]> git.ipfire.org Git - thirdparty/systemd.git/blob - src/core/umount.c
c3decf307578c998ef85d72904c1aafa7c8a9dd2
[thirdparty/systemd.git] / src / core / umount.c
1 /* SPDX-License-Identifier: LGPL-2.1+ */
2 /***
3 This file is part of systemd.
4
5 Copyright 2010 ProFUSION embedded systems
6 ***/
7
8 #include <errno.h>
9 #include <fcntl.h>
10 #include <linux/loop.h>
11 #include <string.h>
12 #include <sys/mount.h>
13 #include <sys/swap.h>
14
15 /* This needs to be after sys/mount.h :( */
16 #include <libmount.h>
17
18 #include "libudev.h"
19
20 #include "alloc-util.h"
21 #include "blockdev-util.h"
22 #include "def.h"
23 #include "escape.h"
24 #include "fd-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"
34 #include "umount.h"
35 #include "util.h"
36 #include "virt.h"
37
38 DEFINE_TRIVIAL_CLEANUP_FUNC(struct libmnt_table*, mnt_free_table);
39 DEFINE_TRIVIAL_CLEANUP_FUNC(struct libmnt_iter*, mnt_free_iter);
40
41 static void mount_point_free(MountPoint **head, MountPoint *m) {
42 assert(head);
43 assert(m);
44
45 LIST_REMOVE(mount_point, *head, m);
46
47 free(m->path);
48 free(m->remount_options);
49 free(m);
50 }
51
52 void mount_points_list_free(MountPoint **head) {
53 assert(head);
54
55 while (*head)
56 mount_point_free(head, *head);
57 }
58
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;
62 int r;
63
64 assert(head);
65
66 t = mnt_new_table();
67 i = mnt_new_iter(MNT_ITER_FORWARD);
68 if (!t || !i)
69 return log_oom();
70
71 r = mnt_table_parse_mtab(t, mountinfo);
72 if (r < 0)
73 return log_error_errno(r, "Failed to parse %s: %m", mountinfo);
74
75 for (;;) {
76 struct libmnt_fs *fs;
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;
81 bool try_remount_ro;
82 MountPoint *m;
83
84 r = mnt_table_next_fs(t, i, &fs);
85 if (r == 1)
86 break;
87 if (r < 0)
88 return log_error_errno(r, "Failed to get next entry from %s: %m", mountinfo);
89
90 path = mnt_fs_get_target(fs);
91 if (!path)
92 continue;
93
94 if (cunescape(path, UNESCAPE_RELAX, &p) < 0)
95 return log_oom();
96
97 options = mnt_fs_get_options(fs);
98 fstype = mnt_fs_get_fstype(fs);
99
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"))
112 continue;
113
114 /* If we are in a container, don't attempt to
115 * read-only mount anything as that brings no real
116 * benefits, but might confuse the host, as we remount
117 * the superblock here, not the bind mount.
118 *
119 * If the filesystem is a network fs, also skip the
120 * remount. It brings no value (we cannot leave
121 * a "dirty fs") and could hang if the network is down.
122 * Note that umount2() is more careful and will not
123 * hang because of the network being down. */
124 try_remount_ro = detect_container() <= 0 &&
125 !fstype_is_network(fstype) &&
126 !fstype_is_api_vfs(fstype) &&
127 !fstype_is_ro(fstype) &&
128 !fstab_test_yes_no_option(options, "ro\0rw\0");
129
130 if (try_remount_ro) {
131 /* mount(2) states that mount flags and options need to be exactly the same
132 * as they were when the filesystem was mounted, except for the desired
133 * changes. So we reconstruct both here and adjust them for the later
134 * remount call too. */
135
136 r = mnt_fs_get_propagation(fs, &remount_flags);
137 if (r < 0) {
138 log_warning_errno(r, "mnt_fs_get_propagation() failed for %s, ignoring: %m", path);
139 continue;
140 }
141
142 r = mount_option_mangle(options, remount_flags, &remount_flags, &remount_options);
143 if (r < 0) {
144 log_warning_errno(r, "mount_option_mangle failed for %s, ignoring: %m", path);
145 continue;
146 }
147
148 /* MS_BIND is special. If it is provided it will only make the mount-point
