]> git.ipfire.org Git - thirdparty/systemd.git/blob - src/core/umount.c
7a72515b2d4f543c2006529014b3ec226bf22020
[thirdparty/systemd.git] / src / core / umount.c
1 /* SPDX-License-Identifier: LGPL-2.1+ */
2 /***
3 Copyright 2010 ProFUSION embedded systems
4 ***/
5
6 #include <errno.h>
7 #include <fcntl.h>
8 #include <linux/loop.h>
9 #include <string.h>
10 #include <sys/mount.h>
11 #include <sys/swap.h>
12
13 /* This needs to be after sys/mount.h :( */
14 #include <libmount.h>
15
16 #include "libudev.h"
17
18 #include "alloc-util.h"
19 #include "blockdev-util.h"
20 #include "def.h"
21 #include "escape.h"
22 #include "fd-util.h"
23 #include "fstab-util.h"
24 #include "linux-3.13/dm-ioctl.h"
25 #include "mount-setup.h"
26 #include "mount-util.h"
27 #include "path-util.h"
28 #include "process-util.h"
29 #include "signal-util.h"
30 #include "string-util.h"
31 #include "udev-util.h"
32 #include "umount.h"
33 #include "util.h"
34 #include "virt.h"
35
36 DEFINE_TRIVIAL_CLEANUP_FUNC(struct libmnt_table*, mnt_free_table);
37 DEFINE_TRIVIAL_CLEANUP_FUNC(struct libmnt_iter*, mnt_free_iter);
38
39 static void mount_point_free(MountPoint **head, MountPoint *m) {
40 assert(head);
41 assert(m);
42
43 LIST_REMOVE(mount_point, *head, m);
44
45 free(m->path);
46 free(m->remount_options);
47 free(m);
48 }
49
50 void mount_points_list_free(MountPoint **head) {
51 assert(head);
52
53 while (*head)
54 mount_point_free(head, *head);
55 }
56
57 int mount_points_list_get(const char *mountinfo, MountPoint **head) {
58 _cleanup_(mnt_free_tablep) struct libmnt_table *t = NULL;
59 _cleanup_(mnt_free_iterp) struct libmnt_iter *i = NULL;
60 int r;
61
62 assert(head);
63
64 t = mnt_new_table();
65 i = mnt_new_iter(MNT_ITER_FORWARD);
66 if (!t || !i)
67 return log_oom();
68
69 r = mnt_table_parse_mtab(t, mountinfo);
70 if (r < 0)
71 return log_error_errno(r, "Failed to parse %s: %m", mountinfo);
72
73 for (;;) {
74 struct libmnt_fs *fs;
75 const char *path, *options, *fstype;
76 _cleanup_free_ char *p = NULL;
77 unsigned long remount_flags = 0u;
78 _cleanup_free_ char *remount_options = NULL;
79 bool try_remount_ro;
80 MountPoint *m;
81
82 r = mnt_table_next_fs(t, i, &fs);
83 if (r == 1)
84 break;
85 if (r < 0)
86 return log_error_errno(r, "Failed to get next entry from %s: %m", mountinfo);
87
88 path = mnt_fs_get_target(fs);
89 if (!path)
90 continue;
91
92 if (cunescape(path, UNESCAPE_RELAX, &p) < 0)
93 return log_oom();
94
95 options = mnt_fs_get_options(fs);
96 fstype = mnt_fs_get_fstype(fs);
97
98 /* Ignore mount points we can't unmount because they
99 * are API or because we are keeping them open (like
100 * /dev/console). Also, ignore all mounts below API
101 * file systems, since they are likely virtual too,
102 * and hence not worth spending time on. Also, in
103 * unprivileged containers we might lack the rights to
104 * unmount these things, hence don't bother. */
105 if (mount_point_is_api(p) ||
106 mount_point_ignore(p) ||
107 path_startswith(p, "/dev") ||
108 path_startswith(p, "/sys") ||
109 path_startswith(p, "/proc"))
110 continue;
111
112 /* If we are in a container, don't attempt to
113 * read-only mount anything as that brings no real
114 * benefits, but might confuse the host, as we remount
115 * the superblock here, not the bind mount.
