]>
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" |
28db6fbf | 35 | #include "constants.h" |
19d22d43 | 36 | #include "core-varlink.h" |
786d19fd | 37 | #include "creds-util.h" |
400f1a33 LP |
38 | #include "dbus-job.h" |
39 | #include "dbus-manager.h" | |
40 | #include "dbus-unit.h" | |
41 | #include "dbus.h" | |
d063a527 | 42 | #include "dirent-util.h" |
400f1a33 | 43 | #include "env-util.h" |
4f5dd394 | 44 | #include "escape.h" |
ec75e8e0 | 45 | #include "event-util.h" |
89711996 | 46 | #include "exec-util.h" |
d3070fbd | 47 | #include "execute.h" |
400f1a33 | 48 | #include "exit-status.h" |
3ffd4af2 | 49 | #include "fd-util.h" |
0d39fa9c | 50 | #include "fileio.h" |
385093b7 | 51 | #include "generator-setup.h" |
60918275 | 52 | #include "hashmap.h" |
baa6a42d | 53 | #include "initrd-util.h" |
9e5fd717 | 54 | #include "inotify-util.h" |
5cfa33e0 | 55 | #include "install.h" |
19d22d43 | 56 | #include "io-util.h" |
d3070fbd | 57 | #include "label.h" |
d904afc7 | 58 | #include "load-fragment.h" |
786d19fd | 59 | #include "locale-setup.h" |
16354eff | 60 | #include "log.h" |
400f1a33 | 61 | #include "macro.h" |
3ffd4af2 | 62 | #include "manager.h" |
2a341bb9 | 63 | #include "manager-dump.h" |
a01ba4b2 | 64 | #include "manager-serialize.h" |
0a970718 | 65 | #include "memory-util.h" |
35cd0ba5 | 66 | #include "mkdir-label.h" |
ca6ce62d | 67 | #include "mount-util.h" |
4bd03515 | 68 | #include "os-util.h" |
6bedfcbb | 69 | #include "parse-util.h" |
400f1a33 LP |
70 | #include "path-lookup.h" |
71 | #include "path-util.h" | |
72 | #include "process-util.h" | |
6bb00842 | 73 | #include "psi-util.h" |
ea430986 | 74 | #include "ratelimit.h" |
31ce987c | 75 | #include "rlimit-util.h" |
c6878637 | 76 | #include "rm-rf.h" |
45ae2f72 | 77 | #include "selinux-util.h" |
400f1a33 | 78 | #include "signal-util.h" |
57b7a260 | 79 | #include "socket-util.h" |
514f4ef5 | 80 | #include "special.h" |
8fcde012 | 81 | #include "stat-util.h" |
8b43440b | 82 | #include "string-table.h" |
07630cea | 83 | #include "string-util.h" |
400f1a33 | 84 | #include "strv.h" |
dd1db3c2 | 85 | #include "strxcpyx.h" |
6123dfaa | 86 | #include "sysctl-util.h" |
a6ecbf83 | 87 | #include "syslog-util.h" |
400f1a33 LP |
88 | #include "terminal-util.h" |
89 | #include "time-util.h" | |
90 | #include "transaction.h" | |
63e8df04 | 91 | #include "uid-range.h" |
affb60b1 | 92 | #include "umask-util.h" |
400f1a33 | 93 | #include "unit-name.h" |
00d9ef85 | 94 | #include "user-util.h" |
5dc4c17f | 95 | #include "virt.h" |
e96d6be7 | 96 | #include "watchdog.h" |
60918275 | 97 | |
a47806fa | 98 | #define NOTIFY_RCVBUF_SIZE (8*1024*1024) |
d8fdc620 | 99 | #define CGROUPS_AGENT_RCVBUF_SIZE (8*1024*1024) |
a47806fa | 100 | |
03b717a3 | 101 | /* Initial delay and the interval for printing status messages about running jobs */ |
1b4154a8 ZJS |
102 | #define JOBS_IN_PROGRESS_WAIT_USEC (2*USEC_PER_SEC) |
103 | #define JOBS_IN_PROGRESS_QUIET_WAIT_USEC (25*USEC_PER_SEC) | |
fd08a840 | 104 | #define JOBS_IN_PROGRESS_PERIOD_USEC (USEC_PER_SEC / 3) |
03b717a3 MS |
105 | #define JOBS_IN_PROGRESS_PERIOD_DIVISOR 3 |
106 | ||
5238e957 | 107 | /* 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 |
108 | * the queue gets more empty. */ |
109 | #define MANAGER_BUS_BUSY_THRESHOLD 1024LU | |
110 | ||
111 | /* How many units and jobs to process of the bus queue before returning to the event loop. */ | |
112 | #define MANAGER_BUS_MESSAGE_BUDGET 100U | |
113 | ||
718db961 | 114 | static int manager_dispatch_notify_fd(sd_event_source *source, int fd, uint32_t revents, void *userdata); |
d8fdc620 | 115 | static int manager_dispatch_cgroups_agent_fd(sd_event_source *source, int fd, uint32_t revents, void *userdata); |
718db961 LP |
116 | static int manager_dispatch_signal_fd(sd_event_source *source, int fd, uint32_t revents, void *userdata); |
117 | static int manager_dispatch_time_change_fd(sd_event_source *source, int fd, uint32_t revents, void *userdata); | |
118 | static int manager_dispatch_idle_pipe_fd(sd_event_source *source, int fd, uint32_t revents, void *userdata); | |
00d9ef85 | 119 | static int manager_dispatch_user_lookup_fd(sd_event_source *source, int fd, uint32_t revents, void *userdata); |
718db961 | 120 | static int manager_dispatch_jobs_in_progress(sd_event_source *source, usec_t usec, void *userdata); |
752b5905 | 121 | static int manager_dispatch_run_queue(sd_event_source *source, void *userdata); |
575b300b | 122 | static int manager_dispatch_sigchld(sd_event_source *source, void *userdata); |
bbf5fd8e | 123 | static int manager_dispatch_timezone_change(sd_event_source *source, const struct inotify_event *event, void *userdata); |
64691d20 | 124 | static int manager_run_environment_generators(Manager *m); |
e801700e | 125 | static int manager_run_generators(Manager *m); |
06a4eb07 | 126 | static void manager_vacuum(Manager *m); |
718db961 | 127 | |
1b4154a8 | 128 | static usec_t manager_watch_jobs_next_time(Manager *m) { |
3889fc6f ZJS |
129 | usec_t timeout; |
130 | ||
131 | if (MANAGER_IS_USER(m)) | |
132 | /* Let the user manager without a timeout show status quickly, so the system manager can make | |
133 | * use of it, if it wants to. */ | |
134 | timeout = JOBS_IN_PROGRESS_WAIT_USEC * 2 / 3; | |
135 | else if (show_status_on(m->show_status)) | |
136 | /* When status is on, just use the usual timeout. */ | |
137 | timeout = JOBS_IN_PROGRESS_WAIT_USEC; | |
138 | else | |
139 | timeout = JOBS_IN_PROGRESS_QUIET_WAIT_USEC; | |
140 | ||
141 | return usec_add(now(CLOCK_MONOTONIC), timeout); | |
1b4154a8 ZJS |
142 | } |
143 | ||
2ae56591 | 144 | static void manager_watch_jobs_in_progress(Manager *m) { |
e5723c89 | 145 | usec_t next; |
cfa9677b | 146 | int r; |
e5723c89 | 147 | |
718db961 | 148 | assert(m); |
03b717a3 | 149 | |
42bf1ae1 FB |
150 | /* We do not want to show the cylon animation if the user |
151 | * needs to confirm service executions otherwise confirmation | |
152 | * messages will be screwed by the cylon animation. */ | |
b0eb2944 | 153 | if (!manager_is_confirm_spawn_disabled(m)) |
42bf1ae1 FB |
154 | return; |
155 | ||
718db961 | 156 | if (m->jobs_in_progress_event_source) |
2ae56591 | 157 | return; |
03b717a3 | 158 | |
1b4154a8 | 159 | next = manager_watch_jobs_next_time(m); |
cfa9677b | 160 | r = sd_event_add_time( |
6a0f1f6d LP |
161 | m->event, |
162 | &m->jobs_in_progress_event_source, | |
163 | CLOCK_MONOTONIC, | |
164 | next, 0, | |
165 | manager_dispatch_jobs_in_progress, m); | |
cfa9677b MM |
166 | if (r < 0) |
167 | return; | |
7dfbe2e3 TG |
168 | |
169 | (void) sd_event_source_set_description(m->jobs_in_progress_event_source, "manager-jobs-in-progress"); | |
03b717a3 MS |
170 | } |
171 | ||
fbd0b64f | 172 | #define CYLON_BUFFER_EXTRA (2*STRLEN(ANSI_RED) + STRLEN(ANSI_HIGHLIGHT_RED) + 2*STRLEN(ANSI_NORMAL)) |
03b717a3 | 173 | |
03b717a3 MS |
174 | static void draw_cylon(char buffer[], size_t buflen, unsigned width, unsigned pos) { |
175 | char *p = buffer; | |
176 | ||
177 | assert(buflen >= CYLON_BUFFER_EXTRA + width + 1); | |
178 | assert(pos <= width+1); /* 0 or width+1 mean that the center light is behind the corner */ | |
179 | ||
180 | if (pos > 1) { | |
6282c859 MS |
181 | if (pos > 2) |
182 | p = mempset(p, ' ', pos-2); | |
64c3610b FB |
183 | if (log_get_show_color()) |
184 | p = stpcpy(p, ANSI_RED); | |
03b717a3 MS |
185 | *p++ = '*'; |
186 | } | |
187 | ||
188 | if (pos > 0 && pos <= width) { | |
64c3610b FB |
189 | if (log_get_show_color()) |
190 | p = stpcpy(p, ANSI_HIGHLIGHT_RED); | |
03b717a3 MS |
191 | *p++ = '*'; |
192 | } | |
193 | ||
64c3610b FB |
194 | if (log_get_show_color()) |
195 | p = stpcpy(p, ANSI_NORMAL); | |
03b717a3 MS |
196 | |
197 | if (pos < width) { | |
64c3610b FB |
198 | if (log_get_show_color()) |
199 | p = stpcpy(p, ANSI_RED); | |
03b717a3 | 200 | *p++ = '*'; |
6282c859 MS |
201 | if (pos < width-1) |
202 | p = mempset(p, ' ', width-1-pos); | |
64c3610b FB |
203 | if (log_get_show_color()) |
204 | strcpy(p, ANSI_NORMAL); | |
03b717a3 | 205 | } |
03b717a3 MS |
206 | } |
207 | ||
1addc46c | 208 | static void manager_flip_auto_status(Manager *m, bool enable, const char *reason) { |
f755e3b7 LP |
209 | assert(m); |
210 | ||
cb8ccb22 ZJS |
211 | if (enable) { |
212 | if (m->show_status == SHOW_STATUS_AUTO) | |
7365a296 | 213 | manager_set_show_status(m, SHOW_STATUS_TEMPORARY, reason); |
cb8ccb22 ZJS |
214 | } else { |
215 | if (m->show_status == SHOW_STATUS_TEMPORARY) | |
7365a296 | 216 | manager_set_show_status(m, SHOW_STATUS_AUTO, reason); |
cb8ccb22 ZJS |
217 | } |
218 | } | |
219 | ||
03b717a3 | 220 | static void manager_print_jobs_in_progress(Manager *m) { |
03b717a3 | 221 | Job *j; |
03b717a3 MS |
222 | unsigned counter = 0, print_nr; |
223 | char cylon[6 + CYLON_BUFFER_EXTRA + 1]; | |
224 | unsigned cylon_pos; | |
4c989f89 | 225 | uint64_t timeout = 0; |
03b717a3 | 226 | |
718db961 | 227 | assert(m); |
9c3349e2 | 228 | assert(m->n_running_jobs > 0); |
718db961 | 229 | |
7365a296 | 230 | manager_flip_auto_status(m, true, "delay"); |
d450b6f2 | 231 | |
03b717a3 MS |
232 | print_nr = (m->jobs_in_progress_iteration / JOBS_IN_PROGRESS_PERIOD_DIVISOR) % m->n_running_jobs; |
233 | ||
90e74a66 | 234 | HASHMAP_FOREACH(j, m->jobs) |
03b717a3 MS |
235 | if (j->state == JOB_RUNNING && counter++ == print_nr) |
236 | break; | |
237 | ||
e970a72e MS |
238 | /* m->n_running_jobs must be consistent with the contents of m->jobs, |
239 | * so the above loop must have succeeded in finding j. */ | |
240 | assert(counter == print_nr + 1); | |
51d122af | 241 | assert(j); |
5a82a91a | 242 | |
03b717a3 MS |
243 | cylon_pos = m->jobs_in_progress_iteration % 14; |
244 | if (cylon_pos >= 8) | |
245 | cylon_pos = 14 - cylon_pos; | |
246 | draw_cylon(cylon, sizeof(cylon), 6, cylon_pos); | |
247 | ||
8bb310c3 ZJS |
248 | m->jobs_in_progress_iteration++; |
249 | ||
3889fc6f | 250 | char job_of_n[STRLEN("( of ) ") + DECIMAL_STR_MAX(unsigned)*2] = ""; |
5291f26d | 251 | if (m->n_running_jobs > 1) |
3889fc6f | 252 | xsprintf(job_of_n, "(%u of %u) ", counter, m->n_running_jobs); |
03b717a3 | 253 | |
4c989f89 | 254 | (void) job_get_timeout(j, &timeout); |
8bb310c3 | 255 | |
04d232d8 ZJS |
256 | /* We want to use enough information for the user to identify previous lines talking about the same |
257 | * unit, but keep the message as short as possible. So if 'Starting foo.service' or 'Starting | |
3889fc6f | 258 | * foo.service - Description' were used, 'foo.service' is enough here. On the other hand, if we used |
04d232d8 ZJS |
259 | * 'Starting Description' before, then we shall also use 'Description' here. So we pass NULL as the |
260 | * second argument to unit_status_string(). */ | |
261 | const char *ident = unit_status_string(j->unit, NULL); | |
262 | ||
3889fc6f | 263 | const char *time = FORMAT_TIMESPAN(now(CLOCK_MONOTONIC) - j->begin_usec, 1*USEC_PER_SEC); |
4c989f89 | 264 | const char *limit = timeout > 0 ? FORMAT_TIMESPAN(timeout - j->begin_usec, 1*USEC_PER_SEC) : "no limit"; |
3889fc6f ZJS |
265 | |
266 | if (m->status_unit_format == STATUS_UNIT_FORMAT_DESCRIPTION) | |
267 | /* When using 'Description', we effectively don't have enough space to show the nested status | |
268 | * without ellipsization, so let's not even try. */ | |
269 | manager_status_printf(m, STATUS_TYPE_EPHEMERAL, cylon, | |
270 | "%sA %s job is running for %s (%s / %s)", | |
271 | job_of_n, | |
272 | job_type_to_string(j->type), | |
273 | ident, | |
274 | time, limit); | |
275 | else { | |
276 | const char *status_text = unit_status_text(j->unit); | |
277 | ||
278 | manager_status_printf(m, STATUS_TYPE_EPHEMERAL, cylon, | |
279 | "%sJob %s/%s running (%s / %s)%s%s", | |
280 | job_of_n, | |
281 | ident, | |
282 | job_type_to_string(j->type), | |
283 | time, limit, | |
284 | status_text ? ": " : "", | |
285 | strempty(status_text)); | |
286 | } | |
287 | ||
288 | sd_notifyf(false, | |
289 | "STATUS=%sUser job %s/%s running (%s / %s)...", | |
290 | job_of_n, | |
291 | ident, | |
292 | job_type_to_string(j->type), | |
293 | time, limit); | |
294 | m->status_ready = false; | |
03b717a3 MS |
295 | } |
296 | ||
e46b13c8 | 297 | static int have_ask_password(void) { |
c2b2df60 | 298 | _cleanup_closedir_ DIR *dir = NULL; |
e46b13c8 ZJS |
299 | |
300 | dir = opendir("/run/systemd/ask-password"); | |
301 | if (!dir) { | |
302 | if (errno == ENOENT) | |
303 | return false; | |
304 | else | |
305 | return -errno; | |
306 | } | |
307 | ||
c7f0d9e5 | 308 | FOREACH_DIRENT_ALL(de, dir, return -errno) |
e46b13c8 ZJS |
309 | if (startswith(de->d_name, "ask.")) |
310 | return true; | |
8fb3f009 | 311 | return false; |
e46b13c8 ZJS |
312 | } |
313 | ||
314 | static int manager_dispatch_ask_password_fd(sd_event_source *source, | |
315 | int fd, uint32_t revents, void *userdata) { | |
99534007 | 316 | Manager *m = ASSERT_PTR(userdata); |
e46b13c8 | 317 | |
665dfe93 | 318 | (void) flush_fd(fd); |
e46b13c8 ZJS |
319 | |
320 | m->have_ask_password = have_ask_password(); | |
321 | if (m->have_ask_password < 0) | |
322 | /* Log error but continue. Negative have_ask_password | |
323 | * is treated as unknown status. */ | |
c33b3297 | 324 | log_error_errno(m->have_ask_password, "Failed to list /run/systemd/ask-password: %m"); |
e46b13c8 ZJS |
325 | |
326 | return 0; | |
327 | } | |
328 | ||
329 | static void manager_close_ask_password(Manager *m) { | |
330 | assert(m); | |
331 | ||
5dcadb4c | 332 | m->ask_password_event_source = sd_event_source_disable_unref(m->ask_password_event_source); |
90990e28 | 333 | m->ask_password_inotify_fd = safe_close(m->ask_password_inotify_fd); |
e46b13c8 ZJS |
334 | m->have_ask_password = -EINVAL; |
335 | } | |
336 | ||
337 | static int manager_check_ask_password(Manager *m) { | |
338 | int r; | |
339 | ||
340 | assert(m); | |
341 | ||
342 | if (!m->ask_password_event_source) { | |
343 | assert(m->ask_password_inotify_fd < 0); | |
344 | ||
17e4b070 | 345 | (void) mkdir_p_label("/run/systemd/ask-password", 0755); |
e46b13c8 ZJS |
346 | |
347 | m->ask_password_inotify_fd = inotify_init1(IN_NONBLOCK|IN_CLOEXEC); | |
4a62c710 | 348 | if (m->ask_password_inotify_fd < 0) |
330b8fb3 | 349 | return log_error_errno(errno, "Failed to create inotify object: %m"); |
e46b13c8 | 350 | |
27c3112d FB |
351 | r = inotify_add_watch_and_warn(m->ask_password_inotify_fd, |
352 | "/run/systemd/ask-password", | |
353 | IN_CREATE|IN_DELETE|IN_MOVE); | |
354 | if (r < 0) { | |
e46b13c8 | 355 | manager_close_ask_password(m); |
27c3112d | 356 | return r; |
e46b13c8 ZJS |
357 | } |
358 | ||
359 | r = sd_event_add_io(m->event, &m->ask_password_event_source, | |
360 | m->ask_password_inotify_fd, EPOLLIN, | |
361 | manager_dispatch_ask_password_fd, m); | |
362 | if (r < 0) { | |
df3d3bdf | 363 | log_error_errno(r, "Failed to add event source for /run/systemd/ask-password: %m"); |
e46b13c8 | 364 | manager_close_ask_password(m); |
df3d3bdf | 365 | return r; |
e46b13c8 ZJS |
366 | } |
367 | ||
7dfbe2e3 TG |
368 | (void) sd_event_source_set_description(m->ask_password_event_source, "manager-ask-password"); |
369 | ||
e46b13c8 ZJS |
370 | /* Queries might have been added meanwhile... */ |
371 | manager_dispatch_ask_password_fd(m->ask_password_event_source, | |
372 | m->ask_password_inotify_fd, EPOLLIN, m); | |
373 | } | |
374 | ||
375 | return m->have_ask_password; | |
376 | } | |
377 | ||
31a7eb86 | 378 | static int manager_watch_idle_pipe(Manager *m) { |
31a7eb86 ZJS |
379 | int r; |
380 | ||
718db961 LP |
381 | assert(m); |
382 | ||
383 | if (m->idle_pipe_event_source) | |
31a7eb86 ZJS |
384 | return 0; |
385 | ||
386 | if (m->idle_pipe[2] < 0) | |
387 | return 0; | |
388 | ||
151b9b96 | 389 | 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 |
390 | if (r < 0) |
391 | return log_error_errno(r, "Failed to watch idle pipe: %m"); | |
31a7eb86 | 392 | |
7dfbe2e3 TG |
393 | (void) sd_event_source_set_description(m->idle_pipe_event_source, "manager-idle-pipe"); |
394 | ||
31a7eb86 | 395 | return 0; |
31a7eb86 ZJS |
396 | } |
397 | ||
718db961 LP |
398 | static void manager_close_idle_pipe(Manager *m) { |
399 | assert(m); | |
31a7eb86 | 400 | |
5dcadb4c | 401 | m->idle_pipe_event_source = sd_event_source_disable_unref(m->idle_pipe_event_source); |
cd72bd8a | 402 | |
3d94f76c LP |
403 | safe_close_pair(m->idle_pipe); |
404 | safe_close_pair(m->idle_pipe + 2); | |
31a7eb86 ZJS |
405 | } |
406 | ||
8742514c | 407 | static int manager_setup_time_change(Manager *m) { |
718db961 | 408 | int r; |
b92bea5d | 409 | |
718db961 | 410 | assert(m); |
8742514c | 411 | |
638cece4 | 412 | if (MANAGER_IS_TEST_RUN(m)) |
0d8c31ff ZJS |
413 | return 0; |
414 | ||
5dcadb4c | 415 | m->time_change_event_source = sd_event_source_disable_unref(m->time_change_event_source); |
7feedd18 | 416 | |
ec75e8e0 | 417 | r = event_add_time_change(m->event, &m->time_change_event_source, manager_dispatch_time_change_fd, m); |
23bbb0de MS |
418 | if (r < 0) |
419 | return log_error_errno(r, "Failed to create time change event source: %m"); | |
8742514c | 420 | |
a5cc7e5a LP |
421 | /* Schedule this slightly earlier than the .timer event sources */ |
422 | r = sd_event_source_set_priority(m->time_change_event_source, SD_EVENT_PRIORITY_NORMAL-1); | |
423 | if (r < 0) | |
424 | return log_error_errno(r, "Failed to set priority of time change event sources: %m"); | |
425 | ||
8742514c LP |
426 | log_debug("Set up TFD_TIMER_CANCEL_ON_SET timerfd."); |
427 | ||
428 | return 0; | |
429 | } | |
430 | ||
bbf5fd8e LP |
431 | static int manager_read_timezone_stat(Manager *m) { |
432 | struct stat st; | |
433 | bool changed; | |
434 | ||
435 | assert(m); | |
436 | ||
437 | /* Read the current stat() data of /etc/localtime so that we detect changes */ | |
438 | if (lstat("/etc/localtime", &st) < 0) { | |
439 | log_debug_errno(errno, "Failed to stat /etc/localtime, ignoring: %m"); | |
440 | changed = m->etc_localtime_accessible; | |
441 | m->etc_localtime_accessible = false; | |
442 | } else { | |
443 | usec_t k; | |
444 | ||
445 | k = timespec_load(&st.st_mtim); | |
446 | changed = !m->etc_localtime_accessible || k != m->etc_localtime_mtime; | |
447 | ||
448 | m->etc_localtime_mtime = k; | |
449 | m->etc_localtime_accessible = true; | |
450 | } | |
451 | ||
452 | return changed; | |
453 | } | |
454 | ||
455 | static int manager_setup_timezone_change(Manager *m) { | |
a5cc7e5a | 456 | _cleanup_(sd_event_source_unrefp) sd_event_source *new_event = NULL; |
bbf5fd8e LP |
457 | int r; |
458 | ||
459 | assert(m); | |
460 | ||
638cece4 | 461 | if (MANAGER_IS_TEST_RUN(m)) |
bbf5fd8e LP |
462 | return 0; |
463 | ||
464 | /* We watch /etc/localtime for three events: change of the link count (which might mean removal from /etc even | |
465 | * though another link might be kept), renames, and file close operations after writing. Note we don't bother | |
466 | * with IN_DELETE_SELF, as that would just report when the inode is removed entirely, i.e. after the link count | |
467 | * went to zero and all fds to it are closed. | |
468 | * | |
469 | * Note that we never follow symlinks here. This is a simplification, but should cover almost all cases | |
470 | * correctly. | |
471 | * | |
472 | * Note that we create the new event source first here, before releasing the old one. This should optimize | |
473 | * behaviour as this way sd-event can reuse the old watch in case the inode didn't change. */ | |
474 | ||
475 | r = sd_event_add_inotify(m->event, &new_event, "/etc/localtime", | |
476 | IN_ATTRIB|IN_MOVE_SELF|IN_CLOSE_WRITE|IN_DONT_FOLLOW, manager_dispatch_timezone_change, m); | |
0cb21d8c LP |
477 | if (r == -ENOENT) { |
478 | /* If the file doesn't exist yet, subscribe to /etc instead, and wait until it is created either by | |
479 | * O_CREATE or by rename() */ | |
480 | ||
481 | log_debug_errno(r, "/etc/localtime doesn't exist yet, watching /etc instead."); | |
bbf5fd8e LP |
482 | r = sd_event_add_inotify(m->event, &new_event, "/etc", |
483 | IN_CREATE|IN_MOVED_TO|IN_ONLYDIR, manager_dispatch_timezone_change, m); | |
0cb21d8c | 484 | } |
bbf5fd8e LP |
485 | if (r < 0) |
486 | return log_error_errno(r, "Failed to create timezone change event source: %m"); | |
487 | ||
a5cc7e5a LP |
488 | /* Schedule this slightly earlier than the .timer event sources */ |
489 | r = sd_event_source_set_priority(new_event, SD_EVENT_PRIORITY_NORMAL-1); | |
490 | if (r < 0) | |
491 | return log_error_errno(r, "Failed to set priority of timezone change event sources: %m"); | |
492 | ||
bbf5fd8e | 493 | sd_event_source_unref(m->timezone_change_event_source); |
a5cc7e5a | 494 | m->timezone_change_event_source = TAKE_PTR(new_event); |
bbf5fd8e LP |
495 | |
496 | return 0; | |
497 | } | |
498 | ||
80876c20 | 499 | static int enable_special_signals(Manager *m) { |
254d1313 | 500 | _cleanup_close_ int fd = -EBADF; |
80876c20 LP |
501 | |
502 | assert(m); | |
503 | ||
638cece4 | 504 | if (MANAGER_IS_TEST_RUN(m)) |
37453b3a EV |
505 | return 0; |
506 | ||
a41b539e | 507 | /* Enable that we get SIGINT on control-alt-del. In containers |
c9999773 LP |
508 | * this will fail with EPERM (older) or EINVAL (newer), so |
509 | * ignore that. */ | |
4c701096 | 510 | if (reboot(RB_DISABLE_CAD) < 0 && !IN_SET(errno, EPERM, EINVAL)) |
56f64d95 | 511 | log_warning_errno(errno, "Failed to enable ctrl-alt-del handling: %m"); |
80876c20 | 512 | |
a41b539e LP |
513 | fd = open_terminal("/dev/tty0", O_RDWR|O_NOCTTY|O_CLOEXEC); |
514 | if (fd < 0) { | |
515 | /* Support systems without virtual console */ | |
516 | if (fd != -ENOENT) | |
56f64d95 | 517 | log_warning_errno(errno, "Failed to open /dev/tty0: %m"); |
a41b539e | 518 | } else { |
80876c20 LP |
519 | /* Enable that we get SIGWINCH on kbrequest */ |
520 | if (ioctl(fd, KDSIGACCEPT, SIGWINCH) < 0) | |
56f64d95 | 521 | log_warning_errno(errno, "Failed to enable kbrequest handling: %m"); |
80876c20 LP |
522 | } |
523 | ||
524 | return 0; | |
525 | } | |
526 | ||
8750ac02 ZJS |
527 | #define RTSIG_IF_AVAILABLE(signum) (signum <= SIGRTMAX ? signum : -1) |
528 | ||
ce578209 | 529 | static int manager_setup_signals(Manager *m) { |
b92bea5d ZJS |
530 | struct sigaction sa = { |
531 | .sa_handler = SIG_DFL, | |
532 | .sa_flags = SA_NOCLDSTOP|SA_RESTART, | |
533 | }; | |
718db961 LP |
534 | sigset_t mask; |
535 | int r; | |
60918275 | 536 | |
ce578209 LP |
537 | assert(m); |
538 | ||
57c0c30e LP |
539 | assert_se(sigaction(SIGCHLD, &sa, NULL) == 0); |
540 | ||
4dffec14 LP |
541 | /* We make liberal use of realtime signals here. On |
542 | * Linux/glibc we have 30 of them (with the exception of Linux | |
543 | * on hppa, see below), between SIGRTMIN+0 ... SIGRTMIN+30 | |
544 | * (aka SIGRTMAX). */ | |
7d793605 | 545 | |
4dffec14 | 546 | assert_se(sigemptyset(&mask) == 0); |
7d793605 LP |
547 | sigset_add_many(&mask, |
548 | SIGCHLD, /* Child died */ | |
549 | SIGTERM, /* Reexecute daemon */ | |
550 | SIGHUP, /* Reload configuration */ | |
54d04cd1 | 551 | SIGUSR1, /* systemd: reconnect to D-Bus */ |
7d793605 LP |
552 | SIGUSR2, /* systemd: dump status */ |
553 | SIGINT, /* Kernel sends us this on control-alt-del */ | |
554 | SIGWINCH, /* Kernel sends us this on kbrequest (alt-arrowup) */ | |
555 | SIGPWR, /* Some kernel drivers and upsd send us this on power failure */ | |
4dffec14 | 556 | |
7d793605 | 557 | SIGRTMIN+0, /* systemd: start default.target */ |
0003d1ab | 558 | SIGRTMIN+1, /* systemd: isolate rescue.target */ |
7d793605 LP |
559 | SIGRTMIN+2, /* systemd: isolate emergency.target */ |
560 | SIGRTMIN+3, /* systemd: start halt.target */ | |
561 | SIGRTMIN+4, /* systemd: start poweroff.target */ | |
562 | SIGRTMIN+5, /* systemd: start reboot.target */ | |
0003d1ab | 563 | SIGRTMIN+6, /* systemd: start kexec.target */ |
4dffec14 LP |
564 | |
565 | /* ... space for more special targets ... */ | |
566 | ||
0003d1ab LP |
567 | SIGRTMIN+13, /* systemd: Immediate halt */ |
568 | SIGRTMIN+14, /* systemd: Immediate poweroff */ | |
569 | SIGRTMIN+15, /* systemd: Immediate reboot */ | |
570 | SIGRTMIN+16, /* systemd: Immediate kexec */ | |
4dffec14 | 571 | |
29e6b0c1 LP |
572 | /* ... space for one more immediate system state change ... */ |
573 | ||
574 | SIGRTMIN+18, /* systemd: control command */ | |
575 | ||
576 | /* ... space ... */ | |
4dffec14 | 577 | |
0658666b LP |
578 | SIGRTMIN+20, /* systemd: enable status messages */ |
579 | SIGRTMIN+21, /* systemd: disable status messages */ | |
253ee27a LP |
580 | SIGRTMIN+22, /* systemd: set log level to LOG_DEBUG */ |
581 | SIGRTMIN+23, /* systemd: set log level to LOG_INFO */ | |
600b704e | 582 | SIGRTMIN+24, /* systemd: Immediate exit (--user only) */ |
463aef23 | 583 | SIGRTMIN+25, /* systemd: reexecute manager */ |
4dffec14 | 584 | |
8750ac02 ZJS |
585 | /* Apparently Linux on hppa had fewer RT signals until v3.18, |
586 | * SIGRTMAX was SIGRTMIN+25, and then SIGRTMIN was lowered, | |
587 | * see commit v3.17-7614-g1f25df2eff. | |
588 | * | |
589 | * We cannot unconditionally make use of those signals here, | |
590 | * so let's use a runtime check. Since these commands are | |
591 | * accessible by different means and only really a safety | |
592 | * net, the missing functionality on hppa shouldn't matter. | |
593 | */ | |
594 | ||
595 | RTSIG_IF_AVAILABLE(SIGRTMIN+26), /* systemd: set log target to journal-or-kmsg */ | |
596 | RTSIG_IF_AVAILABLE(SIGRTMIN+27), /* systemd: set log target to console */ | |
597 | RTSIG_IF_AVAILABLE(SIGRTMIN+28), /* systemd: set log target to kmsg */ | |
