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
| UMOUNT_NOFOLLOW
) < 0) {
62 log_debug_errno(errno
, "Failed to umount %s, ignoring: %m", path
);
66 log_debug("Successfully unmounted %s", path
);
78 static int get_mount_flags(
79 struct libmnt_table
*table
,
87 /* Get the mount flags for the mountpoint at "path" from "table". We have a fallback using statvfs()
88 * in place (which provides us with mostly the same info), but it's just a fallback, since using it
89 * means triggering autofs or NFS mounts, which we'd rather avoid needlessly. */
91 fs
= mnt_table_find_target(table
, path
, MNT_ITER_FORWARD
);
93 log_debug("Could not find '%s' in mount table, ignoring.", path
);
97 opts
= mnt_fs_get_vfs_options(fs
);
103 r
= mnt_optstr_get_flags(opts
, ret
, mnt_get_builtin_optmap(MNT_LINUX_MAP
));
105 log_debug_errno(r
, "Could not get flags for '%s', ignoring: %m", path
);
109 /* MS_RELATIME is default and trying to set it in an unprivileged container causes EPERM */
110 *ret
&= ~MS_RELATIME
;
114 if (statvfs(path
, &buf
) < 0)
117 /* The statvfs() flags and the mount flags mostly have the same values, but for some cases do
118 * not. Hence map the flags manually. (Strictly speaking, ST_RELATIME/MS_RELATIME is the most
119 * prominent one that doesn't match, but that's the one we mask away anyway, see above.) */
122 FLAGS_SET(buf
.f_flag
, ST_RDONLY
) * MS_RDONLY
|
123 FLAGS_SET(buf
.f_flag
, ST_NODEV
) * MS_NODEV
|
124 FLAGS_SET(buf
.f_flag
, ST_NOEXEC
) * MS_NOEXEC
|
125 FLAGS_SET(buf
.f_flag
, ST_NOSUID
) * MS_NOSUID
|
126 FLAGS_SET(buf
.f_flag
, ST_NOATIME
) * MS_NOATIME
|
127 FLAGS_SET(buf
.f_flag
, ST_NODIRATIME
) * MS_NODIRATIME
;
132 /* Use this function only if you do not have direct access to /proc/self/mountinfo but the caller can open it
133 * for you. This is the case when /proc is masked or not mounted. Otherwise, use bind_remount_recursive. */
134 int bind_remount_recursive_with_mountinfo(
136 unsigned long new_flags
,
137 unsigned long flags_mask
,
139 FILE *proc_self_mountinfo
) {
141 _cleanup_set_free_free_ Set
*done
= NULL
;
142 _cleanup_free_
char *simplified
= NULL
;
146 assert(proc_self_mountinfo
);
148 /* Recursively remount a directory (and all its submounts) read-only or read-write. If the directory is already
149 * mounted, we reuse the mount and simply mark it MS_BIND|MS_RDONLY (or remove the MS_RDONLY for read-write
150 * operation). If it isn't we first make it one. Afterwards we apply MS_BIND|MS_RDONLY (or remove MS_RDONLY) to
151 * all submounts we can access, too. When mounts are stacked on the same mount point we only care for each
152 * individual "top-level" mount on each point, as we cannot influence/access the underlying mounts anyway. We
153 * do not have any effect on future submounts that might get propagated, they might be writable. This includes
154 * future submounts that have been triggered via autofs.
