]> git.ipfire.org Git - thirdparty/systemd.git/blob - src/core/umount.c
Split out part of mount-util.c into mountpoint-util.c
[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 "sd-device.h"
17
18 #include "alloc-util.h"
19 #include "blockdev-util.h"
20 #include "def.h"
21 #include "device-util.h"
22 #include "escape.h"
23 #include "fd-util.h"
24 #include "fstab-util.h"
25 #include "linux-3.13/dm-ioctl.h"
26 #include "mount-setup.h"
27 #include "mount-util.h"
28 #include "mountpoint-util.h"
29 #include "path-util.h"
30 #include "process-util.h"
31 #include "signal-util.h"
32 #include "string-util.h"
33 #include "strv.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, *fstype;
78 _cleanup_free_ char *options = NULL;
79 _cleanup_free_ char *p = NULL;
80 unsigned long remount_flags = 0u;
81 _cleanup_free_ char *remount_options = NULL;
82 bool try_remount_ro;
83 MountPoint *m;
84
85 r = mnt_table_next_fs(t, i, &fs);
86 if (r == 1)
87 break;
88 if (r < 0)
89 return log_error_errno(r, "Failed to get next entry from %s: %m", mountinfo);
90
91 path = mnt_fs_get_target(fs);
92 if (!path)
93 continue;
94
95 if (cunescape(path, UNESCAPE_RELAX, &p) < 0)
96 return log_oom();
97
98 fstype = mnt_fs_get_fstype(fs);
99
100 /* Combine the generic VFS options with the FS-specific
101 * options. Duplicates are not a problem here, because the only
102 * options that should come up twice are typically ro/rw, which
103 * are turned into MS_RDONLY or the invertion of it.
104 *
105 * Even if there are duplicates later in mount_option_mangle()
106 * it shouldn't hurt anyways as they override each other.
107 */
108 if (!strextend_with_separator(&options, ",",
109 mnt_fs_get_vfs_options(fs),
110 NULL))
111 return log_oom();
112 if (!strextend_with_separator(&options, ",",
113 mnt_fs_get_fs_options(fs),
114 NULL))
115 return log_oom();
116
117 /* Ignore mount points we can't unmount because they
118 * are API or because we are keeping them open (like
119 * /dev/console). Also, ignore all mounts below API
120 * file systems, since they are likely virtual too,
121 * and hence not worth spending time on. Also, in
122 * unprivileged containers we might lack the rights to
123 * unmount these things, hence don't bother. */
124 if (mount_point_is_api(p) ||
125 mount_point_ignore(p) ||
126 PATH_STARTSWITH_SET(p, "/dev", "/sys", "/proc"))
127 continue;
128
129 /* If we are in a container, don't attempt to
130 * read-only mount anything as that brings no real
131 * benefits, but might confuse the host, as we remount
132 * the superblock here, not the bind mount.
133 *
134 * If the filesystem is a network fs, also skip the
135 * remount. It brings no value (we cannot leave
136 * a "dirty fs") and could hang if the network is down.
137 * Note that umount2() is more careful and will not
138 * hang because of the network being down. */
139 try_remount_ro = detect_container() <= 0 &&
140 !fstype_is_network(fstype) &&
141 !fstype_is_api_vfs(fstype) &&
142 !fstype_is_ro(fstype) &&
143 !fstab_test_yes_no_option(options, "ro\0rw\0");
144
145 if (try_remount_ro) {
146 /* mount(2) states that mount flags and options need to be exactly the same
147 * as they were when the filesystem was mounted, except for the desired
148 * changes. So we reconstruct both here and adjust them for the later
149 * remount call too. */
150
151 r = mnt_fs_get_propagation(fs, &remount_flags);
152 if (r < 0) {
153 log_warning_errno(r, "mnt_fs_get_propagation() failed for %s, ignoring: %m", path);
154 continue;
155 }
156
157 r = mount_option_mangle(options, remount_flags, &remount_flags, &remount_options);