149 * read-only. If left out, the super block itself is remounted, which we want. */
150 remount_flags = (remount_flags|MS_REMOUNT|MS_RDONLY) & ~MS_BIND;
151 }
152
153 m = new0(MountPoint, 1);
154 if (!m)
155 return log_oom();
156
157 free_and_replace(m->path, p);
158 free_and_replace(m->remount_options, remount_options);
159 m->remount_flags = remount_flags;
160 m->try_remount_ro = try_remount_ro;
161
162 LIST_PREPEND(mount_point, *head, m);
163 }
164
165 return 0;
166 }
167
168 int swap_list_get(const char *swaps, MountPoint **head) {
169 _cleanup_(mnt_free_tablep) struct libmnt_table *t = NULL;
170 _cleanup_(mnt_free_iterp) struct libmnt_iter *i = NULL;
171 int r;
172
173 assert(head);
174
175 t = mnt_new_table();
176 i = mnt_new_iter(MNT_ITER_FORWARD);
177 if (!t || !i)
178 return log_oom();
179
180 r = mnt_table_parse_swaps(t, swaps);
181 if (r < 0)
182 return log_error_errno(r, "Failed to parse %s: %m", swaps);
183
184 for (;;) {
185 struct libmnt_fs *fs;
186
187 MountPoint *swap;
188 const char *source;
189 _cleanup_free_ char *d = NULL;
190
191 r = mnt_table_next_fs(t, i, &fs);
192 if (r == 1)
193 break;
194 if (r < 0)
195 return log_error_errno(r, "Failed to get next entry from %s: %m", swaps);
196
197 source = mnt_fs_get_source(fs);
198 if (!source)
199 continue;
200
201 r = cunescape(source, UNESCAPE_RELAX, &d);
202 if (r < 0)
203 return r;
204
205 swap = new0(MountPoint, 1);
206 if (!swap)
207 return -ENOMEM;
208
209 free_and_replace(swap->path, d);
210 LIST_PREPEND(mount_point, *head, swap);
211 }
212
213 return 0;
214 }
215
216 static int loopback_list_get(MountPoint **head) {
217 _cleanup_(udev_enumerate_unrefp) struct udev_enumerate *e = NULL;
218 struct udev_list_entry *item = NULL, *first = NULL;
219 _cleanup_(udev_unrefp) struct udev *udev = NULL;
220 int r;
221
222 assert(head);
223
224 udev = udev_new();
225 if (!udev)
226 return -ENOMEM;
227
228 e = udev_enumerate_new(udev);
229 if (!e)
230 return -ENOMEM;
231
232 r = udev_enumerate_add_match_subsystem(e, "block");
233 if (r < 0)
234 return r;
235
236 r = udev_enumerate_add_match_sysname(e, "loop*");
237 if (r < 0)
238 return r;
239
240 r = udev_enumerate_add_match_sysattr(e, "loop/backing_file", NULL);
241 if (r < 0)
242 return r;
243
244 r = udev_enumerate_scan_devices(e);
245 if (r < 0)
246 return r;
247
248 first = udev_enumerate_get_list_entry(e);
249 udev_list_entry_foreach(item, first) {
250 _cleanup_(udev_device_unrefp) struct udev_device *d;
251 const char *dn;
252 _cleanup_free_ MountPoint *lb = NULL;
253
254 d = udev_device_new_from_syspath(udev, udev_list_entry_get_name(item));
255 if (!d)
256 return -ENOMEM;
257
258 dn = udev_device_get_devnode(d);
259 if (!dn)
260 continue;
261
262 lb = new0(MountPoint, 1);
263 if (!lb)
264 return -ENOMEM;
265
266 r = free_and_strdup(&lb->path, dn);
267 if (r < 0)
268 return r;
269
270 LIST_PREPEND(mount_point, *head, lb);
271 lb = NULL;
272 }
273
274 return 0;
275 }
276
277 static int dm_list_get(MountPoint **head) {
278 _cleanup_(udev_enumerate_unrefp) struct udev_enumerate *e = NULL;
279 struct udev_list_entry *item = NULL, *first = NULL;
280 _cleanup_(udev_unrefp) struct udev *udev = NULL;
281 int r;
282
283 assert(head);
284
285 udev = udev_new();
286 if (!udev)
287 return -ENOMEM;
288
289 e = udev_enumerate_new(udev);
290 if (!e)
291 return -ENOMEM;
292
293 r = udev_enumerate_add_match_subsystem(e, "block");
294 if (r < 0)
295 return r;
296
297 r = udev_enumerate_add_match_sysname(e, "dm-*");
298 if (r < 0)
299 return r;
300
301 r = udev_enumerate_scan_devices(e);
302 if (r < 0)
303 return r;
304
305 first = udev_enumerate_get_list_entry(e);
306 udev_list_entry_foreach(item, first) {
