]> git.ipfire.org Git - thirdparty/systemd.git/blob - src/cgroup.c
cgroup: optionally mount a specific cgroup controllers together, and add cpu+cpuacct...
[thirdparty/systemd.git] / src / cgroup.c
1 /*-*- Mode: C; c-basic-offset: 8; indent-tabs-mode: nil -*-*/
2
3 /***
4 This file is part of systemd.
5
6 Copyright 2010 Lennart Poettering
7
8 systemd is free software; you can redistribute it and/or modify it
9 under the terms of the GNU General Public License as published by
10 the Free Software Foundation; either version 2 of the License, or
11 (at your option) any later version.
12
13 systemd is distributed in the hope that it will be useful, but
14 WITHOUT ANY WARRANTY; without even the implied warranty of
15 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
16 General Public License for more details.
17
18 You should have received a copy of the GNU General Public License
19 along with systemd; If not, see <http://www.gnu.org/licenses/>.
20 ***/
21
22 #include <errno.h>
23 #include <assert.h>
24 #include <unistd.h>
25 #include <sys/types.h>
26 #include <signal.h>
27 #include <sys/mount.h>
28 #include <fcntl.h>
29
30 #include "cgroup.h"
31 #include "cgroup-util.h"
32 #include "log.h"
33
34 int cgroup_bonding_realize(CGroupBonding *b) {
35 int r;
36
37 assert(b);
38 assert(b->path);
39 assert(b->controller);
40
41 if (b->realized)
42 return 0;
43
44 r = cg_create(b->controller, b->path);
45 if (r < 0) {
46 log_warning("Failed to create cgroup %s:%s: %s", b->controller, b->path, strerror(-r));
47 return r;
48 }
49
50 b->realized = true;
51
52 return 0;
53 }
54
55 int cgroup_bonding_realize_list(CGroupBonding *first) {
56 CGroupBonding *b;
57 int r;
58
59 LIST_FOREACH(by_unit, b, first)
60 if ((r = cgroup_bonding_realize(b)) < 0 && b->essential)
61 return r;
62
63 return 0;
64 }
65
66 void cgroup_bonding_free(CGroupBonding *b, bool remove_or_trim) {
67 assert(b);
68
69 if (b->unit) {
70 CGroupBonding *f;
71
72 LIST_REMOVE(CGroupBonding, by_unit, b->unit->meta.cgroup_bondings, b);
73
74 if (streq(b->controller, SYSTEMD_CGROUP_CONTROLLER)) {
75 assert_se(f = hashmap_get(b->unit->meta.manager->cgroup_bondings, b->path));
76 LIST_REMOVE(CGroupBonding, by_path, f, b);
77
78 if (f)
79 hashmap_replace(b->unit->meta.manager->cgroup_bondings, b->path, f);
80 else
81 hashmap_remove(b->unit->meta.manager->cgroup_bondings, b->path);
82 }
83 }
84
85 if (b->realized && b->ours && remove_or_trim) {
86
87 if (cgroup_bonding_is_empty(b) > 0)
88 cg_delete(b->controller, b->path);
89 else
90 cg_trim(b->controller, b->path, false);
91 }
92
93 free(b->controller);
94 free(b->path);
95 free(b);
96 }
97
98 void cgroup_bonding_free_list(CGroupBonding *first, bool remove_or_trim) {
99 CGroupBonding *b, *n;
100
101 LIST_FOREACH_SAFE(by_unit, b, n, first)
102 cgroup_bonding_free(b, remove_or_trim);
103 }
104
105 void cgroup_bonding_trim(CGroupBonding *b, bool delete_root) {
106 assert(b);
107
108 if (b->realized && b->ours)
109 cg_trim(b->controller, b->path, delete_root);
110 }
111
112 void cgroup_bonding_trim_list(CGroupBonding *first, bool delete_root) {
113 CGroupBonding *b;
114
115 LIST_FOREACH(by_unit, b, first)
