]>
Commit | Line | Data |
---|---|---|
f13467ec | 1 | /* SPDX-License-Identifier: GPL-2.0-or-later */ |
7fafc032 | 2 | /* |
810adae9 LP |
3 | * Copyright © 2004 Chris Friesen <chris_friesen@sympatico.ca> |
4 | * Copyright © 2009 Canonical Ltd. | |
5 | * Copyright © 2009 Scott James Remnant <scott@netsplit.com> | |
7fafc032 KS |
6 | */ |
7 | ||
7fafc032 | 8 | #include <errno.h> |
618234a5 LP |
9 | #include <fcntl.h> |
10 | #include <getopt.h> | |
618234a5 LP |
11 | #include <stdbool.h> |
12 | #include <stddef.h> | |
7fafc032 KS |
13 | #include <stdio.h> |
14 | #include <stdlib.h> | |
618234a5 | 15 | #include <sys/epoll.h> |
3ebdb81e | 16 | #include <sys/file.h> |
618234a5 LP |
17 | #include <sys/inotify.h> |
18 | #include <sys/ioctl.h> | |
19 | #include <sys/mount.h> | |
1e03b754 | 20 | #include <sys/prctl.h> |
1e03b754 | 21 | #include <sys/signalfd.h> |
dc117daa | 22 | #include <sys/stat.h> |
618234a5 LP |
23 | #include <sys/time.h> |
24 | #include <sys/wait.h> | |
25 | #include <unistd.h> | |
7fafc032 | 26 | |
392ef7a2 | 27 | #include "sd-daemon.h" |
693d371d | 28 | #include "sd-event.h" |
8314de1d | 29 | |
b5efdb8a | 30 | #include "alloc-util.h" |
a1f4fd38 | 31 | #include "cgroup-setup.h" |
194bbe33 | 32 | #include "cgroup-util.h" |
618234a5 | 33 | #include "cpu-set-util.h" |
5ba2dc25 | 34 | #include "dev-setup.h" |
7f2e3a14 | 35 | #include "device-monitor-private.h" |
abde5ea8 | 36 | #include "device-private.h" |
70068602 | 37 | #include "device-util.h" |
6d63048a | 38 | #include "event-util.h" |
3ffd4af2 | 39 | #include "fd-util.h" |
a5c32cff | 40 | #include "fileio.h" |
f97b34a6 | 41 | #include "format-util.h" |
f4f15635 | 42 | #include "fs-util.h" |
a505965d | 43 | #include "hashmap.h" |
9e5fd717 | 44 | #include "inotify-util.h" |
c004493c | 45 | #include "io-util.h" |
eefc66aa | 46 | #include "limits-util.h" |
40a57716 | 47 | #include "list.h" |
0c5a109a | 48 | #include "main-func.h" |
5ea78a39 | 49 | #include "mkdir.h" |
618234a5 | 50 | #include "netlink-util.h" |
6bedfcbb | 51 | #include "parse-util.h" |
a1f4fd38 | 52 | #include "path-util.h" |
294bf0c3 | 53 | #include "pretty-print.h" |
4e731273 | 54 | #include "proc-cmdline.h" |
618234a5 LP |
55 | #include "process-util.h" |
56 | #include "selinux-util.h" | |
57 | #include "signal-util.h" | |
8f328d36 | 58 | #include "socket-util.h" |
07630cea | 59 | #include "string-util.h" |
49fe5c09 | 60 | #include "strv.h" |
5ea78a39 | 61 | #include "strxcpyx.h" |
46f0fbd8 | 62 | #include "syslog-util.h" |
63e2d171 | 63 | #include "udevd.h" |
07a26e42 | 64 | #include "udev-builtin.h" |
7d68eb1b | 65 | #include "udev-ctrl.h" |
25de7aa7 | 66 | #include "udev-event.h" |
618234a5 | 67 | #include "udev-util.h" |
70068602 | 68 | #include "udev-watch.h" |
ee104e11 | 69 | #include "user-util.h" |
47350c5f | 70 | #include "version.h" |
7fafc032 | 71 | |
88bd5a32 FB |
72 | #define WORKER_NUM_MAX 2048U |
73 | ||
bba7a484 TG |
74 | static bool arg_debug = false; |
75 | static int arg_daemonize = false; | |
c4d44cba | 76 | static ResolveNameTiming arg_resolve_name_timing = RESOLVE_NAME_EARLY; |
216e8bbe | 77 | static unsigned arg_children_max = 0; |
6b92f429 | 78 | static usec_t arg_exec_delay_usec = 0; |
bba7a484 | 79 | static usec_t arg_event_timeout_usec = 180 * USEC_PER_SEC; |
e2099267 | 80 | static int arg_timeout_signal = SIGKILL; |
95ac5230 | 81 | static bool arg_blockdev_read_only = false; |
c0c6806b | 82 | |
d9239923 YW |
83 | typedef struct Event Event; |
84 | typedef struct Worker Worker; | |
85 | ||
c0c6806b | 86 | typedef struct Manager { |
693d371d | 87 | sd_event *event; |
c0c6806b | 88 | Hashmap *workers; |
d9239923 | 89 | LIST_HEAD(Event, events); |
a1f4fd38 | 90 | char *cgroup; |
cb49a4f2 | 91 | pid_t pid; /* the process that originally allocated the manager object */ |
1a0bd015 | 92 | int log_level; |
c0c6806b | 93 | |
9a07157d | 94 | UdevRules *rules; |
9b5150b6 | 95 | Hashmap *properties; |
c0c6806b | 96 | |
0bed242c YW |
97 | sd_netlink *rtnl; |
98 | ||
7f2e3a14 | 99 | sd_device_monitor *monitor; |
e0d61dac | 100 | UdevCtrl *ctrl; |
e237d8cb TG |
101 | int worker_watch[2]; |
102 | ||
df7ee959 YW |
103 | /* used by udev-watch */ |
104 | int inotify_fd; | |
693d371d | 105 | sd_event_source *inotify_event; |
df7ee959 | 106 | |
eca195ec | 107 | sd_event_source *kill_workers_event; |
693d371d | 108 | |
7c4c7e89 TG |
109 | usec_t last_usec; |
110 | ||
481f24d1 YW |
111 | bool stop_exec_queue; |
112 | bool exit; | |
c0c6806b | 113 | } Manager; |
1e03b754 | 114 | |
d9239923 | 115 | typedef enum EventState { |
912541b0 KS |
116 | EVENT_UNDEF, |
117 | EVENT_QUEUED, | |
118 | EVENT_RUNNING, | |
d9239923 | 119 | } EventState; |
1e03b754 | 120 | |
d9239923 | 121 | typedef struct Event { |
cb49a4f2 | 122 | Manager *manager; |
d9239923 YW |
123 | Worker *worker; |
124 | EventState state; | |
eb546b35 YW |
125 | |
126 | sd_device *dev; | |
127 | sd_device *dev_kernel; /* clone of originally received device */ | |
128 | ||
129 | uint64_t seqnum; | |
a1fa99d8 | 130 | uint64_t blocker_seqnum; |
d8f462b4 YW |
131 | |
132 | sd_event_source *timeout_warning_event; | |
133 | sd_event_source *timeout_event; | |
eb546b35 | 134 | |
d9239923 YW |
135 | LIST_FIELDS(Event, event); |
136 | } Event; | |
1e03b754 | 137 | |
d9239923 | 138 | typedef enum WorkerState { |
912541b0 KS |
139 | WORKER_UNDEF, |
140 | WORKER_RUNNING, | |
141 | WORKER_IDLE, | |
142 | WORKER_KILLED, | |
f257a8fc | 143 | WORKER_KILLING, |
d9239923 | 144 | } WorkerState; |
1e03b754 | 145 | |
d9239923 | 146 | typedef struct Worker { |
c0c6806b | 147 | Manager *manager; |
912541b0 | 148 | pid_t pid; |
7f2e3a14 | 149 | sd_device_monitor *monitor; |
d9239923 YW |
150 | WorkerState state; |
151 | Event *event; | |
152 | } Worker; | |
1e03b754 KS |
153 | |
154 | /* passed from worker to main process */ | |
d9239923 YW |
155 | typedef struct WorkerMessage { |
156 | } WorkerMessage; | |
1e03b754 | 157 | |
5393c528 | 158 | static Event *event_free(Event *event) { |
c6aa11f2 | 159 | if (!event) |
5393c528 | 160 | return NULL; |
ba47b71c | 161 | |
40a57716 | 162 | assert(event->manager); |
c6aa11f2 | 163 | |
40a57716 | 164 | LIST_REMOVE(event, event->manager->events, event); |
eb546b35 YW |
165 | sd_device_unref(event->dev); |
166 | sd_device_unref(event->dev_kernel); | |
c6aa11f2 | 167 | |
d8f462b4 YW |
168 | sd_event_source_unref(event->timeout_warning_event); |
169 | sd_event_source_unref(event->timeout_event); | |
693d371d | 170 | |
c6aa11f2 TG |
171 | if (event->worker) |
172 | event->worker->event = NULL; | |
173 | ||
ba47b71c YW |
174 | /* only clean up the queue from the process that created it */ |
175 | if (LIST_IS_EMPTY(event->manager->events) && | |
176 | event->manager->pid == getpid_cached()) | |
6be97d67 YW |
177 | if (unlink("/run/udev/queue") < 0 && errno != ENOENT) |
178 | log_warning_errno(errno, "Failed to unlink /run/udev/queue, ignoring: %m"); | |
cb49a4f2 | 179 | |
5393c528 | 180 | return mfree(event); |
aa8734ff | 181 | } |
7a770250 | 182 | |
419ec631 | 183 | static void event_queue_cleanup(Manager *manager, EventState match_state) { |
80a226b2 | 184 | LIST_FOREACH(event, event, manager->events) { |
419ec631 YW |
185 | if (match_state != EVENT_UNDEF && match_state != event->state) |
186 | continue; | |
187 | ||
188 | event_free(event); | |
189 | } | |
190 | } | |
191 | ||
d9239923 | 192 | static Worker *worker_free(Worker *worker) { |
c6aa11f2 | 193 | if (!worker) |
75db809a | 194 | return NULL; |
bc113de9 | 195 | |
c0c6806b TG |
196 | assert(worker->manager); |
197 | ||
4a0b58c4 | 198 | hashmap_remove(worker->manager->workers, PID_TO_PTR(worker->pid)); |
7f2e3a14 | 199 | sd_device_monitor_unref(worker->monitor); |
c6aa11f2 TG |
200 | event_free(worker->event); |
201 | ||
75db809a | 202 | return mfree(worker); |
ff2c503d KS |
203 | } |
204 | ||
d9239923 YW |
205 | DEFINE_TRIVIAL_CLEANUP_FUNC(Worker*, worker_free); |
206 | DEFINE_PRIVATE_HASH_OPS_WITH_VALUE_DESTRUCTOR(worker_hash_op, void, trivial_hash_func, trivial_compare_func, Worker, worker_free); | |
fc465079 | 207 | |
419ec631 YW |
208 | static void manager_clear_for_worker(Manager *manager) { |
209 | assert(manager); | |
210 | ||
211 | manager->inotify_event = sd_event_source_unref(manager->inotify_event); | |
212 | manager->kill_workers_event = sd_event_source_unref(manager->kill_workers_event); | |
213 | ||
214 | manager->event = sd_event_unref(manager->event); | |
215 | ||
216 | manager->workers = hashmap_free(manager->workers); | |
217 | event_queue_cleanup(manager, EVENT_UNDEF); | |
218 | ||
219 | manager->monitor = sd_device_monitor_unref(manager->monitor); | |
220 | manager->ctrl = udev_ctrl_unref(manager->ctrl); | |
221 | ||
222 | manager->worker_watch[READ_END] = safe_close(manager->worker_watch[READ_END]); | |
223 | } | |
224 | ||
225 | static Manager* manager_free(Manager *manager) { | |
226 | if (!manager) | |
227 | return NULL; | |
228 | ||
229 | udev_builtin_exit(); | |
230 | ||
419ec631 YW |
231 | manager_clear_for_worker(manager); |
232 | ||
233 | sd_netlink_unref(manager->rtnl); | |
234 | ||
235 | hashmap_free_free_free(manager->properties); | |
236 | udev_rules_free(manager->rules); | |
237 | ||
238 | safe_close(manager->inotify_fd); | |
239 | safe_close_pair(manager->worker_watch); | |
240 | ||
a1f4fd38 | 241 | free(manager->cgroup); |
419ec631 YW |
242 | return mfree(manager); |
243 | } | |
244 | ||
245 | DEFINE_TRIVIAL_CLEANUP_FUNC(Manager*, manager_free); | |
246 | ||
d9239923 YW |
247 | static int worker_new(Worker **ret, Manager *manager, sd_device_monitor *worker_monitor, pid_t pid) { |
248 | _cleanup_(worker_freep) Worker *worker = NULL; | |
a505965d | 249 | int r; |
3a19b32a TG |
250 | |
251 | assert(ret); | |
c0c6806b | 252 | assert(manager); |
3a19b32a TG |
253 | assert(worker_monitor); |
254 | assert(pid > 1); | |
255 | ||
d4053464 YW |
256 | /* close monitor, but keep address around */ |
257 | device_monitor_disconnect(worker_monitor); | |
258 | ||
d9239923 | 259 | worker = new(Worker, 1); |
3a19b32a TG |
260 | if (!worker) |
261 | return -ENOMEM; | |
262 | ||
d9239923 | 263 | *worker = (Worker) { |
d4053464 YW |
264 | .manager = manager, |
265 | .monitor = sd_device_monitor_ref(worker_monitor), | |
266 | .pid = pid, | |
267 | }; | |
a505965d | 268 | |
92a74c47 | 269 | r = hashmap_ensure_put(&manager->workers, &worker_hash_op, PID_TO_PTR(pid), worker); |
a505965d TG |
270 | if (r < 0) |
271 | return r; | |
272 | ||
ae2a15bc | 273 | *ret = TAKE_PTR(worker); |
