]>
Commit | Line | Data |
---|---|---|
db9ecf05 | 1 | /* SPDX-License-Identifier: LGPL-2.1-or-later */ |
a7334b09 | 2 | |
60918275 | 3 | #include <errno.h> |
400f1a33 LP |
4 | #include <fcntl.h> |
5 | #include <linux/kd.h> | |
e46b13c8 | 6 | #include <sys/epoll.h> |
400f1a33 | 7 | #include <sys/inotify.h> |
e1414003 | 8 | #include <sys/ioctl.h> |
ca6ce62d | 9 | #include <sys/mount.h> |
400f1a33 | 10 | #include <sys/reboot.h> |
8742514c | 11 | #include <sys/timerfd.h> |
40efaaed | 12 | #include <sys/utsname.h> |
400f1a33 LP |
13 | #include <sys/wait.h> |
14 | #include <unistd.h> | |
830f6caa | 15 | |
349cc4a5 | 16 | #if HAVE_AUDIT |
4927fcae | 17 | #include <libaudit.h> |
830f6caa | 18 | #endif |
60918275 | 19 | |
718db961 | 20 | #include "sd-daemon.h" |
718db961 | 21 | #include "sd-messages.h" |
3536f49e | 22 | #include "sd-path.h" |
81527be1 | 23 | |
57b7a260 | 24 | #include "all-units.h" |
d68c645b | 25 | #include "alloc-util.h" |
400f1a33 LP |
26 | #include "audit-fd.h" |
27 | #include "boot-timestamps.h" | |
28 | #include "bus-common-errors.h" | |
29 | #include "bus-error.h" | |
30 | #include "bus-kernel.h" | |
31 | #include "bus-util.h" | |
00d9ef85 | 32 | #include "clean-ipc.h" |
af6b0ecc | 33 | #include "clock-util.h" |
29e6b0c1 | 34 | #include "common-signal.h" |
08951245 | 35 | #include "confidential-virt.h" |
28db6fbf | 36 | #include "constants.h" |
19d22d43 | 37 | #include "core-varlink.h" |
786d19fd | 38 | #include "creds-util.h" |
400f1a33 LP |
39 | #include "dbus-job.h" |
40 | #include "dbus-manager.h" | |
41 | #include "dbus-unit.h" | |
42 | #include "dbus.h" | |
d063a527 | 43 | #include "dirent-util.h" |
400f1a33 | 44 | #include "env-util.h" |
4f5dd394 | 45 | #include "escape.h" |
ec75e8e0 | 46 | #include "event-util.h" |
89711996 | 47 | #include "exec-util.h" |
d3070fbd | 48 | #include "execute.h" |
400f1a33 | 49 | #include "exit-status.h" |
3ffd4af2 | 50 | #include "fd-util.h" |
0d39fa9c | 51 | #include "fileio.h" |
385093b7 | 52 | #include "generator-setup.h" |
60918275 | 53 | #include "hashmap.h" |
baa6a42d | 54 | #include "initrd-util.h" |
9e5fd717 | 55 | #include "inotify-util.h" |
5cfa33e0 | 56 | #include "install.h" |
19d22d43 | 57 | #include "io-util.h" |
0690160e | 58 | #include "label-util.h" |
d904afc7 | 59 | #include "load-fragment.h" |
786d19fd | 60 | #include "locale-setup.h" |
16354eff | 61 | #include "log.h" |
400f1a33 | 62 | #include "macro.h" |
3ffd4af2 | 63 | #include "manager.h" |
2a341bb9 | 64 | #include "manager-dump.h" |
a01ba4b2 | 65 | #include "manager-serialize.h" |
0a970718 | 66 | #include "memory-util.h" |
35cd0ba5 | 67 | #include "mkdir-label.h" |
ca6ce62d | 68 | #include "mount-util.h" |
4bd03515 | 69 | #include "os-util.h" |
6bedfcbb | 70 | #include "parse-util.h" |
400f1a33 LP |
71 | #include "path-lookup.h" |
72 | #include "path-util.h" | |
73 | #include "process-util.h" | |
6bb00842 | 74 | #include "psi-util.h" |
ea430986 | 75 | #include "ratelimit.h" |
31ce987c | 76 | #include "rlimit-util.h" |
c6878637 | 77 | #include "rm-rf.h" |
45ae2f72 | 78 | #include "selinux-util.h" |
400f1a33 | 79 | #include "signal-util.h" |
57b7a260 | 80 | #include "socket-util.h" |
514f4ef5 | 81 | #include "special.h" |
8fcde012 | 82 | #include "stat-util.h" |
8b43440b | 83 | #include "string-table.h" |
07630cea | 84 | #include "string-util.h" |
400f1a33 | 85 | #include "strv.h" |
dd1db3c2 | 86 | #include "strxcpyx.h" |
6123dfaa | 87 | #include "sysctl-util.h" |
a6ecbf83 | 88 | #include "syslog-util.h" |
400f1a33 LP |
89 | #include "terminal-util.h" |
90 | #include "time-util.h" | |
91 | #include "transaction.h" | |
63e8df04 | 92 | #include "uid-range.h" |
affb60b1 | 93 | #include "umask-util.h" |
400f1a33 | 94 | #include "unit-name.h" |
00d9ef85 | 95 | #include "user-util.h" |
5dc4c17f | 96 | #include "virt.h" |
e96d6be7 | 97 | #include "watchdog.h" |
60918275 | 98 | |
a47806fa | 99 | #define NOTIFY_RCVBUF_SIZE (8*1024*1024) |
d8fdc620 | 100 | #define CGROUPS_AGENT_RCVBUF_SIZE (8*1024*1024) |
a47806fa | 101 | |
03b717a3 | 102 | /* Initial delay and the interval for printing status messages about running jobs */ |
1b4154a8 ZJS |
103 | #define JOBS_IN_PROGRESS_WAIT_USEC (2*USEC_PER_SEC) |
104 | #define JOBS_IN_PROGRESS_QUIET_WAIT_USEC (25*USEC_PER_SEC) | |
fd08a840 | 105 | #define JOBS_IN_PROGRESS_PERIOD_USEC (USEC_PER_SEC / 3) |
03b717a3 MS |
106 | #define JOBS_IN_PROGRESS_PERIOD_DIVISOR 3 |
107 | ||
5238e957 | 108 | /* If there are more than 1K bus messages queue across our API and direct buses, then let's not add more on top until |
e0a08581 LP |
109 | * the queue gets more empty. */ |
110 | #define MANAGER_BUS_BUSY_THRESHOLD 1024LU | |
111 | ||
112 | /* How many units and jobs to process of the bus queue before returning to the event loop. */ | |
113 | #define MANAGER_BUS_MESSAGE_BUDGET 100U | |
114 | ||
ea09a416 LP |
115 | #define DEFAULT_TASKS_MAX ((TasksMax) { 15U, 100U }) /* 15% */ |
116 | ||
718db961 | 117 | static int manager_dispatch_notify_fd(sd_event_source *source, int fd, uint32_t revents, void *userdata); |
d8fdc620 | 118 | static int manager_dispatch_cgroups_agent_fd(sd_event_source *source, int fd, uint32_t revents, void *userdata); |
718db961 LP |
119 | static int manager_dispatch_signal_fd(sd_event_source *source, int fd, uint32_t revents, void *userdata); |
120 | static int manager_dispatch_time_change_fd(sd_event_source *source, int fd, uint32_t revents, void *userdata); | |
121 | static int manager_dispatch_idle_pipe_fd(sd_event_source *source, int fd, uint32_t revents, void *userdata); | |
00d9ef85 | 122 | static int manager_dispatch_user_lookup_fd(sd_event_source *source, int fd, uint32_t revents, void *userdata); |
718db961 | 123 | static int manager_dispatch_jobs_in_progress(sd_event_source *source, usec_t usec, void *userdata); |
752b5905 | 124 | static int manager_dispatch_run_queue(sd_event_source *source, void *userdata); |
575b300b | 125 | static int manager_dispatch_sigchld(sd_event_source *source, void *userdata); |
bbf5fd8e | 126 | static int manager_dispatch_timezone_change(sd_event_source *source, const struct inotify_event *event, void *userdata); |
64691d20 | 127 | static int manager_run_environment_generators(Manager *m); |
e801700e | 128 | static int manager_run_generators(Manager *m); |
06a4eb07 | 129 | static void manager_vacuum(Manager *m); |
718db961 | 130 | |
1b4154a8 | 131 | static usec_t manager_watch_jobs_next_time(Manager *m) { |
3889fc6f ZJS |
132 | usec_t timeout; |
133 | ||
134 | if (MANAGER_IS_USER(m)) | |
135 | /* Let the user manager without a timeout show status quickly, so the system manager can make | |
136 | * use of it, if it wants to. */ | |
137 | timeout = JOBS_IN_PROGRESS_WAIT_USEC * 2 / 3; | |
138 | else if (show_status_on(m->show_status)) | |
139 | /* When status is on, just use the usual timeout. */ | |
140 | timeout = JOBS_IN_PROGRESS_WAIT_USEC; | |
141 | else | |
142 | timeout = JOBS_IN_PROGRESS_QUIET_WAIT_USEC; | |
143 | ||
144 | return usec_add(now(CLOCK_MONOTONIC), timeout); | |
1b4154a8 ZJS |
145 | } |
146 | ||
2ae56591 | 147 | static void manager_watch_jobs_in_progress(Manager *m) { |
e5723c89 | 148 | usec_t next; |
cfa9677b | 149 | int r; |
e5723c89 | 150 | |
718db961 | 151 | assert(m); |
03b717a3 | 152 | |
42bf1ae1 FB |
153 | /* We do not want to show the cylon animation if the user |
154 | * needs to confirm service executions otherwise confirmation | |
155 | * messages will be screwed by the cylon animation. */ | |
b0eb2944 | 156 | if (!manager_is_confirm_spawn_disabled(m)) |
42bf1ae1 FB |
157 | return; |
158 | ||
718db961 | 159 | if (m->jobs_in_progress_event_source) |
2ae56591 | 160 | return; |
03b717a3 | 161 | |
1b4154a8 | 162 | next = manager_watch_jobs_next_time(m); |
cfa9677b | 163 | r = sd_event_add_time( |
6a0f1f6d LP |
164 | m->event, |
165 | &m->jobs_in_progress_event_source, | |
166 | CLOCK_MONOTONIC, | |
167 | next, 0, | |
168 | manager_dispatch_jobs_in_progress, m); | |
cfa9677b MM |
169 | if (r < 0) |
170 | return; | |
7dfbe2e3 TG |
171 | |
172 | (void) sd_event_source_set_description(m->jobs_in_progress_event_source, "manager-jobs-in-progress"); | |
03b717a3 MS |
173 | } |
174 | ||
fbd0b64f | 175 | #define CYLON_BUFFER_EXTRA (2*STRLEN(ANSI_RED) + STRLEN(ANSI_HIGHLIGHT_RED) + 2*STRLEN(ANSI_NORMAL)) |
03b717a3 | 176 | |
03b717a3 MS |
177 | static void draw_cylon(char buffer[], size_t buflen, unsigned width, unsigned pos) { |
178 | char *p = buffer; | |
179 | ||
180 | assert(buflen >= CYLON_BUFFER_EXTRA + width + 1); | |
181 | assert(pos <= width+1); /* 0 or width+1 mean that the center light is behind the corner */ | |
182 | ||
183 | if (pos > 1) { | |
6282c859 MS |
184 | if (pos > 2) |
185 | p = mempset(p, ' ', pos-2); | |
64c3610b FB |
186 | if (log_get_show_color()) |
187 | p = stpcpy(p, ANSI_RED); | |
03b717a3 MS |
188 | *p++ = '*'; |
189 | } | |
190 | ||
191 | if (pos > 0 && pos <= width) { | |
64c3610b FB |
192 | if (log_get_show_color()) |
193 | p = stpcpy(p, ANSI_HIGHLIGHT_RED); | |
03b717a3 MS |
194 | *p++ = '*'; |
195 | } | |
196 | ||
64c3610b FB |
197 | if (log_get_show_color()) |
198 | p = stpcpy(p, ANSI_NORMAL); | |
03b717a3 MS |
199 | |
200 | if (pos < width) { | |
64c3610b FB |
201 | if (log_get_show_color()) |
202 | p = stpcpy(p, ANSI_RED); | |
03b717a3 | 203 | *p++ = '*'; |
6282c859 MS |
204 | if (pos < width-1) |
205 | p = mempset(p, ' ', width-1-pos); | |
64c3610b FB |
206 | if (log_get_show_color()) |
207 | strcpy(p, ANSI_NORMAL); | |
03b717a3 | 208 | } |
03b717a3 MS |
209 | } |
210 | ||
1addc46c | 211 | static void manager_flip_auto_status(Manager *m, bool enable, const char *reason) { |
f755e3b7 LP |
212 | assert(m); |
213 | ||
cb8ccb22 ZJS |
214 | if (enable) { |
215 | if (m->show_status == SHOW_STATUS_AUTO) | |
7365a296 | 216 | manager_set_show_status(m, SHOW_STATUS_TEMPORARY, reason); |
cb8ccb22 ZJS |
217 | } else { |
218 | if (m->show_status == SHOW_STATUS_TEMPORARY) | |
7365a296 | 219 | manager_set_show_status(m, SHOW_STATUS_AUTO, reason); |
cb8ccb22 ZJS |
220 | } |
221 | } | |
222 | ||
03b717a3 | 223 | static void manager_print_jobs_in_progress(Manager *m) { |
03b717a3 | 224 | Job *j; |
03b717a3 MS |
225 | unsigned counter = 0, print_nr; |
226 | char cylon[6 + CYLON_BUFFER_EXTRA + 1]; | |
227 | unsigned cylon_pos; | |
4c989f89 | 228 | uint64_t timeout = 0; |
03b717a3 | 229 | |
718db961 | 230 | assert(m); |
9c3349e2 | 231 | assert(m->n_running_jobs > 0); |
718db961 | 232 | |
7365a296 | 233 | manager_flip_auto_status(m, true, "delay"); |
d450b6f2 | 234 | |
03b717a3 MS |
235 | print_nr = (m->jobs_in_progress_iteration / JOBS_IN_PROGRESS_PERIOD_DIVISOR) % m->n_running_jobs; |
236 | ||
90e74a66 | 237 | HASHMAP_FOREACH(j, m->jobs) |
03b717a3 MS |
238 | if (j->state == JOB_RUNNING && counter++ == print_nr) |
239 | break; | |
240 | ||
e970a72e MS |
241 | /* m->n_running_jobs must be consistent with the contents of m->jobs, |
242 | * so the above loop must have succeeded in finding j. */ | |
243 | assert(counter == print_nr + 1); | |
51d122af | 244 | assert(j); |
5a82a91a | 245 | |
03b717a3 MS |
246 | cylon_pos = m->jobs_in_progress_iteration % 14; |
247 | if (cylon_pos >= 8) | |
248 | cylon_pos = 14 - cylon_pos; | |
249 | draw_cylon(cylon, sizeof(cylon), 6, cylon_pos); | |
250 | ||
8bb310c3 ZJS |
251 | m->jobs_in_progress_iteration++; |
252 | ||
3889fc6f | 253 | char job_of_n[STRLEN("( of ) ") + DECIMAL_STR_MAX(unsigned)*2] = ""; |
5291f26d | 254 | if (m->n_running_jobs > 1) |
3889fc6f | 255 | xsprintf(job_of_n, "(%u of %u) ", counter, m->n_running_jobs); |
03b717a3 | 256 | |
4c989f89 | 257 | (void) job_get_timeout(j, &timeout); |
8bb310c3 | 258 | |
04d232d8 ZJS |
259 | /* We want to use enough information for the user to identify previous lines talking about the same |
260 | * unit, but keep the message as short as possible. So if 'Starting foo.service' or 'Starting | |
3889fc6f | 261 | * foo.service - Description' were used, 'foo.service' is enough here. On the other hand, if we used |
04d232d8 ZJS |
262 | * 'Starting Description' before, then we shall also use 'Description' here. So we pass NULL as the |
263 | * second argument to unit_status_string(). */ | |
264 | const char *ident = unit_status_string(j->unit, NULL); | |
265 | ||
3889fc6f | 266 | const char *time = FORMAT_TIMESPAN(now(CLOCK_MONOTONIC) - j->begin_usec, 1*USEC_PER_SEC); |
4c989f89 | 267 | const char *limit = timeout > 0 ? FORMAT_TIMESPAN(timeout - j->begin_usec, 1*USEC_PER_SEC) : "no limit"; |
3889fc6f ZJS |
268 | |
269 | if (m->status_unit_format == STATUS_UNIT_FORMAT_DESCRIPTION) | |
270 | /* When using 'Description', we effectively don't have enough space to show the nested status | |
271 | * without ellipsization, so let's not even try. */ | |
272 | manager_status_printf(m, STATUS_TYPE_EPHEMERAL, cylon, | |
273 | "%sA %s job is running for %s (%s / %s)", | |
274 | job_of_n, | |
275 | job_type_to_string(j->type), | |
276 | ident, | |
277 | time, limit); | |
278 | else { | |
279 | const char *status_text = unit_status_text(j->unit); | |
280 | ||
281 | manager_status_printf(m, STATUS_TYPE_EPHEMERAL, cylon, | |
282 | "%sJob %s/%s running (%s / %s)%s%s", | |
283 | job_of_n, | |
284 | ident, | |
285 | job_type_to_string(j->type), | |
286 | time, limit, | |
287 | status_text ? ": " : "", | |
288 | strempty(status_text)); | |
289 | } | |
290 | ||
291 | sd_notifyf(false, | |
292 | "STATUS=%sUser job %s/%s running (%s / %s)...", | |
293 | job_of_n, | |
294 | ident, | |
295 | job_type_to_string(j->type), | |
296 | time, limit); | |
297 | m->status_ready = false; | |
03b717a3 MS |
298 | } |
299 | ||
e46b13c8 | 300 | static int have_ask_password(void) { |
c2b2df60 | 301 | _cleanup_closedir_ DIR *dir = NULL; |
e46b13c8 ZJS |
302 | |
303 | dir = opendir("/run/systemd/ask-password"); | |
304 | if (!dir) { | |
305 | if (errno == ENOENT) | |
306 | return false; | |
307 | else | |
308 | return -errno; | |
309 | } | |
310 | ||
c7f0d9e5 | 311 | FOREACH_DIRENT_ALL(de, dir, return -errno) |
e46b13c8 ZJS |
312 | if (startswith(de->d_name, "ask.")) |
313 | return true; | |
8fb3f009 | 314 | return false; |
e46b13c8 ZJS |
315 | } |
316 | ||
317 | static int manager_dispatch_ask_password_fd(sd_event_source *source, | |
318 | int fd, uint32_t revents, void *userdata) { | |
99534007 | 319 | Manager *m = ASSERT_PTR(userdata); |
e46b13c8 | 320 | |
665dfe93 | 321 | (void) flush_fd(fd); |
e46b13c8 ZJS |
322 | |
323 | m->have_ask_password = have_ask_password(); | |
324 | if (m->have_ask_password < 0) | |
325 | /* Log error but continue. Negative have_ask_password | |
326 | * is treated as unknown status. */ | |
c33b3297 | 327 | log_error_errno(m->have_ask_password, "Failed to list /run/systemd/ask-password: %m"); |
e46b13c8 ZJS |
328 | |
329 | return 0; | |
330 | } | |
331 | ||
332 | static void manager_close_ask_password(Manager *m) { | |
333 | assert(m); | |
334 | ||
5dcadb4c | 335 | m->ask_password_event_source = sd_event_source_disable_unref(m->ask_password_event_source); |
90990e28 | 336 | m->ask_password_inotify_fd = safe_close(m->ask_password_inotify_fd); |
e46b13c8 ZJS |
337 | m->have_ask_password = -EINVAL; |
338 | } | |
339 | ||
340 | static int manager_check_ask_password(Manager *m) { | |
341 | int r; | |
342 | ||
343 | assert(m); | |
344 | ||
345 | if (!m->ask_password_event_source) { | |
346 | assert(m->ask_password_inotify_fd < 0); | |
347 | ||
17e4b070 | 348 | (void) mkdir_p_label("/run/systemd/ask-password", 0755); |
e46b13c8 ZJS |
349 | |
350 | m->ask_password_inotify_fd = inotify_init1(IN_NONBLOCK|IN_CLOEXEC); | |
4a62c710 | 351 | if (m->ask_password_inotify_fd < 0) |
330b8fb3 | 352 | return log_error_errno(errno, "Failed to create inotify object: %m"); |
e46b13c8 | 353 | |
27c3112d FB |
354 | r = inotify_add_watch_and_warn(m->ask_password_inotify_fd, |
355 | "/run/systemd/ask-password", | |
356 | IN_CREATE|IN_DELETE|IN_MOVE); | |
357 | if (r < 0) { | |
e46b13c8 | 358 | manager_close_ask_password(m); |
27c3112d | 359 | return r; |
e46b13c8 ZJS |
360 | } |
361 | ||
362 | r = sd_event_add_io(m->event, &m->ask_password_event_source, | |
363 | m->ask_password_inotify_fd, EPOLLIN, | |
364 | manager_dispatch_ask_password_fd, m); | |
365 | if (r < 0) { | |
df3d3bdf | 366 | log_error_errno(r, "Failed to add event source for /run/systemd/ask-password: %m"); |
e46b13c8 | 367 | manager_close_ask_password(m); |
df3d3bdf | 368 | return r; |
e46b13c8 ZJS |
369 | } |
370 | ||
7dfbe2e3 TG |
371 | (void) sd_event_source_set_description(m->ask_password_event_source, "manager-ask-password"); |
372 | ||
e46b13c8 ZJS |
373 | /* Queries might have been added meanwhile... */ |
374 | manager_dispatch_ask_password_fd(m->ask_password_event_source, | |
375 | m->ask_password_inotify_fd, EPOLLIN, m); | |
376 | } | |
377 | ||
378 | return m->have_ask_password; | |
379 | } | |
380 | ||
31a7eb86 | 381 | static int manager_watch_idle_pipe(Manager *m) { |
31a7eb86 ZJS |
382 | int r; |
383 | ||
718db961 LP |
384 | assert(m); |
385 | ||
386 | if (m->idle_pipe_event_source) | |
31a7eb86 ZJS |
387 | return 0; |
388 | ||
389 | if (m->idle_pipe[2] < 0) | |
390 | return 0; | |
391 | ||
151b9b96 | 392 | r = sd_event_add_io(m->event, &m->idle_pipe_event_source, m->idle_pipe[2], EPOLLIN, manager_dispatch_idle_pipe_fd, m); |
23bbb0de MS |
393 | if (r < 0) |
394 | return log_error_errno(r, "Failed to watch idle pipe: %m"); | |
31a7eb86 | 395 | |
7dfbe2e3 TG |
396 | (void) sd_event_source_set_description(m->idle_pipe_event_source, "manager-idle-pipe"); |
397 | ||
31a7eb86 | 398 | return 0; |
31a7eb86 ZJS |
399 | } |
400 | ||
718db961 LP |
401 | static void manager_close_idle_pipe(Manager *m) { |
402 | assert(m); | |
31a7eb86 | 403 | |
5dcadb4c | 404 | m->idle_pipe_event_source = sd_event_source_disable_unref(m->idle_pipe_event_source); |
cd72bd8a | 405 | |
3d94f76c LP |
406 | safe_close_pair(m->idle_pipe); |
407 | safe_close_pair(m->idle_pipe + 2); | |
31a7eb86 ZJS |
408 | } |
409 | ||
8742514c | 410 | static int manager_setup_time_change(Manager *m) { |
718db961 | 411 | int r; |
b92bea5d | 412 | |
718db961 | 413 | assert(m); |
8742514c | 414 | |
638cece4 | 415 | if (MANAGER_IS_TEST_RUN(m)) |
0d8c31ff ZJS |
416 | return 0; |
417 | ||
5dcadb4c | 418 | m->time_change_event_source = sd_event_source_disable_unref(m->time_change_event_source); |
7feedd18 | 419 | |
ec75e8e0 | 420 | r = event_add_time_change(m->event, &m->time_change_event_source, manager_dispatch_time_change_fd, m); |
23bbb0de MS |
421 | if (r < 0) |
422 | return log_error_errno(r, "Failed to create time change event source: %m"); | |
8742514c | 423 | |
a5cc7e5a LP |
424 | /* Schedule this slightly earlier than the .timer event sources */ |
425 | r = sd_event_source_set_priority(m->time_change_event_source, SD_EVENT_PRIORITY_NORMAL-1); | |
426 | if (r < 0) | |
427 | return log_error_errno(r, "Failed to set priority of time change event sources: %m"); | |
428 | ||
8742514c LP |
429 | log_debug("Set up TFD_TIMER_CANCEL_ON_SET timerfd."); |
430 | ||
431 | return 0; | |
432 | } | |
433 | ||
bbf5fd8e LP |
434 | static int manager_read_timezone_stat(Manager *m) { |
435 | struct stat st; | |
436 | bool changed; | |
437 | ||
438 | assert(m); | |
439 | ||
440 | /* Read the current stat() data of /etc/localtime so that we detect changes */ | |
441 | if (lstat("/etc/localtime", &st) < 0) { | |
442 | log_debug_errno(errno, "Failed to stat /etc/localtime, ignoring: %m"); | |
443 | changed = m->etc_localtime_accessible; | |
444 | m->etc_localtime_accessible = false; | |
445 | } else { | |
446 | usec_t k; | |
447 | ||
448 | k = timespec_load(&st.st_mtim); | |
449 | changed = !m->etc_localtime_accessible || k != m->etc_localtime_mtime; | |
450 | ||
451 | m->etc_localtime_mtime = k; | |
452 | m->etc_localtime_accessible = true; | |
453 | } | |
454 | ||
455 | return changed; | |
456 | } | |
457 | ||
458 | static int manager_setup_timezone_change(Manager *m) { | |
a5cc7e5a | 459 | _cleanup_(sd_event_source_unrefp) sd_event_source *new_event = NULL; |
bbf5fd8e LP |
460 | int r; |
461 | ||
462 | assert(m); | |
463 | ||
638cece4 | 464 | if (MANAGER_IS_TEST_RUN(m)) |
bbf5fd8e LP |
465 | return 0; |
466 | ||
467 | /* We watch /etc/localtime for three events: change of the link count (which might mean removal from /etc even | |
468 | * though another link might be kept), renames, and file close operations after writing. Note we don't bother | |
469 | * with IN_DELETE_SELF, as that would just report when the inode is removed entirely, i.e. after the link count | |
470 | * went to zero and all fds to it are closed. | |
471 | * | |
472 | * Note that we never follow symlinks here. This is a simplification, but should cover almost all cases | |
473 | * correctly. | |
474 | * | |
475 | * Note that we create the new event source first here, before releasing the old one. This should optimize | |
476 | * behaviour as this way sd-event can reuse the old watch in case the inode didn't change. */ | |
477 | ||
478 | r = sd_event_add_inotify(m->event, &new_event, "/etc/localtime", | |
479 | IN_ATTRIB|IN_MOVE_SELF|IN_CLOSE_WRITE|IN_DONT_FOLLOW, manager_dispatch_timezone_change, m); | |
0cb21d8c LP |
480 | if (r == -ENOENT) { |
481 | /* If the file doesn't exist yet, subscribe to /etc instead, and wait until it is created either by | |
482 | * O_CREATE or by rename() */ | |
483 | ||
484 | log_debug_errno(r, "/etc/localtime doesn't exist yet, watching /etc instead."); | |
bbf5fd8e LP |
485 | r = sd_event_add_inotify(m->event, &new_event, "/etc", |
486 | IN_CREATE|IN_MOVED_TO|IN_ONLYDIR, manager_dispatch_timezone_change, m); | |
0cb21d8c | 487 | } |
bbf5fd8e LP |
488 | if (r < 0) |
489 | return log_error_errno(r, "Failed to create timezone change event source: %m"); | |
490 | ||
a5cc7e5a LP |
491 | /* Schedule this slightly earlier than the .timer event sources */ |
492 | r = sd_event_source_set_priority(new_event, SD_EVENT_PRIORITY_NORMAL-1); | |
493 | if (r < 0) | |
494 | return log_error_errno(r, "Failed to set priority of timezone change event sources: %m"); | |
495 | ||
bbf5fd8e | 496 | sd_event_source_unref(m->timezone_change_event_source); |
a5cc7e5a | 497 | m->timezone_change_event_source = TAKE_PTR(new_event); |
bbf5fd8e LP |
498 | |
499 | return 0; | |
500 | } | |
501 | ||
80876c20 | 502 | static int enable_special_signals(Manager *m) { |
254d1313 | 503 | _cleanup_close_ int fd = -EBADF; |
80876c20 LP |
504 | |
505 | assert(m); | |
506 | ||
638cece4 | 507 | if (MANAGER_IS_TEST_RUN(m)) |
37453b3a EV |
508 | return 0; |
509 | ||
a41b539e | 510 | /* Enable that we get SIGINT on control-alt-del. In containers |
c9999773 LP |
511 | * this will fail with EPERM (older) or EINVAL (newer), so |
512 | * ignore that. */ | |
4c701096 | 513 | if (reboot(RB_DISABLE_CAD) < 0 && !IN_SET(errno, EPERM, EINVAL)) |
56f64d95 | 514 | log_warning_errno(errno, "Failed to enable ctrl-alt-del handling: %m"); |
80876c20 | 515 | |
a41b539e LP |
516 | fd = open_terminal("/dev/tty0", O_RDWR|O_NOCTTY|O_CLOEXEC); |
517 | if (fd < 0) { | |
518 | /* Support systems without virtual console */ | |
519 | if (fd != -ENOENT) | |
56f64d95 | 520 | log_warning_errno(errno, "Failed to open /dev/tty0: %m"); |
a41b539e | 521 | } else { |
80876c20 LP |
522 | /* Enable that we get SIGWINCH on kbrequest */ |
523 | if (ioctl(fd, KDSIGACCEPT, SIGWINCH) < 0) | |
56f64d95 | 524 | log_warning_errno(errno, "Failed to enable kbrequest handling: %m"); |
80876c20 LP |
525 | } |
526 | ||
527 | return 0; | |
528 | } | |
529 | ||
8750ac02 ZJS |
530 | #define RTSIG_IF_AVAILABLE(signum) (signum <= SIGRTMAX ? signum : -1) |
531 | ||
ce578209 | 532 | static int manager_setup_signals(Manager *m) { |
b92bea5d ZJS |
533 | struct sigaction sa = { |
534 | .sa_handler = SIG_DFL, | |
535 | .sa_flags = SA_NOCLDSTOP|SA_RESTART, | |
536 | }; | |
718db961 LP |
537 | sigset_t mask; |
538 | int r; | |
60918275 | 539 | |
ce578209 LP |
540 | assert(m); |
541 | ||
57c0c30e LP |
542 | assert_se(sigaction(SIGCHLD, &sa, NULL) == 0); |
543 | ||
4dffec14 LP |
544 | /* We make liberal use of realtime signals here. On |
545 | * Linux/glibc we have 30 of them (with the exception of Linux | |
546 | * on hppa, see below), between SIGRTMIN+0 ... SIGRTMIN+30 | |
547 | * (aka SIGRTMAX). */ | |
7d793605 | 548 | |
4dffec14 | 549 | assert_se(sigemptyset(&mask) == 0); |
7d793605 LP |
550 | sigset_add_many(&mask, |
551 | SIGCHLD, /* Child died */ | |
552 | SIGTERM, /* Reexecute daemon */ | |
553 | SIGHUP, /* Reload configuration */ | |
54d04cd1 | 554 | SIGUSR1, /* systemd: reconnect to D-Bus */ |
7d793605 LP |
555 | SIGUSR2, /* systemd: dump status */ |
556 | SIGINT, /* Kernel sends us this on control-alt-del */ | |
557 | SIGWINCH, /* Kernel sends us this on kbrequest (alt-arrowup) */ | |
558 | SIGPWR, /* Some kernel drivers and upsd send us this on power failure */ | |
4dffec14 | 559 | |
7d793605 | 560 | SIGRTMIN+0, /* systemd: start default.target */ |
0003d1ab | 561 | SIGRTMIN+1, /* systemd: isolate rescue.target */ |
7d793605 LP |
562 | SIGRTMIN+2, /* systemd: isolate emergency.target */ |
563 | SIGRTMIN+3, /* systemd: start halt.target */ | |
564 | SIGRTMIN+4, /* systemd: start poweroff.target */ | |
565 | SIGRTMIN+5, /* systemd: start reboot.target */ | |
0003d1ab | 566 | SIGRTMIN+6, /* systemd: start kexec.target */ |
13ffc607 | 567 | SIGRTMIN+7, /* systemd: start soft-reboot.target */ |
4dffec14 LP |
568 | |
569 | /* ... space for more special targets ... */ | |
570 | ||
0003d1ab LP |
571 | SIGRTMIN+13, /* systemd: Immediate halt */ |
572 | SIGRTMIN+14, /* systemd: Immediate poweroff */ | |
573 | SIGRTMIN+15, /* systemd: Immediate reboot */ | |
574 | SIGRTMIN+16, /* systemd: Immediate kexec */ | |
13ffc607 | 575 | SIGRTMIN+17, /* systemd: Immediate soft-reboot */ |
29e6b0c1 LP |
576 | SIGRTMIN+18, /* systemd: control command */ |
577 | ||
578 | /* ... space ... */ | |
4dffec14 | 579 | |
0658666b LP |
580 | SIGRTMIN+20, /* systemd: enable status messages */ |
581 | SIGRTMIN+21, /* systemd: disable status messages */ | |
253ee27a LP |
582 | SIGRTMIN+22, /* systemd: set log level to LOG_DEBUG */ |
583 | SIGRTMIN+23, /* systemd: set log level to LOG_INFO */ | |
600b704e | 584 | SIGRTMIN+24, /* systemd: Immediate exit (--user only) */ |
463aef23 | 585 | SIGRTMIN+25, /* systemd: reexecute manager */ |
4dffec14 | 586 | |
8750ac02 ZJS |
587 | /* Apparently Linux on hppa had fewer RT signals until v3.18, |