116 cgroup_bonding_trim(b, delete_root);
117 }
118
119 int cgroup_bonding_install(CGroupBonding *b, pid_t pid) {
120 int r;
121
122 assert(b);
123 assert(pid >= 0);
124
125 if ((r = cg_create_and_attach(b->controller, b->path, pid)) < 0)
126 return r;
127
128 b->realized = true;
129 return 0;
130 }
131
132 int cgroup_bonding_install_list(CGroupBonding *first, pid_t pid) {
133 CGroupBonding *b;
134 int r;
135
136 LIST_FOREACH(by_unit, b, first)
137 if ((r = cgroup_bonding_install(b, pid)) < 0 && b->essential)
138 return r;
139
140 return 0;
141 }
142
143 int cgroup_bonding_set_group_access(CGroupBonding *b, mode_t mode, uid_t uid, gid_t gid) {
144 assert(b);
145
146 if (!b->realized)
147 return -EINVAL;
148
149 return cg_set_group_access(b->controller, b->path, mode, uid, gid);
150 }
151
152 int cgroup_bonding_set_group_access_list(CGroupBonding *first, mode_t mode, uid_t uid, gid_t gid) {
153 CGroupBonding *b;
154 int r;
155
156 LIST_FOREACH(by_unit, b, first) {
157 r = cgroup_bonding_set_group_access(b, mode, uid, gid);
158 if (r < 0)
159 return r;
160 }
161
162 return 0;
163 }
164
165 int cgroup_bonding_set_task_access(CGroupBonding *b, mode_t mode, uid_t uid, gid_t gid) {
166 assert(b);
167
168 if (!b->realized)
169 return -EINVAL;
170
171 return cg_set_task_access(b->controller, b->path, mode, uid, gid);
172 }
173
174 int cgroup_bonding_set_task_access_list(CGroupBonding *first, mode_t mode, uid_t uid, gid_t gid) {
175 CGroupBonding *b;
176 int r;
177
178 LIST_FOREACH(by_unit, b, first) {
179 r = cgroup_bonding_set_task_access(b, mode, uid, gid);
180 if (r < 0)
181 return r;
182 }
183
184 return 0;
185 }
186
187 int cgroup_bonding_kill(CGroupBonding *b, int sig, bool sigcont, Set *s) {
188 assert(b);
189 assert(sig >= 0);
190
191 /* Don't kill cgroups that aren't ours */
192 if (!b->ours)
193 return 0;
194
195 return cg_kill_recursive(b->controller, b->path, sig, sigcont, true, false, s);
196 }
197
198 int cgroup_bonding_kill_list(CGroupBonding *first, int sig, bool sigcont, Set *s) {
199 CGroupBonding *b;
200 Set *allocated_set = NULL;
201 int ret = -EAGAIN, r;
202
203 if (!s)
204 if (!(s = allocated_set = set_new(trivial_hash_func, trivial_compare_func)))
205 return -ENOMEM;
206
207 LIST_FOREACH(by_unit, b, first) {
208 if ((r = cgroup_bonding_kill(b, sig, sigcont, s)) < 0) {
209 if (r == -EAGAIN || r == -ESRCH)
210 continue;
211
212 ret = r;
213 goto finish;
214 }
215
216 if (ret < 0 || r > 0)
217 ret = r;
218 }
219
220 finish:
221 if (allocated_set)
222 set_free(allocated_set);
223
224 return ret;
225 }
226
227 /* Returns 1 if the group is empty, 0 if it is not, -EAGAIN if we
228 * cannot know */
229 int cgroup_bonding_is_empty(CGroupBonding *b) {
230 int r;
231
232 assert(b);
233
234 if ((r = cg_is_empty_recursive(b->controller, b->path, true)) < 0)
235 return r;
236
237 /* If it is empty it is empty */
238 if (r > 0)
239 return 1;
240
241 /* It's not only us using this cgroup, so we just don't know */
242 return b->ours ? 0 : -EAGAIN;
243 }
244
245 int cgroup_bonding_is_empty_list(CGroupBonding *first) {
246 CGroupBonding *b;
247
248 LIST_FOREACH(by_unit, b, first) {