116 *
117 * If the filesystem is a network fs, also skip the
118 * remount. It brings no value (we cannot leave
119 * a "dirty fs") and could hang if the network is down.
120 * Note that umount2() is more careful and will not
121 * hang because of the network being down. */
122 try_remount_ro = detect_container() <= 0 &&
123 !fstype_is_network(fstype) &&
124 !fstype_is_api_vfs(fstype) &&
125 !fstype_is_ro(fstype) &&
126 !fstab_test_yes_no_option(options, "ro\0rw\0");
127
128 if (try_remount_ro) {
129 /* mount(2) states that mount flags and options need to be exactly the same
130 * as they were when the filesystem was mounted, except for the desired
131 * changes. So we reconstruct both here and adjust them for the later
132 * remount call too. */
133
134 r = mnt_fs_get_propagation(fs, &remount_flags);
135 if (r < 0) {
136 log_warning_errno(r, "mnt_fs_get_propagation() failed for %s, ignoring: %m", path);
137 continue;
138 }
139
140 r = mount_option_mangle(options, remount_flags, &remount_flags, &remount_options);
141 if (r < 0) {
142 log_warning_errno(r, "mount_option_mangle failed for %s, ignoring: %m", path);
143 continue;
144 }
145
146 /* MS_BIND is special. If it is provided it will only make the mount-point
147 * read-only. If left out, the super block itself is remounted, which we want. */
148 remount_flags = (remount_flags|MS_REMOUNT|MS_RDONLY) & ~MS_BIND;
149 }
150
151 m = new0(MountPoint, 1);
152 if (!m)
153 return log_oom();
154
155 free_and_replace(m->path, p);
156 free_and_replace(m->remount_options, remount_options);
157 m->remount_flags = remount_flags;
158 m->try_remount_ro = try_remount_ro;
159
160 LIST_PREPEND(mount_point, *head, m);
161 }
162
163 return 0;
164 }
165
166 int swap_list_get(const char *swaps, MountPoint **head) {
167 _cleanup_(mnt_free_tablep) struct libmnt_table *t = NULL;
168 _cleanup_(mnt_free_iterp) struct libmnt_iter *i = NULL;
169 int r;
170
171 assert(head);
172
173 t = mnt_new_table();
174 i = mnt_new_iter(MNT_ITER_FORWARD);
175 if (!t || !i)
176 return log_oom();
177
178 r = mnt_table_parse_swaps(t, swaps);
179 if (r < 0)
180 return log_error_errno(r, "Failed to parse %s: %m", swaps);
181
182 for (;;) {
183 struct libmnt_fs *fs;
184
185 MountPoint *swap;
186 const char *source;
187 _cleanup_free_ char *d = NULL;
188
189 r = mnt_table_next_fs(t, i, &fs);
190 if (r == 1)
191 break;
192 if (r < 0)
193 return log_error_errno(r, "Failed to get next entry from %s: %m", swaps);
194
195 source = mnt_fs_get_source(fs);
196 if (!source)
197 continue;
198
199 r = cunescape(source, UNESCAPE_RELAX, &d);
200 if (r < 0)
201 return r;
202
203 swap = new0(MountPoint, 1);
204 if (!swap)
205 return -ENOMEM;
206
207 free_and_replace(swap->path, d);
208 LIST_PREPEND(mount_point, *head, swap);
209 }
210
211 return 0;
212 }
213
214 static int loopback_list_get(MountPoint **head) {
215 _cleanup_(udev_enumerate_unrefp) struct udev_enumerate *e = NULL;
216 struct udev_list_entry *item = NULL, *first = NULL;
217 _cleanup_(udev_unrefp) struct udev *udev = NULL;
218 int r;
219
220 assert(head);
221
222 udev = udev_new();
223 if (!udev)
224 return -ENOMEM;
225
226 e = udev_enumerate_new(udev);
227 if (!e)
228 return -ENOMEM;
229
230 r = udev_enumerate_add_match_subsystem(e, "block");
231 if (r < 0)
232 return r;
233
