self/fdinfo/<fd>. This is almost as good as
99 * name_to_handle_at(), however, does not return the
100 * opaque file handle. The opaque file handle is pretty useful
101 * to detect the root directory, which we should always
102 * consider a mount point. Hence we use this only as
103 * fallback. Exporting the mnt_id in fdinfo is a pretty recent
106 * As last fallback we do traditional fstat() based st_dev
107 * comparisons. This is how things were traditionally done,
108 * but unionfs breaks this since it exposes file
109 * systems with a variety of st_dev reported. Also, btrfs
110 * subvolumes have different st_dev, even though they aren't
111 * real mounts of their own. */
113 r = name_to_handle_at(fd, filename, &h.handle, &mount_id, flags);
115 if (IN_SET(errno, ENOSYS, EACCES, EPERM))
116 /* This kernel does not support name_to_handle_at() at all, or the syscall was blocked (maybe
117 * through seccomp, because we are running inside of a container?): fall back to simpler
119 goto fallback_fdinfo;
120 else if (errno == EOPNOTSUPP)
121 /* This kernel or file system does not support
122 * name_to_handle_at(), hence let's see if the
123 * upper fs supports it (in which case it is a
124 * mount point), otherwise fallback to the
125 * traditional stat() logic */
131 r = name_to_handle_at(fd, "", &h_parent.handle, &mount_id_parent, AT_EMPTY_PATH);
133 if (errno == EOPNOTSUPP) {
135 /* Neither parent nor child do name_to_handle_at()?
136 We have no choice but to fall back. */
137 goto fallback_fdinfo;
139 /* The parent can't do name_to_handle_at() but the
140 * directory we are interested in can?
141 * If so, it must be a mount point. */
147 /* The parent can do name_to_handle_at() but the
148 * directory we are interested in can't? If so, it
149 * must be a mount point. */
153 /* If the file handle for the directory we are
154 * interested in and its parent are identical, we
155 * assume this is the root directory, which is a mount
158 if (h.handle.handle_bytes == h_parent.handle.handle_bytes &&
159 h.handle.handle_type == h_parent.handle.handle_type &&
160 memcmp(h.handle.f_handle, h_parent.handle.f_handle, h.handle.handle_bytes) == 0)
163 return mount_id != mount_id_parent;
166 r = fd_fdinfo_mnt_id(fd, filename, flags, &mount_id);
167 if (IN_SET(r, -EOPNOTSUPP, -EACCES, -EPERM))
172 r = fd_fdinfo_mnt_id(fd, "", AT_EMPTY_PATH, &mount_id_parent);
176 if (mount_id != mount_id_parent)
179 /* Hmm, so, the mount ids are the same. This leaves one
180 * special case though for the root file system. For that,
181 * let's see if the parent directory has the same inode as we
182 * are interested in. Hence, let's also do fstat() checks now,
183 * too, but avoid the st_dev comparisons, since they aren't
184 * that useful on unionfs mounts. */
185 check_st_dev = false;
188 /* yay for fstatat() taking a different set of flags than the other
190 if (flags & AT_SYMLINK_FOLLOW)
191 flags &= ~AT_SYMLINK_FOLLOW;
193 flags |= AT_SYMLINK_NOFOLLOW;
194 if (fstatat(fd, filename, &a, flags) < 0)
197 if (fstatat(fd, "", &b, AT_EMPTY_PATH) < 0)
200 /* A directory with same device and inode as its parent? Must
201 * be the root directory */
202 if (a.st_dev == b.st_dev &&
203 a.st_ino == b.st_ino)