158 if (r < 0) {
159 log_warning_errno(r, "mount_option_mangle failed for %s, ignoring: %m", path);
160 continue;
161 }
162
163 /* MS_BIND is special. If it is provided it will only make the mount-point
164 * read-only. If left out, the super block itself is remounted, which we want. */
165 remount_flags = (remount_flags|MS_REMOUNT|MS_RDONLY) & ~MS_BIND;
166 }
167
168 m = new0(MountPoint, 1);
169 if (!m)
170 return log_oom();
171
172 free_and_replace(m->path, p);
173 free_and_replace(m->remount_options, remount_options);
174 m->remount_flags = remount_flags;
175 m->try_remount_ro = try_remount_ro;
176
177 LIST_PREPEND(mount_point, *head, m);
178 }
179
180 return 0;
181 }
182
183 int swap_list_get(const char *swaps, MountPoint **head) {
184 _cleanup_(mnt_free_tablep) struct libmnt_table *t = NULL;
185 _cleanup_(mnt_free_iterp) struct libmnt_iter *i = NULL;
186 int r;
187
188 assert(head);
189
190 t = mnt_new_table();
191 i = mnt_new_iter(MNT_ITER_FORWARD);
192 if (!t || !i)
193 return log_oom();
194
195 r = mnt_table_parse_swaps(t, swaps);
196 if (r < 0)
197 return log_error_errno(r, "Failed to parse %s: %m", swaps);
198
199 for (;;) {
200 struct libmnt_fs *fs;
201
202 MountPoint *swap;
203 const char *source;
204 _cleanup_free_ char *d = NULL;
205
206 r = mnt_table_next_fs(t, i, &fs);
207 if (r == 1)
208 break;
209 if (r < 0)
210 return log_error_errno(r, "Failed to get next entry from %s: %m", swaps);
211
212 source = mnt_fs_get_source(fs);
213 if (!source)
214 continue;
215
216 r = cunescape(source, UNESCAPE_RELAX, &d);
217 if (r < 0)
218 return r;
219
220 swap = new0(MountPoint, 1);
221 if (!swap)
222 return -ENOMEM;
223
224 free_and_replace(swap->path, d);
225 LIST_PREPEND(mount_point, *head, swap);
226 }
227
228 return 0;
229 }
230
231 static int loopback_list_get(MountPoint **head) {
232 _cleanup_(sd_device_enumerator_unrefp) sd_device_enumerator *e = NULL;
233 sd_device *d;
234 int r;
235
236 assert(head);
237
238 r = sd_device_enumerator_new(&e);
239 if (r < 0)
240 return r;
241
242 r = sd_device_enumerator_allow_uninitialized(e);
243 if (r < 0)
244 return r;
245
246 r = sd_device_enumerator_add_match_subsystem(e, "block", true);
247 if (r < 0)
248 return r;
249
250 r = sd_device_enumerator_add_match_sysname(e, "loop*");
251 if (r < 0)
252 return r;
253
254 r = sd_device_enumerator_add_match_sysattr(e, "loop/backing_file", NULL, true);
255 if (r < 0)
256 return r;
257
258 FOREACH_DEVICE(e, d) {
259 _cleanup_free_ char *p = NULL;
260 const char *dn;
261 MountPoint *lb;
262
263 if (sd_device_get_devname(d, &dn) < 0)
264 continue;
265
266 p = strdup(dn);
267 if (!p)
268 return -ENOMEM;
269
270 lb = new(MountPoint, 1);
271 if (!lb)
272 return -ENOMEM;
273
274 *lb = (MountPoint) {
275 .path = TAKE_PTR(p),
276 };
277
278 LIST_PREPEND(mount_point, *head, lb);
279 }
280
281 return 0;
282 }
283
284 static int dm_list_get(MountPoint **head) {
285 _cleanup_(sd_device_enumerator_unrefp) sd_device_enumerator *e = NULL;
286 sd_device *d;
287 int r;
288
289 assert(head);
290
291 r = sd_device_enumerator_new(&e);
292 if (r < 0)
293 return r;
294
295 r = sd_device_enumerator_allow_uninitialized(e);
296 if (r < 0)
297 return r;
298
299 r = sd_device_enumerator_add_match_subsystem(e, "block", true);
300 if (r < 0)
301 return r;
302
303 r = sd_device_enumerator_add_match_sysname(e, "dm-*");
304 if (r < 0)
305 return r;
306
307 FOREACH_DEVICE(e, d) {
308 _cleanup_free_ char *p = NULL;
309 const char *dn;
310 MountPoint *m;
311 dev_t devnum;
312
313 if (sd_device_get_devnum(d, &devnum) < 0 ||