234 r = udev_enumerate_add_match_sysname(e, "loop*");
235 if (r < 0)
236 return r;
237
238 r = udev_enumerate_add_match_sysattr(e, "loop/backing_file", NULL);
239 if (r < 0)
240 return r;
241
242 r = udev_enumerate_scan_devices(e);
243 if (r < 0)
244 return r;
245
246 first = udev_enumerate_get_list_entry(e);
247 udev_list_entry_foreach(item, first) {
248 _cleanup_(udev_device_unrefp) struct udev_device *d;
249 const char *dn;
250 _cleanup_free_ MountPoint *lb = NULL;
251
252 d = udev_device_new_from_syspath(udev, udev_list_entry_get_name(item));
253 if (!d)
254 return -ENOMEM;
255
256 dn = udev_device_get_devnode(d);
257 if (!dn)
258 continue;
259
260 lb = new0(MountPoint, 1);
261 if (!lb)
262 return -ENOMEM;
263
264 r = free_and_strdup(&lb->path, dn);
265 if (r < 0)
266 return r;
267
268 LIST_PREPEND(mount_point, *head, lb);
269 lb = NULL;
270 }
271
272 return 0;
273 }
274
275 static int dm_list_get(MountPoint **head) {
276 _cleanup_(udev_enumerate_unrefp) struct udev_enumerate *e = NULL;
277 struct udev_list_entry *item = NULL, *first = NULL;
278 _cleanup_(udev_unrefp) struct udev *udev = NULL;
279 int r;
280
281 assert(head);
282
283 udev = udev_new();
284 if (!udev)
285 return -ENOMEM;
286
287 e = udev_enumerate_new(udev);
288 if (!e)
289 return -ENOMEM;
290
291 r = udev_enumerate_add_match_subsystem(e, "block");
292 if (r < 0)
293 return r;
294
295 r = udev_enumerate_add_match_sysname(e, "dm-*");
296 if (r < 0)
297 return r;
298
299 r = udev_enumerate_scan_devices(e);
300 if (r < 0)
301 return r;
302
303 first = udev_enumerate_get_list_entry(e);
304 udev_list_entry_foreach(item, first) {
305 _cleanup_(udev_device_unrefp) struct udev_device *d;
306 dev_t devnum;
307 const char *dn;
308 _cleanup_free_ MountPoint *m = NULL;
309
310 d = udev_device_new_from_syspath(udev, udev_list_entry_get_name(item));
311 if (!d)
312 return -ENOMEM;
313
314 devnum = udev_device_get_devnum(d);
315 dn = udev_device_get_devnode(d);
316 if (major(devnum) == 0 || !dn)
317 continue;
318
319 m = new0(MountPoint, 1);
320 if (!m)
321 return -ENOMEM;
322
323 m->devnum = devnum;
324 r = free_and_strdup(&m->path, dn);
325 if (r < 0)
326 return r;
327
328 LIST_PREPEND(mount_point, *head, m);
329 m = NULL;
330 }
331
332 return 0;
333 }
334
335 static int delete_loopback(const char *device) {
336 _cleanup_close_ int fd = -1;
337 int r;
338
339 assert(device);
340
341 fd = open(device, O_RDONLY|O_CLOEXEC);
342 if (fd < 0)
343 return errno == ENOENT ? 0 : -errno;
344
345 r = ioctl(fd, LOOP_CLR_FD, 0);
346 if (r >= 0)
347 return 1;
348
349 /* ENXIO: not bound, so no error */
350 if (errno == ENXIO)
351 return 0;
352
353 return -errno;
354 }
355
356 static int delete_dm(dev_t devnum) {
357
358 struct dm_ioctl dm = {
359 .version = {
360 DM_VERSION_MAJOR,
361 DM_VERSION_MINOR,
362 DM_VERSION_PATCHLEVEL
363 },
364 .data_size = sizeof(dm),
365 .dev = devnum,
366 };
367
368 _cleanup_close_ int fd = -1;
369
370 assert(major(devnum) != 0);
371
372 fd = open("/dev/mapper/control", O_RDWR|O_CLOEXEC);
373 if (fd < 0)
374 return -errno;
375
376 if (ioctl(fd, DM_DEV_REMOVE, &dm) < 0)
377 return -errno;
378
379 return 0;
380 }
381
382 static bool nonunmountable_path(const char *path) {
383 return path_equal(path, "/")
384 #if ! HAVE_SPLIT_USR
385 || path_equal(path, "/usr")
386 #endif
387 || path_startswith(path, "/run/initramfs");