3a19b32a TG |
274 | |
275 | return 0; | |
276 | } | |
277 | ||
419ec631 YW |
278 | static void manager_kill_workers(Manager *manager, bool force) { |
279 | Worker *worker; | |
4fa4d885 | 280 | |
419ec631 | 281 | assert(manager); |
4fa4d885 | 282 | |
419ec631 YW |
283 | HASHMAP_FOREACH(worker, manager->workers) { |
284 | if (worker->state == WORKER_KILLED) | |
285 | continue; | |
4fa4d885 | 286 | |
419ec631 YW |
287 | if (worker->state == WORKER_RUNNING && !force) { |
288 | worker->state = WORKER_KILLING; | |
289 | continue; | |
290 | } | |
4fa4d885 | 291 | |
419ec631 YW |
292 | worker->state = WORKER_KILLED; |
293 | (void) kill(worker->pid, SIGTERM); | |
294 | } | |
4fa4d885 TG |
295 | } |
296 | ||
419ec631 YW |
297 | static void manager_exit(Manager *manager) { |
298 | assert(manager); | |
693d371d | 299 | |
419ec631 | 300 | manager->exit = true; |
693d371d | 301 | |
419ec631 YW |
302 | sd_notify(false, |
303 | "STOPPING=1\n" | |
304 | "STATUS=Starting shutdown..."); | |
39c19cf1 | 305 | |
419ec631 YW |
306 | /* close sources of new events and discard buffered events */ |
307 | manager->ctrl = udev_ctrl_unref(manager->ctrl); | |
0bed242c | 308 | |
0bed242c | 309 | manager->inotify_event = sd_event_source_unref(manager->inotify_event); |
419ec631 | 310 | manager->inotify_fd = safe_close(manager->inotify_fd); |
0bed242c | 311 | |
7f2e3a14 | 312 | manager->monitor = sd_device_monitor_unref(manager->monitor); |
0bed242c | 313 | |
419ec631 YW |
314 | /* discard queued events and kill workers */ |
315 | event_queue_cleanup(manager, EVENT_QUEUED); | |
316 | manager_kill_workers(manager, true); | |
0bed242c YW |
317 | } |
318 | ||
4bf4f50f ZJS |
319 | static void notify_ready(void) { |
320 | int r; | |
321 | ||
322 | r = sd_notifyf(false, | |
323 | "READY=1\n" | |
324 | "STATUS=Processing with %u children at max", arg_children_max); | |
325 | if (r < 0) | |
326 | log_warning_errno(r, "Failed to send readiness notification, ignoring: %m"); | |
327 | } | |
328 | ||
419ec631 YW |
329 | /* reload requested, HUP signal received, rules changed, builtin changed */ |
330 | static void manager_reload(Manager *manager) { | |
419ec631 | 331 | assert(manager); |
b2d21d93 | 332 | |
419ec631 YW |
333 | sd_notify(false, |
334 | "RELOADING=1\n" | |
335 | "STATUS=Flushing configuration..."); | |
76e62a4d | 336 | |
419ec631 YW |
337 | manager_kill_workers(manager, false); |
338 | manager->rules = udev_rules_free(manager->rules); | |
339 | udev_builtin_exit(); | |
693d371d | 340 | |
4bf4f50f | 341 | notify_ready(); |
419ec631 | 342 | } |
e237d8cb | 343 | |
419ec631 YW |
344 | static int on_kill_workers_event(sd_event_source *s, uint64_t usec, void *userdata) { |
345 | Manager *manager = userdata; | |
e237d8cb | 346 | |
419ec631 | 347 | assert(manager); |
e237d8cb | 348 | |
419ec631 YW |
349 | log_debug("Cleanup idle workers"); |
350 | manager_kill_workers(manager, false); | |
e237d8cb | 351 | |
419ec631 YW |
352 | return 1; |
353 | } | |
e237d8cb | 354 | |
9a73bd7c | 355 | static int worker_send_message(int fd) { |
d9239923 | 356 | WorkerMessage message = {}; |
9a73bd7c TG |
357 | |
358 | return loop_write(fd, &message, sizeof(message), false); | |
359 | } | |
360 | ||
b97897e3 | 361 | static int worker_lock_block_device(sd_device *dev, int *ret_fd) { |
0bed242c | 362 | _cleanup_close_ int fd = -1; |
b97897e3 YW |
363 | const char *val; |
364 | int r; | |
fee854ee | 365 | |
0bed242c YW |
366 | assert(dev); |
367 | assert(ret_fd); | |
368 | ||
a1130022 LP |
369 | /* Take a shared lock on the device node; this establishes a concept of device "ownership" to |
370 | * serialize device access. External processes holding an exclusive lock will cause udev to skip the | |
371 | * event handling; in the case udev acquired the lock, the external process can block until udev has | |
372 | * finished its event handling. */ | |
373 | ||
374 | if (device_for_action(dev, SD_DEVICE_REMOVE)) | |
0bed242c YW |
375 | return 0; |
376 | ||
b97897e3 YW |
377 | r = sd_device_get_subsystem(dev, &val); |
378 | if (r < 0) | |
379 | return log_device_debug_errno(dev, r, "Failed to get subsystem: %m"); | |
380 | ||
381 | if (!streq(val, "block")) | |
0bed242c | 382 | return 0; |
fee854ee | 383 | |
b97897e3 YW |
384 | r = sd_device_get_sysname(dev, &val); |
385 | if (r < 0) | |
386 | return log_device_debug_errno(dev, r, "Failed to get sysname: %m"); | |
387 | ||
49fe5c09 | 388 | if (STARTSWITH_SET(val, "dm-", "md", "drbd")) |
0bed242c YW |
389 | return 0; |
390 | ||
b97897e3 YW |
391 | r = sd_device_get_devtype(dev, &val); |
392 | if (r < 0 && r != -ENOENT) | |
393 | return log_device_debug_errno(dev, r, "Failed to get devtype: %m"); | |
394 | if (r >= 0 && streq(val, "partition")) { | |
395 | r = sd_device_get_parent(dev, &dev); | |
396 | if (r < 0) | |
397 | return log_device_debug_errno(dev, r, "Failed to get parent device: %m"); | |
398 | } | |
0bed242c | 399 | |
b97897e3 YW |
400 | r = sd_device_get_devname(dev, &val); |
401 | if (r == -ENOENT) | |
0bed242c | 402 | return 0; |
b97897e3 YW |
403 | if (r < 0) |
404 | return log_device_debug_errno(dev, r, "Failed to get devname: %m"); | |
0bed242c | 405 | |
b97897e3 YW |
406 | fd = open(val, O_RDONLY|O_CLOEXEC|O_NOFOLLOW|O_NONBLOCK); |
407 | if (fd < 0) { | |
408 | log_device_debug_errno(dev, errno, "Failed to open '%s', ignoring: %m", val); | |
0bed242c | 409 | return 0; |
b97897e3 | 410 | } |
0bed242c YW |
411 | |
412 | if (flock(fd, LOCK_SH|LOCK_NB) < 0) | |
b97897e3 | 413 | return log_device_debug_errno(dev, errno, "Failed to flock(%s): %m", val); |
0bed242c YW |
414 | |
415 | *ret_fd = TAKE_FD(fd); | |
416 | return 1; | |
fee854ee RK |
417 | } |
418 | ||
95ac5230 LP |
419 | static int worker_mark_block_device_read_only(sd_device *dev) { |
420 | _cleanup_close_ int fd = -1; | |
421 | const char *val; | |
422 | int state = 1, r; | |
423 | ||
424 | assert(dev); | |
425 | ||
426 | if (!arg_blockdev_read_only) | |
427 | return 0; | |
428 | ||
429 | /* Do this only once, when the block device is new. If the device is later retriggered let's not | |
430 | * toggle the bit again, so that people can boot up with full read-only mode and then unset the bit | |
431 | * for specific devices only. */ | |
a1130022 | 432 | if (!device_for_action(dev, SD_DEVICE_ADD)) |
95ac5230 LP |
433 | return 0; |
434 | ||
435 | r = sd_device_get_subsystem(dev, &val); | |
436 | if (r < 0) | |
437 | return log_device_debug_errno(dev, r, "Failed to get subsystem: %m"); | |
438 | ||
439 | if (!streq(val, "block")) | |
440 | return 0; | |
441 | ||
442 | r = sd_device_get_sysname(dev, &val); | |
443 | if (r < 0) | |
444 | return log_device_debug_errno(dev, r, "Failed to get sysname: %m"); | |
445 | ||
446 | /* Exclude synthetic devices for now, this is supposed to be a safety feature to avoid modification | |
447 | * of physical devices, and what sits on top of those doesn't really matter if we don't allow the | |
cb713f16 | 448 | * underlying block devices to receive changes. */ |
95ac5230 LP |
449 | if (STARTSWITH_SET(val, "dm-", "md", "drbd", "loop", "nbd", "zram")) |
450 | return 0; | |
451 | ||
452 | r = sd_device_get_devname(dev, &val); | |
453 | if (r == -ENOENT) | |
454 | return 0; | |
455 | if (r < 0) | |
456 | return log_device_debug_errno(dev, r, "Failed to get devname: %m"); | |
457 | ||
458 | fd = open(val, O_RDONLY|O_CLOEXEC|O_NOFOLLOW|O_NONBLOCK); | |
459 | if (fd < 0) | |
460 | return log_device_debug_errno(dev, errno, "Failed to open '%s', ignoring: %m", val); | |
461 | ||
462 | if (ioctl(fd, BLKROSET, &state) < 0) | |
463 | return log_device_warning_errno(dev, errno, "Failed to mark block device '%s' read-only: %m", val); | |
464 | ||
465 | log_device_info(dev, "Successfully marked block device '%s' read-only.", val); | |
466 | return 0; | |
467 | } | |
468 | ||
abde5ea8 | 469 | static int worker_process_device(Manager *manager, sd_device *dev) { |
2e088715 | 470 | _cleanup_(udev_event_freep) UdevEvent *udev_event = NULL; |
0bed242c YW |
471 | _cleanup_close_ int fd_lock = -1; |
472 | int r; | |
912541b0 | 473 | |
0bed242c YW |
474 | assert(manager); |
475 | assert(dev); | |
476 | ||
b2d9e58f | 477 | log_device_uevent(dev, "Processing device"); |
abde5ea8 | 478 | |
1a0bd015 | 479 | udev_event = udev_event_new(dev, arg_exec_delay_usec, manager->rtnl, manager->log_level); |
0bed242c YW |
480 | if (!udev_event) |
481 | return -ENOMEM; | |
482 | ||
abde5ea8 | 483 | r = worker_lock_block_device(dev, &fd_lock); |
b8380cc6 LP |
484 | if (r == -EAGAIN) { |
485 | /* So this is a block device and the device is locked currently via the BSD advisory locks — | |
486 | * someone else is exclusively using it. This means we don't run our udev rules now, to not | |
487 | * interfere. However we want to know when the device is unlocked again, and retrigger the | |
488 | * device again then, so that the rules are run eventually. For that we use IN_CLOSE_WRITE | |
489 | * inotify watches (which isn't exactly the same as waiting for the BSD locks to release, but | |
490 | * not totally off, as long as unlock+close() is done together, as it usually is). | |
491 | * | |
492 | * (The user-facing side of this: https://systemd.io/BLOCK_DEVICE_LOCKING) | |
493 | * | |
494 | * There's a bit of a chicken and egg problem here for this however: inotify watching is | |
495 | * supposed to be enabled via an option set via udev rules (OPTIONS+="watch"). If we skip the | |
496 | * udev rules here however (as we just said we do), we would thus never see that specific | |
497 | * udev rule, and thus never turn on inotify watching. But in order to catch up eventually | |
ad337e55 | 498 | * and run them we need the inotify watching: hence a classic chicken and egg problem. |
b8380cc6 LP |
499 | * |
500 | * Our way out here: if we see the block device locked, unconditionally watch the device via | |
501 | * inotify, regardless of any explicit request via OPTIONS+="watch". Thus, a device that is | |
502 | * currently locked via the BSD file locks will be treated as if we ran a single udev rule | |
503 | * only for it: the one that turns on inotify watching for it. If we eventually see the | |
504 | * inotify IN_CLOSE_WRITE event, and then run the rules after all and we then realize that | |
505 | * this wasn't actually requested (i.e. no OPTIONS+="watch" set) we'll simply turn off the | |
506 | * watching again (see below). Effectively this means: inotify watching is now enabled either | |
