]> git.ipfire.org Git - thirdparty/systemd.git/blob - src/core/mount.c
core: port unit_fork_helper_process() and unit_fork_and_watch_rm_rf() to PidRef
[thirdparty/systemd.git] / src / core / mount.c
1 /* SPDX-License-Identifier: LGPL-2.1-or-later */
2
3 #include <errno.h>
4 #include <signal.h>
5 #include <stdio.h>
6 #include <sys/epoll.h>
7
8 #include "sd-messages.h"
9
10 #include "alloc-util.h"
11 #include "dbus-mount.h"
12 #include "dbus-unit.h"
13 #include "device.h"
14 #include "exit-status.h"
15 #include "format-util.h"
16 #include "fs-util.h"
17 #include "fstab-util.h"
18 #include "initrd-util.h"
19 #include "libmount-util.h"
20 #include "log.h"
21 #include "manager.h"
22 #include "mkdir-label.h"
23 #include "mount-setup.h"
24 #include "mount.h"
25 #include "mountpoint-util.h"
26 #include "parse-util.h"
27 #include "path-util.h"
28 #include "process-util.h"
29 #include "serialize.h"
30 #include "special.h"
31 #include "stat-util.h"
32 #include "string-table.h"
33 #include "string-util.h"
34 #include "strv.h"
35 #include "unit-name.h"
36 #include "unit.h"
37
38 #define RETRY_UMOUNT_MAX 32
39
40 static const UnitActiveState state_translation_table[_MOUNT_STATE_MAX] = {
41 [MOUNT_DEAD] = UNIT_INACTIVE,
42 [MOUNT_MOUNTING] = UNIT_ACTIVATING,
43 [MOUNT_MOUNTING_DONE] = UNIT_ACTIVATING,
44 [MOUNT_MOUNTED] = UNIT_ACTIVE,
45 [MOUNT_REMOUNTING] = UNIT_RELOADING,
46 [MOUNT_UNMOUNTING] = UNIT_DEACTIVATING,
47 [MOUNT_REMOUNTING_SIGTERM] = UNIT_RELOADING,
48 [MOUNT_REMOUNTING_SIGKILL] = UNIT_RELOADING,
49 [MOUNT_UNMOUNTING_SIGTERM] = UNIT_DEACTIVATING,
50 [MOUNT_UNMOUNTING_SIGKILL] = UNIT_DEACTIVATING,
51 [MOUNT_FAILED] = UNIT_FAILED,
52 [MOUNT_CLEANING] = UNIT_MAINTENANCE,
53 };
54
55 static int mount_dispatch_timer(sd_event_source *source, usec_t usec, void *userdata);
56 static int mount_dispatch_io(sd_event_source *source, int fd, uint32_t revents, void *userdata);
57 static void mount_enter_dead(Mount *m, MountResult f);
58 static void mount_enter_mounted(Mount *m, MountResult f);
59 static void mount_cycle_clear(Mount *m);
60 static int mount_process_proc_self_mountinfo(Manager *m);
61
62 static bool MOUNT_STATE_WITH_PROCESS(MountState state) {
63 return IN_SET(state,
64 MOUNT_MOUNTING,
65 MOUNT_MOUNTING_DONE,
66 MOUNT_REMOUNTING,
67 MOUNT_REMOUNTING_SIGTERM,
68 MOUNT_REMOUNTING_SIGKILL,
69 MOUNT_UNMOUNTING,
70 MOUNT_UNMOUNTING_SIGTERM,
71 MOUNT_UNMOUNTING_SIGKILL,
72 MOUNT_CLEANING);
73 }
74
75 static MountParameters* get_mount_parameters_fragment(Mount *m) {
76 assert(m);
77
78 if (m->from_fragment)
79 return &m->parameters_fragment;
80
81 return NULL;
82 }
83
84 static MountParameters* get_mount_parameters(Mount *m) {
85 assert(m);
86
87 if (m->from_proc_self_mountinfo)
88 return &m->parameters_proc_self_mountinfo;
89
90 return get_mount_parameters_fragment(m);
91 }
92
93 static bool mount_is_network(const MountParameters *p) {
94 assert(p);
95
96 if (fstab_test_option(p->options, "_netdev\0"))
97 return true;
98
99 if (p->fstype && fstype_is_network(p->fstype))
100 return true;
101
102 return false;
103 }
104
105 static bool mount_is_nofail(const Mount *m) {
106 assert(m);
107
108 if (!m->from_fragment)
109 return false;
110
111 return fstab_test_yes_no_option(m->parameters_fragment.options, "nofail\0" "fail\0");
112 }
113
114 static bool mount_is_loop(const MountParameters *p) {
115 assert(p);
116
117 if (fstab_test_option(p->options, "loop\0"))
118 return true;
119
120 return false;
121 }
122
123 static bool mount_is_bind(const MountParameters *p) {
124 assert(p);
125 return fstab_is_bind(p->options, p->fstype);
126 }
127
128 static bool mount_is_bound_to_device(Mount *m) {
129 const MountParameters *p;
130
131 assert(m);
132
133 /* Determines whether to place a Requires= or BindsTo= dependency on the backing device unit. We do
134 * this by checking for the x-systemd.device-bound mount option. Iff it is set we use BindsTo=,
135 * otherwise Requires=. But note that we might combine the latter with StopPropagatedFrom=, see
136 * below. */
137
138 p = get_mount_parameters(m);
139 if (!p)
140 return false;
141
142 return fstab_test_option(p->options, "x-systemd.device-bound\0");
143 }
144
145 static bool mount_propagate_stop(Mount *m) {
146 assert(m);
147
148 if (mount_is_bound_to_device(m)) /* If we are using BindsTo= the stop propagation is implicit, no need to bother */
149 return false;
150
151 return m->from_fragment; /* let's propagate stop whenever this is an explicitly configured unit,
152 * otherwise let's not bother. */
153 }
154
155 static bool mount_needs_quota(const MountParameters *p) {
156 assert(p);
157
158 if (p->fstype && !fstype_needs_quota(p->fstype))
159 return false;
160
161 if (mount_is_bind(p))
162 return false;
163
164 return fstab_test_option(p->options,
165 "usrquota\0" "grpquota\0" "quota\0" "usrjquota\0" "grpjquota\0");
166 }
167
168 static void mount_init(Unit *u) {
169 Mount *m = MOUNT(u);
170
171 assert(m);
172 assert(u);
173 assert(u->load_state == UNIT_STUB);
174
175 m->timeout_usec = u->manager->defaults.timeout_start_usec;
176
177 m->exec_context.std_output = u->manager->defaults.std_output;
178 m->exec_context.std_error = u->manager->defaults.std_error;
179
180 m->directory_mode = 0755;
181
182 /* We need to make sure that /usr/bin/mount is always called
183 * in the same process group as us, so that the autofs kernel
184 * side doesn't send us another mount request while we are
185 * already trying to comply its last one. */
186 m->exec_context.same_pgrp = true;
187
188 m->control_pid = PIDREF_NULL;
189 m->control_command_id = _MOUNT_EXEC_COMMAND_INVALID;
190
191 u->ignore_on_isolate = true;
192 }
193
194 static int mount_arm_timer(Mount *m, usec_t usec) {
195 int r;
196
197 assert(m);
198
199 if (usec == USEC_INFINITY)
200 return sd_event_source_set_enabled(m->timer_event_source, SD_EVENT_OFF);
201
202 if (m->timer_event_source) {
203 r = sd_event_source_set_time(m->timer_event_source, usec);
204 if (r < 0)
205 return r;
206
207 return sd_event_source_set_enabled(m->timer_event_source, SD_EVENT_ONESHOT);
208 }
209
210 r = sd_event_add_time(
211 UNIT(m)->manager->event,
212 &m->timer_event_source,
213 CLOCK_MONOTONIC,
214 usec, 0,
215 mount_dispatch_timer, m);
216 if (r < 0)
217 return r;
218
219 (void) sd_event_source_set_description(m->timer_event_source, "mount-timer");
220
221 return 0;
222 }
223
224 static void mount_unwatch_control_pid(Mount *m) {
225 assert(m);
226
227 if (!pidref_is_set(&m->control_pid))
228 return;
229
230 unit_unwatch_pid(UNIT(m), m->control_pid.pid);
231 pidref_done(&m->control_pid);
232 }
233
234 static void mount_parameters_done(MountParameters *p) {
235 assert(p);
236
237 p->what = mfree(p->what);
238 p->options = mfree(p->options);
239 p->fstype = mfree(p->fstype);
240 }
241
242 static void mount_done(Unit *u) {
243 Mount *m = MOUNT(u);
244
245 assert(m);
246
247 m->where = mfree(m->where);
248
249 mount_parameters_done(&m->parameters_proc_self_mountinfo);
250 mount_parameters_done(&m->parameters_fragment);
251
252 m->exec_runtime = exec_runtime_free(m->exec_runtime);
253 exec_command_done_array(m->exec_command, _MOUNT_EXEC_COMMAND_MAX);
254 m->control_command = NULL;
255
256 mount_unwatch_control_pid(m);
257
258 m->timer_event_source = sd_event_source_disable_unref(m->timer_event_source);
259 }
260
261 static int update_parameters_proc_self_mountinfo(
262 Mount *m,
263 const char *what,
264 const char *options,
265 const char *fstype) {
266
267 MountParameters *p;
268 int r, q, w;
269
270 p = &m->parameters_proc_self_mountinfo;
271
272 r = free_and_strdup(&p->what, what);
273 if (r < 0)
274 return r;
275
276 q = free_and_strdup(&p->options, options);
277 if (q < 0)
278 return q;
279
280 w = free_and_strdup(&p->fstype, fstype);
281 if (w < 0)
282 return w;
283
284 return r > 0 || q > 0 || w > 0;
285 }
286
287 static int mount_add_mount_dependencies(Mount *m) {
288 MountParameters *pm;
289 Unit *other;
290 Set *s;
291 int r;
292
293 assert(m);
294
295 if (!path_equal(m->where, "/")) {
296 _cleanup_free_ char *parent = NULL;
297
298 /* Adds in links to other mount points that might lie further up in the hierarchy */
299
300 r = path_extract_directory(m->where, &parent);
301 if (r < 0)
302 return r;
303
304 r = unit_require_mounts_for(UNIT(m), parent, UNIT_DEPENDENCY_IMPLICIT);
305 if (r < 0)
306 return r;
307 }
308
309 /* Adds in dependencies to other mount points that might be needed for the source path (if this is a bind mount
310 * or a loop mount) to be available. */
311 pm = get_mount_parameters_fragment(m);
312 if (pm && pm->what &&
313 path_is_absolute(pm->what) &&
314 (mount_is_bind(pm) || mount_is_loop(pm) || !mount_is_network(pm))) {
315
316 r = unit_require_mounts_for(UNIT(m), pm->what, UNIT_DEPENDENCY_FILE);
317 if (r < 0)
318 return r;
319 }
320
321 /* Adds in dependencies to other units that use this path or paths further down in the hierarchy */