588 | * SIGRTMAX was SIGRTMIN+25, and then SIGRTMIN was lowered, | |
589 | * see commit v3.17-7614-g1f25df2eff. | |
590 | * | |
591 | * We cannot unconditionally make use of those signals here, | |
592 | * so let's use a runtime check. Since these commands are | |
593 | * accessible by different means and only really a safety | |
594 | * net, the missing functionality on hppa shouldn't matter. | |
595 | */ | |
596 | ||
597 | RTSIG_IF_AVAILABLE(SIGRTMIN+26), /* systemd: set log target to journal-or-kmsg */ | |
598 | RTSIG_IF_AVAILABLE(SIGRTMIN+27), /* systemd: set log target to console */ | |
599 | RTSIG_IF_AVAILABLE(SIGRTMIN+28), /* systemd: set log target to kmsg */ | |
600 | RTSIG_IF_AVAILABLE(SIGRTMIN+29), /* systemd: set log target to syslog-or-kmsg (obsolete) */ | |
4dffec14 LP |
601 | |
602 | /* ... one free signal here SIGRTMIN+30 ... */ | |
7d793605 | 603 | -1); |
ce578209 LP |
604 | assert_se(sigprocmask(SIG_SETMASK, &mask, NULL) == 0); |
605 | ||
718db961 LP |
606 | m->signal_fd = signalfd(-1, &mask, SFD_NONBLOCK|SFD_CLOEXEC); |
607 | if (m->signal_fd < 0) | |
ce578209 LP |
608 | return -errno; |
609 | ||
151b9b96 | 610 | r = sd_event_add_io(m->event, &m->signal_event_source, m->signal_fd, EPOLLIN, manager_dispatch_signal_fd, m); |
718db961 LP |
611 | if (r < 0) |
612 | return r; | |
ce578209 | 613 | |
7dfbe2e3 TG |
614 | (void) sd_event_source_set_description(m->signal_event_source, "manager-signal"); |
615 | ||
d8fdc620 LP |
616 | /* Process signals a bit earlier than the rest of things, but later than notify_fd processing, so that the |
617 | * notify processing can still figure out to which process/service a message belongs, before we reap the | |
618 | * process. Also, process this before handling cgroup notifications, so that we always collect child exit | |
619 | * status information before detecting that there's no process in a cgroup. */ | |
620 | r = sd_event_source_set_priority(m->signal_event_source, SD_EVENT_PRIORITY_NORMAL-6); | |
29083707 LP |
621 | if (r < 0) |
622 | return r; | |
623 | ||
463d0d15 | 624 | if (MANAGER_IS_SYSTEM(m)) |
80876c20 | 625 | return enable_special_signals(m); |
e1414003 | 626 | |
ce578209 LP |
627 | return 0; |
628 | } | |
629 | ||
1ad6e8b3 | 630 | static char** sanitize_environment(char **l) { |
f069efb4 | 631 | |
47cf8ff2 | 632 | /* Let's remove some environment variables that we need ourselves to communicate with our clients */ |
f069efb4 | 633 | strv_env_unset_many( |
1ad6e8b3 | 634 | l, |
8047ac8f LP |
635 | "CACHE_DIRECTORY", |
636 | "CONFIGURATION_DIRECTORY", | |
bb0c0d6f | 637 | "CREDENTIALS_DIRECTORY", |
47cf8ff2 LP |
638 | "EXIT_CODE", |
639 | "EXIT_STATUS", | |
640 | "INVOCATION_ID", | |
641 | "JOURNAL_STREAM", | |
642 | "LISTEN_FDNAMES", | |
643 | "LISTEN_FDS", | |
644 | "LISTEN_PID", | |
8047ac8f | 645 | "LOGS_DIRECTORY", |
c792a2e5 | 646 | "LOG_NAMESPACE", |
f069efb4 LP |
647 | "MAINPID", |
648 | "MANAGERPID", | |
6bb00842 LP |
649 | "MEMORY_PRESSURE_WATCH", |
650 | "MEMORY_PRESSURE_WRITE", | |
c792a2e5 LP |
651 | "MONITOR_EXIT_CODE", |
652 | "MONITOR_EXIT_STATUS", | |
653 | "MONITOR_INVOCATION_ID", | |
654 | "MONITOR_SERVICE_RESULT", | |
655 | "MONITOR_UNIT", | |
47cf8ff2 | 656 | "NOTIFY_SOCKET", |
dcf3c3c3 | 657 | "PIDFILE", |
47cf8ff2 LP |
658 | "REMOTE_ADDR", |
659 | "REMOTE_PORT", | |
8047ac8f | 660 | "RUNTIME_DIRECTORY", |
47cf8ff2 | 661 | "SERVICE_RESULT", |
8047ac8f | 662 | "STATE_DIRECTORY", |
c792a2e5 LP |
663 | "SYSTEMD_EXEC_PID", |
664 | "TRIGGER_PATH", | |
665 | "TRIGGER_TIMER_MONOTONIC_USEC", | |
666 | "TRIGGER_TIMER_REALTIME_USEC", | |
667 | "TRIGGER_UNIT", | |
f069efb4 LP |
668 | "WATCHDOG_PID", |
669 | "WATCHDOG_USEC", | |
670 | NULL); | |
47cf8ff2 LP |
671 | |
672 | /* Let's order the environment alphabetically, just to make it pretty */ | |
305757d8 | 673 | return strv_sort(l); |
f069efb4 LP |
674 | } |
675 | ||
79a224c4 | 676 | int manager_default_environment(Manager *m) { |
db11487d ZJS |
677 | int r; |
678 | ||
71ecc858 LP |
679 | assert(m); |
680 | ||
1ad6e8b3 LP |
681 | m->transient_environment = strv_free(m->transient_environment); |
682 | ||
463d0d15 | 683 | if (MANAGER_IS_SYSTEM(m)) { |
c792a2e5 LP |
684 | /* The system manager always starts with a clean environment for its children. It does not |
685 | * import the kernel's or the parents' exported variables. | |
e21fea24 | 686 | * |
c792a2e5 LP |
687 | * The initial passed environment is untouched to keep /proc/self/environ valid; it is used |
688 | * for tagging the init process inside containers. */ | |
1ad6e8b3 | 689 | m->transient_environment = strv_new("PATH=" DEFAULT_PATH); |
db11487d ZJS |
690 | if (!m->transient_environment) |
691 | return log_oom(); | |
e21fea24 KS |
692 | |
693 | /* Import locale variables LC_*= from configuration */ | |
1ad6e8b3 | 694 | (void) locale_setup(&m->transient_environment); |
db11487d | 695 | } else { |
aaf057c4 | 696 | /* The user manager passes its own environment along to its children, except for $PATH. */ |
1ad6e8b3 | 697 | m->transient_environment = strv_copy(environ); |
db11487d ZJS |
698 | if (!m->transient_environment) |
699 | return log_oom(); | |
700 | ||
aaf057c4 | 701 | r = strv_env_replace_strdup(&m->transient_environment, "PATH=" DEFAULT_USER_PATH); |
db11487d ZJS |
702 | if (r < 0) |
703 | return log_oom(); | |
db11487d | 704 | } |
8b55b8c4 | 705 | |
1ad6e8b3 | 706 | sanitize_environment(m->transient_environment); |
e21fea24 | 707 | return 0; |
71ecc858 LP |
708 | } |
709 | ||
3536f49e YW |
710 | static int manager_setup_prefix(Manager *m) { |
711 | struct table_entry { | |
712 | uint64_t type; | |
713 | const char *suffix; | |
714 | }; | |
715 | ||
72fd1768 | 716 | static const struct table_entry paths_system[_EXEC_DIRECTORY_TYPE_MAX] = { |
9978e631 ZJS |
717 | [EXEC_DIRECTORY_RUNTIME] = { SD_PATH_SYSTEM_RUNTIME, NULL }, |
718 | [EXEC_DIRECTORY_STATE] = { SD_PATH_SYSTEM_STATE_PRIVATE, NULL }, | |
719 | [EXEC_DIRECTORY_CACHE] = { SD_PATH_SYSTEM_STATE_CACHE, NULL }, | |
720 | [EXEC_DIRECTORY_LOGS] = { SD_PATH_SYSTEM_STATE_LOGS, NULL }, | |
3536f49e YW |
721 | [EXEC_DIRECTORY_CONFIGURATION] = { SD_PATH_SYSTEM_CONFIGURATION, NULL }, |
722 | }; | |
723 | ||
72fd1768 | 724 | static const struct table_entry paths_user[_EXEC_DIRECTORY_TYPE_MAX] = { |
9978e631 | 725 | [EXEC_DIRECTORY_RUNTIME] = { SD_PATH_USER_RUNTIME, NULL }, |
f9c91932 | 726 | [EXEC_DIRECTORY_STATE] = { SD_PATH_USER_STATE_PRIVATE, NULL }, |
9978e631 | 727 | [EXEC_DIRECTORY_CACHE] = { SD_PATH_USER_STATE_CACHE, NULL }, |
f9c91932 | 728 | [EXEC_DIRECTORY_LOGS] = { SD_PATH_USER_STATE_PRIVATE, "log" }, |
9978e631 | 729 | [EXEC_DIRECTORY_CONFIGURATION] = { SD_PATH_USER_CONFIGURATION, NULL }, |
3536f49e YW |
730 | }; |
731 | ||
3536f49e YW |
732 | assert(m); |
733 | ||
24b45970 ZJS |
734 | const struct table_entry *p = MANAGER_IS_SYSTEM(m) ? paths_system : paths_user; |
735 | int r; | |
3536f49e | 736 | |
24b45970 | 737 | for (ExecDirectoryType i = 0; i < _EXEC_DIRECTORY_TYPE_MAX; i++) { |
51327bcc | 738 | r = sd_path_lookup(p[i].type, p[i].suffix, &m->prefix[i]); |
3536f49e | 739 | if (r < 0) |
998df7ce ZJS |
740 | return log_warning_errno(r, "Failed to lookup %s path: %m", |
741 | exec_directory_type_to_string(i)); | |
3536f49e YW |
742 | } |
743 | ||
744 | return 0; | |
745 | } | |
746 | ||
e8630e69 ZJS |
747 | static void manager_free_unit_name_maps(Manager *m) { |
748 | m->unit_id_map = hashmap_free(m->unit_id_map); | |
749 | m->unit_name_map = hashmap_free(m->unit_name_map); | |
3fb2326f | 750 | m->unit_path_cache = set_free(m->unit_path_cache); |
c2911d48 | 751 | m->unit_cache_timestamp_hash = 0; |
e8630e69 ZJS |
752 | } |
753 | ||
279d81dd LP |
754 | static int manager_setup_run_queue(Manager *m) { |
755 | int r; | |
756 | ||
757 | assert(m); | |
758 | assert(!m->run_queue_event_source); | |
759 | ||
760 | r = sd_event_add_defer(m->event, &m->run_queue_event_source, manager_dispatch_run_queue, m); | |
761 | if (r < 0) | |
762 | return r; | |
763 | ||
764 | r = sd_event_source_set_priority(m->run_queue_event_source, SD_EVENT_PRIORITY_IDLE); | |
765 | if (r < 0) | |
766 | return r; | |
767 | ||
768 | r = sd_event_source_set_enabled(m->run_queue_event_source, SD_EVENT_OFF); | |
769 | if (r < 0) | |
770 | return r; | |
771 | ||
772 | (void) sd_event_source_set_description(m->run_queue_event_source, "manager-run-queue"); | |
773 | ||
774 | return 0; | |
775 | } | |
776 | ||
575b300b LP |
777 | static int manager_setup_sigchld_event_source(Manager *m) { |
778 | int r; | |
779 | ||
780 | assert(m); | |
781 | assert(!m->sigchld_event_source); | |
782 | ||
783 | r = sd_event_add_defer(m->event, &m->sigchld_event_source, manager_dispatch_sigchld, m); | |
784 | if (r < 0) | |
785 | return r; | |
786 | ||
787 | r = sd_event_source_set_priority(m->sigchld_event_source, SD_EVENT_PRIORITY_NORMAL-7); | |
788 | if (r < 0) | |
789 | return r; | |
790 | ||
791 | r = sd_event_source_set_enabled(m->sigchld_event_source, SD_EVENT_OFF); | |
792 | if (r < 0) | |
793 | return r; | |
794 | ||
795 | (void) sd_event_source_set_description(m->sigchld_event_source, "manager-sigchld"); | |
796 | ||
797 | return 0; | |
798 | } | |
799 | ||
29e6b0c1 LP |
800 | int manager_setup_memory_pressure_event_source(Manager *m) { |
801 | int r; | |
802 | ||
803 | assert(m); | |
804 | ||
805 | m->memory_pressure_event_source = sd_event_source_disable_unref(m->memory_pressure_event_source); | |
806 | ||
807 | r = sd_event_add_memory_pressure(m->event, &m->memory_pressure_event_source, NULL, NULL); | |
808 | if (r < 0) | |
809 | log_full_errno(ERRNO_IS_NOT_SUPPORTED(r) || ERRNO_IS_PRIVILEGE(r) || (r == -EHOSTDOWN) ? LOG_DEBUG : LOG_NOTICE, r, | |
810 | "Failed to establish memory pressure event source, ignoring: %m"); | |
c9e120e0 | 811 | else if (m->defaults.memory_pressure_threshold_usec != USEC_INFINITY) { |
6bb00842 LP |
812 | |
813 | /* If there's a default memory pressure threshold set, also apply it to the service manager itself */ | |
814 | r = sd_event_source_set_memory_pressure_period( | |
815 | m->memory_pressure_event_source, | |
c9e120e0 | 816 | m->defaults.memory_pressure_threshold_usec, |
6bb00842 LP |
817 | MEMORY_PRESSURE_DEFAULT_WINDOW_USEC); |
818 | if (r < 0) | |
819 | log_warning_errno(r, "Failed to adjust memory pressure threshold, ignoring: %m"); | |
820 | } | |
29e6b0c1 LP |
821 | |
822 | return 0; | |
823 | } | |
824 | ||
2ad591a3 LP |
825 | static int manager_find_credentials_dirs(Manager *m) { |
826 | const char *e; | |
827 | int r; | |
828 | ||
829 | assert(m); | |
830 | ||
831 | r = get_credentials_dir(&e); | |
832 | if (r < 0) { | |
833 | if (r != -ENXIO) | |
834 | log_debug_errno(r, "Failed to determine credentials directory, ignoring: %m"); | |
835 | } else { | |
836 | m->received_credentials_directory = strdup(e); | |
837 | if (!m->received_credentials_directory) | |
838 | return -ENOMEM; | |
839 | } | |
840 | ||
841 | r = get_encrypted_credentials_dir(&e); | |
842 | if (r < 0) { | |
843 | if (r != -ENXIO) | |
844 | log_debug_errno(r, "Failed to determine encrypted credentials directory, ignoring: %m"); | |
845 | } else { | |
846 | m->received_encrypted_credentials_directory = strdup(e); | |
847 | if (!m->received_encrypted_credentials_directory) | |
848 | return -ENOMEM; | |
849 | } | |
850 | ||
851 | return 0; | |
852 | } | |
853 | ||
d35fe8c0 | 854 | void manager_set_switching_root(Manager *m, bool switching_root) { |
7fa49280 LP |
855 | assert(m); |
856 | ||
d35fe8c0 FB |
857 | m->switching_root = MANAGER_IS_SYSTEM(m) && switching_root; |
858 | } | |
859 | ||
4870133b | 860 | int manager_new(RuntimeScope runtime_scope, ManagerTestRunFlags test_run_flags, Manager **_m) { |
c70cac54 | 861 | _cleanup_(manager_freep) Manager *m = NULL; |
e3dd987c | 862 | int r; |
8e274523 LP |
863 | |
864 | assert(_m); | |
4870133b | 865 | assert(IN_SET(runtime_scope, RUNTIME_SCOPE_SYSTEM, RUNTIME_SCOPE_USER)); |
ce578209 | 866 | |
3ad228ce | 867 | m = new(Manager, 1); |
915b3753 | 868 | if (!m) |
8e274523 | 869 | return -ENOMEM; |
60918275 | 870 | |
3ad228ce | 871 | *m = (Manager) { |
4870133b | 872 | .runtime_scope = runtime_scope, |
3ad228ce LP |
873 | .objective = _MANAGER_OBJECTIVE_INVALID, |
874 | ||
36cf4507 ZJS |
875 | .status_unit_format = STATUS_UNIT_FORMAT_DEFAULT, |
876 | ||
3ad228ce LP |
877 | .original_log_level = -1, |
878 | .original_log_target = _LOG_TARGET_INVALID, | |
879 | ||
986935cf FB |
880 | .watchdog_overridden[WATCHDOG_RUNTIME] = USEC_INFINITY, |
881 | .watchdog_overridden[WATCHDOG_REBOOT] = USEC_INFINITY, | |
882 | .watchdog_overridden[WATCHDOG_KEXEC] = USEC_INFINITY, | |
5717062e | 883 | .watchdog_overridden[WATCHDOG_PRETIMEOUT] = USEC_INFINITY, |
986935cf | 884 | |
44a41954 FB |
885 | .show_status_overridden = _SHOW_STATUS_INVALID, |
886 | ||
254d1313 ZJS |
887 | .notify_fd = -EBADF, |
888 | .cgroups_agent_fd = -EBADF, | |
889 | .signal_fd = -EBADF, | |
19ee48a6 | 890 | .user_lookup_fds = PIPE_EBADF, |
254d1313 ZJS |
891 | .private_listen_fd = -EBADF, |
892 | .dev_autofs_fd = -EBADF, | |
893 | .cgroup_inotify_fd = -EBADF, | |
894 | .pin_cgroupfs_fd = -EBADF, | |
895 | .ask_password_inotify_fd = -EBADF, | |
896 | .idle_pipe = { -EBADF, -EBADF, -EBADF, -EBADF}, | |
3ad228ce LP |
897 | |
898 | /* start as id #1, so that we can leave #0 around as "null-like" value */ | |
899 | .current_job_id = 1, | |
900 | ||
901 | .have_ask_password = -EINVAL, /* we don't know */ | |
902 | .first_boot = -1, | |
903 | .test_run_flags = test_run_flags, | |
afcfaa69 | 904 | |
d9365956 LB |
905 | .dump_ratelimit = { |
906 | .interval = 10 * USEC_PER_MINUTE, | |
907 | .burst = 10, | |
908 | }, | |
3ad228ce | 909 | }; |
80876c20 | 910 | |
ea09a416 LP |
911 | unit_defaults_init(&m->defaults, runtime_scope); |
912 | ||
349cc4a5 | 913 | #if ENABLE_EFI |
463d0d15 | 914 | if (MANAGER_IS_SYSTEM(m) && detect_container() <= 0) |
9f9f0342 LP |
915 | boot_timestamps(m->timestamps + MANAGER_TIMESTAMP_USERSPACE, |
916 | m->timestamps + MANAGER_TIMESTAMP_FIRMWARE, | |
917 | m->timestamps + MANAGER_TIMESTAMP_LOADER); | |
463d0d15 LP |
918 | #endif |
919 | ||
f2341e0a | 920 | /* Prepare log fields we can use for structured logging */ |
463d0d15 LP |
921 | if (MANAGER_IS_SYSTEM(m)) { |
922 | m->unit_log_field = "UNIT="; | |
923 | m->unit_log_format_string = "UNIT=%s"; | |
4b58153d LP |
924 | |
925 | m->invocation_log_field = "INVOCATION_ID="; | |
f1c50bec | 926 | m->invocation_log_format_string = "INVOCATION_ID=%s"; |
463d0d15 LP |
927 | } else { |
928 | m->unit_log_field = "USER_UNIT="; | |
929 | m->unit_log_format_string = "USER_UNIT=%s"; | |
4b58153d LP |
930 | |
931 | m->invocation_log_field = "USER_INVOCATION_ID="; | |
f1c50bec | 932 | m->invocation_log_format_string = "USER_INVOCATION_ID=%s"; |
463d0d15 | 933 | } |
f2341e0a | 934 | |
2e5c94b9 | 935 | /* Reboot immediately if the user hits C-A-D more often than 7x per 2s */ |
7d1e61ca | 936 | m->ctrl_alt_del_ratelimit = (const RateLimit) { .interval = 2 * USEC_PER_SEC, .burst = 7 }; |
2e5c94b9 | 937 | |
e21fea24 KS |
938 | r = manager_default_environment(m); |
939 | if (r < 0) | |
c70cac54 | 940 | return r; |
1137a57c | 941 | |
d5099efc | 942 | r = hashmap_ensure_allocated(&m->units, &string_hash_ops); |
718db961 | 943 | if (r < 0) |
c70cac54 | 944 | return r; |
60918275 | 945 | |
548f6937 | 946 | r = hashmap_ensure_allocated(&m->cgroup_unit, &path_hash_ops); |
718db961 | 947 | if (r < 0) |
c70cac54 | 948 | return r; |
9152c765 | 949 | |
d5099efc | 950 | r = hashmap_ensure_allocated(&m->watch_bus, &string_hash_ops); |
718db961 | 951 | if (r < 0) |
c70cac54 | 952 | return r; |
05e343b7 | 953 | |
da8e1782 MO |
954 | r = prioq_ensure_allocated(&m->run_queue, compare_job_priority); |
955 | if (r < 0) | |
956 | return r; | |
957 | ||
e8112e67 | 958 | r = manager_setup_prefix(m); |
718db961 | 959 | if (r < 0) |
c70cac54 | 960 | return r; |
8742514c | 961 | |
2ad591a3 LP |
962 | r = manager_find_credentials_dirs(m); |
963 | if (r < 0) | |
964 | return r; | |
bb0c0d6f | 965 | |
e8112e67 | 966 | r = sd_event_default(&m->event); |
8742514c | 967 | if (r < 0) |
c70cac54 | 968 | return r; |
9152c765 | 969 | |
e8112e67 | 970 | r = manager_setup_run_queue(m); |
a1d32bac | 971 | if (r < 0) |
c70cac54 | 972 | return r; |
8e274523 | 973 | |
5dd2f5ff | 974 | if (FLAGS_SET(test_run_flags, MANAGER_TEST_RUN_MINIMAL)) { |
e8112e67 ZJS |
975 | m->cgroup_root = strdup(""); |
976 | if (!m->cgroup_root) | |
977 | return -ENOMEM; | |
978 | } else { | |
979 | r = manager_setup_signals(m); | |
980 | if (r < 0) | |
981 | return r; | |
8c47c732 | 982 | |
e8112e67 ZJS |
983 | r = manager_setup_cgroup(m); |
984 | if (r < 0) | |
985 | return r; | |
575b300b | 986 | |
e8112e67 ZJS |
987 | r = manager_setup_time_change(m); |
988 | if (r < 0) | |
989 | return r; | |
9670d583 | 990 | |
bbf5fd8e LP |
991 | r = manager_read_timezone_stat(m); |
992 | if (r < 0) | |
993 | return r; | |
994 | ||
ea5c5f68 | 995 | (void) manager_setup_timezone_change(m); |
bbf5fd8e | 996 | |
e8112e67 ZJS |
997 | r = manager_setup_sigchld_event_source(m); |
998 | if (r < 0) | |
999 | return r; | |
b1994387 | 1000 | |
29e6b0c1 LP |
1001 | r = manager_setup_memory_pressure_event_source(m); |
1002 | if (r < 0) | |
1003 | return r; | |
1004 | ||
b1994387 | 1005 | #if HAVE_LIBBPF |
ba187c9c | 1006 | if (MANAGER_IS_SYSTEM(m) && lsm_bpf_supported(/* initialize = */ true)) { |
b1994387 ILG |
1007 | r = lsm_bpf_setup(m); |
1008 | if (r < 0) | |
299d9417 | 1009 | log_warning_errno(r, "Failed to setup LSM BPF, ignoring: %m"); |
b1994387 ILG |
1010 | } |
1011 | #endif | |
e8112e67 | 1012 | } |
e27fe688 | 1013 | |
2f8c48b6 AZ |
1014 | if (test_run_flags == 0) { |
1015 | if (MANAGER_IS_SYSTEM(m)) | |
1016 | r = mkdir_label("/run/systemd/units", 0755); | |
1017 | else { | |
1018 | _cleanup_free_ char *units_path = NULL; | |
1019 | r = xdg_user_runtime_dir(&units_path, "/systemd/units"); | |
1020 | if (r < 0) | |
1021 | return r; | |
1022 | r = mkdir_p_label(units_path, 0755); | |
1023 | } | |
1024 | ||
d3070fbd | 1025 | if (r < 0 && r != -EEXIST) |
c70cac54 | 1026 | return r; |
d3070fbd LP |
1027 | } |
1028 | ||
232f6754 ZJS |
1029 | /* Note that we do not set up the notify fd here. We do that after deserialization, |
1030 | * since they might have gotten serialized across the reexec. */ | |
3536f49e | 1031 | |
1cc6c93a YW |
1032 | *_m = TAKE_PTR(m); |
1033 | ||
8e274523 | 1034 | return 0; |
60918275 LP |
1035 | } |
1036 | ||
d86f9d52 | 1037 | static int manager_setup_notify(Manager *m) { |
7181dbdb | 1038 | int r; |
d86f9d52 | 1039 | |
638cece4 | 1040 | if (MANAGER_IS_TEST_RUN(m)) |
0d8c31ff ZJS |
1041 | return 0; |
1042 | ||
d86f9d52 | 1043 | if (m->notify_fd < 0) { |
254d1313 | 1044 | _cleanup_close_ int fd = -EBADF; |
f36a9d59 ZJS |
1045 | union sockaddr_union sa; |
1046 | socklen_t sa_len; | |
d86f9d52 LP |
1047 | |
1048 | /* First free all secondary fields */ | |
a1e58e8e | 1049 | m->notify_socket = mfree(m->notify_socket); |
5dcadb4c | 1050 | m->notify_event_source = sd_event_source_disable_unref(m->notify_event_source); |
d86f9d52 LP |
1051 | |
1052 | fd = socket(AF_UNIX, SOCK_DGRAM|SOCK_CLOEXEC|SOCK_NONBLOCK, 0); | |
4a62c710 MS |
1053 | if (fd < 0) |
1054 | return log_error_errno(errno, "Failed to allocate notification socket: %m"); | |
d86f9d52 | 1055 | |
28e7e934 | 1056 | fd_increase_rxbuf(fd, NOTIFY_RCVBUF_SIZE); |
a47806fa | 1057 | |
b910cc72 | 1058 | m->notify_socket = path_join(m->prefix[EXEC_DIRECTORY_RUNTIME], "systemd/notify"); |
498e87d6 LP |
1059 | if (!m->notify_socket) |
1060 | return log_oom(); | |
1061 | ||
f36a9d59 ZJS |
1062 | r = sockaddr_un_set_path(&sa.un, m->notify_socket); |
1063 | if (r < 0) | |
1064 | return log_error_errno(r, "Notify socket '%s' not valid for AF_UNIX socket address, refusing.", | |
1065 | m->notify_socket); | |
1066 | sa_len = r; | |
15a3e96f | 1067 | |
498e87d6 | 1068 | (void) mkdir_parents_label(m->notify_socket, 0755); |
fbda85b0 | 1069 | (void) sockaddr_un_unlink(&sa.un); |
7181dbdb | 1070 | |
45ae2f72 | 1071 | r = mac_selinux_bind(fd, &sa.sa, sa_len); |
4a62c710 | 1072 | if (r < 0) |
45ae2f72 | 1073 | return log_error_errno(r, "bind(%s) failed: %m", m->notify_socket); |
d86f9d52 | 1074 | |
2ff48e98 | 1075 | r = setsockopt_int(fd, SOL_SOCKET, SO_PASSCRED, true); |
4a62c710 | 1076 | if (r < 0) |
2ff48e98 | 1077 | return log_error_errno(r, "SO_PASSCRED failed: %m"); |
d86f9d52 | 1078 | |
c10d6bdb | 1079 | m->notify_fd = TAKE_FD(fd); |
d86f9d52 LP |
1080 | |
1081 | log_debug("Using notification socket %s", m->notify_socket); | |
1082 | } | |
1083 | ||
1084 | if (!m->notify_event_source) { | |
151b9b96 | 1085 | r = sd_event_add_io(m->event, &m->notify_event_source, m->notify_fd, EPOLLIN, manager_dispatch_notify_fd, m); |
895b3a7b MS |
1086 | if (r < 0) |
1087 | return log_error_errno(r, "Failed to allocate notify event source: %m"); | |
d86f9d52 | 1088 | |
d8fdc620 LP |
1089 | /* Process notification messages a bit earlier than SIGCHLD, so that we can still identify to which |
1090 | * service an exit message belongs. */ | |
575b300b | 1091 | r = sd_event_source_set_priority(m->notify_event_source, SD_EVENT_PRIORITY_NORMAL-8); |
23bbb0de MS |
1092 | if (r < 0) |
1093 | return log_error_errno(r, "Failed to set priority of notify event source: %m"); | |
7dfbe2e3 TG |
1094 | |
1095 | (void) sd_event_source_set_description(m->notify_event_source, "manager-notify"); | |
d86f9d52 LP |
1096 | } |
1097 | ||
1098 | return 0; | |
1099 | } | |
1100 | ||
d8fdc620 LP |
1101 | static int manager_setup_cgroups_agent(Manager *m) { |
1102 | ||
1103 | static const union sockaddr_union sa = { | |
1104 | .un.sun_family = AF_UNIX, | |
1105 | .un.sun_path = "/run/systemd/cgroups-agent", | |
1106 | }; | |
1107 | int r; | |
1108 | ||
1109 | /* This creates a listening socket we receive cgroups agent messages on. We do not use D-Bus for delivering | |
1110 | * these messages from the cgroups agent binary to PID 1, as the cgroups agent binary is very short-living, and | |
1111 | * each instance of it needs a new D-Bus connection. Since D-Bus connections are SOCK_STREAM/AF_UNIX, on | |
1112 | * overloaded systems the backlog of the D-Bus socket becomes relevant, as not more than the configured number | |
1113 | * of D-Bus connections may be queued until the kernel will start dropping further incoming connections, | |
1114 | * possibly resulting in lost cgroups agent messages. To avoid this, we'll use a private SOCK_DGRAM/AF_UNIX | |
1115 | * socket, where no backlog is relevant as communication may take place without an actual connect() cycle, and | |
1116 | * we thus won't lose messages. | |
1117 | * | |
1118 | * Note that PID 1 will forward the agent message to system bus, so that the user systemd instance may listen | |
1119 | * to it. The system instance hence listens on this special socket, but the user instances listen on the system | |
1120 | * bus for these messages. */ | |
1121 | ||
638cece4 | 1122 | if (MANAGER_IS_TEST_RUN(m)) |
d8fdc620 LP |
1123 | return 0; |
1124 | ||
1125 | if (!MANAGER_IS_SYSTEM(m)) | |
1126 | return 0; | |
1127 | ||
c22800e4 | 1128 | r = cg_unified_controller(SYSTEMD_CGROUP_CONTROLLER); |
b4cccbc1 LP |
1129 | if (r < 0) |
1130 | return log_error_errno(r, "Failed to determine whether unified cgroups hierarchy is used: %m"); | |
1131 | if (r > 0) /* We don't need this anymore on the unified hierarchy */ | |
d8fdc620 LP |
1132 | return 0; |
1133 | ||
1134 | if (m->cgroups_agent_fd < 0) { | |
254d1313 | 1135 | _cleanup_close_ int fd = -EBADF; |
d8fdc620 LP |
1136 | |
1137 | /* First free all secondary fields */ | |
5dcadb4c | 1138 | m->cgroups_agent_event_source = sd_event_source_disable_unref(m->cgroups_agent_event_source); |
d8fdc620 LP |
1139 | |
1140 | fd = socket(AF_UNIX, SOCK_DGRAM|SOCK_CLOEXEC|SOCK_NONBLOCK, 0); | |
1141 | if (fd < 0) | |
1142 | return log_error_errno(errno, "Failed to allocate cgroups agent socket: %m"); | |
1143 | ||
28e7e934 | 1144 | fd_increase_rxbuf(fd, CGROUPS_AGENT_RCVBUF_SIZE); |
d8fdc620 | 1145 | |
155b6876 | 1146 | (void) sockaddr_un_unlink(&sa.un); |
d8fdc620 LP |
1147 | |
1148 | /* Only allow root to connect to this socket */ | |
2053593f | 1149 | WITH_UMASK(0077) |
fc2fffe7 | 1150 | r = bind(fd, &sa.sa, SOCKADDR_UN_LEN(sa.un)); |
d8fdc620 LP |
1151 | if (r < 0) |
1152 | return log_error_errno(errno, "bind(%s) failed: %m", sa.un.sun_path); | |
1153 | ||
0b7e8b32 | 1154 | m->cgroups_agent_fd = TAKE_FD(fd); |
d8fdc620 LP |
1155 | } |
1156 | ||
1157 | if (!m->cgroups_agent_event_source) { | |
1158 | r = sd_event_add_io(m->event, &m->cgroups_agent_event_source, m->cgroups_agent_fd, EPOLLIN, manager_dispatch_cgroups_agent_fd, m); | |
1159 | if (r < 0) | |
1160 | return log_error_errno(r, "Failed to allocate cgroups agent event source: %m"); | |
1161 | ||
cbe83389 LP |
1162 | /* Process cgroups notifications early. Note that when the agent notification is received |
1163 | * we'll just enqueue the unit in the cgroup empty queue, hence pick a high priority than | |
1164 | * that. Also see handling of cgroup inotify for the unified cgroup stuff. */ | |
1165 | r = sd_event_source_set_priority(m->cgroups_agent_event_source, SD_EVENT_PRIORITY_NORMAL-9); | |
d8fdc620 LP |
1166 | if (r < 0) |
1167 | return log_error_errno(r, "Failed to set priority of cgroups agent event source: %m"); | |
1168 | ||
1169 | (void) sd_event_source_set_description(m->cgroups_agent_event_source, "manager-cgroups-agent"); | |
1170 | } | |
1171 | ||
1172 | return 0; | |
1173 | } | |
1174 | ||
00d9ef85 LP |
1175 | static int manager_setup_user_lookup_fd(Manager *m) { |
1176 | int r; | |
1177 | ||
1178 | assert(m); | |
1179 | ||
1180 | /* Set up the socket pair used for passing UID/GID resolution results from forked off processes to PID | |
1181 | * 1. Background: we can't do name lookups (NSS) from PID 1, since it might involve IPC and thus activation, | |
1182 | * and we might hence deadlock on ourselves. Hence we do all user/group lookups asynchronously from the forked | |
1183 | * off processes right before executing the binaries to start. In order to be able to clean up any IPC objects | |
1184 | * created by a unit (see RemoveIPC=) we need to know in PID 1 the used UID/GID of the executed processes, | |
1185 | * hence we establish this communication channel so that forked off processes can pass their UID/GID | |
