1 /* SPDX-License-Identifier: LGPL-2.1+ */
8 #include <sys/statvfs.h>
11 #include "alloc-util.h"
12 #include "extract-word.h"
17 #include "libmount-util.h"
18 #include "mount-util.h"
19 #include "mountpoint-util.h"
20 #include "parse-util.h"
21 #include "path-util.h"
23 #include "stdio-util.h"
24 #include "string-util.h"
27 int umount_recursive(const char *prefix
, int flags
) {
31 /* Try to umount everything recursively below a
32 * directory. Also, take care of stacked mounts, and keep
33 * unmounting them until they are gone. */
36 _cleanup_(mnt_free_tablep
) struct libmnt_table
*table
= NULL
;
37 _cleanup_(mnt_free_iterp
) struct libmnt_iter
*iter
= NULL
;
41 r
= libmount_parse("/proc/self/mountinfo", NULL
, &table
, &iter
);
43 return log_debug_errno(r
, "Failed to parse /proc/self/mountinfo: %m");
49 r
= mnt_table_next_fs(table
, iter
, &fs
);
53 return log_debug_errno(r
, "Failed to get next entry from /proc/self/mountinfo: %m");
55 path
= mnt_fs_get_target(fs
);
59 if (!path_startswith(path
, prefix
))
62 if (umount2(path
, flags
) < 0) {
63 r
= log_debug_errno(errno
, "Failed to umount %s: %m", path
);
67 log_debug("Successfully unmounted %s", path
);
80 /* Get the mount flags for the mountpoint at "path" from "table" */
81 static int get_mount_flags(const char *path
, unsigned long *flags
, struct libmnt_table
*table
) {
82 struct statvfs buf
= {};
83 struct libmnt_fs
*fs
= NULL
;
84 const char *opts
= NULL
;
87 fs
= mnt_table_find_target(table
, path
, MNT_ITER_FORWARD
);
89 log_warning("Could not find '%s' in mount table", path
);
93 opts
= mnt_fs_get_vfs_options(fs
);
94 r
= mnt_optstr_get_flags(opts
, flags
, mnt_get_builtin_optmap(MNT_LINUX_MAP
));
96 log_warning_errno(r
, "Could not get flags for '%s': %m", path
);
100 /* relatime is default and trying to set it in an unprivileged container causes EPERM */
101 *flags
&= ~MS_RELATIME
;
105 if (statvfs(path
, &buf
) < 0)
112 /* Use this function only if you do not have direct access to /proc/self/mountinfo but the caller can open it
113 * for you. This is the case when /proc is masked or not mounted. Otherwise, use bind_remount_recursive. */
114 int bind_remount_recursive_with_mountinfo(
116 unsigned long new_flags
,
117 unsigned long flags_mask
,
119 FILE *proc_self_mountinfo
) {
121 _cleanup_set_free_free_ Set
*done
= NULL
;
122 _cleanup_free_
char *cleaned
= NULL
;
125 assert(proc_self_mountinfo
);
127 /* Recursively remount a directory (and all its submounts) read-only or read-write. If the directory is already
128 * mounted, we reuse the mount and simply mark it MS_BIND|MS_RDONLY (or remove the MS_RDONLY for read-write
129 * operation). If it isn't we first make it one. Afterwards we apply MS_BIND|MS_RDONLY (or remove MS_RDONLY) to
130 * all submounts we can access, too. When mounts are stacked on the same mount point we only care for each
131 * individual "top-level" mount on each point, as we cannot influence/access the underlying mounts anyway. We
132 * do not have any effect on future submounts that might get propagated, they migt be writable. This includes
133 * future submounts that have been triggered via autofs.