307 _cleanup_(udev_device_unrefp) struct udev_device *d;
308 dev_t devnum;
309 const char *dn;
310 _cleanup_free_ MountPoint *m = NULL;
311
312 d = udev_device_new_from_syspath(udev, udev_list_entry_get_name(item));
313 if (!d)
314 return -ENOMEM;
315
316 devnum = udev_device_get_devnum(d);
317 dn = udev_device_get_devnode(d);
318 if (major(devnum) == 0 || !dn)
319 continue;
320
321 m = new0(MountPoint, 1);
322 if (!m)
323 return -ENOMEM;
324
325 m->devnum = devnum;
326 r = free_and_strdup(&m->path, dn);
327 if (r < 0)
328 return r;
329
330 LIST_PREPEND(mount_point, *head, m);
331 m = NULL;
332 }
333
334 return 0;
335 }
336
337 static int delete_loopback(const char *device) {
338 _cleanup_close_ int fd = -1;
339 int r;
340
341 assert(device);
342
343 fd = open(device, O_RDONLY|O_CLOEXEC);
344 if (fd < 0)
345 return errno == ENOENT ? 0 : -errno;
346
347 r = ioctl(fd, LOOP_CLR_FD, 0);
348 if (r >= 0)
349 return 1;
350
351 /* ENXIO: not bound, so no error */
352 if (errno == ENXIO)
353 return 0;
354
355 return -errno;
356 }
357
358 static int delete_dm(dev_t devnum) {
359
360 struct dm_ioctl dm = {
361 .version = {
362 DM_VERSION_MAJOR,
363 DM_VERSION_MINOR,
364 DM_VERSION_PATCHLEVEL
365 },
366 .data_size = sizeof(dm),
367 .dev = devnum,
368 };
369
370 _cleanup_close_ int fd = -1;
371
372 assert(major(devnum) != 0);
373
374 fd = open("/dev/mapper/control", O_RDWR|O_CLOEXEC);
375 if (fd < 0)
376 return -errno;
377
378 if (ioctl(fd, DM_DEV_REMOVE, &dm) < 0)
379 return -errno;
380
381 return 0;
382 }
383
384 static bool nonunmountable_path(const char *path) {
385 return path_equal(path, "/")
386 #if ! HAVE_SPLIT_USR
387 || path_equal(path, "/usr")
388 #endif
389 || path_startswith(path, "/run/initramfs");
390 }
391
392 static int remount_with_timeout(MountPoint *m, int umount_log_level) {
393 pid_t pid;
394 int r;
395
396 BLOCK_SIGNALS(SIGCHLD);
397
398 assert(m);
399
400 /* Due to the possiblity of a remount operation hanging, we
401 * fork a child process and set a timeout. If the timeout
402 * lapses, the assumption is that that particular remount
403 * failed. */
404 r = safe_fork("(sd-remount)", FORK_RESET_SIGNALS|FORK_CLOSE_ALL_FDS|FORK_LOG|FORK_REOPEN_LOG, &pid);
405 if (r < 0)
406 return r;
407 if (r == 0) {
408 log_info("Remounting '%s' read-only in with options '%s'.", m->path, m->remount_options);
409
410 /* Start the mount operation here in the child */
411 r = mount(NULL, m->path, NULL, m->remount_flags, m->remount_options);
412 if (r < 0)
413 log_full_errno(umount_log_level, errno, "Failed to remount '%s' read-only: %m", m->path);
414
415 _exit(r < 0 ? EXIT_FAILURE : EXIT_SUCCESS);
416 }
417
418 r = wait_for_terminate_with_timeout(pid, DEFAULT_TIMEOUT_USEC);
419 if (r == -ETIMEDOUT) {
420 log_error_errno(r, "Remounting '%s' timed out, issuing SIGKILL to PID " PID_FMT ".", m->path, pid);
421 (void) kill(pid, SIGKILL);
422 } else if (r == -EPROTO)
423 log_debug_errno(r, "Remounting '%s' failed abnormally, child process " PID_FMT " aborted or exited non-zero.", m->path, pid);
424 else if (r < 0)
425 log_error_errno(r, "Remounting '%s' failed unexpectedly, couldn't wait for child process " PID_FMT ": %m", m->path, pid);
426
427 return r;
428 }
429
430 static int umount_with_timeout(MountPoint *m, int umount_log_level) {
431 pid_t pid;
432 int r;
433
434 BLOCK_SIGNALS(SIGCHLD);
435
436 assert(m);
437
438 /* Due to the possiblity of a umount operation hanging, we
439 * fork a child process and set a timeout. If the timeout
440 * lapses, the assumption is that that particular umount
441 * failed. */
442 r = safe_fork("(sd-umount)", FORK_RESET_SIGNALS|FORK_CLOSE_ALL_FDS|FORK_LOG|FORK_REOPEN_LOG, &pid);