598 | RTSIG_IF_AVAILABLE(SIGRTMIN+29), /* systemd: set log target to syslog-or-kmsg (obsolete) */ | |
4dffec14 LP |
599 | |
600 | /* ... one free signal here SIGRTMIN+30 ... */ | |
7d793605 | 601 | -1); |
ce578209 LP |
602 | assert_se(sigprocmask(SIG_SETMASK, &mask, NULL) == 0); |
603 | ||
718db961 LP |
604 | m->signal_fd = signalfd(-1, &mask, SFD_NONBLOCK|SFD_CLOEXEC); |
605 | if (m->signal_fd < 0) | |
ce578209 LP |
606 | return -errno; |
607 | ||
151b9b96 | 608 | r = sd_event_add_io(m->event, &m->signal_event_source, m->signal_fd, EPOLLIN, manager_dispatch_signal_fd, m); |
718db961 LP |
609 | if (r < 0) |
610 | return r; | |
ce578209 | 611 | |
7dfbe2e3 TG |
612 | (void) sd_event_source_set_description(m->signal_event_source, "manager-signal"); |
613 | ||
d8fdc620 LP |
614 | /* Process signals a bit earlier than the rest of things, but later than notify_fd processing, so that the |
615 | * notify processing can still figure out to which process/service a message belongs, before we reap the | |
616 | * process. Also, process this before handling cgroup notifications, so that we always collect child exit | |
617 | * status information before detecting that there's no process in a cgroup. */ | |
618 | r = sd_event_source_set_priority(m->signal_event_source, SD_EVENT_PRIORITY_NORMAL-6); | |
29083707 LP |
619 | if (r < 0) |
620 | return r; | |
621 | ||
463d0d15 | 622 | if (MANAGER_IS_SYSTEM(m)) |
80876c20 | 623 | return enable_special_signals(m); |
e1414003 | 624 | |
ce578209 LP |
625 | return 0; |
626 | } | |
627 | ||
1ad6e8b3 | 628 | static char** sanitize_environment(char **l) { |
f069efb4 | 629 | |
47cf8ff2 | 630 | /* Let's remove some environment variables that we need ourselves to communicate with our clients */ |
f069efb4 | 631 | strv_env_unset_many( |
1ad6e8b3 | 632 | l, |
8047ac8f LP |
633 | "CACHE_DIRECTORY", |
634 | "CONFIGURATION_DIRECTORY", | |
bb0c0d6f | 635 | "CREDENTIALS_DIRECTORY", |
47cf8ff2 LP |
636 | "EXIT_CODE", |
637 | "EXIT_STATUS", | |
638 | "INVOCATION_ID", | |
639 | "JOURNAL_STREAM", | |
640 | "LISTEN_FDNAMES", | |
641 | "LISTEN_FDS", | |
642 | "LISTEN_PID", | |
8047ac8f | 643 | "LOGS_DIRECTORY", |
c792a2e5 | 644 | "LOG_NAMESPACE", |
f069efb4 LP |
645 | "MAINPID", |
646 | "MANAGERPID", | |
6bb00842 LP |
647 | "MEMORY_PRESSURE_WATCH", |
648 | "MEMORY_PRESSURE_WRITE", | |
c792a2e5 LP |
649 | "MONITOR_EXIT_CODE", |
650 | "MONITOR_EXIT_STATUS", | |
651 | "MONITOR_INVOCATION_ID", | |
652 | "MONITOR_SERVICE_RESULT", | |
653 | "MONITOR_UNIT", | |
47cf8ff2 | 654 | "NOTIFY_SOCKET", |
dcf3c3c3 | 655 | "PIDFILE", |
47cf8ff2 LP |
656 | "REMOTE_ADDR", |
657 | "REMOTE_PORT", | |
8047ac8f | 658 | "RUNTIME_DIRECTORY", |
47cf8ff2 | 659 | "SERVICE_RESULT", |
8047ac8f | 660 | "STATE_DIRECTORY", |
c792a2e5 LP |
661 | "SYSTEMD_EXEC_PID", |
662 | "TRIGGER_PATH", | |
663 | "TRIGGER_TIMER_MONOTONIC_USEC", | |
664 | "TRIGGER_TIMER_REALTIME_USEC", | |
665 | "TRIGGER_UNIT", | |
f069efb4 LP |
666 | "WATCHDOG_PID", |
667 | "WATCHDOG_USEC", | |
668 | NULL); | |
47cf8ff2 LP |
669 | |
670 | /* Let's order the environment alphabetically, just to make it pretty */ | |
305757d8 | 671 | return strv_sort(l); |
f069efb4 LP |
672 | } |
673 | ||
79a224c4 | 674 | int manager_default_environment(Manager *m) { |
db11487d ZJS |
675 | int r; |
676 | ||
71ecc858 LP |
677 | assert(m); |
678 | ||
1ad6e8b3 LP |
679 | m->transient_environment = strv_free(m->transient_environment); |
680 | ||
463d0d15 | 681 | if (MANAGER_IS_SYSTEM(m)) { |
c792a2e5 LP |
682 | /* The system manager always starts with a clean environment for its children. It does not |
683 | * import the kernel's or the parents' exported variables. | |
e21fea24 | 684 | * |
c792a2e5 LP |
685 | * The initial passed environment is untouched to keep /proc/self/environ valid; it is used |
686 | * for tagging the init process inside containers. */ | |
1ad6e8b3 | 687 | m->transient_environment = strv_new("PATH=" DEFAULT_PATH); |
db11487d ZJS |
688 | if (!m->transient_environment) |
689 | return log_oom(); | |
e21fea24 KS |
690 | |
691 | /* Import locale variables LC_*= from configuration */ | |
1ad6e8b3 | 692 | (void) locale_setup(&m->transient_environment); |
db11487d | 693 | } else { |
aaf057c4 | 694 | /* The user manager passes its own environment along to its children, except for $PATH. */ |
1ad6e8b3 | 695 | m->transient_environment = strv_copy(environ); |
db11487d ZJS |
696 | if (!m->transient_environment) |
697 | return log_oom(); | |
698 | ||
aaf057c4 | 699 | r = strv_env_replace_strdup(&m->transient_environment, "PATH=" DEFAULT_USER_PATH); |
db11487d ZJS |
700 | if (r < 0) |
701 | return log_oom(); | |
db11487d | 702 | } |
8b55b8c4 | 703 | |
1ad6e8b3 | 704 | sanitize_environment(m->transient_environment); |
e21fea24 | 705 | return 0; |
71ecc858 LP |
706 | } |
707 | ||
3536f49e YW |
708 | static int manager_setup_prefix(Manager *m) { |
709 | struct table_entry { | |
710 | uint64_t type; | |
711 | const char *suffix; | |
712 | }; | |
713 | ||
72fd1768 | 714 | static const struct table_entry paths_system[_EXEC_DIRECTORY_TYPE_MAX] = { |
9978e631 ZJS |
715 | [EXEC_DIRECTORY_RUNTIME] = { SD_PATH_SYSTEM_RUNTIME, NULL }, |
716 | [EXEC_DIRECTORY_STATE] = { SD_PATH_SYSTEM_STATE_PRIVATE, NULL }, | |
717 | [EXEC_DIRECTORY_CACHE] = { SD_PATH_SYSTEM_STATE_CACHE, NULL }, | |
718 | [EXEC_DIRECTORY_LOGS] = { SD_PATH_SYSTEM_STATE_LOGS, NULL }, | |
3536f49e YW |
719 | [EXEC_DIRECTORY_CONFIGURATION] = { SD_PATH_SYSTEM_CONFIGURATION, NULL }, |
720 | }; | |
721 | ||
72fd1768 | 722 | static const struct table_entry paths_user[_EXEC_DIRECTORY_TYPE_MAX] = { |
9978e631 ZJS |
723 | [EXEC_DIRECTORY_RUNTIME] = { SD_PATH_USER_RUNTIME, NULL }, |
724 | [EXEC_DIRECTORY_STATE] = { SD_PATH_USER_CONFIGURATION, NULL }, | |
725 | [EXEC_DIRECTORY_CACHE] = { SD_PATH_USER_STATE_CACHE, NULL }, | |
726 | [EXEC_DIRECTORY_LOGS] = { SD_PATH_USER_CONFIGURATION, "log" }, | |
727 | [EXEC_DIRECTORY_CONFIGURATION] = { SD_PATH_USER_CONFIGURATION, NULL }, | |
3536f49e YW |
728 | }; |
729 | ||
3536f49e YW |
730 | assert(m); |
731 | ||
24b45970 ZJS |
732 | const struct table_entry *p = MANAGER_IS_SYSTEM(m) ? paths_system : paths_user; |
733 | int r; | |
3536f49e | 734 | |
24b45970 | 735 | for (ExecDirectoryType i = 0; i < _EXEC_DIRECTORY_TYPE_MAX; i++) { |
51327bcc | 736 | r = sd_path_lookup(p[i].type, p[i].suffix, &m->prefix[i]); |
3536f49e | 737 | if (r < 0) |
998df7ce ZJS |
738 | return log_warning_errno(r, "Failed to lookup %s path: %m", |
739 | exec_directory_type_to_string(i)); | |
3536f49e YW |
740 | } |
741 | ||
742 | return 0; | |
743 | } | |
744 | ||
e8630e69 ZJS |
745 | static void manager_free_unit_name_maps(Manager *m) { |
746 | m->unit_id_map = hashmap_free(m->unit_id_map); | |
747 | m->unit_name_map = hashmap_free(m->unit_name_map); | |
3fb2326f | 748 | m->unit_path_cache = set_free(m->unit_path_cache); |
c2911d48 | 749 | m->unit_cache_timestamp_hash = 0; |
e8630e69 ZJS |
750 | } |
751 | ||
279d81dd LP |
752 | static int manager_setup_run_queue(Manager *m) { |
753 | int r; | |
754 | ||
755 | assert(m); | |
756 | assert(!m->run_queue_event_source); | |
757 | ||
758 | r = sd_event_add_defer(m->event, &m->run_queue_event_source, manager_dispatch_run_queue, m); | |
759 | if (r < 0) | |
760 | return r; | |
761 | ||
762 | r = sd_event_source_set_priority(m->run_queue_event_source, SD_EVENT_PRIORITY_IDLE); | |
763 | if (r < 0) | |
764 | return r; | |
765 | ||
766 | r = sd_event_source_set_enabled(m->run_queue_event_source, SD_EVENT_OFF); | |
767 | if (r < 0) | |
768 | return r; | |
769 | ||
770 | (void) sd_event_source_set_description(m->run_queue_event_source, "manager-run-queue"); | |
771 | ||
772 | return 0; | |
773 | } | |
774 | ||
575b300b LP |
775 | static int manager_setup_sigchld_event_source(Manager *m) { |
776 | int r; | |
777 | ||
778 | assert(m); | |
779 | assert(!m->sigchld_event_source); | |
780 | ||
781 | r = sd_event_add_defer(m->event, &m->sigchld_event_source, manager_dispatch_sigchld, m); | |
782 | if (r < 0) | |
783 | return r; | |
784 | ||
785 | r = sd_event_source_set_priority(m->sigchld_event_source, SD_EVENT_PRIORITY_NORMAL-7); | |
786 | if (r < 0) | |
787 | return r; | |
788 | ||
789 | r = sd_event_source_set_enabled(m->sigchld_event_source, SD_EVENT_OFF); | |
790 | if (r < 0) | |
791 | return r; | |
792 | ||
793 | (void) sd_event_source_set_description(m->sigchld_event_source, "manager-sigchld"); | |
794 | ||
795 | return 0; | |
796 | } | |
797 | ||
29e6b0c1 LP |
798 | int manager_setup_memory_pressure_event_source(Manager *m) { |
799 | int r; | |
800 | ||
801 | assert(m); | |
802 | ||
803 | m->memory_pressure_event_source = sd_event_source_disable_unref(m->memory_pressure_event_source); | |
804 | ||
805 | r = sd_event_add_memory_pressure(m->event, &m->memory_pressure_event_source, NULL, NULL); | |
806 | if (r < 0) | |
807 | log_full_errno(ERRNO_IS_NOT_SUPPORTED(r) || ERRNO_IS_PRIVILEGE(r) || (r == -EHOSTDOWN) ? LOG_DEBUG : LOG_NOTICE, r, | |
808 | "Failed to establish memory pressure event source, ignoring: %m"); | |
6bb00842 LP |
809 | else if (m->default_memory_pressure_threshold_usec != USEC_INFINITY) { |
810 | ||
811 | /* If there's a default memory pressure threshold set, also apply it to the service manager itself */ | |
812 | r = sd_event_source_set_memory_pressure_period( | |
813 | m->memory_pressure_event_source, | |
814 | m->default_memory_pressure_threshold_usec, | |
815 | MEMORY_PRESSURE_DEFAULT_WINDOW_USEC); | |
816 | if (r < 0) | |
817 | log_warning_errno(r, "Failed to adjust memory pressure threshold, ignoring: %m"); | |
818 | } | |
29e6b0c1 LP |
819 | |
820 | return 0; | |
821 | } | |
822 | ||
2ad591a3 LP |
823 | static int manager_find_credentials_dirs(Manager *m) { |
824 | const char *e; | |
825 | int r; | |
826 | ||
827 | assert(m); | |
828 | ||
829 | r = get_credentials_dir(&e); | |
830 | if (r < 0) { | |
831 | if (r != -ENXIO) | |
832 | log_debug_errno(r, "Failed to determine credentials directory, ignoring: %m"); | |
833 | } else { | |
834 | m->received_credentials_directory = strdup(e); | |
835 | if (!m->received_credentials_directory) | |
836 | return -ENOMEM; | |
837 | } | |
838 | ||
839 | r = get_encrypted_credentials_dir(&e); | |
840 | if (r < 0) { | |
841 | if (r != -ENXIO) | |
842 | log_debug_errno(r, "Failed to determine encrypted credentials directory, ignoring: %m"); | |
843 | } else { | |
844 | m->received_encrypted_credentials_directory = strdup(e); | |
845 | if (!m->received_encrypted_credentials_directory) | |
846 | return -ENOMEM; | |
847 | } | |
848 | ||
849 | return 0; | |
850 | } | |
851 | ||
d35fe8c0 | 852 | void manager_set_switching_root(Manager *m, bool switching_root) { |
7fa49280 LP |
853 | assert(m); |
854 | ||
d35fe8c0 FB |
855 | m->switching_root = MANAGER_IS_SYSTEM(m) && switching_root; |
856 | } | |
857 | ||
4870133b | 858 | int manager_new(RuntimeScope runtime_scope, ManagerTestRunFlags test_run_flags, Manager **_m) { |
c70cac54 | 859 | _cleanup_(manager_freep) Manager *m = NULL; |
e3dd987c | 860 | int r; |
8e274523 LP |
861 | |
862 | assert(_m); | |
4870133b | 863 | assert(IN_SET(runtime_scope, RUNTIME_SCOPE_SYSTEM, RUNTIME_SCOPE_USER)); |
ce578209 | 864 | |
3ad228ce | 865 | m = new(Manager, 1); |
915b3753 | 866 | if (!m) |
8e274523 | 867 | return -ENOMEM; |
60918275 | 868 | |
3ad228ce | 869 | *m = (Manager) { |
4870133b | 870 | .runtime_scope = runtime_scope, |
3ad228ce LP |
871 | .objective = _MANAGER_OBJECTIVE_INVALID, |
872 | ||
36cf4507 ZJS |
873 | .status_unit_format = STATUS_UNIT_FORMAT_DEFAULT, |
874 | ||
3ad228ce LP |
875 | .default_timer_accuracy_usec = USEC_PER_MINUTE, |
876 | .default_memory_accounting = MEMORY_ACCOUNTING_DEFAULT, | |
877 | .default_tasks_accounting = true, | |
3a0f06c4 | 878 | .default_tasks_max = TASKS_MAX_UNSET, |
4870133b LP |
879 | .default_timeout_start_usec = manager_default_timeout(runtime_scope), |
880 | .default_timeout_stop_usec = manager_default_timeout(runtime_scope), | |
3ad228ce | 881 | .default_restart_usec = DEFAULT_RESTART_USEC, |
4870133b | 882 | .default_device_timeout_usec = manager_default_timeout(runtime_scope), |
3ad228ce LP |
883 | |
884 | .original_log_level = -1, | |
885 | .original_log_target = _LOG_TARGET_INVALID, | |
886 | ||
986935cf FB |
887 | .watchdog_overridden[WATCHDOG_RUNTIME] = USEC_INFINITY, |
888 | .watchdog_overridden[WATCHDOG_REBOOT] = USEC_INFINITY, | |
889 | .watchdog_overridden[WATCHDOG_KEXEC] = USEC_INFINITY, | |
5717062e | 890 | .watchdog_overridden[WATCHDOG_PRETIMEOUT] = USEC_INFINITY, |
986935cf | 891 | |
44a41954 FB |
892 | .show_status_overridden = _SHOW_STATUS_INVALID, |
893 | ||
254d1313 ZJS |
894 | .notify_fd = -EBADF, |
895 | .cgroups_agent_fd = -EBADF, | |
896 | .signal_fd = -EBADF, | |
19ee48a6 | 897 | .user_lookup_fds = PIPE_EBADF, |
254d1313 ZJS |
898 | .private_listen_fd = -EBADF, |
899 | .dev_autofs_fd = -EBADF, | |
900 | .cgroup_inotify_fd = -EBADF, | |
901 | .pin_cgroupfs_fd = -EBADF, | |
902 | .ask_password_inotify_fd = -EBADF, | |
903 | .idle_pipe = { -EBADF, -EBADF, -EBADF, -EBADF}, | |
3ad228ce LP |
904 | |
905 | /* start as id #1, so that we can leave #0 around as "null-like" value */ | |
906 | .current_job_id = 1, | |
907 | ||
908 | .have_ask_password = -EINVAL, /* we don't know */ | |
909 | .first_boot = -1, | |
910 | .test_run_flags = test_run_flags, | |
afcfaa69 LP |
911 | |
912 | .default_oom_policy = OOM_STOP, | |
6bb00842 LP |
913 | |
914 | .default_memory_pressure_watch = CGROUP_PRESSURE_WATCH_AUTO, | |
915 | .default_memory_pressure_threshold_usec = USEC_INFINITY, | |
d9365956 LB |
916 | |
917 | .dump_ratelimit = { | |
918 | .interval = 10 * USEC_PER_MINUTE, | |
919 | .burst = 10, | |
920 | }, | |
3ad228ce | 921 | }; |
80876c20 | 922 | |
349cc4a5 | 923 | #if ENABLE_EFI |
463d0d15 | 924 | if (MANAGER_IS_SYSTEM(m) && detect_container() <= 0) |
9f9f0342 LP |
925 | boot_timestamps(m->timestamps + MANAGER_TIMESTAMP_USERSPACE, |
926 | m->timestamps + MANAGER_TIMESTAMP_FIRMWARE, | |
927 | m->timestamps + MANAGER_TIMESTAMP_LOADER); | |
463d0d15 LP |
928 | #endif |
929 | ||
f2341e0a | 930 | /* Prepare log fields we can use for structured logging */ |
463d0d15 LP |
931 | if (MANAGER_IS_SYSTEM(m)) { |
932 | m->unit_log_field = "UNIT="; | |
933 | m->unit_log_format_string = "UNIT=%s"; | |
4b58153d LP |
934 | |
935 | m->invocation_log_field = "INVOCATION_ID="; | |
f1c50bec | 936 | m->invocation_log_format_string = "INVOCATION_ID=%s"; |
463d0d15 LP |
937 | } else { |
938 | m->unit_log_field = "USER_UNIT="; | |
939 | m->unit_log_format_string = "USER_UNIT=%s"; | |
4b58153d LP |
940 | |
941 | m->invocation_log_field = "USER_INVOCATION_ID="; | |
f1c50bec | 942 | m->invocation_log_format_string = "USER_INVOCATION_ID=%s"; |
463d0d15 | 943 | } |
f2341e0a | 944 | |
2e5c94b9 | 945 | /* Reboot immediately if the user hits C-A-D more often than 7x per 2s */ |
7d1e61ca | 946 | m->ctrl_alt_del_ratelimit = (const RateLimit) { .interval = 2 * USEC_PER_SEC, .burst = 7 }; |
2e5c94b9 | 947 | |
e21fea24 KS |
948 | r = manager_default_environment(m); |
949 | if (r < 0) | |
c70cac54 | 950 | return r; |
1137a57c | 951 | |
d5099efc | 952 | r = hashmap_ensure_allocated(&m->units, &string_hash_ops); |
718db961 | 953 | if (r < 0) |
c70cac54 | 954 | return r; |
60918275 | 955 | |
548f6937 | 956 | r = hashmap_ensure_allocated(&m->cgroup_unit, &path_hash_ops); |
718db961 | 957 | if (r < 0) |
c70cac54 | 958 | return r; |
9152c765 | 959 | |
d5099efc | 960 | r = hashmap_ensure_allocated(&m->watch_bus, &string_hash_ops); |
718db961 | 961 | if (r < 0) |
c70cac54 | 962 | return r; |
05e343b7 | 963 | |
da8e1782 MO |
964 | r = prioq_ensure_allocated(&m->run_queue, compare_job_priority); |
965 | if (r < 0) | |
966 | return r; | |
967 | ||
e8112e67 | 968 | r = manager_setup_prefix(m); |
718db961 | 969 | if (r < 0) |
c70cac54 | 970 | return r; |
8742514c | 971 | |
2ad591a3 LP |
972 | r = manager_find_credentials_dirs(m); |
973 | if (r < 0) | |
974 | return r; | |
bb0c0d6f | 975 | |
e8112e67 | 976 | r = sd_event_default(&m->event); |
8742514c | 977 | if (r < 0) |
c70cac54 | 978 | return r; |
9152c765 | 979 | |
e8112e67 | 980 | r = manager_setup_run_queue(m); |
a1d32bac | 981 | if (r < 0) |
c70cac54 | 982 | return r; |
8e274523 | 983 | |
5dd2f5ff | 984 | if (FLAGS_SET(test_run_flags, MANAGER_TEST_RUN_MINIMAL)) { |
e8112e67 ZJS |
985 | m->cgroup_root = strdup(""); |
986 | if (!m->cgroup_root) | |
987 | return -ENOMEM; | |
988 | } else { | |
989 | r = manager_setup_signals(m); | |
990 | if (r < 0) | |
991 | return r; | |
8c47c732 | 992 | |
e8112e67 ZJS |
993 | r = manager_setup_cgroup(m); |
994 | if (r < 0) | |
995 | return r; | |
575b300b | 996 | |
e8112e67 ZJS |
997 | r = manager_setup_time_change(m); |
998 | if (r < 0) | |
999 | return r; | |
9670d583 | 1000 | |
bbf5fd8e LP |
1001 | r = manager_read_timezone_stat(m); |
1002 | if (r < 0) | |
1003 | return r; | |
1004 | ||
ea5c5f68 | 1005 | (void) manager_setup_timezone_change(m); |
bbf5fd8e | 1006 | |
e8112e67 ZJS |
1007 | r = manager_setup_sigchld_event_source(m); |
1008 | if (r < 0) | |
1009 | return r; | |
b1994387 | 1010 | |
29e6b0c1 LP |
1011 | r = manager_setup_memory_pressure_event_source(m); |
1012 | if (r < 0) | |
1013 | return r; | |
1014 | ||
b1994387 | 1015 | #if HAVE_LIBBPF |
ba187c9c | 1016 | if (MANAGER_IS_SYSTEM(m) && lsm_bpf_supported(/* initialize = */ true)) { |
b1994387 ILG |
1017 | r = lsm_bpf_setup(m); |
1018 | if (r < 0) | |
299d9417 | 1019 | log_warning_errno(r, "Failed to setup LSM BPF, ignoring: %m"); |
b1994387 ILG |
1020 | } |
1021 | #endif | |
e8112e67 | 1022 | } |
e27fe688 | 1023 | |
2f8c48b6 AZ |
1024 | if (test_run_flags == 0) { |
1025 | if (MANAGER_IS_SYSTEM(m)) | |
1026 | r = mkdir_label("/run/systemd/units", 0755); | |
1027 | else { | |
1028 | _cleanup_free_ char *units_path = NULL; | |
1029 | r = xdg_user_runtime_dir(&units_path, "/systemd/units"); | |
1030 | if (r < 0) | |
1031 | return r; | |
1032 | r = mkdir_p_label(units_path, 0755); | |
1033 | } | |
1034 | ||
d3070fbd | 1035 | if (r < 0 && r != -EEXIST) |
c70cac54 | 1036 | return r; |
d3070fbd LP |
1037 | } |
1038 | ||
e27fe688 LP |
1039 | m->taint_usr = |
1040 | !in_initrd() && | |
db55bbf2 | 1041 | dir_is_empty("/usr", /* ignore_hidden_or_backup= */ false) > 0; |
e27fe688 | 1042 | |
232f6754 ZJS |
1043 | /* Note that we do not set up the notify fd here. We do that after deserialization, |
1044 | * since they might have gotten serialized across the reexec. */ | |
3536f49e | 1045 | |
1cc6c93a YW |
1046 | *_m = TAKE_PTR(m); |
1047 | ||
8e274523 | 1048 | return 0; |
60918275 LP |
1049 | } |
1050 | ||
d86f9d52 | 1051 | static int manager_setup_notify(Manager *m) { |
7181dbdb | 1052 | int r; |
d86f9d52 | 1053 | |
638cece4 | 1054 | if (MANAGER_IS_TEST_RUN(m)) |
0d8c31ff ZJS |
1055 | return 0; |
1056 | ||
d86f9d52 | 1057 | if (m->notify_fd < 0) { |
254d1313 | 1058 | _cleanup_close_ int fd = -EBADF; |
f36a9d59 ZJS |
1059 | union sockaddr_union sa; |
1060 | socklen_t sa_len; | |
d86f9d52 LP |
1061 | |
1062 | /* First free all secondary fields */ | |
a1e58e8e | 1063 | m->notify_socket = mfree(m->notify_socket); |
5dcadb4c | 1064 | m->notify_event_source = sd_event_source_disable_unref(m->notify_event_source); |
d86f9d52 LP |
1065 | |
1066 | fd = socket(AF_UNIX, SOCK_DGRAM|SOCK_CLOEXEC|SOCK_NONBLOCK, 0); | |
4a62c710 MS |
1067 | if (fd < 0) |
1068 | return log_error_errno(errno, "Failed to allocate notification socket: %m"); | |
d86f9d52 | 1069 | |
28e7e934 | 1070 | fd_increase_rxbuf(fd, NOTIFY_RCVBUF_SIZE); |
a47806fa | 1071 | |
b910cc72 | 1072 | m->notify_socket = path_join(m->prefix[EXEC_DIRECTORY_RUNTIME], "systemd/notify"); |
498e87d6 LP |
1073 | if (!m->notify_socket) |
1074 | return log_oom(); | |
1075 | ||
f36a9d59 ZJS |
1076 | r = sockaddr_un_set_path(&sa.un, m->notify_socket); |
1077 | if (r < 0) | |
1078 | return log_error_errno(r, "Notify socket '%s' not valid for AF_UNIX socket address, refusing.", | |
1079 | m->notify_socket); | |
1080 | sa_len = r; | |
15a3e96f | 1081 | |
498e87d6 | 1082 | (void) mkdir_parents_label(m->notify_socket, 0755); |
fbda85b0 | 1083 | (void) sockaddr_un_unlink(&sa.un); |
7181dbdb | 1084 | |
45ae2f72 | 1085 | r = mac_selinux_bind(fd, &sa.sa, sa_len); |
4a62c710 | 1086 | if (r < 0) |
45ae2f72 | 1087 | return log_error_errno(r, "bind(%s) failed: %m", m->notify_socket); |
d86f9d52 | 1088 | |
2ff48e98 | 1089 | r = setsockopt_int(fd, SOL_SOCKET, SO_PASSCRED, true); |
4a62c710 | 1090 | if (r < 0) |
2ff48e98 | 1091 | return log_error_errno(r, "SO_PASSCRED failed: %m"); |
d86f9d52 | 1092 | |
c10d6bdb | 1093 | m->notify_fd = TAKE_FD(fd); |
d86f9d52 LP |
1094 | |
1095 | log_debug("Using notification socket %s", m->notify_socket); | |
1096 | } | |
1097 | ||
1098 | if (!m->notify_event_source) { | |
151b9b96 | 1099 | r = sd_event_add_io(m->event, &m->notify_event_source, m->notify_fd, EPOLLIN, manager_dispatch_notify_fd, m); |
895b3a7b MS |
1100 | if (r < 0) |
1101 | return log_error_errno(r, "Failed to allocate notify event source: %m"); | |
d86f9d52 | 1102 | |
d8fdc620 LP |
1103 | /* Process notification messages a bit earlier than SIGCHLD, so that we can still identify to which |
1104 | * service an exit message belongs. */ | |
575b300b | 1105 | r = sd_event_source_set_priority(m->notify_event_source, SD_EVENT_PRIORITY_NORMAL-8); |
23bbb0de MS |
1106 | if (r < 0) |
1107 | return log_error_errno(r, "Failed to set priority of notify event source: %m"); | |
7dfbe2e3 TG |
1108 | |
1109 | (void) sd_event_source_set_description(m->notify_event_source, "manager-notify"); | |
d86f9d52 LP |
1110 | } |
1111 | ||
1112 | return 0; | |
1113 | } | |
1114 | ||
d8fdc620 LP |
1115 | static int manager_setup_cgroups_agent(Manager *m) { |
1116 | ||
1117 | static const union sockaddr_union sa = { | |
1118 | .un.sun_family = AF_UNIX, | |
1119 | .un.sun_path = "/run/systemd/cgroups-agent", | |
1120 | }; | |
1121 | int r; | |
1122 | ||
1123 | /* This creates a listening socket we receive cgroups agent messages on. We do not use D-Bus for delivering | |
1124 | * these messages from the cgroups agent binary to PID 1, as the cgroups agent binary is very short-living, and | |
1125 | * each instance of it needs a new D-Bus connection. Since D-Bus connections are SOCK_STREAM/AF_UNIX, on | |
1126 | * overloaded systems the backlog of the D-Bus socket becomes relevant, as not more than the configured number | |
1127 | * of D-Bus connections may be queued until the kernel will start dropping further incoming connections, | |
1128 | * possibly resulting in lost cgroups agent messages. To avoid this, we'll use a private SOCK_DGRAM/AF_UNIX | |
1129 | * socket, where no backlog is relevant as communication may take place without an actual connect() cycle, and | |
1130 | * we thus won't lose messages. | |
1131 | * | |
1132 | * Note that PID 1 will forward the agent message to system bus, so that the user systemd instance may listen | |
1133 | * to it. The system instance hence listens on this special socket, but the user instances listen on the system | |
1134 | * bus for these messages. */ | |
1135 | ||
638cece4 | 1136 | if (MANAGER_IS_TEST_RUN(m)) |
d8fdc620 LP |
1137 | return 0; |
1138 | ||
1139 | if (!MANAGER_IS_SYSTEM(m)) | |
1140 | return 0; | |
1141 | ||
c22800e4 | 1142 | r = cg_unified_controller(SYSTEMD_CGROUP_CONTROLLER); |
b4cccbc1 LP |
1143 | if (r < 0) |
1144 | return log_error_errno(r, "Failed to determine whether unified cgroups hierarchy is used: %m"); | |
1145 | if (r > 0) /* We don't need this anymore on the unified hierarchy */ | |
d8fdc620 LP |
1146 | return 0; |
1147 | ||
1148 | if (m->cgroups_agent_fd < 0) { | |
254d1313 | 1149 | _cleanup_close_ int fd = -EBADF; |
d8fdc620 LP |
1150 | |
1151 | /* First free all secondary fields */ | |
5dcadb4c | 1152 | m->cgroups_agent_event_source = sd_event_source_disable_unref(m->cgroups_agent_event_source); |
d8fdc620 LP |
1153 | |
1154 | fd = socket(AF_UNIX, SOCK_DGRAM|SOCK_CLOEXEC|SOCK_NONBLOCK, 0); | |
1155 | if (fd < 0) | |
1156 | return log_error_errno(errno, "Failed to allocate cgroups agent socket: %m"); | |
1157 | ||
28e7e934 | 1158 | fd_increase_rxbuf(fd, CGROUPS_AGENT_RCVBUF_SIZE); |
d8fdc620 | 1159 | |
155b6876 | 1160 | (void) sockaddr_un_unlink(&sa.un); |
d8fdc620 LP |
1161 | |
1162 | /* Only allow root to connect to this socket */ | |
2053593f | 1163 | WITH_UMASK(0077) |
fc2fffe7 | 1164 | r = bind(fd, &sa.sa, SOCKADDR_UN_LEN(sa.un)); |
d8fdc620 LP |
1165 | if (r < 0) |
1166 | return log_error_errno(errno, "bind(%s) failed: %m", sa.un.sun_path); | |
1167 | ||
0b7e8b32 | 1168 | m->cgroups_agent_fd = TAKE_FD(fd); |
d8fdc620 LP |
1169 | } |
1170 | ||
1171 | if (!m->cgroups_agent_event_source) { | |
1172 | r = sd_event_add_io(m->event, &m->cgroups_agent_event_source, m->cgroups_agent_fd, EPOLLIN, manager_dispatch_cgroups_agent_fd, m); | |
1173 | if (r < 0) | |
1174 | return log_error_errno(r, "Failed to allocate cgroups agent event source: %m"); | |
1175 | ||
cbe83389 LP |
1176 | /* Process cgroups notifications early. Note that when the agent notification is received |
1177 | * we'll just enqueue the unit in the cgroup empty queue, hence pick a high priority than | |
1178 | * that. Also see handling of cgroup inotify for the unified cgroup stuff. */ | |
1179 | r = sd_event_source_set_priority(m->cgroups_agent_event_source, SD_EVENT_PRIORITY_NORMAL-9); | |
d8fdc620 LP |
1180 | if (r < 0) |
1181 | return log_error_errno(r, "Failed to set priority of cgroups agent event source: %m"); | |