1186 | * information back to PID 1. The forked off processes send their resolved UID/GID to PID 1 in a simple | |
1187 | * datagram, along with their unit name, so that we can share one communication socket pair among all units for | |
1188 | * this purpose. | |
1189 | * | |
1190 | * You might wonder why we need a communication channel for this that is independent of the usual notification | |
1191 | * socket scheme (i.e. $NOTIFY_SOCKET). The primary difference is about trust: data sent via the $NOTIFY_SOCKET | |
1192 | * channel is only accepted if it originates from the right unit and if reception was enabled for it. The user | |
1193 | * lookup socket OTOH is only accessible by PID 1 and its children until they exec(), and always available. | |
1194 | * | |
1195 | * Note that this function is called under two circumstances: when we first initialize (in which case we | |
1196 | * allocate both the socket pair and the event source to listen on it), and when we deserialize after a reload | |
1197 | * (in which case the socket pair already exists but we still need to allocate the event source for it). */ | |
1198 | ||
1199 | if (m->user_lookup_fds[0] < 0) { | |
1200 | ||
1201 | /* Free all secondary fields */ | |
1202 | safe_close_pair(m->user_lookup_fds); | |
5dcadb4c | 1203 | m->user_lookup_event_source = sd_event_source_disable_unref(m->user_lookup_event_source); |
00d9ef85 LP |
1204 | |
1205 | if (socketpair(AF_UNIX, SOCK_DGRAM|SOCK_CLOEXEC, 0, m->user_lookup_fds) < 0) | |
1206 | return log_error_errno(errno, "Failed to allocate user lookup socket: %m"); | |
1207 | ||
28e7e934 | 1208 | (void) fd_increase_rxbuf(m->user_lookup_fds[0], NOTIFY_RCVBUF_SIZE); |
00d9ef85 LP |
1209 | } |
1210 | ||
1211 | if (!m->user_lookup_event_source) { | |
1212 | r = sd_event_add_io(m->event, &m->user_lookup_event_source, m->user_lookup_fds[0], EPOLLIN, manager_dispatch_user_lookup_fd, m); | |
1213 | if (r < 0) | |
1214 | return log_error_errno(errno, "Failed to allocate user lookup event source: %m"); | |
1215 | ||
1216 | /* Process even earlier than the notify event source, so that we always know first about valid UID/GID | |
1217 | * resolutions */ | |
67ae4e8d | 1218 | r = sd_event_source_set_priority(m->user_lookup_event_source, SD_EVENT_PRIORITY_NORMAL-11); |
00d9ef85 | 1219 | if (r < 0) |
5238e957 | 1220 | return log_error_errno(errno, "Failed to set priority of user lookup event source: %m"); |
00d9ef85 LP |
1221 | |
1222 | (void) sd_event_source_set_description(m->user_lookup_event_source, "user-lookup"); | |
1223 | } | |
1224 | ||
1225 | return 0; | |
1226 | } | |
1227 | ||
23a177ef | 1228 | static unsigned manager_dispatch_cleanup_queue(Manager *m) { |
595ed347 | 1229 | Unit *u; |
23a177ef LP |
1230 | unsigned n = 0; |
1231 | ||
1232 | assert(m); | |
1233 | ||
595ed347 MS |
1234 | while ((u = m->cleanup_queue)) { |
1235 | assert(u->in_cleanup_queue); | |
23a177ef | 1236 | |
595ed347 | 1237 | unit_free(u); |
23a177ef LP |
1238 | n++; |
1239 | } | |
1240 | ||
1241 | return n; | |
1242 | } | |
1243 | ||
6ac62d61 LP |
1244 | static unsigned manager_dispatch_release_resources_queue(Manager *m) { |
1245 | unsigned n = 0; | |
1246 | Unit *u; | |
1247 | ||
1248 | assert(m); | |
1249 | ||
52e3671b | 1250 | while ((u = LIST_POP(release_resources_queue, m->release_resources_queue))) { |
6ac62d61 | 1251 | assert(u->in_release_resources_queue); |
6ac62d61 LP |
1252 | u->in_release_resources_queue = false; |
1253 | ||
1254 | n++; | |
1255 | ||
1256 | unit_release_resources(u); | |
1257 | } | |
1258 | ||
1259 | return n; | |
1260 | } | |
1261 | ||
eced69b3 | 1262 | enum { |
35b8ca3a | 1263 | GC_OFFSET_IN_PATH, /* This one is on the path we were traveling */ |
eced69b3 LP |
1264 | GC_OFFSET_UNSURE, /* No clue */ |
1265 | GC_OFFSET_GOOD, /* We still need this unit */ | |
1266 | GC_OFFSET_BAD, /* We don't need this unit anymore */ | |
1267 | _GC_OFFSET_MAX | |
1268 | }; | |
1269 | ||
00d9ef85 | 1270 | static void unit_gc_mark_good(Unit *u, unsigned gc_marker) { |
4892084f LN |
1271 | Unit *other; |
1272 | ||
1273 | u->gc_marker = gc_marker + GC_OFFSET_GOOD; | |
1274 | ||
1275 | /* Recursively mark referenced units as GOOD as well */ | |
15ed3c3a | 1276 | UNIT_FOREACH_DEPENDENCY(other, u, UNIT_ATOM_REFERENCES) |
4892084f LN |
1277 | if (other->gc_marker == gc_marker + GC_OFFSET_UNSURE) |
1278 | unit_gc_mark_good(other, gc_marker); | |
1279 | } | |
1280 | ||
eced69b3 | 1281 | static void unit_gc_sweep(Unit *u, unsigned gc_marker) { |
701cc384 | 1282 | Unit *other; |
eced69b3 | 1283 | bool is_bad; |
701cc384 LP |
1284 | |
1285 | assert(u); | |
1286 | ||
4c701096 YW |
1287 | if (IN_SET(u->gc_marker - gc_marker, |
1288 | GC_OFFSET_GOOD, GC_OFFSET_BAD, GC_OFFSET_UNSURE, GC_OFFSET_IN_PATH)) | |
701cc384 LP |
1289 | return; |
1290 | ||
ac155bb8 | 1291 | if (u->in_cleanup_queue) |
701cc384 LP |
1292 | goto bad; |
1293 | ||
f2f725e5 | 1294 | if (!unit_may_gc(u)) |
701cc384 LP |
1295 | goto good; |
1296 | ||
ac155bb8 | 1297 | u->gc_marker = gc_marker + GC_OFFSET_IN_PATH; |
eced69b3 LP |
1298 | |
1299 | is_bad = true; | |
1300 | ||
15ed3c3a | 1301 | UNIT_FOREACH_DEPENDENCY(other, u, UNIT_ATOM_REFERENCED_BY) { |
701cc384 LP |
1302 | unit_gc_sweep(other, gc_marker); |
1303 | ||
ac155bb8 | 1304 | if (other->gc_marker == gc_marker + GC_OFFSET_GOOD) |
701cc384 | 1305 | goto good; |
eced69b3 | 1306 | |
ac155bb8 | 1307 | if (other->gc_marker != gc_marker + GC_OFFSET_BAD) |
eced69b3 | 1308 | is_bad = false; |
701cc384 LP |
1309 | } |
1310 | ||
ef3fc326 ZJS |
1311 | LIST_FOREACH(refs_by_target, ref, u->refs_by_target) { |
1312 | unit_gc_sweep(ref->source, gc_marker); | |
2641f02e | 1313 | |
ef3fc326 ZJS |
1314 | if (ref->source->gc_marker == gc_marker + GC_OFFSET_GOOD) |
1315 | goto good; | |
2641f02e | 1316 | |
ef3fc326 ZJS |
1317 | if (ref->source->gc_marker != gc_marker + GC_OFFSET_BAD) |
1318 | is_bad = false; | |
2641f02e | 1319 | } |
701cc384 | 1320 | |
eced69b3 LP |
1321 | if (is_bad) |
1322 | goto bad; | |
1323 | ||
1324 | /* We were unable to find anything out about this entry, so | |
1325 | * let's investigate it later */ | |
ac155bb8 | 1326 | u->gc_marker = gc_marker + GC_OFFSET_UNSURE; |
eced69b3 LP |
1327 | unit_add_to_gc_queue(u); |
1328 | return; | |
1329 | ||
701cc384 | 1330 | bad: |
eced69b3 LP |
1331 | /* We definitely know that this one is not useful anymore, so |
1332 | * let's mark it for deletion */ | |
ac155bb8 | 1333 | u->gc_marker = gc_marker + GC_OFFSET_BAD; |
eced69b3 | 1334 | unit_add_to_cleanup_queue(u); |
701cc384 LP |
1335 | return; |
1336 | ||
1337 | good: | |
4892084f | 1338 | unit_gc_mark_good(u, gc_marker); |
701cc384 LP |
1339 | } |
1340 | ||
c5a97ed1 LP |
1341 | static unsigned manager_dispatch_gc_unit_queue(Manager *m) { |
1342 | unsigned n = 0, gc_marker; | |
595ed347 | 1343 | Unit *u; |
701cc384 LP |
1344 | |
1345 | assert(m); | |
1346 | ||
cf1265e1 | 1347 | /* log_debug("Running GC..."); */ |
701cc384 | 1348 | |
eced69b3 LP |
1349 | m->gc_marker += _GC_OFFSET_MAX; |
1350 | if (m->gc_marker + _GC_OFFSET_MAX <= _GC_OFFSET_MAX) | |
c9c0cadb | 1351 | m->gc_marker = 1; |
701cc384 | 1352 | |
eced69b3 LP |
1353 | gc_marker = m->gc_marker; |
1354 | ||
52e3671b | 1355 | while ((u = LIST_POP(gc_queue, m->gc_unit_queue))) { |
595ed347 | 1356 | assert(u->in_gc_queue); |
701cc384 | 1357 | |
595ed347 | 1358 | unit_gc_sweep(u, gc_marker); |
eced69b3 | 1359 | |
595ed347 | 1360 | u->in_gc_queue = false; |
701cc384 LP |
1361 | |
1362 | n++; | |
1363 | ||
4c701096 YW |
1364 | if (IN_SET(u->gc_marker - gc_marker, |
1365 | GC_OFFSET_BAD, GC_OFFSET_UNSURE)) { | |
cc3bc3e6 | 1366 | if (u->id) |
f2341e0a | 1367 | log_unit_debug(u, "Collecting."); |
595ed347 MS |
1368 | u->gc_marker = gc_marker + GC_OFFSET_BAD; |
1369 | unit_add_to_cleanup_queue(u); | |
701cc384 LP |
1370 | } |
1371 | } | |
1372 | ||
701cc384 LP |
1373 | return n; |
1374 | } | |
1375 | ||
c5a97ed1 LP |
1376 | static unsigned manager_dispatch_gc_job_queue(Manager *m) { |
1377 | unsigned n = 0; | |
1378 | Job *j; | |
1379 | ||
1380 | assert(m); | |
1381 | ||
52e3671b | 1382 | while ((j = LIST_POP(gc_queue, m->gc_job_queue))) { |
c5a97ed1 | 1383 | assert(j->in_gc_queue); |
c5a97ed1 LP |
1384 | j->in_gc_queue = false; |
1385 | ||
1386 | n++; | |
1387 | ||
2ab3050f | 1388 | if (!job_may_gc(j)) |
c5a97ed1 LP |
1389 | continue; |
1390 | ||
1391 | log_unit_debug(j->unit, "Collecting job."); | |
1392 | (void) job_finish_and_invalidate(j, JOB_COLLECTED, false, false); | |
1393 | } | |
1394 | ||
1395 | return n; | |
1396 | } | |
1397 | ||
7223d500 LB |
1398 | static int manager_ratelimit_requeue(sd_event_source *s, uint64_t usec, void *userdata) { |
1399 | Unit *u = userdata; | |
1400 | ||
1401 | assert(u); | |
1402 | assert(s == u->auto_start_stop_event_source); | |
1403 | ||
1404 | u->auto_start_stop_event_source = sd_event_source_unref(u->auto_start_stop_event_source); | |
1405 | ||
1406 | /* Re-queue to all queues, if the rate limit hit we might have been throttled on any of them. */ | |
1407 | unit_submit_to_stop_when_unneeded_queue(u); | |
1408 | unit_submit_to_start_when_upheld_queue(u); | |
1409 | unit_submit_to_stop_when_bound_queue(u); | |
1410 | ||
1411 | return 0; | |
1412 | } | |
1413 | ||
1414 | static int manager_ratelimit_check_and_queue(Unit *u) { | |
1415 | int r; | |
1416 | ||
1417 | assert(u); | |
1418 | ||
1419 | if (ratelimit_below(&u->auto_start_stop_ratelimit)) | |
1420 | return 1; | |
1421 | ||
1422 | /* Already queued, no need to requeue */ | |
1423 | if (u->auto_start_stop_event_source) | |
1424 | return 0; | |
1425 | ||
1426 | r = sd_event_add_time( | |
1427 | u->manager->event, | |
1428 | &u->auto_start_stop_event_source, | |
1429 | CLOCK_MONOTONIC, | |
1430 | ratelimit_end(&u->auto_start_stop_ratelimit), | |
1431 | 0, | |
1432 | manager_ratelimit_requeue, | |
1433 | u); | |
1434 | if (r < 0) | |
1435 | return log_unit_error_errno(u, r, "Failed to queue timer on event loop: %m"); | |
1436 | ||
1437 | return 0; | |
1438 | } | |
1439 | ||
a3c1168a LP |
1440 | static unsigned manager_dispatch_stop_when_unneeded_queue(Manager *m) { |
1441 | unsigned n = 0; | |
1442 | Unit *u; | |
1443 | int r; | |
1444 | ||
1445 | assert(m); | |
1446 | ||
52e3671b | 1447 | while ((u = LIST_POP(stop_when_unneeded_queue, m->stop_when_unneeded_queue))) { |
a3c1168a | 1448 | _cleanup_(sd_bus_error_free) sd_bus_error error = SD_BUS_ERROR_NULL; |
a3c1168a LP |
1449 | |
1450 | assert(u->in_stop_when_unneeded_queue); | |
a3c1168a LP |
1451 | u->in_stop_when_unneeded_queue = false; |
1452 | ||
1453 | n++; | |
1454 | ||
1455 | if (!unit_is_unneeded(u)) | |
1456 | continue; | |
1457 | ||
1458 | log_unit_debug(u, "Unit is not needed anymore."); | |
1459 | ||
1460 | /* If stopping a unit fails continuously we might enter a stop loop here, hence stop acting on the | |
1461 | * service being unnecessary after a while. */ | |
1462 | ||
7223d500 LB |
1463 | r = manager_ratelimit_check_and_queue(u); |
1464 | if (r <= 0) { | |
1465 | log_unit_warning(u, | |
1466 | "Unit not needed anymore, but not stopping since we tried this too often recently.%s", | |
1467 | r == 0 ? " Will retry later." : ""); | |
a3c1168a LP |
1468 | continue; |
1469 | } | |
1470 | ||
1471 | /* Ok, nobody needs us anymore. Sniff. Then let's commit suicide */ | |
50cbaba4 | 1472 | r = manager_add_job(u->manager, JOB_STOP, u, JOB_FAIL, NULL, &error, NULL); |
a3c1168a LP |
1473 | if (r < 0) |
1474 | log_unit_warning_errno(u, r, "Failed to enqueue stop job, ignoring: %s", bus_error_message(&error, r)); | |
1475 | } | |
1476 | ||
1477 | return n; | |
1478 | } | |
1479 | ||
0bc488c9 LP |
1480 | static unsigned manager_dispatch_start_when_upheld_queue(Manager *m) { |
1481 | unsigned n = 0; | |
1482 | Unit *u; | |
1483 | int r; | |
1484 | ||
1485 | assert(m); | |
1486 | ||
52e3671b | 1487 | while ((u = LIST_POP(start_when_upheld_queue, m->start_when_upheld_queue))) { |
0bc488c9 LP |
1488 | _cleanup_(sd_bus_error_free) sd_bus_error error = SD_BUS_ERROR_NULL; |
1489 | Unit *culprit = NULL; | |
1490 | ||
1491 | assert(u->in_start_when_upheld_queue); | |
0bc488c9 LP |
1492 | u->in_start_when_upheld_queue = false; |
1493 | ||
1494 | n++; | |
1495 | ||
1496 | if (!unit_is_upheld_by_active(u, &culprit)) | |
1497 | continue; | |
1498 | ||
1499 | log_unit_debug(u, "Unit is started because upheld by active unit %s.", culprit->id); | |
1500 | ||
1501 | /* If stopping a unit fails continuously we might enter a stop loop here, hence stop acting on the | |
1502 | * service being unnecessary after a while. */ | |
1503 | ||
7223d500 LB |
1504 | r = manager_ratelimit_check_and_queue(u); |
1505 | if (r <= 0) { | |
1506 | log_unit_warning(u, | |
1507 | "Unit needs to be started because active unit %s upholds it, but not starting since we tried this too often recently.%s", | |
1508 | culprit->id, | |
1509 | r == 0 ? " Will retry later." : ""); | |
0bc488c9 LP |
1510 | continue; |
1511 | } | |
1512 | ||
1513 | r = manager_add_job(u->manager, JOB_START, u, JOB_FAIL, NULL, &error, NULL); | |
1514 | if (r < 0) | |
1515 | log_unit_warning_errno(u, r, "Failed to enqueue start job, ignoring: %s", bus_error_message(&error, r)); | |
1516 | } | |
1517 | ||
1518 | return n; | |
1519 | } | |
1520 | ||
56c59592 LP |
1521 | static unsigned manager_dispatch_stop_when_bound_queue(Manager *m) { |
1522 | unsigned n = 0; | |
1523 | Unit *u; | |
1524 | int r; | |
1525 | ||
1526 | assert(m); | |
1527 | ||
52e3671b | 1528 | while ((u = LIST_POP(stop_when_bound_queue, m->stop_when_bound_queue))) { |
56c59592 LP |
1529 | _cleanup_(sd_bus_error_free) sd_bus_error error = SD_BUS_ERROR_NULL; |
1530 | Unit *culprit = NULL; | |
1531 | ||
1532 | assert(u->in_stop_when_bound_queue); | |
56c59592 LP |
1533 | u->in_stop_when_bound_queue = false; |
1534 | ||
1535 | n++; | |
1536 | ||
1537 | if (!unit_is_bound_by_inactive(u, &culprit)) | |
1538 | continue; | |
1539 | ||
1540 | log_unit_debug(u, "Unit is stopped because bound to inactive unit %s.", culprit->id); | |
1541 | ||
1542 | /* If stopping a unit fails continuously we might enter a stop loop here, hence stop acting on the | |
1543 | * service being unnecessary after a while. */ | |
1544 | ||
7223d500 LB |
1545 | r = manager_ratelimit_check_and_queue(u); |
1546 | if (r <= 0) { | |
1547 | log_unit_warning(u, | |
1548 | "Unit needs to be stopped because it is bound to inactive unit %s it, but not stopping since we tried this too often recently.%s", | |
1549 | culprit->id, | |
1550 | r == 0 ? " Will retry later." : ""); | |
56c59592 LP |
1551 | continue; |
1552 | } | |
1553 | ||
1554 | r = manager_add_job(u->manager, JOB_STOP, u, JOB_REPLACE, NULL, &error, NULL); | |
1555 | if (r < 0) | |
1556 | log_unit_warning_errno(u, r, "Failed to enqueue stop job, ignoring: %s", bus_error_message(&error, r)); | |
1557 | } | |
1558 | ||
1559 | return n; | |
1560 | } | |
1561 | ||
a16e1123 | 1562 | static void manager_clear_jobs_and_units(Manager *m) { |
a16e1123 | 1563 | Unit *u; |
60918275 LP |
1564 | |
1565 | assert(m); | |
1566 | ||
87f0e418 LP |
1567 | while ((u = hashmap_first(m->units))) |
1568 | unit_free(u); | |
964e0949 LP |
1569 | |
1570 | manager_dispatch_cleanup_queue(m); | |
1571 | ||
1572 | assert(!m->load_queue); | |
da8e1782 | 1573 | assert(prioq_isempty(m->run_queue)); |
964e0949 LP |
1574 | assert(!m->dbus_unit_queue); |
1575 | assert(!m->dbus_job_queue); | |
1576 | assert(!m->cleanup_queue); | |
c5a97ed1 LP |
1577 | assert(!m->gc_unit_queue); |
1578 | assert(!m->gc_job_queue); | |
13e72103 MK |
1579 | assert(!m->cgroup_realize_queue); |
1580 | assert(!m->cgroup_empty_queue); | |
1581 | assert(!m->cgroup_oom_queue); | |
1582 | assert(!m->target_deps_queue); | |
a3c1168a | 1583 | assert(!m->stop_when_unneeded_queue); |
0bc488c9 | 1584 | assert(!m->start_when_upheld_queue); |
56c59592 | 1585 | assert(!m->stop_when_bound_queue); |
cb3c6aec | 1586 | assert(!m->release_resources_queue); |
964e0949 | 1587 | |
964e0949 LP |
1588 | assert(hashmap_isempty(m->jobs)); |
1589 | assert(hashmap_isempty(m->units)); | |
9e9e2b72 MS |
1590 | |
1591 | m->n_on_console = 0; | |
1592 | m->n_running_jobs = 0; | |
1e75824c MK |
1593 | m->n_installed_jobs = 0; |
1594 | m->n_failed_jobs = 0; | |
a16e1123 LP |
1595 | } |
1596 | ||
06d8d842 | 1597 | Manager* manager_free(Manager *m) { |
06d8d842 ZJS |
1598 | if (!m) |
1599 | return NULL; | |
a16e1123 LP |
1600 | |
1601 | manager_clear_jobs_and_units(m); | |
23a177ef | 1602 | |
24b45970 | 1603 | for (UnitType c = 0; c < _UNIT_TYPE_MAX; c++) |
7824bbeb LP |
1604 | if (unit_vtable[c]->shutdown) |
1605 | unit_vtable[c]->shutdown(m); | |
1606 | ||
86036b26 | 1607 | /* Keep the cgroup hierarchy in place except when we know we are going down for good */ |
13ffc607 | 1608 | manager_shutdown_cgroup(m, /* delete= */ IN_SET(m->objective, MANAGER_EXIT, MANAGER_REBOOT, MANAGER_POWEROFF, MANAGER_HALT, MANAGER_KEXEC)); |
8e274523 | 1609 | |
07a78643 | 1610 | lookup_paths_flush_generator(&m->lookup_paths); |
5a1e9937 | 1611 | |
5e8d1c9a | 1612 | bus_done(m); |
19d22d43 | 1613 | manager_varlink_done(m); |
ea430986 | 1614 | |
e76506b7 DDM |
1615 | exec_shared_runtime_vacuum(m); |
1616 | hashmap_free(m->exec_shared_runtime_by_id); | |
e8a565cb | 1617 | |
29206d46 LP |
1618 | dynamic_user_vacuum(m, false); |
1619 | hashmap_free(m->dynamic_users); | |
1620 | ||
87f0e418 | 1621 | hashmap_free(m->units); |
4b58153d | 1622 | hashmap_free(m->units_by_invocation_id); |
60918275 | 1623 | hashmap_free(m->jobs); |
62a76913 | 1624 | hashmap_free(m->watch_pids); |
495e75ed | 1625 | hashmap_free(m->watch_pids_more); |
05e343b7 | 1626 | hashmap_free(m->watch_bus); |
9152c765 | 1627 | |
da8e1782 MO |
1628 | prioq_free(m->run_queue); |
1629 | ||
95ae05c0 | 1630 | set_free(m->startup_units); |
f755e3b7 LP |
1631 | set_free(m->failed_units); |
1632 | ||
718db961 | 1633 | sd_event_source_unref(m->signal_event_source); |
575b300b | 1634 | sd_event_source_unref(m->sigchld_event_source); |
718db961 | 1635 | sd_event_source_unref(m->notify_event_source); |
d8fdc620 | 1636 | sd_event_source_unref(m->cgroups_agent_event_source); |
718db961 | 1637 | sd_event_source_unref(m->time_change_event_source); |
bbf5fd8e | 1638 | sd_event_source_unref(m->timezone_change_event_source); |
718db961 | 1639 | sd_event_source_unref(m->jobs_in_progress_event_source); |
752b5905 | 1640 | sd_event_source_unref(m->run_queue_event_source); |
00d9ef85 | 1641 | sd_event_source_unref(m->user_lookup_event_source); |
29e6b0c1 | 1642 | sd_event_source_unref(m->memory_pressure_event_source); |
718db961 | 1643 | |
03e334a1 LP |
1644 | safe_close(m->signal_fd); |
1645 | safe_close(m->notify_fd); | |
d8fdc620 | 1646 | safe_close(m->cgroups_agent_fd); |
00d9ef85 | 1647 | safe_close_pair(m->user_lookup_fds); |
718db961 | 1648 | |
e46b13c8 ZJS |
1649 | manager_close_ask_password(m); |
1650 | ||
718db961 LP |
1651 | manager_close_idle_pipe(m); |
1652 | ||
1653 | sd_event_unref(m->event); | |
60918275 | 1654 | |
c952c6ec LP |
1655 | free(m->notify_socket); |
1656 | ||
84e3543e | 1657 | lookup_paths_free(&m->lookup_paths); |
1ad6e8b3 LP |
1658 | strv_free(m->transient_environment); |
1659 | strv_free(m->client_environment); | |
036643a2 | 1660 | |
4ad49000 | 1661 | hashmap_free(m->cgroup_unit); |
e8630e69 | 1662 | manager_free_unit_name_maps(m); |
33be102a | 1663 | |
664f88a7 LP |
1664 | free(m->switch_root); |
1665 | free(m->switch_root_init); | |
1666 | ||
c9e120e0 | 1667 | unit_defaults_done(&m->defaults); |
c93ff2e9 | 1668 | |
a57f7e2c LP |
1669 | assert(hashmap_isempty(m->units_requiring_mounts_for)); |
1670 | hashmap_free(m->units_requiring_mounts_for); | |
1671 | ||
00d9ef85 LP |
1672 | hashmap_free(m->uid_refs); |
1673 | hashmap_free(m->gid_refs); | |
1674 | ||
24b45970 | 1675 | for (ExecDirectoryType dt = 0; dt < _EXEC_DIRECTORY_TYPE_MAX; dt++) |
35aba85a | 1676 | m->prefix[dt] = mfree(m->prefix[dt]); |
2ad591a3 LP |
1677 | free(m->received_credentials_directory); |
1678 | free(m->received_encrypted_credentials_directory); | |
35aba85a | 1679 | |
aff3a9e1 LB |
1680 | free(m->watchdog_pretimeout_governor); |
1681 | free(m->watchdog_pretimeout_governor_overridden); | |
1682 | ||
dc7d69b3 TM |
1683 | m->fw_ctx = fw_ctx_free(m->fw_ctx); |
1684 | ||
b1994387 ILG |
1685 | #if BPF_FRAMEWORK |
1686 | lsm_bpf_destroy(m->restrict_fs); | |
1687 | #endif | |
1688 | ||
6b430fdb | 1689 | return mfree(m); |
60918275 LP |
1690 | } |
1691 | ||
04eb582a | 1692 | static void manager_enumerate_perpetual(Manager *m) { |
04eb582a LP |
1693 | assert(m); |
1694 | ||
1545051c | 1695 | if (FLAGS_SET(m->test_run_flags, MANAGER_TEST_RUN_MINIMAL)) |
b1d5246d ZJS |
1696 | return; |
1697 | ||
04eb582a | 1698 | /* Let's ask every type to load all units from disk/kernel that it might know */ |
24b45970 | 1699 | for (UnitType c = 0; c < _UNIT_TYPE_MAX; c++) { |
04eb582a LP |
1700 | if (!unit_type_supported(c)) { |
1701 | log_debug("Unit type .%s is not supported on this system.", unit_type_to_string(c)); | |
1702 | continue; | |
1703 | } | |
1704 | ||
1705 | if (unit_vtable[c]->enumerate_perpetual) | |
1706 | unit_vtable[c]->enumerate_perpetual(m); | |
1707 | } | |
1708 | } | |
1709 | ||
a1113e08 | 1710 | static void manager_enumerate(Manager *m) { |
f50e0a01 LP |
1711 | assert(m); |
1712 | ||
1545051c | 1713 | if (FLAGS_SET(m->test_run_flags, MANAGER_TEST_RUN_MINIMAL)) |
b1d5246d ZJS |
1714 | return; |
1715 | ||
04eb582a | 1716 | /* Let's ask every type to load all units from disk/kernel that it might know */ |
24b45970 | 1717 | for (UnitType c = 0; c < _UNIT_TYPE_MAX; c++) { |
1c2e9646 | 1718 | if (!unit_type_supported(c)) { |
03afec3c | 1719 | log_debug("Unit type .%s is not supported on this system.", unit_type_to_string(c)); |
0faacd47 | 1720 | continue; |
a57f7e2c | 1721 | } |
f50e0a01 | 1722 | |
94b01dae ZJS |
1723 | if (unit_vtable[c]->enumerate) |
1724 | unit_vtable[c]->enumerate(m); | |
0faacd47 LP |
1725 | } |
1726 | ||
f50e0a01 | 1727 | manager_dispatch_load_queue(m); |
a16e1123 LP |
1728 | } |
1729 | ||
007c6337 | 1730 | static void manager_coldplug(Manager *m) { |
a16e1123 LP |
1731 | Unit *u; |
1732 | char *k; | |
007c6337 | 1733 | int r; |
a16e1123 LP |
1734 | |
1735 | assert(m); | |
f50e0a01 | 1736 | |
28e5e1e9 | 1737 | log_debug("Invoking unit coldplug() handlers%s", special_glyph(SPECIAL_GLYPH_ELLIPSIS)); |
f0831ed2 LP |
1738 | |
1739 | /* Let's place the units back into their deserialized state */ | |
90e74a66 | 1740 | HASHMAP_FOREACH_KEY(u, k, m->units) { |
f50e0a01 LP |
1741 | |
1742 | /* ignore aliases */ | |
ac155bb8 | 1743 | if (u->id != k) |
f50e0a01 LP |
1744 | continue; |
1745 | ||
007c6337 LP |
1746 | r = unit_coldplug(u); |
1747 | if (r < 0) | |
1748 | log_warning_errno(r, "We couldn't coldplug %s, proceeding anyway: %m", u->id); | |
f50e0a01 | 1749 | } |
a16e1123 LP |
1750 | } |
1751 | ||
f0831ed2 | 1752 | static void manager_catchup(Manager *m) { |
f0831ed2 LP |
1753 | Unit *u; |
1754 | char *k; | |
1755 | ||
1756 | assert(m); | |
1757 | ||
28e5e1e9 | 1758 | log_debug("Invoking unit catchup() handlers%s", special_glyph(SPECIAL_GLYPH_ELLIPSIS)); |
f0831ed2 LP |
1759 | |
1760 | /* Let's catch up on any state changes that happened while we were reloading/reexecing */ | |
90e74a66 | 1761 | HASHMAP_FOREACH_KEY(u, k, m->units) { |
f0831ed2 LP |
1762 | |
1763 | /* ignore aliases */ | |
1764 | if (u->id != k) | |
1765 | continue; | |
1766 | ||
1767 | unit_catchup(u); | |
1768 | } | |
1769 | } | |
1770 | ||
9ff1a6f1 | 1771 | static void manager_distribute_fds(Manager *m, FDSet *fds) { |
9ff1a6f1 | 1772 | Unit *u; |
9588bc32 LP |
1773 | |
1774 | assert(m); | |
1775 | ||
90e74a66 | 1776 | HASHMAP_FOREACH(u, m->units) { |
9588bc32 LP |
1777 | |
1778 | if (fdset_size(fds) <= 0) | |
1779 | break; | |
1780 | ||
9ff1a6f1 LP |
1781 | if (!UNIT_VTABLE(u)->distribute_fds) |
1782 | continue; | |
9588bc32 | 1783 | |
9ff1a6f1 LP |
1784 | UNIT_VTABLE(u)->distribute_fds(u, fds); |
1785 | } | |
9588bc32 LP |
1786 | } |
1787 | ||
2b680534 | 1788 | static bool manager_dbus_is_running(Manager *m, bool deserialized) { |
8559b3b7 LP |
1789 | Unit *u; |
1790 | ||
1791 | assert(m); | |
1792 | ||
1793 | /* This checks whether the dbus instance we are supposed to expose our APIs on is up. We check both the socket | |
1794 | * and the service unit. If the 'deserialized' parameter is true we'll check the deserialized state of the unit | |
1795 | * rather than the current one. */ | |
1796 | ||
638cece4 | 1797 | if (MANAGER_IS_TEST_RUN(m)) |
8559b3b7 LP |
1798 | return false; |
1799 | ||
8559b3b7 LP |
1800 | u = manager_get_unit(m, SPECIAL_DBUS_SOCKET); |
1801 | if (!u) | |
1802 | return false; | |
1803 | if ((deserialized ? SOCKET(u)->deserialized_state : SOCKET(u)->state) != SOCKET_RUNNING) | |
1804 | return false; | |
1805 | ||
1806 | u = manager_get_unit(m, SPECIAL_DBUS_SERVICE); | |
1807 | if (!u) | |
1808 | return false; | |
845824ac | 1809 | if (!IN_SET((deserialized ? SERVICE(u)->deserialized_state : SERVICE(u)->state), |
1810 | SERVICE_RUNNING, | |
1811 | SERVICE_RELOAD, | |
1812 | SERVICE_RELOAD_NOTIFY, | |
1813 | SERVICE_RELOAD_SIGNAL)) | |
8559b3b7 LP |
1814 | return false; |
1815 | ||
1816 | return true; | |
1817 | } | |
1818 | ||
9d4c195c LP |
1819 | static void manager_setup_bus(Manager *m) { |
1820 | assert(m); | |
1821 | ||
1822 | /* Let's set up our private bus connection now, unconditionally */ | |
1823 | (void) bus_init_private(m); | |
1824 | ||
1825 | /* If we are in --user mode also connect to the system bus now */ | |
1826 | if (MANAGER_IS_USER(m)) | |
1827 | (void) bus_init_system(m); | |
1828 | ||
1829 | /* Let's connect to the bus now, but only if the unit is supposed to be up */ | |
2b680534 | 1830 | if (manager_dbus_is_running(m, MANAGER_IS_RELOADING(m))) { |
9d4c195c LP |
1831 | (void) bus_init_api(m); |
1832 | ||
1833 | if (MANAGER_IS_SYSTEM(m)) | |
1834 | (void) bus_init_system(m); | |
1835 | } | |
1836 | } | |
1837 | ||
159f1e76 LP |
1838 | static void manager_preset_all(Manager *m) { |
1839 | int r; | |
1840 | ||
1841 | assert(m); | |
1842 | ||
1843 | if (m->first_boot <= 0) | |
1844 | return; | |
1845 | ||
1846 | if (!MANAGER_IS_SYSTEM(m)) | |
1847 | return; | |
1848 | ||
638cece4 | 1849 | if (MANAGER_IS_TEST_RUN(m)) |
159f1e76 LP |
1850 | return; |
1851 | ||
1852 | /* If this is the first boot, and we are in the host system, then preset everything */ | |