314 sd_device_get_devname(d, &dn) < 0)
315 continue;
316
317 p = strdup(dn);
318 if (!p)
319 return -ENOMEM;
320
321 m = new(MountPoint, 1);
322 if (!m)
323 return -ENOMEM;
324
325 *m = (MountPoint) {
326 .path = TAKE_PTR(p),
327 .devnum = devnum,
328 };
329
330 LIST_PREPEND(mount_point, *head, m);
331 }
332
333 return 0;
334 }
335
336 static int delete_loopback(const char *device) {
337 _cleanup_close_ int fd = -1;
338 int r;
339
340 assert(device);
341
342 fd = open(device, O_RDONLY|O_CLOEXEC);
343 if (fd < 0)
344 return errno == ENOENT ? 0 : -errno;
345
346 r = ioctl(fd, LOOP_CLR_FD, 0);
347 if (r >= 0)
348 return 1;
349
350 /* ENXIO: not bound, so no error */
351 if (errno == ENXIO)
352 return 0;
353
354 return -errno;
355 }
356
357 static int delete_dm(dev_t devnum) {
358
359 struct dm_ioctl dm = {
360 .version = {
361 DM_VERSION_MAJOR,
362 DM_VERSION_MINOR,
363 DM_VERSION_PATCHLEVEL
364 },
365 .data_size = sizeof(dm),
366 .dev = devnum,
367 };
368
369 _cleanup_close_ int fd = -1;
370
371 assert(major(devnum) != 0);
372
373 fd = open("/dev/mapper/control", O_RDWR|O_CLOEXEC);
374 if (fd < 0)
375 return -errno;
376
377 if (ioctl(fd, DM_DEV_REMOVE, &dm) < 0)
378 return -errno;
379
380 return 0;
381 }
382
383 static bool nonunmountable_path(const char *path) {
384 return path_equal(path, "/")
385 #if ! HAVE_SPLIT_USR
386 || path_equal(path, "/usr")
387 #endif
388 || path_startswith(path, "/run/initramfs");
389 }
390
391 static int remount_with_timeout(MountPoint *m, int umount_log_level) {
392 pid_t pid;
393 int r;
394
395 BLOCK_SIGNALS(SIGCHLD);
396
397 assert(m);
398
399 /* Due to the possiblity of a remount operation hanging, we
400 * fork a child process and set a timeout. If the timeout
401 * lapses, the assumption is that that particular remount
402 * failed. */
403 r = safe_fork("(sd-remount)", FORK_RESET_SIGNALS|FORK_CLOSE_ALL_FDS|FORK_LOG|FORK_REOPEN_LOG, &pid);
404 if (r < 0)
405 return r;
406 if (r == 0) {
407 log_info("Remounting '%s' read-only in with options '%s'.", m->path, m->remount_options);
408
409 /* Start the mount operation here in the child */
410 r = mount(NULL, m->path, NULL, m->remount_flags, m->remount_options);
411 if (r < 0)
412 log_full_errno(umount_log_level, errno, "Failed to remount '%s' read-only: %m", m->path);
413
414 _exit(r < 0 ? EXIT_FAILURE : EXIT_SUCCESS);
415 }
416
417 r = wait_for_terminate_with_timeout(pid, DEFAULT_TIMEOUT_USEC);
418 if (r == -ETIMEDOUT) {
419 log_error_errno(r, "Remounting '%s' timed out, issuing SIGKILL to PID " PID_FMT ".", m->path, pid);
420 (void) kill(pid, SIGKILL);
421 } else if (r == -EPROTO)
422 log_debug_errno(r, "Remounting '%s' failed abnormally, child process " PID_FMT " aborted or exited non-zero.", m->path, pid);
423 else if (r < 0)
424 log_error_errno(r, "Remounting '%s' failed unexpectedly, couldn't wait for child process " PID_FMT ": %m", m->path, pid);
425
426 return r;
427 }
428
429 static int umount_with_timeout(MountPoint *m, int umount_log_level) {
430 pid_t pid;
431 int r;
432
433 BLOCK_SIGNALS(SIGCHLD);
434
435 assert(m);
436
437 /* Due to the possiblity of a umount operation hanging, we
438 * fork a child process and set a timeout. If the timeout
439 * lapses, the assumption is that that particular umount
440 * failed. */
441 r = safe_fork("(sd-umount)", FORK_RESET_SIGNALS|FORK_CLOSE_ALL_FDS|FORK_LOG|FORK_REOPEN_LOG, &pid);
442 if (r < 0)
443 return r;
444 if (r == 0) {
445 log_info("Unmounting '%s'.", m->path);
446
447 /* Start the mount operation here in the child Using MNT_FORCE