322 s = manager_get_units_requiring_mounts_for(UNIT(m)->manager, m->where);
323 SET_FOREACH(other, s) {
324
325 if (other->load_state != UNIT_LOADED)
326 continue;
327
328 if (other == UNIT(m))
329 continue;
330
331 r = unit_add_dependency(other, UNIT_AFTER, UNIT(m), true, UNIT_DEPENDENCY_PATH);
332 if (r < 0)
333 return r;
334
335 if (UNIT(m)->fragment_path) {
336 /* If we have fragment configuration, then make this dependency required */
337 r = unit_add_dependency(other, UNIT_REQUIRES, UNIT(m), true, UNIT_DEPENDENCY_PATH);
338 if (r < 0)
339 return r;
340 }
341 }
342
343 return 0;
344 }
345
346 static int mount_add_device_dependencies(Mount *m) {
347 UnitDependencyMask mask;
348 MountParameters *p;
349 UnitDependency dep;
350 int r;
351
352 assert(m);
353
354 log_unit_trace(UNIT(m), "Processing implicit device dependencies");
355
356 p = get_mount_parameters(m);
357 if (!p) {
358 log_unit_trace(UNIT(m), "Missing mount parameters, skipping implicit device dependencies");
359 return 0;
360 }
361
362 if (!p->what) {
363 log_unit_trace(UNIT(m), "Missing mount source, skipping implicit device dependencies");
364 return 0;
365 }
366
367 if (mount_is_bind(p)) {
368 log_unit_trace(UNIT(m), "Mount unit is a bind mount, skipping implicit device dependencies");
369 return 0;
370 }
371
372 if (!is_device_path(p->what)) {
373 log_unit_trace(UNIT(m), "Mount source is not a device path, skipping implicit device dependencies");
374 return 0;
375 }
376
377 /* /dev/root is a really weird thing, it's not a real device, but just a path the kernel exports for
378 * the root file system specified on the kernel command line. Ignore it here. */
379 if (PATH_IN_SET(p->what, "/dev/root", "/dev/nfs")) {
380 log_unit_trace(UNIT(m), "Mount source is in /dev/root or /dev/nfs, skipping implicit device dependencies");
381 return 0;
382 }
383
384 if (path_equal(m->where, "/")) {
385 log_unit_trace(UNIT(m), "Mount destination is '/', skipping implicit device dependencies");
386 return 0;
387 }
388
389 /* Mount units from /proc/self/mountinfo are not bound to devices by default since they're subject to
390 * races when mounts are established by other tools with different backing devices than what we
391 * maintain. The user can still force this to be a BindsTo= dependency with an appropriate option (or
392 * udev property) so the mount units are automatically stopped when the device disappears
393 * suddenly. */
394 dep = mount_is_bound_to_device(m) ? UNIT_BINDS_TO : UNIT_REQUIRES;
395
396 /* We always use 'what' from /proc/self/mountinfo if mounted */
397 mask = m->from_proc_self_mountinfo ? UNIT_DEPENDENCY_MOUNTINFO : UNIT_DEPENDENCY_MOUNT_FILE;
398
399 r = unit_add_node_dependency(UNIT(m), p->what, dep, mask);
400 if (r < 0)
401 return r;
402 if (r > 0)
403 log_unit_trace(UNIT(m), "Added %s dependency on %s", unit_dependency_to_string(dep), p->what);
404
405 if (mount_propagate_stop(m)) {
406 r = unit_add_node_dependency(UNIT(m), p->what, UNIT_STOP_PROPAGATED_FROM, mask);
407 if (r < 0)
408 return r;
409 if (r > 0)
410 log_unit_trace(UNIT(m), "Added %s dependency on %s",
411 unit_dependency_to_string(UNIT_STOP_PROPAGATED_FROM), p->what);
412 }
413
414 r = unit_add_blockdev_dependency(UNIT(m), p->what, mask);
415 if (r > 0)
416 log_unit_trace(UNIT(m), "Added %s dependency on %s", unit_dependency_to_string(UNIT_AFTER), p->what);
417
418 return 0;
419 }
420
421 static int mount_add_quota_dependencies(Mount *m) {
422 MountParameters *p;
423 int r;
424
425 assert(m);
426
427 if (!MANAGER_IS_SYSTEM(UNIT(m)->manager))
428 return 0;
429
430 p = get_mount_parameters_fragment(m);
431 if (!p)
432 return 0;
433
434 if (!mount_needs_quota(p))
435 return 0;
436
437 r = unit_add_two_dependencies_by_name(UNIT(m), UNIT_BEFORE, UNIT_WANTS, SPECIAL_QUOTACHECK_SERVICE,
438 /* add_reference= */ true, UNIT_DEPENDENCY_FILE);
439 if (r < 0)
440 return r;
441
442 r = unit_add_two_dependencies_by_name(UNIT(m), UNIT_BEFORE, UNIT_WANTS, SPECIAL_QUOTAON_SERVICE,
443 /* add_reference= */true, UNIT_DEPENDENCY_FILE);
444 if (r < 0)
445 return r;
446
447 return 0;
448 }
449
450 static bool mount_is_extrinsic(Unit *u) {
451 MountParameters *p;
452 Mount *m = MOUNT(u);
453 assert(m);
454
455 /* Returns true for all units that are "magic" and should be excluded from the usual
456 * start-up and shutdown dependencies. We call them "extrinsic" here, as they are generally
457 * mounted outside of the systemd dependency logic. We shouldn't attempt to manage them
458 * ourselves but it's fine if the user operates on them with us. */
459
460 /* We only automatically manage mounts if we are in system mode */
461 if (MANAGER_IS_USER(u->manager))
462 return true;
463
464 p = get_mount_parameters(m);
465 if (p && fstab_is_extrinsic(m->where, p->options))
466 return true;
467
468 return false;
469 }
470
471 static bool mount_is_credentials(Mount *m) {
472 const char *e;
473
474 assert(m);
475
476 /* Returns true if this is a credentials mount. We don't want automatic dependencies on credential
477 * mounts, since they are managed by us for even the earliest services, and we never want anything to
478 * be ordered before them hence. */
479
480 e = path_startswith(m->where, UNIT(m)->manager->prefix[EXEC_DIRECTORY_RUNTIME]);
481 if (!e)
482 return false;
483
484 return !isempty(path_startswith(e, "credentials"));
485 }
486
487 static int mount_add_default_ordering_dependencies(Mount *m, MountParameters *p, UnitDependencyMask mask) {
488 const char *after, *before, *e;
489 int r;
490
491 assert(m);
492
493 e = path_startswith(m->where, "/sysroot");
494 if (e && in_initrd()) {
495 /* All mounts under /sysroot need to happen later, at initrd-fs.target time. IOW,
496 * it's not technically part of the basic initrd filesystem itself, and so
497 * shouldn't inherit the default Before=local-fs.target dependency. However,
498 * these mounts still need to start after local-fs-pre.target, as a sync point
499 * for things like systemd-hibernate-resume.service that should start before
500 * any mounts. */
501
502 after = SPECIAL_LOCAL_FS_PRE_TARGET;
503 before = isempty(e) ? SPECIAL_INITRD_ROOT_FS_TARGET : SPECIAL_INITRD_FS_TARGET;
504
505 } else if (in_initrd() && path_startswith(m->where, "/sysusr/usr")) {
506 after = SPECIAL_LOCAL_FS_PRE_TARGET;
507 before = SPECIAL_INITRD_USR_FS_TARGET;
508
509 } else if (mount_is_credentials(m))
510 after = before = NULL;
511
512 else if (mount_is_network(p)) {
513 after = SPECIAL_REMOTE_FS_PRE_TARGET;
514 before = SPECIAL_REMOTE_FS_TARGET;
515
516 } else {
517 after = SPECIAL_LOCAL_FS_PRE_TARGET;
518 before = SPECIAL_LOCAL_FS_TARGET;
519 }
520
521 if (before && !mount_is_nofail(m)) {
522 r = unit_add_dependency_by_name(UNIT(m), UNIT_BEFORE, before, /* add_reference= */ true, mask);
523 if (r < 0)
524 return r;
525 }
526
527 if (after) {
528 r = unit_add_dependency_by_name(UNIT(m), UNIT_AFTER, after, /* add_reference= */ true, mask);
529 if (r < 0)
530 return r;
531 }
532
533 r = unit_add_two_dependencies_by_name(UNIT(m), UNIT_BEFORE, UNIT_CONFLICTS, SPECIAL_UMOUNT_TARGET,
534 /* add_reference= */ true, mask);
535 if (r < 0)
536 return r;
537
538 /* If this is a tmpfs mount then we have to unmount it before we try to deactivate swaps */
539 if (streq_ptr(p->fstype, "tmpfs") && !mount_is_credentials(m)) {
540 r = unit_add_dependency_by_name(UNIT(m), UNIT_AFTER, SPECIAL_SWAP_TARGET,
541 /* add_reference= */ true, mask);
542 if (r < 0)
543 return r;
544 }
545
546 return 0;
547 }
548
549 static int mount_add_default_network_dependencies(Mount *m, MountParameters *p, UnitDependencyMask mask) {
550 int r;
551
552 assert(m);
553
554 if (!mount_is_network(p))
555 return 0;
556
557 /* We order ourselves after network.target. This is primarily useful at shutdown: services that take
558 * down the network should order themselves before network.target, so that they are shut down only
559 * after this mount unit is stopped. */
560
561 r = unit_add_dependency_by_name(UNIT(m), UNIT_AFTER, SPECIAL_NETWORK_TARGET,
562 /* add_reference= */ true, mask);
563 if (r < 0)
564 return r;
565
566 /* We pull in network-online.target, and order ourselves after it. This is useful at start-up to
567 * actively pull in tools that want to be started before we start mounting network file systems, and
568 * whose purpose it is to delay this until the network is "up". */
569
570 return unit_add_two_dependencies_by_name(UNIT(m), UNIT_WANTS, UNIT_AFTER, SPECIAL_NETWORK_ONLINE_TARGET,
571 /* add_reference= */ true, mask);
572 }
573
574 static int mount_add_default_dependencies(Mount *m) {
575 UnitDependencyMask mask;
576 MountParameters *p;
577 int r;
578
579 assert(m);
580
581 if (!UNIT(m)->default_dependencies)
582 return 0;
583
584 /* We do not add any default dependencies to /, /usr or /run/initramfs/, since they are
585 * guaranteed to stay mounted the whole time, since our system is on it. Also, don't