156 * If the "deny_list" parameter is specified it may contain a list of subtrees to exclude from the
157 * remount operation. Note that we'll ignore the deny list for the top-level path. */
159 simplified
= strdup(prefix
);
163 path_simplify(simplified
, false);
165 done
= set_new(&path_hash_ops
);
170 _cleanup_set_free_free_ Set
*todo
= NULL
;
171 _cleanup_(mnt_free_tablep
) struct libmnt_table
*table
= NULL
;
172 _cleanup_(mnt_free_iterp
) struct libmnt_iter
*iter
= NULL
;
173 bool top_autofs
= false;
175 unsigned long orig_flags
;
177 todo
= set_new(&path_hash_ops
);
181 rewind(proc_self_mountinfo
);
183 r
= libmount_parse("/proc/self/mountinfo", proc_self_mountinfo
, &table
, &iter
);
185 return log_debug_errno(r
, "Failed to parse /proc/self/mountinfo: %m");
188 struct libmnt_fs
*fs
;
189 const char *path
, *type
;
191 r
= mnt_table_next_fs(table
, iter
, &fs
);
195 return log_debug_errno(r
, "Failed to get next entry from /proc/self/mountinfo: %m");
197 path
= mnt_fs_get_target(fs
);
198 type
= mnt_fs_get_fstype(fs
);
202 if (!path_startswith(path
, simplified
))
205 /* Ignore this mount if it is deny-listed, but only if it isn't the top-level mount
206 * we shall operate on. */
207 if (!path_equal(path
, simplified
)) {
208 bool deny_listed
= false;
211 STRV_FOREACH(i
, deny_list
) {
212 if (path_equal(*i
, simplified
))
215 if (!path_startswith(*i
, simplified
))
218 if (path_startswith(path
, *i
)) {
220 log_debug("Not remounting %s deny-listed by %s, called for %s",
221 path
, *i
, simplified
);
229 /* Let's ignore autofs mounts. If they aren't
230 * triggered yet, we want to avoid triggering
231 * them, as we don't make any guarantees for
232 * future submounts anyway. If they are
233 * already triggered, then we will find
234 * another entry for this. */
235 if (streq(type
, "autofs")) {
236 top_autofs
= top_autofs
|| path_equal(path
, simplified
);
240 if (!set_contains(done
, path
)) {
241 r
= set_put_strdup(&todo
, path
);
247 /* If we have no submounts to process anymore and if
248 * the root is either already done, or an autofs, we
250 if (set_isempty(todo
) &&
251 (top_autofs
|| set_contains(done
, simplified
)))
254 if (!set_contains(done
, simplified
) &&
255 !set_contains(todo
, simplified
)) {
256 /* The prefix directory itself is not yet a mount, make it one. */
257 if (mount(simplified
, simplified
, NULL
, MS_BIND
|MS_REC
, NULL
) < 0)
261 (void) get_mount_flags(table
, simplified
, &orig_flags
);
263 if (mount(NULL
, simplified
, NULL
, (orig_flags
& ~flags_mask
)|MS_BIND
|MS_REMOUNT
|new_flags
, NULL
) < 0)
266 log_debug("Made top-level directory %s a mount point.", prefix
);
268 r
= set_put_strdup(&done
, simplified
);
273 while ((x
= set_steal_first(todo
))) {
275 r
= set_consume(done
, x
);
276 if (IN_SET(r
, 0, -EEXIST
))
281 /* Deal with mount points that are obstructed by a later mount */
282 r
= path_is_mount_point(x
, NULL
, 0);
283 if (IN_SET(r
, 0, -ENOENT
))
285 if (IN_SET(r
, -EACCES
, -EPERM
)) {
286 /* Even if root user invoke this, submounts under private FUSE or NFS mount points
287 * may not be acceessed. E.g.,
289 * $ bindfs --no-allow-other ~/mnt/mnt ~/mnt/mnt
290 * $ bindfs --no-allow-other ~/mnt ~/mnt
292 * Then, root user cannot access the mount point ~/mnt/mnt.