507 | * a) when the udev rules say so, or b) while the device is locked. | |
508 | * | |
509 | * Worst case scenario hence: in the (unlikely) case someone locked the device and we clash | |
510 | * with that we might do inotify watching for a brief moment for a device where we actually | |
511 | * weren't supposed to. But that shouldn't be too bad, in particular as BSD locks being taken | |
512 | * on a block device is kinda an indication that the inotify logic is desired too, to some | |
513 | * degree — they go hand-in-hand after all. */ | |
514 | ||
515 | log_device_debug(dev, "Block device is currently locked, installing watch to wait until the lock is released."); | |
df7ee959 | 516 | (void) udev_watch_begin(manager->inotify_fd, dev); |
b8380cc6 LP |
517 | |
518 | /* Now the watch is installed, let's lock the device again, maybe in the meantime things changed */ | |
519 | r = worker_lock_block_device(dev, &fd_lock); | |
520 | } | |
b6aab8ef | 521 | if (r < 0) |
0bed242c | 522 | return r; |
912541b0 | 523 | |
95ac5230 LP |
524 | (void) worker_mark_block_device_read_only(dev); |
525 | ||
0bed242c | 526 | /* apply rules, create node, symlinks */ |
df7ee959 YW |
527 | r = udev_event_execute_rules( |
528 | udev_event, | |
529 | manager->inotify_fd, | |
530 | arg_event_timeout_usec, | |
531 | arg_timeout_signal, | |
532 | manager->properties, | |
533 | manager->rules); | |
99058cd6 YW |
534 | if (r < 0) |
535 | return r; | |
536 | ||
e2099267 | 537 | udev_event_execute_run(udev_event, arg_event_timeout_usec, arg_timeout_signal); |
2dd9f98d | 538 | |
0bed242c YW |
539 | if (!manager->rtnl) |
540 | /* in case rtnl was initialized */ | |
541 | manager->rtnl = sd_netlink_ref(udev_event->rtnl); | |
912541b0 | 542 | |
0d77bc4a YW |
543 | r = udev_event_process_inotify_watch(udev_event, manager->inotify_fd); |
544 | if (r < 0) | |
545 | return r; | |
912541b0 | 546 | |
b2d9e58f | 547 | log_device_uevent(dev, "Device processed"); |
0bed242c YW |
548 | return 0; |
549 | } | |
912541b0 | 550 | |
e2130348 YW |
551 | static int worker_device_monitor_handler(sd_device_monitor *monitor, sd_device *dev, void *userdata) { |
552 | Manager *manager = userdata; | |
553 | int r; | |
554 | ||
555 | assert(dev); | |
556 | assert(manager); | |
557 | ||
558 | r = worker_process_device(manager, dev); | |
5abee64e LP |
559 | if (r == -EAGAIN) |
560 | /* if we couldn't acquire the flock(), then proceed quietly */ | |
561 | log_device_debug_errno(dev, r, "Device currently locked, not processing."); | |
562 | else { | |
563 | if (r < 0) | |
564 | log_device_warning_errno(dev, r, "Failed to process device, ignoring: %m"); | |
e2130348 | 565 | |
5abee64e LP |
566 | /* send processed event back to libudev listeners */ |
567 | r = device_monitor_send_device(monitor, NULL, dev); | |
568 | if (r < 0) | |
569 | log_device_warning_errno(dev, r, "Failed to send device, ignoring: %m"); | |
570 | } | |
e2130348 YW |
571 | |
572 | /* send udevd the result of the event execution */ | |
573 | r = worker_send_message(manager->worker_watch[WRITE_END]); | |
574 | if (r < 0) | |
575 | log_device_warning_errno(dev, r, "Failed to send signal to main daemon, ignoring: %m"); | |
576 | ||
1a0bd015 | 577 | /* Reset the log level, as it might be changed by "OPTIONS=log_level=". */ |
3cc6b14a | 578 | log_set_max_level(manager->log_level); |
1a0bd015 | 579 | |
e2130348 YW |
580 | return 1; |
581 | } | |
582 | ||
7f2e3a14 YW |
583 | static int worker_main(Manager *_manager, sd_device_monitor *monitor, sd_device *first_device) { |
584 | _cleanup_(sd_device_unrefp) sd_device *dev = first_device; | |
0bed242c | 585 | _cleanup_(manager_freep) Manager *manager = _manager; |
44dcf454 | 586 | int r; |
145dae7e | 587 | |
0bed242c YW |
588 | assert(manager); |
589 | assert(monitor); | |
590 | assert(dev); | |
912541b0 | 591 | |
44ee03d1 | 592 | assert_se(unsetenv("NOTIFY_SOCKET") == 0); |
3b64e4d4 | 593 | |
e2130348 YW |
594 | assert_se(sigprocmask_many(SIG_BLOCK, NULL, SIGTERM, -1) >= 0); |
595 | ||
596 | /* Reset OOM score, we only protect the main daemon. */ | |
597 | r = set_oom_score_adjust(0); | |
598 | if (r < 0) | |
599 | log_debug_errno(r, "Failed to reset OOM score, ignoring: %m"); | |
912541b0 | 600 | |
7802194a | 601 | /* Clear unnecessary data in Manager object. */ |
e2130348 | 602 | manager_clear_for_worker(manager); |
3ebdb81e | 603 | |
e2130348 YW |
604 | r = sd_event_new(&manager->event); |
605 | if (r < 0) | |
606 | return log_error_errno(r, "Failed to allocate event loop: %m"); | |
912541b0 | 607 | |
e2130348 YW |
608 | r = sd_event_add_signal(manager->event, NULL, SIGTERM, NULL, NULL); |
609 | if (r < 0) | |
610 | return log_error_errno(r, "Failed to set SIGTERM event: %m"); | |
4c83d994 | 611 | |
e2130348 YW |
612 | r = sd_device_monitor_attach_event(monitor, manager->event); |
613 | if (r < 0) | |
614 | return log_error_errno(r, "Failed to attach event loop to device monitor: %m"); | |
912541b0 | 615 | |
e2130348 | 616 | r = sd_device_monitor_start(monitor, worker_device_monitor_handler, manager); |
0bed242c | 617 | if (r < 0) |
e2130348 | 618 | return log_error_errno(r, "Failed to start device monitor: %m"); |
912541b0 | 619 | |
e2130348 | 620 | (void) sd_event_source_set_description(sd_device_monitor_get_event_source(monitor), "worker-device-monitor"); |
b66f29a1 | 621 | |
e2130348 YW |
622 | /* Process first device */ |
623 | (void) worker_device_monitor_handler(monitor, dev, manager); | |
912541b0 | 624 | |
e2130348 YW |
625 | r = sd_event_loop(manager->event); |
626 | if (r < 0) | |
627 | return log_error_errno(r, "Event loop failed: %m"); | |
628 | ||
44dcf454 | 629 | return 0; |
0bed242c YW |
630 | } |
631 | ||
419ec631 YW |
632 | static int on_event_timeout(sd_event_source *s, uint64_t usec, void *userdata) { |
633 | Event *event = userdata; | |
634 | ||
635 | assert(event); | |
636 | assert(event->worker); | |
637 | ||
638 | kill_and_sigcont(event->worker->pid, arg_timeout_signal); | |
639 | event->worker->state = WORKER_KILLED; | |
640 | ||
641 | log_device_error(event->dev, "Worker ["PID_FMT"] processing SEQNUM=%"PRIu64" killed", event->worker->pid, event->seqnum); | |
642 | ||
643 | return 1; | |
644 | } | |
645 | ||
646 | static int on_event_timeout_warning(sd_event_source *s, uint64_t usec, void *userdata) { | |
647 | Event *event = userdata; | |
648 | ||
649 | assert(event); | |
650 | assert(event->worker); | |
651 | ||
652 | log_device_warning(event->dev, "Worker ["PID_FMT"] processing SEQNUM=%"PRIu64" is taking a long time", event->worker->pid, event->seqnum); | |
653 | ||
654 | return 1; | |
655 | } | |
656 | ||
657 | static void worker_attach_event(Worker *worker, Event *event) { | |
658 | sd_event *e; | |
659 | ||
660 | assert(worker); | |
661 | assert(worker->manager); | |
662 | assert(event); | |
663 | assert(!event->worker); | |
664 | assert(!worker->event); | |
665 | ||
666 | worker->state = WORKER_RUNNING; | |
667 | worker->event = event; | |
668 | event->state = EVENT_RUNNING; | |
669 | event->worker = worker; | |
670 | ||
671 | e = worker->manager->event; | |
672 | ||
673 | (void) sd_event_add_time_relative(e, &event->timeout_warning_event, CLOCK_MONOTONIC, | |
674 | udev_warn_timeout(arg_event_timeout_usec), USEC_PER_SEC, | |
675 | on_event_timeout_warning, event); | |
676 | ||
677 | (void) sd_event_add_time_relative(e, &event->timeout_event, CLOCK_MONOTONIC, | |
678 | arg_event_timeout_usec, USEC_PER_SEC, | |
679 | on_event_timeout, event); | |
680 | } | |
681 | ||
d9239923 | 682 | static int worker_spawn(Manager *manager, Event *event) { |
7f2e3a14 | 683 | _cleanup_(sd_device_monitor_unrefp) sd_device_monitor *worker_monitor = NULL; |
d9239923 | 684 | Worker *worker; |
0bed242c | 685 | pid_t pid; |
7443654e | 686 | int r; |
0bed242c YW |
687 | |
688 | /* listen for new events */ | |
7f2e3a14 YW |
689 | r = device_monitor_new_full(&worker_monitor, MONITOR_GROUP_NONE, -1); |
690 | if (r < 0) | |
691 | return r; | |
7443654e | 692 | |
0bed242c | 693 | /* allow the main daemon netlink address to send devices to the worker */ |
7f2e3a14 YW |
694 | r = device_monitor_allow_unicast_sender(worker_monitor, manager->monitor); |
695 | if (r < 0) | |
696 | return log_error_errno(r, "Worker: Failed to set unicast sender: %m"); | |
697 | ||
698 | r = device_monitor_enable_receiving(worker_monitor); | |
0bed242c | 699 | if (r < 0) |
7f2e3a14 | 700 | return log_error_errno(r, "Worker: Failed to enable receiving of device: %m"); |
0bed242c | 701 | |
ff86c92e | 702 | r = safe_fork(NULL, FORK_DEATHSIG, &pid); |
7443654e YW |
703 | if (r < 0) { |
704 | event->state = EVENT_QUEUED; | |
705 | return log_error_errno(r, "Failed to fork() worker: %m"); | |
706 | } | |
707 | if (r == 0) { | |
b428efa5 MS |
708 | DEVICE_TRACE_POINT(worker_spawned, event->dev, getpid()); |
709 | ||
7443654e | 710 | /* Worker process */ |
eb546b35 | 711 | r = worker_main(manager, worker_monitor, sd_device_ref(event->dev)); |
baa30fbc | 712 | log_close(); |
8b46c3fc | 713 | _exit(r < 0 ? EXIT_FAILURE : EXIT_SUCCESS); |
912541b0 | 714 | } |
e03c7cc2 | 715 | |
7443654e YW |
716 | r = worker_new(&worker, manager, worker_monitor, pid); |
717 | if (r < 0) | |
718 | return log_error_errno(r, "Failed to create worker object: %m"); | |
e03c7cc2 | 719 | |
7443654e | 720 | worker_attach_event(worker, event); |
39c19cf1 | 721 | |
eb546b35 | 722 | log_device_debug(event->dev, "Worker ["PID_FMT"] is forked for processing SEQNUM=%"PRIu64".", pid, event->seqnum); |
7443654e | 723 | return 0; |
7fafc032 KS |
724 | } |
725 | ||
f2a5412b | 726 | static int event_run(Event *event) { |
5406c368 | 727 | static bool log_children_max_reached = true; |
f2a5412b | 728 | Manager *manager; |
d9239923 | 729 | Worker *worker; |
7f2e3a14 | 730 | int r; |
912541b0 | 731 | |
c0c6806b | 732 | assert(event); |
f2a5412b | 733 | assert(event->manager); |
c0c6806b | 734 | |
b2d9e58f | 735 | log_device_uevent(event->dev, "Device ready for processing"); |
7600dbb8 | 736 | |
f2a5412b | 737 | manager = event->manager; |
90e74a66 | 738 | HASHMAP_FOREACH(worker, manager->workers) { |
912541b0 KS |
739 | if (worker->state != WORKER_IDLE) |
740 | continue; | |
741 | ||
eb546b35 | 742 | r = device_monitor_send_device(manager->monitor, worker->monitor, event->dev); |
7f2e3a14 | 743 | if (r < 0) { |
eb546b35 | 744 | log_device_error_errno(event->dev, r, "Worker ["PID_FMT"] did not accept message, killing the worker: %m", |