586 * bother with anything mounted below virtual file systems, it's also going to be virtual,
587 * and hence not worth the effort. */
588 if (mount_is_extrinsic(UNIT(m)))
589 return 0;
590
591 p = get_mount_parameters(m);
592 if (!p)
593 return 0;
594
595 mask = m->from_proc_self_mountinfo ? UNIT_DEPENDENCY_MOUNTINFO : UNIT_DEPENDENCY_MOUNT_FILE;
596
597 r = mount_add_default_ordering_dependencies(m, p, mask);
598 if (r < 0)
599 return r;
600
601 r = mount_add_default_network_dependencies(m, p, mask);
602 if (r < 0)
603 return r;
604
605 return 0;
606 }
607
608 static int mount_verify(Mount *m) {
609 _cleanup_free_ char *e = NULL;
610 MountParameters *p;
611 int r;
612
613 assert(m);
614 assert(UNIT(m)->load_state == UNIT_LOADED);
615
616 if (!m->from_fragment && !m->from_proc_self_mountinfo && !UNIT(m)->perpetual)
617 return -ENOENT;
618
619 r = unit_name_from_path(m->where, ".mount", &e);
620 if (r < 0)
621 return log_unit_error_errno(UNIT(m), r, "Failed to generate unit name from mount path: %m");
622
623 if (!unit_has_name(UNIT(m), e))
624 return log_unit_error_errno(UNIT(m), SYNTHETIC_ERRNO(ENOEXEC), "Where= setting doesn't match unit name. Refusing.");
625
626 if (mount_point_is_api(m->where) || mount_point_ignore(m->where))
627 return log_unit_error_errno(UNIT(m), SYNTHETIC_ERRNO(ENOEXEC), "Cannot create mount unit for API file system %s. Refusing.", m->where);
628
629 p = get_mount_parameters_fragment(m);
630 if (p && !p->what && !UNIT(m)->perpetual)
631 return log_unit_error_errno(UNIT(m), SYNTHETIC_ERRNO(ENOEXEC),
632 "What= setting is missing. Refusing.");
633
634 if (m->exec_context.pam_name && m->kill_context.kill_mode != KILL_CONTROL_GROUP)
635 return log_unit_error_errno(UNIT(m), SYNTHETIC_ERRNO(ENOEXEC), "Unit has PAM enabled. Kill mode must be set to control-group'. Refusing.");
636
637 return 0;
638 }
639
640 static int mount_add_non_exec_dependencies(Mount *m) {
641 int r;
642
643 assert(m);
644
645 /* We may be called due to this mount appearing in /proc/self/mountinfo, hence we clear all existing
646 * dependencies that were initialized from the unit file but whose final value really depends on the
647 * content of /proc/self/mountinfo. Some (such as m->where) might have become stale now. */
648 unit_remove_dependencies(UNIT(m), UNIT_DEPENDENCY_MOUNTINFO | UNIT_DEPENDENCY_MOUNT_FILE);
649
650 if (!m->where)
651 return 0;
652
653 /* Adds in all dependencies directly responsible for ordering the mount, as opposed to dependencies
654 * resulting from the ExecContext and such. */
655
656 r = mount_add_device_dependencies(m);
657 if (r < 0)
658 return r;
659
660 r = mount_add_mount_dependencies(m);
661 if (r < 0)
662 return r;
663
664 r = mount_add_quota_dependencies(m);
665 if (r < 0)
666 return r;
667
668 r = mount_add_default_dependencies(m);
669 if (r < 0)
670 return r;
671
672 return 0;
673 }
674
675 static int mount_add_extras(Mount *m) {
676 Unit *u = UNIT(m);
677 int r;
678
679 assert(m);
680
681 /* Note: this call might be called after we already have been loaded once (and even when it has already been
682 * activated), in case data from /proc/self/mountinfo has changed. This means all code here needs to be ready
683 * to run with an already set up unit. */
684
685 if (u->fragment_path)
686 m->from_fragment = true;
687
688 if (!m->where) {
689 r = unit_name_to_path(u->id, &m->where);
690 if (r == -ENAMETOOLONG)
691 log_unit_error_errno(u, r, "Failed to derive mount point path from unit name, because unit name is hashed. "
692 "Set \"Where=\" in the unit file explicitly.");
693 if (r < 0)
694 return r;
695 }
696
697 path_simplify(m->where);
698
699 if (!u->description) {
700 r = unit_set_description(u, m->where);
701 if (r < 0)
702 return r;
703 }
704
705 r = unit_patch_contexts(u);
706 if (r < 0)
707 return r;
708
709 r = unit_add_exec_dependencies(u, &m->exec_context);
710 if (r < 0)
711 return r;
712
713 r = unit_set_default_slice(u);
714 if (r < 0)
715 return r;
716
717 r = mount_add_non_exec_dependencies(m);
718 if (r < 0)
719 return r;
720
721 return 0;
722 }
723
724 static void mount_load_root_mount(Unit *u) {
725 assert(u);
726
727 if (!unit_has_name(u, SPECIAL_ROOT_MOUNT))
728 return;
729
730 u->perpetual = true;
731 u->default_dependencies = false;
732
733 /* The stdio/kmsg bridge socket is on /, in order to avoid a dep loop, don't use kmsg logging for -.mount */
734 MOUNT(u)->exec_context.std_output = EXEC_OUTPUT_NULL;
735 MOUNT(u)->exec_context.std_input = EXEC_INPUT_NULL;
736
737 if (!u->description)
738 u->description = strdup("Root Mount");
739 }
740
741 static int mount_load(Unit *u) {
742 Mount *m = MOUNT(u);
743 int r, q = 0;
744
745 assert(m);
746 assert(u);
747 assert(u->load_state == UNIT_STUB);
748
749 mount_load_root_mount(u);
750
751 bool fragment_optional = m->from_proc_self_mountinfo || u->perpetual;
752 r = unit_load_fragment_and_dropin(u, !fragment_optional);
753
754 /* Add in some extras. Note we do this in all cases (even if we failed to load the unit) when announced by the
755 * kernel, because we need some things to be set up no matter what when the kernel establishes a mount and thus
756 * we need to update the state in our unit to track it. After all, consider that we don't allow changing the
757 * 'slice' field for a unit once it is active. */
758 if (u->load_state == UNIT_LOADED || m->from_proc_self_mountinfo || u->perpetual)
759 q = mount_add_extras(m);
760
761 if (r < 0)
762 return r;
763 if (q < 0)
764 return q;
765 if (u->load_state != UNIT_LOADED)
766 return 0;
767
768 return mount_verify(m);
769 }
770
771 static void mount_set_state(Mount *m, MountState state) {
772 MountState old_state;
773 assert(m);
774
775 if (m->state != state)
776 bus_unit_send_pending_change_signal(UNIT(m), false);
777
778 old_state = m->state;
779 m->state = state;
780
781 if (!MOUNT_STATE_WITH_PROCESS(state)) {
782 m->timer_event_source = sd_event_source_disable_unref(m->timer_event_source);
783 mount_unwatch_control_pid(m);
784 m->control_command = NULL;
785 m->control_command_id = _MOUNT_EXEC_COMMAND_INVALID;
786 }
787
788 if (state != old_state)
789 log_unit_debug(UNIT(m), "Changed %s -> %s", mount_state_to_string(old_state), mount_state_to_string(state));
790
791 unit_notify(UNIT(m), state_translation_table[old_state], state_translation_table[state], m->reload_result == MOUNT_SUCCESS);
792 }
793
794 static int mount_coldplug(Unit *u) {
795 Mount *m = MOUNT(u);
796 int r;
797
798 assert(m);
799 assert(m->state == MOUNT_DEAD);
800
801 if (m->deserialized_state == m->state)
802 return 0;
803
804 if (pidref_is_set(&m->control_pid) &&
805 pid_is_unwaited(m->control_pid.pid) &&
806 MOUNT_STATE_WITH_PROCESS(m->deserialized_state)) {
807
808 r = unit_watch_pid(UNIT(m), m->control_pid.pid, /* exclusive= */ false);
809 if (r < 0)
810 return r;
811
812 r = mount_arm_timer(m, usec_add(u->state_change_timestamp.monotonic, m->timeout_usec));
813 if (r < 0)
814 return r;
815 }
816
817 if (!IN_SET(m->deserialized_state, MOUNT_DEAD, MOUNT_FAILED))
818 (void) unit_setup_exec_runtime(u);
819
820 mount_set_state(m, m->deserialized_state);
821 return 0;
822 }
823
824 static void mount_catchup(Unit *u) {
825 Mount *m = MOUNT(ASSERT_PTR(u));
826
827 assert(m);
828
829 /* Adjust the deserialized state. See comments in mount_process_proc_self_mountinfo(). */
830 if (m->from_proc_self_mountinfo)
831 switch (m->state) {
832 case MOUNT_DEAD:
833 case MOUNT_FAILED:
834 assert(!pidref_is_set(&m->control_pid));
835 (void) unit_acquire_invocation_id(u);
836 mount_cycle_clear(m);
837 mount_enter_mounted(m, MOUNT_SUCCESS);
838 break;
839 case MOUNT_MOUNTING:
840 assert(pidref_is_set(&m->control_pid));
841 mount_set_state(m, MOUNT_MOUNTING_DONE);
842 break;
843 default:
844 break;
845 }
846 else
847 switch (m->state) {
848 case MOUNT_MOUNTING_DONE:
849 assert(pidref_is_set(&m->control_pid));
850 mount_set_state(m, MOUNT_MOUNTING);
851 break;
852 case MOUNT_MOUNTED:
853 assert(!pidref_is_set(&m->control_pid));
854 mount_enter_dead(m, MOUNT_SUCCESS);
855 break;
856 default:
857 break;
858 }
859 }
860
861 static void mount_dump(Unit *u, FILE *f, const char *prefix) {
862 Mount *m = MOUNT(u);
863 MountParameters *p;
864
865 assert(m);
866 assert(f);
867
868 p = get_mount_parameters(m);
869
870 fprintf(f,
871 "%sMount State: %s\n"
872 "%sResult: %s\n"
873 "%sClean Result: %s\n"
874 "%sWhere: %s\n"
875 "%sWhat: %s\n"
876 "%sFile System Type: %s\n"
877 "%sOptions: %s\n"
878 "%sFrom /proc/self/mountinfo: %s\n"
879 "%sFrom fragment: %s\n"
880 "%sExtrinsic: %s\n"
881 "%sDirectoryMode: %04o\n"
882 "%sSloppyOptions: %s\n"
883 "%sLazyUnmount: %s\n"
884 "%sForceUnmount: %s\n"
885 "%sReadWriteOnly: %s\n"
886 "%sTimeoutSec: %s\n",
887 prefix, mount_state_to_string(m->state),
888 prefix, mount_result_to_string(m->result),
889 prefix, mount_result_to_string(m->clean_result),
890 prefix, m->where,
891 prefix, p ? strna(p->what) : "n/a",
892 prefix, p ? strna(p->fstype) : "n/a",