293 * In such cases, the submounts are ignored, as we have no way to manage them. */
294 log_debug_errno(r
, "Failed to determine '%s' is mount point or not, ignoring: %m", x
);
300 /* Try to reuse the original flag set */
302 (void) get_mount_flags(table
, x
, &orig_flags
);
304 if (mount(NULL
, x
, NULL
, (orig_flags
& ~flags_mask
)|MS_BIND
|MS_REMOUNT
|new_flags
, NULL
) < 0)
307 log_debug("Remounted %s read-only.", x
);
312 int bind_remount_recursive(
314 unsigned long new_flags
,
315 unsigned long flags_mask
,
318 _cleanup_fclose_
FILE *proc_self_mountinfo
= NULL
;
321 r
= fopen_unlocked("/proc/self/mountinfo", "re", &proc_self_mountinfo
);
325 return bind_remount_recursive_with_mountinfo(prefix
, new_flags
, flags_mask
, deny_list
, proc_self_mountinfo
);
328 int bind_remount_one_with_mountinfo(
330 unsigned long new_flags
,
331 unsigned long flags_mask
,
332 FILE *proc_self_mountinfo
) {
334 _cleanup_(mnt_free_tablep
) struct libmnt_table
*table
= NULL
;
335 unsigned long orig_flags
= 0;
339 assert(proc_self_mountinfo
);
341 rewind(proc_self_mountinfo
);
343 table
= mnt_new_table();
347 r
= mnt_table_parse_stream(table
, proc_self_mountinfo
, "/proc/self/mountinfo");
351 /* Try to reuse the original flag set */
352 (void) get_mount_flags(table
, path
, &orig_flags
);
354 if (mount(NULL
, path
, NULL
, (orig_flags
& ~flags_mask
)|MS_BIND
|MS_REMOUNT
|new_flags
, NULL
) < 0)
360 int mount_move_root(const char *path
) {
366 if (mount(path
, "/", NULL
, MS_MOVE
, NULL
) < 0)
378 int repeat_unmount(const char *path
, int flags
) {
383 /* If there are multiple mounts on a mount point, this
384 * removes them all */
387 if (umount2(path
, flags
) < 0) {
399 int mode_to_inaccessible_node(
400 const char *runtime_dir
,
404 /* This function maps a node type to a corresponding inaccessible file node. These nodes are created
405 * during early boot by PID 1. In some cases we lacked the privs to create the character and block
406 * devices (maybe because we run in an userns environment, or miss CAP_SYS_MKNOD, or run with a
407 * devices policy that excludes device nodes with major and minor of 0), but that's fine, in that
408 * case we use an AF_UNIX file node instead, which is not the same, but close enough for most
409 * uses. And most importantly, the kernel allows bind mounts from socket nodes to any non-directory
410 * file nodes, and that's the most important thing that matters.
412 * Note that the runtime directory argument shall be the top-level runtime directory, i.e. /run/ if
413 * we operate in system context and $XDG_RUNTIME_DIR if we operate in user context. */
415 _cleanup_free_
char *d
= NULL
;
416 const char *node
= NULL
;
417 bool fallback
= false;
422 runtime_dir
= "/run";
424 switch(mode
& S_IFMT
) {
426 node
= "/systemd/inaccessible/reg";
430 node
= "/systemd/inaccessible/dir";
434 node
= "/systemd/inaccessible/chr";
439 node
= "/systemd/inaccessible/blk";
444 node
= "/systemd/inaccessible/fifo";
448 node
= "/systemd/inaccessible/sock";
454 d
= path_join(runtime_dir
, node
);
458 if (fallback
&& access(d
, F_OK
) < 0) {
460 d
= path_join(runtime_dir
, "/systemd/inaccessible/sock");
469 #define FLAG(name) (flags & name ? STRINGIFY(name) "|" : "")
470 static char* mount_flags_to_string(long unsigned flags
) {
472 _cleanup_free_
char *y
= NULL
;
