1 /* SPDX-License-Identifier: LGPL-2.1+ */
3 This file is part of systemd.
5 Copyright 2010 ProFUSION embedded systems
10 #include <linux/loop.h>
12 #include <sys/mount.h>
15 /* This needs to be after sys/mount.h :( */
20 #include "alloc-util.h"
21 #include "blockdev-util.h"
25 #include "fstab-util.h"
26 #include "linux-3.13/dm-ioctl.h"
27 #include "mount-setup.h"
28 #include "mount-util.h"
29 #include "path-util.h"
30 #include "process-util.h"
31 #include "signal-util.h"
32 #include "string-util.h"
33 #include "udev-util.h"
38 DEFINE_TRIVIAL_CLEANUP_FUNC(struct libmnt_table
*, mnt_free_table
);
39 DEFINE_TRIVIAL_CLEANUP_FUNC(struct libmnt_iter
*, mnt_free_iter
);
41 static void mount_point_free(MountPoint
**head
, MountPoint
*m
) {
45 LIST_REMOVE(mount_point
, *head
, m
);
48 free(m
->remount_options
);
52 void mount_points_list_free(MountPoint
**head
) {
56 mount_point_free(head
, *head
);
59 int mount_points_list_get(const char *mountinfo
, MountPoint
**head
) {
60 _cleanup_(mnt_free_tablep
) struct libmnt_table
*t
= NULL
;
61 _cleanup_(mnt_free_iterp
) struct libmnt_iter
*i
= NULL
;
67 i
= mnt_new_iter(MNT_ITER_FORWARD
);
71 r
= mnt_table_parse_mtab(t
, mountinfo
);
73 return log_error_errno(r
, "Failed to parse %s: %m", mountinfo
);
77 const char *path
, *options
, *fstype
;
78 _cleanup_free_
char *p
= NULL
;
79 unsigned long remount_flags
= 0u;
80 _cleanup_free_
char *remount_options
= NULL
;
84 r
= mnt_table_next_fs(t
, i
, &fs
);
88 return log_error_errno(r
, "Failed to get next entry from %s: %m", mountinfo
);
90 path
= mnt_fs_get_target(fs
);
94 if (cunescape(path
, UNESCAPE_RELAX
, &p
) < 0)
97 options
= mnt_fs_get_options(fs
);
98 fstype
= mnt_fs_get_fstype(fs
);
100 /* Ignore mount points we can't unmount because they
101 * are API or because we are keeping them open (like
102 * /dev/console). Also, ignore all mounts below API
103 * file systems, since they are likely virtual too,
104 * and hence not worth spending time on. Also, in
105 * unprivileged containers we might lack the rights to
106 * unmount these things, hence don't bother. */
107 if (mount_point_is_api(p
) ||
108 mount_point_ignore(p
) ||
109 path_startswith(p
, "/dev") ||
110 path_startswith(p
, "/sys") ||
111 path_startswith(p
, "/proc"))
115 /* If we are in a container, don't attempt to
116 * read-only mount anything as that brings no real
117 * benefits, but might confuse the host, as we remount
118 * the superblock here, not the bind mount.
120 * If the filesystem is a network fs, also skip the
121 * remount. It brings no value (we cannot leave
122 * a "dirty fs") and could hang if the network is down.