893 prefix, p ? strna(p->options) : "n/a",
894 prefix, yes_no(m->from_proc_self_mountinfo),
895 prefix, yes_no(m->from_fragment),
896 prefix, yes_no(mount_is_extrinsic(u)),
897 prefix, m->directory_mode,
898 prefix, yes_no(m->sloppy_options),
899 prefix, yes_no(m->lazy_unmount),
900 prefix, yes_no(m->force_unmount),
901 prefix, yes_no(m->read_write_only),
902 prefix, FORMAT_TIMESPAN(m->timeout_usec, USEC_PER_SEC));
903
904 if (pidref_is_set(&m->control_pid))
905 fprintf(f,
906 "%sControl PID: "PID_FMT"\n",
907 prefix, m->control_pid.pid);
908
909 exec_context_dump(&m->exec_context, f, prefix);
910 kill_context_dump(&m->kill_context, f, prefix);
911 cgroup_context_dump(UNIT(m), f, prefix);
912 }
913
914 static int mount_spawn(Mount *m, ExecCommand *c, PidRef *ret_pid) {
915
916 _cleanup_(exec_params_clear) ExecParameters exec_params = {
917 .flags = EXEC_APPLY_SANDBOXING|EXEC_APPLY_CHROOT|EXEC_APPLY_TTY_STDIN,
918 .stdin_fd = -EBADF,
919 .stdout_fd = -EBADF,
920 .stderr_fd = -EBADF,
921 .exec_fd = -EBADF,
922 };
923 _cleanup_(pidref_done) PidRef pidref = PIDREF_NULL;
924 pid_t pid;
925 int r;
926
927 assert(m);
928 assert(c);
929 assert(ret_pid);
930
931 r = unit_prepare_exec(UNIT(m));
932 if (r < 0)
933 return r;
934
935 r = mount_arm_timer(m, usec_add(now(CLOCK_MONOTONIC), m->timeout_usec));
936 if (r < 0)
937 return r;
938
939 r = unit_set_exec_params(UNIT(m), &exec_params);
940 if (r < 0)
941 return r;
942
943 r = exec_spawn(UNIT(m),
944 c,
945 &m->exec_context,
946 &exec_params,
947 m->exec_runtime,
948 &m->cgroup_context,
949 &pid);
950 if (r < 0)
951 return r;
952
953 r = pidref_set_pid(&pidref, pid);
954 if (r < 0)
955 return r;
956
957 r = unit_watch_pid(UNIT(m), pidref.pid, /* exclusive= */ true);
958 if (r < 0)
959 return r;
960
961 *ret_pid = TAKE_PIDREF(pidref);
962 return 0;
963 }
964
965 static void mount_enter_dead(Mount *m, MountResult f) {
966 assert(m);
967
968 if (m->result == MOUNT_SUCCESS)
969 m->result = f;
970
971 unit_log_result(UNIT(m), m->result == MOUNT_SUCCESS, mount_result_to_string(m->result));
972 unit_warn_leftover_processes(UNIT(m), unit_log_leftover_process_stop);
973
974 mount_set_state(m, m->result != MOUNT_SUCCESS ? MOUNT_FAILED : MOUNT_DEAD);
975
976 m->exec_runtime = exec_runtime_destroy(m->exec_runtime);
977
978 unit_destroy_runtime_data(UNIT(m), &m->exec_context);
979
980 unit_unref_uid_gid(UNIT(m), true);
981
982 /* Any dependencies based on /proc/self/mountinfo are now stale. Let's re-generate dependencies from
983 * .mount unit. */
984 (void) mount_add_non_exec_dependencies(m);
985 }
986
987 static void mount_enter_mounted(Mount *m, MountResult f) {
988 assert(m);
989
990 if (m->result == MOUNT_SUCCESS)
991 m->result = f;
992
993 mount_set_state(m, MOUNT_MOUNTED);
994 }
995
996 static void mount_enter_dead_or_mounted(Mount *m, MountResult f) {
997 assert(m);
998
999 /* Enter DEAD or MOUNTED state, depending on what the kernel currently says about the mount point. We use this
1000 * whenever we executed an operation, so that our internal state reflects what the kernel says again, after all
1001 * ultimately we just mirror the kernel's internal state on this. */
1002
1003 if (m->from_proc_self_mountinfo)
1004 mount_enter_mounted(m, f);
1005 else
1006 mount_enter_dead(m, f);
1007 }
1008
1009 static int state_to_kill_operation(MountState state) {
1010 switch (state) {
1011
1012 case MOUNT_REMOUNTING_SIGTERM:
1013 return KILL_RESTART;
1014
1015 case MOUNT_UNMOUNTING_SIGTERM:
1016 return KILL_TERMINATE;
1017
1018 case MOUNT_REMOUNTING_SIGKILL:
1019 case MOUNT_UNMOUNTING_SIGKILL:
1020 return KILL_KILL;
1021
1022 default:
1023 return _KILL_OPERATION_INVALID;
1024 }
1025 }
1026
1027 static void mount_enter_signal(Mount *m, MountState state, MountResult f) {
1028 int r;
1029
1030 assert(m);
1031
1032 if (m->result == MOUNT_SUCCESS)
1033 m->result = f;
1034
1035 r = unit_kill_context(
1036 UNIT(m),
1037 &m->kill_context,
1038 state_to_kill_operation(state),
1039 /* main_pid= */ NULL,
1040 &m->control_pid,
1041 /* main_pid_alien= */ false);
1042 if (r < 0)
1043 goto fail;
1044
1045 if (r > 0) {
1046 r = mount_arm_timer(m, usec_add(now(CLOCK_MONOTONIC), m->timeout_usec));
1047 if (r < 0)
1048 goto fail;
1049
1050 mount_set_state(m, state);
1051 } else if (state == MOUNT_REMOUNTING_SIGTERM && m->kill_context.send_sigkill)
1052 mount_enter_signal(m, MOUNT_REMOUNTING_SIGKILL, MOUNT_SUCCESS);
1053 else if (IN_SET(state, MOUNT_REMOUNTING_SIGTERM, MOUNT_REMOUNTING_SIGKILL))
1054 mount_enter_mounted(m, MOUNT_SUCCESS);
1055 else if (state == MOUNT_UNMOUNTING_SIGTERM && m->kill_context.send_sigkill)
1056 mount_enter_signal(m, MOUNT_UNMOUNTING_SIGKILL, MOUNT_SUCCESS);
1057 else
1058 mount_enter_dead_or_mounted(m, MOUNT_SUCCESS);
1059
1060 return;
1061
1062 fail:
1063 log_unit_warning_errno(UNIT(m), r, "Failed to kill processes: %m");
1064 mount_enter_dead_or_mounted(m, MOUNT_FAILURE_RESOURCES);
1065 }
1066
1067 static void mount_enter_unmounting(Mount *m) {
1068 int r;
1069
1070 assert(m);
1071
1072 /* Start counting our attempts */
1073 if (!IN_SET(m->state,
1074 MOUNT_UNMOUNTING,
1075 MOUNT_UNMOUNTING_SIGTERM,
1076 MOUNT_UNMOUNTING_SIGKILL))
1077 m->n_retry_umount = 0;
1078
1079 m->control_command_id = MOUNT_EXEC_UNMOUNT;
1080 m->control_command = m->exec_command + MOUNT_EXEC_UNMOUNT;
1081
1082 r = exec_command_set(m->control_command, UMOUNT_PATH, m->where, "-c", NULL);
1083 if (r >= 0 && m->lazy_unmount)
1084 r = exec_command_append(m->control_command, "-l", NULL);
1085 if (r >= 0 && m->force_unmount)
1086 r = exec_command_append(m->control_command, "-f", NULL);
1087 if (r < 0)
1088 goto fail;
1089
1090 mount_unwatch_control_pid(m);
1091
1092 r = mount_spawn(m, m->control_command, &m->control_pid);
1093 if (r < 0)
1094 goto fail;
1095
1096 mount_set_state(m, MOUNT_UNMOUNTING);
1097
1098 return;
1099
1100 fail:
1101 log_unit_warning_errno(UNIT(m), r, "Failed to run 'umount' task: %m");
1102 mount_enter_dead_or_mounted(m, MOUNT_FAILURE_RESOURCES);
1103 }
1104
1105 static void mount_enter_mounting(Mount *m) {
1106 int r;
1107 MountParameters *p;
1108 bool source_is_dir = true;
1109
1110 assert(m);
1111
1112 r = unit_fail_if_noncanonical(UNIT(m), m->where);
1113 if (r < 0)
1114 goto fail;
1115
1116 p = get_mount_parameters_fragment(m);
1117 if (p && mount_is_bind(p)) {
1118 r = is_dir(p->what, /* follow = */ true);
1119 if (r < 0 && r != -ENOENT)
1120 log_unit_info_errno(UNIT(m), r, "Failed to determine type of bind mount source '%s', ignoring: %m", p->what);
1121 else if (r == 0)
1122 source_is_dir = false;
1123 }
1124
1125 if (source_is_dir)
1126 (void) mkdir_p_label(m->where, m->directory_mode);
1127 else
1128 (void) touch_file(m->where, /* parents = */ true, USEC_INFINITY, UID_INVALID, GID_INVALID, MODE_INVALID);
1129
1130 if (source_is_dir)
1131 unit_warn_if_dir_nonempty(UNIT(m), m->where);
1132 unit_warn_leftover_processes(UNIT(m), unit_log_leftover_process_start);
1133
1134 m->control_command_id = MOUNT_EXEC_MOUNT;
1135 m->control_command = m->exec_command + MOUNT_EXEC_MOUNT;
1136
1137 /* Create the source directory for bind-mounts if needed */
1138 if (p && mount_is_bind(p)) {
1139 r = mkdir_p_label(p->what, m->directory_mode);
1140 /* mkdir_p_label() can return -EEXIST if the target path exists and is not a directory - which is
1141 * totally OK, in case the user wants us to overmount a non-directory inode. Also -EROFS can be
1142 * returned on read-only filesystem. Moreover, -EACCES (and also maybe -EPERM?) may be returned
1143 * when the path is on NFS. See issue #24120. All such errors will be logged in the debug level. */
1144 if (r < 0 && r != -EEXIST)
1145 log_unit_full_errno(UNIT(m),
1146 (r == -EROFS || ERRNO_IS_PRIVILEGE(r)) ? LOG_DEBUG : LOG_WARNING,
1147 r, "Failed to make bind mount source '%s', ignoring: %m", p->what);
1148 }
1149
1150 if (p) {
1151 _cleanup_free_ char *opts = NULL;
1152
1153 r = fstab_filter_options(p->options, "nofail\0" "noauto\0" "auto\0", NULL, NULL, NULL, &opts);
1154 if (r < 0)
1155 goto fail;
1156
1157 r = exec_command_set(m->control_command, MOUNT_PATH, p->what, m->where, NULL);
1158 if (r >= 0 && m->sloppy_options)
1159 r = exec_command_append(m->control_command, "-s", NULL);
1160 if (r >= 0 && m->read_write_only)
1161 r = exec_command_append(m->control_command, "-w", NULL);
1162 if (r >= 0 && p->fstype)
1163 r = exec_command_append(m->control_command, "-t", p->fstype, NULL);
1164 if (r >= 0 && !isempty(opts))
1165 r = exec_command_append(m->control_command, "-o", opts, NULL);
1166 } else
1167 r = -ENOENT;
1168 if (r < 0)
1169 goto fail;
1170
1171 mount_unwatch_control_pid(m);
1172
1173 r = mount_spawn(m, m->control_command, &m->control_pid);
1174 if (r < 0)
1175 goto fail;
1176
1177 mount_set_state(m, MOUNT_MOUNTING);
1178
1179 return;
1180
1181 fail:
1182 log_unit_warning_errno(UNIT(m), r, "Failed to run 'mount' task: %m");
1183 mount_enter_dead_or_mounted(m, MOUNT_FAILURE_RESOURCES);