473 long unsigned overflow
;
475 overflow
= flags
& ~(MS_RDONLY
|
500 if (flags
== 0 || overflow
!= 0)
501 if (asprintf(&y
, "%lx", overflow
) < 0)
504 x
= strjoin(FLAG(MS_RDONLY
),
508 FLAG(MS_SYNCHRONOUS
),
526 FLAG(MS_STRICTATIME
),
532 x
[strlen(x
) - 1] = '\0'; /* truncate the last | */
542 const char *options
) {
544 _cleanup_free_
char *fl
= NULL
, *o
= NULL
;
548 r
= mount_option_mangle(options
, flags
, &f
, &o
);
550 return log_full_errno(error_log_level
, r
,
551 "Failed to mangle mount options %s: %m",
554 fl
= mount_flags_to_string(f
);
556 if ((f
& MS_REMOUNT
) && !what
&& !type
)
557 log_debug("Remounting %s (%s \"%s\")...",
558 where
, strnull(fl
), strempty(o
));
559 else if (!what
&& !type
)
560 log_debug("Mounting %s (%s \"%s\")...",
561 where
, strnull(fl
), strempty(o
));
562 else if ((f
& MS_BIND
) && !type
)
563 log_debug("Bind-mounting %s on %s (%s \"%s\")...",
564 what
, where
, strnull(fl
), strempty(o
));
565 else if (f
& MS_MOVE
)
566 log_debug("Moving mount %s → %s (%s \"%s\")...",
567 what
, where
, strnull(fl
), strempty(o
));
569 log_debug("Mounting %s on %s (%s \"%s\")...",
570 strna(type
), where
, strnull(fl
), strempty(o
));
571 if (mount(what
, where
, type
, f
, o
) < 0)
572 return log_full_errno(error_log_level
, errno
,
573 "Failed to mount %s (type %s) on %s (%s \"%s\"): %m",
574 strna(what
), strna(type
), where
, strnull(fl
), strempty(o
));
578 int umount_verbose(const char *what
) {
579 log_debug("Umounting %s...", what
);
580 if (umount(what
) < 0)
581 return log_error_errno(errno
, "Failed to unmount %s: %m", what
);
585 int mount_option_mangle(
587 unsigned long mount_flags
,
588 unsigned long *ret_mount_flags
,
589 char **ret_remaining_options
) {
591 const struct libmnt_optmap
*map
;
592 _cleanup_free_
char *ret
= NULL
;
596 /* This extracts mount flags from the mount options, and store
597 * non-mount-flag options to '*ret_remaining_options'.
599 * "rw,nosuid,nodev,relatime,size=1630748k,mode=700,uid=1000,gid=1000"
600 * is split to MS_NOSUID|MS_NODEV|MS_RELATIME and
601 * "size=1630748k,mode=700,uid=1000,gid=1000".
602 * See more examples in test-mount-utils.c.
604 * Note that if 'options' does not contain any non-mount-flag options,
605 * then '*ret_remaining_options' is set to NULL instead of empty string.
606 * Note that this does not check validity of options stored in
607 * '*ret_remaining_options'.
608 * Note that if 'options' is NULL, then this just copies 'mount_flags'
609 * to '*ret_mount_flags'. */
611 assert(ret_mount_flags
);
612 assert(ret_remaining_options
);
614 map
= mnt_get_builtin_optmap(MNT_LINUX_MAP
);
620 _cleanup_free_
char *word
= NULL
;
621 const struct libmnt_optmap
*ent
;
623 r
= extract_first_word(&p
, &word
, ",", EXTRACT_UNQUOTE
);
629 for (ent
= map
; ent
->name
; ent
++) {
630 /* All entries in MNT_LINUX_MAP do not take any argument.
631 * Thus, ent->name does not contain "=" or "[=]". */
632 if (!streq(word
, ent
->name
))
635 if (!(ent
->mask
& MNT_INVERT
))
636 mount_flags
|= ent
->id
;
637 else if (mount_flags
& ent
->id
)
638 mount_flags
^= ent
->id
;
643 /* If 'word' is not a mount flag, then store it in '*ret_remaining_options'. */
644 if (!ent
->name
&& !strextend_with_separator(&ret
, ",", word
, NULL
))
648 *ret_mount_flags
= mount_flags
;
649 *ret_remaining_options
= TAKE_PTR(ret
);