448 * causes some filesystems (e.g. FUSE and NFS and other network
449 * filesystems) to abort any pending requests and return -EIO
450 * rather than blocking indefinitely. If the filesysten is
451 * "busy", this may allow processes to die, thus making the
452 * filesystem less busy so the unmount might succeed (rather
453 * then return EBUSY).*/
454 r = umount2(m->path, MNT_FORCE);
455 if (r < 0)
456 log_full_errno(umount_log_level, errno, "Failed to unmount %s: %m", m->path);
457
458 _exit(r < 0 ? EXIT_FAILURE : EXIT_SUCCESS);
459 }
460
461 r = wait_for_terminate_with_timeout(pid, DEFAULT_TIMEOUT_USEC);
462 if (r == -ETIMEDOUT) {
463 log_error_errno(r, "Unmounting '%s' timed out, issuing SIGKILL to PID " PID_FMT ".", m->path, pid);
464 (void) kill(pid, SIGKILL);
465 } else if (r == -EPROTO)
466 log_debug_errno(r, "Unmounting '%s' failed abnormally, child process " PID_FMT " aborted or exited non-zero.", m->path, pid);
467 else if (r < 0)
468 log_error_errno(r, "Unmounting '%s' failed unexpectedly, couldn't wait for child process " PID_FMT ": %m", m->path, pid);
469
470 return r;
471 }
472
473 /* This includes remounting readonly, which changes the kernel mount options.
474 * Therefore the list passed to this function is invalidated, and should not be reused. */
475 static int mount_points_list_umount(MountPoint **head, bool *changed, int umount_log_level) {
476 MountPoint *m;
477 int n_failed = 0;
478
479 assert(head);
480 assert(changed);
481
482 LIST_FOREACH(mount_point, m, *head) {
483 if (m->try_remount_ro) {
484 /* We always try to remount directories
485 * read-only first, before we go on and umount
486 * them.
487 *
488 * Mount points can be stacked. If a mount
489 * point is stacked below / or /usr, we
490 * cannot umount or remount it directly,
491 * since there is no way to refer to the
492 * underlying mount. There's nothing we can do
493 * about it for the general case, but we can
494 * do something about it if it is aliased
495 * somehwere else via a bind mount. If we
496 * explicitly remount the super block of that
497 * alias read-only we hence should be
498 * relatively safe regarding keeping a dirty fs
499 * we cannot otherwise see.
500 *
501 * Since the remount can hang in the instance of
502 * remote filesystems, we remount asynchronously
503 * and skip the subsequent umount if it fails. */
504 if (remount_with_timeout(m, umount_log_level) < 0) {
505 /* Remount failed, but try unmounting anyway,
506 * unless this is a mount point we want to skip. */
507 if (nonunmountable_path(m->path)) {
508 n_failed++;
509 continue;
510 }
511 }
512 }
513
514 /* Skip / and /usr since we cannot unmount that
515 * anyway, since we are running from it. They have
516 * already been remounted ro. */
517 if (nonunmountable_path(m->path))
518 continue;
519
520 /* Trying to umount */
521 if (umount_with_timeout(m, umount_log_level) < 0)
522 n_failed++;
523 else
524 *changed = true;
525 }
526
527 return n_failed;
528 }
529
530 static int swap_points_list_off(MountPoint **head, bool *changed) {
531 MountPoint *m, *n;
532 int n_failed = 0;
533
534 assert(head);
535 assert(changed);
536
537 LIST_FOREACH_SAFE(mount_point, m, n, *head) {
538 log_info("Deactivating swap %s.", m->path);
539 if (swapoff(m->path) == 0) {
540 *changed = true;
541 mount_point_free(head, m);
542 } else {
543 log_warning_errno(errno, "Could not deactivate swap %s: %m", m->path);
544 n_failed++;
545 }
546 }
547
548 return n_failed;
549 }
550
551 static int loopback_points_list_detach(MountPoint **head, bool *changed, int umount_log_level) {
552 MountPoint *m, *n;