7f2e3a14 | 745 | worker->pid); |
cb542e84 | 746 | (void) kill(worker->pid, SIGKILL); |
912541b0 KS |
747 | worker->state = WORKER_KILLED; |
748 | continue; | |
749 | } | |
39c19cf1 | 750 | worker_attach_event(worker, event); |
f2a5412b | 751 | return 1; /* event is now processing. */ |
912541b0 KS |
752 | } |
753 | ||
c0c6806b | 754 | if (hashmap_size(manager->workers) >= arg_children_max) { |
5406c368 FB |
755 | /* Avoid spamming the debug logs if the limit is already reached and |
756 | * many events still need to be processed */ | |
757 | if (log_children_max_reached && arg_children_max > 1) { | |
044497e2 | 758 | log_debug("Maximum number (%u) of children reached.", hashmap_size(manager->workers)); |
5406c368 FB |
759 | log_children_max_reached = false; |
760 | } | |
f2a5412b | 761 | return 0; /* no free worker */ |
912541b0 KS |
762 | } |
763 | ||
5406c368 FB |
764 | /* Re-enable the debug message for the next batch of events */ |
765 | log_children_max_reached = true; | |
766 | ||
2df2152c | 767 | /* fork with up-to-date SELinux label database, so the child inherits the up-to-date db |
f2a5412b | 768 | * and, until the next SELinux policy changes, we safe further reloads in future children */ |
2df2152c CG |
769 | mac_selinux_maybe_reload(); |
770 | ||
912541b0 | 771 | /* start new worker and pass initial device */ |
f2a5412b YW |
772 | r = worker_spawn(manager, event); |
773 | if (r < 0) | |
774 | return r; | |
775 | ||
776 | return 1; /* event is now processing. */ | |
1e03b754 KS |
777 | } |
778 | ||
a1fa99d8 | 779 | static int event_is_blocked(Event *event) { |
eb546b35 YW |
780 | const char *subsystem, *devpath, *devpath_old = NULL; |
781 | dev_t devnum = makedev(0, 0); | |
03677889 | 782 | Event *loop_event = NULL; |
eb546b35 YW |
783 | size_t devpath_len; |
784 | int r, ifindex = 0; | |
785 | bool is_block; | |
786 | ||
a1fa99d8 YW |
787 | /* lookup event for identical, parent, child device */ |
788 | ||
044ac33c YW |
789 | assert(event); |
790 | assert(event->manager); | |
791 | assert(event->blocker_seqnum <= event->seqnum); | |
792 | ||
793 | if (event->blocker_seqnum == event->seqnum) | |
794 | /* we have checked previously and no blocker found */ | |
795 | return false; | |
796 | ||
03677889 YW |
797 | LIST_FOREACH(event, e, event->manager->events) { |
798 | loop_event = e; | |
799 | ||
044ac33c YW |
800 | /* we already found a later event, earlier cannot block us, no need to check again */ |
801 | if (loop_event->seqnum < event->blocker_seqnum) | |
802 | continue; | |
803 | ||
804 | /* event we checked earlier still exists, no need to check again */ | |
805 | if (loop_event->seqnum == event->blocker_seqnum) | |
806 | return true; | |
807 | ||
808 | /* found ourself, no later event can block us */ | |
809 | if (loop_event->seqnum >= event->seqnum) | |
810 | goto no_blocker; | |
811 | ||
812 | /* found event we have not checked */ | |
813 | break; | |
814 | } | |
815 | ||
816 | assert(loop_event); | |
817 | assert(loop_event->seqnum > event->blocker_seqnum && | |
818 | loop_event->seqnum < event->seqnum); | |
819 | ||
eb546b35 YW |
820 | r = sd_device_get_subsystem(event->dev, &subsystem); |
821 | if (r < 0) | |
822 | return r; | |
823 | ||
824 | is_block = streq(subsystem, "block"); | |
825 | ||
826 | r = sd_device_get_devpath(event->dev, &devpath); | |
827 | if (r < 0) | |
828 | return r; | |
829 | ||
830 | devpath_len = strlen(devpath); | |
831 | ||
832 | r = sd_device_get_property_value(event->dev, "DEVPATH_OLD", &devpath_old); | |
833 | if (r < 0 && r != -ENOENT) | |
834 | return r; | |
835 | ||
836 | r = sd_device_get_devnum(event->dev, &devnum); | |
837 | if (r < 0 && r != -ENOENT) | |
838 | return r; | |
839 | ||
840 | r = sd_device_get_ifindex(event->dev, &ifindex); | |
841 | if (r < 0 && r != -ENOENT) | |
842 | return r; | |
912541b0 KS |
843 | |
844 | /* check if queue contains events we depend on */ | |
03677889 | 845 | LIST_FOREACH(event, e, loop_event) { |
eb546b35 YW |
846 | size_t loop_devpath_len, common; |
847 | const char *loop_devpath; | |
848 | ||
03677889 YW |
849 | loop_event = e; |
850 | ||
912541b0 KS |
851 | /* found ourself, no later event can block us */ |
852 | if (loop_event->seqnum >= event->seqnum) | |
044ac33c | 853 | goto no_blocker; |
912541b0 KS |
854 | |
855 | /* check major/minor */ | |
eb546b35 YW |
856 | if (major(devnum) != 0) { |
857 | const char *s; | |
858 | dev_t d; | |
859 | ||
860 | if (sd_device_get_subsystem(loop_event->dev, &s) < 0) | |
861 | continue; | |
862 | ||
863 | if (sd_device_get_devnum(loop_event->dev, &d) >= 0 && | |
864 | devnum == d && is_block == streq(s, "block")) | |
a1fa99d8 | 865 | break; |
eb546b35 | 866 | } |
912541b0 KS |
867 | |
868 | /* check network device ifindex */ | |
eb546b35 YW |
869 | if (ifindex > 0) { |
870 | int i; | |
871 | ||
872 | if (sd_device_get_ifindex(loop_event->dev, &i) >= 0 && | |
873 | ifindex == i) | |
a1fa99d8 | 874 | break; |
eb546b35 YW |
875 | } |
876 | ||
877 | if (sd_device_get_devpath(loop_event->dev, &loop_devpath) < 0) | |
878 | continue; | |
912541b0 KS |
879 | |
880 | /* check our old name */ | |
0bd0407e | 881 | if (devpath_old && streq(devpath_old, loop_devpath)) |
a1fa99d8 | 882 | break; |
912541b0 | 883 | |
eb546b35 YW |
884 | loop_devpath_len = strlen(loop_devpath); |
885 | ||
912541b0 | 886 | /* compare devpath */ |
eb546b35 | 887 | common = MIN(devpath_len, loop_devpath_len); |
912541b0 KS |
888 | |
889 | /* one devpath is contained in the other? */ | |
eb546b35 | 890 | if (!strneq(devpath, loop_devpath, common)) |
912541b0 KS |
891 | continue; |
892 | ||
893 | /* identical device event found */ | |
baa461fc | 894 | if (devpath_len == loop_devpath_len) |
a1fa99d8 | 895 | break; |
912541b0 KS |
896 | |
897 | /* parent device event found */ | |
0bd0407e | 898 | if (devpath[common] == '/') |
a1fa99d8 | 899 | break; |
912541b0 KS |
900 | |
901 | /* child device event found */ | |
0bd0407e | 902 | if (loop_devpath[common] == '/') |
a1fa99d8 | 903 | break; |
912541b0 KS |
904 | } |
905 | ||
a1fa99d8 | 906 | assert(loop_event); |
0bd0407e | 907 | |
7600dbb8 UKK |
908 | log_device_debug(event->dev, "SEQNUM=%" PRIu64 " blocked by SEQNUM=%" PRIu64, |
909 | event->seqnum, loop_event->seqnum); | |
910 | ||
a1fa99d8 | 911 | event->blocker_seqnum = loop_event->seqnum; |
0bd0407e | 912 | return true; |
044ac33c YW |
913 | |
914 | no_blocker: | |
915 | event->blocker_seqnum = event->seqnum; | |
916 | return false; | |
7fafc032 KS |
917 | } |
918 | ||
0744e74c | 919 | static int event_queue_start(Manager *manager) { |
693d371d | 920 | usec_t usec; |
0725c4b9 | 921 | int r; |
8ab44e3f | 922 | |
c0c6806b TG |
923 | assert(manager); |
924 | ||
40a57716 | 925 | if (LIST_IS_EMPTY(manager->events) || |
7c4c7e89 | 926 | manager->exit || manager->stop_exec_queue) |
0744e74c | 927 | return 0; |
7c4c7e89 | 928 | |
3285baa8 | 929 | assert_se(sd_event_now(manager->event, CLOCK_MONOTONIC, &usec) >= 0); |
38a03f06 LP |
930 | /* check for changed config, every 3 seconds at most */ |
931 | if (manager->last_usec == 0 || | |
92fd70ad | 932 | usec > usec_add(manager->last_usec, 3 * USEC_PER_SEC)) { |
38a03f06 | 933 | if (udev_rules_check_timestamp(manager->rules) || |
2024ed61 | 934 | udev_builtin_validate()) |
38a03f06 | 935 | manager_reload(manager); |
693d371d | 936 | |
38a03f06 | 937 | manager->last_usec = usec; |
7c4c7e89 TG |
938 | } |
939 | ||
0725c4b9 YW |
940 | r = event_source_disable(manager->kill_workers_event); |
941 | if (r < 0) | |
942 | log_warning_errno(r, "Failed to disable event source for cleaning up idle workers, ignoring: %m"); | |
eca195ec | 943 | |
2024ed61 | 944 | udev_builtin_init(); |
7c4c7e89 TG |
945 | |
946 | if (!manager->rules) { | |
c238a1f5 | 947 | r = udev_rules_load(&manager->rules, arg_resolve_name_timing); |
0744e74c YW |
948 | if (r < 0) |
949 | return log_warning_errno(r, "Failed to read udev rules: %m"); | |
7c4c7e89 TG |
950 | } |
951 | ||
80a226b2 | 952 | LIST_FOREACH(event, event, manager->events) { |
912541b0 KS |
953 | if (event->state != EVENT_QUEUED) |
954 | continue; | |
0bc74ea7 | 955 | |
c6f78234 YW |
956 | /* do not start event if parent or child event is still running or queued */ |
957 | r = event_is_blocked(event); | |
958 | if (r < 0) { | |
959 | sd_device_action_t a = _SD_DEVICE_ACTION_INVALID; | |
960 | ||
961 | (void) sd_device_get_action(event->dev, &a); | |
962 | log_device_warning_errno(event->dev, r, | |
963 | "Failed to check event dependency, " | |
964 | "skipping event (SEQNUM=%"PRIu64", ACTION=%s)", | |
965 | event->seqnum, | |
966 | strna(device_action_to_string(a))); | |
967 | ||
968 | event_free(event); | |
969 | return r; | |
970 | } | |
971 | if (r > 0) | |
912541b0 | 972 | continue; |
fc465079 | 973 | |
f2a5412b | 974 | r = event_run(event); |
5f4bca9d | 975 | if (r <= 0) |
f2a5412b | 976 | return r; |
912541b0 | 977 | } |
0744e74c YW |
978 | |
979 | return 0; | |
1e03b754 KS |
980 | } |
981 | ||
419ec631 YW |
982 | static int event_queue_insert(Manager *manager, sd_device *dev) { |
983 | _cleanup_(sd_device_unrefp) sd_device *clone = NULL; | |
984 | Event *event; | |
985 | uint64_t seqnum; | |
986 | int r; | |
ff2c503d | 987 | |
419ec631 YW |
988 | assert(manager); |
989 | assert(dev); | |
ff2c503d | 990 | |
419ec631 YW |
991 | /* only one process can add events to the queue */ |
992 | assert(manager->pid == getpid_cached()); | |
993 | ||
994 | /* We only accepts devices received by device monitor. */ | |
995 | r = sd_device_get_seqnum(dev, &seqnum); | |
996 | if (r < 0) | |
997 | return r; | |
998 | ||
999 | /* Save original device to restore the state on failures. */ | |
1000 | r = device_shallow_clone(dev, &clone); | |
1001 | if (r < 0) | |
1002 | return r; | |
1003 | ||
1004 | r = device_copy_properties(clone, dev); | |
1005 | if (r < 0) | |
1006 | return r; | |
1007 | ||
1008 | event = new(Event, 1); | |
1009 | if (!event) | |
1010 | return -ENOMEM; | |
1011 | ||
1012 | *event = (Event) { | |
1013 | .manager = manager, | |
1014 | .dev = sd_device_ref(dev), | |
1015 | .dev_kernel = TAKE_PTR(clone), | |
1016 | .seqnum = seqnum, | |
1017 | .state = EVENT_QUEUED, | |
1018 | }; | |
1019 | ||
1020 | if (LIST_IS_EMPTY(manager->events)) { | |
1021 | r = touch("/run/udev/queue"); | |
1022 | if (r < 0) | |
6be97d67 | 1023 | log_warning_errno(r, "Failed to touch /run/udev/queue, ignoring: %m"); |
912541b0 | 1024 | } |
419ec631 YW |
1025 | |