206 return check_st_dev && (a.st_dev != b.st_dev);
209 /* flags can be AT_SYMLINK_FOLLOW or 0 */
210 int path_is_mount_point(const char *t, const char *root, int flags) {
211 _cleanup_free_ char *canonical = NULL, *parent = NULL;
212 _cleanup_close_ int fd = -1;
217 if (path_equal(t, "/"))
220 /* we need to resolve symlinks manually, we can't just rely on
221 * fd_is_mount_point() to do that for us; if we have a structure like
222 * /bin -> /usr/bin/ and /usr is a mount point, then the parent that we
223 * look at needs to be /usr, not /. */
224 if (flags & AT_SYMLINK_FOLLOW) {
225 r = chase_symlinks(t, root, 0, &canonical);
232 parent = dirname_malloc(t);
236 fd = openat(AT_FDCWD, parent, O_DIRECTORY|O_CLOEXEC|O_PATH);
240 return fd_is_mount_point(fd, basename(t), flags);
243 #if 0 /// UNNEEDED by elogind
244 int umount_recursive(const char *prefix, int flags) {
248 /* Try to umount everything recursively below a
249 * directory. Also, take care of stacked mounts, and keep
250 * unmounting them until they are gone. */
253 _cleanup_fclose_ FILE *proc_self_mountinfo = NULL;
258 proc_self_mountinfo = fopen("/proc/self/mountinfo", "re");
259 if (!proc_self_mountinfo)
263 _cleanup_free_ char *path = NULL, *p = NULL;
266 k = fscanf(proc_self_mountinfo,
267 "%*s " /* (1) mount id */
268 "%*s " /* (2) parent id */
269 "%*s " /* (3) major:minor */
270 "%*s " /* (4) root */
271 "%ms " /* (5) mount point */
272 "%*s" /* (6) mount options */
273 "%*[^-]" /* (7) optional fields */
274 "- " /* (8) separator */
275 "%*s " /* (9) file system type */
276 "%*s" /* (10) mount source */
277 "%*s" /* (11) mount options 2 */
278 "%*[^\n]", /* some rubbish at the end */
287 r = cunescape(path, UNESCAPE_RELAX, &p);
291 if (!path_startswith(p, prefix))
294 if (umount2(p, flags) < 0) {
295 r = log_debug_errno(errno, "Failed to umount %s: %m", p);
299 log_debug("Successfully unmounted %s", p);
312 static int get_mount_flags(const char *path, unsigned long *flags) {
315 if (statvfs(path, &buf) < 0)
321 /* Use this function only if do you have direct access to /proc/self/mountinfo
322 * and need the caller to open it for you. This is the case when /proc is
323 * masked or not mounted. Otherwise, use bind_remount_recursive. */
324 int bind_remount_recursive_with_mountinfo(const char *prefix, bool ro, char **blacklist, FILE *proc_self_mountinfo) {
325 _cleanup_set_free_free_ Set *done = NULL;
326 _cleanup_free_ char *cleaned = NULL;
329 assert(proc_self_mountinfo);
331 /* Recursively remount a directory (and all its submounts) read-only or read-write. If the directory is already
332 * mounted, we reuse the mount and simply mark it MS_BIND|MS_RDONLY (or remove the MS_RDONLY for read-write
333 * operation). If it isn't we first make it one. Afterwards we apply MS_BIND|MS_RDONLY (or remove MS_RDONLY) to
334 * all submounts we can access, too. When mounts are stacked on the same mount point we only care for each
335 * individual "top-level" mount on each point, as we cannot influence/access the underlying mounts anyway. We
336 * do not have any effect on future submounts that might get propagated, they migt be writable. This includes
337 * future submounts that have been triggered via autofs.