123 * Note that umount2() is more careful and will not
124 * hang because of the network being down. */
125 try_remount_ro
= detect_container() <= 0 &&
126 !fstype_is_network(fstype
) &&
127 !fstype_is_api_vfs(fstype
) &&
128 !fstype_is_ro(fstype
) &&
129 !fstab_test_yes_no_option(options
, "ro\0rw\0");
131 if (try_remount_ro
) {
132 /* mount(2) states that mount flags and options need to be exactly the same
133 * as they were when the filesystem was mounted, except for the desired
134 * changes. So we reconstruct both here and adjust them for the later
135 * remount call too. */
137 r
= mnt_fs_get_propagation(fs
, &remount_flags
);
139 log_warning_errno(r
, "mnt_fs_get_propagation() failed for %s, ignoring: %m", path
);
143 r
= mount_option_mangle(options
, remount_flags
, &remount_flags
, &remount_options
);
145 log_warning_errno(r
, "mount_option_mangle failed for %s, ignoring: %m", path
);
149 /* MS_BIND is special. If it is provided it will only make the mount-point
150 * read-only. If left out, the super block itself is remounted, which we want. */
151 remount_flags
= (remount_flags
|MS_REMOUNT
|MS_RDONLY
) & ~MS_BIND
;
154 m
= new0(MountPoint
, 1);
158 free_and_replace(m
->path
, p
);
159 free_and_replace(m
->remount_options
, remount_options
);
160 m
->remount_flags
= remount_flags
;
161 m
->try_remount_ro
= try_remount_ro
;
163 LIST_PREPEND(mount_point
, *head
, m
);
169 int swap_list_get(const char *swaps
, MountPoint
**head
) {
170 _cleanup_(mnt_free_tablep
) struct libmnt_table
*t
= NULL
;
171 _cleanup_(mnt_free_iterp
) struct libmnt_iter
*i
= NULL
;
177 i
= mnt_new_iter(MNT_ITER_FORWARD
);
181 r
= mnt_table_parse_swaps(t
, swaps
);
183 return log_error_errno(r
, "Failed to parse %s: %m", swaps
);
186 struct libmnt_fs
*fs
;
190 _cleanup_free_
char *d
= NULL
;
192 r
= mnt_table_next_fs(t
, i
, &fs
);
196 return log_error_errno(r
, "Failed to get next entry from %s: %m", swaps
);
198 source
= mnt_fs_get_source(fs
);
202 r
= cunescape(source
, UNESCAPE_RELAX
, &d
);
206 swap
= new0(MountPoint
, 1);
210 free_and_replace(swap
->path
, d
);
211 LIST_PREPEND(mount_point
, *head
, swap
);
217 static int loopback_list_get(MountPoint
**head
) {
218 _cleanup_udev_enumerate_unref_
struct udev_enumerate
*e
= NULL
;
219 struct udev_list_entry
*item
= NULL
, *first
= NULL
;
220 _cleanup_udev_unref_
struct udev
*udev
= NULL
;
229 e
= udev_enumerate_new(udev
);
233 r
= udev_enumerate_add_match_subsystem(e
, "block");
237 r
= udev_enumerate_add_match_sysname(e
, "loop*");
241 r
= udev_enumerate_add_match_sysattr(e
, "loop/backing_file", NULL
);
245 r
= udev_enumerate_scan_devices(e
);
249 first
= udev_enumerate_get_list_entry(e
);
250 udev_list_entry_foreach(item
, first
) {
251 _cleanup_udev_device_unref_
struct udev_device
*d
;
253 _cleanup_free_ MountPoint
*lb
= NULL
;
255 d
= udev_device_new_from_syspath(udev
, udev_list_entry_get_name(item
));
259 dn
= udev_device_get_devnode(d
);
263 lb
= new0(MountPoint
, 1);
267 r
= free_and_strdup(&lb
->path
, dn
);
271 LIST_PREPEND(mount_point
, *head
, lb
);
278 static int dm_list_get(MountPoint
**head
) {
279 _cleanup_udev_enumerate_unref_
struct udev_enumerate
*e
= NULL
;
280 struct udev_list_entry
*item
= NULL
, *first
= NULL
;
281 _cleanup_udev_unref_
struct udev
*udev
= NULL
;
290 e
= udev_enumerate_new(udev
);
294 r
= udev_enumerate_add_match_subsystem(e
, "block");
298 r
= udev_enumerate_add_match_sysname(e
, "dm-*");
302 r
= udev_enumerate_scan_devices(e
);
306 first
= udev_enumerate_get_list_entry(e
);
307 udev_list_entry_foreach(item
, first
) {
308 _cleanup_udev_device_unref_
struct udev_device
*d
;
311 _cleanup_free_ MountPoint
*m
= NULL
;
313 d
= udev_device_new_from_syspath(udev
, udev_list_entry_get_name(item
));
317 devnum
= udev_device_get_devnum(d
);
318 dn
= udev_device_get_devnode(d
);
319 if (major(devnum
) == 0 || !dn
)
322 m
= new0(MountPoint
, 1);
327 r
= free_and_strdup(&m
->path
, dn
);
331 LIST_PREPEND(mount_point
, *head
, m
);
338 static int delete_loopback(const char *device
) {
339 _cleanup_close_
int fd
= -1;
344 fd
= open(device
, O_RDONLY
|O_CLOEXEC
);
346 return errno
== ENOENT
? 0 : -errno
;
348 r
= ioctl(fd
, LOOP_CLR_FD
, 0);
352 /* ENXIO: not bound, so no error */
359 static int delete_dm(dev_t devnum
) {
361 struct dm_ioctl dm