1026 | LIST_APPEND(event, manager->events, event); | |
1027 | ||
1028 | log_device_uevent(dev, "Device is queued"); | |
1029 | ||
1030 | return 0; | |
1031 | } | |
1032 | ||
1033 | static int on_uevent(sd_device_monitor *monitor, sd_device *dev, void *userdata) { | |
1034 | Manager *manager = userdata; | |
1035 | int r; | |
1036 | ||
1037 | assert(manager); | |
1038 | ||
1039 | DEVICE_TRACE_POINT(kernel_uevent_received, dev); | |
1040 | ||
1041 | device_ensure_usec_initialized(dev, NULL); | |
1042 | ||
1043 | r = event_queue_insert(manager, dev); | |
1044 | if (r < 0) { | |
1045 | log_device_error_errno(dev, r, "Failed to insert device into event queue: %m"); | |
1046 | return 1; | |
1047 | } | |
1048 | ||
1049 | /* we have fresh events, try to schedule them */ | |
1050 | event_queue_start(manager); | |
1051 | ||
1052 | return 1; | |
ff2c503d KS |
1053 | } |
1054 | ||
e82e8fa5 | 1055 | static int on_worker(sd_event_source *s, int fd, uint32_t revents, void *userdata) { |
c0c6806b TG |
1056 | Manager *manager = userdata; |
1057 | ||
1058 | assert(manager); | |
1059 | ||
912541b0 | 1060 | for (;;) { |
d9239923 | 1061 | WorkerMessage msg; |
979558f3 TG |
1062 | struct iovec iovec = { |
1063 | .iov_base = &msg, | |
1064 | .iov_len = sizeof(msg), | |
1065 | }; | |
fb29cdbe | 1066 | CMSG_BUFFER_TYPE(CMSG_SPACE(sizeof(struct ucred))) control; |
979558f3 TG |
1067 | struct msghdr msghdr = { |
1068 | .msg_iov = &iovec, | |
1069 | .msg_iovlen = 1, | |
1070 | .msg_control = &control, | |
1071 | .msg_controllen = sizeof(control), | |
1072 | }; | |
912541b0 | 1073 | ssize_t size; |
371d72e0 | 1074 | struct ucred *ucred; |
d9239923 | 1075 | Worker *worker; |
912541b0 | 1076 | |
3691bcf3 LP |
1077 | size = recvmsg_safe(fd, &msghdr, MSG_DONTWAIT); |
1078 | if (size == -EINTR) | |
1079 | continue; | |
1080 | if (size == -EAGAIN) | |
1081 | /* nothing more to read */ | |
1082 | break; | |
1083 | if (size < 0) | |
1084 | return log_error_errno(size, "Failed to receive message: %m"); | |
1085 | ||
1086 | cmsg_close_all(&msghdr); | |
979558f3 | 1087 | |
d9239923 | 1088 | if (size != sizeof(WorkerMessage)) { |
d4e98880 | 1089 | log_warning("Ignoring worker message with invalid size %zi bytes", size); |
e82e8fa5 | 1090 | continue; |
979558f3 TG |
1091 | } |
1092 | ||
371d72e0 | 1093 | ucred = CMSG_FIND_DATA(&msghdr, SOL_SOCKET, SCM_CREDENTIALS, struct ucred); |
979558f3 | 1094 | if (!ucred || ucred->pid <= 0) { |
d4e98880 | 1095 | log_warning("Ignoring worker message without valid PID"); |
979558f3 TG |
1096 | continue; |
1097 | } | |
912541b0 KS |
1098 | |
1099 | /* lookup worker who sent the signal */ | |
4a0b58c4 | 1100 | worker = hashmap_get(manager->workers, PID_TO_PTR(ucred->pid)); |
a505965d | 1101 | if (!worker) { |
044497e2 | 1102 | log_debug("Worker ["PID_FMT"] returned, but is no longer tracked", ucred->pid); |
a505965d | 1103 | continue; |
912541b0 | 1104 | } |
c0bbfd72 | 1105 | |
f257a8fc YG |
1106 | if (worker->state == WORKER_KILLING) { |
1107 | worker->state = WORKER_KILLED; | |
1108 | (void) kill(worker->pid, SIGTERM); | |
1109 | } else if (worker->state != WORKER_KILLED) | |
a505965d TG |
1110 | worker->state = WORKER_IDLE; |
1111 | ||
1112 | /* worker returned */ | |
1113 | event_free(worker->event); | |
912541b0 | 1114 | } |
e82e8fa5 | 1115 | |
8302fe5a TG |
1116 | /* we have free workers, try to schedule events */ |
1117 | event_queue_start(manager); | |
1118 | ||
e82e8fa5 TG |
1119 | return 1; |
1120 | } | |
1121 | ||
3b47c739 | 1122 | /* receive the udevd message from userspace */ |
e0d61dac | 1123 | static int on_ctrl_msg(UdevCtrl *uctrl, UdevCtrlMessageType type, const UdevCtrlMessageValue *value, void *userdata) { |
c0c6806b | 1124 | Manager *manager = userdata; |
d02c6f54 | 1125 | int r; |
912541b0 | 1126 | |
d02c6f54 | 1127 | assert(value); |
c0c6806b | 1128 | assert(manager); |
e4f66b77 | 1129 | |
d02c6f54 YW |
1130 | switch (type) { |
1131 | case UDEV_CTRL_SET_LOG_LEVEL: | |
64a3494c | 1132 | log_debug("Received udev control message (SET_LOG_LEVEL), setting log_level=%i", value->intval); |
3cc6b14a | 1133 | log_set_max_level(value->intval); |
1a0bd015 | 1134 | manager->log_level = value->intval; |
f257a8fc | 1135 | manager_kill_workers(manager, false); |
d02c6f54 YW |
1136 | break; |
1137 | case UDEV_CTRL_STOP_EXEC_QUEUE: | |
044497e2 | 1138 | log_debug("Received udev control message (STOP_EXEC_QUEUE)"); |
c0c6806b | 1139 | manager->stop_exec_queue = true; |
d02c6f54 YW |
1140 | break; |
1141 | case UDEV_CTRL_START_EXEC_QUEUE: | |
044497e2 | 1142 | log_debug("Received udev control message (START_EXEC_QUEUE)"); |
c0c6806b | 1143 | manager->stop_exec_queue = false; |
8302fe5a | 1144 | event_queue_start(manager); |
d02c6f54 YW |
1145 | break; |
1146 | case UDEV_CTRL_RELOAD: | |
044497e2 | 1147 | log_debug("Received udev control message (RELOAD)"); |
62d43dac | 1148 | manager_reload(manager); |
d02c6f54 YW |
1149 | break; |
1150 | case UDEV_CTRL_SET_ENV: { | |
d7ac0952 FS |
1151 | _unused_ _cleanup_free_ char *old_val = NULL; |
1152 | _cleanup_free_ char *key = NULL, *val = NULL, *old_key = NULL; | |
d02c6f54 | 1153 | const char *eq; |
9b5150b6 | 1154 | |
d02c6f54 | 1155 | eq = strchr(value->buf, '='); |
9b5150b6 | 1156 | if (!eq) { |
d02c6f54 | 1157 | log_error("Invalid key format '%s'", value->buf); |
9b5150b6 YW |
1158 | return 1; |
1159 | } | |
1160 | ||
d02c6f54 | 1161 | key = strndup(value->buf, eq - value->buf); |
9b5150b6 YW |
1162 | if (!key) { |
1163 | log_oom(); | |
1164 | return 1; | |
1165 | } | |
1166 | ||
1167 | old_val = hashmap_remove2(manager->properties, key, (void **) &old_key); | |
1168 | ||
1169 | r = hashmap_ensure_allocated(&manager->properties, &string_hash_ops); | |
1170 | if (r < 0) { | |
1171 | log_oom(); | |
1172 | return 1; | |
912541b0 | 1173 | } |
9b5150b6 YW |
1174 | |
1175 | eq++; | |
f053fc33 | 1176 | if (isempty(eq)) { |
044497e2 | 1177 | log_debug("Received udev control message (ENV), unsetting '%s'", key); |
9b5150b6 YW |
1178 | |
1179 | r = hashmap_put(manager->properties, key, NULL); | |
1180 | if (r < 0) { | |
1181 | log_oom(); | |
1182 | return 1; | |
1183 | } | |
1184 | } else { | |
1185 | val = strdup(eq); | |
1186 | if (!val) { | |
1187 | log_oom(); | |
1188 | return 1; | |
1189 | } | |
1190 | ||
044497e2 | 1191 | log_debug("Received udev control message (ENV), setting '%s=%s'", key, val); |
9b5150b6 YW |
1192 | |
1193 | r = hashmap_put(manager->properties, key, val); | |
1194 | if (r < 0) { | |
1195 | log_oom(); | |
1196 | return 1; | |
1197 | } | |
1198 | } | |
1199 | ||
1200 | key = val = NULL; | |
f257a8fc | 1201 | manager_kill_workers(manager, false); |
d02c6f54 | 1202 | break; |
912541b0 | 1203 | } |
d02c6f54 YW |
1204 | case UDEV_CTRL_SET_CHILDREN_MAX: |
1205 | if (value->intval <= 0) { | |
1206 | log_debug("Received invalid udev control message (SET_MAX_CHILDREN, %i), ignoring.", value->intval); | |
1207 | return 0; | |
1208 | } | |
912541b0 | 1209 | |
d02c6f54 YW |
1210 | log_debug("Received udev control message (SET_MAX_CHILDREN), setting children_max=%i", value->intval); |
1211 | arg_children_max = value->intval; | |
1ef72b55 | 1212 | |
4bf4f50f | 1213 | notify_ready(); |
d02c6f54 YW |
1214 | break; |
1215 | case UDEV_CTRL_PING: | |
d30f43ee | 1216 | log_debug("Received udev control message (PING)"); |
d02c6f54 YW |
1217 | break; |
1218 | case UDEV_CTRL_EXIT: | |
044497e2 | 1219 | log_debug("Received udev control message (EXIT)"); |
62d43dac | 1220 | manager_exit(manager); |
d02c6f54 YW |
1221 | break; |
1222 | default: | |
1223 | log_debug("Received unknown udev control message, ignoring"); | |
912541b0 | 1224 | } |
e4f66b77 | 1225 | |
e82e8fa5 | 1226 | return 1; |
88f4b648 | 1227 | } |
4a231017 | 1228 | |
25677a05 | 1229 | static int synthesize_change_one(sd_device *dev, sd_device *target) { |
0584b17a ZJS |
1230 | int r; |
1231 | ||
25677a05 YW |
1232 | if (DEBUG_LOGGING) { |
1233 | const char *syspath = NULL; | |
1234 | (void) sd_device_get_syspath(target, &syspath); | |
1235 | log_device_debug(dev, "device is closed, synthesising 'change' on %s", strna(syspath)); | |
1236 | } | |
1237 | ||
1238 | r = sd_device_trigger(target, SD_DEVICE_CHANGE); | |
0584b17a | 1239 | if (r < 0) |
25677a05 YW |
1240 | return log_device_debug_errno(target, r, "Failed to trigger 'change' uevent: %m"); |
1241 | ||
b428efa5 MS |
1242 | DEVICE_TRACE_POINT(synthetic_change_event, dev); |
1243 | ||
0584b17a ZJS |
1244 | return 0; |
1245 | } | |
1246 | ||
70068602 | 1247 | static int synthesize_change(sd_device *dev) { |
25677a05 | 1248 | const char *subsystem, *sysname, *devtype; |
f3a740a5 | 1249 | int r; |
edd32000 | 1250 | |
70068602 YW |
1251 | r = sd_device_get_subsystem(dev, &subsystem); |
1252 | if (r < 0) | |
1253 | return r; | |
1254 | ||
25677a05 | 1255 | r = sd_device_get_devtype(dev, &devtype); |
70068602 YW |
1256 | if (r < 0) |
1257 | return r; | |
1258 | ||
25677a05 | 1259 | r = sd_device_get_sysname(dev, &sysname); |
70068602 YW |
1260 | if (r < 0) |
1261 | return r; | |
1262 | ||
25677a05 YW |
1263 | if (streq_ptr(subsystem, "block") && |
1264 | streq_ptr(devtype, "disk") && | |
70068602 YW |
1265 | !startswith(sysname, "dm-")) { |
1266 | _cleanup_(sd_device_enumerator_unrefp) sd_device_enumerator *e = NULL; | |
1267 | bool part_table_read = false, has_partitions = false; | |
25677a05 | 1268 | const char *devname; |
70068602 | 1269 | sd_device *d; |
ede34445 | 1270 | int fd; |
f3a740a5 | 1271 | |
25677a05 YW |
1272 | r = sd_device_get_devname(dev, &devname); |
1273 | if (r < 0) | |
1274 | return r; | |
1275 | ||
1276 | /* Try to re-read the partition table. This only succeeds if none of the devices is | |
1277 | * busy. The kernel returns 0 if no partition table is found, and we will not get an | |
1278 | * event for the disk. */ | |
70068602 | 1279 | fd = open(devname, O_RDONLY|O_CLOEXEC|O_NOFOLLOW|O_NONBLOCK); |
ede34445 | 1280 | if (fd >= 0) { |
02ba8fb3 KS |
1281 | r = flock(fd, LOCK_EX|LOCK_NB); |
1282 | if (r >= 0) | |
1283 | r = ioctl(fd, BLKRRPART, 0); | |
1284 | ||
ede34445 KS |
1285 | close(fd); |
1286 | if (r >= 0) | |
e9fc29f4 | 1287 | part_table_read = true; |
ede34445 KS |
1288 | } |
1289 | ||
e9fc29f4 | 1290 | /* search for partitions */ |
70068602 | 1291 | r = sd_device_enumerator_new(&e); |
f3a740a5 KS |
1292 | if (r < 0) |
1293 | return r; | |
1294 | ||
70068602 | 1295 | r = sd_device_enumerator_allow_uninitialized(e); |
f3a740a5 KS |
1296 | if (r < 0) |
1297 | return r; | |
1298 | ||