1182 | ||
1183 | (void) sd_event_source_set_description(m->cgroups_agent_event_source, "manager-cgroups-agent"); | |
1184 | } | |
1185 | ||
1186 | return 0; | |
1187 | } | |
1188 | ||
00d9ef85 LP |
1189 | static int manager_setup_user_lookup_fd(Manager *m) { |
1190 | int r; | |
1191 | ||
1192 | assert(m); | |
1193 | ||
1194 | /* Set up the socket pair used for passing UID/GID resolution results from forked off processes to PID | |
1195 | * 1. Background: we can't do name lookups (NSS) from PID 1, since it might involve IPC and thus activation, | |
1196 | * and we might hence deadlock on ourselves. Hence we do all user/group lookups asynchronously from the forked | |
1197 | * off processes right before executing the binaries to start. In order to be able to clean up any IPC objects | |
1198 | * created by a unit (see RemoveIPC=) we need to know in PID 1 the used UID/GID of the executed processes, | |
1199 | * hence we establish this communication channel so that forked off processes can pass their UID/GID | |
1200 | * information back to PID 1. The forked off processes send their resolved UID/GID to PID 1 in a simple | |
1201 | * datagram, along with their unit name, so that we can share one communication socket pair among all units for | |
1202 | * this purpose. | |
1203 | * | |
1204 | * You might wonder why we need a communication channel for this that is independent of the usual notification | |
1205 | * socket scheme (i.e. $NOTIFY_SOCKET). The primary difference is about trust: data sent via the $NOTIFY_SOCKET | |
1206 | * channel is only accepted if it originates from the right unit and if reception was enabled for it. The user | |
1207 | * lookup socket OTOH is only accessible by PID 1 and its children until they exec(), and always available. | |
1208 | * | |
1209 | * Note that this function is called under two circumstances: when we first initialize (in which case we | |
1210 | * allocate both the socket pair and the event source to listen on it), and when we deserialize after a reload | |
1211 | * (in which case the socket pair already exists but we still need to allocate the event source for it). */ | |
1212 | ||
1213 | if (m->user_lookup_fds[0] < 0) { | |
1214 | ||
1215 | /* Free all secondary fields */ | |
1216 | safe_close_pair(m->user_lookup_fds); | |
5dcadb4c | 1217 | m->user_lookup_event_source = sd_event_source_disable_unref(m->user_lookup_event_source); |
00d9ef85 LP |
1218 | |
1219 | if (socketpair(AF_UNIX, SOCK_DGRAM|SOCK_CLOEXEC, 0, m->user_lookup_fds) < 0) | |
1220 | return log_error_errno(errno, "Failed to allocate user lookup socket: %m"); | |
1221 | ||
28e7e934 | 1222 | (void) fd_increase_rxbuf(m->user_lookup_fds[0], NOTIFY_RCVBUF_SIZE); |
00d9ef85 LP |
1223 | } |
1224 | ||
1225 | if (!m->user_lookup_event_source) { | |
1226 | r = sd_event_add_io(m->event, &m->user_lookup_event_source, m->user_lookup_fds[0], EPOLLIN, manager_dispatch_user_lookup_fd, m); | |
1227 | if (r < 0) | |
1228 | return log_error_errno(errno, "Failed to allocate user lookup event source: %m"); | |
1229 | ||
1230 | /* Process even earlier than the notify event source, so that we always know first about valid UID/GID | |
1231 | * resolutions */ | |
67ae4e8d | 1232 | r = sd_event_source_set_priority(m->user_lookup_event_source, SD_EVENT_PRIORITY_NORMAL-11); |
00d9ef85 | 1233 | if (r < 0) |
5238e957 | 1234 | return log_error_errno(errno, "Failed to set priority of user lookup event source: %m"); |
00d9ef85 LP |
1235 | |
1236 | (void) sd_event_source_set_description(m->user_lookup_event_source, "user-lookup"); | |
1237 | } | |
1238 | ||
1239 | return 0; | |
1240 | } | |
1241 | ||
23a177ef | 1242 | static unsigned manager_dispatch_cleanup_queue(Manager *m) { |
595ed347 | 1243 | Unit *u; |
23a177ef LP |
1244 | unsigned n = 0; |
1245 | ||
1246 | assert(m); | |
1247 | ||
595ed347 MS |
1248 | while ((u = m->cleanup_queue)) { |
1249 | assert(u->in_cleanup_queue); | |
23a177ef | 1250 | |
595ed347 | 1251 | unit_free(u); |
23a177ef LP |
1252 | n++; |
1253 | } | |
1254 | ||
1255 | return n; | |
1256 | } | |
1257 | ||
6ac62d61 LP |
1258 | static unsigned manager_dispatch_release_resources_queue(Manager *m) { |
1259 | unsigned n = 0; | |
1260 | Unit *u; | |
1261 | ||
1262 | assert(m); | |
1263 | ||
1264 | while ((u = m->release_resources_queue)) { | |
1265 | assert(u->in_release_resources_queue); | |
1266 | ||
1267 | LIST_REMOVE(release_resources_queue, m->release_resources_queue, u); | |
1268 | u->in_release_resources_queue = false; | |
1269 | ||
1270 | n++; | |
1271 | ||
1272 | unit_release_resources(u); | |
1273 | } | |
1274 | ||
1275 | return n; | |
1276 | } | |
1277 | ||
eced69b3 | 1278 | enum { |
35b8ca3a | 1279 | GC_OFFSET_IN_PATH, /* This one is on the path we were traveling */ |
eced69b3 LP |
1280 | GC_OFFSET_UNSURE, /* No clue */ |
1281 | GC_OFFSET_GOOD, /* We still need this unit */ | |
1282 | GC_OFFSET_BAD, /* We don't need this unit anymore */ | |
1283 | _GC_OFFSET_MAX | |
1284 | }; | |
1285 | ||
00d9ef85 | 1286 | static void unit_gc_mark_good(Unit *u, unsigned gc_marker) { |
4892084f LN |
1287 | Unit *other; |
1288 | ||
1289 | u->gc_marker = gc_marker + GC_OFFSET_GOOD; | |
1290 | ||
1291 | /* Recursively mark referenced units as GOOD as well */ | |
15ed3c3a | 1292 | UNIT_FOREACH_DEPENDENCY(other, u, UNIT_ATOM_REFERENCES) |
4892084f LN |
1293 | if (other->gc_marker == gc_marker + GC_OFFSET_UNSURE) |
1294 | unit_gc_mark_good(other, gc_marker); | |
1295 | } | |
1296 | ||
eced69b3 | 1297 | static void unit_gc_sweep(Unit *u, unsigned gc_marker) { |
701cc384 | 1298 | Unit *other; |
eced69b3 | 1299 | bool is_bad; |
701cc384 LP |
1300 | |
1301 | assert(u); | |
1302 | ||
4c701096 YW |
1303 | if (IN_SET(u->gc_marker - gc_marker, |
1304 | GC_OFFSET_GOOD, GC_OFFSET_BAD, GC_OFFSET_UNSURE, GC_OFFSET_IN_PATH)) | |
701cc384 LP |
1305 | return; |
1306 | ||
ac155bb8 | 1307 | if (u->in_cleanup_queue) |
701cc384 LP |
1308 | goto bad; |
1309 | ||
f2f725e5 | 1310 | if (!unit_may_gc(u)) |
701cc384 LP |
1311 | goto good; |
1312 | ||
ac155bb8 | 1313 | u->gc_marker = gc_marker + GC_OFFSET_IN_PATH; |
eced69b3 LP |
1314 | |
1315 | is_bad = true; | |
1316 | ||
15ed3c3a | 1317 | UNIT_FOREACH_DEPENDENCY(other, u, UNIT_ATOM_REFERENCED_BY) { |
701cc384 LP |
1318 | unit_gc_sweep(other, gc_marker); |
1319 | ||
ac155bb8 | 1320 | if (other->gc_marker == gc_marker + GC_OFFSET_GOOD) |
701cc384 | 1321 | goto good; |
eced69b3 | 1322 | |
ac155bb8 | 1323 | if (other->gc_marker != gc_marker + GC_OFFSET_BAD) |
eced69b3 | 1324 | is_bad = false; |
701cc384 LP |
1325 | } |
1326 | ||
ef3fc326 ZJS |
1327 | LIST_FOREACH(refs_by_target, ref, u->refs_by_target) { |
1328 | unit_gc_sweep(ref->source, gc_marker); | |
2641f02e | 1329 | |
ef3fc326 ZJS |
1330 | if (ref->source->gc_marker == gc_marker + GC_OFFSET_GOOD) |
1331 | goto good; | |
2641f02e | 1332 | |
ef3fc326 ZJS |
1333 | if (ref->source->gc_marker != gc_marker + GC_OFFSET_BAD) |
1334 | is_bad = false; | |
2641f02e | 1335 | } |
701cc384 | 1336 | |
eced69b3 LP |
1337 | if (is_bad) |
1338 | goto bad; | |
1339 | ||
1340 | /* We were unable to find anything out about this entry, so | |
1341 | * let's investigate it later */ | |
ac155bb8 | 1342 | u->gc_marker = gc_marker + GC_OFFSET_UNSURE; |
eced69b3 LP |
1343 | unit_add_to_gc_queue(u); |
1344 | return; | |
1345 | ||
701cc384 | 1346 | bad: |
eced69b3 LP |
1347 | /* We definitely know that this one is not useful anymore, so |
1348 | * let's mark it for deletion */ | |
ac155bb8 | 1349 | u->gc_marker = gc_marker + GC_OFFSET_BAD; |
eced69b3 | 1350 | unit_add_to_cleanup_queue(u); |
701cc384 LP |
1351 | return; |
1352 | ||
1353 | good: | |
4892084f | 1354 | unit_gc_mark_good(u, gc_marker); |
701cc384 LP |
1355 | } |
1356 | ||
c5a97ed1 LP |
1357 | static unsigned manager_dispatch_gc_unit_queue(Manager *m) { |
1358 | unsigned n = 0, gc_marker; | |
595ed347 | 1359 | Unit *u; |
701cc384 LP |
1360 | |
1361 | assert(m); | |
1362 | ||
cf1265e1 | 1363 | /* log_debug("Running GC..."); */ |
701cc384 | 1364 | |
eced69b3 LP |
1365 | m->gc_marker += _GC_OFFSET_MAX; |
1366 | if (m->gc_marker + _GC_OFFSET_MAX <= _GC_OFFSET_MAX) | |
c9c0cadb | 1367 | m->gc_marker = 1; |
701cc384 | 1368 | |
eced69b3 LP |
1369 | gc_marker = m->gc_marker; |
1370 | ||
c5a97ed1 | 1371 | while ((u = m->gc_unit_queue)) { |
595ed347 | 1372 | assert(u->in_gc_queue); |
701cc384 | 1373 | |
595ed347 | 1374 | unit_gc_sweep(u, gc_marker); |
eced69b3 | 1375 | |
c5a97ed1 | 1376 | LIST_REMOVE(gc_queue, m->gc_unit_queue, u); |
595ed347 | 1377 | u->in_gc_queue = false; |
701cc384 LP |
1378 | |
1379 | n++; | |
1380 | ||
4c701096 YW |
1381 | if (IN_SET(u->gc_marker - gc_marker, |
1382 | GC_OFFSET_BAD, GC_OFFSET_UNSURE)) { | |
cc3bc3e6 | 1383 | if (u->id) |
f2341e0a | 1384 | log_unit_debug(u, "Collecting."); |
595ed347 MS |
1385 | u->gc_marker = gc_marker + GC_OFFSET_BAD; |
1386 | unit_add_to_cleanup_queue(u); | |
701cc384 LP |
1387 | } |
1388 | } | |
1389 | ||
701cc384 LP |
1390 | return n; |
1391 | } | |
1392 | ||
c5a97ed1 LP |
1393 | static unsigned manager_dispatch_gc_job_queue(Manager *m) { |
1394 | unsigned n = 0; | |
1395 | Job *j; | |
1396 | ||
1397 | assert(m); | |
1398 | ||
1399 | while ((j = m->gc_job_queue)) { | |
1400 | assert(j->in_gc_queue); | |
1401 | ||
1402 | LIST_REMOVE(gc_queue, m->gc_job_queue, j); | |
1403 | j->in_gc_queue = false; | |
1404 | ||
1405 | n++; | |
1406 | ||
2ab3050f | 1407 | if (!job_may_gc(j)) |
c5a97ed1 LP |
1408 | continue; |
1409 | ||
1410 | log_unit_debug(j->unit, "Collecting job."); | |
1411 | (void) job_finish_and_invalidate(j, JOB_COLLECTED, false, false); | |
1412 | } | |
1413 | ||
1414 | return n; | |
1415 | } | |
1416 | ||
7223d500 LB |
1417 | static int manager_ratelimit_requeue(sd_event_source *s, uint64_t usec, void *userdata) { |
1418 | Unit *u = userdata; | |
1419 | ||
1420 | assert(u); | |
1421 | assert(s == u->auto_start_stop_event_source); | |
1422 | ||
1423 | u->auto_start_stop_event_source = sd_event_source_unref(u->auto_start_stop_event_source); | |
1424 | ||
1425 | /* Re-queue to all queues, if the rate limit hit we might have been throttled on any of them. */ | |
1426 | unit_submit_to_stop_when_unneeded_queue(u); | |
1427 | unit_submit_to_start_when_upheld_queue(u); | |
1428 | unit_submit_to_stop_when_bound_queue(u); | |
1429 | ||
1430 | return 0; | |
1431 | } | |
1432 | ||
1433 | static int manager_ratelimit_check_and_queue(Unit *u) { | |
1434 | int r; | |
1435 | ||
1436 | assert(u); | |
1437 | ||
1438 | if (ratelimit_below(&u->auto_start_stop_ratelimit)) | |
1439 | return 1; | |
1440 | ||
1441 | /* Already queued, no need to requeue */ | |
1442 | if (u->auto_start_stop_event_source) | |
1443 | return 0; | |
1444 | ||
1445 | r = sd_event_add_time( | |
1446 | u->manager->event, | |
1447 | &u->auto_start_stop_event_source, | |
1448 | CLOCK_MONOTONIC, | |
1449 | ratelimit_end(&u->auto_start_stop_ratelimit), | |
1450 | 0, | |
1451 | manager_ratelimit_requeue, | |
1452 | u); | |
1453 | if (r < 0) | |
1454 | return log_unit_error_errno(u, r, "Failed to queue timer on event loop: %m"); | |
1455 | ||
1456 | return 0; | |
1457 | } | |
1458 | ||
a3c1168a LP |
1459 | static unsigned manager_dispatch_stop_when_unneeded_queue(Manager *m) { |
1460 | unsigned n = 0; | |
1461 | Unit *u; | |
1462 | int r; | |
1463 | ||
1464 | assert(m); | |
1465 | ||
1466 | while ((u = m->stop_when_unneeded_queue)) { | |
1467 | _cleanup_(sd_bus_error_free) sd_bus_error error = SD_BUS_ERROR_NULL; | |
a3c1168a LP |
1468 | |
1469 | assert(u->in_stop_when_unneeded_queue); | |
1470 | LIST_REMOVE(stop_when_unneeded_queue, m->stop_when_unneeded_queue, u); | |
1471 | u->in_stop_when_unneeded_queue = false; | |
1472 | ||
1473 | n++; | |
1474 | ||
1475 | if (!unit_is_unneeded(u)) | |
1476 | continue; | |
1477 | ||
1478 | log_unit_debug(u, "Unit is not needed anymore."); | |
1479 | ||
1480 | /* If stopping a unit fails continuously we might enter a stop loop here, hence stop acting on the | |
1481 | * service being unnecessary after a while. */ | |
1482 | ||
7223d500 LB |
1483 | r = manager_ratelimit_check_and_queue(u); |
1484 | if (r <= 0) { | |
1485 | log_unit_warning(u, | |
1486 | "Unit not needed anymore, but not stopping since we tried this too often recently.%s", | |
1487 | r == 0 ? " Will retry later." : ""); | |
a3c1168a LP |
1488 | continue; |
1489 | } | |
1490 | ||
1491 | /* Ok, nobody needs us anymore. Sniff. Then let's commit suicide */ | |
50cbaba4 | 1492 | r = manager_add_job(u->manager, JOB_STOP, u, JOB_FAIL, NULL, &error, NULL); |
a3c1168a LP |
1493 | if (r < 0) |
1494 | log_unit_warning_errno(u, r, "Failed to enqueue stop job, ignoring: %s", bus_error_message(&error, r)); | |
1495 | } | |
1496 | ||
1497 | return n; | |
1498 | } | |
1499 | ||
0bc488c9 LP |
1500 | static unsigned manager_dispatch_start_when_upheld_queue(Manager *m) { |
1501 | unsigned n = 0; | |
1502 | Unit *u; | |
1503 | int r; | |
1504 | ||
1505 | assert(m); | |
1506 | ||
1507 | while ((u = m->start_when_upheld_queue)) { | |
1508 | _cleanup_(sd_bus_error_free) sd_bus_error error = SD_BUS_ERROR_NULL; | |
1509 | Unit *culprit = NULL; | |
1510 | ||
1511 | assert(u->in_start_when_upheld_queue); | |
1512 | LIST_REMOVE(start_when_upheld_queue, m->start_when_upheld_queue, u); | |
1513 | u->in_start_when_upheld_queue = false; | |
1514 | ||
1515 | n++; | |
1516 | ||
1517 | if (!unit_is_upheld_by_active(u, &culprit)) | |
1518 | continue; | |
1519 | ||
1520 | log_unit_debug(u, "Unit is started because upheld by active unit %s.", culprit->id); | |
1521 | ||
1522 | /* If stopping a unit fails continuously we might enter a stop loop here, hence stop acting on the | |
1523 | * service being unnecessary after a while. */ | |
1524 | ||
7223d500 LB |
1525 | r = manager_ratelimit_check_and_queue(u); |
1526 | if (r <= 0) { | |
1527 | log_unit_warning(u, | |
1528 | "Unit needs to be started because active unit %s upholds it, but not starting since we tried this too often recently.%s", | |
1529 | culprit->id, | |
1530 | r == 0 ? " Will retry later." : ""); | |
0bc488c9 LP |
1531 | continue; |
1532 | } | |
1533 | ||
1534 | r = manager_add_job(u->manager, JOB_START, u, JOB_FAIL, NULL, &error, NULL); | |
1535 | if (r < 0) | |
1536 | log_unit_warning_errno(u, r, "Failed to enqueue start job, ignoring: %s", bus_error_message(&error, r)); | |
1537 | } | |
1538 | ||
1539 | return n; | |
1540 | } | |
1541 | ||
56c59592 LP |
1542 | static unsigned manager_dispatch_stop_when_bound_queue(Manager *m) { |
1543 | unsigned n = 0; | |
1544 | Unit *u; | |
1545 | int r; | |
1546 | ||
1547 | assert(m); | |
1548 | ||
1549 | while ((u = m->stop_when_bound_queue)) { | |
1550 | _cleanup_(sd_bus_error_free) sd_bus_error error = SD_BUS_ERROR_NULL; | |
1551 | Unit *culprit = NULL; | |
1552 | ||
1553 | assert(u->in_stop_when_bound_queue); | |
1554 | LIST_REMOVE(stop_when_bound_queue, m->stop_when_bound_queue, u); | |
1555 | u->in_stop_when_bound_queue = false; | |
1556 | ||
1557 | n++; | |
1558 | ||
1559 | if (!unit_is_bound_by_inactive(u, &culprit)) | |
1560 | continue; | |
1561 | ||
1562 | log_unit_debug(u, "Unit is stopped because bound to inactive unit %s.", culprit->id); | |
1563 | ||
1564 | /* If stopping a unit fails continuously we might enter a stop loop here, hence stop acting on the | |
1565 | * service being unnecessary after a while. */ | |
1566 | ||
7223d500 LB |
1567 | r = manager_ratelimit_check_and_queue(u); |
1568 | if (r <= 0) { | |
1569 | log_unit_warning(u, | |
1570 | "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", | |
1571 | culprit->id, | |
1572 | r == 0 ? " Will retry later." : ""); | |
56c59592 LP |
1573 | continue; |
1574 | } | |
1575 | ||
1576 | r = manager_add_job(u->manager, JOB_STOP, u, JOB_REPLACE, NULL, &error, NULL); | |
1577 | if (r < 0) | |
1578 | log_unit_warning_errno(u, r, "Failed to enqueue stop job, ignoring: %s", bus_error_message(&error, r)); | |
1579 | } | |
1580 | ||
1581 | return n; | |
1582 | } | |
1583 | ||
a16e1123 | 1584 | static void manager_clear_jobs_and_units(Manager *m) { |
a16e1123 | 1585 | Unit *u; |
60918275 LP |
1586 | |
1587 | assert(m); | |
1588 | ||
87f0e418 LP |
1589 | while ((u = hashmap_first(m->units))) |
1590 | unit_free(u); | |
964e0949 LP |
1591 | |
1592 | manager_dispatch_cleanup_queue(m); | |
1593 | ||
1594 | assert(!m->load_queue); | |
da8e1782 | 1595 | assert(prioq_isempty(m->run_queue)); |
964e0949 LP |
1596 | assert(!m->dbus_unit_queue); |
1597 | assert(!m->dbus_job_queue); | |
1598 | assert(!m->cleanup_queue); | |
c5a97ed1 LP |
1599 | assert(!m->gc_unit_queue); |
1600 | assert(!m->gc_job_queue); | |
13e72103 MK |
1601 | assert(!m->cgroup_realize_queue); |
1602 | assert(!m->cgroup_empty_queue); | |
1603 | assert(!m->cgroup_oom_queue); | |
1604 | assert(!m->target_deps_queue); | |
a3c1168a | 1605 | assert(!m->stop_when_unneeded_queue); |
0bc488c9 | 1606 | assert(!m->start_when_upheld_queue); |
56c59592 | 1607 | assert(!m->stop_when_bound_queue); |
cb3c6aec | 1608 | assert(!m->release_resources_queue); |
964e0949 | 1609 | |
964e0949 LP |
1610 | assert(hashmap_isempty(m->jobs)); |
1611 | assert(hashmap_isempty(m->units)); | |
9e9e2b72 MS |
1612 | |
1613 | m->n_on_console = 0; | |
1614 | m->n_running_jobs = 0; | |
1e75824c MK |
1615 | m->n_installed_jobs = 0; |
1616 | m->n_failed_jobs = 0; | |
a16e1123 LP |
1617 | } |
1618 | ||
06d8d842 | 1619 | Manager* manager_free(Manager *m) { |
06d8d842 ZJS |
1620 | if (!m) |
1621 | return NULL; | |
a16e1123 LP |
1622 | |
1623 | manager_clear_jobs_and_units(m); | |
23a177ef | 1624 | |
24b45970 | 1625 | for (UnitType c = 0; c < _UNIT_TYPE_MAX; c++) |
7824bbeb LP |
1626 | if (unit_vtable[c]->shutdown) |
1627 | unit_vtable[c]->shutdown(m); | |
1628 | ||
86036b26 LP |
1629 | /* Keep the cgroup hierarchy in place except when we know we are going down for good */ |
1630 | manager_shutdown_cgroup(m, IN_SET(m->objective, MANAGER_EXIT, MANAGER_REBOOT, MANAGER_POWEROFF, MANAGER_HALT, MANAGER_KEXEC)); | |
8e274523 | 1631 | |
07a78643 | 1632 | lookup_paths_flush_generator(&m->lookup_paths); |
5a1e9937 | 1633 | |
5e8d1c9a | 1634 | bus_done(m); |
19d22d43 | 1635 | manager_varlink_done(m); |
ea430986 | 1636 | |
e76506b7 DDM |
1637 | exec_shared_runtime_vacuum(m); |
1638 | hashmap_free(m->exec_shared_runtime_by_id); | |
e8a565cb | 1639 | |
29206d46 LP |
1640 | dynamic_user_vacuum(m, false); |
1641 | hashmap_free(m->dynamic_users); | |
1642 | ||
87f0e418 | 1643 | hashmap_free(m->units); |
4b58153d | 1644 | hashmap_free(m->units_by_invocation_id); |
60918275 | 1645 | hashmap_free(m->jobs); |
62a76913 | 1646 | hashmap_free(m->watch_pids); |
05e343b7 | 1647 | hashmap_free(m->watch_bus); |
9152c765 | 1648 | |
da8e1782 MO |
1649 | prioq_free(m->run_queue); |
1650 | ||
95ae05c0 | 1651 | set_free(m->startup_units); |
f755e3b7 LP |
1652 | set_free(m->failed_units); |
1653 | ||
718db961 | 1654 | sd_event_source_unref(m->signal_event_source); |
575b300b | 1655 | sd_event_source_unref(m->sigchld_event_source); |
718db961 | 1656 | sd_event_source_unref(m->notify_event_source); |
d8fdc620 | 1657 | sd_event_source_unref(m->cgroups_agent_event_source); |
718db961 | 1658 | sd_event_source_unref(m->time_change_event_source); |
bbf5fd8e | 1659 | sd_event_source_unref(m->timezone_change_event_source); |
718db961 | 1660 | sd_event_source_unref(m->jobs_in_progress_event_source); |
752b5905 | 1661 | sd_event_source_unref(m->run_queue_event_source); |
00d9ef85 | 1662 | sd_event_source_unref(m->user_lookup_event_source); |
29e6b0c1 | 1663 | sd_event_source_unref(m->memory_pressure_event_source); |
718db961 | 1664 | |
03e334a1 LP |
1665 | safe_close(m->signal_fd); |
1666 | safe_close(m->notify_fd); | |
d8fdc620 | 1667 | safe_close(m->cgroups_agent_fd); |
00d9ef85 | 1668 | safe_close_pair(m->user_lookup_fds); |
718db961 | 1669 | |
e46b13c8 ZJS |
1670 | manager_close_ask_password(m); |
1671 | ||
718db961 LP |
1672 | manager_close_idle_pipe(m); |
1673 | ||
1674 | sd_event_unref(m->event); | |
60918275 | 1675 | |
c952c6ec LP |
1676 | free(m->notify_socket); |
1677 | ||
84e3543e | 1678 | lookup_paths_free(&m->lookup_paths); |
1ad6e8b3 LP |
1679 | strv_free(m->transient_environment); |
1680 | strv_free(m->client_environment); | |
036643a2 | 1681 | |
4ad49000 | 1682 | hashmap_free(m->cgroup_unit); |
e8630e69 | 1683 | manager_free_unit_name_maps(m); |
33be102a | 1684 | |
664f88a7 LP |
1685 | free(m->switch_root); |
1686 | free(m->switch_root_init); | |
1687 | ||
aa5ae971 ŁS |
1688 | free(m->default_smack_process_label); |
1689 | ||
31ce987c | 1690 | rlimit_free_all(m->rlimit); |
c93ff2e9 | 1691 | |
a57f7e2c LP |
1692 | assert(hashmap_isempty(m->units_requiring_mounts_for)); |
1693 | hashmap_free(m->units_requiring_mounts_for); | |
1694 | ||
00d9ef85 LP |
1695 | hashmap_free(m->uid_refs); |
1696 | hashmap_free(m->gid_refs); | |
1697 | ||
24b45970 | 1698 | for (ExecDirectoryType dt = 0; dt < _EXEC_DIRECTORY_TYPE_MAX; dt++) |
35aba85a | 1699 | m->prefix[dt] = mfree(m->prefix[dt]); |
2ad591a3 LP |
1700 | free(m->received_credentials_directory); |
1701 | free(m->received_encrypted_credentials_directory); | |
35aba85a | 1702 | |
aff3a9e1 LB |
1703 | free(m->watchdog_pretimeout_governor); |
1704 | free(m->watchdog_pretimeout_governor_overridden); | |
1705 | ||
b1994387 ILG |
1706 | #if BPF_FRAMEWORK |
1707 | lsm_bpf_destroy(m->restrict_fs); | |
1708 | #endif | |
1709 | ||
6b430fdb | 1710 | return mfree(m); |
60918275 LP |
1711 | } |
1712 | ||
04eb582a | 1713 | static void manager_enumerate_perpetual(Manager *m) { |
04eb582a LP |
1714 | assert(m); |
1715 | ||
1545051c | 1716 | if (FLAGS_SET(m->test_run_flags, MANAGER_TEST_RUN_MINIMAL)) |
b1d5246d ZJS |
1717 | return; |
1718 | ||
04eb582a | 1719 | /* Let's ask every type to load all units from disk/kernel that it might know */ |
24b45970 | 1720 | for (UnitType c = 0; c < _UNIT_TYPE_MAX; c++) { |
04eb582a LP |
1721 | if (!unit_type_supported(c)) { |
1722 | log_debug("Unit type .%s is not supported on this system.", unit_type_to_string(c)); | |
1723 | continue; | |
1724 | } | |
1725 | ||
1726 | if (unit_vtable[c]->enumerate_perpetual) | |
1727 | unit_vtable[c]->enumerate_perpetual(m); | |
1728 | } | |
1729 | } | |
1730 | ||
a1113e08 | 1731 | static void manager_enumerate(Manager *m) { |
f50e0a01 LP |
1732 | assert(m); |
1733 | ||
1545051c | 1734 | if (FLAGS_SET(m->test_run_flags, MANAGER_TEST_RUN_MINIMAL)) |
b1d5246d ZJS |
1735 | return; |
1736 | ||
04eb582a | 1737 | /* Let's ask every type to load all units from disk/kernel that it might know */ |
24b45970 | 1738 | for (UnitType c = 0; c < _UNIT_TYPE_MAX; c++) { |
1c2e9646 | 1739 | if (!unit_type_supported(c)) { |
03afec3c | 1740 | log_debug("Unit type .%s is not supported on this system.", unit_type_to_string(c)); |
0faacd47 | 1741 | continue; |
a57f7e2c | 1742 | } |
f50e0a01 | 1743 | |
94b01dae ZJS |
1744 | if (unit_vtable[c]->enumerate) |
1745 | unit_vtable[c]->enumerate(m); | |
0faacd47 LP |
1746 | } |
1747 | ||
f50e0a01 | 1748 | manager_dispatch_load_queue(m); |
a16e1123 LP |
1749 | } |
1750 | ||
007c6337 | 1751 | static void manager_coldplug(Manager *m) { |
a16e1123 LP |
1752 | Unit *u; |
1753 | char *k; | |
007c6337 | 1754 | int r; |
a16e1123 LP |
1755 | |
1756 | assert(m); | |
f50e0a01 | 1757 | |
28e5e1e9 | 1758 | log_debug("Invoking unit coldplug() handlers%s", special_glyph(SPECIAL_GLYPH_ELLIPSIS)); |
f0831ed2 LP |
1759 | |
1760 | /* Let's place the units back into their deserialized state */ | |
90e74a66 | 1761 | HASHMAP_FOREACH_KEY(u, k, m->units) { |
f50e0a01 LP |
1762 | |
1763 | /* ignore aliases */ | |
ac155bb8 | 1764 | if (u->id != k) |
f50e0a01 LP |
1765 | continue; |
1766 | ||
007c6337 LP |
1767 | r = unit_coldplug(u); |
1768 | if (r < 0) | |
1769 | log_warning_errno(r, "We couldn't coldplug %s, proceeding anyway: %m", u->id); | |
f50e0a01 | 1770 | } |
a16e1123 LP |
1771 | } |
1772 | ||
f0831ed2 | 1773 | static void manager_catchup(Manager *m) { |
f0831ed2 LP |
1774 | Unit *u; |
1775 | char *k; | |
1776 | ||
1777 | assert(m); | |
1778 | ||
28e5e1e9 | 1779 | log_debug("Invoking unit catchup() handlers%s", special_glyph(SPECIAL_GLYPH_ELLIPSIS)); |
f0831ed2 LP |
1780 | |
1781 | /* Let's catch up on any state changes that happened while we were reloading/reexecing */ | |
90e74a66 | 1782 | HASHMAP_FOREACH_KEY(u, k, m->units) { |
f0831ed2 LP |
1783 | |
1784 | /* ignore aliases */ | |
1785 | if (u->id != k) | |
1786 | continue; | |
1787 | ||
1788 | unit_catchup(u); | |
1789 | } | |
1790 | } | |
1791 | ||
9ff1a6f1 | 1792 | static void manager_distribute_fds(Manager *m, FDSet *fds) { |
9ff1a6f1 | 1793 | Unit *u; |
9588bc32 LP |
1794 | |
1795 | assert(m); | |
1796 | ||
90e74a66 | 1797 | HASHMAP_FOREACH(u, m->units) { |
9588bc32 LP |
1798 | |
1799 | if (fdset_size(fds) <= 0) | |
1800 | break; | |
1801 | ||
9ff1a6f1 LP |
1802 | if (!UNIT_VTABLE(u)->distribute_fds) |
1803 | continue; | |
9588bc32 | 1804 | |
9ff1a6f1 LP |
1805 | UNIT_VTABLE(u)->distribute_fds(u, fds); |
1806 | } | |
9588bc32 LP |
1807 | } |
1808 | ||
2b680534 | 1809 | static bool manager_dbus_is_running(Manager *m, bool deserialized) { |
8559b3b7 LP |
1810 | Unit *u; |
1811 | ||
1812 | assert(m); | |
1813 | ||
1814 | /* This checks whether the dbus instance we are supposed to expose our APIs on is up. We check both the socket | |
1815 | * and the service unit. If the 'deserialized' parameter is true we'll check the deserialized state of the unit | |
1816 | * rather than the current one. */ | |
1817 | ||