93651582 JL |
1853 | UnitFilePresetMode mode = FIRST_BOOT_FULL_PRESET ? UNIT_FILE_PRESET_FULL : UNIT_FILE_PRESET_ENABLE_ONLY; |
1854 | ||
4870133b | 1855 | r = unit_file_preset_all(RUNTIME_SCOPE_SYSTEM, 0, NULL, mode, NULL, 0); |
159f1e76 LP |
1856 | if (r < 0) |
1857 | log_full_errno(r == -EEXIST ? LOG_NOTICE : LOG_WARNING, r, | |
1858 | "Failed to populate /etc with preset unit settings, ignoring: %m"); | |
1859 | else | |
1860 | log_info("Populated /etc with preset unit settings."); | |
1861 | } | |
1862 | ||
5ce5e1ad LP |
1863 | static void manager_ready(Manager *m) { |
1864 | assert(m); | |
1865 | ||
1866 | /* After having loaded everything, do the final round of catching up with what might have changed */ | |
1867 | ||
1868 | m->objective = MANAGER_OK; /* Tell everyone we are up now */ | |
1869 | ||
1870 | /* It might be safe to log to the journal now and connect to dbus */ | |
1871 | manager_recheck_journal(m); | |
1872 | manager_recheck_dbus(m); | |
1873 | ||
5ce5e1ad LP |
1874 | /* Let's finally catch up with any changes that took place while we were reloading/reexecing */ |
1875 | manager_catchup(m); | |
c6e892bc | 1876 | |
15b9243c | 1877 | /* Create a file which will indicate when the manager started loading units the last time. */ |
4b3ad81b LB |
1878 | if (MANAGER_IS_SYSTEM(m)) |
1879 | (void) touch_file("/run/systemd/systemd-units-load", false, | |
1880 | m->timestamps[MANAGER_TIMESTAMP_UNITS_LOAD].realtime ?: now(CLOCK_REALTIME), | |
1881 | UID_INVALID, GID_INVALID, 0444); | |
5ce5e1ad LP |
1882 | } |
1883 | ||
a01ba4b2 | 1884 | Manager* manager_reloading_start(Manager *m) { |
d147e2b6 | 1885 | m->n_reloading++; |
49fbe940 | 1886 | dual_timestamp_get(m->timestamps + MANAGER_TIMESTAMP_UNITS_LOAD); |
d147e2b6 ZJS |
1887 | return m; |
1888 | } | |
49fbe940 | 1889 | |
a01ba4b2 | 1890 | void manager_reloading_stopp(Manager **m) { |
d147e2b6 ZJS |
1891 | if (*m) { |
1892 | assert((*m)->n_reloading > 0); | |
1893 | (*m)->n_reloading--; | |
1894 | } | |
1895 | } | |
1896 | ||
2a7cf953 | 1897 | int manager_startup(Manager *m, FILE *serialization, FDSet *fds, const char *root) { |
17f01ace | 1898 | int r; |
a16e1123 LP |
1899 | |
1900 | assert(m); | |
1901 | ||
a1f31f47 ZJS |
1902 | /* If we are running in test mode, we still want to run the generators, |
1903 | * but we should not touch the real generator directories. */ | |
4870133b | 1904 | r = lookup_paths_init_or_warn(&m->lookup_paths, m->runtime_scope, |
99aad9a2 ZJS |
1905 | MANAGER_IS_TEST_RUN(m) ? LOOKUP_PATHS_TEMPORARY_GENERATED : 0, |
1906 | root); | |
e801700e | 1907 | if (r < 0) |
99aad9a2 | 1908 | return r; |
5a1e9937 | 1909 | |
d4ee7bd8 | 1910 | dual_timestamp_get(m->timestamps + manager_timestamp_initrd_mangle(MANAGER_TIMESTAMP_GENERATORS_START)); |
eb523bfb LP |
1911 | r = manager_run_environment_generators(m); |
1912 | if (r >= 0) | |
1913 | r = manager_run_generators(m); | |
d4ee7bd8 | 1914 | dual_timestamp_get(m->timestamps + manager_timestamp_initrd_mangle(MANAGER_TIMESTAMP_GENERATORS_FINISH)); |
07719a21 LP |
1915 | if (r < 0) |
1916 | return r; | |
1917 | ||
159f1e76 | 1918 | manager_preset_all(m); |
572986ca | 1919 | |
581fef8d | 1920 | lookup_paths_log(&m->lookup_paths); |
572986ca | 1921 | |
4df7d537 ZJS |
1922 | { |
1923 | /* This block is (optionally) done with the reloading counter bumped */ | |
d7ac0952 | 1924 | _unused_ _cleanup_(manager_reloading_stopp) Manager *reloading = NULL; |
4df7d537 | 1925 | |
45f540a2 ZJS |
1926 | /* Make sure we don't have a left-over from a previous run */ |
1927 | if (!serialization) | |
1928 | (void) rm_rf(m->lookup_paths.transient, 0); | |
1929 | ||
4df7d537 ZJS |
1930 | /* If we will deserialize make sure that during enumeration this is already known, so we increase the |
1931 | * counter here already */ | |
1932 | if (serialization) | |
1933 | reloading = manager_reloading_start(m); | |
1934 | ||
1935 | /* First, enumerate what we can from all config files */ | |
1936 | dual_timestamp_get(m->timestamps + manager_timestamp_initrd_mangle(MANAGER_TIMESTAMP_UNITS_LOAD_START)); | |
1937 | manager_enumerate_perpetual(m); | |
1938 | manager_enumerate(m); | |
1939 | dual_timestamp_get(m->timestamps + manager_timestamp_initrd_mangle(MANAGER_TIMESTAMP_UNITS_LOAD_FINISH)); | |
1940 | ||
1941 | /* Second, deserialize if there is something to deserialize */ | |
1942 | if (serialization) { | |
1943 | r = manager_deserialize(m, serialization, fds); | |
1944 | if (r < 0) | |
1945 | return log_error_errno(r, "Deserialization failed: %m"); | |
1946 | } | |
01e10de3 | 1947 | |
4df7d537 ZJS |
1948 | /* Any fds left? Find some unit which wants them. This is useful to allow container managers to pass |
1949 | * some file descriptors to us pre-initialized. This enables socket-based activation of entire | |
1950 | * containers. */ | |
1951 | manager_distribute_fds(m, fds); | |
d86f9d52 | 1952 | |
4df7d537 ZJS |
1953 | /* We might have deserialized the notify fd, but if we didn't then let's create the bus now */ |
1954 | r = manager_setup_notify(m); | |
1955 | if (r < 0) | |
1956 | /* No sense to continue without notifications, our children would fail anyway. */ | |
1957 | return r; | |
d8fdc620 | 1958 | |
4df7d537 ZJS |
1959 | r = manager_setup_cgroups_agent(m); |
1960 | if (r < 0) | |
1961 | /* Likewise, no sense to continue without empty cgroup notifications. */ | |
1962 | return r; | |
00d9ef85 | 1963 | |
4df7d537 ZJS |
1964 | r = manager_setup_user_lookup_fd(m); |
1965 | if (r < 0) | |
1966 | /* This shouldn't fail, except if things are really broken. */ | |
1967 | return r; | |
8559b3b7 | 1968 | |
4df7d537 ZJS |
1969 | /* Connect to the bus if we are good for it */ |
1970 | manager_setup_bus(m); | |
e3dd987c | 1971 | |
5238e957 | 1972 | /* Now that we are connected to all possible buses, let's deserialize who is tracking us. */ |
4df7d537 ZJS |
1973 | r = bus_track_coldplug(m, &m->subscribed, false, m->deserialized_subscribed); |
1974 | if (r < 0) | |
1975 | log_warning_errno(r, "Failed to deserialized tracked clients, ignoring: %m"); | |
1976 | m->deserialized_subscribed = strv_free(m->deserialized_subscribed); | |
a16e1123 | 1977 | |
19d22d43 LP |
1978 | r = manager_varlink_init(m); |
1979 | if (r < 0) | |
064a5c14 | 1980 | log_warning_errno(r, "Failed to set up Varlink, ignoring: %m"); |
19d22d43 | 1981 | |
4df7d537 ZJS |
1982 | /* Third, fire things up! */ |
1983 | manager_coldplug(m); | |
3ad2afb6 | 1984 | |
4df7d537 ZJS |
1985 | /* Clean up runtime objects */ |
1986 | manager_vacuum(m); | |
71445ae7 | 1987 | |
4df7d537 ZJS |
1988 | if (serialization) |
1989 | /* Let's wait for the UnitNew/JobNew messages being sent, before we notify that the | |
1990 | * reload is finished */ | |
1991 | m->send_reloading_done = true; | |
9f611ad8 LP |
1992 | } |
1993 | ||
5ce5e1ad | 1994 | manager_ready(m); |
f0831ed2 | 1995 | |
d35fe8c0 FB |
1996 | manager_set_switching_root(m, false); |
1997 | ||
17f01ace | 1998 | return 0; |
f50e0a01 LP |
1999 | } |
2000 | ||
50cbaba4 LP |
2001 | int manager_add_job( |
2002 | Manager *m, | |
2003 | JobType type, | |
2004 | Unit *unit, | |
2005 | JobMode mode, | |
2006 | Set *affected_jobs, | |
2007 | sd_bus_error *error, | |
2008 | Job **ret) { | |
2009 | ||
c7e3eb5d | 2010 | _cleanup_(transaction_abort_and_freep) Transaction *tr = NULL; |
50cbaba4 | 2011 | int r; |
e5b5ae50 LP |
2012 | |
2013 | assert(m); | |
2014 | assert(type < _JOB_TYPE_MAX); | |
87f0e418 | 2015 | assert(unit); |
e5b5ae50 | 2016 | assert(mode < _JOB_MODE_MAX); |
60918275 | 2017 | |
7358dc02 | 2018 | if (mode == JOB_ISOLATE && type != JOB_START) |
1b09b81c | 2019 | return sd_bus_error_set(error, SD_BUS_ERROR_INVALID_ARGS, "Isolate is only valid for start."); |
c497c7a9 | 2020 | |
7358dc02 | 2021 | if (mode == JOB_ISOLATE && !unit->allow_isolate) |
1b09b81c | 2022 | return sd_bus_error_set(error, BUS_ERROR_NO_ISOLATION, "Operation refused, unit may not be isolated."); |
2528a7a6 | 2023 | |
1f0f9f21 | 2024 | if (mode == JOB_TRIGGERING && type != JOB_STOP) |
1b09b81c | 2025 | return sd_bus_error_set(error, SD_BUS_ERROR_INVALID_ARGS, "--job-mode=triggering is only valid for stop."); |
1f0f9f21 | 2026 | |
09d04ad3 LP |
2027 | if (mode == JOB_RESTART_DEPENDENCIES && type != JOB_START) |
2028 | return sd_bus_error_set(error, SD_BUS_ERROR_INVALID_ARGS, "--job-mode=restart-dependencies is only valid for start."); | |
2029 | ||
f2341e0a | 2030 | log_unit_debug(unit, "Trying to enqueue job %s/%s/%s", unit->id, job_type_to_string(type), job_mode_to_string(mode)); |
9f04bd52 | 2031 | |
c6497ccb | 2032 | type = job_type_collapse(type, unit); |
e0209d83 | 2033 | |
23ade460 | 2034 | tr = transaction_new(mode == JOB_REPLACE_IRREVERSIBLY); |
7527cb52 MS |
2035 | if (!tr) |
2036 | return -ENOMEM; | |
11dd41ce | 2037 | |
b0904249 LP |
2038 | r = transaction_add_job_and_dependencies( |
2039 | tr, | |
2040 | type, | |
2041 | unit, | |
2042 | /* by= */ NULL, | |
2043 | TRANSACTION_MATTERS | | |
2044 | (IN_SET(mode, JOB_IGNORE_DEPENDENCIES, JOB_IGNORE_REQUIREMENTS) ? TRANSACTION_IGNORE_REQUIREMENTS : 0) | | |
09d04ad3 LP |
2045 | (mode == JOB_IGNORE_DEPENDENCIES ? TRANSACTION_IGNORE_ORDER : 0) | |
2046 | (mode == JOB_RESTART_DEPENDENCIES ? TRANSACTION_PROPAGATE_START_AS_RESTART : 0), | |
b0904249 | 2047 | error); |
7527cb52 | 2048 | if (r < 0) |
c7e3eb5d | 2049 | return r; |
c497c7a9 | 2050 | |
7527cb52 MS |
2051 | if (mode == JOB_ISOLATE) { |
2052 | r = transaction_add_isolate_jobs(tr, m); | |
2053 | if (r < 0) | |
c7e3eb5d | 2054 | return r; |
7527cb52 MS |
2055 | } |
2056 | ||
1f0f9f21 KK |
2057 | if (mode == JOB_TRIGGERING) { |
2058 | r = transaction_add_triggering_jobs(tr, unit); | |
2059 | if (r < 0) | |
c7e3eb5d | 2060 | return r; |
1f0f9f21 KK |
2061 | } |
2062 | ||
50cbaba4 | 2063 | r = transaction_activate(tr, m, mode, affected_jobs, error); |
7527cb52 | 2064 | if (r < 0) |
c7e3eb5d | 2065 | return r; |
e5b5ae50 | 2066 | |
f2341e0a | 2067 | log_unit_debug(unit, |
66870f90 ZJS |
2068 | "Enqueued job %s/%s as %u", unit->id, |
2069 | job_type_to_string(type), (unsigned) tr->anchor_job->id); | |
f50e0a01 | 2070 | |
50cbaba4 LP |
2071 | if (ret) |
2072 | *ret = tr->anchor_job; | |
60918275 | 2073 | |
c7e3eb5d | 2074 | tr = transaction_free(tr); |
e5b5ae50 LP |
2075 | return 0; |
2076 | } | |
60918275 | 2077 | |
50cbaba4 | 2078 | int manager_add_job_by_name(Manager *m, JobType type, const char *name, JobMode mode, Set *affected_jobs, sd_bus_error *e, Job **ret) { |
4440b27d | 2079 | Unit *unit = NULL; /* just to appease gcc, initialization is not really necessary */ |
28247076 LP |
2080 | int r; |
2081 | ||
2082 | assert(m); | |
2083 | assert(type < _JOB_TYPE_MAX); | |
2084 | assert(name); | |
2085 | assert(mode < _JOB_MODE_MAX); | |
2086 | ||
c3090674 LP |
2087 | r = manager_load_unit(m, name, NULL, NULL, &unit); |
2088 | if (r < 0) | |
28247076 | 2089 | return r; |
4440b27d | 2090 | assert(unit); |
28247076 | 2091 | |
50cbaba4 | 2092 | return manager_add_job(m, type, unit, mode, affected_jobs, e, ret); |
53f18416 LP |
2093 | } |
2094 | ||
50cbaba4 | 2095 | int manager_add_job_by_name_and_warn(Manager *m, JobType type, const char *name, JobMode mode, Set *affected_jobs, Job **ret) { |
4afd3348 | 2096 | _cleanup_(sd_bus_error_free) sd_bus_error error = SD_BUS_ERROR_NULL; |
53f18416 LP |
2097 | int r; |
2098 | ||
2099 | assert(m); | |
2100 | assert(type < _JOB_TYPE_MAX); | |
2101 | assert(name); | |
2102 | assert(mode < _JOB_MODE_MAX); | |
2103 | ||
50cbaba4 | 2104 | r = manager_add_job_by_name(m, type, name, mode, affected_jobs, &error, ret); |
53f18416 LP |
2105 | if (r < 0) |
2106 | return log_warning_errno(r, "Failed to enqueue %s job for %s: %s", job_mode_to_string(mode), name, bus_error_message(&error, r)); | |
2107 | ||
2108 | return r; | |
28247076 LP |
2109 | } |
2110 | ||
15d167f8 JW |
2111 | int manager_propagate_reload(Manager *m, Unit *unit, JobMode mode, sd_bus_error *e) { |
2112 | int r; | |
32620826 | 2113 | _cleanup_(transaction_abort_and_freep) Transaction *tr = NULL; |
15d167f8 JW |
2114 | |
2115 | assert(m); | |
2116 | assert(unit); | |
2117 | assert(mode < _JOB_MODE_MAX); | |
2118 | assert(mode != JOB_ISOLATE); /* Isolate is only valid for start */ | |
2119 | ||
2120 | tr = transaction_new(mode == JOB_REPLACE_IRREVERSIBLY); | |
2121 | if (!tr) | |
2122 | return -ENOMEM; | |
2123 | ||
2124 | /* We need an anchor job */ | |
b0904249 | 2125 | r = transaction_add_job_and_dependencies(tr, JOB_NOP, unit, NULL, TRANSACTION_IGNORE_REQUIREMENTS|TRANSACTION_IGNORE_ORDER, e); |
15d167f8 | 2126 | if (r < 0) |
32620826 | 2127 | return r; |
15d167f8 JW |
2128 | |
2129 | /* Failure in adding individual dependencies is ignored, so this always succeeds. */ | |
b0904249 LP |
2130 | transaction_add_propagate_reload_jobs( |
2131 | tr, | |
2132 | unit, | |
2133 | tr->anchor_job, | |
3044510d | 2134 | mode == JOB_IGNORE_DEPENDENCIES ? TRANSACTION_IGNORE_ORDER : 0); |
15d167f8 | 2135 | |
50cbaba4 | 2136 | r = transaction_activate(tr, m, mode, NULL, e); |
15d167f8 | 2137 | if (r < 0) |
32620826 | 2138 | return r; |
15d167f8 | 2139 | |
32620826 | 2140 | tr = transaction_free(tr); |
15d167f8 | 2141 | return 0; |
15d167f8 JW |
2142 | } |
2143 | ||
60918275 LP |
2144 | Job *manager_get_job(Manager *m, uint32_t id) { |
2145 | assert(m); | |
2146 | ||
2147 | return hashmap_get(m->jobs, UINT32_TO_PTR(id)); | |
2148 | } | |
2149 | ||
87f0e418 | 2150 | Unit *manager_get_unit(Manager *m, const char *name) { |
60918275 LP |
2151 | assert(m); |
2152 | assert(name); | |
2153 | ||
87f0e418 | 2154 | return hashmap_get(m->units, name); |
60918275 LP |
2155 | } |
2156 | ||
19496554 MS |
2157 | static int manager_dispatch_target_deps_queue(Manager *m) { |
2158 | Unit *u; | |
19496554 MS |
2159 | int r = 0; |
2160 | ||
19496554 MS |
2161 | assert(m); |
2162 | ||
52e3671b | 2163 | while ((u = LIST_POP(target_deps_queue, m->target_deps_queue))) { |
15ed3c3a LP |
2164 | _cleanup_free_ Unit **targets = NULL; |
2165 | int n_targets; | |
2166 | ||
19496554 MS |
2167 | assert(u->in_target_deps_queue); |
2168 | ||
19496554 MS |
2169 | u->in_target_deps_queue = false; |
2170 | ||
15ed3c3a LP |
2171 | /* Take an "atomic" snapshot of dependencies here, as the call below will likely modify the |
2172 | * dependencies, and we can't have it that hash tables we iterate through are modified while | |
2173 | * we are iterating through them. */ | |
2174 | n_targets = unit_get_dependency_array(u, UNIT_ATOM_DEFAULT_TARGET_DEPENDENCIES, &targets); | |
2175 | if (n_targets < 0) | |
2176 | return n_targets; | |
19496554 | 2177 | |
15ed3c3a LP |
2178 | for (int i = 0; i < n_targets; i++) { |
2179 | r = unit_add_default_target_dependency(u, targets[i]); | |
2180 | if (r < 0) | |
2181 | return r; | |
19496554 MS |
2182 | } |
2183 | } | |
2184 | ||
2185 | return r; | |
2186 | } | |
2187 | ||
c1e1601e | 2188 | unsigned manager_dispatch_load_queue(Manager *m) { |
595ed347 | 2189 | Unit *u; |
c1e1601e | 2190 | unsigned n = 0; |
60918275 LP |
2191 | |
2192 | assert(m); | |
2193 | ||
223dabab LP |
2194 | /* Make sure we are not run recursively */ |
2195 | if (m->dispatching_load_queue) | |
c1e1601e | 2196 | return 0; |
223dabab LP |
2197 | |
2198 | m->dispatching_load_queue = true; | |
2199 | ||
87f0e418 | 2200 | /* Dispatches the load queue. Takes a unit from the queue and |
60918275 LP |
2201 | * tries to load its data until the queue is empty */ |
2202 | ||
595ed347 MS |
2203 | while ((u = m->load_queue)) { |
2204 | assert(u->in_load_queue); | |
034c6ed7 | 2205 | |
595ed347 | 2206 | unit_load(u); |
c1e1601e | 2207 | n++; |
60918275 LP |
2208 | } |
2209 | ||
223dabab | 2210 | m->dispatching_load_queue = false; |
19496554 MS |
2211 | |
2212 | /* Dispatch the units waiting for their target dependencies to be added now, as all targets that we know about | |
2213 | * should be loaded and have aliases resolved */ | |
2214 | (void) manager_dispatch_target_deps_queue(m); | |
2215 | ||
c1e1601e | 2216 | return n; |
60918275 LP |
2217 | } |
2218 | ||
81be2388 | 2219 | bool manager_unit_cache_should_retry_load(Unit *u) { |
cda66772 LB |
2220 | assert(u); |
2221 | ||
81be2388 ZJS |
2222 | /* Automatic reloading from disk only applies to units which were not found sometime in the past, and |
2223 | * the not-found stub is kept pinned in the unit graph by dependencies. For units that were | |
2224 | * previously loaded, we don't do automatic reloading, and daemon-reload is necessary to update. */ | |
cda66772 LB |
2225 | if (u->load_state != UNIT_NOT_FOUND) |
2226 | return false; | |
2227 | ||
81be2388 ZJS |
2228 | /* The cache has been updated since the last time we tried to load the unit. There might be new |
2229 | * fragment paths to read. */ | |
c2911d48 | 2230 | if (u->manager->unit_cache_timestamp_hash != u->fragment_not_found_timestamp_hash) |
cda66772 | 2231 | return true; |
d904afc7 | 2232 | |
81be2388 | 2233 | /* The cache needs to be updated because there are modifications on disk. */ |
c2911d48 | 2234 | return !lookup_paths_timestamp_hash_same(&u->manager->lookup_paths, u->manager->unit_cache_timestamp_hash, NULL); |
d904afc7 LB |
2235 | } |
2236 | ||
c2756a68 LP |
2237 | int manager_load_unit_prepare( |
2238 | Manager *m, | |
2239 | const char *name, | |
2240 | const char *path, | |
718db961 | 2241 | sd_bus_error *e, |
4b6a2b3f | 2242 | Unit **ret) { |
c2756a68 | 2243 | |
4b6a2b3f | 2244 | _cleanup_(unit_freep) Unit *cleanup_unit = NULL; |
a99626c1 | 2245 | _cleanup_free_ char *nbuf = NULL; |
60918275 LP |
2246 | int r; |
2247 | ||
2248 | assert(m); | |
4b6a2b3f | 2249 | assert(ret); |
a99626c1 | 2250 | assert(name || path); |
60918275 | 2251 | |
526e3cbb | 2252 | /* This will prepare the unit for loading, but not actually load anything from disk. */ |
0301abf4 | 2253 | |
526e3cbb | 2254 | if (path && !path_is_absolute(path)) |
718db961 | 2255 | return sd_bus_error_setf(e, SD_BUS_ERROR_INVALID_ARGS, "Path %s is not absolute.", path); |
9e2f7c11 | 2256 | |
08f46856 | 2257 | if (!name) { |
a99626c1 LP |
2258 | r = path_extract_filename(path, &nbuf); |
2259 | if (r < 0) | |
2260 | return r; | |
2261 | if (r == O_DIRECTORY) | |
2262 | return sd_bus_error_setf(e, SD_BUS_ERROR_INVALID_ARGS, "Path '%s' refers to directory, refusing.", path); | |
2263 | ||
2264 | name = nbuf; | |
08f46856 | 2265 | } |
9e2f7c11 | 2266 | |
4b6a2b3f | 2267 | UnitType t = unit_name_to_type(name); |
7d17cfbc | 2268 | |
5d512d54 LN |
2269 | if (t == _UNIT_TYPE_INVALID || !unit_name_is_valid(name, UNIT_NAME_PLAIN|UNIT_NAME_INSTANCE)) { |
2270 | if (unit_name_is_valid(name, UNIT_NAME_TEMPLATE)) | |
2271 | return sd_bus_error_setf(e, SD_BUS_ERROR_INVALID_ARGS, "Unit name %s is missing the instance name.", name); | |
2272 | ||
718db961 | 2273 | return sd_bus_error_setf(e, SD_BUS_ERROR_INVALID_ARGS, "Unit name %s is not valid.", name); |
5d512d54 | 2274 | } |
60918275 | 2275 | |
4b6a2b3f ZJS |
2276 | Unit *unit = manager_get_unit(m, name); |
2277 | if (unit) { | |
d904afc7 LB |
2278 | /* The time-based cache allows to start new units without daemon-reload, |
2279 | * but if they are already referenced (because of dependencies or ordering) | |
2280 | * then we have to force a load of the fragment. As an optimization, check | |
2281 | * first if anything in the usual paths was modified since the last time | |
7233e91a LB |
2282 | * the cache was loaded. Also check if the last time an attempt to load the |
2283 | * unit was made was before the most recent cache refresh, so that we know | |
81be2388 | 2284 | * we need to try again — even if the cache is current, it might have been |
7233e91a LB |
2285 | * updated in a different context before we had a chance to retry loading |
2286 | * this particular unit. */ | |
4b6a2b3f ZJS |
2287 | if (manager_unit_cache_should_retry_load(unit)) |
2288 | unit->load_state = UNIT_STUB; | |
d904afc7 | 2289 | else { |
4b6a2b3f | 2290 | *ret = unit; |
535b7fcb | 2291 | return 0; /* The unit was already loaded */ |
d904afc7 LB |
2292 | } |
2293 | } else { | |
4b6a2b3f ZJS |
2294 | unit = cleanup_unit = unit_new(m, unit_vtable[t]->object_size); |
2295 | if (!unit) | |
d904afc7 | 2296 | return -ENOMEM; |
034c6ed7 | 2297 | } |
60918275 | 2298 | |
7d17cfbc | 2299 | if (path) { |
4b6a2b3f | 2300 | r = free_and_strdup(&unit->fragment_path, path); |
d904afc7 LB |
2301 | if (r < 0) |
2302 | return r; | |
7d17cfbc | 2303 | } |
0301abf4 | 2304 | |
4b6a2b3f | 2305 | r = unit_add_name(unit, name); |
dc409696 | 2306 | if (r < 0) |
1ffba6fe | 2307 | return r; |
60918275 | 2308 | |
4b6a2b3f ZJS |
2309 | unit_add_to_load_queue(unit); |
2310 | unit_add_to_dbus_queue(unit); | |
2311 | unit_add_to_gc_queue(unit); | |
c1e1601e | 2312 | |
4b6a2b3f ZJS |
2313 | *ret = unit; |
2314 | TAKE_PTR(cleanup_unit); | |
db06e3b6 | 2315 | |
535b7fcb | 2316 | return 1; /* The unit was added the load queue */ |
db06e3b6 LP |
2317 | } |
2318 | ||
c2756a68 LP |
2319 | int manager_load_unit( |
2320 | Manager *m, | |
2321 | const char *name, | |
2322 | const char *path, | |
718db961 | 2323 | sd_bus_error *e, |
4b6a2b3f | 2324 | Unit **ret) { |
db06e3b6 LP |
2325 | int r; |
2326 | ||
2327 | assert(m); | |
4b6a2b3f | 2328 | assert(ret); |
db06e3b6 | 2329 | |
4b6a2b3f | 2330 | /* This will load the unit config, but not actually start any services or anything. */ |
db06e3b6 | 2331 | |
4b6a2b3f | 2332 | r = manager_load_unit_prepare(m, name, path, e, ret); |
535b7fcb | 2333 | if (r <= 0) |
db06e3b6 LP |
2334 | return r; |
2335 | ||
535b7fcb | 2336 | /* Unit was newly loaded */ |
f50e0a01 | 2337 | manager_dispatch_load_queue(m); |
4b6a2b3f | 2338 | *ret = unit_follow_merge(*ret); |
4109ede7 ZJS |
2339 | return 0; |
2340 | } | |
2341 | ||
2342 | int manager_load_startable_unit_or_warn( | |
2343 | Manager *m, | |
2344 | const char *name, | |
2345 | const char *path, | |
2346 | Unit **ret) { | |
2347 | ||
2348 | /* Load a unit, make sure it loaded fully and is not masked. */ | |
2349 | ||
2350 | _cleanup_(sd_bus_error_free) sd_bus_error error = SD_BUS_ERROR_NULL; | |
2351 | Unit *unit; | |
2352 | int r; | |
2353 | ||
2354 | r = manager_load_unit(m, name, path, &error, &unit); | |
2355 | if (r < 0) | |
2356 | return log_error_errno(r, "Failed to load %s %s: %s", | |
8ace1db7 | 2357 | name ? "unit" : "unit file", name ?: path, |
4109ede7 | 2358 | bus_error_message(&error, r)); |
8ace1db7 LP |
2359 | |
2360 | r = bus_unit_validate_load_state(unit, &error); | |
2361 | if (r < 0) | |
2362 | return log_error_errno(r, "%s", bus_error_message(&error, r)); | |
9e2f7c11 | 2363 | |
4109ede7 | 2364 | *ret = unit; |
60918275 LP |
2365 | return 0; |
2366 | } | |
a66d02c3 | 2367 | |
7fad411c LP |
2368 | void manager_clear_jobs(Manager *m) { |
2369 | Job *j; | |
2370 | ||
2371 | assert(m); | |
2372 | ||
7fad411c | 2373 | while ((j = hashmap_first(m->jobs))) |
5273510e | 2374 | /* No need to recurse. We're cancelling all jobs. */ |
833f92ad | 2375 | job_finish_and_invalidate(j, JOB_CANCELED, false, false); |
7fad411c | 2376 | } |
83c60c9f | 2377 | |
495e75ed | 2378 | void manager_unwatch_pidref(Manager *m, PidRef *pid) { |
f75f613d FB |
2379 | assert(m); |
2380 | ||
495e75ed LP |
2381 | for (;;) { |
2382 | Unit *u; | |
f75f613d | 2383 | |
495e75ed LP |
2384 | u = manager_get_unit_by_pidref_watching(m, pid); |
2385 | if (!u) | |
2386 | break; | |
2387 | ||
2388 | unit_unwatch_pidref(u, pid); | |
2389 | } | |
f75f613d FB |
2390 | } |
2391 | ||
752b5905 | 2392 | static int manager_dispatch_run_queue(sd_event_source *source, void *userdata) { |
99534007 | 2393 | Manager *m = ASSERT_PTR(userdata); |
83c60c9f | 2394 | Job *j; |
034c6ed7 | 2395 | |
752b5905 | 2396 | assert(source); |
9152c765 | 2397 | |
da8e1782 | 2398 | while ((j = prioq_peek(m->run_queue))) { |
ac1135be | 2399 | assert(j->installed); |
034c6ed7 LP |
2400 | assert(j->in_run_queue); |
2401 | ||
6e64994d | 2402 | (void) job_run_and_invalidate(j); |
9152c765 | 2403 | } |
034c6ed7 | 2404 | |
a0b64226 | 2405 | if (m->n_running_jobs > 0) |
03b717a3 MS |
2406 | manager_watch_jobs_in_progress(m); |
2407 | ||
31a7eb86 ZJS |
2408 | if (m->n_on_console > 0) |
2409 | manager_watch_idle_pipe(m); | |
2410 | ||
752b5905 | 2411 | return 1; |
c1e1601e LP |
2412 | } |
2413 | ||
b0c4b282 LP |
2414 | void manager_trigger_run_queue(Manager *m) { |
2415 | int r; | |
2416 | ||
2417 | assert(m); | |
2418 | ||
2419 | r = sd_event_source_set_enabled( | |
2420 | m->run_queue_event_source, | |
2421 | prioq_isempty(m->run_queue) ? SD_EVENT_OFF : SD_EVENT_ONESHOT); | |
2422 | if (r < 0) | |
2423 | log_warning_errno(r, "Failed to enable job run queue event source, ignoring: %m"); | |
2424 | } | |
2425 | ||
9588bc32 | 2426 | static unsigned manager_dispatch_dbus_queue(Manager *m) { |
e0a08581 | 2427 | unsigned n = 0, budget; |
595ed347 | 2428 | Unit *u; |
e0a08581 | 2429 | Job *j; |
c1e1601e LP |
2430 | |
2431 | assert(m); | |
2432 | ||
b8d381c4 LP |
2433 | /* When we are reloading, let's not wait with generating signals, since we need to exit the manager as quickly |
2434 | * as we can. There's no point in throttling generation of signals in that case. */ | |
2435 | if (MANAGER_IS_RELOADING(m) || m->send_reloading_done || m->pending_reload_message) | |
f5fbe71d | 2436 | budget = UINT_MAX; /* infinite budget in this case */ |
b8d381c4 LP |
2437 | else { |
2438 | /* Anything to do at all? */ | |
2439 | if (!m->dbus_unit_queue && !m->dbus_job_queue) | |
2440 | return 0; | |
2441 | ||
2442 | /* Do we have overly many messages queued at the moment? If so, let's not enqueue more on top, let's | |
2443 | * sit this cycle out, and process things in a later cycle when the queues got a bit emptier. */ | |
2444 | if (manager_bus_n_queued_write(m) > MANAGER_BUS_BUSY_THRESHOLD) | |
2445 | return 0; | |
2446 | ||
2447 | /* Only process a certain number of units/jobs per event loop iteration. Even if the bus queue wasn't | |
2448 | * overly full before this call we shouldn't increase it in size too wildly in one step, and we | |
2449 | * shouldn't monopolize CPU time with generating these messages. Note the difference in counting of | |
2450 | * this "budget" and the "threshold" above: the "budget" is decreased only once per generated message, | |
5238e957 BB |
2451 | * regardless how many buses/direct connections it is enqueued on, while the "threshold" is applied to |