1184 }
1185
1186 static void mount_set_reload_result(Mount *m, MountResult result) {
1187 assert(m);
1188
1189 /* Only store the first error we encounter */
1190 if (m->reload_result != MOUNT_SUCCESS)
1191 return;
1192
1193 m->reload_result = result;
1194 }
1195
1196 static void mount_enter_remounting(Mount *m) {
1197 int r;
1198 MountParameters *p;
1199
1200 assert(m);
1201
1202 /* Reset reload result when we are about to start a new remount operation */
1203 m->reload_result = MOUNT_SUCCESS;
1204
1205 m->control_command_id = MOUNT_EXEC_REMOUNT;
1206 m->control_command = m->exec_command + MOUNT_EXEC_REMOUNT;
1207
1208 p = get_mount_parameters_fragment(m);
1209 if (p) {
1210 const char *o;
1211
1212 if (p->options)
1213 o = strjoina("remount,", p->options);
1214 else
1215 o = "remount";
1216
1217 r = exec_command_set(m->control_command, MOUNT_PATH,
1218 p->what, m->where,
1219 "-o", o, NULL);
1220 if (r >= 0 && m->sloppy_options)
1221 r = exec_command_append(m->control_command, "-s", NULL);
1222 if (r >= 0 && m->read_write_only)
1223 r = exec_command_append(m->control_command, "-w", NULL);
1224 if (r >= 0 && p->fstype)
1225 r = exec_command_append(m->control_command, "-t", p->fstype, NULL);
1226 } else
1227 r = -ENOENT;
1228 if (r < 0)
1229 goto fail;
1230
1231 mount_unwatch_control_pid(m);
1232
1233 r = mount_spawn(m, m->control_command, &m->control_pid);
1234 if (r < 0)
1235 goto fail;
1236
1237 mount_set_state(m, MOUNT_REMOUNTING);
1238
1239 return;
1240
1241 fail:
1242 log_unit_warning_errno(UNIT(m), r, "Failed to run 'remount' task: %m");
1243 mount_set_reload_result(m, MOUNT_FAILURE_RESOURCES);
1244 mount_enter_dead_or_mounted(m, MOUNT_SUCCESS);
1245 }
1246
1247 static void mount_cycle_clear(Mount *m) {
1248 assert(m);
1249
1250 /* Clear all state we shall forget for this new cycle */
1251
1252 m->result = MOUNT_SUCCESS;
1253 m->reload_result = MOUNT_SUCCESS;
1254 exec_command_reset_status_array(m->exec_command, _MOUNT_EXEC_COMMAND_MAX);
1255 UNIT(m)->reset_accounting = true;
1256 }
1257
1258 static int mount_start(Unit *u) {
1259 Mount *m = MOUNT(u);
1260 int r;
1261
1262 assert(m);
1263
1264 /* We cannot fulfill this request right now, try again later
1265 * please! */
1266 if (IN_SET(m->state,
1267 MOUNT_UNMOUNTING,
1268 MOUNT_UNMOUNTING_SIGTERM,
1269 MOUNT_UNMOUNTING_SIGKILL,
1270 MOUNT_CLEANING))
1271 return -EAGAIN;
1272
1273 /* Already on it! */
1274 if (IN_SET(m->state, MOUNT_MOUNTING, MOUNT_MOUNTING_DONE))
1275 return 0;
1276
1277 assert(IN_SET(m->state, MOUNT_DEAD, MOUNT_FAILED));
1278
1279 r = unit_acquire_invocation_id(u);
1280 if (r < 0)
1281 return r;
1282
1283 mount_cycle_clear(m);
1284 mount_enter_mounting(m);
1285
1286 return 1;
1287 }
1288
1289 static int mount_stop(Unit *u) {
1290 Mount *m = MOUNT(u);
1291
1292 assert(m);
1293
1294 /* When we directly call umount() for a path, then the state of the corresponding mount unit may be
1295 * outdated. Let's re-read mountinfo now and update the state. */
1296 if (m->invalidated_state)
1297 (void) mount_process_proc_self_mountinfo(u->manager);
1298
1299 switch (m->state) {
1300
1301 case MOUNT_UNMOUNTING:
1302 case MOUNT_UNMOUNTING_SIGKILL:
1303 case MOUNT_UNMOUNTING_SIGTERM:
1304 /* Already on it */
1305 return 0;
1306
1307 case MOUNT_MOUNTING:
1308 case MOUNT_MOUNTING_DONE:
1309 case MOUNT_REMOUNTING:
1310 /* If we are still waiting for /bin/mount, we go directly into kill mode. */
1311 mount_enter_signal(m, MOUNT_UNMOUNTING_SIGTERM, MOUNT_SUCCESS);
1312 return 0;
1313
1314 case MOUNT_REMOUNTING_SIGTERM:
1315 /* If we are already waiting for a hung remount, convert this to the matching unmounting state */
1316 mount_set_state(m, MOUNT_UNMOUNTING_SIGTERM);
1317 return 0;
1318
1319 case MOUNT_REMOUNTING_SIGKILL:
1320 /* as above */
1321 mount_set_state(m, MOUNT_UNMOUNTING_SIGKILL);
1322 return 0;
1323
1324 case MOUNT_MOUNTED:
1325 mount_enter_unmounting(m);
1326 return 1;
1327
1328 case MOUNT_CLEANING:
1329 /* If we are currently cleaning, then abort it, brutally. */
1330 mount_enter_signal(m, MOUNT_UNMOUNTING_SIGKILL, MOUNT_SUCCESS);
1331 return 0;
1332
1333 case MOUNT_DEAD:
1334 case MOUNT_FAILED:
1335 /* The mount has just been unmounted by somebody else. */
1336 return 0;
1337
1338 default:
1339 assert_not_reached();
1340 }
1341 }
1342
1343 static int mount_reload(Unit *u) {
1344 Mount *m = MOUNT(u);
1345
1346 assert(m);
1347 assert(m->state == MOUNT_MOUNTED);
1348
1349 mount_enter_remounting(m);
1350
1351 return 1;
1352 }
1353
1354 static int mount_serialize(Unit *u, FILE *f, FDSet *fds) {
1355 Mount *m = MOUNT(u);
1356
1357 assert(m);
1358 assert(f);
1359 assert(fds);
1360
1361 (void) serialize_item(f, "state", mount_state_to_string(m->state));
1362 (void) serialize_item(f, "result", mount_result_to_string(m->result));
1363 (void) serialize_item(f, "reload-result", mount_result_to_string(m->reload_result));
1364 (void) serialize_item_format(f, "n-retry-umount", "%u", m->n_retry_umount);
1365
1366 if (pidref_is_set(&m->control_pid))
1367 (void) serialize_item_format(f, "control-pid", PID_FMT, m->control_pid.pid);
1368
1369 if (m->control_command_id >= 0)
1370 (void) serialize_item(f, "control-command", mount_exec_command_to_string(m->control_command_id));
1371
1372 return 0;
1373 }
1374
1375 static int mount_deserialize_item(Unit *u, const char *key, const char *value, FDSet *fds) {
1376 Mount *m = MOUNT(u);
1377 int r;
1378
1379 assert(m);
1380 assert(u);
1381 assert(key);
1382 assert(value);
1383 assert(fds);
1384
1385 if (streq(key, "state")) {
1386 MountState state;
1387
1388 state = mount_state_from_string(value);
1389 if (state < 0)
1390 log_unit_debug_errno(u, state, "Failed to parse state value: %s", value);
1391 else
1392 m->deserialized_state = state;
1393
1394 } else if (streq(key, "result")) {
1395 MountResult f;
1396
1397 f = mount_result_from_string(value);
1398 if (f < 0)
1399 log_unit_debug_errno(u, f, "Failed to parse result value: %s", value);
1400 else if (f != MOUNT_SUCCESS)
1401 m->result = f;
1402
1403 } else if (streq(key, "reload-result")) {
1404 MountResult f;
1405
1406 f = mount_result_from_string(value);
1407 if (f < 0)
1408 log_unit_debug_errno(u, f, "Failed to parse reload result value: %s", value);
1409 else if (f != MOUNT_SUCCESS)
1410 m->reload_result = f;
1411
1412 } else if (streq(key, "n-retry-umount")) {
1413
1414 r = safe_atou(value, &m->n_retry_umount);
1415 if (r < 0)
1416 log_unit_debug_errno(u, r, "Failed to parse n-retry-umount value: %s", value);
1417
1418 } else if (streq(key, "control-pid")) {
1419
1420 pidref_done(&m->control_pid);
1421 r = pidref_set_pidstr(&m->control_pid, value);
1422 if (r < 0)
1423 log_debug_errno(r, "Failed to set control PID to '%s': %m", value);
1424
1425 } else if (streq(key, "control-command")) {
1426 MountExecCommand id;
1427
1428 id = mount_exec_command_from_string(value);
1429 if (id < 0)
1430 log_unit_debug_errno(u, id, "Failed to parse exec-command value: %s", value);
1431 else {
1432 m->control_command_id = id;
1433 m->control_command = m->exec_command + id;
1434 }
1435 } else
1436 log_unit_debug(u, "Unknown serialization key: %s", key);
1437
1438 return 0;
1439 }
1440
1441 static UnitActiveState mount_active_state(Unit *u) {
1442 assert(u);
1443
1444 return state_translation_table[MOUNT(u)->state];
1445 }
1446
1447 static const char *mount_sub_state_to_string(Unit *u) {
1448 assert(u);
1449
1450 return mount_state_to_string(MOUNT(u)->state);
1451 }
1452
1453 static bool mount_may_gc(Unit *u) {
1454 Mount *m = MOUNT(u);
1455
1456 assert(m);
1457
1458 if (m->from_proc_self_mountinfo)
1459 return false;
1460
1461 return true;
1462 }
1463
1464 static void mount_sigchld_event(Unit *u, pid_t pid, int code, int status) {
1465 Mount *m = MOUNT(u);
1466 MountResult f;
1467
1468 assert(m);
1469 assert(pid >= 0);
1470
1471 if (pid != m->control_pid.pid)
1472 return;
1473
1474 /* So here's the thing, we really want to know before /usr/bin/mount or /usr/bin/umount exit whether
1475 * they established/remove a mount. This is important when mounting, but even more so when unmounting
1476 * since we need to deal with nested mounts and otherwise cannot safely determine whether to repeat
1477 * the unmounts. In theory, the kernel fires /proc/self/mountinfo changes off before returning from
1478 * the mount() or umount() syscalls, and thus we should see the changes to the proc file before we
1479 * process the waitid() for the /usr/bin/(u)mount processes. However, this is unfortunately racy: we
1480 * have to waitid() for processes using P_ALL (since we need to reap unexpected children that got
1481 * reparented to PID 1), but when using P_ALL we might end up reaping processes that terminated just
1482 * instants ago, i.e. already after our last event loop iteration (i.e. after the last point we might
1483 * have noticed /proc/self/mountinfo events via epoll). This means event loop priorities for