249 int r;
250
251 if ((r = cgroup_bonding_is_empty(b)) < 0) {
252 /* If this returned -EAGAIN, then we don't know if the
253 * group is empty, so let's see if another group can
254 * tell us */
255
256 if (r != -EAGAIN)
257 return r;
258 } else
259 return r;
260 }
261
262 return -EAGAIN;
263 }
264
265 int manager_setup_cgroup(Manager *m) {
266 char *current = NULL, *path = NULL;
267 int r;
268 char suffix[32];
269
270 assert(m);
271
272 /* 0. Be nice to Ingo Molnar #628004 */
273 if (path_is_mount_point("/sys/fs/cgroup/systemd", false) <= 0) {
274 log_warning("No control group support available, not creating root group.");
275 return 0;
276 }
277
278 /* 1. Determine hierarchy */
279 if ((r = cg_get_by_pid(SYSTEMD_CGROUP_CONTROLLER, 0, &current)) < 0) {
280 log_error("Cannot determine cgroup we are running in: %s", strerror(-r));
281 goto finish;
282 }
283
284 if (m->running_as == MANAGER_SYSTEM)
285 strcpy(suffix, "/system");
286 else {
287 snprintf(suffix, sizeof(suffix), "/systemd-%lu", (unsigned long) getpid());
288 char_array_0(suffix);
289 }
290
291 free(m->cgroup_hierarchy);
292 if (endswith(current, suffix)) {
293 /* We probably got reexecuted and can continue to use our root cgroup */
294 m->cgroup_hierarchy = current;
295 current = NULL;
296
297 } else {
298 /* We need a new root cgroup */
299 m->cgroup_hierarchy = NULL;
300 if (asprintf(&m->cgroup_hierarchy, "%s%s", streq(current, "/") ? "" : current, suffix) < 0) {
301 log_error("Out of memory");
302 r = -ENOMEM;
303 goto finish;
304 }
305 }
306
307 /* 2. Show data */
308 if ((r = cg_get_path(SYSTEMD_CGROUP_CONTROLLER, m->cgroup_hierarchy, NULL, &path)) < 0) {
309 log_error("Cannot find cgroup mount point: %s", strerror(-r));
310 goto finish;
311 }
312
313 log_debug("Using cgroup controller " SYSTEMD_CGROUP_CONTROLLER ". File system hierarchy is at %s.", path);
314
315 /* 3. Install agent */
316 if ((r = cg_install_release_agent(SYSTEMD_CGROUP_CONTROLLER, SYSTEMD_CGROUP_AGENT_PATH)) < 0)
317 log_warning("Failed to install release agent, ignoring: %s", strerror(-r));
318 else if (r > 0)
319 log_debug("Installed release agent.");
320 else
321 log_debug("Release agent already installed.");
322
323 /* 4. Realize the group */
324 if ((r = cg_create_and_attach(SYSTEMD_CGROUP_CONTROLLER, m->cgroup_hierarchy, 0)) < 0) {
325 log_error("Failed to create root cgroup hierarchy: %s", strerror(-r));
326 goto finish;
327 }
328
329 /* 5. And pin it, so that it cannot be unmounted */
330 if (m->pin_cgroupfs_fd >= 0)
331 close_nointr_nofail(m->pin_cgroupfs_fd);
332
333 if ((m->pin_cgroupfs_fd = open(path, O_RDONLY|O_CLOEXEC|O_DIRECTORY|O_NOCTTY|O_NONBLOCK)) < 0) {
334 log_error("Failed to open pin file: %m");
335 r = -errno;
336 goto finish;
337 }
338
339 log_debug("Created root group.");
340
341 finish:
342 free(current);
343 free(path);
344
345 return r;
346 }
347
348 void manager_shutdown_cgroup(Manager *m, bool delete) {
349 assert(m);
350
351 if (delete && m->cgroup_hierarchy)
352 cg_delete(SYSTEMD_CGROUP_CONTROLLER, m->cgroup_hierarchy);
353
354 if (m->pin_cgroupfs_fd >= 0) {
355 close_nointr_nofail(m->pin_cgroupfs_fd);