638cece4 | 1818 | if (MANAGER_IS_TEST_RUN(m)) |
8559b3b7 LP |
1819 | return false; |
1820 | ||
8559b3b7 LP |
1821 | u = manager_get_unit(m, SPECIAL_DBUS_SOCKET); |
1822 | if (!u) | |
1823 | return false; | |
1824 | if ((deserialized ? SOCKET(u)->deserialized_state : SOCKET(u)->state) != SOCKET_RUNNING) | |
1825 | return false; | |
1826 | ||
1827 | u = manager_get_unit(m, SPECIAL_DBUS_SERVICE); | |
1828 | if (!u) | |
1829 | return false; | |
845824ac | 1830 | if (!IN_SET((deserialized ? SERVICE(u)->deserialized_state : SERVICE(u)->state), |
1831 | SERVICE_RUNNING, | |
1832 | SERVICE_RELOAD, | |
1833 | SERVICE_RELOAD_NOTIFY, | |
1834 | SERVICE_RELOAD_SIGNAL)) | |
8559b3b7 LP |
1835 | return false; |
1836 | ||
1837 | return true; | |
1838 | } | |
1839 | ||
9d4c195c LP |
1840 | static void manager_setup_bus(Manager *m) { |
1841 | assert(m); | |
1842 | ||
1843 | /* Let's set up our private bus connection now, unconditionally */ | |
1844 | (void) bus_init_private(m); | |
1845 | ||
1846 | /* If we are in --user mode also connect to the system bus now */ | |
1847 | if (MANAGER_IS_USER(m)) | |
1848 | (void) bus_init_system(m); | |
1849 | ||
1850 | /* Let's connect to the bus now, but only if the unit is supposed to be up */ | |
2b680534 | 1851 | if (manager_dbus_is_running(m, MANAGER_IS_RELOADING(m))) { |
9d4c195c LP |
1852 | (void) bus_init_api(m); |
1853 | ||
1854 | if (MANAGER_IS_SYSTEM(m)) | |
1855 | (void) bus_init_system(m); | |
1856 | } | |
1857 | } | |
1858 | ||
159f1e76 LP |
1859 | static void manager_preset_all(Manager *m) { |
1860 | int r; | |
1861 | ||
1862 | assert(m); | |
1863 | ||
1864 | if (m->first_boot <= 0) | |
1865 | return; | |
1866 | ||
1867 | if (!MANAGER_IS_SYSTEM(m)) | |
1868 | return; | |
1869 | ||
638cece4 | 1870 | if (MANAGER_IS_TEST_RUN(m)) |
159f1e76 LP |
1871 | return; |
1872 | ||
1873 | /* If this is the first boot, and we are in the host system, then preset everything */ | |
93651582 JL |
1874 | UnitFilePresetMode mode = FIRST_BOOT_FULL_PRESET ? UNIT_FILE_PRESET_FULL : UNIT_FILE_PRESET_ENABLE_ONLY; |
1875 | ||
4870133b | 1876 | r = unit_file_preset_all(RUNTIME_SCOPE_SYSTEM, 0, NULL, mode, NULL, 0); |
159f1e76 LP |
1877 | if (r < 0) |
1878 | log_full_errno(r == -EEXIST ? LOG_NOTICE : LOG_WARNING, r, | |
1879 | "Failed to populate /etc with preset unit settings, ignoring: %m"); | |
1880 | else | |
1881 | log_info("Populated /etc with preset unit settings."); | |
1882 | } | |
1883 | ||
5ce5e1ad LP |
1884 | static void manager_ready(Manager *m) { |
1885 | assert(m); | |
1886 | ||
1887 | /* After having loaded everything, do the final round of catching up with what might have changed */ | |
1888 | ||
1889 | m->objective = MANAGER_OK; /* Tell everyone we are up now */ | |
1890 | ||
1891 | /* It might be safe to log to the journal now and connect to dbus */ | |
1892 | manager_recheck_journal(m); | |
1893 | manager_recheck_dbus(m); | |
1894 | ||
5ce5e1ad LP |
1895 | /* Let's finally catch up with any changes that took place while we were reloading/reexecing */ |
1896 | manager_catchup(m); | |
c6e892bc | 1897 | |
15b9243c | 1898 | /* Create a file which will indicate when the manager started loading units the last time. */ |
4b3ad81b LB |
1899 | if (MANAGER_IS_SYSTEM(m)) |
1900 | (void) touch_file("/run/systemd/systemd-units-load", false, | |
1901 | m->timestamps[MANAGER_TIMESTAMP_UNITS_LOAD].realtime ?: now(CLOCK_REALTIME), | |
1902 | UID_INVALID, GID_INVALID, 0444); | |
5ce5e1ad LP |
1903 | } |
1904 | ||
a01ba4b2 | 1905 | Manager* manager_reloading_start(Manager *m) { |
d147e2b6 | 1906 | m->n_reloading++; |
49fbe940 | 1907 | dual_timestamp_get(m->timestamps + MANAGER_TIMESTAMP_UNITS_LOAD); |
d147e2b6 ZJS |
1908 | return m; |
1909 | } | |
49fbe940 | 1910 | |
a01ba4b2 | 1911 | void manager_reloading_stopp(Manager **m) { |
d147e2b6 ZJS |
1912 | if (*m) { |
1913 | assert((*m)->n_reloading > 0); | |
1914 | (*m)->n_reloading--; | |
1915 | } | |
1916 | } | |
1917 | ||
2a7cf953 | 1918 | int manager_startup(Manager *m, FILE *serialization, FDSet *fds, const char *root) { |
17f01ace | 1919 | int r; |
a16e1123 LP |
1920 | |
1921 | assert(m); | |
1922 | ||
a1f31f47 ZJS |
1923 | /* If we are running in test mode, we still want to run the generators, |
1924 | * but we should not touch the real generator directories. */ | |
4870133b | 1925 | r = lookup_paths_init_or_warn(&m->lookup_paths, m->runtime_scope, |
99aad9a2 ZJS |
1926 | MANAGER_IS_TEST_RUN(m) ? LOOKUP_PATHS_TEMPORARY_GENERATED : 0, |
1927 | root); | |
e801700e | 1928 | if (r < 0) |
99aad9a2 | 1929 | return r; |
5a1e9937 | 1930 | |
d4ee7bd8 | 1931 | dual_timestamp_get(m->timestamps + manager_timestamp_initrd_mangle(MANAGER_TIMESTAMP_GENERATORS_START)); |
eb523bfb LP |
1932 | r = manager_run_environment_generators(m); |
1933 | if (r >= 0) | |
1934 | r = manager_run_generators(m); | |
d4ee7bd8 | 1935 | dual_timestamp_get(m->timestamps + manager_timestamp_initrd_mangle(MANAGER_TIMESTAMP_GENERATORS_FINISH)); |
07719a21 LP |
1936 | if (r < 0) |
1937 | return r; | |
1938 | ||
159f1e76 | 1939 | manager_preset_all(m); |
572986ca | 1940 | |
581fef8d | 1941 | lookup_paths_log(&m->lookup_paths); |
572986ca | 1942 | |
4df7d537 ZJS |
1943 | { |
1944 | /* This block is (optionally) done with the reloading counter bumped */ | |
d7ac0952 | 1945 | _unused_ _cleanup_(manager_reloading_stopp) Manager *reloading = NULL; |
4df7d537 | 1946 | |
45f540a2 ZJS |
1947 | /* Make sure we don't have a left-over from a previous run */ |
1948 | if (!serialization) | |
1949 | (void) rm_rf(m->lookup_paths.transient, 0); | |
1950 | ||
4df7d537 ZJS |
1951 | /* If we will deserialize make sure that during enumeration this is already known, so we increase the |
1952 | * counter here already */ | |
1953 | if (serialization) | |
1954 | reloading = manager_reloading_start(m); | |
1955 | ||
1956 | /* First, enumerate what we can from all config files */ | |
1957 | dual_timestamp_get(m->timestamps + manager_timestamp_initrd_mangle(MANAGER_TIMESTAMP_UNITS_LOAD_START)); | |
1958 | manager_enumerate_perpetual(m); | |
1959 | manager_enumerate(m); | |
1960 | dual_timestamp_get(m->timestamps + manager_timestamp_initrd_mangle(MANAGER_TIMESTAMP_UNITS_LOAD_FINISH)); | |
1961 | ||
1962 | /* Second, deserialize if there is something to deserialize */ | |
1963 | if (serialization) { | |
1964 | r = manager_deserialize(m, serialization, fds); | |
1965 | if (r < 0) | |
1966 | return log_error_errno(r, "Deserialization failed: %m"); | |
1967 | } | |
01e10de3 | 1968 | |
4df7d537 ZJS |
1969 | /* Any fds left? Find some unit which wants them. This is useful to allow container managers to pass |
1970 | * some file descriptors to us pre-initialized. This enables socket-based activation of entire | |
1971 | * containers. */ | |
1972 | manager_distribute_fds(m, fds); | |
d86f9d52 | 1973 | |
4df7d537 ZJS |
1974 | /* We might have deserialized the notify fd, but if we didn't then let's create the bus now */ |
1975 | r = manager_setup_notify(m); | |
1976 | if (r < 0) | |
1977 | /* No sense to continue without notifications, our children would fail anyway. */ | |
1978 | return r; | |
d8fdc620 | 1979 | |
4df7d537 ZJS |
1980 | r = manager_setup_cgroups_agent(m); |
1981 | if (r < 0) | |
1982 | /* Likewise, no sense to continue without empty cgroup notifications. */ | |
1983 | return r; | |
00d9ef85 | 1984 | |
4df7d537 ZJS |
1985 | r = manager_setup_user_lookup_fd(m); |
1986 | if (r < 0) | |
1987 | /* This shouldn't fail, except if things are really broken. */ | |
1988 | return r; | |
8559b3b7 | 1989 | |
4df7d537 ZJS |
1990 | /* Connect to the bus if we are good for it */ |
1991 | manager_setup_bus(m); | |
e3dd987c | 1992 | |
5238e957 | 1993 | /* Now that we are connected to all possible buses, let's deserialize who is tracking us. */ |
4df7d537 ZJS |
1994 | r = bus_track_coldplug(m, &m->subscribed, false, m->deserialized_subscribed); |
1995 | if (r < 0) | |
1996 | log_warning_errno(r, "Failed to deserialized tracked clients, ignoring: %m"); | |
1997 | m->deserialized_subscribed = strv_free(m->deserialized_subscribed); | |
a16e1123 | 1998 | |
19d22d43 LP |
1999 | r = manager_varlink_init(m); |
2000 | if (r < 0) | |
064a5c14 | 2001 | log_warning_errno(r, "Failed to set up Varlink, ignoring: %m"); |
19d22d43 | 2002 | |
4df7d537 ZJS |
2003 | /* Third, fire things up! */ |
2004 | manager_coldplug(m); | |
3ad2afb6 | 2005 | |
4df7d537 ZJS |
2006 | /* Clean up runtime objects */ |
2007 | manager_vacuum(m); | |
71445ae7 | 2008 | |
4df7d537 ZJS |
2009 | if (serialization) |
2010 | /* Let's wait for the UnitNew/JobNew messages being sent, before we notify that the | |
2011 | * reload is finished */ | |
2012 | m->send_reloading_done = true; | |
9f611ad8 LP |
2013 | } |
2014 | ||
5ce5e1ad | 2015 | manager_ready(m); |
f0831ed2 | 2016 | |
d35fe8c0 FB |
2017 | manager_set_switching_root(m, false); |
2018 | ||
17f01ace | 2019 | return 0; |
f50e0a01 LP |
2020 | } |
2021 | ||
50cbaba4 LP |
2022 | int manager_add_job( |
2023 | Manager *m, | |
2024 | JobType type, | |
2025 | Unit *unit, | |
2026 | JobMode mode, | |
2027 | Set *affected_jobs, | |
2028 | sd_bus_error *error, | |
2029 | Job **ret) { | |
2030 | ||
c7e3eb5d | 2031 | _cleanup_(transaction_abort_and_freep) Transaction *tr = NULL; |
50cbaba4 | 2032 | int r; |
e5b5ae50 LP |
2033 | |
2034 | assert(m); | |
2035 | assert(type < _JOB_TYPE_MAX); | |
87f0e418 | 2036 | assert(unit); |
e5b5ae50 | 2037 | assert(mode < _JOB_MODE_MAX); |
60918275 | 2038 | |
7358dc02 | 2039 | if (mode == JOB_ISOLATE && type != JOB_START) |
1b09b81c | 2040 | return sd_bus_error_set(error, SD_BUS_ERROR_INVALID_ARGS, "Isolate is only valid for start."); |
c497c7a9 | 2041 | |
7358dc02 | 2042 | if (mode == JOB_ISOLATE && !unit->allow_isolate) |
1b09b81c | 2043 | return sd_bus_error_set(error, BUS_ERROR_NO_ISOLATION, "Operation refused, unit may not be isolated."); |
2528a7a6 | 2044 | |
1f0f9f21 | 2045 | if (mode == JOB_TRIGGERING && type != JOB_STOP) |
1b09b81c | 2046 | return sd_bus_error_set(error, SD_BUS_ERROR_INVALID_ARGS, "--job-mode=triggering is only valid for stop."); |
1f0f9f21 | 2047 | |
f2341e0a | 2048 | log_unit_debug(unit, "Trying to enqueue job %s/%s/%s", unit->id, job_type_to_string(type), job_mode_to_string(mode)); |
9f04bd52 | 2049 | |
c6497ccb | 2050 | type = job_type_collapse(type, unit); |
e0209d83 | 2051 | |
23ade460 | 2052 | tr = transaction_new(mode == JOB_REPLACE_IRREVERSIBLY); |
7527cb52 MS |
2053 | if (!tr) |
2054 | return -ENOMEM; | |
11dd41ce | 2055 | |
4bd29fe5 | 2056 | r = transaction_add_job_and_dependencies(tr, type, unit, NULL, true, false, |
3742095b | 2057 | IN_SET(mode, JOB_IGNORE_DEPENDENCIES, JOB_IGNORE_REQUIREMENTS), |
50cbaba4 | 2058 | mode == JOB_IGNORE_DEPENDENCIES, error); |
7527cb52 | 2059 | if (r < 0) |
c7e3eb5d | 2060 | return r; |
c497c7a9 | 2061 | |
7527cb52 MS |
2062 | if (mode == JOB_ISOLATE) { |
2063 | r = transaction_add_isolate_jobs(tr, m); | |
2064 | if (r < 0) | |
c7e3eb5d | 2065 | return r; |
7527cb52 MS |
2066 | } |
2067 | ||
1f0f9f21 KK |
2068 | if (mode == JOB_TRIGGERING) { |
2069 | r = transaction_add_triggering_jobs(tr, unit); | |
2070 | if (r < 0) | |
c7e3eb5d | 2071 | return r; |
1f0f9f21 KK |
2072 | } |
2073 | ||
50cbaba4 | 2074 | r = transaction_activate(tr, m, mode, affected_jobs, error); |
7527cb52 | 2075 | if (r < 0) |
c7e3eb5d | 2076 | return r; |
e5b5ae50 | 2077 | |
f2341e0a | 2078 | log_unit_debug(unit, |
66870f90 ZJS |
2079 | "Enqueued job %s/%s as %u", unit->id, |
2080 | job_type_to_string(type), (unsigned) tr->anchor_job->id); | |
f50e0a01 | 2081 | |
50cbaba4 LP |
2082 | if (ret) |
2083 | *ret = tr->anchor_job; | |
60918275 | 2084 | |
c7e3eb5d | 2085 | tr = transaction_free(tr); |
e5b5ae50 LP |
2086 | return 0; |
2087 | } | |
60918275 | 2088 | |
50cbaba4 | 2089 | 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 | 2090 | Unit *unit = NULL; /* just to appease gcc, initialization is not really necessary */ |
28247076 LP |
2091 | int r; |
2092 | ||
2093 | assert(m); | |
2094 | assert(type < _JOB_TYPE_MAX); | |
2095 | assert(name); | |
2096 | assert(mode < _JOB_MODE_MAX); | |
2097 | ||
c3090674 LP |
2098 | r = manager_load_unit(m, name, NULL, NULL, &unit); |
2099 | if (r < 0) | |
28247076 | 2100 | return r; |
4440b27d | 2101 | assert(unit); |
28247076 | 2102 | |
50cbaba4 | 2103 | return manager_add_job(m, type, unit, mode, affected_jobs, e, ret); |
53f18416 LP |
2104 | } |
2105 | ||
50cbaba4 | 2106 | int manager_add_job_by_name_and_warn(Manager *m, JobType type, const char *name, JobMode mode, Set *affected_jobs, Job **ret) { |
4afd3348 | 2107 | _cleanup_(sd_bus_error_free) sd_bus_error error = SD_BUS_ERROR_NULL; |
53f18416 LP |
2108 | int r; |
2109 | ||
2110 | assert(m); | |
2111 | assert(type < _JOB_TYPE_MAX); | |
2112 | assert(name); | |
2113 | assert(mode < _JOB_MODE_MAX); | |
2114 | ||
50cbaba4 | 2115 | r = manager_add_job_by_name(m, type, name, mode, affected_jobs, &error, ret); |
53f18416 LP |
2116 | if (r < 0) |
2117 | return log_warning_errno(r, "Failed to enqueue %s job for %s: %s", job_mode_to_string(mode), name, bus_error_message(&error, r)); | |
2118 | ||
2119 | return r; | |
28247076 LP |
2120 | } |
2121 | ||
15d167f8 JW |
2122 | int manager_propagate_reload(Manager *m, Unit *unit, JobMode mode, sd_bus_error *e) { |
2123 | int r; | |
32620826 | 2124 | _cleanup_(transaction_abort_and_freep) Transaction *tr = NULL; |
15d167f8 JW |
2125 | |
2126 | assert(m); | |
2127 | assert(unit); | |
2128 | assert(mode < _JOB_MODE_MAX); | |
2129 | assert(mode != JOB_ISOLATE); /* Isolate is only valid for start */ | |
2130 | ||
2131 | tr = transaction_new(mode == JOB_REPLACE_IRREVERSIBLY); | |
2132 | if (!tr) | |
2133 | return -ENOMEM; | |
2134 | ||
2135 | /* We need an anchor job */ | |
2136 | r = transaction_add_job_and_dependencies(tr, JOB_NOP, unit, NULL, false, false, true, true, e); | |
2137 | if (r < 0) | |
32620826 | 2138 | return r; |
15d167f8 JW |
2139 | |
2140 | /* Failure in adding individual dependencies is ignored, so this always succeeds. */ | |
2141 | transaction_add_propagate_reload_jobs(tr, unit, tr->anchor_job, mode == JOB_IGNORE_DEPENDENCIES, e); | |
2142 | ||
50cbaba4 | 2143 | r = transaction_activate(tr, m, mode, NULL, e); |
15d167f8 | 2144 | if (r < 0) |
32620826 | 2145 | return r; |
15d167f8 | 2146 | |
32620826 | 2147 | tr = transaction_free(tr); |
15d167f8 | 2148 | return 0; |
15d167f8 JW |
2149 | } |
2150 | ||
60918275 LP |
2151 | Job *manager_get_job(Manager *m, uint32_t id) { |
2152 | assert(m); | |
2153 | ||
2154 | return hashmap_get(m->jobs, UINT32_TO_PTR(id)); | |
2155 | } | |
2156 | ||
87f0e418 | 2157 | Unit *manager_get_unit(Manager *m, const char *name) { |
60918275 LP |
2158 | assert(m); |
2159 | assert(name); | |
2160 | ||
87f0e418 | 2161 | return hashmap_get(m->units, name); |
60918275 LP |
2162 | } |
2163 | ||
19496554 MS |
2164 | static int manager_dispatch_target_deps_queue(Manager *m) { |
2165 | Unit *u; | |
19496554 MS |
2166 | int r = 0; |
2167 | ||
19496554 MS |
2168 | assert(m); |
2169 | ||
2170 | while ((u = m->target_deps_queue)) { | |
15ed3c3a LP |
2171 | _cleanup_free_ Unit **targets = NULL; |
2172 | int n_targets; | |
2173 | ||
19496554 MS |
2174 | assert(u->in_target_deps_queue); |
2175 | ||
2176 | LIST_REMOVE(target_deps_queue, u->manager->target_deps_queue, u); | |
2177 | u->in_target_deps_queue = false; | |
2178 | ||
15ed3c3a LP |
2179 | /* Take an "atomic" snapshot of dependencies here, as the call below will likely modify the |
2180 | * dependencies, and we can't have it that hash tables we iterate through are modified while | |
2181 | * we are iterating through them. */ | |
2182 | n_targets = unit_get_dependency_array(u, UNIT_ATOM_DEFAULT_TARGET_DEPENDENCIES, &targets); | |
2183 | if (n_targets < 0) | |
2184 | return n_targets; | |
19496554 | 2185 | |
15ed3c3a LP |
2186 | for (int i = 0; i < n_targets; i++) { |
2187 | r = unit_add_default_target_dependency(u, targets[i]); | |
2188 | if (r < 0) | |
2189 | return r; | |
19496554 MS |
2190 | } |
2191 | } | |
2192 | ||
2193 | return r; | |
2194 | } | |
2195 | ||
c1e1601e | 2196 | unsigned manager_dispatch_load_queue(Manager *m) { |
595ed347 | 2197 | Unit *u; |
c1e1601e | 2198 | unsigned n = 0; |
60918275 LP |
2199 | |
2200 | assert(m); | |
2201 | ||
223dabab LP |
2202 | /* Make sure we are not run recursively */ |
2203 | if (m->dispatching_load_queue) | |
c1e1601e | 2204 | return 0; |
223dabab LP |
2205 | |
2206 | m->dispatching_load_queue = true; | |
2207 | ||
87f0e418 | 2208 | /* Dispatches the load queue. Takes a unit from the queue and |
60918275 LP |
2209 | * tries to load its data until the queue is empty */ |
2210 | ||
595ed347 MS |
2211 | while ((u = m->load_queue)) { |
2212 | assert(u->in_load_queue); | |
034c6ed7 | 2213 | |
595ed347 | 2214 | unit_load(u); |
c1e1601e | 2215 | n++; |
60918275 LP |
2216 | } |
2217 | ||
223dabab | 2218 | m->dispatching_load_queue = false; |
19496554 MS |
2219 | |
2220 | /* Dispatch the units waiting for their target dependencies to be added now, as all targets that we know about | |
2221 | * should be loaded and have aliases resolved */ | |
2222 | (void) manager_dispatch_target_deps_queue(m); | |
2223 | ||
c1e1601e | 2224 | return n; |
60918275 LP |
2225 | } |
2226 | ||
81be2388 | 2227 | bool manager_unit_cache_should_retry_load(Unit *u) { |
cda66772 LB |
2228 | assert(u); |
2229 | ||
81be2388 ZJS |
2230 | /* Automatic reloading from disk only applies to units which were not found sometime in the past, and |
2231 | * the not-found stub is kept pinned in the unit graph by dependencies. For units that were | |
2232 | * previously loaded, we don't do automatic reloading, and daemon-reload is necessary to update. */ | |
cda66772 LB |
2233 | if (u->load_state != UNIT_NOT_FOUND) |
2234 | return false; | |
2235 | ||
81be2388 ZJS |
2236 | /* The cache has been updated since the last time we tried to load the unit. There might be new |
2237 | * fragment paths to read. */ | |
c2911d48 | 2238 | if (u->manager->unit_cache_timestamp_hash != u->fragment_not_found_timestamp_hash) |
cda66772 | 2239 | return true; |
d904afc7 | 2240 | |
81be2388 | 2241 | /* The cache needs to be updated because there are modifications on disk. */ |
c2911d48 | 2242 | return !lookup_paths_timestamp_hash_same(&u->manager->lookup_paths, u->manager->unit_cache_timestamp_hash, NULL); |
d904afc7 LB |
2243 | } |
2244 | ||
c2756a68 LP |
2245 | int manager_load_unit_prepare( |
2246 | Manager *m, | |
2247 | const char *name, | |
2248 | const char *path, | |
718db961 | 2249 | sd_bus_error *e, |
4b6a2b3f | 2250 | Unit **ret) { |
c2756a68 | 2251 | |
4b6a2b3f | 2252 | _cleanup_(unit_freep) Unit *cleanup_unit = NULL; |
a99626c1 | 2253 | _cleanup_free_ char *nbuf = NULL; |
60918275 LP |
2254 | int r; |
2255 | ||
2256 | assert(m); | |
4b6a2b3f | 2257 | assert(ret); |
a99626c1 | 2258 | assert(name || path); |
60918275 | 2259 | |
526e3cbb | 2260 | /* This will prepare the unit for loading, but not actually load anything from disk. */ |
0301abf4 | 2261 | |
526e3cbb | 2262 | if (path && !path_is_absolute(path)) |
718db961 | 2263 | return sd_bus_error_setf(e, SD_BUS_ERROR_INVALID_ARGS, "Path %s is not absolute.", path); |
9e2f7c11 | 2264 | |
08f46856 | 2265 | if (!name) { |
a99626c1 LP |
2266 | r = path_extract_filename(path, &nbuf); |
2267 | if (r < 0) | |
2268 | return r; | |
2269 | if (r == O_DIRECTORY) | |
2270 | return sd_bus_error_setf(e, SD_BUS_ERROR_INVALID_ARGS, "Path '%s' refers to directory, refusing.", path); | |
2271 | ||
2272 | name = nbuf; | |
08f46856 | 2273 | } |
9e2f7c11 | 2274 | |
4b6a2b3f | 2275 | UnitType t = unit_name_to_type(name); |
7d17cfbc | 2276 | |
5d512d54 LN |
2277 | if (t == _UNIT_TYPE_INVALID || !unit_name_is_valid(name, UNIT_NAME_PLAIN|UNIT_NAME_INSTANCE)) { |
2278 | if (unit_name_is_valid(name, UNIT_NAME_TEMPLATE)) | |
2279 | return sd_bus_error_setf(e, SD_BUS_ERROR_INVALID_ARGS, "Unit name %s is missing the instance name.", name); | |
2280 | ||
718db961 | 2281 | return sd_bus_error_setf(e, SD_BUS_ERROR_INVALID_ARGS, "Unit name %s is not valid.", name); |
5d512d54 | 2282 | } |
60918275 | 2283 | |
4b6a2b3f ZJS |
2284 | Unit *unit = manager_get_unit(m, name); |
2285 | if (unit) { | |
d904afc7 LB |
2286 | /* The time-based cache allows to start new units without daemon-reload, |
2287 | * but if they are already referenced (because of dependencies or ordering) | |
2288 | * then we have to force a load of the fragment. As an optimization, check | |
2289 | * first if anything in the usual paths was modified since the last time | |
7233e91a LB |
2290 | * the cache was loaded. Also check if the last time an attempt to load the |
2291 | * unit was made was before the most recent cache refresh, so that we know | |
81be2388 | 2292 | * we need to try again — even if the cache is current, it might have been |
7233e91a LB |
2293 | * updated in a different context before we had a chance to retry loading |
2294 | * this particular unit. */ | |
4b6a2b3f ZJS |
2295 | if (manager_unit_cache_should_retry_load(unit)) |
2296 | unit->load_state = UNIT_STUB; | |
d904afc7 | 2297 | else { |
4b6a2b3f | 2298 | *ret = unit; |
535b7fcb | 2299 | return 0; /* The unit was already loaded */ |
d904afc7 LB |
2300 | } |
2301 | } else { | |
4b6a2b3f ZJS |
2302 | unit = cleanup_unit = unit_new(m, unit_vtable[t]->object_size); |
2303 | if (!unit) | |
d904afc7 | 2304 | return -ENOMEM; |
034c6ed7 | 2305 | } |
60918275 | 2306 | |
7d17cfbc | 2307 | if (path) { |
4b6a2b3f | 2308 | r = free_and_strdup(&unit->fragment_path, path); |
d904afc7 LB |
2309 | if (r < 0) |
2310 | return r; | |
7d17cfbc | 2311 | } |
0301abf4 | 2312 | |
4b6a2b3f | 2313 | r = unit_add_name(unit, name); |
dc409696 | 2314 | if (r < 0) |
1ffba6fe | 2315 | return r; |
60918275 | 2316 | |
4b6a2b3f ZJS |
2317 | unit_add_to_load_queue(unit); |
2318 | unit_add_to_dbus_queue(unit); | |
2319 | unit_add_to_gc_queue(unit); | |
c1e1601e | 2320 | |
4b6a2b3f ZJS |
2321 | *ret = unit; |
2322 | TAKE_PTR(cleanup_unit); | |
db06e3b6 | 2323 | |
535b7fcb | 2324 | return 1; /* The unit was added the load queue */ |
db06e3b6 LP |
2325 | } |
2326 | ||
c2756a68 LP |
2327 | int manager_load_unit( |
2328 | Manager *m, | |
2329 | const char *name, | |
2330 | const char *path, | |
718db961 | 2331 | sd_bus_error *e, |
4b6a2b3f | 2332 | Unit **ret) { |
db06e3b6 LP |
2333 | int r; |
2334 | ||
2335 | assert(m); | |
4b6a2b3f | 2336 | assert(ret); |
db06e3b6 | 2337 | |
4b6a2b3f | 2338 | /* This will load the unit config, but not actually start any services or anything. */ |
db06e3b6 | 2339 | |
4b6a2b3f | 2340 | r = manager_load_unit_prepare(m, name, path, e, ret); |
535b7fcb | 2341 | if (r <= 0) |
db06e3b6 LP |
2342 | return r; |
2343 | ||
535b7fcb | 2344 | /* Unit was newly loaded */ |
f50e0a01 | 2345 | manager_dispatch_load_queue(m); |
4b6a2b3f | 2346 | *ret = unit_follow_merge(*ret); |
4109ede7 ZJS |
2347 | return 0; |
2348 | } | |
2349 | ||
2350 | int manager_load_startable_unit_or_warn( | |
2351 | Manager *m, | |
2352 | const char *name, | |
2353 | const char *path, | |
2354 | Unit **ret) { | |
2355 | ||
2356 | /* Load a unit, make sure it loaded fully and is not masked. */ | |
2357 | ||
2358 | _cleanup_(sd_bus_error_free) sd_bus_error error = SD_BUS_ERROR_NULL; | |
2359 | Unit *unit; | |
2360 | int r; | |
2361 | ||
2362 | r = manager_load_unit(m, name, path, &error, &unit); | |
2363 | if (r < 0) | |
2364 | return log_error_errno(r, "Failed to load %s %s: %s", | |
8ace1db7 | 2365 | name ? "unit" : "unit file", name ?: path, |
4109ede7 | 2366 | bus_error_message(&error, r)); |
8ace1db7 LP |
2367 | |
2368 | r = bus_unit_validate_load_state(unit, &error); | |
2369 | if (r < 0) | |
2370 | return log_error_errno(r, "%s", bus_error_message(&error, r)); | |
9e2f7c11 | 2371 | |
4109ede7 | 2372 | *ret = unit; |
60918275 LP |
2373 | return 0; |
2374 | } | |
a66d02c3 | 2375 | |
7fad411c LP |
2376 | void manager_clear_jobs(Manager *m) { |
2377 | Job *j; | |
2378 | ||
2379 | assert(m); | |
2380 | ||
7fad411c | 2381 | while ((j = hashmap_first(m->jobs))) |
5273510e | 2382 | /* No need to recurse. We're cancelling all jobs. */ |
833f92ad | 2383 | job_finish_and_invalidate(j, JOB_CANCELED, false, false); |
7fad411c | 2384 | } |
83c60c9f | 2385 | |
f75f613d FB |
2386 | void manager_unwatch_pid(Manager *m, pid_t pid) { |
2387 | assert(m); | |
2388 | ||
2389 | /* First let's drop the unit keyed as "pid". */ | |
2390 | (void) hashmap_remove(m->watch_pids, PID_TO_PTR(pid)); | |
2391 | ||
2392 | /* Then, let's also drop the array keyed by -pid. */ | |
2393 | free(hashmap_remove(m->watch_pids, PID_TO_PTR(-pid))); | |
2394 | } | |
2395 | ||
752b5905 | 2396 | static int manager_dispatch_run_queue(sd_event_source *source, void *userdata) { |
99534007 | 2397 | Manager *m = ASSERT_PTR(userdata); |
83c60c9f | 2398 | Job *j; |
034c6ed7 | 2399 | |
752b5905 | 2400 | assert(source); |
9152c765 | 2401 | |
da8e1782 | 2402 | while ((j = prioq_peek(m->run_queue))) { |
ac1135be | 2403 | assert(j->installed); |
034c6ed7 LP |
2404 | assert(j->in_run_queue); |
2405 | ||
6e64994d | 2406 | (void) job_run_and_invalidate(j); |
9152c765 | 2407 | } |
034c6ed7 | 2408 | |
a0b64226 | 2409 | if (m->n_running_jobs > 0) |
03b717a3 MS |
2410 | manager_watch_jobs_in_progress(m); |
2411 | ||
31a7eb86 ZJS |
2412 | if (m->n_on_console > 0) |
2413 | manager_watch_idle_pipe(m); | |
2414 | ||
752b5905 | 2415 | return 1; |
c1e1601e LP |
2416 | } |
2417 | ||
b0c4b282 LP |