1484 * processing SIGCHLD vs. /proc/self/mountinfo IO events are not as relevant as we want. To fix that
1485 * race, let's explicitly scan /proc/self/mountinfo before we start processing /usr/bin/(u)mount
1486 * dying. It's ugly, but it makes our ordering systematic again, and makes sure we always see
1487 * /proc/self/mountinfo changes before our mount/umount exits. */
1488 (void) mount_process_proc_self_mountinfo(u->manager);
1489
1490 pidref_done(&m->control_pid);
1491
1492 if (is_clean_exit(code, status, EXIT_CLEAN_COMMAND, NULL))
1493 f = MOUNT_SUCCESS;
1494 else if (code == CLD_EXITED)
1495 f = MOUNT_FAILURE_EXIT_CODE;
1496 else if (code == CLD_KILLED)
1497 f = MOUNT_FAILURE_SIGNAL;
1498 else if (code == CLD_DUMPED)
1499 f = MOUNT_FAILURE_CORE_DUMP;
1500 else
1501 assert_not_reached();
1502
1503 if (IN_SET(m->state, MOUNT_REMOUNTING, MOUNT_REMOUNTING_SIGKILL, MOUNT_REMOUNTING_SIGTERM))
1504 mount_set_reload_result(m, f);
1505 else if (m->result == MOUNT_SUCCESS)
1506 m->result = f;
1507
1508 if (m->control_command) {
1509 exec_status_exit(&m->control_command->exec_status, &m->exec_context, pid, code, status);
1510
1511 m->control_command = NULL;
1512 m->control_command_id = _MOUNT_EXEC_COMMAND_INVALID;
1513 }
1514
1515 unit_log_process_exit(
1516 u,
1517 "Mount process",
1518 mount_exec_command_to_string(m->control_command_id),
1519 f == MOUNT_SUCCESS,
1520 code, status);
1521
1522 /* Note that due to the io event priority logic, we can be sure the new mountinfo is loaded
1523 * before we process the SIGCHLD for the mount command. */
1524
1525 switch (m->state) {
1526
1527 case MOUNT_MOUNTING:
1528 /* Our mount point has not appeared in mountinfo. Something went wrong. */
1529
1530 if (f == MOUNT_SUCCESS) {
1531 /* Either /bin/mount has an unexpected definition of success,
1532 * or someone raced us and we lost. */
1533 log_unit_warning(UNIT(m), "Mount process finished, but there is no mount.");
1534 f = MOUNT_FAILURE_PROTOCOL;
1535 }
1536 mount_enter_dead(m, f);
1537 break;
1538
1539 case MOUNT_MOUNTING_DONE:
1540 mount_enter_mounted(m, f);
1541 break;
1542
1543 case MOUNT_REMOUNTING:
1544 case MOUNT_REMOUNTING_SIGTERM:
1545 case MOUNT_REMOUNTING_SIGKILL:
1546 mount_enter_dead_or_mounted(m, MOUNT_SUCCESS);
1547 break;
1548
1549 case MOUNT_UNMOUNTING:
1550
1551 if (f == MOUNT_SUCCESS && m->from_proc_self_mountinfo) {
1552
1553 /* Still a mount point? If so, let's try again. Most likely there were multiple mount points
1554 * stacked on top of each other. We might exceed the timeout specified by the user overall,
1555 * but we will stop as soon as any one umount times out. */
1556
1557 if (m->n_retry_umount < RETRY_UMOUNT_MAX) {
1558 log_unit_debug(u, "Mount still present, trying again.");
1559 m->n_retry_umount++;
1560 mount_enter_unmounting(m);
1561 } else {
1562 log_unit_warning(u, "Mount still present after %u attempts to unmount, giving up.", m->n_retry_umount);
1563 mount_enter_mounted(m, f);
1564 }
1565 } else
1566 mount_enter_dead_or_mounted(m, f);
1567
1568 break;
1569
1570 case MOUNT_UNMOUNTING_SIGKILL:
1571 case MOUNT_UNMOUNTING_SIGTERM:
1572 mount_enter_dead_or_mounted(m, f);
1573 break;
1574
1575 case MOUNT_CLEANING:
1576 if (m->clean_result == MOUNT_SUCCESS)
1577 m->clean_result = f;
1578
1579 mount_enter_dead(m, MOUNT_SUCCESS);
1580 break;
1581
1582 default:
1583 assert_not_reached();
1584 }
1585
1586 /* Notify clients about changed exit status */
1587 unit_add_to_dbus_queue(u);
1588 }
1589
1590 static int mount_dispatch_timer(sd_event_source *source, usec_t usec, void *userdata) {
1591 Mount *m = MOUNT(userdata);
1592
1593 assert(m);
1594 assert(m->timer_event_source == source);
1595
1596 switch (m->state) {
1597
1598 case MOUNT_MOUNTING:
1599 case MOUNT_MOUNTING_DONE:
1600 log_unit_warning(UNIT(m), "Mounting timed out. Terminating.");
1601 mount_enter_signal(m, MOUNT_UNMOUNTING_SIGTERM, MOUNT_FAILURE_TIMEOUT);
1602 break;
1603
1604 case MOUNT_REMOUNTING:
1605 log_unit_warning(UNIT(m), "Remounting timed out. Terminating remount process.");
1606 mount_set_reload_result(m, MOUNT_FAILURE_TIMEOUT);
1607 mount_enter_signal(m, MOUNT_REMOUNTING_SIGTERM, MOUNT_SUCCESS);
1608 break;
1609
1610 case MOUNT_REMOUNTING_SIGTERM:
1611 mount_set_reload_result(m, MOUNT_FAILURE_TIMEOUT);
1612
1613 if (m->kill_context.send_sigkill) {
1614 log_unit_warning(UNIT(m), "Remounting timed out. Killing.");
1615 mount_enter_signal(m, MOUNT_REMOUNTING_SIGKILL, MOUNT_SUCCESS);
1616 } else {
1617 log_unit_warning(UNIT(m), "Remounting timed out. Skipping SIGKILL. Ignoring.");
1618 mount_enter_dead_or_mounted(m, MOUNT_SUCCESS);
1619 }
1620 break;
1621
1622 case MOUNT_REMOUNTING_SIGKILL:
1623 mount_set_reload_result(m, MOUNT_FAILURE_TIMEOUT);
1624
1625 log_unit_warning(UNIT(m), "Mount process still around after SIGKILL. Ignoring.");
1626 mount_enter_dead_or_mounted(m, MOUNT_SUCCESS);
1627 break;
1628
1629 case MOUNT_UNMOUNTING:
1630 log_unit_warning(UNIT(m), "Unmounting timed out. Terminating.");
1631 mount_enter_signal(m, MOUNT_UNMOUNTING_SIGTERM, MOUNT_FAILURE_TIMEOUT);
1632 break;
1633
1634 case MOUNT_UNMOUNTING_SIGTERM:
1635 if (m->kill_context.send_sigkill) {
1636 log_unit_warning(UNIT(m), "Mount process timed out. Killing.");
1637 mount_enter_signal(m, MOUNT_UNMOUNTING_SIGKILL, MOUNT_FAILURE_TIMEOUT);
1638 } else {
1639 log_unit_warning(UNIT(m), "Mount process timed out. Skipping SIGKILL. Ignoring.");
1640 mount_enter_dead_or_mounted(m, MOUNT_FAILURE_TIMEOUT);
1641 }
1642 break;
1643
1644 case MOUNT_UNMOUNTING_SIGKILL:
1645 log_unit_warning(UNIT(m), "Mount process still around after SIGKILL. Ignoring.");
1646 mount_enter_dead_or_mounted(m, MOUNT_FAILURE_TIMEOUT);
1647 break;
1648
1649 case MOUNT_CLEANING:
1650 log_unit_warning(UNIT(m), "Cleaning timed out. killing.");
1651
1652 if (m->clean_result == MOUNT_SUCCESS)
1653 m->clean_result = MOUNT_FAILURE_TIMEOUT;
1654
1655 mount_enter_signal(m, MOUNT_UNMOUNTING_SIGKILL, 0);
1656 break;
1657
1658 default:
1659 assert_not_reached();
1660 }
1661
1662 return 0;
1663 }
1664
1665 static int mount_setup_new_unit(
1666 Manager *m,
1667 const char *name,
1668 const char *what,
1669 const char *where,
1670 const char *options,
1671 const char *fstype,
1672 MountProcFlags *ret_flags,
1673 Unit **ret) {
1674
1675 _cleanup_(unit_freep) Unit *u = NULL;
1676 int r;
1677
1678 assert(m);
1679 assert(name);
1680 assert(ret_flags);
1681 assert(ret);
1682
1683 r = unit_new_for_name(m, sizeof(Mount), name, &u);
1684 if (r < 0)
1685 return r;
1686
1687 r = free_and_strdup(&u->source_path, "/proc/self/mountinfo");
1688 if (r < 0)
1689 return r;
1690
1691 r = free_and_strdup(&MOUNT(u)->where, where);
1692 if (r < 0)
1693 return r;
1694
1695 r = update_parameters_proc_self_mountinfo(MOUNT(u), what, options, fstype);
1696 if (r < 0)
1697 return r;
1698
1699 /* This unit was generated because /proc/self/mountinfo reported it. Remember this, so that by the
1700 * time we load the unit file for it (and thus add in extra deps right after) we know what source to
1701 * attributes the deps to. */
1702 MOUNT(u)->from_proc_self_mountinfo = true;
1703
1704 r = mount_add_non_exec_dependencies(MOUNT(u));
1705 if (r < 0)
1706 return r;
1707
1708 /* We have only allocated the stub now, let's enqueue this unit for loading now, so that everything
1709 * else is loaded in now. */
1710 unit_add_to_load_queue(u);
1711
1712 *ret_flags = MOUNT_PROC_IS_MOUNTED | MOUNT_PROC_JUST_MOUNTED | MOUNT_PROC_JUST_CHANGED;
1713 *ret = TAKE_PTR(u);
1714 return 0;
1715 }
1716
1717 static int mount_setup_existing_unit(
1718 Unit *u,
1719 const char *what,
1720 const char *where,
1721 const char *options,
1722 const char *fstype,
1723 MountProcFlags *ret_flags) {
1724
1725 int r;
1726
1727 assert(u);
1728 assert(ret_flags);
1729
1730 if (!MOUNT(u)->where) {
1731 MOUNT(u)->where = strdup(where);
1732 if (!MOUNT(u)->where)
1733 return -ENOMEM;
1734 }
1735
1736 /* In case we have multiple mounts established on the same mount point, let's merge flags set already
1737 * for the current unit. Note that the flags field is reset on each iteration of reading
1738 * /proc/self/mountinfo, hence we know for sure anything already set here is from the current
1739 * iteration and thus worthy of taking into account. */
1740 MountProcFlags flags =
1741 MOUNT(u)->proc_flags | MOUNT_PROC_IS_MOUNTED;
1742
1743 r = update_parameters_proc_self_mountinfo(MOUNT(u), what, options, fstype);
1744 if (r < 0)
1745 return r;
1746 if (r > 0)
1747 flags |= MOUNT_PROC_JUST_CHANGED;
1748
1749 /* There are two conditions when we consider a mount point just mounted: when we haven't seen it in
1750 * /proc/self/mountinfo before or when MOUNT_MOUNTING is our current state. Why bother with the
1751 * latter? Shouldn't that be covered by the former? No, during reload it is not because we might then
1752 * encounter a new /proc/self/mountinfo in combination with an old mount unit state (since it stems