356 m->pin_cgroupfs_fd = -1;
357 }
358
359 free(m->cgroup_hierarchy);
360 m->cgroup_hierarchy = NULL;
361 }
362
363 int cgroup_notify_empty(Manager *m, const char *group) {
364 CGroupBonding *l, *b;
365
366 assert(m);
367 assert(group);
368
369 if (!(l = hashmap_get(m->cgroup_bondings, group)))
370 return 0;
371
372 LIST_FOREACH(by_path, b, l) {
373 int t;
374
375 if (!b->unit)
376 continue;
377
378 if ((t = cgroup_bonding_is_empty_list(b)) < 0) {
379
380 /* If we don't know, we don't know */
381 if (t != -EAGAIN)
382 log_warning("Failed to check whether cgroup is empty: %s", strerror(errno));
383
384 continue;
385 }
386
387 if (t > 0)
388 if (UNIT_VTABLE(b->unit)->cgroup_notify_empty)
389 UNIT_VTABLE(b->unit)->cgroup_notify_empty(b->unit);
390 }
391
392 return 0;
393 }
394
395 Unit* cgroup_unit_by_pid(Manager *m, pid_t pid) {
396 CGroupBonding *l, *b;
397 char *group = NULL;
398
399 assert(m);
400
401 if (pid <= 1)
402 return NULL;
403
404 if (cg_get_by_pid(SYSTEMD_CGROUP_CONTROLLER, pid, &group) < 0)
405 return NULL;
406
407 l = hashmap_get(m->cgroup_bondings, group);
408
409 if (!l) {
410 char *slash;
411
412 while ((slash = strrchr(group, '/'))) {
413 if (slash == group)
414 break;
415
416 *slash = 0;
417
418 if ((l = hashmap_get(m->cgroup_bondings, group)))
419 break;
420 }
421 }
422
423 free(group);
424
425 LIST_FOREACH(by_path, b, l) {
426
427 if (!b->unit)
428 continue;
429
430 if (b->ours)
431 return b->unit;
432 }
433
434 return NULL;
435 }
436
437 CGroupBonding *cgroup_bonding_find_list(CGroupBonding *first, const char *controller) {
438 CGroupBonding *b;
439
440 assert(controller);
441
442 LIST_FOREACH(by_unit, b, first)
443 if (streq(b->controller, controller))
444 return b;
445
446 return NULL;
447 }
448
449 char *cgroup_bonding_to_string(CGroupBonding *b) {
450 char *r;
451
452 assert(b);
453
454 if (asprintf(&r, "%s:%s", b->controller, b->path) < 0)
455 return NULL;
456
457 return r;
458 }
459
460 pid_t cgroup_bonding_search_main_pid(CGroupBonding *b) {
461 FILE *f;
462 pid_t pid = 0, npid, mypid;
463
464 assert(b);
465
466 if (!b->ours)
467 return 0;
468
469 if (cg_enumerate_processes(b->controller, b->path, &f) < 0)
470 return 0;
471
472 mypid = getpid();
473
474 while (cg_read_pid(f, &npid) > 0) {
475 pid_t ppid;
476
477 if (npid == pid)
478 continue;
479
480 /* Ignore processes that aren't our kids */
481 if (get_parent_of_pid(npid, &ppid) >= 0 && ppid != mypid)
482 continue;
483
484 if (pid != 0) {
485 /* Dang, there's more than one daemonized PID
486 in this group, so we don't know what process
487 is the main process. */
488 pid = 0;
489 break;
490 }
491
492 pid = npid;
493 }
494
495 fclose(f);
496
497 return pid;
498 }
499
500 pid_t cgroup_bonding_search_main_pid_list(CGroupBonding *first) {
501 CGroupBonding *b;
502 pid_t pid;
503
504 /* Try to find a main pid from this cgroup, but checking if
505 * there's only one PID in the cgroup and returning it. Later
506 * on we might want to add additional, smarter heuristics
507 * here. */
508
509 LIST_FOREACH(by_unit, b, first)
510 if ((pid = cgroup_bonding_search_main_pid(b)) != 0)
511 return pid;
512
513 return 0;
514
515 }