= {
365 DM_VERSION_PATCHLEVEL
367 .data_size
= sizeof(dm
),
371 _cleanup_close_
int fd
= -1;
373 assert(major(devnum
) != 0);
375 fd
= open("/dev/mapper/control", O_RDWR
|O_CLOEXEC
);
379 if (ioctl(fd
, DM_DEV_REMOVE
, &dm
) < 0)
385 static bool nonunmountable_path(const char *path
) {
386 return path_equal(path
, "/")
388 || path_equal(path
, "/usr")
390 || path_startswith(path
, "/run/initramfs");
393 static int remount_with_timeout(MountPoint
*m
, int umount_log_level
) {
397 BLOCK_SIGNALS(SIGCHLD
);
401 /* Due to the possiblity of a remount operation hanging, we
402 * fork a child process and set a timeout. If the timeout
403 * lapses, the assumption is that that particular remount
405 r
= safe_fork("(sd-remount)", FORK_RESET_SIGNALS
|FORK_CLOSE_ALL_FDS
|FORK_LOG
|FORK_REOPEN_LOG
, &pid
);
409 log_info("Remounting '%s' read-only in with options '%s'.", m
->path
, m
->remount_options
);
411 /* Start the mount operation here in the child */
412 r
= mount(NULL
, m
->path
, NULL
, m
->remount_flags
, m
->remount_options
);
414 log_full_errno(umount_log_level
, errno
, "Failed to remount '%s' read-only: %m", m
->path
);
416 _exit(r
< 0 ? EXIT_FAILURE
: EXIT_SUCCESS
);
419 r
= wait_for_terminate_with_timeout(pid
, DEFAULT_TIMEOUT_USEC
);
420 if (r
== -ETIMEDOUT
) {
421 log_error_errno(r
, "Remounting '%s' timed out, issuing SIGKILL to PID " PID_FMT
".", m
->path
, pid
);
422 (void) kill(pid
, SIGKILL
);
423 } else if (r
== -EPROTO
)
424 log_debug_errno(r
, "Remounting '%s' failed abnormally, child process " PID_FMT
" aborted or exited non-zero.", m
->path
, pid
);
426 log_error_errno(r
, "Remounting '%s' failed unexpectedly, couldn't wait for child process " PID_FMT
": %m", m
->path
, pid
);
431 static int umount_with_timeout(MountPoint
*m
, int umount_log_level
) {
435 BLOCK_SIGNALS(SIGCHLD
);
439 /* Due to the possiblity of a umount operation hanging, we
440 * fork a child process and set a timeout. If the timeout
441 * lapses, the assumption is that that particular umount
443 r
= safe_fork("(sd-umount)", FORK_RESET_SIGNALS
|FORK_CLOSE_ALL_FDS
|FORK_LOG
|FORK_REOPEN_LOG
, &pid
);
447 log_info("Unmounting '%s'.", m
->path
);
449 /* Start the mount operation here in the child Using MNT_FORCE
450 * causes some filesystems (e.g. FUSE and NFS and other network
451 * filesystems) to abort any pending requests and return -EIO
452 * rather than blocking indefinitely. If the filesysten is
453 * "busy", this may allow processes to die, thus making the
454 * filesystem less busy so the unmount might succeed (rather
455 * then return EBUSY).*/
456 r
= umount2(m
->path
, MNT_FORCE
);
458 log_full_errno(umount_log_level
, errno
, "Failed to unmount %s: %m", m
->path
);
460 _exit(r
< 0 ? EXIT_FAILURE
: EXIT_SUCCESS
);
463 r
= wait_for_terminate_with_timeout(pid
, DEFAULT_TIMEOUT_USEC
);
464 if (r
== -ETIMEDOUT
) {
465 log_error_errno(r
, "Unmounting '%s' timed out, issuing SIGKILL to PID " PID_FMT
".", m
->path
, pid
);
466 (void) kill(pid
, SIGKILL
);
467 } else if (r
== -EPROTO
)
468 log_debug_errno(r
, "Unmounting '%s' failed abnormally, child process " PID_FMT
" aborted or exited non-zero.", m
->path
, pid
);
470 log_error_errno(r
, "Unmounting '%s' failed unexpectedly, couldn't wait for child process " PID_FMT
": %m", m
->path
, pid
);
475 /* This includes remounting readonly, which changes the kernel mount options.
476 * Therefore the list passed to this function is invalidated, and should not be reused. */
477 static int mount_points_list_umount(MountPoint
**head
, bool *changed
, int umount_log_level
) {
484 LIST_FOREACH(mount_point
, m
, *head
) {
485 if (m
->try_remount_ro
) {
486 /* We always try to remount directories
487 * read-only first, before we go on and umount
490 * Mount points can be stacked. If a mount
491 * point is stacked below / or /usr, we
492 * cannot umount or remount it directly,
493 * since there is no way to refer to the
494 * underlying mount. There's nothing we can do
495 * about it for the general case, but we can
496 * do something about it if it is aliased
497 * somehwere else via a bind mount. If we
498 * explicitly remount the super block of that
499 * alias read-only we hence should be
500 * relatively safe regarding keeping a dirty fs
501 * we cannot otherwise see.