553 int n_failed = 0, k;
554 struct stat root_st;
555
556 assert(head);
557 assert(changed);
558
559 k = lstat("/", &root_st);
560
561 LIST_FOREACH_SAFE(mount_point, m, n, *head) {
562 int r;
563 struct stat loopback_st;
564
565 if (k >= 0 &&
566 major(root_st.st_dev) != 0 &&
567 lstat(m->path, &loopback_st) >= 0 &&
568 root_st.st_dev == loopback_st.st_rdev) {
569 n_failed++;
570 continue;
571 }
572
573 log_info("Detaching loopback %s.", m->path);
574 r = delete_loopback(m->path);
575 if (r >= 0) {
576 if (r > 0)
577 *changed = true;
578
579 mount_point_free(head, m);
580 } else {
581 log_full_errno(umount_log_level, errno, "Could not detach loopback %s: %m", m->path);
582 n_failed++;
583 }
584 }
585
586 return n_failed;
587 }
588
589 static int dm_points_list_detach(MountPoint **head, bool *changed, int umount_log_level) {
590 MountPoint *m, *n;
591 int n_failed = 0, r;
592 dev_t rootdev;
593
594 assert(head);
595 assert(changed);
596
597 r = get_block_device("/", &rootdev);
598 if (r <= 0)
599 rootdev = 0;
600
601 LIST_FOREACH_SAFE(mount_point, m, n, *head) {
602
603 if (major(rootdev) != 0 && rootdev == m->devnum) {
604 n_failed ++;
605 continue;
606 }
607
608 log_info("Detaching DM %u:%u.", major(m->devnum), minor(m->devnum));
609 r = delete_dm(m->devnum);
610 if (r >= 0) {
611 *changed = true;
612 mount_point_free(head, m);
613 } else {
614 log_full_errno(umount_log_level, errno, "Could not detach DM %s: %m", m->path);
615 n_failed++;
616 }
617 }
618
619 return n_failed;
620 }
621
622 static int umount_all_once(bool *changed, int umount_log_level) {
623 int r;
624 _cleanup_(mount_points_list_free) LIST_HEAD(MountPoint, mp_list_head);
625
626 assert(changed);
627
628 LIST_HEAD_INIT(mp_list_head);
629 r = mount_points_list_get(NULL, &mp_list_head);
630 if (r < 0)
631 return r;
632
633 return mount_points_list_umount(&mp_list_head, changed, umount_log_level);
634 }
635
636 int umount_all(bool *changed, int umount_log_level) {
637 bool umount_changed;
638 int r;
639
640 assert(changed);
641
642 /* Retry umount, until nothing can be umounted anymore. Mounts are
643 * processed in order, newest first. The retries are needed when
644 * an old mount has been moved, to a path inside a newer mount. */
645 do {
646 umount_changed = false;
647
648 r = umount_all_once(&umount_changed, umount_log_level);
649 if (umount_changed)
650 *changed = true;
651 } while (umount_changed);
652
653 return r;
654 }
655
656 int swapoff_all(bool *changed) {
657 _cleanup_(mount_points_list_free) LIST_HEAD(MountPoint, swap_list_head);
658 int r;
659
660 assert(changed);
661
662 LIST_HEAD_INIT(swap_list_head);
663
664 r = swap_list_get(NULL, &swap_list_head);
665 if (r < 0)
666 return r;
667
668 return swap_points_list_off(&swap_list_head, changed);
669 }
670
671 int loopback_detach_all(bool *changed, int umount_log_level) {
672 _cleanup_(mount_points_list_free) LIST_HEAD(MountPoint, loopback_list_head);
673 int r;
674
675 assert(changed);
676
677 LIST_HEAD_INIT(loopback_list_head);
678
679 r = loopback_list_get(&loopback_list_head);
680 if (r < 0)
681 return r;
682
683 return loopback_points_list_detach(&loopback_list_head, changed, umount_log_level);
684 }
685
686 int dm_detach_all(bool *changed, int umount_log_level) {
687 _cleanup_(mount_points_list_free) LIST_HEAD(MountPoint, dm_list_head);
688 int r;
689
690 assert(changed);
691
692 LIST_HEAD_INIT(dm_list_head);
693
694 r = dm_list_get(&dm_list_head);
695 if (r < 0)
696 return r;
697
698 return dm_points_list_detach(&dm_list_head, changed, umount_log_level);
699 }