1 /* SPDX-License-Identifier: LGPL-2.1+ */
3 This file is part of systemd.
5 Copyright 2010 Lennart Poettering
7 systemd is free software; you can redistribute it and/or modify it
8 under the terms of the GNU Lesser General Public License as published by
9 the Free Software Foundation; either version 2.1 of the License, or
10 (at your option) any later version.
12 systemd is distributed in the hope that it will be useful, but
13 WITHOUT ANY WARRANTY; without even the implied warranty of
14 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
15 Lesser General Public License for more details.
17 You should have received a copy of the GNU Lesser General Public License
18 along with systemd; If not, see <http://www.gnu.org/licenses/>.
22 #include <stdio_ext.h>
25 #include <sys/mount.h>
27 #include <sys/statvfs.h>
30 #include "alloc-util.h"
36 #include "mount-util.h"
37 #include "parse-util.h"
38 #include "path-util.h"
40 #include "stdio-util.h"
41 #include "string-util.h"
44 /* This is the original MAX_HANDLE_SZ definition from the kernel, when the API was introduced. We use that in place of
45 * any more currently defined value to future-proof things: if the size is increased in the API headers, and our code
46 * is recompiled then it would cease working on old kernels, as those refuse any sizes larger than this value with
47 * EINVAL right-away. Hence, let's disconnect ourselves from any such API changes, and stick to the original definition
48 * from when it was introduced. We use it as a start value only anyway (see below), and hence should be able to deal
49 * with large file handles anyway. */
50 #define ORIGINAL_MAX_HANDLE_SZ 128
52 int name_to_handle_at_loop(
55 struct file_handle
**ret_handle
,
59 _cleanup_free_
struct file_handle
*h
= NULL
;
60 size_t n
= ORIGINAL_MAX_HANDLE_SZ
;
62 /* We need to invoke name_to_handle_at() in a loop, given that it might return EOVERFLOW when the specified
63 * buffer is too small. Note that in contrast to what the docs might suggest, MAX_HANDLE_SZ is only good as a
64 * start value, it is not an upper bound on the buffer size required.
66 * This improves on raw name_to_handle_at() also in one other regard: ret_handle and ret_mnt_id can be passed
67 * as NULL if there's no interest in either. */
72 h
= malloc0(offsetof(struct file_handle
, f_handle
) + n
);
78 if (name_to_handle_at(fd
, path
, h
, &mnt_id
, flags
) >= 0) {
90 if (errno
!= EOVERFLOW
)
93 if (!ret_handle
&& ret_mnt_id
&& mnt_id
>= 0) {
95 /* As it appears, name_to_handle_at() fills in mnt_id even when it returns EOVERFLOW when the
96 * buffer is too small, but that's undocumented. Hence, let's make use of this if it appears to
97 * be filled in, and the caller was interested in only the mount ID an nothing else. */
103 /* If name_to_handle_at() didn't increase the byte size, then this EOVERFLOW is caused by something
104 * else (apparently EOVERFLOW is returned for untriggered nfs4 mounts sometimes), not by the too small
105 * buffer. In that case propagate EOVERFLOW */
106 if (h
->handle_bytes
<= n
)
109 /* The buffer was too small. Size the new buffer by what name_to_handle_at() returned. */
111 if (offsetof(struct file_handle
, f_handle
) + n
< n
) /* check for addition overflow */
118 static int fd_fdinfo_mnt_id(int fd
, const char *filename
, int flags
, int *mnt_id
) {
119 char path
[STRLEN("/proc/self/fdinfo/") + DECIMAL_STR_MAX(int)];
120 _cleanup_free_
char *fdinfo
= NULL
;
121 _cleanup_close_
int subfd
= -1;
125 if ((flags
& AT_EMPTY_PATH
) && isempty(filename
))
126 xsprintf(path
, "/proc/self/fdinfo/%i", fd
);
128 subfd
= openat(fd
, filename
, O_CLOEXEC
|O_PATH
);
132 xsprintf(path
, "/proc/self/fdinfo/%i", subfd
);
135 r
= read_full_file(path
, &fdinfo
, NULL
);
136 if (r
== -ENOENT
) /* The fdinfo directory is a relatively new addition */
141 p
= startswith(fdinfo
, "mnt_id:");
143 p
= strstr(fdinfo
, "\nmnt_id:");
144 if (!p
) /* The mnt_id field is a relatively new addition */
150 p
+= strspn(p
, WHITESPACE
);
151 p
[strcspn(p
, WHITESPACE
)] = 0;
153 return safe_atoi(p
, mnt_id
);
156 int fd_is_mount_point(int fd
, const char *filename
, int flags
) {
157 _cleanup_free_
struct file_handle
*h
= NULL
, *h_parent
= NULL
;
158 int mount_id
= -1, mount_id_parent
= -1;
159 bool nosupp
= false, check_st_dev
= true;
166 /* First we will try the name_to_handle_at() syscall, which
167 * tells us the mount id and an opaque file "handle". It is
168 * not supported everywhere though (kernel compile-time
169 * option, not all file systems are hooked up). If it works
170 * the mount id is usually good enough to tell us whether
171 * something is a mount point.