339 * If the "blacklist" parameter is specified it may contain a list of subtrees to exclude from the
340 * remount operation. Note that we'll ignore the blacklist for the top-level path. */
342 cleaned = strdup(prefix);
346 path_kill_slashes(cleaned);
348 done = set_new(&string_hash_ops);
353 _cleanup_set_free_free_ Set *todo = NULL;
354 bool top_autofs = false;
356 unsigned long orig_flags;
358 todo = set_new(&string_hash_ops);
362 rewind(proc_self_mountinfo);
365 _cleanup_free_ char *path = NULL, *p = NULL, *type = NULL;
368 k = fscanf(proc_self_mountinfo,
369 "%*s " /* (1) mount id */
370 "%*s " /* (2) parent id */
371 "%*s " /* (3) major:minor */
372 "%*s " /* (4) root */
373 "%ms " /* (5) mount point */
374 "%*s" /* (6) mount options (superblock) */
375 "%*[^-]" /* (7) optional fields */
376 "- " /* (8) separator */
377 "%ms " /* (9) file system type */
378 "%*s" /* (10) mount source */
379 "%*s" /* (11) mount options (bind mount) */
380 "%*[^\n]", /* some rubbish at the end */
390 r = cunescape(path, UNESCAPE_RELAX, &p);
394 if (!path_startswith(p, cleaned))
397 /* Ignore this mount if it is blacklisted, but only if it isn't the top-level mount we shall
399 if (!path_equal(cleaned, p)) {
400 bool blacklisted = false;
403 STRV_FOREACH(i, blacklist) {
405 if (path_equal(*i, cleaned))
408 if (!path_startswith(*i, cleaned))
411 if (path_startswith(p, *i)) {
413 log_debug("Not remounting %s, because blacklisted by %s, called for %s", p, *i, cleaned);
421 /* Let's ignore autofs mounts. If they aren't
422 * triggered yet, we want to avoid triggering
423 * them, as we don't make any guarantees for
424 * future submounts anyway. If they are
425 * already triggered, then we will find
426 * another entry for this. */
427 if (streq(type, "autofs")) {
428 top_autofs = top_autofs || path_equal(cleaned, p);
432 if (!set_contains(done, p)) {
433 r = set_consume(todo, p);
442 /* If we have no submounts to process anymore and if
443 * the root is either already done, or an autofs, we
445 if (set_isempty(todo) &&
446 (top_autofs || set_contains(done, cleaned)))
449 if (!set_contains(done, cleaned) &&
450 !set_contains(todo, cleaned)) {
451 /* The prefix directory itself is not yet a mount, make it one. */
452 if (mount(cleaned, cleaned, NULL, MS_BIND|MS_REC, NULL) < 0)
456 (void) get_mount_flags(cleaned, &orig_flags);
457 orig_flags &= ~MS_RDONLY;
459 if (mount(NULL, prefix, NULL, orig_flags|MS_BIND|MS_REMOUNT|(ro ? MS_RDONLY : 0), NULL) < 0)
462 log_debug("Made top-level directory %s a mount point.", prefix);
468 r = set_consume(done, x);
473 while ((x = set_steal_first(todo))) {
475 r = set_consume(done, x);
476 if (IN_SET(r, 0, -EEXIST))
481 /* Deal with mount points that are obstructed by a later mount */
482 r = path_is_mount_point(x, NULL, 0);
483 if (IN_SET(r, 0, -ENOENT))
488 /* Try to reuse the original flag set */
490 (void) get_mount_flags(x, &orig_flags);
491 orig_flags &= ~MS_RDONLY;
493 if (mount(NULL, x, NULL, orig_flags|MS_BIND|MS_REMOUNT|(ro ? MS_RDONLY : 0), NULL) < 0)
496 log_debug("Remounted %s read-only.", x);
501 int bind_remount_recursive(const char *prefix, bool ro, char **blacklist) {
502 _cleanup_fclose_ FILE *proc_self_mountinfo = NULL;
504 proc_self_mountinfo = fopen("/proc/self/mountinfo", "re");
505 if (!proc_self_mountinfo)
508 return bind_remount_recursive_with_mountinfo(prefix, ro, blacklist, proc_self_mountinfo);
511 int mount_move_root(const char *path) {
517 if (mount(path, "/", NULL, MS_MOVE, NULL) < 0)
529 bool fstype_is_network(const char *fstype) {
532 x = startswith(fstype, "fuse.");
536 return STR_IN_SET(fstype,
548 "pvfs2", /* OrangeFS */
553 bool fstype_is_api_vfs(const char *fstype) {
554 return STR_IN_SET(fstype,
577 bool fstype_is_ro(const char *fstype) {
578 /* All Linux file systems that are necessarily read-only */
579 return STR_IN_SET(fstype,
585 bool fstype_can_discard(const char *fstype) {
586 return STR_IN_SET(fstype,
593 int repeat_unmount(const char *path, int flags) {
598 /* If there are multiple mounts on a mount point, this
599 * removes them all */
602 if (umount2(path, flags) < 0) {
615 const char* mode_to_inaccessible_node(mode_t mode) {
616 /* This function maps a node type to the correspondent inaccessible node type.