443 if (r < 0)
444 return r;
445 if (r == 0) {
446 log_info("Unmounting '%s'.", m->path);
447
448 /* Start the mount operation here in the child Using MNT_FORCE
449 * causes some filesystems (e.g. FUSE and NFS and other network
450 * filesystems) to abort any pending requests and return -EIO
451 * rather than blocking indefinitely. If the filesysten is
452 * "busy", this may allow processes to die, thus making the
453 * filesystem less busy so the unmount might succeed (rather
454 * then return EBUSY).*/
455 r = umount2(m->path, MNT_FORCE);
456 if (r < 0)
457 log_full_errno(umount_log_level, errno, "Failed to unmount %s: %m", m->path);
458
459 _exit(r < 0 ? EXIT_FAILURE : EXIT_SUCCESS);
460 }
461
462 r = wait_for_terminate_with_timeout(pid, DEFAULT_TIMEOUT_USEC);
463 if (r == -ETIMEDOUT) {
464 log_error_errno(r, "Unmounting '%s' timed out, issuing SIGKILL to PID " PID_FMT ".", m->path, pid);
465 (void) kill(pid, SIGKILL);
466 } else if (r == -EPROTO)
467 log_debug_errno(r, "Unmounting '%s' failed abnormally, child process " PID_FMT " aborted or exited non-zero.", m->path, pid);
468 else if (r < 0)
469 log_error_errno(r, "Unmounting '%s' failed unexpectedly, couldn't wait for child process " PID_FMT ": %m", m->path, pid);
470
471 return r;
472 }
473
474 /* This includes remounting readonly, which changes the kernel mount options.
475 * Therefore the list passed to this function is invalidated, and should not be reused. */
476 static int mount_points_list_umount(MountPoint **head, bool *changed, int umount_log_level) {
477 MountPoint *m;
478 int n_failed = 0;
479
480 assert(head);
481 assert(changed);
482
483 LIST_FOREACH(mount_point, m, *head) {
484 if (m->try_remount_ro) {
485 /* We always try to remount directories
486 * read-only first, before we go on and umount
487 * them.
488 *
489 * Mount points can be stacked. If a mount
490 * point is stacked below / or /usr, we
491 * cannot umount or remount it directly,
492 * since there is no way to refer to the
493 * underlying mount. There's nothing we can do
494 * about it for the general case, but we can
495 * do something about it if it is aliased
496 * somehwere else via a bind mount. If we
497 * explicitly remount the super block of that
498 * alias read-only we hence should be
499 * relatively safe regarding keeping a dirty fs
500 * we cannot otherwise see.
501 *
502 * Since the remount can hang in the instance of
503 * remote filesystems, we remount asynchronously
504 * and skip the subsequent umount if it fails. */
505 if (remount_with_timeout(m, umount_log_level) < 0) {
506 /* Remount failed, but try unmounting anyway,
507 * unless this is a mount point we want to skip. */
508 if (nonunmountable_path(m->path)) {
509 n_failed++;
510 continue;
511 }
512 }
513 }
514
515 /* Skip / and /usr since we cannot unmount that
516 * anyway, since we are running from it. They have
517 * already been remounted ro. */
518 if (nonunmountable_path(m->path))
519 continue;
520
521 /* Trying to umount */
522 if (umount_with_timeout(m, umount_log_level) < 0)
523 n_failed++;
524 else
525 *changed = true;
526 }
527
528 return n_failed;
529 }
530
531 static int swap_points_list_off(MountPoint **head, bool *changed) {
532 MountPoint *m, *n;
533 int n_failed = 0;
534
535 assert(head);
536 assert(changed);
537
538 LIST_FOREACH_SAFE(mount_point, m, n, *head) {
539 log_info("Deactivating swap %s.", m->path);
540 if (swapoff(m->path) == 0) {
541 *changed = true;
542 mount_point_free(head, m);
543 } else {
544 log_warning_errno(errno, "Could not deactivate swap %s: %m", m->path);
545 n_failed++;
546 }
547 }
548
549 return n_failed;
550 }
551