2452 | * each queued instance of bus message, i.e. if the same message is enqueued to five buses/direct | |
b8d381c4 LP |
2453 | * connections it will be counted five times. This difference in counting ("references" |
2454 | * vs. "instances") is primarily a result of the fact that it's easier to implement it this way, | |
2455 | * however it also reflects the thinking that the "threshold" should put a limit on used queue memory, | |
2456 | * i.e. space, while the "budget" should put a limit on time. Also note that the "threshold" is | |
2457 | * currently chosen much higher than the "budget". */ | |
2458 | budget = MANAGER_BUS_MESSAGE_BUDGET; | |
2459 | } | |
e0a08581 | 2460 | |
b8d381c4 | 2461 | while (budget != 0 && (u = m->dbus_unit_queue)) { |
e0a08581 | 2462 | |
595ed347 | 2463 | assert(u->in_dbus_queue); |
c1e1601e | 2464 | |
595ed347 | 2465 | bus_unit_send_change_signal(u); |
b8d381c4 LP |
2466 | n++; |
2467 | ||
f5fbe71d | 2468 | if (budget != UINT_MAX) |
b8d381c4 | 2469 | budget--; |
c1e1601e LP |
2470 | } |
2471 | ||
b8d381c4 | 2472 | while (budget != 0 && (j = m->dbus_job_queue)) { |
c1e1601e LP |
2473 | assert(j->in_dbus_queue); |
2474 | ||
2475 | bus_job_send_change_signal(j); | |
b8d381c4 LP |
2476 | n++; |
2477 | ||
f5fbe71d | 2478 | if (budget != UINT_MAX) |
b8d381c4 | 2479 | budget--; |
c1e1601e LP |
2480 | } |
2481 | ||
b8d381c4 | 2482 | if (m->send_reloading_done) { |
71445ae7 | 2483 | m->send_reloading_done = false; |
718db961 | 2484 | bus_manager_send_reloading(m, false); |
b8d381c4 | 2485 | n++; |
71445ae7 LP |
2486 | } |
2487 | ||
b8d381c4 | 2488 | if (m->pending_reload_message) { |
209de525 | 2489 | bus_send_pending_reload_message(m); |
e0a08581 LP |
2490 | n++; |
2491 | } | |
718db961 | 2492 | |
c1e1601e | 2493 | return n; |
9152c765 LP |
2494 | } |
2495 | ||
d8fdc620 LP |
2496 | static int manager_dispatch_cgroups_agent_fd(sd_event_source *source, int fd, uint32_t revents, void *userdata) { |
2497 | Manager *m = userdata; | |
60473f0c | 2498 | char buf[PATH_MAX]; |
d8fdc620 LP |
2499 | ssize_t n; |
2500 | ||
2501 | n = recv(fd, buf, sizeof(buf), 0); | |
2502 | if (n < 0) | |
2503 | return log_error_errno(errno, "Failed to read cgroups agent message: %m"); | |
2504 | if (n == 0) { | |
2505 | log_error("Got zero-length cgroups agent message, ignoring."); | |
2506 | return 0; | |
2507 | } | |
2508 | if ((size_t) n >= sizeof(buf)) { | |
2509 | log_error("Got overly long cgroups agent message, ignoring."); | |
2510 | return 0; | |
2511 | } | |
2512 | ||
2513 | if (memchr(buf, 0, n)) { | |
2514 | log_error("Got cgroups agent message with embedded NUL byte, ignoring."); | |
2515 | return 0; | |
2516 | } | |
2517 | buf[n] = 0; | |
2518 | ||
2519 | manager_notify_cgroup_empty(m, buf); | |
d5f15326 | 2520 | (void) bus_forward_agent_released(m, buf); |
d8fdc620 LP |
2521 | |
2522 | return 0; | |
2523 | } | |
2524 | ||
5151b4cc | 2525 | static bool manager_process_barrier_fd(char * const *tags, FDSet *fds) { |
4f07ddfa KKD |
2526 | |
2527 | /* nothing else must be sent when using BARRIER=1 */ | |
5151b4cc | 2528 | if (strv_contains(tags, "BARRIER=1")) { |
46276454 | 2529 | if (strv_length(tags) != 1) |
5151b4cc | 2530 | log_warning("Extra notification messages sent with BARRIER=1, ignoring everything."); |
46276454 LP |
2531 | else if (fdset_size(fds) != 1) |
2532 | log_warning("Got incorrect number of fds with BARRIER=1, closing them."); | |
5151b4cc BR |
2533 | |
2534 | /* Drop the message if BARRIER=1 was found */ | |
4f07ddfa | 2535 | return true; |
5151b4cc | 2536 | } |
4f07ddfa KKD |
2537 | |
2538 | return false; | |
2539 | } | |
2540 | ||
db256aab LP |
2541 | static void manager_invoke_notify_message( |
2542 | Manager *m, | |
2543 | Unit *u, | |
2544 | const struct ucred *ucred, | |
5151b4cc | 2545 | char * const *tags, |
db256aab LP |
2546 | FDSet *fds) { |
2547 | ||
5ba6985b LP |
2548 | assert(m); |
2549 | assert(u); | |
db256aab | 2550 | assert(ucred); |
5151b4cc | 2551 | assert(tags); |
5ba6985b | 2552 | |
62a76913 | 2553 | if (u->notifygen == m->notifygen) /* Already invoked on this same unit in this same iteration? */ |
5ba6985b | 2554 | return; |
62a76913 LP |
2555 | u->notifygen = m->notifygen; |
2556 | ||
5151b4cc | 2557 | if (UNIT_VTABLE(u)->notify_message) |
db256aab | 2558 | UNIT_VTABLE(u)->notify_message(u, ucred, tags, fds); |
62a76913 | 2559 | |
5151b4cc BR |
2560 | else if (DEBUG_LOGGING) { |
2561 | _cleanup_free_ char *buf = NULL, *x = NULL, *y = NULL; | |
a86b7675 | 2562 | |
5151b4cc BR |
2563 | buf = strv_join(tags, ", "); |
2564 | if (buf) | |
2565 | x = ellipsize(buf, 20, 90); | |
a86b7675 | 2566 | if (x) |
da5fb861 LP |
2567 | y = cescape(x); |
2568 | ||
a86b7675 ZJS |
2569 | log_unit_debug(u, "Got notification message \"%s\", ignoring.", strnull(y)); |
2570 | } | |
5ba6985b LP |
2571 | } |
2572 | ||
718db961 | 2573 | static int manager_dispatch_notify_fd(sd_event_source *source, int fd, uint32_t revents, void *userdata) { |
3d0b8a55 | 2574 | |
b215b0ed | 2575 | _cleanup_fdset_free_ FDSet *fds = NULL; |
99534007 | 2576 | Manager *m = ASSERT_PTR(userdata); |
b215b0ed DH |
2577 | char buf[NOTIFY_BUFFER_MAX+1]; |
2578 | struct iovec iovec = { | |
2579 | .iov_base = buf, | |
2580 | .iov_len = sizeof(buf)-1, | |
2581 | }; | |
fb29cdbe LP |
2582 | CMSG_BUFFER_TYPE(CMSG_SPACE(sizeof(struct ucred)) + |
2583 | CMSG_SPACE(sizeof(int) * NOTIFY_FD_MAX)) control; | |
b215b0ed DH |
2584 | struct msghdr msghdr = { |
2585 | .msg_iov = &iovec, | |
2586 | .msg_iovlen = 1, | |
2587 | .msg_control = &control, | |
2588 | .msg_controllen = sizeof(control), | |
2589 | }; | |
2590 | ||
2591 | struct cmsghdr *cmsg; | |
2592 | struct ucred *ucred = NULL; | |
62a76913 | 2593 | _cleanup_free_ Unit **array_copy = NULL; |
5151b4cc | 2594 | _cleanup_strv_free_ char **tags = NULL; |
62a76913 | 2595 | Unit *u1, *u2, **array; |
b215b0ed | 2596 | int r, *fd_array = NULL; |
da6053d0 | 2597 | size_t n_fds = 0; |
62a76913 | 2598 | bool found = false; |
8c47c732 LP |
2599 | ssize_t n; |
2600 | ||
718db961 LP |
2601 | assert(m->notify_fd == fd); |
2602 | ||
2603 | if (revents != EPOLLIN) { | |
2604 | log_warning("Got unexpected poll event for notify fd."); | |
2605 | return 0; | |
2606 | } | |
8c47c732 | 2607 | |
3691bcf3 | 2608 | n = recvmsg_safe(m->notify_fd, &msghdr, MSG_DONTWAIT|MSG_CMSG_CLOEXEC|MSG_TRUNC); |
882321a1 ZJS |
2609 | if (ERRNO_IS_NEG_TRANSIENT(n)) |
2610 | return 0; /* Spurious wakeup, try again */ | |
2611 | if (n == -EXFULL) { | |
2612 | log_warning("Got message with truncated control data (too many fds sent?), ignoring."); | |
2613 | return 0; | |
8add30a0 | 2614 | } |
882321a1 ZJS |
2615 | if (n < 0) |
2616 | /* If this is any other, real error, then stop processing this socket. This of course means | |
2617 | * we won't take notification messages anymore, but that's still better than busy looping: | |
2618 | * being woken up over and over again, but being unable to actually read the message from the | |
2619 | * socket. */ | |
2620 | return log_error_errno(n, "Failed to receive notification message: %m"); | |
a354329f | 2621 | |
882321a1 | 2622 | CMSG_FOREACH(cmsg, &msghdr) |
b215b0ed | 2623 | if (cmsg->cmsg_level == SOL_SOCKET && cmsg->cmsg_type == SCM_RIGHTS) { |
a354329f | 2624 | |
3691bcf3 | 2625 | assert(!fd_array); |
b1d02191 | 2626 | fd_array = CMSG_TYPED_DATA(cmsg, int); |
b215b0ed | 2627 | n_fds = (cmsg->cmsg_len - CMSG_LEN(0)) / sizeof(int); |
a354329f | 2628 | |
b215b0ed DH |
2629 | } else if (cmsg->cmsg_level == SOL_SOCKET && |
2630 | cmsg->cmsg_type == SCM_CREDENTIALS && | |
2631 | cmsg->cmsg_len == CMSG_LEN(sizeof(struct ucred))) { | |
a354329f | 2632 | |
3691bcf3 | 2633 | assert(!ucred); |
b1d02191 | 2634 | ucred = CMSG_TYPED_DATA(cmsg, struct ucred); |
a354329f LP |
2635 | } |
2636 | ||
b215b0ed DH |
2637 | if (n_fds > 0) { |
2638 | assert(fd_array); | |
a354329f | 2639 | |
b215b0ed DH |
2640 | r = fdset_new_array(&fds, fd_array, n_fds); |
2641 | if (r < 0) { | |
2642 | close_many(fd_array, n_fds); | |
9987750e FB |
2643 | log_oom(); |
2644 | return 0; | |
a354329f | 2645 | } |
b215b0ed | 2646 | } |
8c47c732 | 2647 | |
15e23e8c | 2648 | if (!ucred || !pid_is_valid(ucred->pid)) { |
b215b0ed DH |
2649 | log_warning("Received notify message without valid credentials. Ignoring."); |
2650 | return 0; | |
2651 | } | |
8c47c732 | 2652 | |
045a3d59 | 2653 | if ((size_t) n >= sizeof(buf) || (msghdr.msg_flags & MSG_TRUNC)) { |
b215b0ed DH |
2654 | log_warning("Received notify message exceeded maximum size. Ignoring."); |
2655 | return 0; | |
2656 | } | |
8c47c732 | 2657 | |
8f41e6b6 ZJS |
2658 | /* As extra safety check, let's make sure the string we get doesn't contain embedded NUL bytes. |
2659 | * We permit one trailing NUL byte in the message, but don't expect it. */ | |
875ca88d LP |
2660 | if (n > 1 && memchr(buf, 0, n-1)) { |
2661 | log_warning("Received notify message with embedded NUL bytes. Ignoring."); | |
2662 | return 0; | |
2663 | } | |
2664 | ||
8f41e6b6 | 2665 | /* Make sure it's NUL-terminated, then parse it to obtain the tags list. */ |
b215b0ed | 2666 | buf[n] = 0; |
5151b4cc BR |
2667 | tags = strv_split_newlines(buf); |
2668 | if (!tags) { | |
2669 | log_oom(); | |
2670 | return 0; | |
2671 | } | |
8c47c732 | 2672 | |
8f41e6b6 | 2673 | /* Possibly a barrier fd, let's see. */ |
7b0a1267 LP |
2674 | if (manager_process_barrier_fd(tags, fds)) { |
2675 | log_debug("Received barrier notification message from PID " PID_FMT ".", ucred->pid); | |
4f07ddfa | 2676 | return 0; |
7b0a1267 | 2677 | } |
4f07ddfa | 2678 | |
62a76913 LP |
2679 | /* Increase the generation counter used for filtering out duplicate unit invocations. */ |
2680 | m->notifygen++; | |
2681 | ||
495e75ed LP |
2682 | /* Generate lookup key from the PID (we have no pidfd here, after all) */ |
2683 | PidRef pidref = PIDREF_MAKE_FROM_PID(ucred->pid); | |
2684 | ||
62a76913 | 2685 | /* Notify every unit that might be interested, which might be multiple. */ |
495e75ed LP |
2686 | u1 = manager_get_unit_by_pidref_cgroup(m, &pidref); |
2687 | u2 = hashmap_get(m->watch_pids, &pidref); | |
2688 | array = hashmap_get(m->watch_pids_more, &pidref); | |
62a76913 LP |
2689 | if (array) { |
2690 | size_t k = 0; | |
5ba6985b | 2691 | |
62a76913 LP |
2692 | while (array[k]) |
2693 | k++; | |
5ba6985b | 2694 | |
62a76913 LP |
2695 | array_copy = newdup(Unit*, array, k+1); |
2696 | if (!array_copy) | |
2697 | log_oom(); | |
2698 | } | |
8f41e6b6 ZJS |
2699 | /* And now invoke the per-unit callbacks. Note that manager_invoke_notify_message() will handle |
2700 | * duplicate units make sure we only invoke each unit's handler once. */ | |
62a76913 | 2701 | if (u1) { |
5151b4cc | 2702 | manager_invoke_notify_message(m, u1, ucred, tags, fds); |
62a76913 LP |
2703 | found = true; |
2704 | } | |
2705 | if (u2) { | |
5151b4cc | 2706 | manager_invoke_notify_message(m, u2, ucred, tags, fds); |
62a76913 LP |
2707 | found = true; |
2708 | } | |
2709 | if (array_copy) | |
2710 | for (size_t i = 0; array_copy[i]; i++) { | |
5151b4cc | 2711 | manager_invoke_notify_message(m, array_copy[i], ucred, tags, fds); |
62a76913 LP |
2712 | found = true; |
2713 | } | |
8c47c732 | 2714 | |
62a76913 LP |
2715 | if (!found) |
2716 | log_warning("Cannot find unit for notify message of PID "PID_FMT", ignoring.", ucred->pid); | |
a354329f | 2717 | |
b215b0ed | 2718 | if (fdset_size(fds) > 0) |
5fd2c135 | 2719 | log_warning("Got extra auxiliary fds with notification message, closing them."); |
8c47c732 LP |
2720 | |
2721 | return 0; | |
2722 | } | |
2723 | ||
62a76913 LP |
2724 | static void manager_invoke_sigchld_event( |
2725 | Manager *m, | |
2726 | Unit *u, | |
2727 | const siginfo_t *si) { | |
36f20ae3 | 2728 | |
5ba6985b LP |
2729 | assert(m); |
2730 | assert(u); | |
2731 | assert(si); | |
2732 | ||
62a76913 LP |
2733 | /* Already invoked the handler of this unit in this iteration? Then don't process this again */ |
2734 | if (u->sigchldgen == m->sigchldgen) | |
2735 | return; | |
2736 | u->sigchldgen = m->sigchldgen; | |
5ba6985b | 2737 | |
62a76913 | 2738 | log_unit_debug(u, "Child "PID_FMT" belongs to %s.", si->si_pid, u->id); |
5ba6985b | 2739 | unit_unwatch_pid(u, si->si_pid); |
e57051f5 | 2740 | |
62a76913 LP |
2741 | if (UNIT_VTABLE(u)->sigchld_event) |
2742 | UNIT_VTABLE(u)->sigchld_event(u, si->si_pid, si->si_code, si->si_status); | |
5ba6985b LP |
2743 | } |
2744 | ||
575b300b | 2745 | static int manager_dispatch_sigchld(sd_event_source *source, void *userdata) { |
99534007 | 2746 | Manager *m = ASSERT_PTR(userdata); |
575b300b LP |
2747 | siginfo_t si = {}; |
2748 | int r; | |
2749 | ||
2750 | assert(source); | |
9152c765 | 2751 | |
8f41e6b6 ZJS |
2752 | /* First we call waitid() for a PID and do not reap the zombie. That way we can still access |
2753 | * /proc/$PID for it while it is a zombie. */ | |
9152c765 | 2754 | |
575b300b | 2755 | if (waitid(P_ALL, 0, &si, WEXITED|WNOHANG|WNOWAIT) < 0) { |
acbb0225 | 2756 | |
8afabc50 AJ |
2757 | if (errno != ECHILD) |
2758 | log_error_errno(errno, "Failed to peek for child with waitid(), ignoring: %m"); | |
acbb0225 | 2759 | |
8afabc50 | 2760 | goto turn_off; |
575b300b | 2761 | } |
4112df16 | 2762 | |
575b300b LP |
2763 | if (si.si_pid <= 0) |
2764 | goto turn_off; | |
2765 | ||
2766 | if (IN_SET(si.si_code, CLD_EXITED, CLD_KILLED, CLD_DUMPED)) { | |
62a76913 | 2767 | _cleanup_free_ Unit **array_copy = NULL; |
575b300b | 2768 | _cleanup_free_ char *name = NULL; |
62a76913 | 2769 | Unit *u1, *u2, **array; |
575b300b LP |
2770 | |
2771 | (void) get_process_comm(si.si_pid, &name); | |
2772 | ||
2773 | log_debug("Child "PID_FMT" (%s) died (code=%s, status=%i/%s)", | |
2774 | si.si_pid, strna(name), | |
2775 | sigchld_code_to_string(si.si_code), | |
2776 | si.si_status, | |
2777 | strna(si.si_code == CLD_EXITED | |
2778 | ? exit_status_to_string(si.si_status, EXIT_STATUS_FULL) | |
2779 | : signal_to_string(si.si_status))); | |
2780 | ||
62a76913 LP |
2781 | /* Increase the generation counter used for filtering out duplicate unit invocations */ |
2782 | m->sigchldgen++; | |
2783 | ||
495e75ed LP |
2784 | /* We look this up by a PidRef that only consists of the PID. After all we couldn't create a |
2785 | * pidfd here any more even if we wanted (since the process just exited). */ | |
2786 | PidRef pidref = PIDREF_MAKE_FROM_PID(si.si_pid); | |
2787 | ||
62a76913 | 2788 | /* And now figure out the unit this belongs to, it might be multiple... */ |
495e75ed LP |
2789 | u1 = manager_get_unit_by_pidref_cgroup(m, &pidref); |
2790 | u2 = hashmap_get(m->watch_pids, &pidref); | |
2791 | array = hashmap_get(m->watch_pids_more, &pidref); | |
62a76913 LP |
2792 | if (array) { |
2793 | size_t n = 0; | |
2794 | ||
5238e957 | 2795 | /* Count how many entries the array has */ |
62a76913 LP |
2796 | while (array[n]) |
2797 | n++; | |
2798 | ||
2799 | /* Make a copy of the array so that we don't trip up on the array changing beneath us */ | |
2800 | array_copy = newdup(Unit*, array, n+1); | |
2801 | if (!array_copy) | |
2802 | log_oom(); | |
2803 | } | |
2804 | ||
2805 | /* Finally, execute them all. Note that u1, u2 and the array might contain duplicates, but | |
2806 | * that's fine, manager_invoke_sigchld_event() will ensure we only invoke the handlers once for | |
2807 | * each iteration. */ | |
2ba6ae6b ZJS |
2808 | if (u1) { |
2809 | /* We check for oom condition, in case we got SIGCHLD before the oom notification. | |
2810 | * We only do this for the cgroup the PID belonged to. */ | |
2811 | (void) unit_check_oom(u1); | |
2812 | ||
288bd406 | 2813 | /* We check if systemd-oomd performed a kill so that we log and notify appropriately */ |
fe8d22fb AZ |
2814 | (void) unit_check_oomd_kill(u1); |
2815 | ||
62a76913 | 2816 | manager_invoke_sigchld_event(m, u1, &si); |
2ba6ae6b | 2817 | } |
62a76913 LP |
2818 | if (u2) |
2819 | manager_invoke_sigchld_event(m, u2, &si); | |
2820 | if (array_copy) | |
2821 | for (size_t i = 0; array_copy[i]; i++) | |
2822 | manager_invoke_sigchld_event(m, array_copy[i], &si); | |
575b300b | 2823 | } |
9152c765 | 2824 | |
575b300b LP |
2825 | /* And now, we actually reap the zombie. */ |
2826 | if (waitid(P_PID, si.si_pid, &si, WEXITED) < 0) { | |
2827 | log_error_errno(errno, "Failed to dequeue child, ignoring: %m"); | |
2828 | return 0; | |
2829 | } | |
9152c765 | 2830 | |
575b300b | 2831 | return 0; |
8c47c732 | 2832 | |
575b300b LP |
2833 | turn_off: |
2834 | /* All children processed for now, turn off event source */ | |
4112df16 | 2835 | |
575b300b LP |
2836 | r = sd_event_source_set_enabled(m->sigchld_event_source, SD_EVENT_OFF); |
2837 | if (r < 0) | |
2838 | return log_error_errno(r, "Failed to disable SIGCHLD event source: %m"); | |
9152c765 LP |
2839 | |
2840 | return 0; | |
2841 | } | |
2842 | ||
5f968096 ZJS |
2843 | static void manager_start_special(Manager *m, const char *name, JobMode mode) { |
2844 | Job *job; | |
398ef8ba | 2845 | |
5f968096 ZJS |
2846 | if (manager_add_job_by_name_and_warn(m, JOB_START, name, mode, NULL, &job) < 0) |
2847 | return; | |
1e001f52 | 2848 | |
5f968096 ZJS |
2849 | const char *s = unit_status_string(job->unit, NULL); |
2850 | ||
2851 | log_info("Activating special unit %s...", s); | |
3889fc6f ZJS |
2852 | |
2853 | sd_notifyf(false, | |
2854 | "STATUS=Activating special unit %s...", s); | |
2855 | m->status_ready = false; | |
28247076 LP |
2856 | } |
2857 | ||
24dd31c1 LN |
2858 | static void manager_handle_ctrl_alt_del(Manager *m) { |
2859 | /* If the user presses C-A-D more than | |
2860 | * 7 times within 2s, we reboot/shutdown immediately, | |
2861 | * unless it was disabled in system.conf */ | |
2862 | ||
7994ac1d | 2863 | if (ratelimit_below(&m->ctrl_alt_del_ratelimit) || m->cad_burst_action == EMERGENCY_ACTION_NONE) |
5f968096 | 2864 | manager_start_special(m, SPECIAL_CTRL_ALT_DEL_TARGET, JOB_REPLACE_IRREVERSIBLY); |
ae8c7939 | 2865 | else |
7af67e9a | 2866 | emergency_action(m, m->cad_burst_action, EMERGENCY_ACTION_WARN, NULL, -1, |
ae8c7939 | 2867 | "Ctrl-Alt-Del was pressed more than 7 times within 2s"); |
24dd31c1 LN |
2868 | } |
2869 | ||
718db961 | 2870 | static int manager_dispatch_signal_fd(sd_event_source *source, int fd, uint32_t revents, void *userdata) { |
99534007 | 2871 | Manager *m = ASSERT_PTR(userdata); |
9152c765 LP |
2872 | ssize_t n; |
2873 | struct signalfd_siginfo sfsi; | |
dacd6cee | 2874 | int r; |
9152c765 | 2875 | |
718db961 LP |
2876 | assert(m->signal_fd == fd); |
2877 | ||
2878 | if (revents != EPOLLIN) { | |
2879 | log_warning("Got unexpected events from signal file descriptor."); | |
2880 | return 0; | |
2881 | } | |
9152c765 | 2882 | |
575b300b | 2883 | n = read(m->signal_fd, &sfsi, sizeof(sfsi)); |
8add30a0 YW |
2884 | if (n < 0) { |
2885 | if (ERRNO_IS_TRANSIENT(errno)) | |
575b300b | 2886 | return 0; |
9152c765 | 2887 | |
575b300b LP |
2888 | /* We return an error here, which will kill this handler, |
2889 | * to avoid a busy loop on read error. */ | |
2890 | return log_error_errno(errno, "Reading from signal fd failed: %m"); | |
2891 | } | |
8add30a0 | 2892 | if (n != sizeof(sfsi)) { |
c0f86d66 | 2893 | log_warning("Truncated read from signal fd (%zi bytes), ignoring!", n); |
8add30a0 YW |
2894 | return 0; |
2895 | } | |
9152c765 | 2896 | |
575b300b LP |
2897 | log_received_signal(sfsi.ssi_signo == SIGCHLD || |
2898 | (sfsi.ssi_signo == SIGTERM && MANAGER_IS_USER(m)) | |
2899 | ? LOG_DEBUG : LOG_INFO, | |
2900 | &sfsi); | |
1e001f52 | 2901 | |
575b300b | 2902 | switch (sfsi.ssi_signo) { |
b9cd2ec1 | 2903 | |
575b300b LP |
2904 | case SIGCHLD: |
2905 | r = sd_event_source_set_enabled(m->sigchld_event_source, SD_EVENT_ON); | |
2906 | if (r < 0) | |
8afabc50 | 2907 | log_warning_errno(r, "Failed to enable SIGCHLD event source, ignoring: %m"); |
b9cd2ec1 | 2908 | |
575b300b | 2909 | break; |
84e9af1e | 2910 | |
575b300b LP |
2911 | case SIGTERM: |
2912 | if (MANAGER_IS_SYSTEM(m)) { | |
ba0c7754 | 2913 | /* This is for compatibility with the original sysvinit */ |
c52b19d6 LP |
2914 | if (verify_run_space_and_log("Refusing to reexecute") < 0) |
2915 | break; | |
2916 | ||
2917 | m->objective = MANAGER_REEXECUTE; | |
a1b256b0 | 2918 | break; |
575b300b | 2919 | } |
84e9af1e | 2920 | |
575b300b LP |
2921 | _fallthrough_; |
2922 | case SIGINT: | |
2923 | if (MANAGER_IS_SYSTEM(m)) | |
2924 | manager_handle_ctrl_alt_del(m); | |
2925 | else | |
5f968096 | 2926 | manager_start_special(m, SPECIAL_EXIT_TARGET, JOB_REPLACE_IRREVERSIBLY); |
575b300b | 2927 | break; |
84e9af1e | 2928 | |
575b300b | 2929 | case SIGWINCH: |
ba0c7754 | 2930 | /* This is a nop on non-init */ |
575b300b | 2931 | if (MANAGER_IS_SYSTEM(m)) |
5f968096 | 2932 | manager_start_special(m, SPECIAL_KBREQUEST_TARGET, JOB_REPLACE); |
84e9af1e | 2933 | |
575b300b | 2934 | break; |
84e9af1e | 2935 | |
575b300b | 2936 | case SIGPWR: |
ba0c7754 | 2937 | /* This is a nop on non-init */ |
575b300b | 2938 | if (MANAGER_IS_SYSTEM(m)) |
5f968096 | 2939 | manager_start_special(m, SPECIAL_SIGPWR_TARGET, JOB_REPLACE); |
6632c602 | 2940 | |
575b300b | 2941 | break; |
57ee42ce | 2942 | |
8559b3b7 | 2943 | case SIGUSR1: |
2b680534 | 2944 | if (manager_dbus_is_running(m, false)) { |
575b300b | 2945 | log_info("Trying to reconnect to bus..."); |
575b300b | 2946 | |
8559b3b7 LP |
2947 | (void) bus_init_api(m); |
2948 | ||
2949 | if (MANAGER_IS_SYSTEM(m)) | |
2950 | (void) bus_init_system(m); | |
5f968096 ZJS |
2951 | } else |
2952 | manager_start_special(m, SPECIAL_DBUS_SERVICE, JOB_REPLACE); | |
57ee42ce | 2953 | |
575b300b | 2954 | break; |
575b300b LP |
2955 | |
2956 | case SIGUSR2: { | |
2957 | _cleanup_free_ char *dump = NULL; | |
2958 | ||
d1d8786c | 2959 | r = manager_get_dump_string(m, /* patterns= */ NULL, &dump); |
575b300b LP |
2960 | if (r < 0) { |
2961 | log_warning_errno(errno, "Failed to acquire manager dump: %m"); | |
57ee42ce LP |
2962 | break; |
2963 | } | |
2964 | ||
575b300b LP |
2965 | log_dump(LOG_INFO, dump); |
2966 | break; | |
2967 | } | |
2149e37c | 2968 | |
575b300b | 2969 | case SIGHUP: |
c52b19d6 LP |
2970 | if (verify_run_space_and_log("Refusing to reload") < 0) |
2971 | break; | |
2972 | ||
2973 | m->objective = MANAGER_RELOAD; | |
575b300b LP |
2974 | break; |
2975 | ||
2976 | default: { | |
2977 | ||
2978 | /* Starting SIGRTMIN+0 */ | |
2979 | static const struct { | |
2980 | const char *target; | |
2981 | JobMode mode; | |
2982 | } target_table[] = { | |
13ffc607 LP |
2983 | [0] = { SPECIAL_DEFAULT_TARGET, JOB_ISOLATE }, |
2984 | [1] = { SPECIAL_RESCUE_TARGET, JOB_ISOLATE }, | |
2985 | [2] = { SPECIAL_EMERGENCY_TARGET, JOB_ISOLATE }, | |
2986 | [3] = { SPECIAL_HALT_TARGET, JOB_REPLACE_IRREVERSIBLY }, | |
2987 | [4] = { SPECIAL_POWEROFF_TARGET, JOB_REPLACE_IRREVERSIBLY }, | |
2988 | [5] = { SPECIAL_REBOOT_TARGET, JOB_REPLACE_IRREVERSIBLY }, | |
2989 | [6] = { SPECIAL_KEXEC_TARGET, JOB_REPLACE_IRREVERSIBLY }, | |
2990 | [7] = { SPECIAL_SOFT_REBOOT_TARGET, JOB_REPLACE_IRREVERSIBLY }, | |
575b300b LP |
2991 | }; |
2992 | ||
2993 | /* Starting SIGRTMIN+13, so that target halt and system halt are 10 apart */ | |
af41e508 | 2994 | static const ManagerObjective objective_table[] = { |
575b300b LP |
2995 | [0] = MANAGER_HALT, |
2996 | [1] = MANAGER_POWEROFF, | |
2997 | [2] = MANAGER_REBOOT, | |
2998 | [3] = MANAGER_KEXEC, | |
13ffc607 | 2999 | [4] = MANAGER_SOFT_REBOOT, |
575b300b | 3000 | }; |
b2cdc666 | 3001 | |
575b300b LP |
3002 | if ((int) sfsi.ssi_signo >= SIGRTMIN+0 && |
3003 | (int) sfsi.ssi_signo < SIGRTMIN+(int) ELEMENTSOF(target_table)) { | |
3004 | int idx = (int) sfsi.ssi_signo - SIGRTMIN; | |
5f968096 | 3005 | manager_start_special(m, target_table[idx].target, target_table[idx].mode); |
1005d14f | 3006 | break; |
2149e37c | 3007 | } |
1005d14f | 3008 | |
575b300b | 3009 | if ((int) sfsi.ssi_signo >= SIGRTMIN+13 && |
af41e508 LP |
3010 | (int) sfsi.ssi_signo < SIGRTMIN+13+(int) ELEMENTSOF(objective_table)) { |
3011 | m->objective = objective_table[sfsi.ssi_signo - SIGRTMIN - 13]; | |
575b300b LP |
3012 | break; |
3013 | } | |
3014 | ||
3015 | switch (sfsi.ssi_signo - SIGRTMIN) { | |
3016 | ||
29e6b0c1 LP |
3017 | case 18: { |
3018 | bool generic = false; | |
3019 | ||
3020 | if (sfsi.ssi_code != SI_QUEUE) | |
3021 | generic = true; | |
3022 | else { | |
3023 | /* Override a few select commands by our own PID1-specific logic */ | |
3024 | ||
3025 | switch (sfsi.ssi_int) { | |
3026 | ||
3027 | case _COMMON_SIGNAL_COMMAND_LOG_LEVEL_BASE..._COMMON_SIGNAL_COMMAND_LOG_LEVEL_END: | |
3028 | manager_override_log_level(m, sfsi.ssi_int - _COMMON_SIGNAL_COMMAND_LOG_LEVEL_BASE); | |
3029 | break; | |
3030 | ||
3031 | case COMMON_SIGNAL_COMMAND_CONSOLE: | |
3032 | manager_override_log_target(m, LOG_TARGET_CONSOLE); | |
3033 | break; | |
3034 | ||
3035 | case COMMON_SIGNAL_COMMAND_JOURNAL: | |
3036 | manager_override_log_target(m, LOG_TARGET_JOURNAL); | |
3037 | break; | |
3038 | ||
3039 | case COMMON_SIGNAL_COMMAND_KMSG: | |
3040 | manager_override_log_target(m, LOG_TARGET_KMSG); | |
3041 | break; | |
3042 | ||
3043 | case COMMON_SIGNAL_COMMAND_NULL: | |
3044 | manager_override_log_target(m, LOG_TARGET_NULL); | |
3045 | break; | |
3046 | ||
0112c37c LB |
3047 | case MANAGER_SIGNAL_COMMAND_DUMP_JOBS: { |
3048 | _cleanup_free_ char *dump_jobs = NULL; | |
3049 | ||
3050 | r = manager_get_dump_jobs_string(m, /* patterns= */ NULL, " ", &dump_jobs); | |
3051 | if (r < 0) { | |
3052 | log_warning_errno(errno, "Failed to acquire manager jobs dump: %m"); | |
3053 | break; | |
3054 | } | |
3055 | ||
3056 | log_dump(LOG_INFO, dump_jobs); | |
3057 | break; | |
3058 | } | |
3059 | ||
29e6b0c1 LP |
3060 | default: |
3061 | generic = true; | |
3062 | } | |
3063 | } | |
3064 | ||
3065 | if (generic) | |
3066 | return sigrtmin18_handler(source, &sfsi, NULL); | |
3067 | ||
3068 | break; | |
3069 | } | |
3070 | ||
575b300b | 3071 | case 20: |
43bba15a | 3072 | manager_override_show_status(m, SHOW_STATUS_YES, "signal"); |
a16e1123 LP |
3073 | break; |
3074 | ||
575b300b | 3075 | case 21: |
43bba15a | 3076 | manager_override_show_status(m, SHOW_STATUS_NO, "signal"); |
575b300b | 3077 | break; |
7d793605 | 3078 | |
575b300b | 3079 | case 22: |
a6ecbf83 | 3080 | manager_override_log_level(m, LOG_DEBUG); |
575b300b LP |
3081 | break; |
3082 | ||
3083 | case 23: | |