173 * If that didn't work we will try to read the mount id from
174 * /proc/self/fdinfo/<fd>. This is almost as good as
175 * name_to_handle_at(), however, does not return the
176 * opaque file handle. The opaque file handle is pretty useful
177 * to detect the root directory, which we should always
178 * consider a mount point. Hence we use this only as
179 * fallback. Exporting the mnt_id in fdinfo is a pretty recent
182 * As last fallback we do traditional fstat() based st_dev
183 * comparisons. This is how things were traditionally done,
184 * but unionfs breaks this since it exposes file
185 * systems with a variety of st_dev reported. Also, btrfs
186 * subvolumes have different st_dev, even though they aren't
187 * real mounts of their own. */
189 r
= name_to_handle_at_loop(fd
, filename
, &h
, &mount_id
, flags
);
190 if (IN_SET(r
, -ENOSYS
, -EACCES
, -EPERM
, -EOVERFLOW
, -EINVAL
))
191 /* This kernel does not support name_to_handle_at() at all (ENOSYS), or the syscall was blocked
192 * (EACCES/EPERM; maybe through seccomp, because we are running inside of a container?), or the mount
193 * point is not triggered yet (EOVERFLOW, think nfs4), or some general name_to_handle_at() flakiness
194 * (EINVAL): fall back to simpler logic. */
195 goto fallback_fdinfo
;
196 else if (r
== -EOPNOTSUPP
)
197 /* This kernel or file system does not support name_to_handle_at(), hence let's see if the upper fs
198 * supports it (in which case it is a mount point), otherwise fallback to the traditional stat()
204 r
= name_to_handle_at_loop(fd
, "", &h_parent
, &mount_id_parent
, AT_EMPTY_PATH
);
205 if (r
== -EOPNOTSUPP
) {
207 /* Neither parent nor child do name_to_handle_at()? We have no choice but to fall back. */
208 goto fallback_fdinfo
;
210 /* The parent can't do name_to_handle_at() but the directory we are interested in can? If so,
211 * it must be a mount point. */
216 /* The parent can do name_to_handle_at() but the
217 * directory we are interested in can't? If so, it
218 * must be a mount point. */
222 /* If the file handle for the directory we are
223 * interested in and its parent are identical, we
224 * assume this is the root directory, which is a mount
227 if (h
->handle_bytes
== h_parent
->handle_bytes
&&
228 h
->handle_type
== h_parent
->handle_type
&&
229 memcmp(h
->f_handle
, h_parent
->f_handle
, h
->handle_bytes
) == 0)
232 return mount_id
!= mount_id_parent
;
235 r
= fd_fdinfo_mnt_id(fd
, filename
, flags
, &mount_id
);
236 if (IN_SET(r
, -EOPNOTSUPP
, -EACCES
, -EPERM
))
241 r
= fd_fdinfo_mnt_id(fd
, "", AT_EMPTY_PATH
, &mount_id_parent
);
245 if (mount_id
!= mount_id_parent
)
248 /* Hmm, so, the mount ids are the same. This leaves one
249 * special case though for the root file system. For that,
250 * let's see if the parent directory has the same inode as we
251 * are interested in. Hence, let's also do fstat() checks now,
252 * too, but avoid the st_dev comparisons, since they aren't
253 * that useful on unionfs mounts. */
254 check_st_dev
= false;
257 /* yay for fstatat() taking a different set of flags than the other
259 if (flags
& AT_SYMLINK_FOLLOW
)
260 flags
&= ~AT_SYMLINK_FOLLOW
;
262 flags
|= AT_SYMLINK_NOFOLLOW
;
263 if (fstatat(fd
, filename
, &a
, flags
) < 0)
266 if (fstatat(fd
, "", &b
, AT_EMPTY_PATH
) < 0)
269 /* A directory with same device and inode as its parent? Must
270 * be the root directory */
271 if (a
.st_dev
== b
.st_dev
&&
272 a
.st_ino
== b
.st_ino
)
275 return check_st_dev
&& (a
.st_dev
!= b
.st_dev
);
278 /* flags can be AT_SYMLINK_FOLLOW or 0 */
279 int path_is_mount_point(const char *t