552 static int loopback_points_list_detach(MountPoint **head, bool *changed, int umount_log_level) {
553 MountPoint *m, *n;
554 int n_failed = 0, k;
555 struct stat root_st;
556
557 assert(head);
558 assert(changed);
559
560 k = lstat("/", &root_st);
561
562 LIST_FOREACH_SAFE(mount_point, m, n, *head) {
563 int r;
564 struct stat loopback_st;
565
566 if (k >= 0 &&
567 major(root_st.st_dev) != 0 &&
568 lstat(m->path, &loopback_st) >= 0 &&
569 root_st.st_dev == loopback_st.st_rdev) {
570 n_failed++;
571 continue;
572 }
573
574 log_info("Detaching loopback %s.", m->path);
575 r = delete_loopback(m->path);
576 if (r >= 0) {
577 if (r > 0)
578 *changed = true;
579
580 mount_point_free(head, m);
581 } else {
582 log_full_errno(umount_log_level, errno, "Could not detach loopback %s: %m", m->path);
583 n_failed++;
584 }
585 }
586
587 return n_failed;
588 }
589
590 static int dm_points_list_detach(MountPoint **head, bool *changed, int umount_log_level) {
591 MountPoint *m, *n;
592 int n_failed = 0, r;
593 dev_t rootdev;
594
595 assert(head);
596 assert(changed);
597
598 r = get_block_device("/", &rootdev);
599 if (r <= 0)
600 rootdev = 0;
601
602 LIST_FOREACH_SAFE(mount_point, m, n, *head) {
603
604 if (major(rootdev) != 0 && rootdev == m->devnum) {
605 n_failed ++;
606 continue;
607 }
608
609 log_info("Detaching DM %u:%u.", major(m->devnum), minor(m->devnum));
610 r = delete_dm(m->devnum);
611 if (r >= 0) {
612 *changed = true;
613 mount_point_free(head, m);
614 } else {
615 log_full_errno(umount_log_level, errno, "Could not detach DM %s: %m", m->path);
616 n_failed++;
617 }
618 }
619
620 return n_failed;
621 }
622
623 static int umount_all_once(bool *changed, int umount_log_level) {
624 int r;
625 _cleanup_(mount_points_list_free) LIST_HEAD(MountPoint, mp_list_head);
626
627 assert(changed);
628
629 LIST_HEAD_INIT(mp_list_head);
630 r = mount_points_list_get(NULL, &mp_list_head);
631 if (r < 0)
632 return r;
633
634 return mount_points_list_umount(&mp_list_head, changed, umount_log_level);
635 }
636
637 int umount_all(bool *changed, int umount_log_level) {
638 bool umount_changed;
639 int r;
640
641 assert(changed);
642
643 /* Retry umount, until nothing can be umounted anymore. Mounts are
644 * processed in order, newest first. The retries are needed when
645 * an old mount has been moved, to a path inside a newer mount. */
646 do {
647 umount_changed = false;
648
649 r = umount_all_once(&umount_changed, umount_log_level);
650 if (umount_changed)
651 *changed = true;
652 } while (umount_changed);
653
654 return r;
655 }
656
657 int swapoff_all(bool *changed) {
658 _cleanup_(mount_points_list_free) LIST_HEAD(MountPoint, swap_list_head);
659 int r;
660
661 assert(changed);
662
663 LIST_HEAD_INIT(swap_list_head);
664
665 r = swap_list_get(NULL, &swap_list_head);
666 if (r < 0)
667 return r;
668
669 return swap_points_list_off(&swap_list_head, changed);
670 }
671
672 int loopback_detach_all(bool *changed, int umount_log_level) {
673 _cleanup_(mount_points_list_free) LIST_HEAD(MountPoint, loopback_list_head);
674 int r;
675
676 assert(changed);
677
678 LIST_HEAD_INIT(loopback_list_head);
679
680 r = loopback_list_get(&loopback_list_head);
681 if (r < 0)
682 return r;
683
684 return loopback_points_list_detach(&loopback_list_head, changed, umount_log_level);
685 }
686
687 int dm_detach_all(bool *changed, int umount_log_level) {
688 _cleanup_(mount_points_list_free) LIST_HEAD(MountPoint, dm_list_head);
689 int r;
690
691 assert(changed);
692
693 LIST_HEAD_INIT(dm_list_head);
694
695 r = dm_list_get(&dm_list_head);
696 if (r < 0)
697 return r;
698
699 return dm_points_list_detach(&dm_list_head, changed, umount_log_level);
700 }