70068602 | 1299 | r = sd_device_enumerator_add_match_parent(e, dev); |
47a3fa0f TA |
1300 | if (r < 0) |
1301 | return r; | |
e9fc29f4 | 1302 | |
70068602 YW |
1303 | r = sd_device_enumerator_add_match_subsystem(e, "block", true); |
1304 | if (r < 0) | |
1305 | return r; | |
e9fc29f4 | 1306 | |
70068602 YW |
1307 | FOREACH_DEVICE(e, d) { |
1308 | const char *t; | |
e9fc29f4 | 1309 | |
25677a05 | 1310 | if (sd_device_get_devtype(d, &t) < 0 || !streq(t, "partition")) |
e9fc29f4 KS |
1311 | continue; |
1312 | ||
1313 | has_partitions = true; | |
1314 | break; | |
1315 | } | |
1316 | ||
25677a05 YW |
1317 | /* We have partitions and re-read the table, the kernel already sent out a "change" |
1318 | * event for the disk, and "remove/add" for all partitions. */ | |
e9fc29f4 KS |
1319 | if (part_table_read && has_partitions) |
1320 | return 0; | |
1321 | ||
25677a05 YW |
1322 | /* We have partitions but re-reading the partition table did not work, synthesize |
1323 | * "change" for the disk and all partitions. */ | |
1324 | (void) synthesize_change_one(dev, dev); | |
e9fc29f4 | 1325 | |
70068602 | 1326 | FOREACH_DEVICE(e, d) { |
25677a05 | 1327 | const char *t; |
f3a740a5 | 1328 | |
25677a05 | 1329 | if (sd_device_get_devtype(d, &t) < 0 || !streq(t, "partition")) |
f3a740a5 KS |
1330 | continue; |
1331 | ||
25677a05 | 1332 | (void) synthesize_change_one(dev, d); |
f3a740a5 | 1333 | } |
ede34445 | 1334 | |
0584b17a | 1335 | } else |
25677a05 | 1336 | (void) synthesize_change_one(dev, dev); |
ede34445 | 1337 | |
f3a740a5 | 1338 | return 0; |
edd32000 KS |
1339 | } |
1340 | ||
e82e8fa5 | 1341 | static int on_inotify(sd_event_source *s, int fd, uint32_t revents, void *userdata) { |
c0c6806b | 1342 | Manager *manager = userdata; |
0254e944 | 1343 | union inotify_event_buffer buffer; |
f7c1ad4f LP |
1344 | struct inotify_event *e; |
1345 | ssize_t l; | |
0725c4b9 | 1346 | int r; |
912541b0 | 1347 | |
c0c6806b | 1348 | assert(manager); |
e82e8fa5 | 1349 | |
0725c4b9 YW |
1350 | r = event_source_disable(manager->kill_workers_event); |
1351 | if (r < 0) | |
1352 | log_warning_errno(r, "Failed to disable event source for cleaning up idle workers, ignoring: %m"); | |
eca195ec | 1353 | |
e82e8fa5 | 1354 | l = read(fd, &buffer, sizeof(buffer)); |
f7c1ad4f | 1355 | if (l < 0) { |
8add30a0 | 1356 | if (ERRNO_IS_TRANSIENT(errno)) |
e82e8fa5 | 1357 | return 1; |
912541b0 | 1358 | |
f7c1ad4f | 1359 | return log_error_errno(errno, "Failed to read inotify fd: %m"); |
912541b0 KS |
1360 | } |
1361 | ||
f7c1ad4f | 1362 | FOREACH_INOTIFY_EVENT(e, buffer, l) { |
70068602 YW |
1363 | _cleanup_(sd_device_unrefp) sd_device *dev = NULL; |
1364 | const char *devnode; | |
1365 | ||
e7f781e4 YW |
1366 | r = device_new_from_watch_handle(&dev, e->wd); |
1367 | if (r < 0) { | |
1368 | log_debug_errno(r, "Failed to create sd_device object from watch handle, ignoring: %m"); | |
70068602 | 1369 | continue; |
e7f781e4 | 1370 | } |
912541b0 | 1371 | |
70068602 | 1372 | if (sd_device_get_devname(dev, &devnode) < 0) |
edd32000 | 1373 | continue; |
912541b0 | 1374 | |
7fe3324c | 1375 | log_device_debug(dev, "Inotify event: %x for %s", e->mask, devnode); |
da143134 | 1376 | if (e->mask & IN_CLOSE_WRITE) |
e7f781e4 YW |
1377 | (void) synthesize_change(dev); |
1378 | ||
1379 | /* Do not handle IN_IGNORED here. It should be handled by worker in 'remove' uevent; | |
1380 | * udev_event_execute_rules() -> event_execute_rules_on_remove() -> udev_watch_end(). */ | |
912541b0 KS |
1381 | } |
1382 | ||
e82e8fa5 | 1383 | return 1; |
bd284db1 SJR |
1384 | } |
1385 | ||
0561329d | 1386 | static int on_sigterm(sd_event_source *s, const struct signalfd_siginfo *si, void *userdata) { |
c0c6806b TG |
1387 | Manager *manager = userdata; |
1388 | ||
1389 | assert(manager); | |
1390 | ||
62d43dac | 1391 | manager_exit(manager); |
912541b0 | 1392 | |
e82e8fa5 TG |
1393 | return 1; |
1394 | } | |
912541b0 | 1395 | |
0561329d | 1396 | static int on_sighup(sd_event_source *s, const struct signalfd_siginfo *si, void *userdata) { |
c0c6806b TG |
1397 | Manager *manager = userdata; |
1398 | ||
1399 | assert(manager); | |
1400 | ||
62d43dac | 1401 | manager_reload(manager); |
912541b0 | 1402 | |
e82e8fa5 TG |
1403 | return 1; |
1404 | } | |
912541b0 | 1405 | |
e82e8fa5 | 1406 | static int on_sigchld(sd_event_source *s, const struct signalfd_siginfo *si, void *userdata) { |
c0c6806b | 1407 | Manager *manager = userdata; |
0725c4b9 | 1408 | int r; |
c0c6806b TG |
1409 | |
1410 | assert(manager); | |
1411 | ||
e82e8fa5 TG |
1412 | for (;;) { |
1413 | pid_t pid; | |
1414 | int status; | |
d9239923 | 1415 | Worker *worker; |
d1317d02 | 1416 | |
e82e8fa5 TG |
1417 | pid = waitpid(-1, &status, WNOHANG); |
1418 | if (pid <= 0) | |
f29328d6 | 1419 | break; |
e82e8fa5 | 1420 | |
4a0b58c4 | 1421 | worker = hashmap_get(manager->workers, PID_TO_PTR(pid)); |
e82e8fa5 | 1422 | if (!worker) { |
dc7faf2a | 1423 | log_warning("Worker ["PID_FMT"] is unknown, ignoring", pid); |
f29328d6 | 1424 | continue; |
912541b0 | 1425 | } |
e82e8fa5 TG |
1426 | |
1427 | if (WIFEXITED(status)) { | |
1428 | if (WEXITSTATUS(status) == 0) | |
dc7faf2a | 1429 | log_debug("Worker ["PID_FMT"] exited", pid); |
e82e8fa5 | 1430 | else |
dc7faf2a | 1431 | log_warning("Worker ["PID_FMT"] exited with return code %i", pid, WEXITSTATUS(status)); |
33ad742a | 1432 | } else if (WIFSIGNALED(status)) |
dc7faf2a | 1433 | log_warning("Worker ["PID_FMT"] terminated by signal %i (%s)", pid, WTERMSIG(status), signal_to_string(WTERMSIG(status))); |
33ad742a | 1434 | else if (WIFSTOPPED(status)) { |
dc7faf2a | 1435 | log_info("Worker ["PID_FMT"] stopped", pid); |
f29328d6 | 1436 | continue; |
e82e8fa5 | 1437 | } else if (WIFCONTINUED(status)) { |
dc7faf2a | 1438 | log_info("Worker ["PID_FMT"] continued", pid); |
f29328d6 | 1439 | continue; |
e82e8fa5 | 1440 | } else |
dc7faf2a | 1441 | log_warning("Worker ["PID_FMT"] exit with status 0x%04x", pid, status); |
e82e8fa5 | 1442 | |
05e6d9c6 | 1443 | if ((!WIFEXITED(status) || WEXITSTATUS(status) != 0) && worker->event) { |
eb546b35 | 1444 | log_device_error(worker->event->dev, "Worker ["PID_FMT"] failed", pid); |
dc7faf2a | 1445 | |
05e6d9c6 | 1446 | /* delete state from disk */ |
eb546b35 YW |
1447 | device_delete_db(worker->event->dev); |
1448 | device_tag_index(worker->event->dev, NULL, false); | |
dc7faf2a | 1449 | |
030f4571 MW |
1450 | if (manager->monitor) { |
1451 | /* forward kernel event without amending it */ | |
1452 | r = device_monitor_send_device(manager->monitor, NULL, worker->event->dev_kernel); | |
1453 | if (r < 0) | |
1454 | log_device_error_errno(worker->event->dev_kernel, r, "Failed to send back device to kernel: %m"); | |
1455 | } | |
e82e8fa5 TG |
1456 | } |
1457 | ||
1458 | worker_free(worker); | |
912541b0 | 1459 | } |
e82e8fa5 | 1460 | |
8302fe5a TG |
1461 | /* we can start new workers, try to schedule events */ |
1462 | event_queue_start(manager); | |
1463 | ||
eca195ec | 1464 | /* Disable unnecessary cleanup event */ |
0725c4b9 YW |
1465 | if (hashmap_isempty(manager->workers)) { |
1466 | r = event_source_disable(manager->kill_workers_event); | |
1467 | if (r < 0) | |
1468 | log_warning_errno(r, "Failed to disable event source for cleaning up idle workers, ignoring: %m"); | |
1469 | } | |
eca195ec | 1470 | |
e82e8fa5 | 1471 | return 1; |
f27125f9 | 1472 | } |
1473 | ||
693d371d TG |
1474 | static int on_post(sd_event_source *s, void *userdata) { |
1475 | Manager *manager = userdata; | |
693d371d TG |
1476 | |
1477 | assert(manager); | |
1478 | ||
b6107f01 YW |
1479 | if (!LIST_IS_EMPTY(manager->events)) |
1480 | return 1; | |
1481 | ||
1482 | /* There are no pending events. Let's cleanup idle process. */ | |
1483 | ||
1484 | if (!hashmap_isempty(manager->workers)) { | |
1485 | /* There are idle workers */ | |
6d63048a YW |
1486 | (void) event_reset_time(manager->event, &manager->kill_workers_event, CLOCK_MONOTONIC, |
1487 | now(CLOCK_MONOTONIC) + 3 * USEC_PER_SEC, USEC_PER_SEC, | |
1488 | on_kill_workers_event, manager, 0, "kill-workers-event", false); | |
b6107f01 | 1489 | return 1; |
693d371d TG |
1490 | } |
1491 | ||
b6107f01 YW |
1492 | /* There are no idle workers. */ |
1493 | ||
1494 | if (manager->exit) | |
1495 | return sd_event_exit(manager->event, 0); | |
1496 | ||
42670846 YW |
1497 | if (manager->cgroup) |
1498 | /* cleanup possible left-over processes in our cgroup */ | |
1499 | (void) cg_kill(SYSTEMD_CGROUP_CONTROLLER, manager->cgroup, SIGKILL, CGROUP_IGNORE_SELF, NULL, NULL, NULL); | |
b6107f01 | 1500 | |
693d371d TG |
1501 | return 1; |
1502 | } | |
1503 | ||
c4b69e99 | 1504 | static int listen_fds(int *ret_ctrl, int *ret_netlink) { |
fcff1e72 | 1505 | int ctrl_fd = -1, netlink_fd = -1; |
c4b69e99 | 1506 | int fd, n; |
912541b0 | 1507 | |
c4b69e99 YW |
1508 | assert(ret_ctrl); |
1509 | assert(ret_netlink); | |
fcff1e72 | 1510 | |
912541b0 | 1511 | n = sd_listen_fds(true); |
fcff1e72 TG |
1512 | if (n < 0) |
1513 | return n; | |
912541b0 KS |
1514 | |
1515 | for (fd = SD_LISTEN_FDS_START; fd < n + SD_LISTEN_FDS_START; fd++) { | |
c52cff07 | 1516 | if (sd_is_socket(fd, AF_LOCAL, SOCK_SEQPACKET, -1) > 0) { |
fcff1e72 TG |
1517 | if (ctrl_fd >= 0) |
1518 | return -EINVAL; | |
1519 | ctrl_fd = fd; | |
912541b0 KS |
1520 | continue; |
1521 | } | |
1522 | ||
c52cff07 | 1523 | if (sd_is_socket(fd, AF_NETLINK, SOCK_RAW, -1) > 0) { |
fcff1e72 TG |
1524 | if (netlink_fd >= 0) |
1525 | return -EINVAL; | |
1526 | netlink_fd = fd; | |
912541b0 KS |
1527 | continue; |
1528 | } | |
1529 | ||
fcff1e72 | 1530 | return -EINVAL; |
912541b0 KS |
1531 | } |
1532 | ||
c4b69e99 YW |
1533 | *ret_ctrl = ctrl_fd; |
1534 | *ret_netlink = netlink_fd; | |
912541b0 | 1535 | |
912541b0 | 1536 | return 0; |
7459bcdc KS |
1537 | } |
1538 | ||
e6f86cac | 1539 | /* |
3f85ef0f | 1540 | * read the kernel command line, in case we need to get into debug mode |
64a3494c | 1541 | * udev.log_level=<level> syslog priority |
1d84ad94 LP |
1542 | * udev.children_max=<number of workers> events are fully serialized if set to 1 |
1543 | * udev.exec_delay=<number of seconds> delay execution of every executed program | |
1544 | * udev.event_timeout=<number of seconds> seconds to wait before terminating an event | |
95ac5230 | 1545 | * udev.blockdev_read_only<=bool> mark all block devices read-only when they appear |
e6f86cac | 1546 | */ |
96287a49 | 1547 | static int parse_proc_cmdline_item(const char *key, const char *value, void *data) { |