388 }
389
390 static int remount_with_timeout(MountPoint *m, int umount_log_level) {
391 pid_t pid;
392 int r;
393
394 BLOCK_SIGNALS(SIGCHLD);
395
396 assert(m);
397
398 /* Due to the possiblity of a remount operation hanging, we
399 * fork a child process and set a timeout. If the timeout
400 * lapses, the assumption is that that particular remount
401 * failed. */
402 r = safe_fork("(sd-remount)", FORK_RESET_SIGNALS|FORK_CLOSE_ALL_FDS|FORK_LOG|FORK_REOPEN_LOG, &pid);
403 if (r < 0)
404 return r;
405 if (r == 0) {
406 log_info("Remounting '%s' read-only in with options '%s'.", m->path, m->remount_options);
407
408 /* Start the mount operation here in the child */
409 r = mount(NULL, m->path, NULL, m->remount_flags, m->remount_options);
410 if (r < 0)
411 log_full_errno(umount_log_level, errno, "Failed to remount '%s' read-only: %m", m->path);
412
413 _exit(r < 0 ? EXIT_FAILURE : EXIT_SUCCESS);
414 }
415
416 r = wait_for_terminate_with_timeout(pid, DEFAULT_TIMEOUT_USEC);
417 if (r == -ETIMEDOUT) {
418 log_error_errno(r, "Remounting '%s' timed out, issuing SIGKILL to PID " PID_FMT ".", m->path, pid);
419 (void) kill(pid, SIGKILL);
420 } else if (r == -EPROTO)
421 log_debug_errno(r, "Remounting '%s' failed abnormally, child process " PID_FMT " aborted or exited non-zero.", m->path, pid);
422 else if (r < 0)
423 log_error_errno(r, "Remounting '%s' failed unexpectedly, couldn't wait for child process " PID_FMT ": %m", m->path, pid);
424
425 return r;
426 }
427
428 static int umount_with_timeout(MountPoint *m, int umount_log_level) {
429 pid_t pid;
430 int r;
431
432 BLOCK_SIGNALS(SIGCHLD);
433
434 assert(m);
435
436 /* Due to the possiblity of a umount operation hanging, we
437 * fork a child process and set a timeout. If the timeout
438 * lapses, the assumption is that that particular umount
439 * failed. */
440 r = safe_fork("(sd-umount)", FORK_RESET_SIGNALS|FORK_CLOSE_ALL_FDS|FORK_LOG|FORK_REOPEN_LOG, &pid);
441 if (r < 0)
442 return r;
443 if (r == 0) {
444 log_info("Unmounting '%s'.", m->path);
445
446 /* Start the mount operation here in the child Using MNT_FORCE
447 * causes some filesystems (e.g. FUSE and NFS and other network
448 * filesystems) to abort any pending requests and return -EIO
449 * rather than blocking indefinitely. If the filesysten is
450 * "busy", this may allow processes to die, thus making the
451 * filesystem less busy so the unmount might succeed (rather
452 * then return EBUSY).*/
453 r = umount2(m->path, MNT_FORCE);
454 if (r < 0)
455 log_full_errno(umount_log_level, errno, "Failed to unmount %s: %m", m->path);
456
457 _exit(r < 0 ? EXIT_FAILURE : EXIT_SUCCESS);
458 }
459
460 r = wait_for_terminate_with_timeout(pid, DEFAULT_TIMEOUT_USEC);
461 if (r == -ETIMEDOUT) {
462 log_error_errno(r, "Unmounting '%s' timed out, issuing SIGKILL to PID " PID_FMT ".", m->path, pid);
463 (void) kill(pid, SIGKILL);
464 } else if (r == -EPROTO)
465 log_debug_errno(r, "Unmounting '%s' failed abnormally, child process " PID_FMT " aborted or exited non-zero.", m->path, pid);
466 else if (r < 0)
467 log_error_errno(r, "Unmounting '%s' failed unexpectedly, couldn't wait for child process " PID_FMT ": %m", m->path, pid);
468
469 return r;
470 }
471
472 /* This includes remounting readonly, which changes the kernel mount options.
