1 /* SPDX-License-Identifier: LGPL-2.1+ */
7 #include <sys/statvfs.h>
10 #include "alloc-util.h"
11 #include "extract-word.h"
16 #include "libmount-util.h"
17 #include "mount-util.h"
18 #include "mountpoint-util.h"
19 #include "parse-util.h"
20 #include "path-util.h"
22 #include "stdio-util.h"
23 #include "string-util.h"
26 int umount_recursive(const char *prefix
, int flags
) {
30 /* Try to umount everything recursively below a
31 * directory. Also, take care of stacked mounts, and keep
32 * unmounting them until they are gone. */
35 _cleanup_(mnt_free_tablep
) struct libmnt_table
*table
= NULL
;
36 _cleanup_(mnt_free_iterp
) struct libmnt_iter
*iter
= NULL
;
40 r
= libmount_parse("/proc/self/mountinfo", NULL
, &table
, &iter
);
42 return log_debug_errno(r
, "Failed to parse /proc/self/mountinfo: %m");
48 r
= mnt_table_next_fs(table
, iter
, &fs
);
52 return log_debug_errno(r
, "Failed to get next entry from /proc/self/mountinfo: %m");
54 path
= mnt_fs_get_target(fs
);
58 if (!path_startswith(path
, prefix
))
61 if (umount2(path
, flags
) < 0) {
62 r
= log_debug_errno(errno
, "Failed to umount %s: %m", path
);
66 log_debug("Successfully unmounted %s", path
);
79 /* Get the mount flags for the mountpoint at "path" from "table" */
80 static int get_mount_flags(const char *path
, unsigned long *flags
, struct libmnt_table
*table
) {
81 struct statvfs buf
= {};
82 struct libmnt_fs
*fs
= NULL
;
83 const char *opts
= NULL
;
86 fs
= mnt_table_find_target(table
, path
, MNT_ITER_FORWARD
);
88 log_warning("Could not find '%s' in mount table", path
);
92 opts
= mnt_fs_get_vfs_options(fs
);
93 r
= mnt_optstr_get_flags(opts
, flags
, mnt_get_builtin_optmap(MNT_LINUX_MAP
));
95 log_warning_errno(r
, "Could not get flags for '%s': %m", path
);
99 /* relatime is default and trying to set it in an unprivileged container causes EPERM */
100 *flags
&= ~MS_RELATIME
;
104 if (statvfs(path
, &buf
) < 0)
111 /* Use this function only if you do not have direct access to /proc/self/mountinfo but the caller can open it
112 * for you. This is the case when /proc is masked or not mounted. Otherwise, use bind_remount_recursive. */
113 int bind_remount_recursive_with_mountinfo(
115 unsigned long new_flags
,
116 unsigned long flags_mask
,
118 FILE *proc_self_mountinfo
) {
120 _cleanup_set_free_free_ Set
*done
= NULL
;
121 _cleanup_free_
char *cleaned
= NULL
;
124 assert(proc_self_mountinfo
);
126 /* Recursively remount a directory (and all its submounts) read-only or read-write. If the directory is already
127 * mounted, we reuse the mount and simply mark it MS_BIND|MS_RDONLY (or remove the MS_RDONLY for read-write
128 * operation). If it isn't we first make it one. Afterwards we apply MS_BIND|MS_RDONLY (or remove MS_RDONLY) to
129 * all submounts we can access, too. When mounts are stacked on the same mount point we only care for each
130 * individual "top-level" mount on each point, as we cannot influence/access the underlying mounts anyway. We
131 * do not have any effect on future submounts that might get propagated, they migt be writable. This includes
132 * future submounts that have been triggered via autofs.
