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) {
306 static int get_mount_flags(const char *path, unsigned long *flags) {
309 if (statvfs(path, &buf) < 0)
315 int bind_remount_recursive(const char *prefix, bool ro) {
316 _cleanup_set_free_free_ Set *done = NULL;
317 _cleanup_free_ char *cleaned = NULL;
320 /* Recursively remount a directory (and all its submounts)
321 * read-only or read-write. If the directory is already
322 * mounted, we reuse the mount and simply mark it
323 * 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
325 * apply MS_BIND|MS_RDONLY (or remove MS_RDONLY) to all
326 * submounts we can access, too. When mounts are stacked on
327 * the same mount point we only care for each individual
328 * "top-level" mount on each point, as we cannot
329 * influence/access the underlying mounts anyway. We do not
330 * have any effect on future submounts that might get
331 * propagated, they migt be writable. This includes future
332 * submounts that have been triggered via autofs. */
334 cleaned = strdup(prefix);
338 path_kill_slashes(cleaned);
340 done = set_new(&string_hash_ops);
345 _cleanup_fclose_ FILE *proc_self_mountinfo = NULL;
346 _cleanup_set_free_free_ Set *todo = NULL;
347 bool top_autofs = false;
349 unsigned long orig_flags;
351 todo = set_new(&string_hash_ops);
355 proc_self_mountinfo = fopen("/proc/self/mountinfo", "re");
356 if (!proc_self_mountinfo)
360 _cleanup_free_ char *path = NULL, *p = NULL, *type = NULL;
363 k = fscanf(proc_self_mountinfo,
364 "%*s " /* (1) mount id */
365 "%*s " /* (2) parent id */
366 "%*s " /* (3) major:minor */
367 "%*s " /* (4) root */
368 "%ms " /* (5) mount point */
369 "%*s" /* (6) mount options (superblock) */
370 "%*[^-]" /* (7) optional fields */
371 "- " /* (8) separator */
372 "%ms " /* (9) file system type */
373 "%*s" /* (10) mount source */
374 "%*s" /* (11) mount options (bind mount) */
375 "%*[^\n]", /* some rubbish at the end */
385 r = cunescape(path, UNESCAPE_RELAX, &p);
389 /* Let's ignore autofs mounts. If they aren't
390 * triggered yet, we want to avoid triggering
391 * them, as we don't make any guarantees for
392 * future submounts anyway. If they are
393 * already triggered, then we will find
394 * another entry for this. */
395 if (streq(type, "autofs")) {
396 top_autofs = top_autofs || path_equal(cleaned, p);
400 if (path_startswith(p, cleaned) &&
401 !set_contains(done, p)) {
403 r = set_consume(todo, p);
413 /* If we have no submounts to process anymore and if
414 * the root is either already done, or an autofs, we
416 if (set_isempty(todo) &&
417 (top_autofs || set_contains(done, cleaned)))
420 if (!set_contains(done, cleaned) &&
421 !set_contains(todo, cleaned)) {
422 /* The prefix directory itself is not yet a
423 * mount, make it one. */
424 if (mount(cleaned, cleaned, NULL, MS_BIND|MS_REC, NULL) < 0)
428 (void) get_mount_flags(cleaned, &orig_flags);
429 orig_flags &= ~MS_RDONLY;
431 if (mount(NULL, prefix, NULL, orig_flags|MS_BIND|MS_REMOUNT|(ro ? MS_RDONLY : 0), NULL) < 0)
438 r = set_consume(done, x);
443 while ((x = set_steal_first(todo))) {
445 r = set_consume(done, x);
446 if (r == -EEXIST || r == 0)
451 /* Deal with mount points that are obstructed by a
453 r = path_is_mount_point(x, 0);
454 if (r == -ENOENT || r == 0)
459 /* Try to reuse the original flag set */
461 (void) get_mount_flags(x, &orig_flags);
462 orig_flags &= ~MS_RDONLY;
464 if (mount(NULL, x, NULL, orig_flags|MS_BIND|MS_REMOUNT|(ro ? MS_RDONLY : 0), NULL) < 0)
471 int mount_move_root(const char *path) {
477 if (mount(path, "/", NULL, MS_MOVE, NULL) < 0)
489 bool fstype_is_network(const char *fstype) {
490 static const char table[] =
502 "pvfs2\0" /* OrangeFS */
508 x = startswith(fstype, "fuse.");
512 return nulstr_contains(table, fstype);
515 int repeat_unmount(const char *path, int flags) {
520 /* If there are multiple mounts on a mount point, this
521 * removes them all */
524 if (umount2(path, flags) < 0) {
537 const char* mode_to_inaccessible_node(mode_t mode) {
538 /* This function maps a node type to the correspondent inaccessible node type.
539 * Character and block inaccessible devices may not be created (because major=0 and minor=0),
540 * in such case we map character and block devices to the inaccessible node type socket. */
541 switch(mode & S_IFMT) {
543 return "/run/systemd/inaccessible/reg";
545 return "/run/systemd/inaccessible/dir";
547 if (access("/run/systemd/inaccessible/chr", F_OK) == 0)
548 return "/run/systemd/inaccessible/chr";
549 return "/run/systemd/inaccessible/sock";
551 if (access("/run/systemd/inaccessible/blk", F_OK) == 0)
552 return "/run/systemd/inaccessible/blk";
553 return "/run/systemd/inaccessible/sock";
555 return "/run/systemd/inaccessible/fifo";
557 return "/run/systemd/inaccessible/sock";