503 * Since the remount can hang in the instance of
504 * remote filesystems, we remount asynchronously
505 * and skip the subsequent umount if it fails. */
506 if (remount_with_timeout(m
, umount_log_level
) < 0) {
507 /* Remount failed, but try unmounting anyway,
508 * unless this is a mount point we want to skip. */
509 if (nonunmountable_path(m
->path
)) {
516 /* Skip / and /usr since we cannot unmount that
517 * anyway, since we are running from it. They have
518 * already been remounted ro. */
519 if (nonunmountable_path(m
->path
))
522 /* Trying to umount */
523 if (umount_with_timeout(m
, umount_log_level
) < 0)
532 static int swap_points_list_off(MountPoint
**head
, bool *changed
) {
539 LIST_FOREACH_SAFE(mount_point
, m
, n
, *head
) {
540 log_info("Deactivating swap %s.", m
->path
);
541 if (swapoff(m
->path
) == 0) {
543 mount_point_free(head
, m
);
545 log_warning_errno(errno
, "Could not deactivate swap %s: %m", m
->path
);
553 static int loopback_points_list_detach(MountPoint
**head
, bool *changed
, int umount_log_level
) {
561 k
= lstat("/", &root_st
);
563 LIST_FOREACH_SAFE(mount_point
, m
, n
, *head
) {
565 struct stat loopback_st
;
568 major(root_st
.st_dev
) != 0 &&
569 lstat(m
->path
, &loopback_st
) >= 0 &&
570 root_st
.st_dev
== loopback_st
.st_rdev
) {
575 log_info("Detaching loopback %s.", m
->path
);
576 r
= delete_loopback(m
->path
);
581 mount_point_free(head
, m
);
583 log_full_errno(umount_log_level
, errno
, "Could not detach loopback %s: %m", m
->path
);
591 static int dm_points_list_detach(MountPoint
**head
, bool *changed
, int umount_log_level
) {
599 r
= get_block_device("/", &rootdev
);
603 LIST_FOREACH_SAFE(mount_point
, m
, n
, *head
) {
605 if (major(rootdev
) != 0 && rootdev
== m
->devnum
) {
610 log_info("Detaching DM %u:%u.", major(m
->devnum
), minor(m
->devnum
));
611 r
= delete_dm(m
->devnum
);
614 mount_point_free(head
, m
);
616 log_full_errno(umount_log_level
, errno
, "Could not detach DM %s: %m", m
->path
);
624 static int umount_all_once(bool *changed
, int umount_log_level
) {
626 _cleanup_(mount_points_list_free
) LIST_HEAD(MountPoint
, mp_list_head
);
630 LIST_HEAD_INIT(mp_list_head
);
631 r
= mount_points_list_get(NULL
, &mp_list_head
);
635 return mount_points_list_umount(&mp_list_head
, changed
, umount_log_level
);
638 int umount_all(bool *changed
, int umount_log_level
) {
644 /* Retry umount, until nothing can be umounted anymore. Mounts are
645 * processed in order, newest first. The retries are needed when
646 * an old mount has been moved, to a path inside a newer mount. */
648 umount_changed
= false;
650 r
= umount_all_once(&umount_changed
, umount_log_level
);
653 } while (umount_changed
);
658 int swapoff_all(bool *changed
) {
659 _cleanup_(mount_points_list_free
) LIST_HEAD(MountPoint
, swap_list_head
);
664 LIST_HEAD_INIT(swap_list_head
);
666 r
= swap_list_get(NULL
, &swap_list_head
);
670 return swap_points_list_off(&swap_list_head
, changed
);
673 int loopback_detach_all(bool *changed
, int umount_log_level
) {
674 _cleanup_(mount_points_list_free
) LIST_HEAD(MountPoint
, loopback_list_head
);
679 LIST_HEAD_INIT(loopback_list_head
);
681 r
= loopback_list_get(&loopback_list_head
);
685 return loopback_points_list_detach(&loopback_list_head
, changed
, umount_log_level
);
688 int dm_detach_all(bool *changed
, int umount_log_level
) {
689 _cleanup_(mount_points_list_free
) LIST_HEAD(MountPoint
, dm_list_head
);
694 LIST_HEAD_INIT(dm_list_head
);
696 r
= dm_list_get(&dm_list_head
);
700 return dm_points_list_detach(&dm_list_head
, changed
, umount_log_level
);