135 * If the "blacklist" parameter is specified it may contain a list of subtrees to exclude from the
136 * remount operation. Note that we'll ignore the blacklist for the top-level path. */
138 cleaned
= strdup(prefix
);
142 path_simplify(cleaned
, false);
144 done
= set_new(&path_hash_ops
);
149 _cleanup_set_free_free_ Set
*todo
= NULL
;
150 _cleanup_(mnt_free_tablep
) struct libmnt_table
*table
= NULL
;
151 _cleanup_(mnt_free_iterp
) struct libmnt_iter
*iter
= NULL
;
152 bool top_autofs
= false;
154 unsigned long orig_flags
;
156 todo
= set_new(&path_hash_ops
);
160 rewind(proc_self_mountinfo
);
162 r
= libmount_parse("/proc/self/mountinfo", proc_self_mountinfo
, &table
, &iter
);
164 return log_debug_errno(r
, "Failed to parse /proc/self/mountinfo: %m");
167 struct libmnt_fs
*fs
;
168 const char *path
, *type
;
170 r
= mnt_table_next_fs(table
, iter
, &fs
);
174 return log_debug_errno(r
, "Failed to get next entry from /proc/self/mountinfo: %m");
176 path
= mnt_fs_get_target(fs
);
177 type
= mnt_fs_get_fstype(fs
);
181 if (!path_startswith(path
, cleaned
))
184 /* Ignore this mount if it is blacklisted, but only if it isn't the top-level mount
185 * we shall operate on. */
186 if (!path_equal(path
, cleaned
)) {
187 bool blacklisted
= false;
190 STRV_FOREACH(i
, blacklist
) {
191 if (path_equal(*i
, cleaned
))
194 if (!path_startswith(*i
, cleaned
))
197 if (path_startswith(path
, *i
)) {
199 log_debug("Not remounting %s blacklisted by %s, called for %s",
208 /* Let's ignore autofs mounts. If they aren't
209 * triggered yet, we want to avoid triggering
210 * them, as we don't make any guarantees for
211 * future submounts anyway. If they are
212 * already triggered, then we will find
213 * another entry for this. */
214 if (streq(type
, "autofs")) {
215 top_autofs
= top_autofs
|| path_equal(path
, cleaned
);
219 if (!set_contains(done
, path
)) {
220 r
= set_put_strdup(todo
, path
);
226 /* If we have no submounts to process anymore and if
227 * the root is either already done, or an autofs, we
229 if (set_isempty(todo
) &&
230 (top_autofs
|| set_contains(done
, cleaned
)))
233 if (!set_contains(done
, cleaned
) &&
234 !set_contains(todo
, cleaned
)) {
235 /* The prefix directory itself is not yet a mount, make it one. */
236 if (mount(cleaned
, cleaned
, NULL
, MS_BIND
|MS_REC
, NULL
) < 0)
240 (void) get_mount_flags(cleaned
, &orig_flags
, table
);
241 orig_flags
&= ~MS_RDONLY
;
243 if (mount(NULL
, cleaned
, NULL
, (orig_flags
& ~flags_mask
)|MS_BIND
|MS_REMOUNT
|new_flags
, NULL
) < 0)
246 log_debug("Made top-level directory %s a mount point.", prefix
);
248 r
= set_put_strdup(done
, cleaned
);
253 while ((x
= set_steal_first(todo
))) {
255 r
= set_consume(done
, x
);
256 if (IN_SET(r
, 0, -EEXIST
))
261 /* Deal with mount points that are obstructed by a later mount */
262 r
= path_is_mount_point(x
, NULL
, 0);
263 if (IN_SET(r
, 0, -ENOENT
))
265 if (IN_SET(r
, -EACCES
, -EPERM
)) {
266 /* Even if root user invoke this, submounts under private FUSE or NFS mount points
267 * may not be acceessed. E.g.,
269 * $ bindfs --no-allow-other ~/mnt/mnt ~/mnt/mnt
270 * $ bindfs --no-allow-other ~/mnt ~/mnt
272 * Then, root user cannot access the mount point ~/mnt/mnt.
273 * In such cases, the submounts are ignored, as we have no way to manage them. */
274 log_debug_errno(r
, "Failed to determine '%s' is mount point or not, ignoring: %m", x
);
280 /* Try to reuse the original flag set */
282 (void) get_mount_flags(x
, &orig_flags
, table
);
283 orig_flags
&= ~MS_RDONLY
;
285 if (mount(NULL
, x
, NULL
, (orig_flags
& ~flags_mask
)|MS_BIND
|MS_REMOUNT
|new_flags
, NULL
) < 0)
288 log_debug("Remounted %s read-only.", x
);
293 int bind_remount_recursive(const char *prefix
, unsigned long new_flags
, unsigned long flags_mask
, char **blacklist
) {
294 _cleanup_fclose_
FILE *proc_self_mountinfo
= NULL
;
297 r
= fopen_unlocked("/proc/self/mountinfo", "re", &proc_self_mountinfo
);
301 return bind_remount_recursive_with_mountinfo(prefix
, new_flags
, flags_mask
, blacklist
, proc_self_mountinfo
);
304 int mount_move_root(const char *path
) {
310 if (mount(path
, "/", NULL
, MS_MOVE
, NULL
) < 0)
322 int repeat_unmount(const char *path
, int flags
) {
327 /* If there are multiple mounts on a mount point, this
328 * removes them all */
331 if (umount2(path
, flags
) < 0) {
343 const char* mode_to_inaccessible_node(mode_t mode
) {
344 /* This function maps a node type to a corresponding inaccessible file node. These nodes are created during