473 * Therefore the list passed to this function is invalidated, and should not be reused. */
474 static int mount_points_list_umount(MountPoint **head, bool *changed, int umount_log_level) {
475 MountPoint *m;
476 int n_failed = 0;
477
478 assert(head);
479 assert(changed);
480
481 LIST_FOREACH(mount_point, m, *head) {
482 if (m->try_remount_ro) {
483 /* We always try to remount directories
484 * read-only first, before we go on and umount
485 * them.
486 *
487 * Mount points can be stacked. If a mount
488 * point is stacked below / or /usr, we
489 * cannot umount or remount it directly,
490 * since there is no way to refer to the
491 * underlying mount. There's nothing we can do
492 * about it for the general case, but we can
493 * do something about it if it is aliased
494 * somehwere else via a bind mount. If we
495 * explicitly remount the super block of that
496 * alias read-only we hence should be
497 * relatively safe regarding keeping a dirty fs
498 * we cannot otherwise see.
499 *
500 * Since the remount can hang in the instance of
501 * remote filesystems, we remount asynchronously
502 * and skip the subsequent umount if it fails. */
503 if (remount_with_timeout(m, umount_log_level) < 0) {
504 /* Remount failed, but try unmounting anyway,
505 * unless this is a mount point we want to skip. */
506 if (nonunmountable_path(m->path)) {
507 n_failed++;
508 continue;
509 }
510 }
511 }
512
513 /* Skip / and /usr since we cannot unmount that
514 * anyway, since we are running from it. They have
515 * already been remounted ro. */
516 if (nonunmountable_path(m->path))
517 continue;
518
519 /* Trying to umount */
520 if (umount_with_timeout(m, umount_log_level) < 0)
521 n_failed++;
522 else
523 *changed = true;
524 }
525
526 return n_failed;
527 }
528
529 static int swap_points_list_off(MountPoint **head, bool *changed) {
530 MountPoint *m, *n;
531 int n_failed = 0;
532
533 assert(head);
534 assert(changed);
535
536 LIST_FOREACH_SAFE(mount_point, m, n, *head) {
537 log_info("Deactivating swap %s.", m->path);
538 if (swapoff(m->path) == 0) {
539 *changed = true;
540 mount_point_free(head, m);
541 } else {
542 log_warning_errno(errno, "Could not deactivate swap %s: %m", m->path);
543 n_failed++;
544 }
545 }
546
547 return n_failed;
548 }
549
550 static int loopback_points_list_detach(MountPoint **head, bool *changed, int umount_log_level) {
551 MountPoint *m, *n;
552 int n_failed = 0, k;
553 struct stat root_st;
554
555 assert(head);
556 assert(changed);
557
558 k = lstat("/", &root_st);
559
560 LIST_FOREACH_SAFE(mount_point, m, n, *head) {
561 int r;
562 struct stat loopback_st;
563
564 if (k >= 0 &&
565 major(root_st.st_dev) != 0 &&
566 lstat(m->path, &loopback_st) >= 0 &&
567 root_st.st_dev == loopback_st.st_rdev) {
568 n_failed++;
569 continue;
570 }
571
572 log_info("Detaching loopback %s.", m->path);
573 r = delete_loopback(m->path);
574 if (r >= 0) {
575 if (r > 0)
576 *changed = true;
577
578 mount_point_free(head, m);