a6ecbf83 | 3084 | manager_restore_original_log_level(m); |
575b300b | 3085 | break; |
0003d1ab | 3086 | |
575b300b LP |
3087 | case 24: |
3088 | if (MANAGER_IS_USER(m)) { | |
af41e508 | 3089 | m->objective = MANAGER_EXIT; |
575b300b | 3090 | return 0; |
0658666b | 3091 | } |
9152c765 | 3092 | |
575b300b LP |
3093 | /* This is a nop on init */ |
3094 | break; | |
3095 | ||
463aef23 FB |
3096 | case 25: |
3097 | m->objective = MANAGER_REEXECUTE; | |
3098 | break; | |
3099 | ||
575b300b LP |
3100 | case 26: |
3101 | case 29: /* compatibility: used to be mapped to LOG_TARGET_SYSLOG_OR_KMSG */ | |
bda7d78b | 3102 | manager_restore_original_log_target(m); |
575b300b LP |
3103 | break; |
3104 | ||
3105 | case 27: | |
bda7d78b | 3106 | manager_override_log_target(m, LOG_TARGET_CONSOLE); |
575b300b LP |
3107 | break; |
3108 | ||
3109 | case 28: | |
bda7d78b | 3110 | manager_override_log_target(m, LOG_TARGET_KMSG); |
575b300b LP |
3111 | break; |
3112 | ||
3113 | default: | |
3114 | log_warning("Got unhandled signal <%s>.", signal_to_string(sfsi.ssi_signo)); | |
3115 | } | |
3116 | }} | |
034c6ed7 LP |
3117 | |
3118 | return 0; | |
3119 | } | |
3120 | ||
718db961 | 3121 | static int manager_dispatch_time_change_fd(sd_event_source *source, int fd, uint32_t revents, void *userdata) { |
99534007 | 3122 | Manager *m = ASSERT_PTR(userdata); |
718db961 | 3123 | Unit *u; |
034c6ed7 | 3124 | |
a80c1575 | 3125 | log_struct(LOG_DEBUG, |
2b044526 | 3126 | "MESSAGE_ID=" SD_MESSAGE_TIME_CHANGE_STR, |
a1230ff9 | 3127 | LOG_MESSAGE("Time has been changed")); |
034c6ed7 | 3128 | |
718db961 | 3129 | /* Restart the watch */ |
7feedd18 | 3130 | (void) manager_setup_time_change(m); |
4e434314 | 3131 | |
90e74a66 | 3132 | HASHMAP_FOREACH(u, m->units) |
718db961 LP |
3133 | if (UNIT_VTABLE(u)->time_change) |
3134 | UNIT_VTABLE(u)->time_change(u); | |
ea430986 | 3135 | |
718db961 LP |
3136 | return 0; |
3137 | } | |
ea430986 | 3138 | |
bbf5fd8e LP |
3139 | static int manager_dispatch_timezone_change( |
3140 | sd_event_source *source, | |
3141 | const struct inotify_event *e, | |
3142 | void *userdata) { | |
3143 | ||
99534007 | 3144 | Manager *m = ASSERT_PTR(userdata); |
bbf5fd8e | 3145 | int changed; |
bbf5fd8e LP |
3146 | Unit *u; |
3147 | ||
bbf5fd8e LP |
3148 | log_debug("inotify event for /etc/localtime"); |
3149 | ||
3150 | changed = manager_read_timezone_stat(m); | |
f20db199 | 3151 | if (changed <= 0) |
bbf5fd8e | 3152 | return changed; |
bbf5fd8e LP |
3153 | |
3154 | /* Something changed, restart the watch, to ensure we watch the new /etc/localtime if it changed */ | |
3155 | (void) manager_setup_timezone_change(m); | |
3156 | ||
3157 | /* Read the new timezone */ | |
3158 | tzset(); | |
3159 | ||
3160 | log_debug("Timezone has been changed (now: %s).", tzname[daylight]); | |
3161 | ||
90e74a66 | 3162 | HASHMAP_FOREACH(u, m->units) |
bbf5fd8e LP |
3163 | if (UNIT_VTABLE(u)->timezone_change) |
3164 | UNIT_VTABLE(u)->timezone_change(u); | |
3165 | ||
3166 | return 0; | |
3167 | } | |
3168 | ||
718db961 | 3169 | static int manager_dispatch_idle_pipe_fd(sd_event_source *source, int fd, uint32_t revents, void *userdata) { |
99534007 | 3170 | Manager *m = ASSERT_PTR(userdata); |
8742514c | 3171 | |
718db961 | 3172 | assert(m->idle_pipe[2] == fd); |
8742514c | 3173 | |
8f41e6b6 ZJS |
3174 | /* There's at least one Type=idle child that just gave up on us waiting for the boot process to |
3175 | * complete. Let's now turn off any further console output if there's at least one service that needs | |
3176 | * console access, so that from now on our own output should not spill into that service's output | |
3177 | * anymore. After all, we support Type=idle only to beautify console output and it generally is set | |
3178 | * on services that want to own the console exclusively without our interference. */ | |
718db961 | 3179 | m->no_console_output = m->n_on_console > 0; |
03b717a3 | 3180 | |
30fd9a2d | 3181 | /* Acknowledge the child's request, and let all other children know too that they shouldn't wait |
8f41e6b6 | 3182 | * any longer by closing the pipes towards them, which is what they are waiting for. */ |
718db961 | 3183 | manager_close_idle_pipe(m); |
03b717a3 | 3184 | |
718db961 LP |
3185 | return 0; |
3186 | } | |
31a7eb86 | 3187 | |
718db961 | 3188 | static int manager_dispatch_jobs_in_progress(sd_event_source *source, usec_t usec, void *userdata) { |
99534007 | 3189 | Manager *m = ASSERT_PTR(userdata); |
fd08a840 | 3190 | int r; |
31a7eb86 | 3191 | |
fd08a840 | 3192 | assert(source); |
9152c765 | 3193 | |
718db961 | 3194 | manager_print_jobs_in_progress(m); |
fd08a840 | 3195 | |
39cf0351 | 3196 | r = sd_event_source_set_time_relative(source, JOBS_IN_PROGRESS_PERIOD_USEC); |
fd08a840 ZJS |
3197 | if (r < 0) |
3198 | return r; | |
3199 | ||
3200 | return sd_event_source_set_enabled(source, SD_EVENT_ONESHOT); | |
9152c765 LP |
3201 | } |
3202 | ||
3203 | int manager_loop(Manager *m) { | |
8c227e7f | 3204 | RateLimit rl = { .interval = 1*USEC_PER_SEC, .burst = 50000 }; |
9152c765 | 3205 | int r; |
9152c765 LP |
3206 | |
3207 | assert(m); | |
1fb70e66 | 3208 | assert(m->objective == MANAGER_OK); /* Ensure manager_startup() has been called */ |
9152c765 | 3209 | |
b0c918b9 LP |
3210 | manager_check_finished(m); |
3211 | ||
575b300b LP |
3212 | /* There might still be some zombies hanging around from before we were exec()'ed. Let's reap them. */ |
3213 | r = sd_event_source_set_enabled(m->sigchld_event_source, SD_EVENT_ON); | |
e96d6be7 | 3214 | if (r < 0) |
575b300b | 3215 | return log_error_errno(r, "Failed to enable SIGCHLD event source: %m"); |
a4312405 | 3216 | |
af41e508 | 3217 | while (m->objective == MANAGER_OK) { |
9152c765 | 3218 | |
ae4a0ec4 | 3219 | (void) watchdog_ping(); |
e96d6be7 | 3220 | |
7994ac1d | 3221 | if (!ratelimit_below(&rl)) { |
ea430986 LP |
3222 | /* Yay, something is going seriously wrong, pause a little */ |
3223 | log_warning("Looping too fast. Throttling execution a little."); | |
3224 | sleep(1); | |
3225 | } | |
3226 | ||
37a8e683 | 3227 | if (manager_dispatch_load_queue(m) > 0) |
23a177ef LP |
3228 | continue; |
3229 | ||
c5a97ed1 LP |
3230 | if (manager_dispatch_gc_job_queue(m) > 0) |
3231 | continue; | |
3232 | ||
3233 | if (manager_dispatch_gc_unit_queue(m) > 0) | |
701cc384 LP |
3234 | continue; |
3235 | ||
cf1265e1 | 3236 | if (manager_dispatch_cleanup_queue(m) > 0) |
c1e1601e | 3237 | continue; |
034c6ed7 | 3238 | |
91a6073e | 3239 | if (manager_dispatch_cgroup_realize_queue(m) > 0) |
c1e1601e LP |
3240 | continue; |
3241 | ||
0bc488c9 LP |
3242 | if (manager_dispatch_start_when_upheld_queue(m) > 0) |
3243 | continue; | |
3244 | ||
56c59592 LP |
3245 | if (manager_dispatch_stop_when_bound_queue(m) > 0) |
3246 | continue; | |
3247 | ||
a3c1168a LP |
3248 | if (manager_dispatch_stop_when_unneeded_queue(m) > 0) |
3249 | continue; | |
3250 | ||
6ac62d61 LP |
3251 | if (manager_dispatch_release_resources_queue(m) > 0) |
3252 | continue; | |
3253 | ||
c1e1601e | 3254 | if (manager_dispatch_dbus_queue(m) > 0) |
ea430986 | 3255 | continue; |
ea430986 | 3256 | |
c5f8a179 | 3257 | /* Sleep for watchdog runtime wait time */ |
12663295 | 3258 | r = sd_event_run(m->event, watchdog_runtime_wait()); |
23bbb0de MS |
3259 | if (r < 0) |
3260 | return log_error_errno(r, "Failed to run event loop: %m"); | |
a16e1123 | 3261 | } |
957ca890 | 3262 | |
af41e508 | 3263 | return m->objective; |
83c60c9f | 3264 | } |
ea430986 | 3265 | |
718db961 | 3266 | int manager_load_unit_from_dbus_path(Manager *m, const char *s, sd_bus_error *e, Unit **_u) { |
ede3a796 | 3267 | _cleanup_free_ char *n = NULL; |
4b58153d | 3268 | sd_id128_t invocation_id; |
ea430986 | 3269 | Unit *u; |
80fbf05e | 3270 | int r; |
ea430986 LP |
3271 | |
3272 | assert(m); | |
3273 | assert(s); | |
3274 | assert(_u); | |
3275 | ||
ede3a796 LP |
3276 | r = unit_name_from_dbus_path(s, &n); |
3277 | if (r < 0) | |
3278 | return r; | |
ea430986 | 3279 | |
da890466 | 3280 | /* Permit addressing units by invocation ID: if the passed bus path is suffixed by a 128-bit ID then |
8f41e6b6 | 3281 | * we use it as invocation ID. */ |
4b58153d LP |
3282 | r = sd_id128_from_string(n, &invocation_id); |
3283 | if (r >= 0) { | |
3284 | u = hashmap_get(m->units_by_invocation_id, &invocation_id); | |
3285 | if (u) { | |
3286 | *_u = u; | |
3287 | return 0; | |
3288 | } | |
3289 | ||
930c124c ZJS |
3290 | return sd_bus_error_setf(e, BUS_ERROR_NO_UNIT_FOR_INVOCATION_ID, |
3291 | "No unit with the specified invocation ID " SD_ID128_FORMAT_STR " known.", | |
3292 | SD_ID128_FORMAT_VAL(invocation_id)); | |
4b58153d LP |
3293 | } |
3294 | ||
00c83b43 | 3295 | /* If this didn't work, we check if this is a unit name */ |
930c124c ZJS |
3296 | if (!unit_name_is_valid(n, UNIT_NAME_PLAIN|UNIT_NAME_INSTANCE)) { |
3297 | _cleanup_free_ char *nn = NULL; | |
3298 | ||
3299 | nn = cescape(n); | |
3300 | return sd_bus_error_setf(e, SD_BUS_ERROR_INVALID_ARGS, | |
3301 | "Unit name %s is neither a valid invocation ID nor unit name.", strnull(nn)); | |
3302 | } | |
00c83b43 | 3303 | |
80fbf05e | 3304 | r = manager_load_unit(m, n, NULL, e, &u); |
80fbf05e MS |
3305 | if (r < 0) |
3306 | return r; | |
ea430986 LP |
3307 | |
3308 | *_u = u; | |
ea430986 LP |
3309 | return 0; |
3310 | } | |
86fbf370 LP |
3311 | |
3312 | int manager_get_job_from_dbus_path(Manager *m, const char *s, Job **_j) { | |
718db961 | 3313 | const char *p; |
86fbf370 | 3314 | unsigned id; |
718db961 | 3315 | Job *j; |
86fbf370 LP |
3316 | int r; |
3317 | ||
3318 | assert(m); | |
3319 | assert(s); | |
3320 | assert(_j); | |
3321 | ||
718db961 LP |
3322 | p = startswith(s, "/org/freedesktop/systemd1/job/"); |
3323 | if (!p) | |
86fbf370 LP |
3324 | return -EINVAL; |
3325 | ||
718db961 | 3326 | r = safe_atou(p, &id); |
8742514c | 3327 | if (r < 0) |
86fbf370 LP |
3328 | return r; |
3329 | ||
8742514c LP |
3330 | j = manager_get_job(m, id); |
3331 | if (!j) | |
86fbf370 LP |
3332 | return -ENOENT; |
3333 | ||
3334 | *_j = j; | |
3335 | ||
3336 | return 0; | |
3337 | } | |
dfcd764e | 3338 | |
4927fcae | 3339 | void manager_send_unit_audit(Manager *m, Unit *u, int type, bool success) { |
e537352b | 3340 | |
349cc4a5 | 3341 | #if HAVE_AUDIT |
2ba11090 | 3342 | _cleanup_free_ char *p = NULL; |
0aa281df | 3343 | const char *msg; |
7410616c | 3344 | int audit_fd, r; |
e537352b | 3345 | |
463d0d15 | 3346 | if (!MANAGER_IS_SYSTEM(m)) |
a1a078ee LP |
3347 | return; |
3348 | ||
c1165f82 LP |
3349 | audit_fd = get_audit_fd(); |
3350 | if (audit_fd < 0) | |
e537352b LP |
3351 | return; |
3352 | ||
bbd3a7ba LP |
3353 | /* Don't generate audit events if the service was already |
3354 | * started and we're just deserializing */ | |
2c289ea8 | 3355 | if (MANAGER_IS_RELOADING(m)) |
bbd3a7ba LP |
3356 | return; |
3357 | ||
7410616c LP |
3358 | r = unit_name_to_prefix_and_instance(u->id, &p); |
3359 | if (r < 0) { | |
d52b8493 | 3360 | log_warning_errno(r, "Failed to extract prefix and instance of unit name, ignoring: %m"); |
e537352b LP |
3361 | return; |
3362 | } | |
3363 | ||
63c372cb | 3364 | msg = strjoina("unit=", p); |
0aa281df | 3365 | if (audit_log_user_comm_message(audit_fd, type, msg, "systemd", NULL, NULL, NULL, success) < 0) { |
d52b8493 LP |
3366 | if (ERRNO_IS_PRIVILEGE(errno)) { |
3367 | /* We aren't allowed to send audit messages? Then let's not retry again. */ | |
3368 | log_debug_errno(errno, "Failed to send audit message, closing audit socket: %m"); | |
c1165f82 | 3369 | close_audit_fd(); |
d52b8493 LP |
3370 | } else |
3371 | log_warning_errno(errno, "Failed to send audit message, ignoring: %m"); | |
391ade86 | 3372 | } |
4927fcae | 3373 | #endif |
e537352b | 3374 | |
e537352b LP |
3375 | } |
3376 | ||
e983b760 | 3377 | void manager_send_unit_plymouth(Manager *m, Unit *u) { |
fc2fffe7 | 3378 | static const union sockaddr_union sa = PLYMOUTH_SOCKET; |
2ba11090 | 3379 | _cleanup_free_ char *message = NULL; |
254d1313 | 3380 | _cleanup_close_ int fd = -EBADF; |
fc2fffe7 | 3381 | int n = 0; |
e983b760 LP |
3382 | |
3383 | /* Don't generate plymouth events if the service was already | |
3384 | * started and we're just deserializing */ | |
2c289ea8 | 3385 | if (MANAGER_IS_RELOADING(m)) |
e983b760 LP |
3386 | return; |
3387 | ||
463d0d15 | 3388 | if (!MANAGER_IS_SYSTEM(m)) |
e983b760 LP |
3389 | return; |
3390 | ||
75f86906 | 3391 | if (detect_container() > 0) |
3772995a LP |
3392 | return; |
3393 | ||
b2bfd121 | 3394 | if (!UNIT_VTABLE(u)->notify_plymouth) |
e983b760 LP |
3395 | return; |
3396 | ||
3397 | /* We set SOCK_NONBLOCK here so that we rather drop the | |
3398 | * message then wait for plymouth */ | |
e62d8c39 ZJS |
3399 | fd = socket(AF_UNIX, SOCK_STREAM|SOCK_CLOEXEC|SOCK_NONBLOCK, 0); |
3400 | if (fd < 0) { | |
56f64d95 | 3401 | log_error_errno(errno, "socket() failed: %m"); |
e983b760 LP |
3402 | return; |
3403 | } | |
3404 | ||
fc2fffe7 | 3405 | if (connect(fd, &sa.sa, SOCKADDR_UN_LEN(sa.un)) < 0) { |
f60a028a | 3406 | if (!IN_SET(errno, EAGAIN, ENOENT) && !ERRNO_IS_DISCONNECT(errno)) |
56f64d95 | 3407 | log_error_errno(errno, "connect() failed: %m"); |
2ba11090 | 3408 | return; |
e983b760 LP |
3409 | } |
3410 | ||
305757d8 ZJS |
3411 | if (asprintf(&message, "U\002%c%s%n", (int) (strlen(u->id) + 1), u->id, &n) < 0) |
3412 | return (void) log_oom(); | |
e983b760 LP |
3413 | |
3414 | errno = 0; | |
2ba11090 | 3415 | if (write(fd, message, n + 1) != n + 1) |
f60a028a | 3416 | if (!IN_SET(errno, EAGAIN, ENOENT) && !ERRNO_IS_DISCONNECT(errno)) |
56f64d95 | 3417 | log_error_errno(errno, "Failed to write Plymouth message: %m"); |
e983b760 LP |
3418 | } |
3419 | ||
986935cf FB |
3420 | usec_t manager_get_watchdog(Manager *m, WatchdogType t) { |
3421 | assert(m); | |
3422 | ||
3423 | if (MANAGER_IS_USER(m)) | |
3424 | return USEC_INFINITY; | |
3425 | ||
902ea119 | 3426 | if (m->watchdog_overridden[t] != USEC_INFINITY) |
986935cf FB |
3427 | return m->watchdog_overridden[t]; |
3428 | ||
3429 | return m->watchdog[t]; | |
3430 | } | |
3431 | ||
3432 | void manager_set_watchdog(Manager *m, WatchdogType t, usec_t timeout) { | |
986935cf FB |
3433 | |
3434 | assert(m); | |
3435 | ||
3436 | if (MANAGER_IS_USER(m)) | |
3437 | return; | |
3438 | ||
3439 | if (m->watchdog[t] == timeout) | |
3440 | return; | |
3441 | ||
902ea119 CK |
3442 | if (m->watchdog_overridden[t] == USEC_INFINITY) { |
3443 | if (t == WATCHDOG_RUNTIME) | |
f16890f8 | 3444 | (void) watchdog_setup(timeout); |
902ea119 | 3445 | else if (t == WATCHDOG_PRETIMEOUT) |
5717062e | 3446 | (void) watchdog_setup_pretimeout(timeout); |
902ea119 | 3447 | } |
986935cf | 3448 | |
61927b9f | 3449 | m->watchdog[t] = timeout; |
986935cf FB |
3450 | } |
3451 | ||
aac47032 | 3452 | void manager_override_watchdog(Manager *m, WatchdogType t, usec_t timeout) { |
902ea119 | 3453 | usec_t usec; |
986935cf FB |
3454 | |
3455 | assert(m); | |
3456 | ||
3457 | if (MANAGER_IS_USER(m)) | |
aac47032 | 3458 | return; |
986935cf FB |
3459 | |
3460 | if (m->watchdog_overridden[t] == timeout) | |
aac47032 | 3461 | return; |
986935cf | 3462 | |
9b5560f3 | 3463 | usec = timeout == USEC_INFINITY ? m->watchdog[t] : timeout; |
902ea119 | 3464 | if (t == WATCHDOG_RUNTIME) |
f16890f8 | 3465 | (void) watchdog_setup(usec); |
902ea119 CK |
3466 | else if (t == WATCHDOG_PRETIMEOUT) |
3467 | (void) watchdog_setup_pretimeout(usec); | |
986935cf | 3468 | |
61927b9f | 3469 | m->watchdog_overridden[t] = timeout; |
986935cf FB |
3470 | } |
3471 | ||
aff3a9e1 LB |
3472 | int manager_set_watchdog_pretimeout_governor(Manager *m, const char *governor) { |
3473 | _cleanup_free_ char *p = NULL; | |
3474 | int r; | |
3475 | ||
3476 | assert(m); | |
3477 | ||
3478 | if (MANAGER_IS_USER(m)) | |
3479 | return 0; | |
3480 | ||
3481 | if (streq_ptr(m->watchdog_pretimeout_governor, governor)) | |
3482 | return 0; | |
3483 | ||
3484 | p = strdup(governor); | |
3485 | if (!p) | |
3486 | return -ENOMEM; | |
3487 | ||
3488 | r = watchdog_setup_pretimeout_governor(governor); | |
3489 | if (r < 0) | |
3490 | return r; | |
3491 | ||
3492 | return free_and_replace(m->watchdog_pretimeout_governor, p); | |
3493 | } | |
3494 | ||
3495 | int manager_override_watchdog_pretimeout_governor(Manager *m, const char *governor) { | |
3496 | _cleanup_free_ char *p = NULL; | |
3497 | int r; | |
3498 | ||
3499 | assert(m); | |
3500 | ||
3501 | if (MANAGER_IS_USER(m)) | |
3502 | return 0; | |
3503 | ||
3504 | if (streq_ptr(m->watchdog_pretimeout_governor_overridden, governor)) | |
3505 | return 0; | |
3506 | ||
3507 | p = strdup(governor); | |
3508 | if (!p) | |
3509 | return -ENOMEM; | |
3510 | ||
3511 | r = watchdog_setup_pretimeout_governor(governor); | |
3512 | if (r < 0) | |
3513 | return r; | |
3514 | ||
3515 | return free_and_replace(m->watchdog_pretimeout_governor_overridden, p); | |
3516 | } | |
3517 | ||
a16e1123 | 3518 | int manager_reload(Manager *m) { |
d7ac0952 | 3519 | _unused_ _cleanup_(manager_reloading_stopp) Manager *reloading = NULL; |
51d122af | 3520 | _cleanup_fdset_free_ FDSet *fds = NULL; |
6a33af40 LP |
3521 | _cleanup_fclose_ FILE *f = NULL; |
3522 | int r; | |
a16e1123 LP |
3523 | |
3524 | assert(m); | |
3525 | ||
07719a21 LP |
3526 | r = manager_open_serialization(m, &f); |
3527 | if (r < 0) | |
6a33af40 | 3528 | return log_error_errno(r, "Failed to create serialization file: %m"); |
38c52d46 | 3529 | |
07719a21 | 3530 | fds = fdset_new(); |
6a33af40 LP |
3531 | if (!fds) |
3532 | return log_oom(); | |
3533 | ||
d147e2b6 ZJS |
3534 | /* We are officially in reload mode from here on. */ |
3535 | reloading = manager_reloading_start(m); | |
a16e1123 | 3536 | |
b3680f49 | 3537 | r = manager_serialize(m, f, fds, false); |
d147e2b6 | 3538 | if (r < 0) |
d68c645b | 3539 | return r; |
a16e1123 | 3540 | |
d147e2b6 ZJS |
3541 | if (fseeko(f, 0, SEEK_SET) < 0) |
3542 | return log_error_errno(errno, "Failed to seek to beginning of serialization: %m"); | |
a16e1123 | 3543 | |
6a33af40 | 3544 | /* 💀 This is the point of no return, from here on there is no way back. 💀 */ |
d147e2b6 | 3545 | reloading = NULL; |
6a33af40 LP |
3546 | |
3547 | bus_manager_send_reloading(m, true); | |
3548 | ||
3549 | /* Start by flushing out all jobs and units, all generated units, all runtime environments, all dynamic users | |
3550 | * and everything else that is worth flushing out. We'll get it all back from the serialization — if we need | |
7802194a | 3551 | * it. */ |
6a33af40 | 3552 | |
a16e1123 | 3553 | manager_clear_jobs_and_units(m); |
07a78643 | 3554 | lookup_paths_flush_generator(&m->lookup_paths); |
84e3543e | 3555 | lookup_paths_free(&m->lookup_paths); |
e76506b7 | 3556 | exec_shared_runtime_vacuum(m); |
29206d46 | 3557 | dynamic_user_vacuum(m, false); |
00d9ef85 LP |
3558 | m->uid_refs = hashmap_free(m->uid_refs); |
3559 | m->gid_refs = hashmap_free(m->gid_refs); | |
2ded0c04 | 3560 | |
4870133b | 3561 | r = lookup_paths_init_or_warn(&m->lookup_paths, m->runtime_scope, 0, NULL); |
6a33af40 | 3562 | if (r < 0) |
99aad9a2 | 3563 | return r; |
5a1e9937 | 3564 | |
6a33af40 LP |
3565 | (void) manager_run_environment_generators(m); |
3566 | (void) manager_run_generators(m); | |
64691d20 | 3567 | |
581fef8d | 3568 | lookup_paths_log(&m->lookup_paths); |
07719a21 | 3569 | |
91e0ee5f | 3570 | /* We flushed out generated files, for which we don't watch mtime, so we should flush the old map. */ |
e8630e69 | 3571 | manager_free_unit_name_maps(m); |
a82b8b3d | 3572 | m->unit_file_state_outdated = false; |
5a1e9937 | 3573 | |
6a33af40 | 3574 | /* First, enumerate what we can from kernel and suchlike */ |
3ad2afb6 | 3575 | manager_enumerate_perpetual(m); |
ba64af90 | 3576 | manager_enumerate(m); |
a16e1123 LP |
3577 | |
3578 | /* Second, deserialize our stored data */ | |
6a33af40 LP |
3579 | r = manager_deserialize(m, f, fds); |
3580 | if (r < 0) | |
3581 | log_warning_errno(r, "Deserialization failed, proceeding anyway: %m"); | |
a16e1123 | 3582 | |
6a33af40 | 3583 | /* We don't need the serialization anymore */ |
62b0cbb3 | 3584 | f = safe_fclose(f); |
a16e1123 | 3585 | |
6a33af40 LP |
3586 | /* Re-register notify_fd as event source, and set up other sockets/communication channels we might need */ |
3587 | (void) manager_setup_notify(m); | |
3588 | (void) manager_setup_cgroups_agent(m); | |
3589 | (void) manager_setup_user_lookup_fd(m); | |
00d9ef85 | 3590 | |
a16e1123 | 3591 | /* Third, fire things up! */ |
007c6337 | 3592 | manager_coldplug(m); |
a16e1123 | 3593 | |
5197be06 LP |
3594 | /* Clean up runtime objects no longer referenced */ |
3595 | manager_vacuum(m); | |
e8a565cb | 3596 | |
3deed59a AA |
3597 | /* Clean up deserialized tracked clients */ |
3598 | m->deserialized_subscribed = strv_free(m->deserialized_subscribed); | |
3599 | ||
6a33af40 | 3600 | /* Consider the reload process complete now. */ |
31dc1ca3 LP |
3601 | assert(m->n_reloading > 0); |
3602 | m->n_reloading--; | |
3603 | ||
5ce5e1ad | 3604 | manager_ready(m); |
8936a5e3 | 3605 | |
71445ae7 | 3606 | m->send_reloading_done = true; |
6a33af40 | 3607 | return 0; |
a16e1123 LP |
3608 | } |
3609 | ||
fdf20a31 | 3610 | void manager_reset_failed(Manager *m) { |
5632e374 | 3611 | Unit *u; |
5632e374 LP |
3612 | |
3613 | assert(m); | |
3614 | ||
90e74a66 | 3615 | HASHMAP_FOREACH(u, m->units) |
fdf20a31 | 3616 | unit_reset_failed(u); |
5632e374 LP |
3617 | } |
3618 | ||
31afa0a4 | 3619 | bool manager_unit_inactive_or_pending(Manager *m, const char *name) { |
8f6df3fa LP |
3620 | Unit *u; |
3621 | ||
3622 | assert(m); | |
3623 | assert(name); | |
3624 | ||
3625 | /* Returns true if the unit is inactive or going down */ | |
bd0af849 ZJS |
3626 | u = manager_get_unit(m, name); |
3627 | if (!u) | |
8f6df3fa LP |
3628 | return true; |
3629 | ||
31afa0a4 | 3630 | return unit_inactive_or_pending(u); |
8f6df3fa LP |
3631 | } |
3632 | ||
d8eb10d6 ZJS |
3633 | static void log_taint_string(Manager *m) { |
3634 | _cleanup_free_ char *taint = NULL; | |
3635 | ||
3636 | assert(m); | |
3637 | ||
3638 | if (MANAGER_IS_USER(m) || m->taint_logged) | |
3639 | return; | |
3640 | ||
3641 | m->taint_logged = true; /* only check for taint once */ | |
3642 | ||
3643 | taint = manager_taint_string(m); | |
3644 | if (isempty(taint)) | |
3645 | return; | |
3646 | ||
3647 | log_struct(LOG_NOTICE, | |
3648 | LOG_MESSAGE("System is tainted: %s", taint), | |
3649 | "TAINT=%s", taint, | |
a1230ff9 | 3650 | "MESSAGE_ID=" SD_MESSAGE_TAINTED_STR); |
d8eb10d6 ZJS |
3651 | } |
3652 | ||
56dacdbc | 3653 | static void manager_notify_finished(Manager *m) { |
915b3753 | 3654 | usec_t firmware_usec, loader_usec, kernel_usec, initrd_usec, userspace_usec, total_usec; |
b0c918b9 | 3655 | |
638cece4 | 3656 | if (MANAGER_IS_TEST_RUN(m)) |
b0c918b9 LP |
3657 | return; |
3658 | ||
463d0d15 | 3659 | if (MANAGER_IS_SYSTEM(m) && detect_container() <= 0) { |
dc3c9f5e ZJS |
3660 | char buf[FORMAT_TIMESPAN_MAX + STRLEN(" (firmware) + ") + FORMAT_TIMESPAN_MAX + STRLEN(" (loader) + ")] |
3661 | = {}; | |
3662 | char *p = buf; | |
3663 | size_t size = sizeof buf; | |
e03ae661 | 3664 | |
9f9f0342 LP |
3665 | /* Note that MANAGER_TIMESTAMP_KERNEL's monotonic value is always at 0, and |
3666 | * MANAGER_TIMESTAMP_FIRMWARE's and MANAGER_TIMESTAMP_LOADER's monotonic value should be considered | |
915b3753 LP |
3667 | * negative values. */ |
3668 | ||
9f9f0342 LP |
3669 | firmware_usec = m->timestamps[MANAGER_TIMESTAMP_FIRMWARE].monotonic - m->timestamps[MANAGER_TIMESTAMP_LOADER].monotonic; |
3670 | loader_usec = m->timestamps[MANAGER_TIMESTAMP_LOADER].monotonic - m->timestamps[MANAGER_TIMESTAMP_KERNEL].monotonic; | |
3671 | userspace_usec = m->timestamps[MANAGER_TIMESTAMP_FINISH].monotonic - m->timestamps[MANAGER_TIMESTAMP_USERSPACE].monotonic; | |
3672 | total_usec = m->timestamps[MANAGER_TIMESTAMP_FIRMWARE].monotonic + m->timestamps[MANAGER_TIMESTAMP_FINISH].monotonic; | |
18fa6b27 | 3673 | |
dd1db3c2 | 3674 | if (firmware_usec > 0) |
5291f26d | 3675 | size = strpcpyf(&p, size, "%s (firmware) + ", FORMAT_TIMESPAN(firmware_usec, USEC_PER_MSEC)); |
dd1db3c2 | 3676 | if (loader_usec > 0) |
5291f26d | 3677 | size = strpcpyf(&p, size, "%s (loader) + ", FORMAT_TIMESPAN(loader_usec, USEC_PER_MSEC)); |
dd1db3c2 | 3678 | |
9f9f0342 | 3679 | if (dual_timestamp_is_set(&m->timestamps[MANAGER_TIMESTAMP_INITRD])) { |
18fa6b27 | 3680 | |
7802194a | 3681 | /* The initrd case on bare-metal */ |
9f9f0342 LP |
3682 | kernel_usec = m->timestamps[MANAGER_TIMESTAMP_INITRD].monotonic - m->timestamps[MANAGER_TIMESTAMP_KERNEL].monotonic; |
3683 | initrd_usec = m->timestamps[MANAGER_TIMESTAMP_USERSPACE].monotonic - m->timestamps[MANAGER_TIMESTAMP_INITRD].monotonic; | |
18fa6b27 | 3684 | |
e12919e8 | 3685 | log_struct(LOG_INFO, |
2b044526 | 3686 | "MESSAGE_ID=" SD_MESSAGE_STARTUP_FINISHED_STR, |
e12919e8 LP |
3687 | "KERNEL_USEC="USEC_FMT, kernel_usec, |
3688 | "INITRD_USEC="USEC_FMT, initrd_usec, | |
3689 | "USERSPACE_USEC="USEC_FMT, userspace_usec, | |
dd1db3c2 YW |
3690 | LOG_MESSAGE("Startup finished in %s%s (kernel) + %s (initrd) + %s (userspace) = %s.", |
3691 | buf, | |
5291f26d ZJS |
3692 | FORMAT_TIMESPAN(kernel_usec, USEC_PER_MSEC), |
3693 | FORMAT_TIMESPAN(initrd_usec, USEC_PER_MSEC), | |
3694 | FORMAT_TIMESPAN(userspace_usec, USEC_PER_MSEC), | |
3695 | FORMAT_TIMESPAN(total_usec, USEC_PER_MSEC))); | |
18fa6b27 | 3696 | } else { |
7802194a | 3697 | /* The initrd-less case on bare-metal */ |
9f9f0342 LP |
3698 | |
3699 | kernel_usec = m->timestamps[MANAGER_TIMESTAMP_USERSPACE].monotonic - m->timestamps[MANAGER_TIMESTAMP_KERNEL].monotonic; | |
18fa6b27 LP |
3700 | initrd_usec = 0; |
3701 | ||
81270860 | 3702 | log_struct(LOG_INFO, |
2b044526 | 3703 | "MESSAGE_ID=" SD_MESSAGE_STARTUP_FINISHED_STR, |
e12919e8 | 3704 | "KERNEL_USEC="USEC_FMT, kernel_usec, |
ccd06097 | 3705 | "USERSPACE_USEC="USEC_FMT, userspace_usec, |
dd1db3c2 YW |
3706 | LOG_MESSAGE("Startup finished in %s%s (kernel) + %s (userspace) = %s.", |
3707 | buf, | |
5291f26d ZJS |
3708 | FORMAT_TIMESPAN(kernel_usec, USEC_PER_MSEC), |
3709 | FORMAT_TIMESPAN(userspace_usec, USEC_PER_MSEC), | |
3710 | FORMAT_TIMESPAN(total_usec, USEC_PER_MSEC))); | |
e12919e8 LP |
3711 | } |
3712 | } else { | |
4adf314b | 3713 | /* The container and --user case */ |
e12919e8 | 3714 | firmware_usec = loader_usec = initrd_usec = kernel_usec = 0; |