134 * If the "blacklist" parameter is specified it may contain a list of subtrees to exclude from the
135 * remount operation. Note that we'll ignore the blacklist for the top-level path. */
137 cleaned
= strdup(prefix
);
141 path_simplify(cleaned
, false);
143 done
= set_new(&path_hash_ops
);
148 _cleanup_set_free_free_ Set
*todo
= NULL
;
149 _cleanup_(mnt_free_tablep
) struct libmnt_table
*table
= NULL
;
150 _cleanup_(mnt_free_iterp
) struct libmnt_iter
*iter
= NULL
;
151 bool top_autofs
= false;
153 unsigned long orig_flags
;
155 todo
= set_new(&path_hash_ops
);
159 rewind(proc_self_mountinfo
);
161 r
= libmount_parse("/proc/self/mountinfo", proc_self_mountinfo
, &table
, &iter
);
163 return log_debug_errno(r
, "Failed to parse /proc/self/mountinfo: %m");
166 struct libmnt_fs
*fs
;
167 const char *path
, *type
;
169 r
= mnt_table_next_fs(table
, iter
, &fs
);
173 return log_debug_errno(r
, "Failed to get next entry from /proc/self/mountinfo: %m");
175 path
= mnt_fs_get_target(fs
);
176 type
= mnt_fs_get_fstype(fs
);
180 if (!path_startswith(path
, cleaned
))
183 /* Ignore this mount if it is blacklisted, but only if it isn't the top-level mount
184 * we shall operate on. */
185 if (!path_equal(path
, cleaned
)) {
186 bool blacklisted
= false;
189 STRV_FOREACH(i
, blacklist
) {
190 if (path_equal(*i
, cleaned
))
193 if (!path_startswith(*i
, cleaned
))
196 if (path_startswith(path
, *i
)) {
198 log_debug("Not remounting %s blacklisted by %s, called for %s",
207 /* Let's ignore autofs mounts. If they aren't
208 * triggered yet, we want to avoid triggering
209 * them, as we don't make any guarantees for
210 * future submounts anyway. If they are
211 * already triggered, then we will find
212 * another entry for this. */
213 if (streq(type
, "autofs")) {
214 top_autofs
= top_autofs
|| path_equal(path
, cleaned
);
218 if (!set_contains(done
, path
)) {
219 r
= set_put_strdup(todo
, path
);
225 /* If we have no submounts to process anymore and if
226 * the root is either already done, or an autofs, we
228 if (set_isempty(todo
) &&
229 (top_autofs
|| set_contains(done
, cleaned
)))
232 if (!set_contains(done
, cleaned
) &&
233 !set_contains(todo
, cleaned
)) {
234 /* The prefix directory itself is not yet a mount, make it one. */
235 if (mount(cleaned
, cleaned
, NULL
, MS_BIND
|MS_REC
, NULL
) < 0)
239 (void) get_mount_flags(cleaned
, &orig_flags
, table
);
240 orig_flags
&= ~MS_RDONLY
;
242 if (mount(NULL
, cleaned
, NULL
, (orig_flags
& ~flags_mask
)|MS_BIND
|MS_REMOUNT
|new_flags
, NULL
) < 0)
245 log_debug("Made top-level directory %s a mount point.", prefix
);
247 r
= set_put_strdup(done
, cleaned
);
252 while ((x
= set_steal_first(todo
))) {
254 r
= set_consume(done
, x
);
255 if (IN_SET(r
, 0, -EEXIST
))
260 /* Deal with mount points that are obstructed by a later mount */
261 r
= path_is_mount_point(x
, NULL
, 0);
262 if (IN_SET(r
, 0, -ENOENT
))
264 if (IN_SET(r
, -EACCES
, -EPERM
)) {
265 /* Even if root user invoke this, submounts under private FUSE or NFS mount points
266 * may not be acceessed. E.g.,
268 * $ bindfs --no-allow-other ~/mnt/mnt ~/mnt/mnt
269 * $ bindfs --no-allow-other ~/mnt ~/mnt
271 * Then, root user cannot access the mount point ~/mnt/mnt.
272 * In such cases, the submounts are ignored, as we have no way to manage them. */
273 log_debug_errno(r
, "Failed to determine '%s' is mount point or not, ignoring: %m", x
);
279 /* Try to reuse the original flag set */
281 (void) get_mount_flags(x
, &orig_flags
, table
);
282 orig_flags
&= ~MS_RDONLY
;
284 if (mount(NULL
, x
, NULL
, (orig_flags
& ~flags_mask
)|MS_BIND
|MS_REMOUNT
|new_flags
, NULL
) < 0)
287 log_debug("Remounted %s read-only.", x
);
292 int bind_remount_recursive(const char *prefix
, unsigned long new_flags
, unsigned long flags_mask
, char **blacklist
) {
293 _cleanup_fclose_
FILE *proc_self_mountinfo
= NULL
;
296 r
= fopen_unlocked("/proc/self/mountinfo", "re", &proc_self_mountinfo
);
300 return bind_remount_recursive_with_mountinfo(prefix
, new_flags
, flags_mask
, blacklist
, proc_self_mountinfo
);
303 int mount_move_root(const char *path
) {
309 if (mount(path
, "/", NULL
, MS_MOVE
, NULL
) < 0)
321 int repeat_unmount(const char *path
, int flags
) {
326 /* If there are multiple mounts on a mount point, this
327 * removes them all */
330 if (umount2(path
, flags
) < 0) {
342 const char* mode_to_inaccessible_node(mode_t mode
) {
343 /* This function maps a node type to a corresponding inaccessible file node. These nodes are created during