2418 | void manager_trigger_run_queue(Manager *m) { |
2419 | int r; | |
2420 | ||
2421 | assert(m); | |
2422 | ||
2423 | r = sd_event_source_set_enabled( | |
2424 | m->run_queue_event_source, | |
2425 | prioq_isempty(m->run_queue) ? SD_EVENT_OFF : SD_EVENT_ONESHOT); | |
2426 | if (r < 0) | |
2427 | log_warning_errno(r, "Failed to enable job run queue event source, ignoring: %m"); | |
2428 | } | |
2429 | ||
9588bc32 | 2430 | static unsigned manager_dispatch_dbus_queue(Manager *m) { |
e0a08581 | 2431 | unsigned n = 0, budget; |
595ed347 | 2432 | Unit *u; |
e0a08581 | 2433 | Job *j; |
c1e1601e LP |
2434 | |
2435 | assert(m); | |
2436 | ||
b8d381c4 LP |
2437 | /* When we are reloading, let's not wait with generating signals, since we need to exit the manager as quickly |
2438 | * as we can. There's no point in throttling generation of signals in that case. */ | |
2439 | if (MANAGER_IS_RELOADING(m) || m->send_reloading_done || m->pending_reload_message) | |
f5fbe71d | 2440 | budget = UINT_MAX; /* infinite budget in this case */ |
b8d381c4 LP |
2441 | else { |
2442 | /* Anything to do at all? */ | |
2443 | if (!m->dbus_unit_queue && !m->dbus_job_queue) | |
2444 | return 0; | |
2445 | ||
2446 | /* Do we have overly many messages queued at the moment? If so, let's not enqueue more on top, let's | |
2447 | * sit this cycle out, and process things in a later cycle when the queues got a bit emptier. */ | |
2448 | if (manager_bus_n_queued_write(m) > MANAGER_BUS_BUSY_THRESHOLD) | |
2449 | return 0; | |
2450 | ||
2451 | /* Only process a certain number of units/jobs per event loop iteration. Even if the bus queue wasn't | |
2452 | * overly full before this call we shouldn't increase it in size too wildly in one step, and we | |
2453 | * shouldn't monopolize CPU time with generating these messages. Note the difference in counting of | |
2454 | * this "budget" and the "threshold" above: the "budget" is decreased only once per generated message, | |
5238e957 BB |
2455 | * regardless how many buses/direct connections it is enqueued on, while the "threshold" is applied to |
2456 | * each queued instance of bus message, i.e. if the same message is enqueued to five buses/direct | |
b8d381c4 LP |
2457 | * connections it will be counted five times. This difference in counting ("references" |
2458 | * vs. "instances") is primarily a result of the fact that it's easier to implement it this way, | |
2459 | * however it also reflects the thinking that the "threshold" should put a limit on used queue memory, | |
2460 | * i.e. space, while the "budget" should put a limit on time. Also note that the "threshold" is | |
2461 | * currently chosen much higher than the "budget". */ | |
2462 | budget = MANAGER_BUS_MESSAGE_BUDGET; | |
2463 | } | |
e0a08581 | 2464 | |
b8d381c4 | 2465 | while (budget != 0 && (u = m->dbus_unit_queue)) { |
e0a08581 | 2466 | |
595ed347 | 2467 | assert(u->in_dbus_queue); |
c1e1601e | 2468 | |
595ed347 | 2469 | bus_unit_send_change_signal(u); |
b8d381c4 LP |
2470 | n++; |
2471 | ||
f5fbe71d | 2472 | if (budget != UINT_MAX) |
b8d381c4 | 2473 | budget--; |
c1e1601e LP |
2474 | } |
2475 | ||
b8d381c4 | 2476 | while (budget != 0 && (j = m->dbus_job_queue)) { |
c1e1601e LP |
2477 | assert(j->in_dbus_queue); |
2478 | ||
2479 | bus_job_send_change_signal(j); | |
b8d381c4 LP |
2480 | n++; |
2481 | ||
f5fbe71d | 2482 | if (budget != UINT_MAX) |
b8d381c4 | 2483 | budget--; |
c1e1601e LP |
2484 | } |
2485 | ||
b8d381c4 | 2486 | if (m->send_reloading_done) { |
71445ae7 | 2487 | m->send_reloading_done = false; |
718db961 | 2488 | bus_manager_send_reloading(m, false); |
b8d381c4 | 2489 | n++; |
71445ae7 LP |
2490 | } |
2491 | ||
b8d381c4 | 2492 | if (m->pending_reload_message) { |
209de525 | 2493 | bus_send_pending_reload_message(m); |
e0a08581 LP |
2494 | n++; |
2495 | } | |
718db961 | 2496 | |
c1e1601e | 2497 | return n; |
9152c765 LP |
2498 | } |
2499 | ||
d8fdc620 LP |
2500 | static int manager_dispatch_cgroups_agent_fd(sd_event_source *source, int fd, uint32_t revents, void *userdata) { |
2501 | Manager *m = userdata; | |
60473f0c | 2502 | char buf[PATH_MAX]; |
d8fdc620 LP |
2503 | ssize_t n; |
2504 | ||
2505 | n = recv(fd, buf, sizeof(buf), 0); | |
2506 | if (n < 0) | |
2507 | return log_error_errno(errno, "Failed to read cgroups agent message: %m"); | |
2508 | if (n == 0) { | |
2509 | log_error("Got zero-length cgroups agent message, ignoring."); | |
2510 | return 0; | |
2511 | } | |
2512 | if ((size_t) n >= sizeof(buf)) { | |
2513 | log_error("Got overly long cgroups agent message, ignoring."); | |
2514 | return 0; | |
2515 | } | |
2516 | ||
2517 | if (memchr(buf, 0, n)) { | |
2518 | log_error("Got cgroups agent message with embedded NUL byte, ignoring."); | |
2519 | return 0; | |
2520 | } | |
2521 | buf[n] = 0; | |
2522 | ||
2523 | manager_notify_cgroup_empty(m, buf); | |
d5f15326 | 2524 | (void) bus_forward_agent_released(m, buf); |
d8fdc620 LP |
2525 | |
2526 | return 0; | |
2527 | } | |
2528 | ||
5151b4cc | 2529 | static bool manager_process_barrier_fd(char * const *tags, FDSet *fds) { |
4f07ddfa KKD |
2530 | |
2531 | /* nothing else must be sent when using BARRIER=1 */ | |
5151b4cc | 2532 | if (strv_contains(tags, "BARRIER=1")) { |
46276454 | 2533 | if (strv_length(tags) != 1) |
5151b4cc | 2534 | log_warning("Extra notification messages sent with BARRIER=1, ignoring everything."); |
46276454 LP |
2535 | else if (fdset_size(fds) != 1) |
2536 | log_warning("Got incorrect number of fds with BARRIER=1, closing them."); | |
5151b4cc BR |
2537 | |
2538 | /* Drop the message if BARRIER=1 was found */ | |
4f07ddfa | 2539 | return true; |
5151b4cc | 2540 | } |
4f07ddfa KKD |
2541 | |
2542 | return false; | |
2543 | } | |
2544 | ||
db256aab LP |
2545 | static void manager_invoke_notify_message( |
2546 | Manager *m, | |
2547 | Unit *u, | |
2548 | const struct ucred *ucred, | |
5151b4cc | 2549 | char * const *tags, |
db256aab LP |
2550 | FDSet *fds) { |
2551 | ||
5ba6985b LP |
2552 | assert(m); |
2553 | assert(u); | |
db256aab | 2554 | assert(ucred); |
5151b4cc | 2555 | assert(tags); |
5ba6985b | 2556 | |
62a76913 | 2557 | if (u->notifygen == m->notifygen) /* Already invoked on this same unit in this same iteration? */ |
5ba6985b | 2558 | return; |
62a76913 LP |
2559 | u->notifygen = m->notifygen; |
2560 | ||
5151b4cc | 2561 | if (UNIT_VTABLE(u)->notify_message) |
db256aab | 2562 | UNIT_VTABLE(u)->notify_message(u, ucred, tags, fds); |
62a76913 | 2563 | |
5151b4cc BR |
2564 | else if (DEBUG_LOGGING) { |
2565 | _cleanup_free_ char *buf = NULL, *x = NULL, *y = NULL; | |
a86b7675 | 2566 | |
5151b4cc BR |
2567 | buf = strv_join(tags, ", "); |
2568 | if (buf) | |
2569 | x = ellipsize(buf, 20, 90); | |
a86b7675 | 2570 | if (x) |
da5fb861 LP |
2571 | y = cescape(x); |
2572 | ||
a86b7675 ZJS |
2573 | log_unit_debug(u, "Got notification message \"%s\", ignoring.", strnull(y)); |
2574 | } | |
5ba6985b LP |
2575 | } |
2576 | ||
718db961 | 2577 | static int manager_dispatch_notify_fd(sd_event_source *source, int fd, uint32_t revents, void *userdata) { |
3d0b8a55 | 2578 | |
b215b0ed | 2579 | _cleanup_fdset_free_ FDSet *fds = NULL; |
99534007 | 2580 | Manager *m = ASSERT_PTR(userdata); |
b215b0ed DH |
2581 | char buf[NOTIFY_BUFFER_MAX+1]; |
2582 | struct iovec iovec = { | |
2583 | .iov_base = buf, | |
2584 | .iov_len = sizeof(buf)-1, | |
2585 | }; | |
fb29cdbe LP |
2586 | CMSG_BUFFER_TYPE(CMSG_SPACE(sizeof(struct ucred)) + |
2587 | CMSG_SPACE(sizeof(int) * NOTIFY_FD_MAX)) control; | |
b215b0ed DH |
2588 | struct msghdr msghdr = { |
2589 | .msg_iov = &iovec, | |
2590 | .msg_iovlen = 1, | |
2591 | .msg_control = &control, | |
2592 | .msg_controllen = sizeof(control), | |
2593 | }; | |
2594 | ||
2595 | struct cmsghdr *cmsg; | |
2596 | struct ucred *ucred = NULL; | |
62a76913 | 2597 | _cleanup_free_ Unit **array_copy = NULL; |
5151b4cc | 2598 | _cleanup_strv_free_ char **tags = NULL; |
62a76913 | 2599 | Unit *u1, *u2, **array; |
b215b0ed | 2600 | int r, *fd_array = NULL; |
da6053d0 | 2601 | size_t n_fds = 0; |
62a76913 | 2602 | bool found = false; |
8c47c732 LP |
2603 | ssize_t n; |
2604 | ||
718db961 LP |
2605 | assert(m->notify_fd == fd); |
2606 | ||
2607 | if (revents != EPOLLIN) { | |
2608 | log_warning("Got unexpected poll event for notify fd."); | |
2609 | return 0; | |
2610 | } | |
8c47c732 | 2611 | |
3691bcf3 | 2612 | n = recvmsg_safe(m->notify_fd, &msghdr, MSG_DONTWAIT|MSG_CMSG_CLOEXEC|MSG_TRUNC); |
8add30a0 YW |
2613 | if (n < 0) { |
2614 | if (ERRNO_IS_TRANSIENT(n)) | |
2615 | return 0; /* Spurious wakeup, try again */ | |
2616 | if (n == -EXFULL) { | |
2617 | log_warning("Got message with truncated control data (too many fds sent?), ignoring."); | |
2618 | return 0; | |
2619 | } | |
3691bcf3 LP |
2620 | /* If this is any other, real error, then let's stop processing this socket. This of course |
2621 | * means we won't take notification messages anymore, but that's still better than busy | |
2622 | * looping around this: being woken up over and over again but being unable to actually read | |
2623 | * the message off the socket. */ | |
2624 | return log_error_errno(n, "Failed to receive notification message: %m"); | |
8add30a0 | 2625 | } |
a354329f | 2626 | |
b215b0ed DH |
2627 | CMSG_FOREACH(cmsg, &msghdr) { |
2628 | if (cmsg->cmsg_level == SOL_SOCKET && cmsg->cmsg_type == SCM_RIGHTS) { | |
a354329f | 2629 | |
3691bcf3 | 2630 | assert(!fd_array); |
b1d02191 | 2631 | fd_array = CMSG_TYPED_DATA(cmsg, int); |
b215b0ed | 2632 | n_fds = (cmsg->cmsg_len - CMSG_LEN(0)) / sizeof(int); |
a354329f | 2633 | |
b215b0ed DH |
2634 | } else if (cmsg->cmsg_level == SOL_SOCKET && |
2635 | cmsg->cmsg_type == SCM_CREDENTIALS && | |
2636 | cmsg->cmsg_len == CMSG_LEN(sizeof(struct ucred))) { | |
a354329f | 2637 | |
3691bcf3 | 2638 | assert(!ucred); |
b1d02191 | 2639 | ucred = CMSG_TYPED_DATA(cmsg, struct ucred); |
a354329f | 2640 | } |
b215b0ed | 2641 | } |
a354329f | 2642 | |
b215b0ed DH |
2643 | if (n_fds > 0) { |
2644 | assert(fd_array); | |
a354329f | 2645 | |
b215b0ed DH |
2646 | r = fdset_new_array(&fds, fd_array, n_fds); |
2647 | if (r < 0) { | |
2648 | close_many(fd_array, n_fds); | |
9987750e FB |
2649 | log_oom(); |
2650 | return 0; | |
a354329f | 2651 | } |
b215b0ed | 2652 | } |
8c47c732 | 2653 | |
15e23e8c | 2654 | if (!ucred || !pid_is_valid(ucred->pid)) { |
b215b0ed DH |
2655 | log_warning("Received notify message without valid credentials. Ignoring."); |
2656 | return 0; | |
2657 | } | |
8c47c732 | 2658 | |
045a3d59 | 2659 | if ((size_t) n >= sizeof(buf) || (msghdr.msg_flags & MSG_TRUNC)) { |
b215b0ed DH |
2660 | log_warning("Received notify message exceeded maximum size. Ignoring."); |
2661 | return 0; | |
2662 | } | |
8c47c732 | 2663 | |
8f41e6b6 ZJS |
2664 | /* As extra safety check, let's make sure the string we get doesn't contain embedded NUL bytes. |
2665 | * We permit one trailing NUL byte in the message, but don't expect it. */ | |
875ca88d LP |
2666 | if (n > 1 && memchr(buf, 0, n-1)) { |
2667 | log_warning("Received notify message with embedded NUL bytes. Ignoring."); | |
2668 | return 0; | |
2669 | } | |
2670 | ||
8f41e6b6 | 2671 | /* Make sure it's NUL-terminated, then parse it to obtain the tags list. */ |
b215b0ed | 2672 | buf[n] = 0; |
5151b4cc BR |
2673 | tags = strv_split_newlines(buf); |
2674 | if (!tags) { | |
2675 | log_oom(); | |
2676 | return 0; | |
2677 | } | |
8c47c732 | 2678 | |
8f41e6b6 | 2679 | /* Possibly a barrier fd, let's see. */ |
7b0a1267 LP |
2680 | if (manager_process_barrier_fd(tags, fds)) { |
2681 | log_debug("Received barrier notification message from PID " PID_FMT ".", ucred->pid); | |
4f07ddfa | 2682 | return 0; |
7b0a1267 | 2683 | } |
4f07ddfa | 2684 | |
62a76913 LP |
2685 | /* Increase the generation counter used for filtering out duplicate unit invocations. */ |
2686 | m->notifygen++; | |
2687 | ||
2688 | /* Notify every unit that might be interested, which might be multiple. */ | |
b215b0ed | 2689 | u1 = manager_get_unit_by_pid_cgroup(m, ucred->pid); |
62a76913 LP |
2690 | u2 = hashmap_get(m->watch_pids, PID_TO_PTR(ucred->pid)); |
2691 | array = hashmap_get(m->watch_pids, PID_TO_PTR(-ucred->pid)); | |
2692 | if (array) { | |
2693 | size_t k = 0; | |
5ba6985b | 2694 | |
62a76913 LP |
2695 | while (array[k]) |
2696 | k++; | |
5ba6985b | 2697 | |
62a76913 LP |
2698 | array_copy = newdup(Unit*, array, k+1); |
2699 | if (!array_copy) | |
2700 | log_oom(); | |
2701 | } | |
8f41e6b6 ZJS |
2702 | /* And now invoke the per-unit callbacks. Note that manager_invoke_notify_message() will handle |
2703 | * duplicate units make sure we only invoke each unit's handler once. */ | |
62a76913 | 2704 | if (u1) { |
5151b4cc | 2705 | manager_invoke_notify_message(m, u1, ucred, tags, fds); |
62a76913 LP |
2706 | found = true; |
2707 | } | |
2708 | if (u2) { | |
5151b4cc | 2709 | manager_invoke_notify_message(m, u2, ucred, tags, fds); |
62a76913 LP |
2710 | found = true; |
2711 | } | |
2712 | if (array_copy) | |
2713 | for (size_t i = 0; array_copy[i]; i++) { | |
5151b4cc | 2714 | manager_invoke_notify_message(m, array_copy[i], ucred, tags, fds); |
62a76913 LP |
2715 | found = true; |
2716 | } | |
8c47c732 | 2717 | |
62a76913 LP |
2718 | if (!found) |
2719 | log_warning("Cannot find unit for notify message of PID "PID_FMT", ignoring.", ucred->pid); | |
a354329f | 2720 | |
b215b0ed | 2721 | if (fdset_size(fds) > 0) |
5fd2c135 | 2722 | log_warning("Got extra auxiliary fds with notification message, closing them."); |
8c47c732 LP |
2723 | |
2724 | return 0; | |
2725 | } | |
2726 | ||
62a76913 LP |
2727 | static void manager_invoke_sigchld_event( |
2728 | Manager *m, | |
2729 | Unit *u, | |
2730 | const siginfo_t *si) { | |
36f20ae3 | 2731 | |
5ba6985b LP |
2732 | assert(m); |
2733 | assert(u); | |
2734 | assert(si); | |
2735 | ||
62a76913 LP |
2736 | /* Already invoked the handler of this unit in this iteration? Then don't process this again */ |
2737 | if (u->sigchldgen == m->sigchldgen) | |
2738 | return; | |
2739 | u->sigchldgen = m->sigchldgen; | |
5ba6985b | 2740 | |
62a76913 | 2741 | log_unit_debug(u, "Child "PID_FMT" belongs to %s.", si->si_pid, u->id); |
5ba6985b | 2742 | unit_unwatch_pid(u, si->si_pid); |
e57051f5 | 2743 | |
62a76913 LP |
2744 | if (UNIT_VTABLE(u)->sigchld_event) |
2745 | UNIT_VTABLE(u)->sigchld_event(u, si->si_pid, si->si_code, si->si_status); | |
5ba6985b LP |
2746 | } |
2747 | ||
575b300b | 2748 | static int manager_dispatch_sigchld(sd_event_source *source, void *userdata) { |
99534007 | 2749 | Manager *m = ASSERT_PTR(userdata); |
575b300b LP |
2750 | siginfo_t si = {}; |
2751 | int r; | |
2752 | ||
2753 | assert(source); | |
9152c765 | 2754 | |
8f41e6b6 ZJS |
2755 | /* First we call waitid() for a PID and do not reap the zombie. That way we can still access |
2756 | * /proc/$PID for it while it is a zombie. */ | |
9152c765 | 2757 | |
575b300b | 2758 | if (waitid(P_ALL, 0, &si, WEXITED|WNOHANG|WNOWAIT) < 0) { |
acbb0225 | 2759 | |
8afabc50 AJ |
2760 | if (errno != ECHILD) |
2761 | log_error_errno(errno, "Failed to peek for child with waitid(), ignoring: %m"); | |
acbb0225 | 2762 | |
8afabc50 | 2763 | goto turn_off; |
575b300b | 2764 | } |
4112df16 | 2765 | |
575b300b LP |
2766 | if (si.si_pid <= 0) |
2767 | goto turn_off; | |
2768 | ||
2769 | if (IN_SET(si.si_code, CLD_EXITED, CLD_KILLED, CLD_DUMPED)) { | |
62a76913 | 2770 | _cleanup_free_ Unit **array_copy = NULL; |
575b300b | 2771 | _cleanup_free_ char *name = NULL; |
62a76913 | 2772 | Unit *u1, *u2, **array; |
575b300b LP |
2773 | |
2774 | (void) get_process_comm(si.si_pid, &name); | |
2775 | ||
2776 | log_debug("Child "PID_FMT" (%s) died (code=%s, status=%i/%s)", | |
2777 | si.si_pid, strna(name), | |
2778 | sigchld_code_to_string(si.si_code), | |
2779 | si.si_status, | |
2780 | strna(si.si_code == CLD_EXITED | |
2781 | ? exit_status_to_string(si.si_status, EXIT_STATUS_FULL) | |
2782 | : signal_to_string(si.si_status))); | |
2783 | ||
62a76913 LP |
2784 | /* Increase the generation counter used for filtering out duplicate unit invocations */ |
2785 | m->sigchldgen++; | |
2786 | ||
2787 | /* And now figure out the unit this belongs to, it might be multiple... */ | |
575b300b | 2788 | u1 = manager_get_unit_by_pid_cgroup(m, si.si_pid); |
62a76913 LP |
2789 | u2 = hashmap_get(m->watch_pids, PID_TO_PTR(si.si_pid)); |
2790 | array = hashmap_get(m->watch_pids, PID_TO_PTR(-si.si_pid)); | |
2791 | if (array) { | |
2792 | size_t n = 0; | |
2793 | ||
5238e957 | 2794 | /* Count how many entries the array has */ |
62a76913 LP |
2795 | while (array[n]) |
2796 | n++; | |
2797 | ||
2798 | /* Make a copy of the array so that we don't trip up on the array changing beneath us */ | |
2799 | array_copy = newdup(Unit*, array, n+1); | |
2800 | if (!array_copy) | |
2801 | log_oom(); | |
2802 | } | |
2803 | ||
2804 | /* Finally, execute them all. Note that u1, u2 and the array might contain duplicates, but | |
2805 | * that's fine, manager_invoke_sigchld_event() will ensure we only invoke the handlers once for | |
2806 | * each iteration. */ | |
2ba6ae6b ZJS |
2807 | if (u1) { |
2808 | /* We check for oom condition, in case we got SIGCHLD before the oom notification. | |
2809 | * We only do this for the cgroup the PID belonged to. */ | |
2810 | (void) unit_check_oom(u1); | |
2811 | ||
288bd406 | 2812 | /* We check if systemd-oomd performed a kill so that we log and notify appropriately */ |
fe8d22fb AZ |
2813 | (void) unit_check_oomd_kill(u1); |
2814 | ||
62a76913 | 2815 | manager_invoke_sigchld_event(m, u1, &si); |
2ba6ae6b | 2816 | } |
62a76913 LP |
2817 | if (u2) |
2818 | manager_invoke_sigchld_event(m, u2, &si); | |
2819 | if (array_copy) | |
2820 | for (size_t i = 0; array_copy[i]; i++) | |
2821 | manager_invoke_sigchld_event(m, array_copy[i], &si); | |
575b300b | 2822 | } |
9152c765 | 2823 | |
575b300b LP |
2824 | /* And now, we actually reap the zombie. */ |
2825 | if (waitid(P_PID, si.si_pid, &si, WEXITED) < 0) { | |
2826 | log_error_errno(errno, "Failed to dequeue child, ignoring: %m"); | |
2827 | return 0; | |
2828 | } | |
9152c765 | 2829 | |
575b300b | 2830 | return 0; |
8c47c732 | 2831 | |
575b300b LP |
2832 | turn_off: |
2833 | /* All children processed for now, turn off event source */ | |
4112df16 | 2834 | |
575b300b LP |
2835 | r = sd_event_source_set_enabled(m->sigchld_event_source, SD_EVENT_OFF); |
2836 | if (r < 0) | |
2837 | return log_error_errno(r, "Failed to disable SIGCHLD event source: %m"); | |
9152c765 LP |
2838 | |
2839 | return 0; | |
2840 | } | |
2841 | ||
5f968096 ZJS |
2842 | static void manager_start_special(Manager *m, const char *name, JobMode mode) { |
2843 | Job *job; | |
398ef8ba | 2844 | |
5f968096 ZJS |
2845 | if (manager_add_job_by_name_and_warn(m, JOB_START, name, mode, NULL, &job) < 0) |
2846 | return; | |
1e001f52 | 2847 | |
5f968096 ZJS |
2848 | const char *s = unit_status_string(job->unit, NULL); |
2849 | ||
2850 | log_info("Activating special unit %s...", s); | |
3889fc6f ZJS |
2851 | |
2852 | sd_notifyf(false, | |
2853 | "STATUS=Activating special unit %s...", s); | |
2854 | m->status_ready = false; | |
28247076 LP |
2855 | } |
2856 | ||
24dd31c1 LN |
2857 | static void manager_handle_ctrl_alt_del(Manager *m) { |
2858 | /* If the user presses C-A-D more than | |
2859 | * 7 times within 2s, we reboot/shutdown immediately, | |
2860 | * unless it was disabled in system.conf */ | |
2861 | ||
7994ac1d | 2862 | if (ratelimit_below(&m->ctrl_alt_del_ratelimit) || m->cad_burst_action == EMERGENCY_ACTION_NONE) |
5f968096 | 2863 | manager_start_special(m, SPECIAL_CTRL_ALT_DEL_TARGET, JOB_REPLACE_IRREVERSIBLY); |
ae8c7939 | 2864 | else |
7af67e9a | 2865 | emergency_action(m, m->cad_burst_action, EMERGENCY_ACTION_WARN, NULL, -1, |
ae8c7939 | 2866 | "Ctrl-Alt-Del was pressed more than 7 times within 2s"); |
24dd31c1 LN |
2867 | } |
2868 | ||
718db961 | 2869 | static int manager_dispatch_signal_fd(sd_event_source *source, int fd, uint32_t revents, void *userdata) { |
99534007 | 2870 | Manager *m = ASSERT_PTR(userdata); |
9152c765 LP |
2871 | ssize_t n; |
2872 | struct signalfd_siginfo sfsi; | |
dacd6cee | 2873 | int r; |
9152c765 | 2874 | |
718db961 LP |
2875 | assert(m->signal_fd == fd); |
2876 | ||
2877 | if (revents != EPOLLIN) { | |
2878 | log_warning("Got unexpected events from signal file descriptor."); | |
2879 | return 0; | |
2880 | } | |
9152c765 | 2881 | |
575b300b | 2882 | n = read(m->signal_fd, &sfsi, sizeof(sfsi)); |
8add30a0 YW |
2883 | if (n < 0) { |
2884 | if (ERRNO_IS_TRANSIENT(errno)) | |
575b300b | 2885 | return 0; |
9152c765 | 2886 | |
575b300b LP |
2887 | /* We return an error here, which will kill this handler, |
2888 | * to avoid a busy loop on read error. */ | |
2889 | return log_error_errno(errno, "Reading from signal fd failed: %m"); | |
2890 | } | |
8add30a0 | 2891 | if (n != sizeof(sfsi)) { |
c0f86d66 | 2892 | log_warning("Truncated read from signal fd (%zi bytes), ignoring!", n); |
8add30a0 YW |
2893 | return 0; |
2894 | } | |
9152c765 | 2895 | |
575b300b LP |
2896 | log_received_signal(sfsi.ssi_signo == SIGCHLD || |
2897 | (sfsi.ssi_signo == SIGTERM && MANAGER_IS_USER(m)) | |
2898 | ? LOG_DEBUG : LOG_INFO, | |
2899 | &sfsi); | |
1e001f52 | 2900 | |
575b300b | 2901 | switch (sfsi.ssi_signo) { |
b9cd2ec1 | 2902 | |
575b300b LP |
2903 | case SIGCHLD: |
2904 | r = sd_event_source_set_enabled(m->sigchld_event_source, SD_EVENT_ON); | |
2905 | if (r < 0) | |
8afabc50 | 2906 | log_warning_errno(r, "Failed to enable SIGCHLD event source, ignoring: %m"); |
b9cd2ec1 | 2907 | |
575b300b | 2908 | break; |
84e9af1e | 2909 | |
575b300b LP |
2910 | case SIGTERM: |
2911 | if (MANAGER_IS_SYSTEM(m)) { | |
ba0c7754 | 2912 | /* This is for compatibility with the original sysvinit */ |
c52b19d6 LP |
2913 | if (verify_run_space_and_log("Refusing to reexecute") < 0) |
2914 | break; | |
2915 | ||
2916 | m->objective = MANAGER_REEXECUTE; | |
a1b256b0 | 2917 | break; |
575b300b | 2918 | } |
84e9af1e | 2919 | |
575b300b LP |
2920 | _fallthrough_; |
2921 | case SIGINT: | |
2922 | if (MANAGER_IS_SYSTEM(m)) | |
2923 | manager_handle_ctrl_alt_del(m); | |
2924 | else | |
5f968096 | 2925 | manager_start_special(m, SPECIAL_EXIT_TARGET, JOB_REPLACE_IRREVERSIBLY); |
575b300b | 2926 | break; |
84e9af1e | 2927 | |
575b300b | 2928 | case SIGWINCH: |
ba0c7754 | 2929 | /* This is a nop on non-init */ |
575b300b | 2930 | if (MANAGER_IS_SYSTEM(m)) |
5f968096 | 2931 | manager_start_special(m, SPECIAL_KBREQUEST_TARGET, JOB_REPLACE); |
84e9af1e | 2932 | |
575b300b | 2933 | break; |
84e9af1e | 2934 | |
575b300b | 2935 | case SIGPWR: |
ba0c7754 | 2936 | /* This is a nop on non-init */ |
575b300b | 2937 | if (MANAGER_IS_SYSTEM(m)) |
5f968096 | 2938 | manager_start_special(m, SPECIAL_SIGPWR_TARGET, JOB_REPLACE); |
6632c602 | 2939 | |
575b300b | 2940 | break; |
57ee42ce | 2941 | |
8559b3b7 | 2942 | case SIGUSR1: |
2b680534 | 2943 | if (manager_dbus_is_running(m, false)) { |
575b300b | 2944 | log_info("Trying to reconnect to bus..."); |
575b300b | 2945 | |
8559b3b7 LP |
2946 | (void) bus_init_api(m); |
2947 | ||
2948 | if (MANAGER_IS_SYSTEM(m)) | |
2949 | (void) bus_init_system(m); | |
5f968096 ZJS |
2950 | } else |
2951 | manager_start_special(m, SPECIAL_DBUS_SERVICE, JOB_REPLACE); | |
57ee42ce | 2952 | |
575b300b | 2953 | break; |
575b300b LP |
2954 | |
2955 | case SIGUSR2: { | |
2956 | _cleanup_free_ char *dump = NULL; | |
2957 | ||
d1d8786c | 2958 | r = manager_get_dump_string(m, /* patterns= */ NULL, &dump); |
575b300b LP |
2959 | if (r < 0) { |
2960 | log_warning_errno(errno, "Failed to acquire manager dump: %m"); | |
57ee42ce LP |
2961 | break; |
2962 | } | |
2963 | ||
575b300b LP |
2964 | log_dump(LOG_INFO, dump); |
2965 | break; | |
2966 | } | |
2149e37c | 2967 | |
575b300b | 2968 | case SIGHUP: |
c52b19d6 LP |
2969 | if (verify_run_space_and_log("Refusing to reload") < 0) |
2970 | break; | |
2971 | ||
2972 | m->objective = MANAGER_RELOAD; | |
575b300b LP |
2973 | break; |
2974 | ||
2975 | default: { | |
2976 | ||
2977 | /* Starting SIGRTMIN+0 */ | |
2978 | static const struct { | |
2979 | const char *target; | |
2980 | JobMode mode; | |
2981 | } target_table[] = { | |
2982 | [0] = { SPECIAL_DEFAULT_TARGET, JOB_ISOLATE }, | |
2983 | [1] = { SPECIAL_RESCUE_TARGET, JOB_ISOLATE }, | |
2984 | [2] = { SPECIAL_EMERGENCY_TARGET, JOB_ISOLATE }, | |
2985 | [3] = { SPECIAL_HALT_TARGET, JOB_REPLACE_IRREVERSIBLY }, | |
2986 | [4] = { SPECIAL_POWEROFF_TARGET, JOB_REPLACE_IRREVERSIBLY }, | |
2987 | [5] = { SPECIAL_REBOOT_TARGET, JOB_REPLACE_IRREVERSIBLY }, | |
2988 | [6] = { SPECIAL_KEXEC_TARGET, JOB_REPLACE_IRREVERSIBLY }, | |
2989 | }; | |
2990 | ||
2991 | /* Starting SIGRTMIN+13, so that target halt and system halt are 10 apart */ | |
af41e508 | 2992 | static const ManagerObjective objective_table[] = { |
575b300b LP |
2993 | [0] = MANAGER_HALT, |
2994 | [1] = MANAGER_POWEROFF, | |
2995 | [2] = MANAGER_REBOOT, | |
2996 | [3] = MANAGER_KEXEC, | |
2997 | }; | |
b2cdc666 | 2998 | |
575b300b LP |
2999 | if ((int) sfsi.ssi_signo >= SIGRTMIN+0 && |
3000 | (int) sfsi.ssi_signo < SIGRTMIN+(int) ELEMENTSOF(target_table)) { | |
3001 | int idx = (int) sfsi.ssi_signo - SIGRTMIN; | |
5f968096 | 3002 | manager_start_special(m, target_table[idx].target, target_table[idx].mode); |
1005d14f | 3003 | break; |
2149e37c | 3004 | } |
1005d14f | 3005 | |
575b300b | 3006 | if ((int) sfsi.ssi_signo >= SIGRTMIN+13 && |
af41e508 LP |
3007 | (int) sfsi.ssi_signo < SIGRTMIN+13+(int) ELEMENTSOF(objective_table)) { |