9f9f0342 | 3715 | total_usec = userspace_usec = m->timestamps[MANAGER_TIMESTAMP_FINISH].monotonic - m->timestamps[MANAGER_TIMESTAMP_USERSPACE].monotonic; |
e12919e8 LP |
3716 | |
3717 | log_struct(LOG_INFO, | |
2b044526 | 3718 | "MESSAGE_ID=" SD_MESSAGE_USER_STARTUP_FINISHED_STR, |
e12919e8 | 3719 | "USERSPACE_USEC="USEC_FMT, userspace_usec, |
e2cc6eca | 3720 | LOG_MESSAGE("Startup finished in %s.", |
5291f26d | 3721 | FORMAT_TIMESPAN(total_usec, USEC_PER_MSEC))); |
18fa6b27 | 3722 | } |
b0c918b9 | 3723 | |
718db961 | 3724 | bus_manager_send_finished(m, firmware_usec, loader_usec, kernel_usec, initrd_usec, userspace_usec, total_usec); |
530345e7 | 3725 | |
d8eb10d6 | 3726 | log_taint_string(m); |
b0c918b9 LP |
3727 | } |
3728 | ||
6d932659 | 3729 | static void user_manager_send_ready(Manager *m) { |
4bf4f50f ZJS |
3730 | int r; |
3731 | ||
4adf314b LP |
3732 | assert(m); |
3733 | ||
3734 | /* We send READY=1 on reaching basic.target only when running in --user mode. */ | |
3735 | if (!MANAGER_IS_USER(m) || m->ready_sent) | |
3736 | return; | |
3737 | ||
d4341b76 ZJS |
3738 | r = sd_notify(false, |
3739 | "READY=1\n" | |
3740 | "STATUS=Reached " SPECIAL_BASIC_TARGET "."); | |
4bf4f50f ZJS |
3741 | if (r < 0) |
3742 | log_warning_errno(r, "Failed to send readiness notification, ignoring: %m"); | |
3743 | ||
6d932659 ZJS |
3744 | m->ready_sent = true; |
3745 | m->status_ready = false; | |
3746 | } | |
3747 | ||
3748 | static void manager_send_ready(Manager *m) { | |
4bf4f50f ZJS |
3749 | int r; |
3750 | ||
6d932659 ZJS |
3751 | if (m->ready_sent && m->status_ready) |
3752 | /* Skip the notification if nothing changed. */ | |
3753 | return; | |
3754 | ||
028f7d3a ZJS |
3755 | r = sd_notify(false, |
3756 | "READY=1\n" | |
3757 | "STATUS=Ready."); | |
4bf4f50f ZJS |
3758 | if (r < 0) |
3759 | log_full_errno(m->ready_sent ? LOG_DEBUG : LOG_WARNING, r, | |
3760 | "Failed to send readiness notification, ignoring: %m"); | |
3761 | ||
6d932659 | 3762 | m->ready_sent = m->status_ready = true; |
4adf314b LP |
3763 | } |
3764 | ||
3765 | static void manager_check_basic_target(Manager *m) { | |
3766 | Unit *u; | |
3767 | ||
3768 | assert(m); | |
3769 | ||
3770 | /* Small shortcut */ | |
3771 | if (m->ready_sent && m->taint_logged) | |
3772 | return; | |
3773 | ||
3774 | u = manager_get_unit(m, SPECIAL_BASIC_TARGET); | |
3775 | if (!u || !UNIT_IS_ACTIVE_OR_RELOADING(unit_active_state(u))) | |
3776 | return; | |
3777 | ||
3778 | /* For user managers, send out READY=1 as soon as we reach basic.target */ | |
6d932659 | 3779 | user_manager_send_ready(m); |
4adf314b LP |
3780 | |
3781 | /* Log the taint string as soon as we reach basic.target */ | |
3782 | log_taint_string(m); | |
3783 | } | |
3784 | ||
56dacdbc | 3785 | void manager_check_finished(Manager *m) { |
56dacdbc ZJS |
3786 | assert(m); |
3787 | ||
2c289ea8 | 3788 | if (MANAGER_IS_RELOADING(m)) |
aad1976f LP |
3789 | return; |
3790 | ||
4259d202 LP |
3791 | /* Verify that we have entered the event loop already, and not left it again. */ |
3792 | if (!MANAGER_IS_RUNNING(m)) | |
9771b62d LP |
3793 | return; |
3794 | ||
4adf314b | 3795 | manager_check_basic_target(m); |
0c2826c6 | 3796 | |
56dacdbc | 3797 | if (hashmap_size(m->jobs) > 0) { |
56dacdbc | 3798 | if (m->jobs_in_progress_event_source) |
2ae56591 | 3799 | /* Ignore any failure, this is only for feedback */ |
1b4154a8 ZJS |
3800 | (void) sd_event_source_set_time(m->jobs_in_progress_event_source, |
3801 | manager_watch_jobs_next_time(m)); | |
56dacdbc ZJS |
3802 | return; |
3803 | } | |
3804 | ||
a4ac27c1 ZJS |
3805 | /* The jobs hashmap tends to grow a lot during boot, and then it's not reused until shutdown. Let's |
3806 | kill the hashmap if it is relatively large. */ | |
3807 | if (hashmap_buckets(m->jobs) > hashmap_size(m->units) / 10) | |
3808 | m->jobs = hashmap_free(m->jobs); | |
3809 | ||
6d932659 ZJS |
3810 | manager_send_ready(m); |
3811 | ||
9c1b17c3 YW |
3812 | /* Notify Type=idle units that we are done now */ |
3813 | manager_close_idle_pipe(m); | |
3814 | ||
6d932659 ZJS |
3815 | if (MANAGER_IS_FINISHED(m)) |
3816 | return; | |
3817 | ||
7365a296 | 3818 | manager_flip_auto_status(m, false, "boot finished"); |
56dacdbc | 3819 | |
56dacdbc | 3820 | /* Turn off confirm spawn now */ |
7d5ceb64 | 3821 | m->confirm_spawn = NULL; |
56dacdbc ZJS |
3822 | |
3823 | /* No need to update ask password status when we're going non-interactive */ | |
3824 | manager_close_ask_password(m); | |
3825 | ||
3826 | /* This is no longer the first boot */ | |
3827 | manager_set_first_boot(m, false); | |
3828 | ||
9f9f0342 | 3829 | dual_timestamp_get(m->timestamps + MANAGER_TIMESTAMP_FINISH); |
56dacdbc ZJS |
3830 | |
3831 | manager_notify_finished(m); | |
3832 | ||
e7ab4d1a | 3833 | manager_invalidate_startup_units(m); |
56dacdbc ZJS |
3834 | } |
3835 | ||
dd0ab174 LP |
3836 | void manager_send_reloading(Manager *m) { |
3837 | assert(m); | |
3838 | ||
3839 | /* Let whoever invoked us know that we are now reloading */ | |
3840 | (void) sd_notifyf(/* unset= */ false, | |
3841 | "RELOADING=1\n" | |
3842 | "MONOTONIC_USEC=" USEC_FMT "\n", now(CLOCK_MONOTONIC)); | |
3843 | ||
3844 | /* And ensure that we'll send READY=1 again as soon as we are ready again */ | |
3845 | m->ready_sent = false; | |
3846 | } | |
3847 | ||
64691d20 | 3848 | static bool generator_path_any(const char* const* paths) { |
64691d20 ZJS |
3849 | bool found = false; |
3850 | ||
3851 | /* Optimize by skipping the whole process by not creating output directories | |
3852 | * if no generators are found. */ | |
2034c8b8 | 3853 | STRV_FOREACH(path, paths) |
64691d20 ZJS |
3854 | if (access(*path, F_OK) == 0) |
3855 | found = true; | |
3856 | else if (errno != ENOENT) | |
3857 | log_warning_errno(errno, "Failed to open generator directory %s: %m", *path); | |
3858 | ||
3859 | return found; | |
3860 | } | |
3861 | ||
64691d20 ZJS |
3862 | static int manager_run_environment_generators(Manager *m) { |
3863 | char **tmp = NULL; /* this is only used in the forked process, no cleanup here */ | |
cccf5703 | 3864 | _cleanup_strv_free_ char **paths = NULL; |
1ad6e8b3 LP |
3865 | void* args[] = { |
3866 | [STDOUT_GENERATE] = &tmp, | |
3867 | [STDOUT_COLLECT] = &tmp, | |
3868 | [STDOUT_CONSUME] = &m->transient_environment, | |
3869 | }; | |
e3b8d063 | 3870 | int r; |
64691d20 | 3871 | |
638cece4 | 3872 | if (MANAGER_IS_TEST_RUN(m) && !(m->test_run_flags & MANAGER_TEST_RUN_ENV_GENERATORS)) |
e0a3da1f ZJS |
3873 | return 0; |
3874 | ||
361cacf4 | 3875 | paths = env_generator_binary_paths(m->runtime_scope); |
cccf5703 BB |
3876 | if (!paths) |
3877 | return log_oom(); | |
64691d20 | 3878 | |
cccf5703 | 3879 | if (!generator_path_any((const char* const*) paths)) |
64691d20 ZJS |
3880 | return 0; |
3881 | ||
2053593f | 3882 | WITH_UMASK(0022) |
cccf5703 | 3883 | r = execute_directories((const char* const*) paths, DEFAULT_TIMEOUT_USEC, gather_environment, |
43f565c6 YW |
3884 | args, NULL, m->transient_environment, |
3885 | EXEC_DIR_PARALLEL | EXEC_DIR_IGNORE_ERRORS | EXEC_DIR_SET_SYSTEMD_EXEC_PID); | |
e3b8d063 | 3886 | return r; |
64691d20 ZJS |
3887 | } |
3888 | ||
82c5db16 LP |
3889 | static int build_generator_environment(Manager *m, char ***ret) { |
3890 | _cleanup_strv_free_ char **nl = NULL; | |
3891 | Virtualization v; | |
08951245 | 3892 | ConfidentialVirtualization cv; |
82c5db16 LP |
3893 | int r; |
3894 | ||
3895 | assert(m); | |
3896 | assert(ret); | |
3897 | ||
3898 | /* Generators oftentimes want to know some basic facts about the environment they run in, in order to | |
3899 | * adjust generated units to that. Let's pass down some bits of information that are easy for us to | |
3900 | * determine (but a bit harder for generator scripts to determine), as environment variables. */ | |
3901 | ||
3902 | nl = strv_copy(m->transient_environment); | |
3903 | if (!nl) | |
3904 | return -ENOMEM; | |
3905 | ||
4870133b | 3906 | r = strv_env_assign(&nl, "SYSTEMD_SCOPE", runtime_scope_to_string(m->runtime_scope)); |
82c5db16 LP |
3907 | if (r < 0) |
3908 | return r; | |
3909 | ||
3910 | if (MANAGER_IS_SYSTEM(m)) { | |
3911 | /* Note that $SYSTEMD_IN_INITRD may be used to override the initrd detection in much of our | |
3912 | * codebase. This is hence more than purely informational. It will shortcut detection of the | |
3913 | * initrd state if generators invoke our own tools. But that's OK, as it would come to the | |
3914 | * same results (hopefully). */ | |
3915 | r = strv_env_assign(&nl, "SYSTEMD_IN_INITRD", one_zero(in_initrd())); | |
3916 | if (r < 0) | |
3917 | return r; | |
3918 | ||
3919 | if (m->first_boot >= 0) { | |
3920 | r = strv_env_assign(&nl, "SYSTEMD_FIRST_BOOT", one_zero(m->first_boot)); | |
3921 | if (r < 0) | |
3922 | return r; | |
3923 | } | |
3924 | } | |
3925 | ||
3926 | v = detect_virtualization(); | |
3927 | if (v < 0) | |
3928 | log_debug_errno(v, "Failed to detect virtualization, ignoring: %m"); | |
3929 | else if (v > 0) { | |
3930 | const char *s; | |
3931 | ||
3932 | s = strjoina(VIRTUALIZATION_IS_VM(v) ? "vm:" : | |
3933 | VIRTUALIZATION_IS_CONTAINER(v) ? "container:" : ":", | |
3934 | virtualization_to_string(v)); | |
3935 | ||
3936 | r = strv_env_assign(&nl, "SYSTEMD_VIRTUALIZATION", s); | |
3937 | if (r < 0) | |
3938 | return r; | |
3939 | } | |
3940 | ||
08951245 DB |
3941 | cv = detect_confidential_virtualization(); |
3942 | if (cv < 0) | |
3943 | log_debug_errno(cv, "Failed to detect confidential virtualization, ignoring: %m"); | |
3944 | else if (cv > 0) { | |
3945 | r = strv_env_assign(&nl, "SYSTEMD_CONFIDENTIAL_VIRTUALIZATION", confidential_virtualization_to_string(cv)); | |
3946 | if (r < 0) | |
3947 | return r; | |
3948 | } | |
3949 | ||
82c5db16 LP |
3950 | r = strv_env_assign(&nl, "SYSTEMD_ARCHITECTURE", architecture_to_string(uname_architecture())); |
3951 | if (r < 0) | |
3952 | return r; | |
3953 | ||
3954 | *ret = TAKE_PTR(nl); | |
3955 | return 0; | |
3956 | } | |
3957 | ||
ca6ce62d ZJS |
3958 | static int manager_execute_generators(Manager *m, char **paths, bool remount_ro) { |
3959 | _cleanup_strv_free_ char **ge = NULL; | |
3960 | const char *argv[] = { | |
3961 | NULL, /* Leave this empty, execute_directory() will fill something in */ | |
3962 | m->lookup_paths.generator, | |
3963 | m->lookup_paths.generator_early, | |
3964 | m->lookup_paths.generator_late, | |
3965 | NULL, | |
3966 | }; | |
3967 | int r; | |
3968 | ||
3969 | r = build_generator_environment(m, &ge); | |
3970 | if (r < 0) | |
3971 | return log_error_errno(r, "Failed to build generator environment: %m"); | |
3972 | ||
3973 | if (remount_ro) { | |
3974 | /* Remount most of the filesystem tree read-only. We leave /sys/ as-is, because our code | |
3975 | * checks whether it is read-only to detect containerized execution environments. We leave | |
3976 | * /run/ as-is too, because that's where our output goes. We also leave /proc/ and /dev/shm/ | |
3977 | * because they're API, and /tmp/ that safe_fork() mounted for us. | |
3978 | */ | |
3979 | r = bind_remount_recursive("/", MS_RDONLY, MS_RDONLY, | |
3980 | STRV_MAKE("/sys", "/run", "/proc", "/dev/shm", "/tmp")); | |
3981 | if (r < 0) | |
3982 | log_warning_errno(r, "Read-only bind remount failed, ignoring: %m"); | |
3983 | } | |
3984 | ||
3985 | BLOCK_WITH_UMASK(0022); | |
3986 | return execute_directories( | |
3987 | (const char* const*) paths, | |
3988 | DEFAULT_TIMEOUT_USEC, | |
3989 | /* callbacks= */ NULL, /* callback_args= */ NULL, | |
3990 | (char**) argv, | |
3991 | ge, | |
3992 | EXEC_DIR_PARALLEL | EXEC_DIR_IGNORE_ERRORS | EXEC_DIR_SET_SYSTEMD_EXEC_PID); | |
3993 | } | |
3994 | ||
e801700e | 3995 | static int manager_run_generators(Manager *m) { |
b8fba0cd | 3996 | ForkFlags flags = FORK_RESET_SIGNALS | FORK_WAIT | FORK_NEW_MOUNTNS | FORK_MOUNTNS_SLAVE; |
ca6ce62d | 3997 | _cleanup_strv_free_ char **paths = NULL; |
07719a21 | 3998 | int r; |
5a1e9937 LP |
3999 | |
4000 | assert(m); | |
4001 | ||
638cece4 | 4002 | if (MANAGER_IS_TEST_RUN(m) && !(m->test_run_flags & MANAGER_TEST_RUN_GENERATORS)) |
e0a3da1f ZJS |
4003 | return 0; |
4004 | ||
4870133b | 4005 | paths = generator_binary_paths(m->runtime_scope); |
e801700e ZJS |
4006 | if (!paths) |
4007 | return log_oom(); | |
5a1e9937 | 4008 | |
64691d20 ZJS |
4009 | if (!generator_path_any((const char* const*) paths)) |
4010 | return 0; | |
5a1e9937 | 4011 | |
cd64fd56 | 4012 | r = lookup_paths_mkdir_generator(&m->lookup_paths); |
7eb4f326 LP |
4013 | if (r < 0) { |
4014 | log_error_errno(r, "Failed to create generator directories: %m"); | |
07719a21 | 4015 | goto finish; |
7eb4f326 | 4016 | } |
5a1e9937 | 4017 | |
ca6ce62d ZJS |
4018 | /* If we are the system manager, we fork and invoke the generators in a sanitized mount namespace. If |
4019 | * we are the user manager, let's just execute the generators directly. We might not have the | |
4020 | * necessary privileges, and the system manager has already mounted /tmp/ and everything else for us. | |
4021 | */ | |
4022 | if (MANAGER_IS_USER(m)) { | |
4023 | r = manager_execute_generators(m, paths, /* remount_ro= */ false); | |
82c5db16 LP |
4024 | goto finish; |
4025 | } | |
5a1e9937 | 4026 | |
b8fba0cd LB |
4027 | /* On some systems /tmp/ doesn't exist, and on some other systems we cannot create it at all. Avoid |
4028 | * trying to mount a private tmpfs on it as there's no one size fits all. */ | |
4029 | if (is_dir("/tmp", /* follow= */ false) > 0) | |
4030 | flags |= FORK_PRIVATE_TMP; | |
4031 | ||
4032 | r = safe_fork("(sd-gens)", flags, NULL); | |
ca6ce62d ZJS |
4033 | if (r == 0) { |
4034 | r = manager_execute_generators(m, paths, /* remount_ro= */ true); | |
4035 | _exit(r >= 0 ? EXIT_SUCCESS : EXIT_FAILURE); | |
4036 | } | |
a2275dcb | 4037 | if (r < 0) { |
46801870 | 4038 | if (!ERRNO_IS_PRIVILEGE(r) && r != -EINVAL) { |
a2275dcb YW |
4039 | log_error_errno(r, "Failed to fork off sandboxing environment for executing generators: %m"); |
4040 | goto finish; | |
4041 | } | |
4042 | ||
4043 | /* Failed to fork with new mount namespace? Maybe, running in a container environment with | |
46801870 YW |
4044 | * seccomp or without capability. |
4045 | * | |
4046 | * We also allow -EINVAL to allow running without CLONE_NEWNS. | |
4047 | * | |
4048 | * Also, when running on non-native userland architecture via systemd-nspawn and | |
4049 | * qemu-user-static QEMU-emulator, clone() with CLONE_NEWNS fails with EINVAL, see | |
4050 | * https://github.com/systemd/systemd/issues/28901. | |
4051 | */ | |
a2275dcb YW |
4052 | log_debug_errno(r, |
4053 | "Failed to fork off sandboxing environment for executing generators. " | |
4054 | "Falling back to execute generators without sandboxing: %m"); | |
4055 | r = manager_execute_generators(m, paths, /* remount_ro= */ false); | |
4056 | } | |
5a1e9937 | 4057 | |
718db961 | 4058 | finish: |
cd64fd56 | 4059 | lookup_paths_trim_generator(&m->lookup_paths); |
e801700e | 4060 | return r; |
5a1e9937 LP |
4061 | } |
4062 | ||
1ad6e8b3 LP |
4063 | int manager_transient_environment_add(Manager *m, char **plus) { |
4064 | char **a; | |
4065 | ||
4066 | assert(m); | |
4067 | ||
4068 | if (strv_isempty(plus)) | |
4069 | return 0; | |
4070 | ||
4ab3d29f | 4071 | a = strv_env_merge(m->transient_environment, plus); |
1ad6e8b3 | 4072 | if (!a) |
2fbbbf9a | 4073 | return log_oom(); |
1ad6e8b3 LP |
4074 | |
4075 | sanitize_environment(a); | |
4076 | ||
4077 | return strv_free_and_replace(m->transient_environment, a); | |
4078 | } | |
4079 | ||
4080 | int manager_client_environment_modify( | |
4081 | Manager *m, | |
4082 | char **minus, | |
4083 | char **plus) { | |
4084 | ||
718db961 | 4085 | char **a = NULL, **b = NULL, **l; |
1ad6e8b3 | 4086 | |
97d0e5f8 | 4087 | assert(m); |
bcd8e6d1 | 4088 | |
1ad6e8b3 LP |
4089 | if (strv_isempty(minus) && strv_isempty(plus)) |
4090 | return 0; | |
4091 | ||
4092 | l = m->client_environment; | |
bcd8e6d1 | 4093 | |
718db961 LP |
4094 | if (!strv_isempty(minus)) { |
4095 | a = strv_env_delete(l, 1, minus); | |
4096 | if (!a) | |
4097 | return -ENOMEM; | |
4098 | ||
4099 | l = a; | |
4100 | } | |
4101 | ||
4102 | if (!strv_isempty(plus)) { | |
4ab3d29f | 4103 | b = strv_env_merge(l, plus); |
aa9f8a30 AH |
4104 | if (!b) { |
4105 | strv_free(a); | |
718db961 | 4106 | return -ENOMEM; |
aa9f8a30 | 4107 | } |
bcd8e6d1 | 4108 | |
718db961 LP |
4109 | l = b; |
4110 | } | |
4111 | ||
1ad6e8b3 LP |
4112 | if (m->client_environment != l) |
4113 | strv_free(m->client_environment); | |
4114 | ||
718db961 LP |
4115 | if (a != l) |
4116 | strv_free(a); | |
4117 | if (b != l) | |
4118 | strv_free(b); | |
4119 | ||
1ad6e8b3 LP |
4120 | m->client_environment = sanitize_environment(l); |
4121 | return 0; | |
4122 | } | |
4123 | ||
4124 | int manager_get_effective_environment(Manager *m, char ***ret) { | |
4125 | char **l; | |
4126 | ||
4127 | assert(m); | |
4128 | assert(ret); | |
4129 | ||
4ab3d29f | 4130 | l = strv_env_merge(m->transient_environment, m->client_environment); |
1ad6e8b3 LP |
4131 | if (!l) |
4132 | return -ENOMEM; | |
f069efb4 | 4133 | |
1ad6e8b3 | 4134 | *ret = l; |
97d0e5f8 UTL |
4135 | return 0; |
4136 | } | |
4137 | ||
bfb27b06 LP |
4138 | int manager_set_unit_defaults(Manager *m, const UnitDefaults *defaults) { |
4139 | _cleanup_free_ char *label = NULL; | |
4140 | struct rlimit *rlimit[_RLIMIT_MAX]; | |
4141 | int r; | |
4142 | ||
aa5ae971 | 4143 | assert(m); |
bfb27b06 | 4144 | assert(defaults); |
aa5ae971 | 4145 | |
bfb27b06 LP |
4146 | if (streq_ptr(defaults->smack_process_label, "/")) |
4147 | label = NULL; | |
4148 | else { | |
4149 | const char *l = defaults->smack_process_label; | |
aa5ae971 | 4150 | #ifdef SMACK_DEFAULT_PROCESS_LABEL |
bfb27b06 LP |
4151 | if (!l) |
4152 | l = SMACK_DEFAULT_PROCESS_LABEL; | |
aa5ae971 | 4153 | #endif |
bfb27b06 LP |
4154 | if (l) { |
4155 | label = strdup(l); | |
4156 | if (!label) | |
4157 | return -ENOMEM; | |
4158 | } else | |
4159 | label = NULL; | |
4160 | } | |
aa5ae971 | 4161 | |
bfb27b06 LP |
4162 | r = rlimit_copy_all(rlimit, defaults->rlimit); |
4163 | if (r < 0) | |
4164 | return r; | |
aa5ae971 | 4165 | |
bfb27b06 LP |
4166 | m->defaults.std_output = defaults->std_output; |
4167 | m->defaults.std_error = defaults->std_error; | |
c93ff2e9 | 4168 | |
bfb27b06 LP |
4169 | m->defaults.restart_usec = defaults->restart_usec; |
4170 | m->defaults.timeout_start_usec = defaults->timeout_start_usec; | |
4171 | m->defaults.timeout_stop_usec = defaults->timeout_stop_usec; | |
4172 | m->defaults.timeout_abort_usec = defaults->timeout_abort_usec; | |
4173 | m->defaults.timeout_abort_set = defaults->timeout_abort_set; | |
4174 | m->defaults.device_timeout_usec = defaults->device_timeout_usec; | |
d9814c76 | 4175 | |
bfb27b06 LP |
4176 | m->defaults.start_limit_interval = defaults->start_limit_interval; |
4177 | m->defaults.start_limit_burst = defaults->start_limit_burst; | |
c93ff2e9 | 4178 | |
bfb27b06 LP |
4179 | m->defaults.cpu_accounting = defaults->cpu_accounting; |
4180 | m->defaults.memory_accounting = defaults->memory_accounting; | |
4181 | m->defaults.io_accounting = defaults->io_accounting; | |
4182 | m->defaults.blockio_accounting = defaults->blockio_accounting; | |
4183 | m->defaults.tasks_accounting = defaults->tasks_accounting; | |
4184 | m->defaults.ip_accounting = defaults->ip_accounting; | |
4185 | ||
4186 | m->defaults.tasks_max = defaults->tasks_max; | |
4187 | m->defaults.timer_accuracy_usec = defaults->timer_accuracy_usec; | |
4188 | ||
4189 | m->defaults.oom_policy = defaults->oom_policy; | |
4190 | m->defaults.oom_score_adjust = defaults->oom_score_adjust; | |
4191 | m->defaults.oom_score_adjust_set = defaults->oom_score_adjust_set; | |
4192 | ||
4193 | m->defaults.memory_pressure_watch = defaults->memory_pressure_watch; | |
4194 | m->defaults.memory_pressure_threshold_usec = defaults->memory_pressure_threshold_usec; | |
4195 | ||
4196 | free_and_replace(m->defaults.smack_process_label, label); | |
4197 | rlimit_free_all(m->defaults.rlimit); | |
4198 | memcpy(m->defaults.rlimit, rlimit, sizeof(struct rlimit*) * _RLIMIT_MAX); | |
c93ff2e9 FC |
4199 | |
4200 | return 0; | |
4201 | } | |
4202 | ||
8559b3b7 LP |
4203 | void manager_recheck_dbus(Manager *m) { |
4204 | assert(m); | |
4205 | ||
8f41e6b6 ZJS |
4206 | /* Connects to the bus if the dbus service and socket are running. If we are running in user mode |
4207 | * this is all it does. In system mode we'll also connect to the system bus (which will most likely | |
4208 | * just reuse the connection of the API bus). That's because the system bus after all runs as service | |
4209 | * of the system instance, while in the user instance we can assume it's already there. */ | |
8559b3b7 | 4210 | |
31dc1ca3 LP |
4211 | if (MANAGER_IS_RELOADING(m)) |
4212 | return; /* don't check while we are reloading… */ | |
4213 | ||
2b680534 | 4214 | if (manager_dbus_is_running(m, false)) { |
8559b3b7 LP |
4215 | (void) bus_init_api(m); |
4216 | ||
4217 | if (MANAGER_IS_SYSTEM(m)) | |
4218 | (void) bus_init_system(m); | |
4219 | } else { | |
4220 | (void) bus_done_api(m); | |
4221 | ||
4222 | if (MANAGER_IS_SYSTEM(m)) | |
4223 | (void) bus_done_system(m); | |
4224 | } | |
4225 | } | |
4226 | ||
d075092f | 4227 | static bool manager_journal_is_running(Manager *m) { |
f1dd0c3f LP |
4228 | Unit *u; |
4229 | ||
4230 | assert(m); | |
4231 | ||
638cece4 | 4232 | if (MANAGER_IS_TEST_RUN(m)) |
7d814a19 LP |
4233 | return false; |
4234 | ||
d075092f | 4235 | /* If we are the user manager we can safely assume that the journal is up */ |
463d0d15 | 4236 | if (!MANAGER_IS_SYSTEM(m)) |
d075092f | 4237 | return true; |
f1dd0c3f | 4238 | |
d075092f | 4239 | /* Check that the socket is not only up, but in RUNNING state */ |
731a676c | 4240 | u = manager_get_unit(m, SPECIAL_JOURNALD_SOCKET); |
d075092f LP |
4241 | if (!u) |
4242 | return false; | |
4243 | if (SOCKET(u)->state != SOCKET_RUNNING) | |
4244 | return false; | |
f1dd0c3f | 4245 | |
d075092f | 4246 | /* Similar, check if the daemon itself is fully up, too */ |
731a676c | 4247 | u = manager_get_unit(m, SPECIAL_JOURNALD_SERVICE); |
d075092f LP |
4248 | if (!u) |
4249 | return false; | |
217677ab | 4250 | if (!IN_SET(SERVICE(u)->state, SERVICE_RELOAD, SERVICE_RUNNING)) |
d075092f LP |
4251 | return false; |
4252 | ||
4253 | return true; | |
4254 | } | |
4255 | ||
6123dfaa ZJS |
4256 | void disable_printk_ratelimit(void) { |
4257 | /* Disable kernel's printk ratelimit. | |
4258 | * | |
4259 | * Logging to /dev/kmsg is most useful during early boot and shutdown, where normal logging | |
4260 | * mechanisms are not available. The semantics of this sysctl are such that any kernel command-line | |
4261 | * setting takes precedence. */ | |
4262 | int r; | |
4263 | ||
4264 | r = sysctl_write("kernel/printk_devkmsg", "on"); | |
4265 | if (r < 0) | |
4266 | log_debug_errno(r, "Failed to set sysctl kernel.printk_devkmsg=on: %m"); | |
4267 | } | |
4268 | ||
d075092f LP |
4269 | void manager_recheck_journal(Manager *m) { |
4270 | ||
4271 | assert(m); | |
4272 | ||
4273 | /* Don't bother with this unless we are in the special situation of being PID 1 */ | |
4274 | if (getpid_cached() != 1) | |
731a676c | 4275 | return; |
f1dd0c3f | 4276 | |
31dc1ca3 LP |
4277 | /* Don't check this while we are reloading, things might still change */ |
4278 | if (MANAGER_IS_RELOADING(m)) | |
4279 | return; | |
4280 | ||
8f41e6b6 ZJS |
4281 | /* The journal is fully and entirely up? If so, let's permit logging to it, if that's configured. If |
4282 | * the journal is down, don't ever log to it, otherwise we might end up deadlocking ourselves as we | |
4283 | * might trigger an activation ourselves we can't fulfill. */ | |
cedf5088 | 4284 | log_set_prohibit_ipc(!manager_journal_is_running(m)); |
cc2b9e6b | 4285 | log_open(); |
f1dd0c3f LP |
4286 | } |
4287 | ||
44a41954 FB |
4288 | static ShowStatus manager_get_show_status(Manager *m) { |
4289 | assert(m); | |
4290 | ||
4291 | if (MANAGER_IS_USER(m)) | |
4292 | return _SHOW_STATUS_INVALID; | |
4293 | ||
4294 | if (m->show_status_overridden != _SHOW_STATUS_INVALID) | |
4295 | return m->show_status_overridden; | |
4296 | ||
4297 | return m->show_status; | |
4298 | } | |
4299 | ||
4300 | bool manager_get_show_status_on(Manager *m) { | |
4301 | assert(m); | |
4302 | ||
4303 | return show_status_on(manager_get_show_status(m)); | |
4304 | } | |
b309078a | 4305 | |
3ceb3471 FB |
4306 | static void set_show_status_marker(bool b) { |
4307 | if (b) | |
4308 | (void) touch("/run/systemd/show-status"); | |
4309 | else | |
4310 | (void) unlink("/run/systemd/show-status"); | |
4311 | } | |
4312 | ||
44a41954 | 4313 | void manager_set_show_status(Manager *m, ShowStatus mode, const char *reason) { |
27d340c7 | 4314 | assert(m); |
44a41954 | 4315 | assert(reason); |
0d066dd1 | 4316 | assert(mode >= 0 && mode < _SHOW_STATUS_MAX); |
27d340c7 | 4317 | |
44a41954 | 4318 | if (MANAGER_IS_USER(m)) |
27d340c7 LP |
4319 | return; |
4320 | ||
ef15d3e1 ZJS |
4321 | if (mode == m->show_status) |
4322 | return; | |
4323 | ||
44a41954 FB |
4324 | if (m->show_status_overridden == _SHOW_STATUS_INVALID) { |
4325 | bool enabled; | |
4326 | ||
4327 | enabled = show_status_on(mode); | |
4328 | log_debug("%s (%s) showing of status (%s).", | |
4329 | enabled ? "Enabling" : "Disabling", | |
4330 | strna(show_status_to_string(mode)), | |
4331 | reason); | |
4332 | ||
3ceb3471 | 4333 | set_show_status_marker(enabled); |
44a41954 FB |
4334 | } |
4335 | ||
4336 | m->show_status = mode; | |
4337 | } | |
4338 | ||
43bba15a | 4339 | void manager_override_show_status(Manager *m, ShowStatus mode, const char *reason) { |
44a41954 FB |
4340 | assert(m); |
4341 | assert(mode < _SHOW_STATUS_MAX); | |
4342 | ||
4343 | if (MANAGER_IS_USER(m)) | |
4344 | return; | |
4345 | ||
4346 | if (mode == m->show_status_overridden) | |
4347 | return; | |
4348 | ||
4349 | m->show_status_overridden = mode; | |
4350 | ||
4351 | if (mode == _SHOW_STATUS_INVALID) | |
4352 | mode = m->show_status; | |
4353 | ||
ef15d3e1 | 4354 | log_debug("%s (%s) showing of status (%s).", |
44a41954 | 4355 | m->show_status_overridden != _SHOW_STATUS_INVALID ? "Overriding" : "Restoring", |
ef15d3e1 ZJS |
4356 | strna(show_status_to_string(mode)), |
4357 | reason); | |
b309078a | 4358 | |
3ceb3471 | 4359 | set_show_status_marker(show_status_on(mode)); |
27d340c7 LP |
4360 | } |
4361 | ||
7d5ceb64 FB |