579 } else {
580 log_full_errno(umount_log_level, errno, "Could not detach loopback %s: %m", m->path);
581 n_failed++;
582 }
583 }
584
585 return n_failed;
586 }
587
588 static int dm_points_list_detach(MountPoint **head, bool *changed, int umount_log_level) {
589 MountPoint *m, *n;
590 int n_failed = 0, r;
591 dev_t rootdev;
592
593 assert(head);
594 assert(changed);
595
596 r = get_block_device("/", &rootdev);
597 if (r <= 0)
598 rootdev = 0;
599
600 LIST_FOREACH_SAFE(mount_point, m, n, *head) {
601
602 if (major(rootdev) != 0 && rootdev == m->devnum) {
603 n_failed ++;
604 continue;
605 }
606
607 log_info("Detaching DM %u:%u.", major(m->devnum), minor(m->devnum));
608 r = delete_dm(m->devnum);
609 if (r >= 0) {
610 *changed = true;
611 mount_point_free(head, m);
612 } else {
613 log_full_errno(umount_log_level, errno, "Could not detach DM %s: %m", m->path);
614 n_failed++;
615 }
616 }
617
618 return n_failed;
619 }
620
621 static int umount_all_once(bool *changed, int umount_log_level) {
622 int r;
623 _cleanup_(mount_points_list_free) LIST_HEAD(MountPoint, mp_list_head);
624
625 assert(changed);
626
627 LIST_HEAD_INIT(mp_list_head);
628 r = mount_points_list_get(NULL, &mp_list_head);
629 if (r < 0)
630 return r;
631
632 return mount_points_list_umount(&mp_list_head, changed, umount_log_level);
633 }
634
635 int umount_all(bool *changed, int umount_log_level) {
636 bool umount_changed;
637 int r;
638
639 assert(changed);
640
641 /* Retry umount, until nothing can be umounted anymore. Mounts are
642 * processed in order, newest first. The retries are needed when
643 * an old mount has been moved, to a path inside a newer mount. */
644 do {
645 umount_changed = false;
646
647 r = umount_all_once(&umount_changed, umount_log_level);
648 if (umount_changed)
649 *changed = true;
650 } while (umount_changed);
651
652 return r;
653 }
654
655 int swapoff_all(bool *changed) {
656 _cleanup_(mount_points_list_free) LIST_HEAD(MountPoint, swap_list_head);
657 int r;
658
659 assert(changed);
660
661 LIST_HEAD_INIT(swap_list_head);
662
663 r = swap_list_get(NULL, &swap_list_head);
664 if (r < 0)
665 return r;
666
667 return swap_points_list_off(&swap_list_head, changed);
668 }
669
670 int loopback_detach_all(bool *changed, int umount_log_level) {
671 _cleanup_(mount_points_list_free) LIST_HEAD(MountPoint, loopback_list_head);
672 int r;
673
674 assert(changed);
675
676 LIST_HEAD_INIT(loopback_list_head);
677
678 r = loopback_list_get(&loopback_list_head);
679 if (r < 0)
680 return r;
681
682 return loopback_points_list_detach(&loopback_list_head, changed, umount_log_level);
683 }
684
685 int dm_detach_all(bool *changed, int umount_log_level) {
686 _cleanup_(mount_points_list_free) LIST_HEAD(MountPoint, dm_list_head);
687 int r;
688
689 assert(changed);
690
691 LIST_HEAD_INIT(dm_list_head);
692
693 r = dm_list_get(&dm_list_head);
694 if (r < 0)
695 return r;
696
697 return dm_points_list_detach(&dm_list_head, changed, umount_log_level);
698 }