3008 | m->objective = objective_table[sfsi.ssi_signo - SIGRTMIN - 13]; | |
575b300b LP |
3009 | break; |
3010 | } | |
3011 | ||
3012 | switch (sfsi.ssi_signo - SIGRTMIN) { | |
3013 | ||
29e6b0c1 LP |
3014 | case 18: { |
3015 | bool generic = false; | |
3016 | ||
3017 | if (sfsi.ssi_code != SI_QUEUE) | |
3018 | generic = true; | |
3019 | else { | |
3020 | /* Override a few select commands by our own PID1-specific logic */ | |
3021 | ||
3022 | switch (sfsi.ssi_int) { | |
3023 | ||
3024 | case _COMMON_SIGNAL_COMMAND_LOG_LEVEL_BASE..._COMMON_SIGNAL_COMMAND_LOG_LEVEL_END: | |
3025 | manager_override_log_level(m, sfsi.ssi_int - _COMMON_SIGNAL_COMMAND_LOG_LEVEL_BASE); | |
3026 | break; | |
3027 | ||
3028 | case COMMON_SIGNAL_COMMAND_CONSOLE: | |
3029 | manager_override_log_target(m, LOG_TARGET_CONSOLE); | |
3030 | break; | |
3031 | ||
3032 | case COMMON_SIGNAL_COMMAND_JOURNAL: | |
3033 | manager_override_log_target(m, LOG_TARGET_JOURNAL); | |
3034 | break; | |
3035 | ||
3036 | case COMMON_SIGNAL_COMMAND_KMSG: | |
3037 | manager_override_log_target(m, LOG_TARGET_KMSG); | |
3038 | break; | |
3039 | ||
3040 | case COMMON_SIGNAL_COMMAND_NULL: | |
3041 | manager_override_log_target(m, LOG_TARGET_NULL); | |
3042 | break; | |
3043 | ||
3044 | default: | |
3045 | generic = true; | |
3046 | } | |
3047 | } | |
3048 | ||
3049 | if (generic) | |
3050 | return sigrtmin18_handler(source, &sfsi, NULL); | |
3051 | ||
3052 | break; | |
3053 | } | |
3054 | ||
575b300b | 3055 | case 20: |
43bba15a | 3056 | manager_override_show_status(m, SHOW_STATUS_YES, "signal"); |
a16e1123 LP |
3057 | break; |
3058 | ||
575b300b | 3059 | case 21: |
43bba15a | 3060 | manager_override_show_status(m, SHOW_STATUS_NO, "signal"); |
575b300b | 3061 | break; |
7d793605 | 3062 | |
575b300b | 3063 | case 22: |
a6ecbf83 | 3064 | manager_override_log_level(m, LOG_DEBUG); |
575b300b LP |
3065 | break; |
3066 | ||
3067 | case 23: | |
a6ecbf83 | 3068 | manager_restore_original_log_level(m); |
575b300b | 3069 | break; |
0003d1ab | 3070 | |
575b300b LP |
3071 | case 24: |
3072 | if (MANAGER_IS_USER(m)) { | |
af41e508 | 3073 | m->objective = MANAGER_EXIT; |
575b300b | 3074 | return 0; |
0658666b | 3075 | } |
9152c765 | 3076 | |
575b300b LP |
3077 | /* This is a nop on init */ |
3078 | break; | |
3079 | ||
463aef23 FB |
3080 | case 25: |
3081 | m->objective = MANAGER_REEXECUTE; | |
3082 | break; | |
3083 | ||
575b300b LP |
3084 | case 26: |
3085 | case 29: /* compatibility: used to be mapped to LOG_TARGET_SYSLOG_OR_KMSG */ | |
bda7d78b | 3086 | manager_restore_original_log_target(m); |
575b300b LP |
3087 | break; |
3088 | ||
3089 | case 27: | |
bda7d78b | 3090 | manager_override_log_target(m, LOG_TARGET_CONSOLE); |
575b300b LP |
3091 | break; |
3092 | ||
3093 | case 28: | |
bda7d78b | 3094 | manager_override_log_target(m, LOG_TARGET_KMSG); |
575b300b LP |
3095 | break; |
3096 | ||
3097 | default: | |
3098 | log_warning("Got unhandled signal <%s>.", signal_to_string(sfsi.ssi_signo)); | |
3099 | } | |
3100 | }} | |
034c6ed7 LP |
3101 | |
3102 | return 0; | |
3103 | } | |
3104 | ||
718db961 | 3105 | static int manager_dispatch_time_change_fd(sd_event_source *source, int fd, uint32_t revents, void *userdata) { |
99534007 | 3106 | Manager *m = ASSERT_PTR(userdata); |
718db961 | 3107 | Unit *u; |
034c6ed7 | 3108 | |
a80c1575 | 3109 | log_struct(LOG_DEBUG, |
2b044526 | 3110 | "MESSAGE_ID=" SD_MESSAGE_TIME_CHANGE_STR, |
a1230ff9 | 3111 | LOG_MESSAGE("Time has been changed")); |
034c6ed7 | 3112 | |
718db961 | 3113 | /* Restart the watch */ |
7feedd18 | 3114 | (void) manager_setup_time_change(m); |
4e434314 | 3115 | |
90e74a66 | 3116 | HASHMAP_FOREACH(u, m->units) |
718db961 LP |
3117 | if (UNIT_VTABLE(u)->time_change) |
3118 | UNIT_VTABLE(u)->time_change(u); | |
ea430986 | 3119 | |
718db961 LP |
3120 | return 0; |
3121 | } | |
ea430986 | 3122 | |
bbf5fd8e LP |
3123 | static int manager_dispatch_timezone_change( |
3124 | sd_event_source *source, | |
3125 | const struct inotify_event *e, | |
3126 | void *userdata) { | |
3127 | ||
99534007 | 3128 | Manager *m = ASSERT_PTR(userdata); |
bbf5fd8e | 3129 | int changed; |
bbf5fd8e LP |
3130 | Unit *u; |
3131 | ||
bbf5fd8e LP |
3132 | log_debug("inotify event for /etc/localtime"); |
3133 | ||
3134 | changed = manager_read_timezone_stat(m); | |
f20db199 | 3135 | if (changed <= 0) |
bbf5fd8e | 3136 | return changed; |
bbf5fd8e LP |
3137 | |
3138 | /* Something changed, restart the watch, to ensure we watch the new /etc/localtime if it changed */ | |
3139 | (void) manager_setup_timezone_change(m); | |
3140 | ||
3141 | /* Read the new timezone */ | |
3142 | tzset(); | |
3143 | ||
3144 | log_debug("Timezone has been changed (now: %s).", tzname[daylight]); | |
3145 | ||
90e74a66 | 3146 | HASHMAP_FOREACH(u, m->units) |
bbf5fd8e LP |
3147 | if (UNIT_VTABLE(u)->timezone_change) |
3148 | UNIT_VTABLE(u)->timezone_change(u); | |
3149 | ||
3150 | return 0; | |
3151 | } | |
3152 | ||
718db961 | 3153 | static int manager_dispatch_idle_pipe_fd(sd_event_source *source, int fd, uint32_t revents, void *userdata) { |
99534007 | 3154 | Manager *m = ASSERT_PTR(userdata); |
8742514c | 3155 | |
718db961 | 3156 | assert(m->idle_pipe[2] == fd); |
8742514c | 3157 | |
8f41e6b6 ZJS |
3158 | /* There's at least one Type=idle child that just gave up on us waiting for the boot process to |
3159 | * complete. Let's now turn off any further console output if there's at least one service that needs | |
3160 | * console access, so that from now on our own output should not spill into that service's output | |
3161 | * anymore. After all, we support Type=idle only to beautify console output and it generally is set | |
3162 | * on services that want to own the console exclusively without our interference. */ | |
718db961 | 3163 | m->no_console_output = m->n_on_console > 0; |
03b717a3 | 3164 | |
30fd9a2d | 3165 | /* Acknowledge the child's request, and let all other children know too that they shouldn't wait |
8f41e6b6 | 3166 | * any longer by closing the pipes towards them, which is what they are waiting for. */ |
718db961 | 3167 | manager_close_idle_pipe(m); |
03b717a3 | 3168 | |
718db961 LP |
3169 | return 0; |
3170 | } | |
31a7eb86 | 3171 | |
718db961 | 3172 | static int manager_dispatch_jobs_in_progress(sd_event_source *source, usec_t usec, void *userdata) { |
99534007 | 3173 | Manager *m = ASSERT_PTR(userdata); |
fd08a840 | 3174 | int r; |
31a7eb86 | 3175 | |
fd08a840 | 3176 | assert(source); |
9152c765 | 3177 | |
718db961 | 3178 | manager_print_jobs_in_progress(m); |
fd08a840 | 3179 | |
39cf0351 | 3180 | r = sd_event_source_set_time_relative(source, JOBS_IN_PROGRESS_PERIOD_USEC); |
fd08a840 ZJS |
3181 | if (r < 0) |
3182 | return r; | |
3183 | ||
3184 | return sd_event_source_set_enabled(source, SD_EVENT_ONESHOT); | |
9152c765 LP |
3185 | } |
3186 | ||
3187 | int manager_loop(Manager *m) { | |
8c227e7f | 3188 | RateLimit rl = { .interval = 1*USEC_PER_SEC, .burst = 50000 }; |
9152c765 | 3189 | int r; |
9152c765 LP |
3190 | |
3191 | assert(m); | |
1fb70e66 | 3192 | assert(m->objective == MANAGER_OK); /* Ensure manager_startup() has been called */ |
9152c765 | 3193 | |
b0c918b9 LP |
3194 | manager_check_finished(m); |
3195 | ||
575b300b LP |
3196 | /* There might still be some zombies hanging around from before we were exec()'ed. Let's reap them. */ |
3197 | r = sd_event_source_set_enabled(m->sigchld_event_source, SD_EVENT_ON); | |
e96d6be7 | 3198 | if (r < 0) |
575b300b | 3199 | return log_error_errno(r, "Failed to enable SIGCHLD event source: %m"); |
a4312405 | 3200 | |
af41e508 | 3201 | while (m->objective == MANAGER_OK) { |
9152c765 | 3202 | |
ae4a0ec4 | 3203 | (void) watchdog_ping(); |
e96d6be7 | 3204 | |
7994ac1d | 3205 | if (!ratelimit_below(&rl)) { |
ea430986 LP |
3206 | /* Yay, something is going seriously wrong, pause a little */ |
3207 | log_warning("Looping too fast. Throttling execution a little."); | |
3208 | sleep(1); | |
3209 | } | |
3210 | ||
37a8e683 | 3211 | if (manager_dispatch_load_queue(m) > 0) |
23a177ef LP |
3212 | continue; |
3213 | ||
c5a97ed1 LP |
3214 | if (manager_dispatch_gc_job_queue(m) > 0) |
3215 | continue; | |
3216 | ||
3217 | if (manager_dispatch_gc_unit_queue(m) > 0) | |
701cc384 LP |
3218 | continue; |
3219 | ||
cf1265e1 | 3220 | if (manager_dispatch_cleanup_queue(m) > 0) |
c1e1601e | 3221 | continue; |
034c6ed7 | 3222 | |
91a6073e | 3223 | if (manager_dispatch_cgroup_realize_queue(m) > 0) |
c1e1601e LP |
3224 | continue; |
3225 | ||
0bc488c9 LP |
3226 | if (manager_dispatch_start_when_upheld_queue(m) > 0) |
3227 | continue; | |
3228 | ||
56c59592 LP |
3229 | if (manager_dispatch_stop_when_bound_queue(m) > 0) |
3230 | continue; | |
3231 | ||
a3c1168a LP |
3232 | if (manager_dispatch_stop_when_unneeded_queue(m) > 0) |
3233 | continue; | |
3234 | ||
6ac62d61 LP |
3235 | if (manager_dispatch_release_resources_queue(m) > 0) |
3236 | continue; | |
3237 | ||
c1e1601e | 3238 | if (manager_dispatch_dbus_queue(m) > 0) |
ea430986 | 3239 | continue; |
ea430986 | 3240 | |
c5f8a179 | 3241 | /* Sleep for watchdog runtime wait time */ |
12663295 | 3242 | r = sd_event_run(m->event, watchdog_runtime_wait()); |
23bbb0de MS |
3243 | if (r < 0) |
3244 | return log_error_errno(r, "Failed to run event loop: %m"); | |
a16e1123 | 3245 | } |
957ca890 | 3246 | |
af41e508 | 3247 | return m->objective; |
83c60c9f | 3248 | } |
ea430986 | 3249 | |
718db961 | 3250 | int manager_load_unit_from_dbus_path(Manager *m, const char *s, sd_bus_error *e, Unit **_u) { |
ede3a796 | 3251 | _cleanup_free_ char *n = NULL; |
4b58153d | 3252 | sd_id128_t invocation_id; |
ea430986 | 3253 | Unit *u; |
80fbf05e | 3254 | int r; |
ea430986 LP |
3255 | |
3256 | assert(m); | |
3257 | assert(s); | |
3258 | assert(_u); | |
3259 | ||
ede3a796 LP |
3260 | r = unit_name_from_dbus_path(s, &n); |
3261 | if (r < 0) | |
3262 | return r; | |
ea430986 | 3263 | |
8f41e6b6 ZJS |
3264 | /* Permit addressing units by invocation ID: if the passed bus path is suffixed by a 128bit ID then |
3265 | * we use it as invocation ID. */ | |
4b58153d LP |
3266 | r = sd_id128_from_string(n, &invocation_id); |
3267 | if (r >= 0) { | |
3268 | u = hashmap_get(m->units_by_invocation_id, &invocation_id); | |
3269 | if (u) { | |
3270 | *_u = u; | |
3271 | return 0; | |
3272 | } | |
3273 | ||
930c124c ZJS |
3274 | return sd_bus_error_setf(e, BUS_ERROR_NO_UNIT_FOR_INVOCATION_ID, |
3275 | "No unit with the specified invocation ID " SD_ID128_FORMAT_STR " known.", | |
3276 | SD_ID128_FORMAT_VAL(invocation_id)); | |
4b58153d LP |
3277 | } |
3278 | ||
00c83b43 | 3279 | /* If this didn't work, we check if this is a unit name */ |
930c124c ZJS |
3280 | if (!unit_name_is_valid(n, UNIT_NAME_PLAIN|UNIT_NAME_INSTANCE)) { |
3281 | _cleanup_free_ char *nn = NULL; | |
3282 | ||
3283 | nn = cescape(n); | |
3284 | return sd_bus_error_setf(e, SD_BUS_ERROR_INVALID_ARGS, | |
3285 | "Unit name %s is neither a valid invocation ID nor unit name.", strnull(nn)); | |
3286 | } | |
00c83b43 | 3287 | |
80fbf05e | 3288 | r = manager_load_unit(m, n, NULL, e, &u); |
80fbf05e MS |
3289 | if (r < 0) |
3290 | return r; | |
ea430986 LP |
3291 | |
3292 | *_u = u; | |
ea430986 LP |
3293 | return 0; |
3294 | } | |
86fbf370 LP |
3295 | |
3296 | int manager_get_job_from_dbus_path(Manager *m, const char *s, Job **_j) { | |
718db961 | 3297 | const char *p; |
86fbf370 | 3298 | unsigned id; |
718db961 | 3299 | Job *j; |
86fbf370 LP |
3300 | int r; |
3301 | ||
3302 | assert(m); | |
3303 | assert(s); | |
3304 | assert(_j); | |
3305 | ||
718db961 LP |
3306 | p = startswith(s, "/org/freedesktop/systemd1/job/"); |
3307 | if (!p) | |
86fbf370 LP |
3308 | return -EINVAL; |
3309 | ||
718db961 | 3310 | r = safe_atou(p, &id); |
8742514c | 3311 | if (r < 0) |
86fbf370 LP |
3312 | return r; |
3313 | ||
8742514c LP |
3314 | j = manager_get_job(m, id); |
3315 | if (!j) | |
86fbf370 LP |
3316 | return -ENOENT; |
3317 | ||
3318 | *_j = j; | |
3319 | ||
3320 | return 0; | |
3321 | } | |
dfcd764e | 3322 | |
4927fcae | 3323 | void manager_send_unit_audit(Manager *m, Unit *u, int type, bool success) { |
e537352b | 3324 | |
349cc4a5 | 3325 | #if HAVE_AUDIT |
2ba11090 | 3326 | _cleanup_free_ char *p = NULL; |
0aa281df | 3327 | const char *msg; |
7410616c | 3328 | int audit_fd, r; |
e537352b | 3329 | |
463d0d15 | 3330 | if (!MANAGER_IS_SYSTEM(m)) |
a1a078ee LP |
3331 | return; |
3332 | ||
c1165f82 LP |
3333 | audit_fd = get_audit_fd(); |
3334 | if (audit_fd < 0) | |
e537352b LP |
3335 | return; |
3336 | ||
bbd3a7ba LP |
3337 | /* Don't generate audit events if the service was already |
3338 | * started and we're just deserializing */ | |
2c289ea8 | 3339 | if (MANAGER_IS_RELOADING(m)) |
bbd3a7ba LP |
3340 | return; |
3341 | ||
ac155bb8 | 3342 | if (u->type != UNIT_SERVICE) |
f1dd0c3f LP |
3343 | return; |
3344 | ||
7410616c LP |
3345 | r = unit_name_to_prefix_and_instance(u->id, &p); |
3346 | if (r < 0) { | |
3347 | log_error_errno(r, "Failed to extract prefix and instance of unit name: %m"); | |
e537352b LP |
3348 | return; |
3349 | } | |
3350 | ||
63c372cb | 3351 | msg = strjoina("unit=", p); |
0aa281df LP |
3352 | if (audit_log_user_comm_message(audit_fd, type, msg, "systemd", NULL, NULL, NULL, success) < 0) { |
3353 | if (errno == EPERM) | |
391ade86 | 3354 | /* We aren't allowed to send audit messages? |
44785992 | 3355 | * Then let's not retry again. */ |
c1165f82 | 3356 | close_audit_fd(); |
0aa281df | 3357 | else |
56f64d95 | 3358 | log_warning_errno(errno, "Failed to send audit message: %m"); |
391ade86 | 3359 | } |
4927fcae | 3360 | #endif |
e537352b | 3361 | |
e537352b LP |
3362 | } |
3363 | ||
e983b760 | 3364 | void manager_send_unit_plymouth(Manager *m, Unit *u) { |
fc2fffe7 | 3365 | static const union sockaddr_union sa = PLYMOUTH_SOCKET; |
2ba11090 | 3366 | _cleanup_free_ char *message = NULL; |
254d1313 | 3367 | _cleanup_close_ int fd = -EBADF; |
fc2fffe7 | 3368 | int n = 0; |
e983b760 LP |
3369 | |
3370 | /* Don't generate plymouth events if the service was already | |
3371 | * started and we're just deserializing */ | |
2c289ea8 | 3372 | if (MANAGER_IS_RELOADING(m)) |
e983b760 LP |
3373 | return; |
3374 | ||
463d0d15 | 3375 | if (!MANAGER_IS_SYSTEM(m)) |
e983b760 LP |
3376 | return; |
3377 | ||
75f86906 | 3378 | if (detect_container() > 0) |
3772995a LP |
3379 | return; |
3380 | ||
ec2ce0c5 | 3381 | if (!IN_SET(u->type, UNIT_SERVICE, UNIT_MOUNT, UNIT_SWAP)) |
e983b760 LP |
3382 | return; |
3383 | ||
3384 | /* We set SOCK_NONBLOCK here so that we rather drop the | |
3385 | * message then wait for plymouth */ | |
e62d8c39 ZJS |
3386 | fd = socket(AF_UNIX, SOCK_STREAM|SOCK_CLOEXEC|SOCK_NONBLOCK, 0); |
3387 | if (fd < 0) { | |
56f64d95 | 3388 | log_error_errno(errno, "socket() failed: %m"); |
e983b760 LP |
3389 | return; |
3390 | } | |
3391 | ||
fc2fffe7 | 3392 | if (connect(fd, &sa.sa, SOCKADDR_UN_LEN(sa.un)) < 0) { |
f60a028a | 3393 | if (!IN_SET(errno, EAGAIN, ENOENT) && !ERRNO_IS_DISCONNECT(errno)) |
56f64d95 | 3394 | log_error_errno(errno, "connect() failed: %m"); |
2ba11090 | 3395 | return; |
e983b760 LP |
3396 | } |
3397 | ||
305757d8 ZJS |
3398 | if (asprintf(&message, "U\002%c%s%n", (int) (strlen(u->id) + 1), u->id, &n) < 0) |
3399 | return (void) log_oom(); | |
e983b760 LP |
3400 | |
3401 | errno = 0; | |
2ba11090 | 3402 | if (write(fd, message, n + 1) != n + 1) |
f60a028a | 3403 | if (!IN_SET(errno, EAGAIN, ENOENT) && !ERRNO_IS_DISCONNECT(errno)) |
56f64d95 | 3404 | log_error_errno(errno, "Failed to write Plymouth message: %m"); |
e983b760 LP |
3405 | } |
3406 | ||
986935cf FB |
3407 | usec_t manager_get_watchdog(Manager *m, WatchdogType t) { |
3408 | assert(m); | |
3409 | ||
3410 | if (MANAGER_IS_USER(m)) | |
3411 | return USEC_INFINITY; | |
3412 | ||
3413 | if (timestamp_is_set(m->watchdog_overridden[t])) | |
3414 | return m->watchdog_overridden[t]; | |
3415 | ||
3416 | return m->watchdog[t]; | |
3417 | } | |
3418 | ||
3419 | void manager_set_watchdog(Manager *m, WatchdogType t, usec_t timeout) { | |
986935cf FB |
3420 | |
3421 | assert(m); | |
3422 | ||
3423 | if (MANAGER_IS_USER(m)) | |
3424 | return; | |
3425 | ||
3426 | if (m->watchdog[t] == timeout) | |
3427 | return; | |
3428 | ||
5717062e | 3429 | if (t == WATCHDOG_RUNTIME) { |
f16890f8 FB |
3430 | if (!timestamp_is_set(m->watchdog_overridden[WATCHDOG_RUNTIME])) |
3431 | (void) watchdog_setup(timeout); | |
5717062e CK |
3432 | } else if (t == WATCHDOG_PRETIMEOUT) |
3433 | if (m->watchdog_overridden[WATCHDOG_PRETIMEOUT] == USEC_INFINITY) | |
3434 | (void) watchdog_setup_pretimeout(timeout); | |
986935cf | 3435 | |
61927b9f | 3436 | m->watchdog[t] = timeout; |
986935cf FB |
3437 | } |
3438 | ||
aac47032 | 3439 | void manager_override_watchdog(Manager *m, WatchdogType t, usec_t timeout) { |
986935cf FB |
3440 | |
3441 | assert(m); | |
3442 | ||
3443 | if (MANAGER_IS_USER(m)) | |
aac47032 | 3444 | return; |
986935cf FB |
3445 | |
3446 | if (m->watchdog_overridden[t] == timeout) | |
aac47032 | 3447 | return; |
986935cf FB |
3448 | |
3449 | if (t == WATCHDOG_RUNTIME) { | |
2628b98f | 3450 | usec_t usec = timestamp_is_set(timeout) ? timeout : m->watchdog[t]; |
986935cf | 3451 | |
f16890f8 | 3452 | (void) watchdog_setup(usec); |
5717062e CK |
3453 | } else if (t == WATCHDOG_PRETIMEOUT) |
3454 | (void) watchdog_setup_pretimeout(timeout); | |
986935cf | 3455 | |
61927b9f | 3456 | m->watchdog_overridden[t] = timeout; |
986935cf FB |
3457 | } |
3458 | ||
aff3a9e1 LB |
3459 | int manager_set_watchdog_pretimeout_governor(Manager *m, const char *governor) { |
3460 | _cleanup_free_ char *p = NULL; | |
3461 | int r; | |
3462 | ||
3463 | assert(m); | |
3464 | ||
3465 | if (MANAGER_IS_USER(m)) | |
3466 | return 0; | |
3467 | ||
3468 | if (streq_ptr(m->watchdog_pretimeout_governor, governor)) | |
3469 | return 0; | |
3470 | ||
3471 | p = strdup(governor); | |
3472 | if (!p) | |
3473 | return -ENOMEM; | |
3474 | ||
3475 | r = watchdog_setup_pretimeout_governor(governor); | |
3476 | if (r < 0) | |
3477 | return r; | |
3478 | ||
3479 | return free_and_replace(m->watchdog_pretimeout_governor, p); | |
3480 | } | |
3481 | ||
3482 | int manager_override_watchdog_pretimeout_governor(Manager *m, const char *governor) { | |
3483 | _cleanup_free_ char *p = NULL; | |
3484 | int r; | |
3485 | ||
3486 | assert(m); | |
3487 | ||
3488 | if (MANAGER_IS_USER(m)) | |
3489 | return 0; | |
3490 | ||
3491 | if (streq_ptr(m->watchdog_pretimeout_governor_overridden, governor)) | |
3492 | return 0; | |
3493 | ||
3494 | p = strdup(governor); | |
3495 | if (!p) | |
3496 | return -ENOMEM; | |
3497 | ||
3498 | r = watchdog_setup_pretimeout_governor(governor); | |
3499 | if (r < 0) | |
3500 | return r; | |
3501 | ||
3502 | return free_and_replace(m->watchdog_pretimeout_governor_overridden, p); | |
3503 | } | |
3504 | ||
a16e1123 | 3505 | int manager_reload(Manager *m) { |
d7ac0952 | 3506 | _unused_ _cleanup_(manager_reloading_stopp) Manager *reloading = NULL; |
51d122af | 3507 | _cleanup_fdset_free_ FDSet *fds = NULL; |
6a33af40 LP |
3508 | _cleanup_fclose_ FILE *f = NULL; |
3509 | int r; | |
a16e1123 LP |
3510 | |
3511 | assert(m); | |
3512 | ||
07719a21 LP |
3513 | r = manager_open_serialization(m, &f); |
3514 | if (r < 0) | |
6a33af40 | 3515 | return log_error_errno(r, "Failed to create serialization file: %m"); |
38c52d46 | 3516 | |
07719a21 | 3517 | fds = fdset_new(); |
6a33af40 LP |
3518 | if (!fds) |
3519 | return log_oom(); | |
3520 | ||
d147e2b6 ZJS |
3521 | /* We are officially in reload mode from here on. */ |
3522 | reloading = manager_reloading_start(m); | |
a16e1123 | 3523 | |
b3680f49 | 3524 | r = manager_serialize(m, f, fds, false); |
d147e2b6 | 3525 | if (r < 0) |
d68c645b | 3526 | return r; |
a16e1123 | 3527 | |
d147e2b6 ZJS |
3528 | if (fseeko(f, 0, SEEK_SET) < 0) |
3529 | return log_error_errno(errno, "Failed to seek to beginning of serialization: %m"); | |
a16e1123 | 3530 | |
6a33af40 | 3531 | /* 💀 This is the point of no return, from here on there is no way back. 💀 */ |
d147e2b6 | 3532 | reloading = NULL; |
6a33af40 LP |
3533 | |
3534 | bus_manager_send_reloading(m, true); | |
3535 | ||
3536 | /* Start by flushing out all jobs and units, all generated units, all runtime environments, all dynamic users | |
3537 | * and everything else that is worth flushing out. We'll get it all back from the serialization — if we need | |
7802194a | 3538 | * it. */ |
6a33af40 | 3539 | |
a16e1123 | 3540 | manager_clear_jobs_and_units(m); |
07a78643 | 3541 | lookup_paths_flush_generator(&m->lookup_paths); |
84e3543e | 3542 | lookup_paths_free(&m->lookup_paths); |
e76506b7 | 3543 | exec_shared_runtime_vacuum(m); |
29206d46 | 3544 | dynamic_user_vacuum(m, false); |
00d9ef85 LP |
3545 | m->uid_refs = hashmap_free(m->uid_refs); |
3546 | m->gid_refs = hashmap_free(m->gid_refs); | |
2ded0c04 | 3547 | |
4870133b | 3548 | r = lookup_paths_init_or_warn(&m->lookup_paths, m->runtime_scope, 0, NULL); |
6a33af40 | 3549 | if (r < 0) |
99aad9a2 | 3550 | return r; |
5a1e9937 | 3551 | |
6a33af40 LP |
3552 | (void) manager_run_environment_generators(m); |
3553 | (void) manager_run_generators(m); | |
64691d20 | 3554 | |
581fef8d | 3555 | lookup_paths_log(&m->lookup_paths); |
07719a21 | 3556 | |
91e0ee5f | 3557 | /* We flushed out generated files, for which we don't watch mtime, so we should flush the old map. */ |
e8630e69 | 3558 | manager_free_unit_name_maps(m); |
5a1e9937 | 3559 | |
6a33af40 | 3560 | /* First, enumerate what we can from kernel and suchlike */ |
3ad2afb6 | 3561 | manager_enumerate_perpetual(m); |
ba64af90 | 3562 | manager_enumerate(m); |
a16e1123 LP |
3563 | |
3564 | /* Second, deserialize our stored data */ | |
6a33af40 LP |
3565 | r = manager_deserialize(m, f, fds); |
3566 | if (r < 0) | |
3567 | log_warning_errno(r, "Deserialization failed, proceeding anyway: %m"); | |
a16e1123 | 3568 | |
6a33af40 | 3569 | /* We don't need the serialization anymore */ |
62b0cbb3 | 3570 | f = safe_fclose(f); |
a16e1123 | 3571 | |
6a33af40 LP |
3572 | /* Re-register notify_fd as event source, and set up other sockets/communication channels we might need */ |
3573 | (void) manager_setup_notify(m); | |
3574 | (void) manager_setup_cgroups_agent(m); | |
3575 | (void) manager_setup_user_lookup_fd(m); | |
00d9ef85 | 3576 | |
a16e1123 | 3577 | /* Third, fire things up! */ |
007c6337 | 3578 | manager_coldplug(m); |
a16e1123 | 3579 | |
5197be06 LP |
3580 | /* Clean up runtime objects no longer referenced */ |
3581 | manager_vacuum(m); | |
e8a565cb | 3582 | |
3deed59a AA |
3583 | /* Clean up deserialized tracked clients */ |
3584 | m->deserialized_subscribed = strv_free(m->deserialized_subscribed); | |
3585 | ||
6a33af40 | 3586 | /* Consider the reload process complete now. */ |
31dc1ca3 LP |
3587 | assert(m->n_reloading > 0); |
3588 | m->n_reloading--; | |
3589 | ||
5ce5e1ad | 3590 | manager_ready(m); |
8936a5e3 | 3591 | |
71445ae7 | 3592 | m->send_reloading_done = true; |
6a33af40 | 3593 | return 0; |
a16e1123 LP |
3594 | } |
3595 | ||
fdf20a31 | 3596 | void manager_reset_failed(Manager *m) { |
5632e374 | 3597 | Unit *u; |
5632e374 LP |
3598 | |
3599 | assert(m); | |
3600 | ||
90e74a66 | 3601 | HASHMAP_FOREACH(u, m->units) |
fdf20a31 | 3602 | unit_reset_failed(u); |
5632e374 LP |
3603 | } |
3604 | ||
31afa0a4 | 3605 | bool manager_unit_inactive_or_pending(Manager *m, const char *name) { |
8f6df3fa LP |
3606 | Unit *u; |
3607 | ||
3608 | assert(m); | |
3609 | assert(name); | |
3610 | ||
3611 | /* Returns true if the unit is inactive or going down */ | |
bd0af849 ZJS |
3612 | u = manager_get_unit(m, name); |
3613 | if (!u) | |
8f6df3fa LP |
3614 | return true; |
3615 | ||
31afa0a4 | 3616 | return unit_inactive_or_pending(u); |
8f6df3fa LP |
3617 | } |
3618 | ||
d8eb10d6 ZJS |
3619 | static void log_taint_string(Manager *m) { |
3620 | _cleanup_free_ char *taint = NULL; | |
3621 | ||
3622 | assert(m); | |
3623 | ||
3624 | if (MANAGER_IS_USER(m) || m->taint_logged) | |
3625 | return; | |
3626 | ||
3627 | m->taint_logged = true; /* only check for taint once */ | |
3628 | ||
3629 | taint = manager_taint_string(m); | |
3630 | if (isempty(taint)) | |
3631 | return; | |
3632 | ||
3633 | log_struct(LOG_NOTICE, | |
3634 | LOG_MESSAGE("System is tainted: %s", taint), | |
3635 | "TAINT=%s", taint, | |
a1230ff9 | 3636 | "MESSAGE_ID=" SD_MESSAGE_TAINTED_STR); |
d8eb10d6 ZJS |
3637 | } |
3638 | ||
56dacdbc | 3639 | static void manager_notify_finished(Manager *m) { |
915b3753 | 3640 | usec_t firmware_usec, loader_usec, kernel_usec, initrd_usec, userspace_usec, total_usec; |
b0c918b9 | 3641 | |
638cece4 | 3642 | if (MANAGER_IS_TEST_RUN(m)) |
b0c918b9 LP |
3643 | return; |
3644 | ||
463d0d15 | 3645 | if (MANAGER_IS_SYSTEM(m) && detect_container() <= 0) { |
dc3c9f5e ZJS |
3646 | char buf[FORMAT_TIMESPAN_MAX + STRLEN(" (firmware) + ") + FORMAT_TIMESPAN_MAX + STRLEN(" (loader) + ")] |
3647 | = {}; | |
3648 | char *p = buf; | |
3649 | size_t size = sizeof buf; | |
e03ae661 | 3650 | |
9f9f0342 LP |
3651 | /* Note that MANAGER_TIMESTAMP_KERNEL's monotonic value is always at 0, and |
3652 | * MANAGER_TIMESTAMP_FIRMWARE's and MANAGER_TIMESTAMP_LOADER's monotonic value should be considered | |
915b3753 LP |
3653 | * negative values. */ |
3654 | ||
9f9f0342 LP |
3655 | firmware_usec = m->timestamps[MANAGER_TIMESTAMP_FIRMWARE].monotonic - m->timestamps[MANAGER_TIMESTAMP_LOADER].monotonic; |
3656 | loader_usec = m->timestamps[MANAGER_TIMESTAMP_LOADER].monotonic - m->timestamps[MANAGER_TIMESTAMP_KERNEL].monotonic; | |