95ac5230 | 1548 | int r; |
e6f86cac | 1549 | |
614a823c | 1550 | assert(key); |
e6f86cac | 1551 | |
64a3494c FB |
1552 | if (proc_cmdline_key_streq(key, "udev.log_level") || |
1553 | proc_cmdline_key_streq(key, "udev.log_priority")) { /* kept for backward compatibility */ | |
1d84ad94 LP |
1554 | |
1555 | if (proc_cmdline_value_missing(key, value)) | |
1556 | return 0; | |
1557 | ||
46f0fbd8 | 1558 | r = log_level_from_string(value); |
92e72467 ZJS |
1559 | if (r >= 0) |
1560 | log_set_max_level(r); | |
1d84ad94 LP |
1561 | |
1562 | } else if (proc_cmdline_key_streq(key, "udev.event_timeout")) { | |
1563 | ||
1564 | if (proc_cmdline_value_missing(key, value)) | |
1565 | return 0; | |
1566 | ||
9d9264ba | 1567 | r = parse_sec(value, &arg_event_timeout_usec); |
1d84ad94 LP |
1568 | |
1569 | } else if (proc_cmdline_key_streq(key, "udev.children_max")) { | |
1570 | ||
1571 | if (proc_cmdline_value_missing(key, value)) | |
1572 | return 0; | |
1573 | ||
020328e1 | 1574 | r = safe_atou(value, &arg_children_max); |
1d84ad94 LP |
1575 | |
1576 | } else if (proc_cmdline_key_streq(key, "udev.exec_delay")) { | |
1577 | ||
1578 | if (proc_cmdline_value_missing(key, value)) | |
1579 | return 0; | |
1580 | ||
6b92f429 | 1581 | r = parse_sec(value, &arg_exec_delay_usec); |
1d84ad94 | 1582 | |
e2099267 | 1583 | } else if (proc_cmdline_key_streq(key, "udev.timeout_signal")) { |
95ac5230 | 1584 | |
e2099267 MS |
1585 | if (proc_cmdline_value_missing(key, value)) |
1586 | return 0; | |
1587 | ||
1588 | r = signal_from_string(value); | |
1589 | if (r > 0) | |
1590 | arg_timeout_signal = r; | |
95ac5230 LP |
1591 | |
1592 | } else if (proc_cmdline_key_streq(key, "udev.blockdev_read_only")) { | |
1593 | ||
1594 | if (!value) | |
1595 | arg_blockdev_read_only = true; | |
1596 | else { | |
1597 | r = parse_boolean(value); | |
1598 | if (r < 0) | |
1599 | log_warning_errno(r, "Failed to parse udev.blockdev-read-only argument, ignoring: %s", value); | |
1600 | else | |
1601 | arg_blockdev_read_only = r; | |
1602 | } | |
1603 | ||
1604 | if (arg_blockdev_read_only) | |
1605 | log_notice("All physical block devices will be marked read-only."); | |
1606 | ||
1607 | return 0; | |
1608 | ||
1609 | } else { | |
1610 | if (startswith(key, "udev.")) | |
1611 | log_warning("Unknown udev kernel command line option \"%s\", ignoring.", key); | |
1612 | ||
1613 | return 0; | |
1614 | } | |
614a823c | 1615 | |
92e72467 ZJS |
1616 | if (r < 0) |
1617 | log_warning_errno(r, "Failed to parse \"%s=%s\", ignoring: %m", key, value); | |
1d84ad94 | 1618 | |
614a823c | 1619 | return 0; |
e6f86cac KS |
1620 | } |
1621 | ||
37ec0fdd LP |
1622 | static int help(void) { |
1623 | _cleanup_free_ char *link = NULL; | |
1624 | int r; | |
1625 | ||
1626 | r = terminal_urlify_man("systemd-udevd.service", "8", &link); | |
1627 | if (r < 0) | |
1628 | return log_oom(); | |
1629 | ||
ed216e1f | 1630 | printf("%s [OPTIONS...]\n\n" |
d1109e12 | 1631 | "Rule-based manager for device events and files.\n\n" |
5ac0162c | 1632 | " -h --help Print this message\n" |
2d19c17e MF |
1633 | " -V --version Print version of the program\n" |
1634 | " -d --daemon Detach and run in the background\n" | |
1635 | " -D --debug Enable debug output\n" | |
1636 | " -c --children-max=INT Set maximum number of workers\n" | |
1637 | " -e --exec-delay=SECONDS Seconds to wait before executing RUN=\n" | |
1638 | " -t --event-timeout=SECONDS Seconds to wait before terminating an event\n" | |
1639 | " -N --resolve-names=early|late|never\n" | |
5ac0162c | 1640 | " When to resolve users and groups\n" |
bc556335 DDM |
1641 | "\nSee the %s for details.\n", |
1642 | program_invocation_short_name, | |
1643 | link); | |
37ec0fdd LP |
1644 | |
1645 | return 0; | |
ed216e1f TG |
1646 | } |
1647 | ||
bba7a484 | 1648 | static int parse_argv(int argc, char *argv[]) { |
e2099267 MS |
1649 | enum { |
1650 | ARG_TIMEOUT_SIGNAL, | |
1651 | }; | |
1652 | ||
912541b0 | 1653 | static const struct option options[] = { |
e2099267 MS |
1654 | { "daemon", no_argument, NULL, 'd' }, |
1655 | { "debug", no_argument, NULL, 'D' }, | |
1656 | { "children-max", required_argument, NULL, 'c' }, | |
1657 | { "exec-delay", required_argument, NULL, 'e' }, | |
1658 | { "event-timeout", required_argument, NULL, 't' }, | |
1659 | { "resolve-names", required_argument, NULL, 'N' }, | |
1660 | { "help", no_argument, NULL, 'h' }, | |
1661 | { "version", no_argument, NULL, 'V' }, | |
1662 | { "timeout-signal", required_argument, NULL, ARG_TIMEOUT_SIGNAL }, | |
912541b0 KS |
1663 | {} |
1664 | }; | |
689a97f5 | 1665 | |
044497e2 | 1666 | int c, r; |
689a97f5 | 1667 | |
bba7a484 TG |
1668 | assert(argc >= 0); |
1669 | assert(argv); | |
912541b0 | 1670 | |
e14b6f21 | 1671 | while ((c = getopt_long(argc, argv, "c:de:Dt:N:hV", options, NULL)) >= 0) { |
bba7a484 | 1672 | switch (c) { |
912541b0 | 1673 | |
912541b0 | 1674 | case 'd': |
bba7a484 | 1675 | arg_daemonize = true; |
912541b0 KS |
1676 | break; |
1677 | case 'c': | |
020328e1 | 1678 | r = safe_atou(optarg, &arg_children_max); |
6f5cf8a8 | 1679 | if (r < 0) |
389f9bf2 | 1680 | log_warning_errno(r, "Failed to parse --children-max= value '%s', ignoring: %m", optarg); |
912541b0 KS |
1681 | break; |
1682 | case 'e': | |
6b92f429 | 1683 | r = parse_sec(optarg, &arg_exec_delay_usec); |
6f5cf8a8 | 1684 | if (r < 0) |
6b92f429 | 1685 | log_warning_errno(r, "Failed to parse --exec-delay= value '%s', ignoring: %m", optarg); |
912541b0 | 1686 | break; |
e2099267 MS |
1687 | case ARG_TIMEOUT_SIGNAL: |
1688 | r = signal_from_string(optarg); | |
1689 | if (r <= 0) | |
1690 | log_warning_errno(r, "Failed to parse --timeout-signal= value '%s', ignoring: %m", optarg); | |
1691 | else | |
1692 | arg_timeout_signal = r; | |
1693 | ||
1694 | break; | |
9719859c | 1695 | case 't': |
9d9264ba | 1696 | r = parse_sec(optarg, &arg_event_timeout_usec); |
f1e8664e | 1697 | if (r < 0) |
9d9264ba | 1698 | log_warning_errno(r, "Failed to parse --event-timeout= value '%s', ignoring: %m", optarg); |
9719859c | 1699 | break; |
912541b0 | 1700 | case 'D': |
bba7a484 | 1701 | arg_debug = true; |
912541b0 | 1702 | break; |
c4d44cba YW |
1703 | case 'N': { |
1704 | ResolveNameTiming t; | |
1705 | ||
1706 | t = resolve_name_timing_from_string(optarg); | |
1707 | if (t < 0) | |
1708 | log_warning("Invalid --resolve-names= value '%s', ignoring.", optarg); | |
1709 | else | |
1710 | arg_resolve_name_timing = t; | |
912541b0 | 1711 | break; |
c4d44cba | 1712 | } |
912541b0 | 1713 | case 'h': |
37ec0fdd | 1714 | return help(); |
912541b0 | 1715 | case 'V': |
681bd2c5 | 1716 | printf("%s\n", GIT_VERSION); |
bba7a484 TG |
1717 | return 0; |
1718 | case '?': | |
1719 | return -EINVAL; | |
912541b0 | 1720 | default: |
04499a70 | 1721 | assert_not_reached(); |
bba7a484 | 1722 | |
912541b0 KS |
1723 | } |
1724 | } | |
1725 | ||
bba7a484 TG |
1726 | return 1; |
1727 | } | |
1728 | ||
a1f4fd38 YW |
1729 | static int create_subcgroup(char **ret) { |
1730 | _cleanup_free_ char *cgroup = NULL, *subcgroup = NULL; | |
1731 | int r; | |
1732 | ||
1733 | if (getppid() != 1) | |
1734 | return log_debug_errno(SYNTHETIC_ERRNO(EOPNOTSUPP), "Not invoked by PID1."); | |
1735 | ||
1736 | r = sd_booted(); | |
1737 | if (r < 0) | |
1738 | return log_debug_errno(r, "Failed to check if systemd is running: %m"); | |
1739 | if (r == 0) | |
1740 | return log_debug_errno(SYNTHETIC_ERRNO(EOPNOTSUPP), "systemd is not running."); | |
1741 | ||
1742 | /* Get our own cgroup, we regularly kill everything udev has left behind. | |
1743 | * We only do this on systemd systems, and only if we are directly spawned | |
1744 | * by PID1. Otherwise we are not guaranteed to have a dedicated cgroup. */ | |
1745 | ||
1746 | r = cg_pid_get_path(SYSTEMD_CGROUP_CONTROLLER, 0, &cgroup); | |
1747 | if (r < 0) { | |
1748 | if (IN_SET(r, -ENOENT, -ENOMEDIUM)) | |
1749 | return log_debug_errno(r, "Dedicated cgroup not found: %m"); | |
1750 | return log_debug_errno(r, "Failed to get cgroup: %m"); | |
1751 | } | |
1752 | ||
1753 | r = cg_get_xattr_bool(SYSTEMD_CGROUP_CONTROLLER, cgroup, "trusted.delegate"); | |
1754 | if (IN_SET(r, 0, -ENODATA)) | |
1755 | return log_debug_errno(SYNTHETIC_ERRNO(EOPNOTSUPP), "The cgroup %s is not delegated to us.", cgroup); | |
1756 | if (r < 0) | |
1757 | return log_debug_errno(r, "Failed to read trusted.delegate attribute: %m"); | |
1758 | ||
1759 | /* We are invoked with our own delegated cgroup tree, let's move us one level down, so that we | |
1760 | * don't collide with the "no processes in inner nodes" rule of cgroups, when the service | |
1761 | * manager invokes the ExecReload= job in the .control/ subcgroup. */ | |
1762 | ||
1763 | subcgroup = path_join(cgroup, "/udev"); | |
1764 | if (!subcgroup) | |
1765 | return log_oom_debug(); | |
1766 | ||
1767 | r = cg_create_and_attach(SYSTEMD_CGROUP_CONTROLLER, subcgroup, 0); | |
1768 | if (r < 0) | |
1769 | return log_debug_errno(r, "Failed to create %s subcgroup: %m", subcgroup); | |
1770 | ||
1771 | log_debug("Created %s subcgroup.", subcgroup); | |
1772 | if (ret) | |
1773 | *ret = TAKE_PTR(subcgroup); | |
1774 | return 0; | |
1775 | } | |
1776 | ||
1777 | static int manager_new(Manager **ret, int fd_ctrl, int fd_uevent) { | |
c0c6806b | 1778 | _cleanup_(manager_freep) Manager *manager = NULL; |
a1f4fd38 | 1779 | _cleanup_free_ char *cgroup = NULL; |
b5af8c8c | 1780 | int r; |
c0c6806b TG |
1781 | |
1782 | assert(ret); | |
1783 | ||
a1f4fd38 YW |
1784 | (void) create_subcgroup(&cgroup); |
1785 | ||
6f19b42f | 1786 | manager = new(Manager, 1); |
c0c6806b TG |
1787 | if (!manager) |
1788 | return log_oom(); | |
1789 | ||
6f19b42f | 1790 | *manager = (Manager) { |
df7ee959 | 1791 | .inotify_fd = -1, |
6f19b42f | 1792 | .worker_watch = { -1, -1 }, |
a1f4fd38 | 1793 | .cgroup = TAKE_PTR(cgroup), |
6f19b42f | 1794 | }; |
e237d8cb | 1795 | |
100bc5bf YW |
1796 | r = udev_ctrl_new_from_fd(&manager->ctrl, fd_ctrl); |
1797 | if (r < 0) | |
1798 | return log_error_errno(r, "Failed to initialize udev control socket: %m"); | |
e237d8cb | 1799 | |
b5af8c8c ZJS |
1800 | r = udev_ctrl_enable_receiving(manager->ctrl); |
1801 | if (r < 0) | |
1802 | return log_error_errno(r, "Failed to bind udev control socket: %m"); | |
c4b69e99 | 1803 | |
7f2e3a14 YW |
1804 | r = device_monitor_new_full(&manager->monitor, MONITOR_GROUP_KERNEL, fd_uevent); |
1805 | if (r < 0) | |
1806 | return log_error_errno(r, "Failed to initialize device monitor: %m"); | |