1753 * from the serialized state), and need to catch up. Since we know that the MOUNT_MOUNTING state is
1754 * reached when we wait for the mount to appear we hence can assume that if we are in it, we are
1755 * actually seeing it established for the first time. */
1756 if (!MOUNT(u)->from_proc_self_mountinfo || MOUNT(u)->state == MOUNT_MOUNTING)
1757 flags |= MOUNT_PROC_JUST_MOUNTED;
1758
1759 MOUNT(u)->from_proc_self_mountinfo = true;
1760
1761 if (IN_SET(u->load_state, UNIT_NOT_FOUND, UNIT_BAD_SETTING, UNIT_ERROR)) {
1762 /* The unit was previously not found or otherwise not loaded. Now that the unit shows up in
1763 * /proc/self/mountinfo we should reconsider it this, hence set it to UNIT_LOADED. */
1764 u->load_state = UNIT_LOADED;
1765 u->load_error = 0;
1766
1767 flags |= MOUNT_PROC_JUST_CHANGED;
1768 }
1769
1770 if (FLAGS_SET(flags, MOUNT_PROC_JUST_CHANGED)) {
1771 /* If things changed, then make sure that all deps are regenerated. Let's
1772 * first remove all automatic deps, and then add in the new ones. */
1773 r = mount_add_non_exec_dependencies(MOUNT(u));
1774 if (r < 0)
1775 return r;
1776 }
1777
1778 *ret_flags = flags;
1779 return 0;
1780 }
1781
1782 static int mount_setup_unit(
1783 Manager *m,
1784 const char *what,
1785 const char *where,
1786 const char *options,
1787 const char *fstype,
1788 bool set_flags) {
1789
1790 _cleanup_free_ char *e = NULL;
1791 MountProcFlags flags;
1792 Unit *u;
1793 int r;
1794
1795 assert(m);
1796 assert(what);
1797 assert(where);
1798 assert(options);
1799 assert(fstype);
1800
1801 /* Ignore API mount points. They should never be referenced in
1802 * dependencies ever. */
1803 if (mount_point_is_api(where) || mount_point_ignore(where))
1804 return 0;
1805
1806 if (streq(fstype, "autofs"))
1807 return 0;
1808
1809 /* probably some kind of swap, ignore */
1810 if (!is_path(where))
1811 return 0;
1812
1813 r = unit_name_from_path(where, ".mount", &e);
1814 if (r < 0)
1815 return log_struct_errno(
1816 LOG_WARNING, r,
1817 "MESSAGE_ID=" SD_MESSAGE_MOUNT_POINT_PATH_NOT_SUITABLE_STR,
1818 "MOUNT_POINT=%s", where,
1819 LOG_MESSAGE("Failed to generate valid unit name from mount point path '%s', ignoring mount point: %m",
1820 where));
1821
1822 u = manager_get_unit(m, e);
1823 if (u)
1824 r = mount_setup_existing_unit(u, what, where, options, fstype, &flags);
1825 else
1826 /* First time we see this mount point meaning that it's not been initiated by a mount unit
1827 * but rather by the sysadmin having called mount(8) directly. */
1828 r = mount_setup_new_unit(m, e, what, where, options, fstype, &flags, &u);
1829 if (r < 0)
1830 return log_warning_errno(r, "Failed to set up mount unit for '%s': %m", where);
1831
1832 /* If the mount changed properties or state, let's notify our clients */
1833 if (flags & (MOUNT_PROC_JUST_CHANGED|MOUNT_PROC_JUST_MOUNTED))
1834 unit_add_to_dbus_queue(u);
1835
1836 if (set_flags)
1837 MOUNT(u)->proc_flags = flags;
1838
1839 return 0;
1840 }
1841
1842 static int mount_load_proc_self_mountinfo(Manager *m, bool set_flags) {
1843 _cleanup_(mnt_free_tablep) struct libmnt_table *table = NULL;
1844 _cleanup_(mnt_free_iterp) struct libmnt_iter *iter = NULL;
1845 int r;
1846
1847 assert(m);
1848
1849 r = libmount_parse(NULL, NULL, &table, &iter);
1850 if (r < 0)
1851 return log_error_errno(r, "Failed to parse /proc/self/mountinfo: %m");
1852
1853 for (;;) {
1854 struct libmnt_fs *fs;
1855 const char *device, *path, *options, *fstype;
1856
1857 r = mnt_table_next_fs(table, iter, &fs);
1858 if (r == 1)
1859 break;
1860 if (r < 0)
1861 return log_error_errno(r, "Failed to get next entry from /proc/self/mountinfo: %m");
1862
1863 device = mnt_fs_get_source(fs);
1864 path = mnt_fs_get_target(fs);
1865 options = mnt_fs_get_options(fs);
1866 fstype = mnt_fs_get_fstype(fs);
1867
1868 if (!device || !path)
1869 continue;
1870
1871 device_found_node(m, device, DEVICE_FOUND_MOUNT, DEVICE_FOUND_MOUNT);
1872
1873 (void) mount_setup_unit(m, device, path, options, fstype, set_flags);
1874 }
1875
1876 return 0;
1877 }
1878
1879 static void mount_shutdown(Manager *m) {
1880 assert(m);
1881
1882 m->mount_event_source = sd_event_source_disable_unref(m->mount_event_source);
1883
1884 mnt_unref_monitor(m->mount_monitor);
1885 m->mount_monitor = NULL;
1886 }
1887
1888 static int mount_get_timeout(Unit *u, usec_t *timeout) {
1889 Mount *m = MOUNT(u);
1890 usec_t t;
1891 int r;
1892
1893 assert(m);
1894 assert(u);
1895
1896 if (!m->timer_event_source)
1897 return 0;
1898
1899 r = sd_event_source_get_time(m->timer_event_source, &t);
1900 if (r < 0)
1901 return r;
1902 if (t == USEC_INFINITY)
1903 return 0;
1904
1905 *timeout = t;
1906 return 1;
1907 }
1908
1909 static void mount_enumerate_perpetual(Manager *m) {
1910 Unit *u;
1911 int r;
1912
1913 assert(m);
1914
1915 /* Whatever happens, we know for sure that the root directory is around, and cannot go away. Let's
1916 * unconditionally synthesize it here and mark it as perpetual. */
1917
1918 u = manager_get_unit(m, SPECIAL_ROOT_MOUNT);
1919 if (!u) {
1920 r = unit_new_for_name(m, sizeof(Mount), SPECIAL_ROOT_MOUNT, &u);
1921 if (r < 0) {
1922 log_error_errno(r, "Failed to allocate the special " SPECIAL_ROOT_MOUNT " unit: %m");
1923 return;
1924 }
1925 }
1926
1927 u->perpetual = true;
1928 MOUNT(u)->deserialized_state = MOUNT_MOUNTED;
1929
1930 unit_add_to_load_queue(u);
1931 unit_add_to_dbus_queue(u);
1932 }
1933
1934 static bool mount_is_mounted(Mount *m) {
1935 assert(m);
1936
1937 return UNIT(m)->perpetual || FLAGS_SET(m->proc_flags, MOUNT_PROC_IS_MOUNTED);
1938 }
1939
1940 static int mount_on_ratelimit_expire(sd_event_source *s, void *userdata) {
1941 Manager *m = ASSERT_PTR(userdata);
1942 Job *j;
1943
1944 /* Let's enqueue all start jobs that were previously skipped because of active ratelimit. */
1945 HASHMAP_FOREACH(j, m->jobs) {
1946 if (j->unit->type != UNIT_MOUNT)
1947 continue;
1948
1949 job_add_to_run_queue(j);
1950 }
1951
1952 /* By entering ratelimited state we made all mount start jobs not runnable, now rate limit is over so
1953 * let's make sure we dispatch them in the next iteration. */
1954 manager_trigger_run_queue(m);
1955
1956 return 0;
1957 }
1958
1959 static void mount_enumerate(Manager *m) {
1960 int r;
1961
1962 assert(m);
1963
1964 mnt_init_debug(0);
1965
1966 if (!m->mount_monitor) {
1967 unsigned mount_rate_limit_burst = 5;
1968 int fd;
1969
1970 m->mount_monitor = mnt_new_monitor();
1971 if (!m->mount_monitor) {
1972 log_oom();
1973 goto fail;
1974 }
1975
1976 r = mnt_monitor_enable_kernel(m->mount_monitor, 1);
1977 if (r < 0) {
1978 log_error_errno(r, "Failed to enable watching of kernel mount events: %m");
1979 goto fail;
1980 }
1981
1982 r = mnt_monitor_enable_userspace(m->mount_monitor, 1, NULL);
1983 if (r < 0) {
1984 log_error_errno(r, "Failed to enable watching of userspace mount events: %m");
1985 goto fail;
1986 }
1987
1988 /* mnt_unref_monitor() will close the fd */
1989 fd = r = mnt_monitor_get_fd(m->mount_monitor);
1990 if (r < 0) {
1991 log_error_errno(r, "Failed to acquire watch file descriptor: %m");
1992 goto fail;
1993 }
1994
1995 r = sd_event_add_io(m->event, &m->mount_event_source, fd, EPOLLIN, mount_dispatch_io, m);
1996 if (r < 0) {
1997 log_error_errno(r, "Failed to watch mount file descriptor: %m");
1998 goto fail;
1999 }
2000
2001 r = sd_event_source_set_priority(m->mount_event_source, SD_EVENT_PRIORITY_NORMAL-10);
2002 if (r < 0) {
2003 log_error_errno(r, "Failed to adjust mount watch priority: %m");
2004 goto fail;
2005 }
2006
2007 /* Let users override the default (5 in 1s), as it stalls the boot sequence on busy systems. */
2008 const char *e = secure_getenv("SYSTEMD_DEFAULT_MOUNT_RATE_LIMIT_BURST");
2009 if (e) {
2010 r = safe_atou(e, &mount_rate_limit_burst);
2011 if (r < 0)
2012 log_debug("Invalid value in $SYSTEMD_DEFAULT_MOUNT_RATE_LIMIT_BURST, ignoring: %s", e);
2013 }
2014
2015 r = sd_event_source_set_ratelimit(m->mount_event_source, 1 * USEC_PER_SEC, mount_rate_limit_burst);
2016 if (r < 0) {
2017 log_error_errno(r, "Failed to enable rate limit for mount events: %m");
2018 goto fail;
2019 }
2020
2021 r = sd_event_source_set_ratelimit_expire_callback(m->mount_event_source, mount_on_ratelimit_expire);
2022 if (r < 0) {
2023 log_error_errno(r, "Failed to enable rate limit for mount events: %m");
2024 goto fail;
2025 }
2026
2027 (void) sd_event_source_set_description(m->mount_event_source, "mount-monitor-dispatch");
2028 }
2029
2030 r = mount_load_proc_self_mountinfo(m, false);
2031 if (r < 0)
2032 goto fail;
2033
2034 return;
2035
2036 fail:
2037 mount_shutdown(m);
2038 }
2039
2040 static int drain_libmount(Manager *m) {
2041 bool rescan = false;
2042 int r;
2043
2044 assert(m);
2045
2046 /* Drain all events and verify that the event is valid.
2047 *
2048 * Note that libmount also monitors /run/mount mkdir if the directory does not exist yet. The mkdir
2049 * may generate event which is irrelevant for us.