4362 | const char *manager_get_confirm_spawn(Manager *m) { |
4363 | static int last_errno = 0; | |
7d5ceb64 FB |
4364 | struct stat st; |
4365 | int r; | |
4366 | ||
ea758432 LP |
4367 | assert(m); |
4368 | ||
7d5ceb64 FB |
4369 | /* Here's the deal: we want to test the validity of the console but don't want |
4370 | * PID1 to go through the whole console process which might block. But we also | |
4371 | * want to warn the user only once if something is wrong with the console so we | |
4372 | * cannot do the sanity checks after spawning our children. So here we simply do | |
4373 | * really basic tests to hopefully trap common errors. | |
4374 | * | |
4375 | * If the console suddenly disappear at the time our children will really it | |
4376 | * then they will simply fail to acquire it and a positive answer will be | |
2aed63f4 | 4377 | * assumed. New children will fall back to /dev/console though. |
7d5ceb64 FB |
4378 | * |
4379 | * Note: TTYs are devices that can come and go any time, and frequently aren't | |
4380 | * available yet during early boot (consider a USB rs232 dongle...). If for any | |
2aed63f4 | 4381 | * reason the configured console is not ready, we fall back to the default |
7d5ceb64 FB |
4382 | * console. */ |
4383 | ||
ea758432 LP |
4384 | if (!m->confirm_spawn || path_equal(m->confirm_spawn, "/dev/console")) |
4385 | return m->confirm_spawn; | |
7d5ceb64 | 4386 | |
ea758432 LP |
4387 | if (stat(m->confirm_spawn, &st) < 0) { |
4388 | r = -errno; | |
7d5ceb64 | 4389 | goto fail; |
ea758432 | 4390 | } |
7d5ceb64 FB |
4391 | |
4392 | if (!S_ISCHR(st.st_mode)) { | |
ea758432 | 4393 | r = -ENOTTY; |
7d5ceb64 FB |
4394 | goto fail; |
4395 | } | |
4396 | ||
4397 | last_errno = 0; | |
ea758432 LP |
4398 | return m->confirm_spawn; |
4399 | ||
7d5ceb64 | 4400 | fail: |
ea758432 LP |
4401 | if (last_errno != r) |
4402 | last_errno = log_warning_errno(r, "Failed to open %s, using default console: %m", m->confirm_spawn); | |
4403 | ||
7d5ceb64 FB |
4404 | return "/dev/console"; |
4405 | } | |
4406 | ||
e2680723 LP |
4407 | void manager_set_first_boot(Manager *m, bool b) { |
4408 | assert(m); | |
4409 | ||
463d0d15 | 4410 | if (!MANAGER_IS_SYSTEM(m)) |
e2680723 LP |
4411 | return; |
4412 | ||
ae2a2c53 LP |
4413 | if (m->first_boot != (int) b) { |
4414 | if (b) | |
4415 | (void) touch("/run/systemd/first-boot"); | |
4416 | else | |
4417 | (void) unlink("/run/systemd/first-boot"); | |
4418 | } | |
e2680723 | 4419 | |
ae2a2c53 | 4420 | m->first_boot = b; |
e2680723 LP |
4421 | } |
4422 | ||
b0eb2944 FB |
4423 | void manager_disable_confirm_spawn(void) { |
4424 | (void) touch("/run/systemd/confirm_spawn_disabled"); | |
4425 | } | |
4426 | ||
4427 | bool manager_is_confirm_spawn_disabled(Manager *m) { | |
4428 | if (!m->confirm_spawn) | |
4429 | return true; | |
4430 | ||
4431 | return access("/run/systemd/confirm_spawn_disabled", F_OK) >= 0; | |
4432 | } | |
4433 | ||
0d6d3cf0 FB |
4434 | static bool manager_should_show_status(Manager *m, StatusType type) { |
4435 | assert(m); | |
4436 | ||
4437 | if (!MANAGER_IS_SYSTEM(m)) | |
4438 | return false; | |
4439 | ||
4440 | if (m->no_console_output) | |
4441 | return false; | |
4442 | ||
4443 | if (!IN_SET(manager_state(m), MANAGER_INITIALIZING, MANAGER_STARTING, MANAGER_STOPPING)) | |
4444 | return false; | |
4445 | ||
4446 | /* If we cannot find out the status properly, just proceed. */ | |
4447 | if (type != STATUS_TYPE_EMERGENCY && manager_check_ask_password(m) > 0) | |
4448 | return false; | |
4449 | ||
4450 | if (type == STATUS_TYPE_NOTICE && m->show_status != SHOW_STATUS_NO) | |
4451 | return true; | |
4452 | ||
44a41954 | 4453 | return manager_get_show_status_on(m); |
0d6d3cf0 FB |
4454 | } |
4455 | ||
127d5fd1 | 4456 | void manager_status_printf(Manager *m, StatusType type, const char *status, const char *format, ...) { |
25cee550 MS |
4457 | va_list ap; |
4458 | ||
cb6531be ZJS |
4459 | /* If m is NULL, assume we're after shutdown and let the messages through. */ |
4460 | ||
0d6d3cf0 | 4461 | if (m && !manager_should_show_status(m, type)) |
25cee550 MS |
4462 | return; |
4463 | ||
03b717a3 MS |
4464 | /* XXX We should totally drop the check for ephemeral here |
4465 | * and thus effectively make 'Type=idle' pointless. */ | |
cb6531be | 4466 | if (type == STATUS_TYPE_EPHEMERAL && m && m->n_on_console > 0) |
03b717a3 MS |
4467 | return; |
4468 | ||
25cee550 | 4469 | va_start(ap, format); |
a885727a | 4470 | status_vprintf(status, SHOW_STATUS_ELLIPSIZE|(type == STATUS_TYPE_EPHEMERAL ? SHOW_STATUS_EPHEMERAL : 0), format, ap); |
25cee550 MS |
4471 | va_end(ap); |
4472 | } | |
4473 | ||
ac19bdd0 | 4474 | Set* manager_get_units_requiring_mounts_for(Manager *m, const char *path) { |
a57f7e2c LP |
4475 | assert(m); |
4476 | assert(path); | |
4477 | ||
ac19bdd0 ZJS |
4478 | if (path_equal(path, "/")) |
4479 | path = ""; | |
a57f7e2c | 4480 | |
ac19bdd0 | 4481 | return hashmap_get(m->units_requiring_mounts_for, path); |
a57f7e2c | 4482 | } |
e66cf1a3 | 4483 | |
5269eb6b | 4484 | int manager_update_failed_units(Manager *m, Unit *u, bool failed) { |
03455c28 | 4485 | unsigned size; |
5269eb6b | 4486 | int r; |
03455c28 LDM |
4487 | |
4488 | assert(m); | |
4489 | assert(u->manager == m); | |
4490 | ||
4491 | size = set_size(m->failed_units); | |
4492 | ||
9fff8981 | 4493 | if (failed) { |
de7fef4b | 4494 | r = set_ensure_put(&m->failed_units, NULL, u); |
5269eb6b LP |
4495 | if (r < 0) |
4496 | return log_oom(); | |
9fff8981 | 4497 | } else |
5269eb6b | 4498 | (void) set_remove(m->failed_units, u); |
03455c28 LDM |
4499 | |
4500 | if (set_size(m->failed_units) != size) | |
4501 | bus_manager_send_change_signal(m); | |
5269eb6b LP |
4502 | |
4503 | return 0; | |
03455c28 LDM |
4504 | } |
4505 | ||
f755e3b7 LP |
4506 | ManagerState manager_state(Manager *m) { |
4507 | Unit *u; | |
4508 | ||
4509 | assert(m); | |
4510 | ||
f9d29f6d JB |
4511 | /* Is the special shutdown target active or queued? If so, we are in shutdown state */ |
4512 | u = manager_get_unit(m, SPECIAL_SHUTDOWN_TARGET); | |
4513 | if (u && unit_active_or_pending(u)) | |
4514 | return MANAGER_STOPPING; | |
4515 | ||
f755e3b7 | 4516 | /* Did we ever finish booting? If not then we are still starting up */ |
49d5666c | 4517 | if (!MANAGER_IS_FINISHED(m)) { |
d81afec1 LP |
4518 | |
4519 | u = manager_get_unit(m, SPECIAL_BASIC_TARGET); | |
4520 | if (!u || !UNIT_IS_ACTIVE_OR_RELOADING(unit_active_state(u))) | |
4521 | return MANAGER_INITIALIZING; | |
4522 | ||
f755e3b7 | 4523 | return MANAGER_STARTING; |
d81afec1 | 4524 | } |
f755e3b7 | 4525 | |
45a7b16b LP |
4526 | if (MANAGER_IS_SYSTEM(m)) { |
4527 | /* Are the rescue or emergency targets active or queued? If so we are in maintenance state */ | |
4528 | u = manager_get_unit(m, SPECIAL_RESCUE_TARGET); | |
4529 | if (u && unit_active_or_pending(u)) | |
4530 | return MANAGER_MAINTENANCE; | |
f755e3b7 | 4531 | |
45a7b16b LP |
4532 | u = manager_get_unit(m, SPECIAL_EMERGENCY_TARGET); |
4533 | if (u && unit_active_or_pending(u)) | |
4534 | return MANAGER_MAINTENANCE; | |
4535 | } | |
f755e3b7 LP |
4536 | |
4537 | /* Are there any failed units? If so, we are in degraded mode */ | |
4538 | if (set_size(m->failed_units) > 0) | |
4539 | return MANAGER_DEGRADED; | |
4540 | ||
4541 | return MANAGER_RUNNING; | |
4542 | } | |
4543 | ||
00d9ef85 | 4544 | static void manager_unref_uid_internal( |
010becd9 | 4545 | Hashmap *uid_refs, |
00d9ef85 LP |
4546 | uid_t uid, |
4547 | bool destroy_now, | |
4548 | int (*_clean_ipc)(uid_t uid)) { | |
4549 | ||
4550 | uint32_t c, n; | |
4551 | ||
00d9ef85 LP |
4552 | assert(uid_is_valid(uid)); |
4553 | assert(_clean_ipc); | |
4554 | ||
8f41e6b6 ZJS |
4555 | /* A generic implementation, covering both manager_unref_uid() and manager_unref_gid(), under the |
4556 | * assumption that uid_t and gid_t are actually defined the same way, with the same validity rules. | |
00d9ef85 | 4557 | * |
da890466 | 4558 | * We store a hashmap where the key is the UID/GID and the value is a 32-bit reference counter, whose |
8f41e6b6 ZJS |
4559 | * highest bit is used as flag for marking UIDs/GIDs whose IPC objects to remove when the last |
4560 | * reference to the UID/GID is dropped. The flag is set to on, once at least one reference from a | |
4561 | * unit where RemoveIPC= is set is added on a UID/GID. It is reset when the UID's/GID's reference | |
4562 | * counter drops to 0 again. */ | |
00d9ef85 LP |
4563 | |
4564 | assert_cc(sizeof(uid_t) == sizeof(gid_t)); | |
4565 | assert_cc(UID_INVALID == (uid_t) GID_INVALID); | |
4566 | ||
4567 | if (uid == 0) /* We don't keep track of root, and will never destroy it */ | |
4568 | return; | |
4569 | ||
010becd9 | 4570 | c = PTR_TO_UINT32(hashmap_get(uid_refs, UID_TO_PTR(uid))); |
00d9ef85 LP |
4571 | |
4572 | n = c & ~DESTROY_IPC_FLAG; | |
4573 | assert(n > 0); | |
4574 | n--; | |
4575 | ||
4576 | if (destroy_now && n == 0) { | |
010becd9 | 4577 | hashmap_remove(uid_refs, UID_TO_PTR(uid)); |
00d9ef85 LP |
4578 | |
4579 | if (c & DESTROY_IPC_FLAG) { | |
4580 | log_debug("%s " UID_FMT " is no longer referenced, cleaning up its IPC.", | |
4581 | _clean_ipc == clean_ipc_by_uid ? "UID" : "GID", | |
4582 | uid); | |
4583 | (void) _clean_ipc(uid); | |
4584 | } | |
4585 | } else { | |
4586 | c = n | (c & DESTROY_IPC_FLAG); | |
010becd9 | 4587 | assert_se(hashmap_update(uid_refs, UID_TO_PTR(uid), UINT32_TO_PTR(c)) >= 0); |
00d9ef85 LP |
4588 | } |
4589 | } | |
4590 | ||
4591 | void manager_unref_uid(Manager *m, uid_t uid, bool destroy_now) { | |
010becd9 | 4592 | manager_unref_uid_internal(m->uid_refs, uid, destroy_now, clean_ipc_by_uid); |
00d9ef85 LP |
4593 | } |
4594 | ||
4595 | void manager_unref_gid(Manager *m, gid_t gid, bool destroy_now) { | |
010becd9 | 4596 | manager_unref_uid_internal(m->gid_refs, (uid_t) gid, destroy_now, clean_ipc_by_gid); |
00d9ef85 LP |
4597 | } |
4598 | ||
4599 | static int manager_ref_uid_internal( | |
00d9ef85 LP |
4600 | Hashmap **uid_refs, |
4601 | uid_t uid, | |
4602 | bool clean_ipc) { | |
4603 | ||
4604 | uint32_t c, n; | |
4605 | int r; | |
4606 | ||
00d9ef85 LP |
4607 | assert(uid_refs); |
4608 | assert(uid_is_valid(uid)); | |
4609 | ||
8f41e6b6 ZJS |
4610 | /* A generic implementation, covering both manager_ref_uid() and manager_ref_gid(), under the |
4611 | * assumption that uid_t and gid_t are actually defined the same way, with the same validity | |
4612 | * rules. */ | |
00d9ef85 LP |
4613 | |
4614 | assert_cc(sizeof(uid_t) == sizeof(gid_t)); | |
4615 | assert_cc(UID_INVALID == (uid_t) GID_INVALID); | |
4616 | ||
4617 | if (uid == 0) /* We don't keep track of root, and will never destroy it */ | |
4618 | return 0; | |
4619 | ||
4620 | r = hashmap_ensure_allocated(uid_refs, &trivial_hash_ops); | |
4621 | if (r < 0) | |
4622 | return r; | |
4623 | ||
4624 | c = PTR_TO_UINT32(hashmap_get(*uid_refs, UID_TO_PTR(uid))); | |
4625 | ||
4626 | n = c & ~DESTROY_IPC_FLAG; | |
4627 | n++; | |
4628 | ||
4629 | if (n & DESTROY_IPC_FLAG) /* check for overflow */ | |
4630 | return -EOVERFLOW; | |
4631 | ||
4632 | c = n | (c & DESTROY_IPC_FLAG) | (clean_ipc ? DESTROY_IPC_FLAG : 0); | |
4633 | ||
4634 | return hashmap_replace(*uid_refs, UID_TO_PTR(uid), UINT32_TO_PTR(c)); | |
4635 | } | |
4636 | ||
4637 | int manager_ref_uid(Manager *m, uid_t uid, bool clean_ipc) { | |
010becd9 | 4638 | return manager_ref_uid_internal(&m->uid_refs, uid, clean_ipc); |
00d9ef85 LP |
4639 | } |
4640 | ||
4641 | int manager_ref_gid(Manager *m, gid_t gid, bool clean_ipc) { | |
010becd9 | 4642 | return manager_ref_uid_internal(&m->gid_refs, (uid_t) gid, clean_ipc); |
00d9ef85 LP |
4643 | } |
4644 | ||
4645 | static void manager_vacuum_uid_refs_internal( | |
010becd9 | 4646 | Hashmap *uid_refs, |
00d9ef85 LP |
4647 | int (*_clean_ipc)(uid_t uid)) { |
4648 | ||
00d9ef85 LP |
4649 | void *p, *k; |
4650 | ||
00d9ef85 LP |
4651 | assert(_clean_ipc); |
4652 | ||
010becd9 | 4653 | HASHMAP_FOREACH_KEY(p, k, uid_refs) { |
00d9ef85 LP |
4654 | uint32_t c, n; |
4655 | uid_t uid; | |
4656 | ||
4657 | uid = PTR_TO_UID(k); | |
4658 | c = PTR_TO_UINT32(p); | |
4659 | ||
4660 | n = c & ~DESTROY_IPC_FLAG; | |
4661 | if (n > 0) | |
4662 | continue; | |
4663 | ||
4664 | if (c & DESTROY_IPC_FLAG) { | |
4665 | log_debug("Found unreferenced %s " UID_FMT " after reload/reexec. Cleaning up.", | |
4666 | _clean_ipc == clean_ipc_by_uid ? "UID" : "GID", | |
4667 | uid); | |
4668 | (void) _clean_ipc(uid); | |
4669 | } | |
4670 | ||
010becd9 | 4671 | assert_se(hashmap_remove(uid_refs, k) == p); |
00d9ef85 LP |
4672 | } |
4673 | } | |
4674 | ||
06a4eb07 | 4675 | static void manager_vacuum_uid_refs(Manager *m) { |
010becd9 | 4676 | manager_vacuum_uid_refs_internal(m->uid_refs, clean_ipc_by_uid); |
00d9ef85 LP |
4677 | } |
4678 | ||
06a4eb07 | 4679 | static void manager_vacuum_gid_refs(Manager *m) { |
010becd9 | 4680 | manager_vacuum_uid_refs_internal(m->gid_refs, clean_ipc_by_gid); |
00d9ef85 LP |
4681 | } |
4682 | ||
06a4eb07 FB |
4683 | static void manager_vacuum(Manager *m) { |
4684 | assert(m); | |
4685 | ||
4686 | /* Release any dynamic users no longer referenced */ | |
4687 | dynamic_user_vacuum(m, true); | |
4688 | ||
4689 | /* Release any references to UIDs/GIDs no longer referenced, and destroy any IPC owned by them */ | |
4690 | manager_vacuum_uid_refs(m); | |
4691 | manager_vacuum_gid_refs(m); | |
4692 | ||
4693 | /* Release any runtimes no longer referenced */ | |
e76506b7 | 4694 | exec_shared_runtime_vacuum(m); |
06a4eb07 FB |
4695 | } |
4696 | ||
00d9ef85 LP |
4697 | int manager_dispatch_user_lookup_fd(sd_event_source *source, int fd, uint32_t revents, void *userdata) { |
4698 | struct buffer { | |
4699 | uid_t uid; | |
4700 | gid_t gid; | |
4701 | char unit_name[UNIT_NAME_MAX+1]; | |
4702 | } _packed_ buffer; | |
4703 | ||
4704 | Manager *m = userdata; | |
4705 | ssize_t l; | |
4706 | size_t n; | |
4707 | Unit *u; | |
4708 | ||
4709 | assert_se(source); | |
4710 | assert_se(m); | |
4711 | ||
8f41e6b6 ZJS |
4712 | /* Invoked whenever a child process succeeded resolving its user/group to use and sent us the |
4713 | * resulting UID/GID in a datagram. We parse the datagram here and pass it off to the unit, so that | |
4714 | * it can add a reference to the UID/GID so that it can destroy the UID/GID's IPC objects when the | |
4715 | * reference counter drops to 0. */ | |
00d9ef85 LP |
4716 | |
4717 | l = recv(fd, &buffer, sizeof(buffer), MSG_DONTWAIT); | |
4718 | if (l < 0) { | |
8add30a0 | 4719 | if (ERRNO_IS_TRANSIENT(errno)) |
00d9ef85 LP |
4720 | return 0; |
4721 | ||
4722 | return log_error_errno(errno, "Failed to read from user lookup fd: %m"); | |
4723 | } | |
4724 | ||
4725 | if ((size_t) l <= offsetof(struct buffer, unit_name)) { | |
4726 | log_warning("Received too short user lookup message, ignoring."); | |
4727 | return 0; | |
4728 | } | |
4729 | ||
4730 | if ((size_t) l > offsetof(struct buffer, unit_name) + UNIT_NAME_MAX) { | |
4731 | log_warning("Received too long user lookup message, ignoring."); | |
4732 | return 0; | |
4733 | } | |
4734 | ||
4735 | if (!uid_is_valid(buffer.uid) && !gid_is_valid(buffer.gid)) { | |
4736 | log_warning("Got user lookup message with invalid UID/GID pair, ignoring."); | |
4737 | return 0; | |
4738 | } | |
4739 | ||
4740 | n = (size_t) l - offsetof(struct buffer, unit_name); | |
4741 | if (memchr(buffer.unit_name, 0, n)) { | |
4742 | log_warning("Received lookup message with embedded NUL character, ignoring."); | |
4743 | return 0; | |
4744 | } | |
4745 | ||
4746 | buffer.unit_name[n] = 0; | |
4747 | u = manager_get_unit(m, buffer.unit_name); | |
4748 | if (!u) { | |
4749 | log_debug("Got user lookup message but unit doesn't exist, ignoring."); | |
4750 | return 0; | |
4751 | } | |
4752 | ||
4753 | log_unit_debug(u, "User lookup succeeded: uid=" UID_FMT " gid=" GID_FMT, buffer.uid, buffer.gid); | |
4754 | ||
4755 | unit_notify_user_lookup(u, buffer.uid, buffer.gid); | |
4756 | return 0; | |
4757 | } | |
4758 | ||
63e8df04 | 4759 | static int short_uid_range(const char *path) { |
8dcc66ce | 4760 | _cleanup_(uid_range_freep) UidRange *p = NULL; |
63e8df04 LP |
4761 | int r; |
4762 | ||
4763 | assert(path); | |
4764 | ||
4765 | /* Taint systemd if we the UID range assigned to this environment doesn't at least cover 0…65534, | |
4766 | * i.e. from root to nobody. */ | |
4767 | ||
8dcc66ce | 4768 | r = uid_range_load_userns(&p, path); |
882321a1 ZJS |
4769 | if (ERRNO_IS_NEG_NOT_SUPPORTED(r)) |
4770 | return false; | |
4771 | if (r < 0) | |
63e8df04 LP |
4772 | return log_debug_errno(r, "Failed to load %s: %m", path); |
4773 | ||
8dcc66ce | 4774 | return !uid_range_covers(p, 0, 65535); |
63e8df04 LP |
4775 | } |
4776 | ||
83fe5d8a ZJS |
4777 | char* manager_taint_string(const Manager *m) { |
4778 | /* Returns a "taint string", e.g. "local-hwclock:var-run-bad". Only things that are detected at | |
4779 | * runtime should be tagged here. For stuff that is known during compilation, emit a warning in the | |
63e8df04 | 4780 | * configuration phase. */ |
198ce932 | 4781 | |
af6b0ecc LP |
4782 | assert(m); |
4783 | ||
b0d3095f | 4784 | const char* stage[12] = {}; |
83fe5d8a | 4785 | size_t n = 0; |
af6b0ecc | 4786 | |
31cd2dd9 LB |
4787 | _cleanup_free_ char *usrbin = NULL; |
4788 | if (readlink_malloc("/bin", &usrbin) < 0 || !PATH_IN_SET(usrbin, "usr/bin", "/usr/bin")) | |
4789 | stage[n++] = "unmerged-usr"; | |
4790 | ||
af6b0ecc | 4791 | if (access("/proc/cgroups", F_OK) < 0) |
83fe5d8a | 4792 | stage[n++] = "cgroups-missing"; |
af6b0ecc | 4793 | |
82f30632 | 4794 | if (cg_all_unified() == 0) |
83fe5d8a | 4795 | stage[n++] = "cgroupsv1"; |
82f30632 | 4796 | |
af6b0ecc | 4797 | if (clock_is_localtime(NULL) > 0) |
83fe5d8a | 4798 | stage[n++] = "local-hwclock"; |
af6b0ecc | 4799 | |
6bfe9b3b | 4800 | if (os_release_support_ended(NULL, /* quiet= */ true, NULL) > 0) |
4bd03515 ZJS |
4801 | stage[n++] = "support-ended"; |
4802 | ||
83fe5d8a ZJS |
4803 | _cleanup_free_ char *destination = NULL; |
4804 | if (readlink_malloc("/var/run", &destination) < 0 || | |
4805 | !PATH_IN_SET(destination, "../run", "/run")) | |
4806 | stage[n++] = "var-run-bad"; | |
af6b0ecc | 4807 | |
83fe5d8a ZJS |
4808 | _cleanup_free_ char *overflowuid = NULL, *overflowgid = NULL; |
4809 | if (read_one_line_file("/proc/sys/kernel/overflowuid", &overflowuid) >= 0 && | |
4810 | !streq(overflowuid, "65534")) | |
4811 | stage[n++] = "overflowuid-not-65534"; | |
4812 | if (read_one_line_file("/proc/sys/kernel/overflowgid", &overflowgid) >= 0 && | |
4813 | !streq(overflowgid, "65534")) | |
4814 | stage[n++] = "overflowgid-not-65534"; | |
90d7464d | 4815 | |
83fe5d8a | 4816 | struct utsname uts; |
40efaaed LP |
4817 | assert_se(uname(&uts) >= 0); |
4818 | if (strverscmp_improved(uts.release, KERNEL_BASELINE_VERSION) < 0) | |
83fe5d8a | 4819 | stage[n++] = "old-kernel"; |
40efaaed | 4820 | |
63e8df04 | 4821 | if (short_uid_range("/proc/self/uid_map") > 0) |
83fe5d8a | 4822 | stage[n++] = "short-uid-range"; |
63e8df04 | 4823 | if (short_uid_range("/proc/self/gid_map") > 0) |
83fe5d8a | 4824 | stage[n++] = "short-gid-range"; |
63e8df04 | 4825 | |
83fe5d8a | 4826 | assert(n < ELEMENTSOF(stage) - 1); /* One extra for NULL terminator */ |
af6b0ecc | 4827 | |
83fe5d8a | 4828 | return strv_join((char**) stage, ":"); |
af6b0ecc LP |
4829 | } |
4830 | ||
adefcf28 LP |
4831 | void manager_ref_console(Manager *m) { |
4832 | assert(m); | |
4833 | ||
4834 | m->n_on_console++; | |
4835 | } | |
4836 | ||
4837 | void manager_unref_console(Manager *m) { | |
4838 | ||
4839 | assert(m->n_on_console > 0); | |
4840 | m->n_on_console--; | |
4841 | ||
4842 | if (m->n_on_console == 0) | |
4843 | m->no_console_output = false; /* unset no_console_output flag, since the console is definitely free now */ | |
4844 | } | |
4845 | ||
a6ecbf83 FB |
4846 | void manager_override_log_level(Manager *m, int level) { |
4847 | _cleanup_free_ char *s = NULL; | |
4848 | assert(m); | |
4849 | ||
4850 | if (!m->log_level_overridden) { | |
4851 | m->original_log_level = log_get_max_level(); | |
4852 | m->log_level_overridden = true; | |
4853 | } | |
4854 | ||
4855 | (void) log_level_to_string_alloc(level, &s); | |
4856 | log_info("Setting log level to %s.", strna(s)); | |
4857 | ||
4858 | log_set_max_level(level); | |
4859 | } | |
4860 | ||
4861 | void manager_restore_original_log_level(Manager *m) { | |
4862 | _cleanup_free_ char *s = NULL; | |
4863 | assert(m); | |
4864 | ||
4865 | if (!m->log_level_overridden) | |
4866 | return; | |
4867 | ||
4868 | (void) log_level_to_string_alloc(m->original_log_level, &s); | |
4869 | log_info("Restoring log level to original (%s).", strna(s)); | |
4870 | ||
4871 | log_set_max_level(m->original_log_level); | |
4872 | m->log_level_overridden = false; | |
4873 | } | |
4874 | ||
bda7d78b FB |
4875 | void manager_override_log_target(Manager *m, LogTarget target) { |
4876 | assert(m); | |
4877 | ||
4878 | if (!m->log_target_overridden) { | |
4879 | m->original_log_target = log_get_target(); | |
4880 | m->log_target_overridden = true; | |
4881 | } | |
4882 | ||
4883 | log_info("Setting log target to %s.", log_target_to_string(target)); | |
4884 | log_set_target(target); | |
4885 | } | |
4886 | ||
4887 | void manager_restore_original_log_target(Manager *m) { | |
4888 | assert(m); | |
4889 | ||
4890 | if (!m->log_target_overridden) | |
4891 | return; | |
4892 | ||
4893 | log_info("Restoring log target to original %s.", log_target_to_string(m->original_log_target)); | |
4894 | ||
4895 | log_set_target(m->original_log_target); | |
4896 | m->log_target_overridden = false; | |
4897 | } | |
4898 | ||
d4ee7bd8 YW |
4899 | ManagerTimestamp manager_timestamp_initrd_mangle(ManagerTimestamp s) { |
4900 | if (in_initrd() && | |
4901 | s >= MANAGER_TIMESTAMP_SECURITY_START && | |
4902 | s <= MANAGER_TIMESTAMP_UNITS_LOAD_FINISH) | |
4903 | return s - MANAGER_TIMESTAMP_SECURITY_START + MANAGER_TIMESTAMP_INITRD_SECURITY_START; | |
4904 | return s; | |
4905 | } | |
4906 | ||
ea09a416 LP |
4907 | void unit_defaults_init(UnitDefaults *defaults, RuntimeScope scope) { |
4908 | assert(defaults); | |
4909 | assert(scope >= 0); | |
4910 | assert(scope < _RUNTIME_SCOPE_MAX); | |
4911 | ||
4912 | *defaults = (UnitDefaults) { | |
4913 | .std_output = EXEC_OUTPUT_JOURNAL, | |
4914 | .std_error = EXEC_OUTPUT_INHERIT, | |
4915 | .restart_usec = DEFAULT_RESTART_USEC, | |
4916 | .timeout_start_usec = manager_default_timeout(scope), | |
4917 | .timeout_stop_usec = manager_default_timeout(scope), | |
4918 | .timeout_abort_usec = manager_default_timeout(scope), | |
4919 | .timeout_abort_set = false, | |
4920 | .device_timeout_usec = manager_default_timeout(scope), | |
4921 | .start_limit_interval = DEFAULT_START_LIMIT_INTERVAL, | |
4922 | .start_limit_burst = DEFAULT_START_LIMIT_BURST, | |
4923 | ||
4924 | /* On 4.15+ with unified hierarchy, CPU accounting is essentially free as it doesn't require the CPU | |
4925 | * controller to be enabled, so the default is to enable it unless we got told otherwise. */ | |
4926 | .cpu_accounting = cpu_accounting_is_cheap(), | |
4927 | .memory_accounting = MEMORY_ACCOUNTING_DEFAULT, | |
4928 | .io_accounting = false, | |
4929 | .blockio_accounting = false, | |
4930 | .tasks_accounting = true, | |
4931 | .ip_accounting = false, | |
4932 | ||
4933 | .tasks_max = DEFAULT_TASKS_MAX, | |
4934 | .timer_accuracy_usec = 1 * USEC_PER_MINUTE, | |
4935 | ||
4936 | .memory_pressure_watch = CGROUP_PRESSURE_WATCH_AUTO, | |
4937 | .memory_pressure_threshold_usec = MEMORY_PRESSURE_DEFAULT_THRESHOLD_USEC, | |
4938 | ||
4939 | .oom_policy = OOM_STOP, | |
4940 | .oom_score_adjust_set = false, | |
4941 | }; | |
4942 | } | |
4943 | ||
c9e120e0 LP |
4944 | void unit_defaults_done(UnitDefaults *defaults) { |
4945 | assert(defaults); | |
4946 | ||
4947 | defaults->smack_process_label = mfree(defaults->smack_process_label); | |
4948 | rlimit_free_all(defaults->rlimit); | |
4949 | } | |
4950 | ||
f755e3b7 | 4951 | static const char *const manager_state_table[_MANAGER_STATE_MAX] = { |
d81afec1 | 4952 | [MANAGER_INITIALIZING] = "initializing", |
48d83e33 ZJS |
4953 | [MANAGER_STARTING] = "starting", |
4954 | [MANAGER_RUNNING] = "running", | |
4955 | [MANAGER_DEGRADED] = "degraded", | |
4956 | [MANAGER_MAINTENANCE] = "maintenance", | |
4957 | [MANAGER_STOPPING] = "stopping", | |
f755e3b7 LP |
4958 | }; |
4959 | ||
4960 | DEFINE_STRING_TABLE_LOOKUP(manager_state, ManagerState); | |
9f9f0342 LP |
4961 | |
4962 | static const char *const manager_timestamp_table[_MANAGER_TIMESTAMP_MAX] = { | |
48d83e33 ZJS |
4963 | [MANAGER_TIMESTAMP_FIRMWARE] = "firmware", |
4964 | [MANAGER_TIMESTAMP_LOADER] = "loader", | |
4965 | [MANAGER_TIMESTAMP_KERNEL] = "kernel", | |
4966 | [MANAGER_TIMESTAMP_INITRD] = "initrd", | |
4967 | [MANAGER_TIMESTAMP_USERSPACE] = "userspace", | |
4968 | [MANAGER_TIMESTAMP_FINISH] = "finish", | |
4969 | [MANAGER_TIMESTAMP_SECURITY_START] = "security-start", | |
4970 | [MANAGER_TIMESTAMP_SECURITY_FINISH] = "security-finish", | |
4971 | [MANAGER_TIMESTAMP_GENERATORS_START] = "generators-start", | |
4972 | [MANAGER_TIMESTAMP_GENERATORS_FINISH] = "generators-finish", | |
4973 | [MANAGER_TIMESTAMP_UNITS_LOAD_START] = "units-load-start", | |
4974 | [MANAGER_TIMESTAMP_UNITS_LOAD_FINISH] = "units-load-finish", | |
49fbe940 | 4975 | [MANAGER_TIMESTAMP_UNITS_LOAD] = "units-load", |
48d83e33 ZJS |
4976 | [MANAGER_TIMESTAMP_INITRD_SECURITY_START] = "initrd-security-start", |
4977 | [MANAGER_TIMESTAMP_INITRD_SECURITY_FINISH] = "initrd-security-finish", | |
4978 | [MANAGER_TIMESTAMP_INITRD_GENERATORS_START] = "initrd-generators-start", | |
d4ee7bd8 | 4979 | [MANAGER_TIMESTAMP_INITRD_GENERATORS_FINISH] = "initrd-generators-finish", |
48d83e33 | 4980 | [MANAGER_TIMESTAMP_INITRD_UNITS_LOAD_START] = "initrd-units-load-start", |
d4ee7bd8 | 4981 | [MANAGER_TIMESTAMP_INITRD_UNITS_LOAD_FINISH] = "initrd-units-load-finish", |
9f9f0342 LP |
4982 | }; |
4983 | ||
4984 | DEFINE_STRING_TABLE_LOOKUP(manager_timestamp, ManagerTimestamp); | |
afcfaa69 LP |
4985 | |
4986 | static const char* const oom_policy_table[_OOM_POLICY_MAX] = { | |
4987 | [OOM_CONTINUE] = "continue", | |
48d83e33 ZJS |
4988 | [OOM_STOP] = "stop", |
4989 | [OOM_KILL] = "kill", | |
afcfaa69 LP |
4990 | }; |
4991 | ||
4992 | DEFINE_STRING_TABLE_LOOKUP(oom_policy, OOMPolicy); |