, const char *root
, int flags
) {
280 _cleanup_free_
char *canonical
= NULL
, *parent
= NULL
;
281 _cleanup_close_
int fd
= -1;
285 assert((flags
& ~AT_SYMLINK_FOLLOW
) == 0);
287 if (path_equal(t
, "/"))
290 /* we need to resolve symlinks manually, we can't just rely on
291 * fd_is_mount_point() to do that for us; if we have a structure like
292 * /bin -> /usr/bin/ and /usr is a mount point, then the parent that we
293 * look at needs to be /usr, not /. */
294 if (flags
& AT_SYMLINK_FOLLOW
) {
295 r
= chase_symlinks(t
, root
, 0, &canonical
);
302 parent
= dirname_malloc(t
);
306 fd
= openat(AT_FDCWD
, parent
, O_DIRECTORY
|O_CLOEXEC
|O_PATH
);
310 return fd_is_mount_point(fd
, last_path_component(t
), flags
);
313 int path_get_mnt_id(const char *path
, int *ret
) {
316 r
= name_to_handle_at_loop(AT_FDCWD
, path
, NULL
, ret
, 0);
317 if (IN_SET(r
, -EOPNOTSUPP
, -ENOSYS
, -EACCES
, -EPERM
, -EOVERFLOW
, -EINVAL
)) /* kernel/fs don't support this, or seccomp blocks access, or untriggered mount, or name_to_handle_at() is flaky */
318 return fd_fdinfo_mnt_id(AT_FDCWD
, path
, 0, ret
);
323 int umount_recursive(const char *prefix
, int flags
) {
327 /* Try to umount everything recursively below a
328 * directory. Also, take care of stacked mounts, and keep
329 * unmounting them until they are gone. */
332 _cleanup_fclose_
FILE *proc_self_mountinfo
= NULL
;
337 proc_self_mountinfo
= fopen("/proc/self/mountinfo", "re");
338 if (!proc_self_mountinfo
)
341 (void) __fsetlocking(proc_self_mountinfo
, FSETLOCKING_BYCALLER
);
344 _cleanup_free_
char *path
= NULL
, *p
= NULL
;
347 k
= fscanf(proc_self_mountinfo
,
348 "%*s " /* (1) mount id */
349 "%*s " /* (2) parent id */
350 "%*s " /* (3) major:minor */
351 "%*s " /* (4) root */
352 "%ms " /* (5) mount point */
353 "%*s" /* (6) mount options */
354 "%*[^-]" /* (7) optional fields */
355 "- " /* (8) separator */
356 "%*s " /* (9) file system type */
357 "%*s" /* (10) mount source */
358 "%*s" /* (11) mount options 2 */
359 "%*[^\n]", /* some rubbish at the end */
368 r
= cunescape(path
, UNESCAPE_RELAX
, &p
);
372 if (!path_startswith(p
, prefix
))
375 if (umount2(p
, flags
) < 0) {
376 r
= log_debug_errno(errno
, "Failed to umount %s: %m", p
);
380 log_debug("Successfully unmounted %s", p
);
393 static int get_mount_flags(const char *path
, unsigned long *flags
) {
396 if (statvfs(path
, &buf
) < 0)
402 /* Use this function only if do you have direct access to /proc/self/mountinfo
403 * and need the caller to open it for you. This is the case when /proc is
404 * masked or not mounted. Otherwise, use bind_remount_recursive. */
405 int bind_remount_recursive_with_mountinfo(const char *prefix
, bool ro
, char **blacklist
, FILE *proc_self_mountinfo
) {
406 _cleanup_set_free_free_ Set
*done
= NULL
;
407 _cleanup_free_
char *cleaned
= NULL
;
410 assert(proc_self_mountinfo
);
412 /* Recursively remount a directory (and all its submounts) read-only or read-write. If the directory is already
413 * mounted, we reuse the mount and simply mark it MS_BIND|MS_RDONLY (or remove the MS_RDONLY for read-write
414 * operation). If it isn't we first make it one. Afterwards we apply MS_BIND|MS_RDONLY (or remove MS_RDONLY) to
415 * all submounts we can access, too. When mounts are stacked on the same mount point we only care for each
416 * individual "top-level" mount on each point, as we cannot influence/access the underlying mounts anyway. We
417 * do not have any effect on future submounts that might get propagated, they migt be writable. This includes
418 * future submounts that have been triggered via autofs.