2050 *
2051 * error: r < 0; valid: r == 0, false positive: r == 1 */
2052 do {
2053 r = mnt_monitor_next_change(m->mount_monitor, NULL, NULL);
2054 if (r < 0)
2055 return log_error_errno(r, "Failed to drain libmount events: %m");
2056 if (r == 0)
2057 rescan = true;
2058 } while (r == 0);
2059
2060 return rescan;
2061 }
2062
2063 static int mount_process_proc_self_mountinfo(Manager *m) {
2064 _cleanup_set_free_ Set *around = NULL, *gone = NULL;
2065 const char *what;
2066 int r;
2067
2068 assert(m);
2069
2070 r = drain_libmount(m);
2071 if (r <= 0)
2072 return r;
2073
2074 r = mount_load_proc_self_mountinfo(m, true);
2075 if (r < 0) {
2076 /* Reset flags, just in case, for later calls */
2077 LIST_FOREACH(units_by_type, u, m->units_by_type[UNIT_MOUNT])
2078 MOUNT(u)->proc_flags = 0;
2079
2080 return 0;
2081 }
2082
2083 manager_dispatch_load_queue(m);
2084
2085 LIST_FOREACH(units_by_type, u, m->units_by_type[UNIT_MOUNT]) {
2086 Mount *mount = MOUNT(u);
2087
2088 mount->invalidated_state = false;
2089
2090 if (!mount_is_mounted(mount)) {
2091
2092 /* A mount point is not around right now. It might be gone, or might never have
2093 * existed. */
2094
2095 if (mount->from_proc_self_mountinfo &&
2096 mount->parameters_proc_self_mountinfo.what)
2097 /* Remember that this device might just have disappeared */
2098 if (set_put_strdup_full(&gone, &path_hash_ops_free, mount->parameters_proc_self_mountinfo.what) < 0)
2099 log_oom(); /* we don't care too much about OOM here... */
2100
2101 mount->from_proc_self_mountinfo = false;
2102 assert_se(update_parameters_proc_self_mountinfo(mount, NULL, NULL, NULL) >= 0);
2103
2104 switch (mount->state) {
2105
2106 case MOUNT_MOUNTED:
2107 /* This has just been unmounted by somebody else, follow the state change. */
2108 mount_enter_dead(mount, MOUNT_SUCCESS);
2109 break;
2110
2111 case MOUNT_MOUNTING_DONE:
2112 /* The mount command may add the corresponding proc mountinfo entry and
2113 * then remove it because of an internal error. E.g., fuse.sshfs seems
2114 * to do that when the connection fails. See #17617. To handle such the
2115 * case, let's once set the state back to mounting. Then, the unit can
2116 * correctly enter the failed state later in mount_sigchld(). */
2117 mount_set_state(mount, MOUNT_MOUNTING);
2118 break;
2119
2120 default:
2121 break;
2122 }
2123
2124 } else if (mount->proc_flags & (MOUNT_PROC_JUST_MOUNTED|MOUNT_PROC_JUST_CHANGED)) {
2125
2126 /* A mount point was added or changed */
2127
2128 switch (mount->state) {
2129
2130 case MOUNT_DEAD:
2131 case MOUNT_FAILED:
2132
2133 /* This has just been mounted by somebody else, follow the state change, but let's
2134 * generate a new invocation ID for this implicitly and automatically. */
2135 (void) unit_acquire_invocation_id(u);
2136 mount_cycle_clear(mount);
2137 mount_enter_mounted(mount, MOUNT_SUCCESS);
2138 break;
2139
2140 case MOUNT_MOUNTING:
2141 mount_set_state(mount, MOUNT_MOUNTING_DONE);
2142 break;
2143
2144 default:
2145 /* Nothing really changed, but let's issue an notification call nonetheless,
2146 * in case somebody is waiting for this. (e.g. file system ro/rw
2147 * remounts.) */
2148 mount_set_state(mount, mount->state);
2149 break;
2150 }
2151 }
2152
2153 if (mount_is_mounted(mount) &&
2154 mount->from_proc_self_mountinfo &&
2155 mount->parameters_proc_self_mountinfo.what)
2156 /* Track devices currently used */
2157 if (set_put_strdup_full(&around, &path_hash_ops_free, mount->parameters_proc_self_mountinfo.what) < 0)
2158 log_oom();
2159
2160 /* Reset the flags for later calls */
2161 mount->proc_flags = 0;
2162 }
2163
2164 SET_FOREACH(what, gone) {
2165 if (set_contains(around, what))
2166 continue;
2167
2168 /* Let the device units know that the device is no longer mounted */
2169 device_found_node(m, what, DEVICE_NOT_FOUND, DEVICE_FOUND_MOUNT);
2170 }
2171
2172 return 0;
2173 }
2174
2175 static int mount_dispatch_io(sd_event_source *source, int fd, uint32_t revents, void *userdata) {
2176 Manager *m = ASSERT_PTR(userdata);
2177
2178 assert(revents & EPOLLIN);
2179
2180 return mount_process_proc_self_mountinfo(m);
2181 }
2182
2183 int mount_invalidate_state_by_path(Manager *manager, const char *path) {
2184 _cleanup_free_ char *name = NULL;
2185 Unit *u;
2186 int r;
2187
2188 assert(manager);
2189 assert(path);
2190
2191 r = unit_name_from_path(path, ".mount", &name);
2192 if (r < 0)
2193 return log_debug_errno(r, "Failed to generate unit name from path \"%s\", ignoring: %m", path);
2194
2195 u = manager_get_unit(manager, name);
2196 if (!u)
2197 return -ENOENT;
2198
2199 MOUNT(u)->invalidated_state = true;
2200 return 0;
2201 }
2202
2203 static void mount_reset_failed(Unit *u) {
2204 Mount *m = MOUNT(u);
2205
2206 assert(m);
2207
2208 if (m->state == MOUNT_FAILED)
2209 mount_set_state(m, MOUNT_DEAD);
2210
2211 m->result = MOUNT_SUCCESS;
2212 m->reload_result = MOUNT_SUCCESS;
2213 m->clean_result = MOUNT_SUCCESS;
2214 }
2215
2216 static PidRef* mount_control_pid(Unit *u) {
2217 return &ASSERT_PTR(MOUNT(u))->control_pid;
2218 }
2219
2220 static int mount_clean(Unit *u, ExecCleanMask mask) {
2221 _cleanup_strv_free_ char **l = NULL;
2222 Mount *m = MOUNT(u);
2223 int r;
2224
2225 assert(m);
2226 assert(mask != 0);
2227
2228 if (m->state != MOUNT_DEAD)
2229 return -EBUSY;
2230
2231 r = exec_context_get_clean_directories(&m->exec_context, u->manager->prefix, mask, &l);
2232 if (r < 0)
2233 return r;
2234
2235 if (strv_isempty(l))
2236 return -EUNATCH;
2237
2238 mount_unwatch_control_pid(m);
2239 m->clean_result = MOUNT_SUCCESS;
2240 m->control_command = NULL;
2241 m->control_command_id = _MOUNT_EXEC_COMMAND_INVALID;
2242
2243 r = mount_arm_timer(m, usec_add(now(CLOCK_MONOTONIC), m->exec_context.timeout_clean_usec));
2244 if (r < 0)
2245 goto fail;
2246
2247 r = unit_fork_and_watch_rm_rf(u, l, &m->control_pid);
2248 if (r < 0)
2249 goto fail;
2250
2251 mount_set_state(m, MOUNT_CLEANING);
2252 return 0;
2253
2254 fail:
2255 log_unit_warning_errno(u, r, "Failed to initiate cleaning: %m");
2256 m->clean_result = MOUNT_FAILURE_RESOURCES;
2257 m->timer_event_source = sd_event_source_disable_unref(m->timer_event_source);
2258 return r;
2259 }
2260
2261 static int mount_can_clean(Unit *u, ExecCleanMask *ret) {
2262 Mount *m = MOUNT(u);
2263
2264 assert(m);
2265
2266 return exec_context_get_clean_mask(&m->exec_context, ret);
2267 }
2268
2269 static int mount_can_start(Unit *u) {
2270 Mount *m = MOUNT(u);
2271 int r;
2272
2273 assert(m);
2274
2275 r = unit_test_start_limit(u);
2276 if (r < 0) {
2277 mount_enter_dead(m, MOUNT_FAILURE_START_LIMIT_HIT);
2278 return r;
2279 }
2280
2281 return 1;
2282 }
2283
2284 static int mount_subsystem_ratelimited(Manager *m) {
2285 assert(m);
2286
2287 if (!m->mount_event_source)
2288 return false;
2289
2290 return sd_event_source_is_ratelimited(m->mount_event_source);
2291 }
2292
2293 static const char* const mount_exec_command_table[_MOUNT_EXEC_COMMAND_MAX] = {
2294 [MOUNT_EXEC_MOUNT] = "ExecMount",
2295 [MOUNT_EXEC_UNMOUNT] = "ExecUnmount",
2296 [MOUNT_EXEC_REMOUNT] = "ExecRemount",
2297 };
2298
2299 DEFINE_STRING_TABLE_LOOKUP(mount_exec_command, MountExecCommand);
2300
2301 static const char* const mount_result_table[_MOUNT_RESULT_MAX] = {
2302 [MOUNT_SUCCESS] = "success",
2303 [MOUNT_FAILURE_RESOURCES] = "resources",
2304 [MOUNT_FAILURE_TIMEOUT] = "timeout",
2305 [MOUNT_FAILURE_EXIT_CODE] = "exit-code",
2306 [MOUNT_FAILURE_SIGNAL] = "signal",
2307 [MOUNT_FAILURE_CORE_DUMP] = "core-dump",
2308 [MOUNT_FAILURE_START_LIMIT_HIT] = "start-limit-hit",
2309 [MOUNT_FAILURE_PROTOCOL] = "protocol",
2310 };
2311
2312 DEFINE_STRING_TABLE_LOOKUP(mount_result, MountResult);
2313
2314 const UnitVTable mount_vtable = {
2315 .object_size = sizeof(Mount),
2316 .exec_context_offset = offsetof(Mount, exec_context),
2317 .cgroup_context_offset = offsetof(Mount, cgroup_context),
2318 .kill_context_offset = offsetof(Mount, kill_context),
2319 .exec_runtime_offset = offsetof(Mount, exec_runtime),
2320
2321 .sections =
2322 "Unit\0"
2323 "Mount\0"
2324 "Install\0",
2325 .private_section = "Mount",
2326
2327 .can_transient = true,
2328 .can_fail = true,
2329 .exclude_from_switch_root_serialization = true,
2330
2331 .init = mount_init,
2332 .load = mount_load,
2333 .done = mount_done,
2334
2335 .coldplug = mount_coldplug,
2336 .catchup = mount_catchup,
2337
2338 .dump = mount_dump,
2339
2340 .start = mount_start,
2341 .stop = mount_stop,
2342 .reload = mount_reload,
2343
2344 .clean = mount_clean,
2345 .can_clean = mount_can_clean,
2346
2347 .serialize = mount_serialize,
2348 .deserialize_item = mount_deserialize_item,
2349
2350 .active_state = mount_active_state,
2351 .sub_state_to_string = mount_sub_state_to_string,
2352
2353 .will_restart = unit_will_restart_default,
2354
2355 .may_gc = mount_may_gc,
2356 .is_extrinsic = mount_is_extrinsic,
2357
2358 .sigchld_event = mount_sigchld_event,
2359
2360 .reset_failed = mount_reset_failed,
2361
2362 .control_pid = mount_control_pid,
2363
2364 .bus_set_property = bus_mount_set_property,
2365 .bus_commit_properties = bus_mount_commit_properties,
2366
2367 .get_timeout = mount_get_timeout,
2368
2369 .enumerate_perpetual = mount_enumerate_perpetual,
2370 .enumerate = mount_enumerate,
2371 .shutdown = mount_shutdown,
2372 .subsystem_ratelimited = mount_subsystem_ratelimited,
2373
2374 .status_message_formats = {
2375 .starting_stopping = {
2376 [0] = "Mounting %s...",
2377 [1] = "Unmounting %s...",
2378 },
2379 .finished_start_job = {
2380 [JOB_DONE] = "Mounted %s.",
2381 [JOB_FAILED] = "Failed to mount %s.",
2382 [JOB_TIMEOUT] = "Timed out mounting %s.",
2383 },
2384 .finished_stop_job = {
2385 [JOB_DONE] = "Unmounted %s.",
2386 [JOB_FAILED] = "Failed unmounting %s.",
2387 [JOB_TIMEOUT] = "Timed out unmounting %s.",
2388 },
2389 },
2390
2391 .can_start = mount_can_start,
2392
2393 .notify_plymouth = true,
2394 };