3657 | userspace_usec = m->timestamps[MANAGER_TIMESTAMP_FINISH].monotonic - m->timestamps[MANAGER_TIMESTAMP_USERSPACE].monotonic; | |
3658 | total_usec = m->timestamps[MANAGER_TIMESTAMP_FIRMWARE].monotonic + m->timestamps[MANAGER_TIMESTAMP_FINISH].monotonic; | |
18fa6b27 | 3659 | |
dd1db3c2 | 3660 | if (firmware_usec > 0) |
5291f26d | 3661 | size = strpcpyf(&p, size, "%s (firmware) + ", FORMAT_TIMESPAN(firmware_usec, USEC_PER_MSEC)); |
dd1db3c2 | 3662 | if (loader_usec > 0) |
5291f26d | 3663 | size = strpcpyf(&p, size, "%s (loader) + ", FORMAT_TIMESPAN(loader_usec, USEC_PER_MSEC)); |
dd1db3c2 | 3664 | |
9f9f0342 | 3665 | if (dual_timestamp_is_set(&m->timestamps[MANAGER_TIMESTAMP_INITRD])) { |
18fa6b27 | 3666 | |
7802194a | 3667 | /* The initrd case on bare-metal */ |
9f9f0342 LP |
3668 | kernel_usec = m->timestamps[MANAGER_TIMESTAMP_INITRD].monotonic - m->timestamps[MANAGER_TIMESTAMP_KERNEL].monotonic; |
3669 | initrd_usec = m->timestamps[MANAGER_TIMESTAMP_USERSPACE].monotonic - m->timestamps[MANAGER_TIMESTAMP_INITRD].monotonic; | |
18fa6b27 | 3670 | |
e12919e8 | 3671 | log_struct(LOG_INFO, |
2b044526 | 3672 | "MESSAGE_ID=" SD_MESSAGE_STARTUP_FINISHED_STR, |
e12919e8 LP |
3673 | "KERNEL_USEC="USEC_FMT, kernel_usec, |
3674 | "INITRD_USEC="USEC_FMT, initrd_usec, | |
3675 | "USERSPACE_USEC="USEC_FMT, userspace_usec, | |
dd1db3c2 YW |
3676 | LOG_MESSAGE("Startup finished in %s%s (kernel) + %s (initrd) + %s (userspace) = %s.", |
3677 | buf, | |
5291f26d ZJS |
3678 | FORMAT_TIMESPAN(kernel_usec, USEC_PER_MSEC), |
3679 | FORMAT_TIMESPAN(initrd_usec, USEC_PER_MSEC), | |
3680 | FORMAT_TIMESPAN(userspace_usec, USEC_PER_MSEC), | |
3681 | FORMAT_TIMESPAN(total_usec, USEC_PER_MSEC))); | |
18fa6b27 | 3682 | } else { |
7802194a | 3683 | /* The initrd-less case on bare-metal */ |
9f9f0342 LP |
3684 | |
3685 | kernel_usec = m->timestamps[MANAGER_TIMESTAMP_USERSPACE].monotonic - m->timestamps[MANAGER_TIMESTAMP_KERNEL].monotonic; | |
18fa6b27 LP |
3686 | initrd_usec = 0; |
3687 | ||
81270860 | 3688 | log_struct(LOG_INFO, |
2b044526 | 3689 | "MESSAGE_ID=" SD_MESSAGE_STARTUP_FINISHED_STR, |
e12919e8 | 3690 | "KERNEL_USEC="USEC_FMT, kernel_usec, |
ccd06097 | 3691 | "USERSPACE_USEC="USEC_FMT, userspace_usec, |
dd1db3c2 YW |
3692 | LOG_MESSAGE("Startup finished in %s%s (kernel) + %s (userspace) = %s.", |
3693 | buf, | |
5291f26d ZJS |
3694 | FORMAT_TIMESPAN(kernel_usec, USEC_PER_MSEC), |
3695 | FORMAT_TIMESPAN(userspace_usec, USEC_PER_MSEC), | |
3696 | FORMAT_TIMESPAN(total_usec, USEC_PER_MSEC))); | |
e12919e8 LP |
3697 | } |
3698 | } else { | |
4adf314b | 3699 | /* The container and --user case */ |
e12919e8 | 3700 | firmware_usec = loader_usec = initrd_usec = kernel_usec = 0; |
9f9f0342 | 3701 | total_usec = userspace_usec = m->timestamps[MANAGER_TIMESTAMP_FINISH].monotonic - m->timestamps[MANAGER_TIMESTAMP_USERSPACE].monotonic; |
e12919e8 LP |
3702 | |
3703 | log_struct(LOG_INFO, | |
2b044526 | 3704 | "MESSAGE_ID=" SD_MESSAGE_USER_STARTUP_FINISHED_STR, |
e12919e8 | 3705 | "USERSPACE_USEC="USEC_FMT, userspace_usec, |
e2cc6eca | 3706 | LOG_MESSAGE("Startup finished in %s.", |
5291f26d | 3707 | FORMAT_TIMESPAN(total_usec, USEC_PER_MSEC))); |
18fa6b27 | 3708 | } |
b0c918b9 | 3709 | |
718db961 | 3710 | bus_manager_send_finished(m, firmware_usec, loader_usec, kernel_usec, initrd_usec, userspace_usec, total_usec); |
530345e7 | 3711 | |
d8eb10d6 | 3712 | log_taint_string(m); |
b0c918b9 LP |
3713 | } |
3714 | ||
6d932659 | 3715 | static void user_manager_send_ready(Manager *m) { |
4bf4f50f ZJS |
3716 | int r; |
3717 | ||
4adf314b LP |
3718 | assert(m); |
3719 | ||
3720 | /* We send READY=1 on reaching basic.target only when running in --user mode. */ | |
3721 | if (!MANAGER_IS_USER(m) || m->ready_sent) | |
3722 | return; | |
3723 | ||
d4341b76 ZJS |
3724 | r = sd_notify(false, |
3725 | "READY=1\n" | |
3726 | "STATUS=Reached " SPECIAL_BASIC_TARGET "."); | |
4bf4f50f ZJS |
3727 | if (r < 0) |
3728 | log_warning_errno(r, "Failed to send readiness notification, ignoring: %m"); | |
3729 | ||
6d932659 ZJS |
3730 | m->ready_sent = true; |
3731 | m->status_ready = false; | |
3732 | } | |
3733 | ||
3734 | static void manager_send_ready(Manager *m) { | |
4bf4f50f ZJS |
3735 | int r; |
3736 | ||
6d932659 ZJS |
3737 | if (m->ready_sent && m->status_ready) |
3738 | /* Skip the notification if nothing changed. */ | |
3739 | return; | |
3740 | ||
028f7d3a ZJS |
3741 | r = sd_notify(false, |
3742 | "READY=1\n" | |
3743 | "STATUS=Ready."); | |
4bf4f50f ZJS |
3744 | if (r < 0) |
3745 | log_full_errno(m->ready_sent ? LOG_DEBUG : LOG_WARNING, r, | |
3746 | "Failed to send readiness notification, ignoring: %m"); | |
3747 | ||
6d932659 | 3748 | m->ready_sent = m->status_ready = true; |
4adf314b LP |
3749 | } |
3750 | ||
3751 | static void manager_check_basic_target(Manager *m) { | |
3752 | Unit *u; | |
3753 | ||
3754 | assert(m); | |
3755 | ||
3756 | /* Small shortcut */ | |
3757 | if (m->ready_sent && m->taint_logged) | |
3758 | return; | |
3759 | ||
3760 | u = manager_get_unit(m, SPECIAL_BASIC_TARGET); | |
3761 | if (!u || !UNIT_IS_ACTIVE_OR_RELOADING(unit_active_state(u))) | |
3762 | return; | |
3763 | ||
3764 | /* For user managers, send out READY=1 as soon as we reach basic.target */ | |
6d932659 | 3765 | user_manager_send_ready(m); |
4adf314b LP |
3766 | |
3767 | /* Log the taint string as soon as we reach basic.target */ | |
3768 | log_taint_string(m); | |
3769 | } | |
3770 | ||
56dacdbc | 3771 | void manager_check_finished(Manager *m) { |
56dacdbc ZJS |
3772 | assert(m); |
3773 | ||
2c289ea8 | 3774 | if (MANAGER_IS_RELOADING(m)) |
aad1976f LP |
3775 | return; |
3776 | ||
4259d202 LP |
3777 | /* Verify that we have entered the event loop already, and not left it again. */ |
3778 | if (!MANAGER_IS_RUNNING(m)) | |
9771b62d LP |
3779 | return; |
3780 | ||
4adf314b | 3781 | manager_check_basic_target(m); |
0c2826c6 | 3782 | |
56dacdbc | 3783 | if (hashmap_size(m->jobs) > 0) { |
56dacdbc | 3784 | if (m->jobs_in_progress_event_source) |
2ae56591 | 3785 | /* Ignore any failure, this is only for feedback */ |
1b4154a8 ZJS |
3786 | (void) sd_event_source_set_time(m->jobs_in_progress_event_source, |
3787 | manager_watch_jobs_next_time(m)); | |
56dacdbc ZJS |
3788 | return; |
3789 | } | |
3790 | ||
a4ac27c1 ZJS |
3791 | /* The jobs hashmap tends to grow a lot during boot, and then it's not reused until shutdown. Let's |
3792 | kill the hashmap if it is relatively large. */ | |
3793 | if (hashmap_buckets(m->jobs) > hashmap_size(m->units) / 10) | |
3794 | m->jobs = hashmap_free(m->jobs); | |
3795 | ||
6d932659 ZJS |
3796 | manager_send_ready(m); |
3797 | ||
9c1b17c3 YW |
3798 | /* Notify Type=idle units that we are done now */ |
3799 | manager_close_idle_pipe(m); | |
3800 | ||
6d932659 ZJS |
3801 | if (MANAGER_IS_FINISHED(m)) |
3802 | return; | |
3803 | ||
7365a296 | 3804 | manager_flip_auto_status(m, false, "boot finished"); |
56dacdbc | 3805 | |
56dacdbc | 3806 | /* Turn off confirm spawn now */ |
7d5ceb64 | 3807 | m->confirm_spawn = NULL; |
56dacdbc ZJS |
3808 | |
3809 | /* No need to update ask password status when we're going non-interactive */ | |
3810 | manager_close_ask_password(m); | |
3811 | ||
3812 | /* This is no longer the first boot */ | |
3813 | manager_set_first_boot(m, false); | |
3814 | ||
9f9f0342 | 3815 | dual_timestamp_get(m->timestamps + MANAGER_TIMESTAMP_FINISH); |
56dacdbc ZJS |
3816 | |
3817 | manager_notify_finished(m); | |
3818 | ||
e7ab4d1a | 3819 | manager_invalidate_startup_units(m); |
56dacdbc ZJS |
3820 | } |
3821 | ||
dd0ab174 LP |
3822 | void manager_send_reloading(Manager *m) { |
3823 | assert(m); | |
3824 | ||
3825 | /* Let whoever invoked us know that we are now reloading */ | |
3826 | (void) sd_notifyf(/* unset= */ false, | |
3827 | "RELOADING=1\n" | |
3828 | "MONOTONIC_USEC=" USEC_FMT "\n", now(CLOCK_MONOTONIC)); | |
3829 | ||
3830 | /* And ensure that we'll send READY=1 again as soon as we are ready again */ | |
3831 | m->ready_sent = false; | |
3832 | } | |
3833 | ||
64691d20 | 3834 | static bool generator_path_any(const char* const* paths) { |
64691d20 ZJS |
3835 | bool found = false; |
3836 | ||
3837 | /* Optimize by skipping the whole process by not creating output directories | |
3838 | * if no generators are found. */ | |
2034c8b8 | 3839 | STRV_FOREACH(path, paths) |
64691d20 ZJS |
3840 | if (access(*path, F_OK) == 0) |
3841 | found = true; | |
3842 | else if (errno != ENOENT) | |
3843 | log_warning_errno(errno, "Failed to open generator directory %s: %m", *path); | |
3844 | ||
3845 | return found; | |
3846 | } | |
3847 | ||
64691d20 ZJS |
3848 | static int manager_run_environment_generators(Manager *m) { |
3849 | char **tmp = NULL; /* this is only used in the forked process, no cleanup here */ | |
cccf5703 | 3850 | _cleanup_strv_free_ char **paths = NULL; |
1ad6e8b3 LP |
3851 | void* args[] = { |
3852 | [STDOUT_GENERATE] = &tmp, | |
3853 | [STDOUT_COLLECT] = &tmp, | |
3854 | [STDOUT_CONSUME] = &m->transient_environment, | |
3855 | }; | |
e3b8d063 | 3856 | int r; |
64691d20 | 3857 | |
638cece4 | 3858 | if (MANAGER_IS_TEST_RUN(m) && !(m->test_run_flags & MANAGER_TEST_RUN_ENV_GENERATORS)) |
e0a3da1f ZJS |
3859 | return 0; |
3860 | ||
361cacf4 | 3861 | paths = env_generator_binary_paths(m->runtime_scope); |
cccf5703 BB |
3862 | if (!paths) |
3863 | return log_oom(); | |
64691d20 | 3864 | |
cccf5703 | 3865 | if (!generator_path_any((const char* const*) paths)) |
64691d20 ZJS |
3866 | return 0; |
3867 | ||
2053593f | 3868 | WITH_UMASK(0022) |
cccf5703 | 3869 | r = execute_directories((const char* const*) paths, DEFAULT_TIMEOUT_USEC, gather_environment, |
43f565c6 YW |
3870 | args, NULL, m->transient_environment, |
3871 | EXEC_DIR_PARALLEL | EXEC_DIR_IGNORE_ERRORS | EXEC_DIR_SET_SYSTEMD_EXEC_PID); | |
e3b8d063 | 3872 | return r; |
64691d20 ZJS |
3873 | } |
3874 | ||
82c5db16 LP |
3875 | static int build_generator_environment(Manager *m, char ***ret) { |
3876 | _cleanup_strv_free_ char **nl = NULL; | |
3877 | Virtualization v; | |
3878 | int r; | |
3879 | ||
3880 | assert(m); | |
3881 | assert(ret); | |
3882 | ||
3883 | /* Generators oftentimes want to know some basic facts about the environment they run in, in order to | |
3884 | * adjust generated units to that. Let's pass down some bits of information that are easy for us to | |
3885 | * determine (but a bit harder for generator scripts to determine), as environment variables. */ | |
3886 | ||
3887 | nl = strv_copy(m->transient_environment); | |
3888 | if (!nl) | |
3889 | return -ENOMEM; | |
3890 | ||
4870133b | 3891 | r = strv_env_assign(&nl, "SYSTEMD_SCOPE", runtime_scope_to_string(m->runtime_scope)); |
82c5db16 LP |
3892 | if (r < 0) |
3893 | return r; | |
3894 | ||
3895 | if (MANAGER_IS_SYSTEM(m)) { | |
3896 | /* Note that $SYSTEMD_IN_INITRD may be used to override the initrd detection in much of our | |
3897 | * codebase. This is hence more than purely informational. It will shortcut detection of the | |
3898 | * initrd state if generators invoke our own tools. But that's OK, as it would come to the | |
3899 | * same results (hopefully). */ | |
3900 | r = strv_env_assign(&nl, "SYSTEMD_IN_INITRD", one_zero(in_initrd())); | |
3901 | if (r < 0) | |
3902 | return r; | |
3903 | ||
3904 | if (m->first_boot >= 0) { | |
3905 | r = strv_env_assign(&nl, "SYSTEMD_FIRST_BOOT", one_zero(m->first_boot)); | |
3906 | if (r < 0) | |
3907 | return r; | |
3908 | } | |
3909 | } | |
3910 | ||
3911 | v = detect_virtualization(); | |
3912 | if (v < 0) | |
3913 | log_debug_errno(v, "Failed to detect virtualization, ignoring: %m"); | |
3914 | else if (v > 0) { | |
3915 | const char *s; | |
3916 | ||
3917 | s = strjoina(VIRTUALIZATION_IS_VM(v) ? "vm:" : | |
3918 | VIRTUALIZATION_IS_CONTAINER(v) ? "container:" : ":", | |
3919 | virtualization_to_string(v)); | |
3920 | ||
3921 | r = strv_env_assign(&nl, "SYSTEMD_VIRTUALIZATION", s); | |
3922 | if (r < 0) | |
3923 | return r; | |
3924 | } | |
3925 | ||
3926 | r = strv_env_assign(&nl, "SYSTEMD_ARCHITECTURE", architecture_to_string(uname_architecture())); | |
3927 | if (r < 0) | |
3928 | return r; | |
3929 | ||
3930 | *ret = TAKE_PTR(nl); | |
3931 | return 0; | |
3932 | } | |
3933 | ||
ca6ce62d ZJS |
3934 | static int manager_execute_generators(Manager *m, char **paths, bool remount_ro) { |
3935 | _cleanup_strv_free_ char **ge = NULL; | |
3936 | const char *argv[] = { | |
3937 | NULL, /* Leave this empty, execute_directory() will fill something in */ | |
3938 | m->lookup_paths.generator, | |
3939 | m->lookup_paths.generator_early, | |
3940 | m->lookup_paths.generator_late, | |
3941 | NULL, | |
3942 | }; | |
3943 | int r; | |
3944 | ||
3945 | r = build_generator_environment(m, &ge); | |
3946 | if (r < 0) | |
3947 | return log_error_errno(r, "Failed to build generator environment: %m"); | |
3948 | ||
3949 | if (remount_ro) { | |
3950 | /* Remount most of the filesystem tree read-only. We leave /sys/ as-is, because our code | |
3951 | * checks whether it is read-only to detect containerized execution environments. We leave | |
3952 | * /run/ as-is too, because that's where our output goes. We also leave /proc/ and /dev/shm/ | |
3953 | * because they're API, and /tmp/ that safe_fork() mounted for us. | |
3954 | */ | |
3955 | r = bind_remount_recursive("/", MS_RDONLY, MS_RDONLY, | |
3956 | STRV_MAKE("/sys", "/run", "/proc", "/dev/shm", "/tmp")); | |
3957 | if (r < 0) | |
3958 | log_warning_errno(r, "Read-only bind remount failed, ignoring: %m"); | |
3959 | } | |
3960 | ||
3961 | BLOCK_WITH_UMASK(0022); | |
3962 | return execute_directories( | |
3963 | (const char* const*) paths, | |
3964 | DEFAULT_TIMEOUT_USEC, | |
3965 | /* callbacks= */ NULL, /* callback_args= */ NULL, | |
3966 | (char**) argv, | |
3967 | ge, | |
3968 | EXEC_DIR_PARALLEL | EXEC_DIR_IGNORE_ERRORS | EXEC_DIR_SET_SYSTEMD_EXEC_PID); | |
3969 | } | |
3970 | ||
e801700e | 3971 | static int manager_run_generators(Manager *m) { |
b8fba0cd | 3972 | ForkFlags flags = FORK_RESET_SIGNALS | FORK_WAIT | FORK_NEW_MOUNTNS | FORK_MOUNTNS_SLAVE; |
ca6ce62d | 3973 | _cleanup_strv_free_ char **paths = NULL; |
07719a21 | 3974 | int r; |
5a1e9937 LP |
3975 | |
3976 | assert(m); | |
3977 | ||
638cece4 | 3978 | if (MANAGER_IS_TEST_RUN(m) && !(m->test_run_flags & MANAGER_TEST_RUN_GENERATORS)) |
e0a3da1f ZJS |
3979 | return 0; |
3980 | ||
4870133b | 3981 | paths = generator_binary_paths(m->runtime_scope); |
e801700e ZJS |
3982 | if (!paths) |
3983 | return log_oom(); | |
5a1e9937 | 3984 | |
64691d20 ZJS |
3985 | if (!generator_path_any((const char* const*) paths)) |
3986 | return 0; | |
5a1e9937 | 3987 | |
cd64fd56 | 3988 | r = lookup_paths_mkdir_generator(&m->lookup_paths); |
7eb4f326 LP |
3989 | if (r < 0) { |
3990 | log_error_errno(r, "Failed to create generator directories: %m"); | |
07719a21 | 3991 | goto finish; |
7eb4f326 | 3992 | } |
5a1e9937 | 3993 | |
ca6ce62d ZJS |
3994 | /* If we are the system manager, we fork and invoke the generators in a sanitized mount namespace. If |
3995 | * we are the user manager, let's just execute the generators directly. We might not have the | |
3996 | * necessary privileges, and the system manager has already mounted /tmp/ and everything else for us. | |
3997 | */ | |
3998 | if (MANAGER_IS_USER(m)) { | |
3999 | r = manager_execute_generators(m, paths, /* remount_ro= */ false); | |
82c5db16 LP |
4000 | goto finish; |
4001 | } | |
5a1e9937 | 4002 | |
b8fba0cd LB |
4003 | /* On some systems /tmp/ doesn't exist, and on some other systems we cannot create it at all. Avoid |
4004 | * trying to mount a private tmpfs on it as there's no one size fits all. */ | |
4005 | if (is_dir("/tmp", /* follow= */ false) > 0) | |
4006 | flags |= FORK_PRIVATE_TMP; | |
4007 | ||
4008 | r = safe_fork("(sd-gens)", flags, NULL); | |
ca6ce62d ZJS |
4009 | if (r == 0) { |
4010 | r = manager_execute_generators(m, paths, /* remount_ro= */ true); | |
4011 | _exit(r >= 0 ? EXIT_SUCCESS : EXIT_FAILURE); | |
4012 | } | |
a2275dcb YW |
4013 | if (r < 0) { |
4014 | if (!ERRNO_IS_PRIVILEGE(r)) { | |
4015 | log_error_errno(r, "Failed to fork off sandboxing environment for executing generators: %m"); | |
4016 | goto finish; | |
4017 | } | |
4018 | ||
4019 | /* Failed to fork with new mount namespace? Maybe, running in a container environment with | |
4020 | * seccomp or without capability. */ | |
4021 | log_debug_errno(r, | |
4022 | "Failed to fork off sandboxing environment for executing generators. " | |
4023 | "Falling back to execute generators without sandboxing: %m"); | |
4024 | r = manager_execute_generators(m, paths, /* remount_ro= */ false); | |
4025 | } | |
5a1e9937 | 4026 | |
718db961 | 4027 | finish: |
cd64fd56 | 4028 | lookup_paths_trim_generator(&m->lookup_paths); |
e801700e | 4029 | return r; |
5a1e9937 LP |
4030 | } |
4031 | ||
1ad6e8b3 LP |
4032 | int manager_transient_environment_add(Manager *m, char **plus) { |
4033 | char **a; | |
4034 | ||
4035 | assert(m); | |
4036 | ||
4037 | if (strv_isempty(plus)) | |
4038 | return 0; | |
4039 | ||
4ab3d29f | 4040 | a = strv_env_merge(m->transient_environment, plus); |
1ad6e8b3 | 4041 | if (!a) |
2fbbbf9a | 4042 | return log_oom(); |
1ad6e8b3 LP |
4043 | |
4044 | sanitize_environment(a); | |
4045 | ||
4046 | return strv_free_and_replace(m->transient_environment, a); | |
4047 | } | |
4048 | ||
4049 | int manager_client_environment_modify( | |
4050 | Manager *m, | |
4051 | char **minus, | |
4052 | char **plus) { | |
4053 | ||
718db961 | 4054 | char **a = NULL, **b = NULL, **l; |
1ad6e8b3 | 4055 | |
97d0e5f8 | 4056 | assert(m); |
bcd8e6d1 | 4057 | |
1ad6e8b3 LP |
4058 | if (strv_isempty(minus) && strv_isempty(plus)) |
4059 | return 0; | |
4060 | ||
4061 | l = m->client_environment; | |
bcd8e6d1 | 4062 | |
718db961 LP |
4063 | if (!strv_isempty(minus)) { |
4064 | a = strv_env_delete(l, 1, minus); | |
4065 | if (!a) | |
4066 | return -ENOMEM; | |
4067 | ||
4068 | l = a; | |
4069 | } | |
4070 | ||
4071 | if (!strv_isempty(plus)) { | |
4ab3d29f | 4072 | b = strv_env_merge(l, plus); |
aa9f8a30 AH |
4073 | if (!b) { |
4074 | strv_free(a); | |
718db961 | 4075 | return -ENOMEM; |
aa9f8a30 | 4076 | } |
bcd8e6d1 | 4077 | |
718db961 LP |
4078 | l = b; |
4079 | } | |
4080 | ||
1ad6e8b3 LP |
4081 | if (m->client_environment != l) |
4082 | strv_free(m->client_environment); | |
4083 | ||
718db961 LP |
4084 | if (a != l) |
4085 | strv_free(a); | |
4086 | if (b != l) | |
4087 | strv_free(b); | |
4088 | ||
1ad6e8b3 LP |
4089 | m->client_environment = sanitize_environment(l); |
4090 | return 0; | |
4091 | } | |
4092 | ||
4093 | int manager_get_effective_environment(Manager *m, char ***ret) { | |
4094 | char **l; | |
4095 | ||
4096 | assert(m); | |
4097 | assert(ret); | |
4098 | ||
4ab3d29f | 4099 | l = strv_env_merge(m->transient_environment, m->client_environment); |
1ad6e8b3 LP |
4100 | if (!l) |
4101 | return -ENOMEM; | |
f069efb4 | 4102 | |
1ad6e8b3 | 4103 | *ret = l; |
97d0e5f8 UTL |
4104 | return 0; |
4105 | } | |
4106 | ||
aa5ae971 ŁS |
4107 | int manager_set_default_smack_process_label(Manager *m, const char *label) { |
4108 | assert(m); | |
4109 | ||
4110 | #ifdef SMACK_DEFAULT_PROCESS_LABEL | |
4111 | if (!label) | |
4112 | return free_and_strdup(&m->default_smack_process_label, SMACK_DEFAULT_PROCESS_LABEL); | |
4113 | #endif | |
4114 | if (streq_ptr(label, "/")) | |
4115 | return free_and_strdup(&m->default_smack_process_label, NULL); | |
4116 | ||
4117 | return free_and_strdup(&m->default_smack_process_label, label); | |
4118 | } | |
4119 | ||
c93ff2e9 | 4120 | int manager_set_default_rlimits(Manager *m, struct rlimit **default_rlimit) { |
c93ff2e9 FC |
4121 | assert(m); |
4122 | ||
5b10116e | 4123 | for (unsigned i = 0; i < _RLIMIT_MAX; i++) { |
d9814c76 EV |
4124 | m->rlimit[i] = mfree(m->rlimit[i]); |
4125 | ||
07719a21 LP |
4126 | if (!default_rlimit[i]) |
4127 | continue; | |
c93ff2e9 | 4128 | |
07719a21 LP |
4129 | m->rlimit[i] = newdup(struct rlimit, default_rlimit[i], 1); |
4130 | if (!m->rlimit[i]) | |
3ce40911 | 4131 | return log_oom(); |
c93ff2e9 FC |
4132 | } |
4133 | ||
4134 | return 0; | |
4135 | } | |
4136 | ||
8559b3b7 LP |
4137 | void manager_recheck_dbus(Manager *m) { |
4138 | assert(m); | |
4139 | ||
8f41e6b6 ZJS |
4140 | /* Connects to the bus if the dbus service and socket are running. If we are running in user mode |
4141 | * this is all it does. In system mode we'll also connect to the system bus (which will most likely | |
4142 | * just reuse the connection of the API bus). That's because the system bus after all runs as service | |
4143 | * of the system instance, while in the user instance we can assume it's already there. */ | |
8559b3b7 | 4144 | |
31dc1ca3 LP |
4145 | if (MANAGER_IS_RELOADING(m)) |
4146 | return; /* don't check while we are reloading… */ | |
4147 | ||
2b680534 | 4148 | if (manager_dbus_is_running(m, false)) { |
8559b3b7 LP |
4149 | (void) bus_init_api(m); |
4150 | ||
4151 | if (MANAGER_IS_SYSTEM(m)) | |
4152 | (void) bus_init_system(m); | |
4153 | } else { | |
4154 | (void) bus_done_api(m); | |
4155 | ||
4156 | if (MANAGER_IS_SYSTEM(m)) | |
4157 | (void) bus_done_system(m); | |
4158 | } | |
4159 | } | |
4160 | ||
d075092f | 4161 | static bool manager_journal_is_running(Manager *m) { |
f1dd0c3f LP |
4162 | Unit *u; |
4163 | ||
4164 | assert(m); | |
4165 | ||
638cece4 | 4166 | if (MANAGER_IS_TEST_RUN(m)) |
7d814a19 LP |
4167 | return false; |
4168 | ||
d075092f | 4169 | /* If we are the user manager we can safely assume that the journal is up */ |
463d0d15 | 4170 | if (!MANAGER_IS_SYSTEM(m)) |
d075092f | 4171 | return true; |
f1dd0c3f | 4172 | |
d075092f | 4173 | /* Check that the socket is not only up, but in RUNNING state */ |
731a676c | 4174 | u = manager_get_unit(m, SPECIAL_JOURNALD_SOCKET); |
d075092f LP |
4175 | if (!u) |
4176 | return false; | |
4177 | if (SOCKET(u)->state != SOCKET_RUNNING) | |
4178 | return false; | |
f1dd0c3f | 4179 | |
d075092f | 4180 | /* Similar, check if the daemon itself is fully up, too */ |
731a676c | 4181 | u = manager_get_unit(m, SPECIAL_JOURNALD_SERVICE); |
d075092f LP |
4182 | if (!u) |
4183 | return false; | |
217677ab | 4184 | if (!IN_SET(SERVICE(u)->state, SERVICE_RELOAD, SERVICE_RUNNING)) |
d075092f LP |
4185 | return false; |
4186 | ||
4187 | return true; | |
4188 | } | |
4189 | ||
6123dfaa ZJS |
4190 | void disable_printk_ratelimit(void) { |
4191 | /* Disable kernel's printk ratelimit. | |
4192 | * | |
4193 | * Logging to /dev/kmsg is most useful during early boot and shutdown, where normal logging | |
4194 | * mechanisms are not available. The semantics of this sysctl are such that any kernel command-line | |
4195 | * setting takes precedence. */ | |
4196 | int r; | |
4197 | ||
4198 | r = sysctl_write("kernel/printk_devkmsg", "on"); | |
4199 | if (r < 0) | |
4200 | log_debug_errno(r, "Failed to set sysctl kernel.printk_devkmsg=on: %m"); | |
4201 | } | |
4202 | ||
d075092f LP |
4203 | void manager_recheck_journal(Manager *m) { |
4204 | ||
4205 | assert(m); | |
4206 | ||
4207 | /* Don't bother with this unless we are in the special situation of being PID 1 */ | |
4208 | if (getpid_cached() != 1) | |
731a676c | 4209 | return; |
f1dd0c3f | 4210 | |
31dc1ca3 LP |
4211 | /* Don't check this while we are reloading, things might still change */ |
4212 | if (MANAGER_IS_RELOADING(m)) | |
4213 | return; | |
4214 | ||
8f41e6b6 ZJS |
4215 | /* The journal is fully and entirely up? If so, let's permit logging to it, if that's configured. If |
4216 | * the journal is down, don't ever log to it, otherwise we might end up deadlocking ourselves as we | |
4217 | * might trigger an activation ourselves we can't fulfill. */ | |
cedf5088 | 4218 | log_set_prohibit_ipc(!manager_journal_is_running(m)); |
cc2b9e6b | 4219 | log_open(); |
f1dd0c3f LP |
4220 | } |
4221 | ||
44a41954 FB |
4222 | static ShowStatus manager_get_show_status(Manager *m) { |
4223 | assert(m); | |
4224 | ||
4225 | if (MANAGER_IS_USER(m)) | |
4226 | return _SHOW_STATUS_INVALID; | |
4227 | ||
4228 | if (m->show_status_overridden != _SHOW_STATUS_INVALID) | |
4229 | return m->show_status_overridden; | |
4230 | ||
4231 | return m->show_status; | |
4232 | } | |
4233 | ||
4234 | bool manager_get_show_status_on(Manager *m) { | |
4235 | assert(m); | |
4236 | ||
4237 | return show_status_on(manager_get_show_status(m)); | |
4238 | } | |
b309078a | 4239 | |
3ceb3471 FB |
4240 | static void set_show_status_marker(bool b) { |
4241 | if (b) | |
4242 | (void) touch("/run/systemd/show-status"); | |
4243 | else | |
4244 | (void) unlink("/run/systemd/show-status"); | |
4245 | } | |
4246 | ||
44a41954 | 4247 | void manager_set_show_status(Manager *m, ShowStatus mode, const char *reason) { |
27d340c7 | 4248 | assert(m); |
44a41954 | 4249 | assert(reason); |
0d066dd1 | 4250 | assert(mode >= 0 && mode < _SHOW_STATUS_MAX); |
27d340c7 | 4251 | |
44a41954 | 4252 | if (MANAGER_IS_USER(m)) |
27d340c7 LP |
4253 | return; |
4254 | ||
ef15d3e1 ZJS |
4255 | if (mode == m->show_status) |
4256 | return; | |
4257 | ||
44a41954 FB |
4258 | if (m->show_status_overridden == _SHOW_STATUS_INVALID) { |
4259 | bool enabled; | |
4260 | ||
4261 | enabled = show_status_on(mode); | |
4262 | log_debug("%s (%s) showing of status (%s).", | |
4263 | enabled ? "Enabling" : "Disabling", | |
4264 | strna(show_status_to_string(mode)), | |
4265 | reason); | |
4266 | ||
3ceb3471 | 4267 | set_show_status_marker(enabled); |