617 * Character and block inaccessible devices may not be created (because major=0 and minor=0),
618 * in such case we map character and block devices to the inaccessible node type socket. */
619 switch(mode & S_IFMT) {
621 return "/run/systemd/inaccessible/reg";
623 return "/run/systemd/inaccessible/dir";
625 if (access("/run/systemd/inaccessible/chr", F_OK) == 0)
626 return "/run/systemd/inaccessible/chr";
627 return "/run/systemd/inaccessible/sock";
629 if (access("/run/systemd/inaccessible/blk", F_OK) == 0)
630 return "/run/systemd/inaccessible/blk";
631 return "/run/systemd/inaccessible/sock";
633 return "/run/systemd/inaccessible/fifo";
635 return "/run/systemd/inaccessible/sock";
640 #if 0 /// UNNEEDED by elogind
641 #define FLAG(name) (flags & name ? STRINGIFY(name) "|" : "")
642 static char* mount_flags_to_string(long unsigned flags) {
644 _cleanup_free_ char *y = NULL;
645 long unsigned overflow;
647 overflow = flags & ~(MS_RDONLY |
672 if (flags == 0 || overflow != 0)
673 if (asprintf(&y, "%lx", overflow) < 0)
676 x = strjoin(FLAG(MS_RDONLY),
680 FLAG(MS_SYNCHRONOUS),
698 FLAG(MS_STRICTATIME),
704 x[strlen(x) - 1] = '\0'; /* truncate the last | */
714 const char *options) {
716 _cleanup_free_ char *fl = NULL;
718 fl = mount_flags_to_string(flags);
720 if ((flags & MS_REMOUNT) && !what && !type)
721 log_debug("Remounting %s (%s \"%s\")...",
722 where, strnull(fl), strempty(options));
723 else if (!what && !type)
724 log_debug("Mounting %s (%s \"%s\")...",
725 where, strnull(fl), strempty(options));
726 else if ((flags & MS_BIND) && !type)
727 log_debug("Bind-mounting %s on %s (%s \"%s\")...",
728 what, where, strnull(fl), strempty(options));
729 else if (flags & MS_MOVE)
730 log_debug("Moving mount %s → %s (%s \"%s\")...",
731 what, where, strnull(fl), strempty(options));
733 log_debug("Mounting %s on %s (%s \"%s\")...",
734 strna(type), where, strnull(fl), strempty(options));
735 if (mount(what, where, type, flags, options) < 0)
736 return log_full_errno(error_log_level, errno,
737 "Failed to mount %s on %s (%s \"%s\"): %m",
738 strna(type), where, strnull(fl), strempty(options));
742 int umount_verbose(const char *what) {
743 log_debug("Umounting %s...", what);
744 if (umount(what) < 0)
745 return log_error_errno(errno, "Failed to unmount %s: %m", what);
750 const char *mount_propagation_flags_to_string(unsigned long flags) {
752 switch (flags & (MS_SHARED|MS_SLAVE|MS_PRIVATE)) {
767 int mount_propagation_flags_from_string(const char *name, unsigned long *ret) {
771 else if (streq(name, "shared"))
773 else if (streq(name, "slave"))
775 else if (streq(name, "private"))