420 * If the "blacklist" parameter is specified it may contain a list of subtrees to exclude from the
421 * remount operation. Note that we'll ignore the blacklist for the top-level path. */
423 cleaned
= strdup(prefix
);
427 path_kill_slashes(cleaned
);
429 done
= set_new(&string_hash_ops
);
434 _cleanup_set_free_free_ Set
*todo
= NULL
;
435 bool top_autofs
= false;
437 unsigned long orig_flags
;
439 todo
= set_new(&string_hash_ops
);
443 rewind(proc_self_mountinfo
);
446 _cleanup_free_
char *path
= NULL
, *p
= NULL
, *type
= NULL
;
449 k
= fscanf(proc_self_mountinfo
,
450 "%*s " /* (1) mount id */
451 "%*s " /* (2) parent id */
452 "%*s " /* (3) major:minor */
453 "%*s " /* (4) root */
454 "%ms " /* (5) mount point */
455 "%*s" /* (6) mount options (superblock) */
456 "%*[^-]" /* (7) optional fields */
457 "- " /* (8) separator */
458 "%ms " /* (9) file system type */
459 "%*s" /* (10) mount source */
460 "%*s" /* (11) mount options (bind mount) */
461 "%*[^\n]", /* some rubbish at the end */
471 r
= cunescape(path
, UNESCAPE_RELAX
, &p
);
475 if (!path_startswith(p
, cleaned
))
478 /* Ignore this mount if it is blacklisted, but only if it isn't the top-level mount we shall
480 if (!path_equal(cleaned
, p
)) {
481 bool blacklisted
= false;
484 STRV_FOREACH(i
, blacklist
) {
486 if (path_equal(*i
, cleaned
))
489 if (!path_startswith(*i
, cleaned
))
492 if (path_startswith(p
, *i
)) {
494 log_debug("Not remounting %s, because blacklisted by %s, called for %s", p
, *i
, cleaned
);
502 /* Let's ignore autofs mounts. If they aren't
503 * triggered yet, we want to avoid triggering
504 * them, as we don't make any guarantees for
505 * future submounts anyway. If they are
506 * already triggered, then we will find
507 * another entry for this. */
508 if (streq(type
, "autofs")) {
509 top_autofs
= top_autofs
|| path_equal(cleaned
, p
);
513 if (!set_contains(done
, p
)) {
514 r
= set_consume(todo
, p
);
523 /* If we have no submounts to process anymore and if
524 * the root is either already done, or an autofs, we
526 if (set_isempty(todo
) &&
527 (top_autofs
|| set_contains(done
, cleaned
)))
530 if (!set_contains(done
, cleaned
) &&
531 !set_contains(todo
, cleaned
)) {
532 /* The prefix directory itself is not yet a mount, make it one. */
533 if (mount(cleaned
, cleaned
, NULL
, MS_BIND
|MS_REC
, NULL
) < 0)
537 (void) get_mount_flags(cleaned
, &orig_flags
);
538 orig_flags
&= ~MS_RDONLY
;
540 if (mount(NULL
, prefix
, NULL
, orig_flags
|MS_BIND
|MS_REMOUNT
|(ro
? MS_RDONLY
: 0), NULL
) < 0)
543 log_debug("Made top-level directory %s a mount point.", prefix
);
549 r
= set_consume(done
, x
);
554 while ((x
= set_steal_first(todo
))) {
556 r
= set_consume(done
, x
);
557 if (IN_SET(r
, 0, -EEXIST
))
562 /* Deal with mount points that are obstructed by a later mount */
563 r
= path_is_mount_point(x
, NULL
, 0);
564 if (IN_SET(r
, 0, -ENOENT
))
569 /* Try to reuse the original flag set */
571 (void) get_mount_flags(x
, &orig_flags
);
572 orig_flags
&= ~MS_RDONLY
;
574 if (mount(NULL
, x
, NULL
, orig_flags
|MS_BIND
|MS_REMOUNT
|(ro
? MS_RDONLY
: 0), NULL
) < 0)
577 log_debug("Remounted %s read-only.", x
);
582 int bind_remount_recursive(const char *prefix
, bool ro
, char **blacklist
) {
583 _cleanup_fclose_
FILE *proc_self_mountinfo
= NULL
;
585 proc_self_mountinfo
= fopen("/proc/self/mountinfo", "re");
586 if (!proc_self_mountinfo
)
589 (void) __fsetlocking(proc_self_mountinfo
, FSETLOCKING_BYCALLER
);
591 return bind_remount_recursive_with_mountinfo(prefix
, ro
, blacklist
, proc_self_mountinfo
);
594 int mount_move_root(const char *path
) {
600 if (mount(path
, "/", NULL
, MS_MOVE
, NULL
) < 0)
612 bool fstype_is_network(const char *fstype
) {
615 x
= startswith(fstype
, "fuse.");
619 return STR_IN_SET(fstype
,
631 "pvfs2", /* OrangeFS */
636 bool fstype_is_api_vfs(const char *fstype
) {
637 return STR_IN_SET(fstype
,
660 bool fstype_is_ro(const char *fstype
) {
661 /* All Linux file systems that are necessarily read-only */
662 return STR_IN_SET(fstype
,
668 bool fstype_can_discard(const char *fstype
) {
669 return STR_IN_SET(fstype
,
676 bool fstype_can_uid_gid(const char *fstype
) {
678 /* All file systems that have a uid=/gid= mount option that fixates the owners of all files and directories,
679 * current and future. */
681 return STR_IN_SET(fstype
,
692 int repeat_unmount(const char *path
, int flags
) {
697 /* If there are multiple mounts on a mount point, this
698 * removes them all */
701 if (umount2(path
, flags
) < 0) {
713 const char* mode_to_inaccessible_node(mode_t mode
) {
714 /* This function maps a node type to the correspondent inaccessible node type.