44a41954 FB |
4268 | } |
4269 | ||
4270 | m->show_status = mode; | |
4271 | } | |
4272 | ||
43bba15a | 4273 | void manager_override_show_status(Manager *m, ShowStatus mode, const char *reason) { |
44a41954 FB |
4274 | assert(m); |
4275 | assert(mode < _SHOW_STATUS_MAX); | |
4276 | ||
4277 | if (MANAGER_IS_USER(m)) | |
4278 | return; | |
4279 | ||
4280 | if (mode == m->show_status_overridden) | |
4281 | return; | |
4282 | ||
4283 | m->show_status_overridden = mode; | |
4284 | ||
4285 | if (mode == _SHOW_STATUS_INVALID) | |
4286 | mode = m->show_status; | |
4287 | ||
ef15d3e1 | 4288 | log_debug("%s (%s) showing of status (%s).", |
44a41954 | 4289 | m->show_status_overridden != _SHOW_STATUS_INVALID ? "Overriding" : "Restoring", |
ef15d3e1 ZJS |
4290 | strna(show_status_to_string(mode)), |
4291 | reason); | |
b309078a | 4292 | |
3ceb3471 | 4293 | set_show_status_marker(show_status_on(mode)); |
27d340c7 LP |
4294 | } |
4295 | ||
7d5ceb64 FB |
4296 | const char *manager_get_confirm_spawn(Manager *m) { |
4297 | static int last_errno = 0; | |
7d5ceb64 FB |
4298 | struct stat st; |
4299 | int r; | |
4300 | ||
ea758432 LP |
4301 | assert(m); |
4302 | ||
7d5ceb64 FB |
4303 | /* Here's the deal: we want to test the validity of the console but don't want |
4304 | * PID1 to go through the whole console process which might block. But we also | |
4305 | * want to warn the user only once if something is wrong with the console so we | |
4306 | * cannot do the sanity checks after spawning our children. So here we simply do | |
4307 | * really basic tests to hopefully trap common errors. | |
4308 | * | |
4309 | * If the console suddenly disappear at the time our children will really it | |
4310 | * then they will simply fail to acquire it and a positive answer will be | |
2aed63f4 | 4311 | * assumed. New children will fall back to /dev/console though. |
7d5ceb64 FB |
4312 | * |
4313 | * Note: TTYs are devices that can come and go any time, and frequently aren't | |
4314 | * available yet during early boot (consider a USB rs232 dongle...). If for any | |
2aed63f4 | 4315 | * reason the configured console is not ready, we fall back to the default |
7d5ceb64 FB |
4316 | * console. */ |
4317 | ||
ea758432 LP |
4318 | if (!m->confirm_spawn || path_equal(m->confirm_spawn, "/dev/console")) |
4319 | return m->confirm_spawn; | |
7d5ceb64 | 4320 | |
ea758432 LP |
4321 | if (stat(m->confirm_spawn, &st) < 0) { |
4322 | r = -errno; | |
7d5ceb64 | 4323 | goto fail; |
ea758432 | 4324 | } |
7d5ceb64 FB |
4325 | |
4326 | if (!S_ISCHR(st.st_mode)) { | |
ea758432 | 4327 | r = -ENOTTY; |
7d5ceb64 FB |
4328 | goto fail; |
4329 | } | |
4330 | ||
4331 | last_errno = 0; | |
ea758432 LP |
4332 | return m->confirm_spawn; |
4333 | ||
7d5ceb64 | 4334 | fail: |
ea758432 LP |
4335 | if (last_errno != r) |
4336 | last_errno = log_warning_errno(r, "Failed to open %s, using default console: %m", m->confirm_spawn); | |
4337 | ||
7d5ceb64 FB |
4338 | return "/dev/console"; |
4339 | } | |
4340 | ||
e2680723 LP |
4341 | void manager_set_first_boot(Manager *m, bool b) { |
4342 | assert(m); | |
4343 | ||
463d0d15 | 4344 | if (!MANAGER_IS_SYSTEM(m)) |
e2680723 LP |
4345 | return; |
4346 | ||
ae2a2c53 LP |
4347 | if (m->first_boot != (int) b) { |
4348 | if (b) | |
4349 | (void) touch("/run/systemd/first-boot"); | |
4350 | else | |
4351 | (void) unlink("/run/systemd/first-boot"); | |
4352 | } | |
e2680723 | 4353 | |
ae2a2c53 | 4354 | m->first_boot = b; |
e2680723 LP |
4355 | } |
4356 | ||
b0eb2944 FB |
4357 | void manager_disable_confirm_spawn(void) { |
4358 | (void) touch("/run/systemd/confirm_spawn_disabled"); | |
4359 | } | |
4360 | ||
4361 | bool manager_is_confirm_spawn_disabled(Manager *m) { | |
4362 | if (!m->confirm_spawn) | |
4363 | return true; | |
4364 | ||
4365 | return access("/run/systemd/confirm_spawn_disabled", F_OK) >= 0; | |
4366 | } | |
4367 | ||
0d6d3cf0 FB |
4368 | static bool manager_should_show_status(Manager *m, StatusType type) { |
4369 | assert(m); | |
4370 | ||
4371 | if (!MANAGER_IS_SYSTEM(m)) | |
4372 | return false; | |
4373 | ||
4374 | if (m->no_console_output) | |
4375 | return false; | |
4376 | ||
4377 | if (!IN_SET(manager_state(m), MANAGER_INITIALIZING, MANAGER_STARTING, MANAGER_STOPPING)) | |
4378 | return false; | |
4379 | ||
4380 | /* If we cannot find out the status properly, just proceed. */ | |
4381 | if (type != STATUS_TYPE_EMERGENCY && manager_check_ask_password(m) > 0) | |
4382 | return false; | |
4383 | ||
4384 | if (type == STATUS_TYPE_NOTICE && m->show_status != SHOW_STATUS_NO) | |
4385 | return true; | |
4386 | ||
44a41954 | 4387 | return manager_get_show_status_on(m); |
0d6d3cf0 FB |
4388 | } |
4389 | ||
127d5fd1 | 4390 | void manager_status_printf(Manager *m, StatusType type, const char *status, const char *format, ...) { |
25cee550 MS |
4391 | va_list ap; |
4392 | ||
cb6531be ZJS |
4393 | /* If m is NULL, assume we're after shutdown and let the messages through. */ |
4394 | ||
0d6d3cf0 | 4395 | if (m && !manager_should_show_status(m, type)) |
25cee550 MS |
4396 | return; |
4397 | ||
03b717a3 MS |
4398 | /* XXX We should totally drop the check for ephemeral here |
4399 | * and thus effectively make 'Type=idle' pointless. */ | |
cb6531be | 4400 | if (type == STATUS_TYPE_EPHEMERAL && m && m->n_on_console > 0) |
03b717a3 MS |
4401 | return; |
4402 | ||
25cee550 | 4403 | va_start(ap, format); |
a885727a | 4404 | status_vprintf(status, SHOW_STATUS_ELLIPSIZE|(type == STATUS_TYPE_EPHEMERAL ? SHOW_STATUS_EPHEMERAL : 0), format, ap); |
25cee550 MS |
4405 | va_end(ap); |
4406 | } | |
4407 | ||
ac19bdd0 | 4408 | Set* manager_get_units_requiring_mounts_for(Manager *m, const char *path) { |
a57f7e2c LP |
4409 | assert(m); |
4410 | assert(path); | |
4411 | ||
ac19bdd0 ZJS |
4412 | if (path_equal(path, "/")) |
4413 | path = ""; | |
a57f7e2c | 4414 | |
ac19bdd0 | 4415 | return hashmap_get(m->units_requiring_mounts_for, path); |
a57f7e2c | 4416 | } |
e66cf1a3 | 4417 | |
5269eb6b | 4418 | int manager_update_failed_units(Manager *m, Unit *u, bool failed) { |
03455c28 | 4419 | unsigned size; |
5269eb6b | 4420 | int r; |
03455c28 LDM |
4421 | |
4422 | assert(m); | |
4423 | assert(u->manager == m); | |
4424 | ||
4425 | size = set_size(m->failed_units); | |
4426 | ||
9fff8981 | 4427 | if (failed) { |
de7fef4b | 4428 | r = set_ensure_put(&m->failed_units, NULL, u); |
5269eb6b LP |
4429 | if (r < 0) |
4430 | return log_oom(); | |
9fff8981 | 4431 | } else |
5269eb6b | 4432 | (void) set_remove(m->failed_units, u); |
03455c28 LDM |
4433 | |
4434 | if (set_size(m->failed_units) != size) | |
4435 | bus_manager_send_change_signal(m); | |
5269eb6b LP |
4436 | |
4437 | return 0; | |
03455c28 LDM |
4438 | } |
4439 | ||
f755e3b7 LP |
4440 | ManagerState manager_state(Manager *m) { |
4441 | Unit *u; | |
4442 | ||
4443 | assert(m); | |
4444 | ||
f9d29f6d JB |
4445 | /* Is the special shutdown target active or queued? If so, we are in shutdown state */ |
4446 | u = manager_get_unit(m, SPECIAL_SHUTDOWN_TARGET); | |
4447 | if (u && unit_active_or_pending(u)) | |
4448 | return MANAGER_STOPPING; | |
4449 | ||
f755e3b7 | 4450 | /* Did we ever finish booting? If not then we are still starting up */ |
49d5666c | 4451 | if (!MANAGER_IS_FINISHED(m)) { |
d81afec1 LP |
4452 | |
4453 | u = manager_get_unit(m, SPECIAL_BASIC_TARGET); | |
4454 | if (!u || !UNIT_IS_ACTIVE_OR_RELOADING(unit_active_state(u))) | |
4455 | return MANAGER_INITIALIZING; | |
4456 | ||
f755e3b7 | 4457 | return MANAGER_STARTING; |
d81afec1 | 4458 | } |
f755e3b7 | 4459 | |
45a7b16b LP |
4460 | if (MANAGER_IS_SYSTEM(m)) { |
4461 | /* Are the rescue or emergency targets active or queued? If so we are in maintenance state */ | |
4462 | u = manager_get_unit(m, SPECIAL_RESCUE_TARGET); | |
4463 | if (u && unit_active_or_pending(u)) | |
4464 | return MANAGER_MAINTENANCE; | |
f755e3b7 | 4465 | |
45a7b16b LP |
4466 | u = manager_get_unit(m, SPECIAL_EMERGENCY_TARGET); |
4467 | if (u && unit_active_or_pending(u)) | |
4468 | return MANAGER_MAINTENANCE; | |
4469 | } | |
f755e3b7 LP |
4470 | |
4471 | /* Are there any failed units? If so, we are in degraded mode */ | |
4472 | if (set_size(m->failed_units) > 0) | |
4473 | return MANAGER_DEGRADED; | |
4474 | ||
4475 | return MANAGER_RUNNING; | |
4476 | } | |
4477 | ||
00d9ef85 | 4478 | static void manager_unref_uid_internal( |
010becd9 | 4479 | Hashmap *uid_refs, |
00d9ef85 LP |
4480 | uid_t uid, |
4481 | bool destroy_now, | |
4482 | int (*_clean_ipc)(uid_t uid)) { | |
4483 | ||
4484 | uint32_t c, n; | |
4485 | ||
00d9ef85 LP |
4486 | assert(uid_is_valid(uid)); |
4487 | assert(_clean_ipc); | |
4488 | ||
8f41e6b6 ZJS |
4489 | /* A generic implementation, covering both manager_unref_uid() and manager_unref_gid(), under the |
4490 | * assumption that uid_t and gid_t are actually defined the same way, with the same validity rules. | |
00d9ef85 | 4491 | * |
8f41e6b6 ZJS |
4492 | * We store a hashmap where the key is the UID/GID and the value is a 32bit reference counter, whose |
4493 | * highest bit is used as flag for marking UIDs/GIDs whose IPC objects to remove when the last | |
4494 | * reference to the UID/GID is dropped. The flag is set to on, once at least one reference from a | |
4495 | * unit where RemoveIPC= is set is added on a UID/GID. It is reset when the UID's/GID's reference | |
4496 | * counter drops to 0 again. */ | |
00d9ef85 LP |
4497 | |
4498 | assert_cc(sizeof(uid_t) == sizeof(gid_t)); | |
4499 | assert_cc(UID_INVALID == (uid_t) GID_INVALID); | |
4500 | ||
4501 | if (uid == 0) /* We don't keep track of root, and will never destroy it */ | |
4502 | return; | |
4503 | ||
010becd9 | 4504 | c = PTR_TO_UINT32(hashmap_get(uid_refs, UID_TO_PTR(uid))); |
00d9ef85 LP |
4505 | |
4506 | n = c & ~DESTROY_IPC_FLAG; | |
4507 | assert(n > 0); | |
4508 | n--; | |
4509 | ||
4510 | if (destroy_now && n == 0) { | |
010becd9 | 4511 | hashmap_remove(uid_refs, UID_TO_PTR(uid)); |
00d9ef85 LP |
4512 | |
4513 | if (c & DESTROY_IPC_FLAG) { | |
4514 | log_debug("%s " UID_FMT " is no longer referenced, cleaning up its IPC.", | |
4515 | _clean_ipc == clean_ipc_by_uid ? "UID" : "GID", | |
4516 | uid); | |
4517 | (void) _clean_ipc(uid); | |
4518 | } | |
4519 | } else { | |
4520 | c = n | (c & DESTROY_IPC_FLAG); | |
010becd9 | 4521 | assert_se(hashmap_update(uid_refs, UID_TO_PTR(uid), UINT32_TO_PTR(c)) >= 0); |
00d9ef85 LP |
4522 | } |
4523 | } | |
4524 | ||
4525 | void manager_unref_uid(Manager *m, uid_t uid, bool destroy_now) { | |
010becd9 | 4526 | manager_unref_uid_internal(m->uid_refs, uid, destroy_now, clean_ipc_by_uid); |
00d9ef85 LP |
4527 | } |
4528 | ||
4529 | void manager_unref_gid(Manager *m, gid_t gid, bool destroy_now) { | |
010becd9 | 4530 | manager_unref_uid_internal(m->gid_refs, (uid_t) gid, destroy_now, clean_ipc_by_gid); |
00d9ef85 LP |
4531 | } |
4532 | ||
4533 | static int manager_ref_uid_internal( | |
00d9ef85 LP |
4534 | Hashmap **uid_refs, |
4535 | uid_t uid, | |
4536 | bool clean_ipc) { | |
4537 | ||
4538 | uint32_t c, n; | |
4539 | int r; | |
4540 | ||
00d9ef85 LP |
4541 | assert(uid_refs); |
4542 | assert(uid_is_valid(uid)); | |
4543 | ||
8f41e6b6 ZJS |
4544 | /* A generic implementation, covering both manager_ref_uid() and manager_ref_gid(), under the |
4545 | * assumption that uid_t and gid_t are actually defined the same way, with the same validity | |
4546 | * rules. */ | |
00d9ef85 LP |
4547 | |
4548 | assert_cc(sizeof(uid_t) == sizeof(gid_t)); | |
4549 | assert_cc(UID_INVALID == (uid_t) GID_INVALID); | |
4550 | ||
4551 | if (uid == 0) /* We don't keep track of root, and will never destroy it */ | |
4552 | return 0; | |
4553 | ||
4554 | r = hashmap_ensure_allocated(uid_refs, &trivial_hash_ops); | |
4555 | if (r < 0) | |
4556 | return r; | |
4557 | ||
4558 | c = PTR_TO_UINT32(hashmap_get(*uid_refs, UID_TO_PTR(uid))); | |
4559 | ||
4560 | n = c & ~DESTROY_IPC_FLAG; | |
4561 | n++; | |
4562 | ||
4563 | if (n & DESTROY_IPC_FLAG) /* check for overflow */ | |
4564 | return -EOVERFLOW; | |
4565 | ||
4566 | c = n | (c & DESTROY_IPC_FLAG) | (clean_ipc ? DESTROY_IPC_FLAG : 0); | |
4567 | ||
4568 | return hashmap_replace(*uid_refs, UID_TO_PTR(uid), UINT32_TO_PTR(c)); | |
4569 | } | |
4570 | ||
4571 | int manager_ref_uid(Manager *m, uid_t uid, bool clean_ipc) { | |
010becd9 | 4572 | return manager_ref_uid_internal(&m->uid_refs, uid, clean_ipc); |
00d9ef85 LP |
4573 | } |
4574 | ||
4575 | int manager_ref_gid(Manager *m, gid_t gid, bool clean_ipc) { | |
010becd9 | 4576 | return manager_ref_uid_internal(&m->gid_refs, (uid_t) gid, clean_ipc); |
00d9ef85 LP |
4577 | } |
4578 | ||
4579 | static void manager_vacuum_uid_refs_internal( | |
010becd9 | 4580 | Hashmap *uid_refs, |
00d9ef85 LP |
4581 | int (*_clean_ipc)(uid_t uid)) { |
4582 | ||
00d9ef85 LP |
4583 | void *p, *k; |
4584 | ||
00d9ef85 LP |
4585 | assert(_clean_ipc); |
4586 | ||
010becd9 | 4587 | HASHMAP_FOREACH_KEY(p, k, uid_refs) { |
00d9ef85 LP |
4588 | uint32_t c, n; |
4589 | uid_t uid; | |
4590 | ||
4591 | uid = PTR_TO_UID(k); | |
4592 | c = PTR_TO_UINT32(p); | |
4593 | ||
4594 | n = c & ~DESTROY_IPC_FLAG; | |
4595 | if (n > 0) | |
4596 | continue; | |
4597 | ||
4598 | if (c & DESTROY_IPC_FLAG) { | |
4599 | log_debug("Found unreferenced %s " UID_FMT " after reload/reexec. Cleaning up.", | |
4600 | _clean_ipc == clean_ipc_by_uid ? "UID" : "GID", | |
4601 | uid); | |
4602 | (void) _clean_ipc(uid); | |
4603 | } | |
4604 | ||
010becd9 | 4605 | assert_se(hashmap_remove(uid_refs, k) == p); |
00d9ef85 LP |
4606 | } |
4607 | } | |
4608 | ||
06a4eb07 | 4609 | static void manager_vacuum_uid_refs(Manager *m) { |
010becd9 | 4610 | manager_vacuum_uid_refs_internal(m->uid_refs, clean_ipc_by_uid); |
00d9ef85 LP |
4611 | } |
4612 | ||
06a4eb07 | 4613 | static void manager_vacuum_gid_refs(Manager *m) { |
010becd9 | 4614 | manager_vacuum_uid_refs_internal(m->gid_refs, clean_ipc_by_gid); |
00d9ef85 LP |
4615 | } |
4616 | ||
06a4eb07 FB |
4617 | static void manager_vacuum(Manager *m) { |
4618 | assert(m); | |
4619 | ||
4620 | /* Release any dynamic users no longer referenced */ | |
4621 | dynamic_user_vacuum(m, true); | |
4622 | ||
4623 | /* Release any references to UIDs/GIDs no longer referenced, and destroy any IPC owned by them */ | |
4624 | manager_vacuum_uid_refs(m); | |
4625 | manager_vacuum_gid_refs(m); | |
4626 | ||
4627 | /* Release any runtimes no longer referenced */ | |
e76506b7 | 4628 | exec_shared_runtime_vacuum(m); |
06a4eb07 FB |
4629 | } |
4630 | ||
00d9ef85 LP |
4631 | int manager_dispatch_user_lookup_fd(sd_event_source *source, int fd, uint32_t revents, void *userdata) { |
4632 | struct buffer { | |
4633 | uid_t uid; | |
4634 | gid_t gid; | |
4635 | char unit_name[UNIT_NAME_MAX+1]; | |
4636 | } _packed_ buffer; | |
4637 | ||
4638 | Manager *m = userdata; | |
4639 | ssize_t l; | |
4640 | size_t n; | |
4641 | Unit *u; | |
4642 | ||
4643 | assert_se(source); | |
4644 | assert_se(m); | |
4645 | ||
8f41e6b6 ZJS |
4646 | /* Invoked whenever a child process succeeded resolving its user/group to use and sent us the |
4647 | * resulting UID/GID in a datagram. We parse the datagram here and pass it off to the unit, so that | |
4648 | * it can add a reference to the UID/GID so that it can destroy the UID/GID's IPC objects when the | |
4649 | * reference counter drops to 0. */ | |
00d9ef85 LP |
4650 | |
4651 | l = recv(fd, &buffer, sizeof(buffer), MSG_DONTWAIT); | |
4652 | if (l < 0) { | |
8add30a0 | 4653 | if (ERRNO_IS_TRANSIENT(errno)) |
00d9ef85 LP |
4654 | return 0; |
4655 | ||
4656 | return log_error_errno(errno, "Failed to read from user lookup fd: %m"); | |
4657 | } | |
4658 | ||
4659 | if ((size_t) l <= offsetof(struct buffer, unit_name)) { | |
4660 | log_warning("Received too short user lookup message, ignoring."); | |
4661 | return 0; | |
4662 | } | |
4663 | ||
4664 | if ((size_t) l > offsetof(struct buffer, unit_name) + UNIT_NAME_MAX) { | |
4665 | log_warning("Received too long user lookup message, ignoring."); | |
4666 | return 0; | |
4667 | } | |
4668 | ||
4669 | if (!uid_is_valid(buffer.uid) && !gid_is_valid(buffer.gid)) { | |
4670 | log_warning("Got user lookup message with invalid UID/GID pair, ignoring."); | |
4671 | return 0; | |
4672 | } | |
4673 | ||
4674 | n = (size_t) l - offsetof(struct buffer, unit_name); | |
4675 | if (memchr(buffer.unit_name, 0, n)) { | |
4676 | log_warning("Received lookup message with embedded NUL character, ignoring."); | |
4677 | return 0; | |
4678 | } | |
4679 | ||
4680 | buffer.unit_name[n] = 0; | |
4681 | u = manager_get_unit(m, buffer.unit_name); | |
4682 | if (!u) { | |
4683 | log_debug("Got user lookup message but unit doesn't exist, ignoring."); | |
4684 | return 0; | |
4685 | } | |
4686 | ||
4687 | log_unit_debug(u, "User lookup succeeded: uid=" UID_FMT " gid=" GID_FMT, buffer.uid, buffer.gid); | |
4688 | ||
4689 | unit_notify_user_lookup(u, buffer.uid, buffer.gid); | |
4690 | return 0; | |
4691 | } | |
4692 | ||
63e8df04 | 4693 | static int short_uid_range(const char *path) { |
8dcc66ce | 4694 | _cleanup_(uid_range_freep) UidRange *p = NULL; |
63e8df04 LP |
4695 | int r; |
4696 | ||
4697 | assert(path); | |
4698 | ||
4699 | /* Taint systemd if we the UID range assigned to this environment doesn't at least cover 0…65534, | |
4700 | * i.e. from root to nobody. */ | |
4701 | ||
8dcc66ce YW |
4702 | r = uid_range_load_userns(&p, path); |
4703 | if (r < 0) { | |
4704 | if (ERRNO_IS_NOT_SUPPORTED(r)) | |
4705 | return false; | |
63e8df04 | 4706 | return log_debug_errno(r, "Failed to load %s: %m", path); |
8dcc66ce | 4707 | } |
63e8df04 | 4708 | |
8dcc66ce | 4709 | return !uid_range_covers(p, 0, 65535); |
63e8df04 LP |
4710 | } |
4711 | ||
83fe5d8a ZJS |
4712 | char* manager_taint_string(const Manager *m) { |
4713 | /* Returns a "taint string", e.g. "local-hwclock:var-run-bad". Only things that are detected at | |
4714 | * runtime should be tagged here. For stuff that is known during compilation, emit a warning in the | |
63e8df04 | 4715 | * configuration phase. */ |
198ce932 | 4716 | |
af6b0ecc LP |
4717 | assert(m); |
4718 | ||
4bd03515 | 4719 | const char* stage[13] = {}; |
83fe5d8a | 4720 | size_t n = 0; |
af6b0ecc LP |
4721 | |
4722 | if (m->taint_usr) | |
83fe5d8a | 4723 | stage[n++] = "split-usr"; |
af6b0ecc | 4724 | |
31cd2dd9 LB |
4725 | _cleanup_free_ char *usrbin = NULL; |
4726 | if (readlink_malloc("/bin", &usrbin) < 0 || !PATH_IN_SET(usrbin, "usr/bin", "/usr/bin")) | |
4727 | stage[n++] = "unmerged-usr"; | |
4728 | ||
af6b0ecc | 4729 | if (access("/proc/cgroups", F_OK) < 0) |
83fe5d8a | 4730 | stage[n++] = "cgroups-missing"; |
af6b0ecc | 4731 | |
82f30632 | 4732 | if (cg_all_unified() == 0) |
83fe5d8a | 4733 | stage[n++] = "cgroupsv1"; |
82f30632 | 4734 | |
af6b0ecc | 4735 | if (clock_is_localtime(NULL) > 0) |
83fe5d8a | 4736 | stage[n++] = "local-hwclock"; |
af6b0ecc | 4737 | |
6bfe9b3b | 4738 | if (os_release_support_ended(NULL, /* quiet= */ true, NULL) > 0) |
4bd03515 ZJS |
4739 | stage[n++] = "support-ended"; |
4740 | ||
83fe5d8a ZJS |
4741 | _cleanup_free_ char *destination = NULL; |
4742 | if (readlink_malloc("/var/run", &destination) < 0 || | |
4743 | !PATH_IN_SET(destination, "../run", "/run")) | |
4744 | stage[n++] = "var-run-bad"; | |
af6b0ecc | 4745 | |
83fe5d8a ZJS |
4746 | _cleanup_free_ char *overflowuid = NULL, *overflowgid = NULL; |
4747 | if (read_one_line_file("/proc/sys/kernel/overflowuid", &overflowuid) >= 0 && | |
4748 | !streq(overflowuid, "65534")) | |
4749 | stage[n++] = "overflowuid-not-65534"; | |
4750 | if (read_one_line_file("/proc/sys/kernel/overflowgid", &overflowgid) >= 0 && | |
4751 | !streq(overflowgid, "65534")) | |
4752 | stage[n++] = "overflowgid-not-65534"; | |
90d7464d | 4753 | |
83fe5d8a | 4754 | struct utsname uts; |
40efaaed LP |
4755 | assert_se(uname(&uts) >= 0); |
4756 | if (strverscmp_improved(uts.release, KERNEL_BASELINE_VERSION) < 0) | |
83fe5d8a | 4757 | stage[n++] = "old-kernel"; |
40efaaed | 4758 | |
63e8df04 | 4759 | if (short_uid_range("/proc/self/uid_map") > 0) |
83fe5d8a | 4760 | stage[n++] = "short-uid-range"; |
63e8df04 | 4761 | if (short_uid_range("/proc/self/gid_map") > 0) |
83fe5d8a | 4762 | stage[n++] = "short-gid-range"; |
63e8df04 | 4763 | |
83fe5d8a | 4764 | assert(n < ELEMENTSOF(stage) - 1); /* One extra for NULL terminator */ |
af6b0ecc | 4765 | |
83fe5d8a | 4766 | return strv_join((char**) stage, ":"); |
af6b0ecc LP |
4767 | } |
4768 | ||
adefcf28 LP |
4769 | void manager_ref_console(Manager *m) { |
4770 | assert(m); | |
4771 | ||
4772 | m->n_on_console++; | |
4773 | } | |
4774 | ||
4775 | void manager_unref_console(Manager *m) { | |
4776 | ||
4777 | assert(m->n_on_console > 0); | |
4778 | m->n_on_console--; | |
4779 | ||
4780 | if (m->n_on_console == 0) | |
4781 | m->no_console_output = false; /* unset no_console_output flag, since the console is definitely free now */ | |
4782 | } | |
4783 | ||
a6ecbf83 FB |
4784 | void manager_override_log_level(Manager *m, int level) { |
4785 | _cleanup_free_ char *s = NULL; | |
4786 | assert(m); | |
4787 | ||
4788 | if (!m->log_level_overridden) { | |
4789 | m->original_log_level = log_get_max_level(); | |
4790 | m->log_level_overridden = true; | |
4791 | } | |
4792 | ||
4793 | (void) log_level_to_string_alloc(level, &s); | |
4794 | log_info("Setting log level to %s.", strna(s)); | |
4795 | ||
4796 | log_set_max_level(level); | |
4797 | } | |
4798 | ||
4799 | void manager_restore_original_log_level(Manager *m) { | |
4800 | _cleanup_free_ char *s = NULL; | |
4801 | assert(m); | |
4802 | ||
4803 | if (!m->log_level_overridden) | |
4804 | return; | |
4805 | ||
4806 | (void) log_level_to_string_alloc(m->original_log_level, &s); | |
4807 | log_info("Restoring log level to original (%s).", strna(s)); | |
4808 | ||
4809 | log_set_max_level(m->original_log_level); | |
4810 | m->log_level_overridden = false; | |
4811 | } | |
4812 | ||
bda7d78b FB |
4813 | void manager_override_log_target(Manager *m, LogTarget target) { |
4814 | assert(m); | |
4815 | ||
4816 | if (!m->log_target_overridden) { | |
4817 | m->original_log_target = log_get_target(); | |
4818 | m->log_target_overridden = true; | |
4819 | } | |
4820 | ||
4821 | log_info("Setting log target to %s.", log_target_to_string(target)); | |
4822 | log_set_target(target); | |
4823 | } | |
4824 | ||
4825 | void manager_restore_original_log_target(Manager *m) { | |
4826 | assert(m); | |
4827 | ||
4828 | if (!m->log_target_overridden) | |
4829 | return; | |
4830 | ||
4831 | log_info("Restoring log target to original %s.", log_target_to_string(m->original_log_target)); | |
4832 | ||
4833 | log_set_target(m->original_log_target); | |
4834 | m->log_target_overridden = false; | |
4835 | } | |
4836 | ||
d4ee7bd8 YW |
4837 | ManagerTimestamp manager_timestamp_initrd_mangle(ManagerTimestamp s) { |
4838 | if (in_initrd() && | |
4839 | s >= MANAGER_TIMESTAMP_SECURITY_START && | |
4840 | s <= MANAGER_TIMESTAMP_UNITS_LOAD_FINISH) | |
4841 | return s - MANAGER_TIMESTAMP_SECURITY_START + MANAGER_TIMESTAMP_INITRD_SECURITY_START; | |
4842 | return s; | |
4843 | } | |
4844 | ||
f755e3b7 | 4845 | static const char *const manager_state_table[_MANAGER_STATE_MAX] = { |
d81afec1 | 4846 | [MANAGER_INITIALIZING] = "initializing", |
48d83e33 ZJS |
4847 | [MANAGER_STARTING] = "starting", |
4848 | [MANAGER_RUNNING] = "running", | |
4849 | [MANAGER_DEGRADED] = "degraded", | |
4850 | [MANAGER_MAINTENANCE] = "maintenance", | |
4851 | [MANAGER_STOPPING] = "stopping", | |
f755e3b7 LP |
4852 | }; |
4853 | ||
4854 | DEFINE_STRING_TABLE_LOOKUP(manager_state, ManagerState); | |
9f9f0342 LP |
4855 | |
4856 | static const char *const manager_timestamp_table[_MANAGER_TIMESTAMP_MAX] = { | |
48d83e33 ZJS |
4857 | [MANAGER_TIMESTAMP_FIRMWARE] = "firmware", |
4858 | [MANAGER_TIMESTAMP_LOADER] = "loader", | |
4859 | [MANAGER_TIMESTAMP_KERNEL] = "kernel", | |
4860 | [MANAGER_TIMESTAMP_INITRD] = "initrd", | |
4861 | [MANAGER_TIMESTAMP_USERSPACE] = "userspace", | |
4862 | [MANAGER_TIMESTAMP_FINISH] = "finish", | |
4863 | [MANAGER_TIMESTAMP_SECURITY_START] = "security-start", | |
4864 | [MANAGER_TIMESTAMP_SECURITY_FINISH] = "security-finish", | |
4865 | [MANAGER_TIMESTAMP_GENERATORS_START] = "generators-start", | |
4866 | [MANAGER_TIMESTAMP_GENERATORS_FINISH] = "generators-finish", | |
4867 | [MANAGER_TIMESTAMP_UNITS_LOAD_START] = "units-load-start", | |
4868 | [MANAGER_TIMESTAMP_UNITS_LOAD_FINISH] = "units-load-finish", | |
49fbe940 | 4869 | [MANAGER_TIMESTAMP_UNITS_LOAD] = "units-load", |
48d83e33 ZJS |
4870 | [MANAGER_TIMESTAMP_INITRD_SECURITY_START] = "initrd-security-start", |
4871 | [MANAGER_TIMESTAMP_INITRD_SECURITY_FINISH] = "initrd-security-finish", | |
4872 | [MANAGER_TIMESTAMP_INITRD_GENERATORS_START] = "initrd-generators-start", | |
d4ee7bd8 | 4873 | [MANAGER_TIMESTAMP_INITRD_GENERATORS_FINISH] = "initrd-generators-finish", |
48d83e33 | 4874 | [MANAGER_TIMESTAMP_INITRD_UNITS_LOAD_START] = "initrd-units-load-start", |
d4ee7bd8 | 4875 | [MANAGER_TIMESTAMP_INITRD_UNITS_LOAD_FINISH] = "initrd-units-load-finish", |
9f9f0342 LP |
4876 | }; |
4877 | ||
4878 | DEFINE_STRING_TABLE_LOOKUP(manager_timestamp, ManagerTimestamp); | |
afcfaa69 LP |
4879 | |
4880 | static const char* const oom_policy_table[_OOM_POLICY_MAX] = { | |
4881 | [OOM_CONTINUE] = "continue", | |
48d83e33 ZJS |
4882 | [OOM_STOP] = "stop", |
4883 | [OOM_KILL] = "kill", | |
afcfaa69 LP |
4884 | }; |
4885 | ||
4886 | DEFINE_STRING_TABLE_LOOKUP(oom_policy, OOMPolicy); |