e237d8cb | 1807 | |
1ffadeaa FB |
1808 | /* Bump receiver buffer, but only if we are not called via socket activation, as in that |
1809 | * case systemd sets the receive buffer size for us, and the value in the .socket unit | |
1810 | * should take full effect. */ | |
e77f52e5 YW |
1811 | if (fd_uevent < 0) { |
1812 | r = sd_device_monitor_set_receive_buffer_size(manager->monitor, 128 * 1024 * 1024); | |
1813 | if (r < 0) | |
1814 | log_warning_errno(r, "Failed to set receive buffer size for device monitor, ignoring: %m"); | |
1815 | } | |
c4b69e99 | 1816 | |
b5af8c8c ZJS |
1817 | r = device_monitor_enable_receiving(manager->monitor); |
1818 | if (r < 0) | |
1819 | return log_error_errno(r, "Failed to bind netlink socket: %m"); | |
1820 | ||
1a0bd015 YW |
1821 | manager->log_level = log_get_max_level(); |
1822 | ||
b5af8c8c ZJS |
1823 | *ret = TAKE_PTR(manager); |
1824 | ||
1825 | return 0; | |
1826 | } | |
1827 | ||
1828 | static int main_loop(Manager *manager) { | |
d02c6f54 | 1829 | int fd_worker, r; |
b5af8c8c | 1830 | |
76e62a4d YW |
1831 | manager->pid = getpid_cached(); |
1832 | ||
e237d8cb TG |
1833 | /* unnamed socket from workers to the main daemon */ |
1834 | r = socketpair(AF_LOCAL, SOCK_DGRAM|SOCK_CLOEXEC, 0, manager->worker_watch); | |
1835 | if (r < 0) | |
044497e2 | 1836 | return log_error_errno(errno, "Failed to create socketpair for communicating with workers: %m"); |
e237d8cb | 1837 | |
693d371d | 1838 | fd_worker = manager->worker_watch[READ_END]; |
e237d8cb | 1839 | |
2ff48e98 | 1840 | r = setsockopt_int(fd_worker, SOL_SOCKET, SO_PASSCRED, true); |
e237d8cb | 1841 | if (r < 0) |
044497e2 | 1842 | return log_error_errno(r, "Failed to enable SO_PASSCRED: %m"); |
e237d8cb | 1843 | |
df7ee959 YW |
1844 | manager->inotify_fd = inotify_init1(IN_CLOEXEC); |
1845 | if (manager->inotify_fd < 0) | |
1846 | return log_error_errno(errno, "Failed to create inotify descriptor: %m"); | |
e237d8cb | 1847 | |
df7ee959 | 1848 | udev_watch_restore(manager->inotify_fd); |
e237d8cb TG |
1849 | |
1850 | /* block and listen to all signals on signalfd */ | |
72c0a2c2 | 1851 | assert_se(sigprocmask_many(SIG_BLOCK, NULL, SIGTERM, SIGINT, SIGHUP, SIGCHLD, -1) >= 0); |
693d371d | 1852 | |
49f997f3 TG |
1853 | r = sd_event_default(&manager->event); |
1854 | if (r < 0) | |
044497e2 | 1855 | return log_error_errno(r, "Failed to allocate event loop: %m"); |
49f997f3 | 1856 | |
693d371d TG |
1857 | r = sd_event_add_signal(manager->event, NULL, SIGINT, on_sigterm, manager); |
1858 | if (r < 0) | |
044497e2 | 1859 | return log_error_errno(r, "Failed to create SIGINT event source: %m"); |
693d371d TG |
1860 | |
1861 | r = sd_event_add_signal(manager->event, NULL, SIGTERM, on_sigterm, manager); | |
1862 | if (r < 0) | |
044497e2 | 1863 | return log_error_errno(r, "Failed to create SIGTERM event source: %m"); |
693d371d TG |
1864 | |
1865 | r = sd_event_add_signal(manager->event, NULL, SIGHUP, on_sighup, manager); | |
1866 | if (r < 0) | |
044497e2 | 1867 | return log_error_errno(r, "Failed to create SIGHUP event source: %m"); |
693d371d TG |
1868 | |
1869 | r = sd_event_add_signal(manager->event, NULL, SIGCHLD, on_sigchld, manager); | |
1870 | if (r < 0) | |
044497e2 | 1871 | return log_error_errno(r, "Failed to create SIGCHLD event source: %m"); |
693d371d TG |
1872 | |
1873 | r = sd_event_set_watchdog(manager->event, true); | |
1874 | if (r < 0) | |
044497e2 | 1875 | return log_error_errno(r, "Failed to create watchdog event source: %m"); |
693d371d | 1876 | |
d02c6f54 YW |
1877 | r = udev_ctrl_attach_event(manager->ctrl, manager->event); |
1878 | if (r < 0) | |
1879 | return log_error_errno(r, "Failed to attach event to udev control: %m"); | |
b5af8c8c | 1880 | |
d02c6f54 | 1881 | r = udev_ctrl_start(manager->ctrl, on_ctrl_msg, manager); |
693d371d | 1882 | if (r < 0) |
d02c6f54 | 1883 | return log_error_errno(r, "Failed to start device monitor: %m"); |
693d371d TG |
1884 | |
1885 | /* This needs to be after the inotify and uevent handling, to make sure | |
1886 | * that the ping is send back after fully processing the pending uevents | |
1887 | * (including the synthetic ones we may create due to inotify events). | |
1888 | */ | |
d02c6f54 | 1889 | r = sd_event_source_set_priority(udev_ctrl_get_event_source(manager->ctrl), SD_EVENT_PRIORITY_IDLE); |
693d371d | 1890 | if (r < 0) |
044497e2 | 1891 | return log_error_errno(r, "Failed to set IDLE event priority for udev control event source: %m"); |
693d371d | 1892 | |
df7ee959 | 1893 | r = sd_event_add_io(manager->event, &manager->inotify_event, manager->inotify_fd, EPOLLIN, on_inotify, manager); |
693d371d | 1894 | if (r < 0) |
044497e2 | 1895 | return log_error_errno(r, "Failed to create inotify event source: %m"); |
693d371d | 1896 | |
f00d2b6d YW |
1897 | r = sd_device_monitor_attach_event(manager->monitor, manager->event); |
1898 | if (r < 0) | |
1899 | return log_error_errno(r, "Failed to attach event to device monitor: %m"); | |
1900 | ||
1901 | r = sd_device_monitor_start(manager->monitor, on_uevent, manager); | |
693d371d | 1902 | if (r < 0) |
f00d2b6d YW |
1903 | return log_error_errno(r, "Failed to start device monitor: %m"); |
1904 | ||
1905 | (void) sd_event_source_set_description(sd_device_monitor_get_event_source(manager->monitor), "device-monitor"); | |
693d371d TG |
1906 | |
1907 | r = sd_event_add_io(manager->event, NULL, fd_worker, EPOLLIN, on_worker, manager); | |
1908 | if (r < 0) | |
044497e2 | 1909 | return log_error_errno(r, "Failed to create worker event source: %m"); |
693d371d TG |
1910 | |
1911 | r = sd_event_add_post(manager->event, NULL, on_post, manager); | |
1912 | if (r < 0) | |
044497e2 | 1913 | return log_error_errno(r, "Failed to create post event source: %m"); |
e237d8cb | 1914 | |
b5af8c8c | 1915 | udev_builtin_init(); |
077fc5e2 | 1916 | |
c238a1f5 | 1917 | r = udev_rules_load(&manager->rules, arg_resolve_name_timing); |
b5af8c8c ZJS |
1918 | if (!manager->rules) |
1919 | return log_error_errno(r, "Failed to read udev rules: %m"); | |
077fc5e2 DH |
1920 | |
1921 | r = udev_rules_apply_static_dev_perms(manager->rules); | |
1922 | if (r < 0) | |
044497e2 | 1923 | log_error_errno(r, "Failed to apply permissions on static device nodes: %m"); |
077fc5e2 | 1924 | |
4bf4f50f | 1925 | notify_ready(); |
077fc5e2 DH |
1926 | |
1927 | r = sd_event_loop(manager->event); | |
44dcf454 | 1928 | if (r < 0) |
044497e2 | 1929 | log_error_errno(r, "Event loop failed: %m"); |
077fc5e2 | 1930 | |
077fc5e2 DH |
1931 | sd_notify(false, |
1932 | "STOPPING=1\n" | |
1933 | "STATUS=Shutting down..."); | |
077fc5e2 DH |
1934 | return r; |
1935 | } | |
1936 | ||
63e2d171 | 1937 | int run_udevd(int argc, char *argv[]) { |
b5af8c8c | 1938 | _cleanup_(manager_freep) Manager *manager = NULL; |
efa1606e | 1939 | int fd_ctrl = -1, fd_uevent = -1; |
e5d7bce1 | 1940 | int r; |
bba7a484 | 1941 | |
bba7a484 | 1942 | log_set_target(LOG_TARGET_AUTO); |
6b413782 | 1943 | log_open(); |
e2099267 | 1944 | udev_parse_config_full(&arg_children_max, &arg_exec_delay_usec, &arg_event_timeout_usec, &arg_resolve_name_timing, &arg_timeout_signal); |
bba7a484 | 1945 | log_parse_environment(); |
6b413782 | 1946 | log_open(); /* Done again to update after reading configuration. */ |
bba7a484 | 1947 | |
bba7a484 TG |
1948 | r = parse_argv(argc, argv); |
1949 | if (r <= 0) | |
0c5a109a | 1950 | return r; |
bba7a484 | 1951 | |
1d84ad94 | 1952 | r = proc_cmdline_parse(parse_proc_cmdline_item, NULL, PROC_CMDLINE_STRIP_RD_PREFIX); |
614a823c | 1953 | if (r < 0) |
044497e2 | 1954 | log_warning_errno(r, "Failed to parse kernel command line, ignoring: %m"); |
912541b0 | 1955 | |
78d3e041 KS |
1956 | if (arg_debug) { |
1957 | log_set_target(LOG_TARGET_CONSOLE); | |
bba7a484 | 1958 | log_set_max_level(LOG_DEBUG); |
78d3e041 | 1959 | } |
bba7a484 | 1960 | |
fba868fa LP |
1961 | r = must_be_root(); |
1962 | if (r < 0) | |
0c5a109a | 1963 | return r; |
912541b0 | 1964 | |
712cebf1 | 1965 | if (arg_children_max == 0) { |
fe56acd8 | 1966 | unsigned long cpu_limit, mem_limit, cpu_count = 1; |
d457ff83 | 1967 | |
fe56acd8 LP |
1968 | r = cpus_in_affinity_mask(); |
1969 | if (r < 0) | |
1970 | log_warning_errno(r, "Failed to determine number of local CPUs, ignoring: %m"); | |
1971 | else | |
1972 | cpu_count = r; | |
88bd5a32 FB |
1973 | |
1974 | cpu_limit = cpu_count * 2 + 16; | |
1975 | mem_limit = MAX(physical_memory() / (128UL*1024*1024), 10U); | |
912541b0 | 1976 | |
88bd5a32 FB |
1977 | arg_children_max = MIN(cpu_limit, mem_limit); |
1978 | arg_children_max = MIN(WORKER_NUM_MAX, arg_children_max); | |
e438c57a | 1979 | |
044497e2 | 1980 | log_debug("Set children_max to %u", arg_children_max); |
d457ff83 | 1981 | } |
912541b0 | 1982 | |
712cebf1 | 1983 | /* set umask before creating any file/directory */ |
712cebf1 | 1984 | umask(022); |
912541b0 | 1985 | |
c3dacc8b | 1986 | r = mac_selinux_init(); |
0c5a109a | 1987 | if (r < 0) |
a9ba0e32 | 1988 | return r; |
912541b0 | 1989 | |
3f692e2e | 1990 | r = RET_NERRNO(mkdir("/run/udev", 0755)); |
0c5a109a ZJS |
1991 | if (r < 0 && r != -EEXIST) |
1992 | return log_error_errno(r, "Failed to create /run/udev: %m"); | |
712cebf1 | 1993 | |
b7f74dd4 | 1994 | r = listen_fds(&fd_ctrl, &fd_uevent); |
0c5a109a ZJS |
1995 | if (r < 0) |
1996 | return log_error_errno(r, "Failed to listen on fds: %m"); | |
b7f74dd4 | 1997 | |
a1f4fd38 | 1998 | r = manager_new(&manager, fd_ctrl, fd_uevent); |
b5af8c8c ZJS |
1999 | if (r < 0) |
2000 | return log_error_errno(r, "Failed to create manager: %m"); | |
2001 | ||
bba7a484 | 2002 | if (arg_daemonize) { |
912541b0 | 2003 | pid_t pid; |
912541b0 | 2004 | |
b5af8c8c | 2005 | log_info("Starting version " GIT_VERSION); |
3cbb2057 | 2006 | |
40e749b5 | 2007 | /* connect /dev/null to stdin, stdout, stderr */ |
c76cf844 AK |
2008 | if (log_get_max_level() < LOG_DEBUG) { |
2009 | r = make_null_stdio(); | |
2010 | if (r < 0) | |
2011 | log_warning_errno(r, "Failed to redirect standard streams to /dev/null: %m"); | |
2012 | } | |
2013 | ||
912541b0 | 2014 | pid = fork(); |
0c5a109a ZJS |
2015 | if (pid < 0) |
2016 | return log_error_errno(errno, "Failed to fork daemon: %m"); | |
2017 | if (pid > 0) | |
2018 | /* parent */ | |
2019 | return 0; | |
912541b0 | 2020 | |
0c5a109a | 2021 | /* child */ |
ece0fe12 | 2022 | (void) setsid(); |
7500cd5e | 2023 | } |
912541b0 | 2024 | |
76e62a4d | 2025 | return main_loop(manager); |
7fafc032 | 2026 | } |