344 * early boot by PID 1. In some cases we lacked the privs to create the character and block devices (maybe
345 * because we run in an userns environment, or miss CAP_SYS_MKNOD, or run with a devices policy that excludes
346 * device nodes with major and minor of 0), but that's fine, in that case we use an AF_UNIX file node instead,
347 * which is not the same, but close enough for most uses. And most importantly, the kernel allows bind mounts
348 * from socket nodes to any non-directory file nodes, and that's the most important thing that matters. */
350 switch(mode
& S_IFMT
) {
352 return "/run/systemd/inaccessible/reg";
355 return "/run/systemd/inaccessible/dir";
358 if (access("/run/systemd/inaccessible/chr", F_OK
) == 0)
359 return "/run/systemd/inaccessible/chr";
360 return "/run/systemd/inaccessible/sock";
363 if (access("/run/systemd/inaccessible/blk", F_OK
) == 0)
364 return "/run/systemd/inaccessible/blk";
365 return "/run/systemd/inaccessible/sock";
368 return "/run/systemd/inaccessible/fifo";
371 return "/run/systemd/inaccessible/sock";
376 #define FLAG(name) (flags & name ? STRINGIFY(name) "|" : "")
377 static char* mount_flags_to_string(long unsigned flags
) {
379 _cleanup_free_
char *y
= NULL
;
380 long unsigned overflow
;
382 overflow
= flags
& ~(MS_RDONLY
|
407 if (flags
== 0 || overflow
!= 0)
408 if (asprintf(&y
, "%lx", overflow
) < 0)
411 x
= strjoin(FLAG(MS_RDONLY
),
415 FLAG(MS_SYNCHRONOUS
),
433 FLAG(MS_STRICTATIME
),
439 x
[strlen(x
) - 1] = '\0'; /* truncate the last | */
449 const char *options
) {
451 _cleanup_free_
char *fl
= NULL
, *o
= NULL
;
455 r
= mount_option_mangle(options
, flags
, &f
, &o
);
457 return log_full_errno(error_log_level
, r
,
458 "Failed to mangle mount options %s: %m",
461 fl
= mount_flags_to_string(f
);
463 if ((f
& MS_REMOUNT
) && !what
&& !type
)
464 log_debug("Remounting %s (%s \"%s\")...",
465 where
, strnull(fl
), strempty(o
));
466 else if (!what
&& !type
)
467 log_debug("Mounting %s (%s \"%s\")...",
468 where
, strnull(fl
), strempty(o
));
469 else if ((f
& MS_BIND
) && !type
)
470 log_debug("Bind-mounting %s on %s (%s \"%s\")...",
471 what
, where
, strnull(fl
), strempty(o
));
472 else if (f
& MS_MOVE
)
473 log_debug("Moving mount %s → %s (%s \"%s\")...",
474 what
, where
, strnull(fl
), strempty(o
));
476 log_debug("Mounting %s on %s (%s \"%s\")...",
477 strna(type
), where
, strnull(fl
), strempty(o
));
478 if (mount(what
, where
, type
, f
, o
) < 0)
479 return log_full_errno(error_log_level
, errno
,
480 "Failed to mount %s (type %s) on %s (%s \"%s\"): %m",
481 strna(what
), strna(type
), where
, strnull(fl
), strempty(o
));
485 int umount_verbose(const char *what
) {
486 log_debug("Umounting %s...", what
);
487 if (umount(what
) < 0)
488 return log_error_errno(errno
, "Failed to unmount %s: %m", what
);
492 int mount_option_mangle(
494 unsigned long mount_flags
,
495 unsigned long *ret_mount_flags
,
496 char **ret_remaining_options
) {
498 const struct libmnt_optmap
*map
;
499 _cleanup_free_
char *ret
= NULL
;
503 /* This extracts mount flags from the mount options, and store
504 * non-mount-flag options to '*ret_remaining_options'.
506 * "rw,nosuid,nodev,relatime,size=1630748k,mode=700,uid=1000,gid=1000"
507 * is split to MS_NOSUID|MS_NODEV|MS_RELATIME and
508 * "size=1630748k,mode=700,uid=1000,gid=1000".
509 * See more examples in test-mount-utils.c.
511 * Note that if 'options' does not contain any non-mount-flag options,
512 * then '*ret_remaining_options' is set to NULL instead of empty string.
513 * Note that this does not check validity of options stored in
514 * '*ret_remaining_options'.
515 * Note that if 'options' is NULL, then this just copies 'mount_flags'
516 * to '*ret_mount_flags'. */
518 assert(ret_mount_flags
);
519 assert(ret_remaining_options
);
521 map
= mnt_get_builtin_optmap(MNT_LINUX_MAP
);
527 _cleanup_free_
char *word
= NULL
;
528 const struct libmnt_optmap
*ent
;
530 r
= extract_first_word(&p
, &word
, ",", EXTRACT_UNQUOTE
);
536 for (ent
= map
; ent
->name
; ent
++) {
537 /* All entries in MNT_LINUX_MAP do not take any argument.
538 * Thus, ent->name does not contain "=" or "[=]". */
539 if (!streq(word
, ent
->name
))
542 if (!(ent
->mask
& MNT_INVERT
))
543 mount_flags
|= ent
->id
;
544 else if (mount_flags
& ent
->id
)
545 mount_flags
^= ent
->id
;
550 /* If 'word' is not a mount flag, then store it in '*ret_remaining_options'. */
551 if (!ent
->name
&& !strextend_with_separator(&ret
, ",", word
, NULL
))
555 *ret_mount_flags
= mount_flags
;
556 *ret_remaining_options
= TAKE_PTR(ret
);