715 * Character and block inaccessible devices may not be created (because major=0 and minor=0),
716 * in such case we map character and block devices to the inaccessible node type socket. */
717 switch(mode
& S_IFMT
) {
719 return "/run/systemd/inaccessible/reg";
721 return "/run/systemd/inaccessible/dir";
723 if (access("/run/systemd/inaccessible/chr", F_OK
) == 0)
724 return "/run/systemd/inaccessible/chr";
725 return "/run/systemd/inaccessible/sock";
727 if (access("/run/systemd/inaccessible/blk", F_OK
) == 0)
728 return "/run/systemd/inaccessible/blk";
729 return "/run/systemd/inaccessible/sock";
731 return "/run/systemd/inaccessible/fifo";
733 return "/run/systemd/inaccessible/sock";
738 #define FLAG(name) (flags & name ? STRINGIFY(name) "|" : "")
739 static char* mount_flags_to_string(long unsigned flags
) {
741 _cleanup_free_
char *y
= NULL
;
742 long unsigned overflow
;
744 overflow
= flags
& ~(MS_RDONLY
|
769 if (flags
== 0 || overflow
!= 0)
770 if (asprintf(&y
, "%lx", overflow
) < 0)
773 x
= strjoin(FLAG(MS_RDONLY
),
777 FLAG(MS_SYNCHRONOUS
),
795 FLAG(MS_STRICTATIME
),
801 x
[strlen(x
) - 1] = '\0'; /* truncate the last | */
811 const char *options
) {
813 _cleanup_free_
char *fl
= NULL
;
815 fl
= mount_flags_to_string(flags
);
817 if ((flags
& MS_REMOUNT
) && !what
&& !type
)
818 log_debug("Remounting %s (%s \"%s\")...",
819 where
, strnull(fl
), strempty(options
));
820 else if (!what
&& !type
)
821 log_debug("Mounting %s (%s \"%s\")...",
822 where
, strnull(fl
), strempty(options
));
823 else if ((flags
& MS_BIND
) && !type
)
824 log_debug("Bind-mounting %s on %s (%s \"%s\")...",
825 what
, where
, strnull(fl
), strempty(options
));
826 else if (flags
& MS_MOVE
)
827 log_debug("Moving mount %s → %s (%s \"%s\")...",
828 what
, where
, strnull(fl
), strempty(options
));
830 log_debug("Mounting %s on %s (%s \"%s\")...",
831 strna(type
), where
, strnull(fl
), strempty(options
));
832 if (mount(what
, where
, type
, flags
, options
) < 0)
833 return log_full_errno(error_log_level
, errno
,
834 "Failed to mount %s on %s (%s \"%s\"): %m",
835 strna(type
), where
, strnull(fl
), strempty(options
));
839 int umount_verbose(const char *what
) {
840 log_debug("Umounting %s...", what
);
841 if (umount(what
) < 0)
842 return log_error_errno(errno
, "Failed to unmount %s: %m", what
);
846 const char *mount_propagation_flags_to_string(unsigned long flags
) {
848 switch (flags
& (MS_SHARED
|MS_SLAVE
|MS_PRIVATE
)) {
863 int mount_propagation_flags_from_string(const char *name
, unsigned long *ret
) {
867 else if (streq(name
, "shared"))
869 else if (streq(name
, "slave"))
871 else if (streq(name
, "private"))