345 * early boot by PID 1. In some cases we lacked the privs to create the character and block devices (maybe
346 * because we run in an userns environment, or miss CAP_SYS_MKNOD, or run with a devices policy that excludes
347 * device nodes with major and minor of 0), but that's fine, in that case we use an AF_UNIX file node instead,
348 * which is not the same, but close enough for most uses. And most importantly, the kernel allows bind mounts
349 * from socket nodes to any non-directory file nodes, and that's the most important thing that matters. */
351 switch(mode
& S_IFMT
) {
353 return "/run/systemd/inaccessible/reg";
356 return "/run/systemd/inaccessible/dir";
359 if (access("/run/systemd/inaccessible/chr", F_OK
) == 0)
360 return "/run/systemd/inaccessible/chr";
361 return "/run/systemd/inaccessible/sock";
364 if (access("/run/systemd/inaccessible/blk", F_OK
) == 0)
365 return "/run/systemd/inaccessible/blk";
366 return "/run/systemd/inaccessible/sock";
369 return "/run/systemd/inaccessible/fifo";
372 return "/run/systemd/inaccessible/sock";
377 #define FLAG(name) (flags & name ? STRINGIFY(name) "|" : "")
378 static char* mount_flags_to_string(long unsigned flags
) {
380 _cleanup_free_
char *y
= NULL
;
381 long unsigned overflow
;
383 overflow
= flags
& ~(MS_RDONLY
|
408 if (flags
== 0 || overflow
!= 0)
409 if (asprintf(&y
, "%lx", overflow
) < 0)
412 x
= strjoin(FLAG(MS_RDONLY
),
416 FLAG(MS_SYNCHRONOUS
),
434 FLAG(MS_STRICTATIME
),
440 x
[strlen(x
) - 1] = '\0'; /* truncate the last | */
450 const char *options
) {
452 _cleanup_free_
char *fl
= NULL
, *o
= NULL
;
456 r
= mount_option_mangle(options
, flags
, &f
, &o
);
458 return log_full_errno(error_log_level
, r
,
459 "Failed to mangle mount options %s: %m",
462 fl
= mount_flags_to_string(f
);
464 if ((f
& MS_REMOUNT
) && !what
&& !type
)
465 log_debug("Remounting %s (%s \"%s\")...",
466 where
, strnull(fl
), strempty(o
));
467 else if (!what
&& !type
)
468 log_debug("Mounting %s (%s \"%s\")...",
469 where
, strnull(fl
), strempty(o
));
470 else if ((f
& MS_BIND
) && !type
)
471 log_debug("Bind-mounting %s on %s (%s \"%s\")...",
472 what
, where
, strnull(fl
), strempty(o
));
473 else if (f
& MS_MOVE
)
474 log_debug("Moving mount %s → %s (%s \"%s\")...",
475 what
, where
, strnull(fl
), strempty(o
));
477 log_debug("Mounting %s on %s (%s \"%s\")...",
478 strna(type
), where
, strnull(fl
), strempty(o
));
479 if (mount(what
, where
, type
, f
, o
) < 0)
480 return log_full_errno(error_log_level
, errno
,
481 "Failed to mount %s (type %s) on %s (%s \"%s\"): %m",
482 strna(what
), strna(type
), where
, strnull(fl
), strempty(o
));
486 int umount_verbose(const char *what
) {
487 log_debug("Umounting %s...", what
);
488 if (umount(what
) < 0)
489 return log_error_errno(errno
, "Failed to unmount %s: %m", what
);
493 int mount_option_mangle(
495 unsigned long mount_flags
,
496 unsigned long *ret_mount_flags
,
497 char **ret_remaining_options
) {
499 const struct libmnt_optmap
*map
;
500 _cleanup_free_
char *ret
= NULL
;
504 /* This extracts mount flags from the mount options, and store
505 * non-mount-flag options to '*ret_remaining_options'.
507 * "rw,nosuid,nodev,relatime,size=1630748k,mode=700,uid=1000,gid=1000"
508 * is split to MS_NOSUID|MS_NODEV|MS_RELATIME and
509 * "size=1630748k,mode=700,uid=1000,gid=1000".
510 * See more examples in test-mount-utils.c.
512 * Note that if 'options' does not contain any non-mount-flag options,
513 * then '*ret_remaining_options' is set to NULL instead of empty string.
514 * Note that this does not check validity of options stored in
515 * '*ret_remaining_options'.
516 * Note that if 'options' is NULL, then this just copies 'mount_flags'
517 * to '*ret_mount_flags'. */
519 assert(ret_mount_flags
);
520 assert(ret_remaining_options
);
522 map
= mnt_get_builtin_optmap(MNT_LINUX_MAP
);
528 _cleanup_free_
char *word
= NULL
;
529 const struct libmnt_optmap
*ent
;
531 r
= extract_first_word(&p
, &word
, ",", EXTRACT_UNQUOTE
);
537 for (ent
= map
; ent
->name
; ent
++) {
538 /* All entries in MNT_LINUX_MAP do not take any argument.
539 * Thus, ent->name does not contain "=" or "[=]". */
540 if (!streq(word
, ent
->name
))
543 if (!(ent
->mask
& MNT_INVERT
))
544 mount_flags
|= ent
->id
;
545 else if (mount_flags
& ent
->id
)
546 mount_flags
^= ent
->id
;
551 /* If 'word' is not a mount flag, then store it in '*ret_remaining_options'. */
552 if (!ent
->name
&& !strextend_with_separator(&ret
, ",", word
, NULL
))
556 *ret_mount_flags
= mount_flags
;
557 *ret_remaining_options
= TAKE_PTR(ret
);