]> git.ipfire.org Git - thirdparty/systemd.git/blob - src/cgroup-util.c
gcc: make gcc shut up
[thirdparty/systemd.git] / src / cgroup-util.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 <unistd.h>
24 #include <signal.h>
25 #include <string.h>
26 #include <stdlib.h>
27 #include <dirent.h>
28 #include <sys/stat.h>
29 #include <sys/types.h>
30
31 #include "cgroup-util.h"
32 #include "log.h"
33 #include "set.h"
34 #include "macro.h"
35 #include "util.h"
36
37 int cg_enumerate_processes(const char *controller, const char *path, FILE **_f) {
38 char *fs;
39 int r;
40 FILE *f;
41
42 assert(controller);
43 assert(path);
44 assert(_f);
45
46 if ((r = cg_get_path(controller, path, "cgroup.procs", &fs)) < 0)
47 return r;
48
49 f = fopen(fs, "re");
50 free(fs);
51
52 if (!f)
53 return -errno;
54
55 *_f = f;
56 return 0;
57 }
58
59 int cg_enumerate_tasks(const char *controller, const char *path, FILE **_f) {
60 char *fs;
61 int r;
62 FILE *f;
63
64 assert(controller);
65 assert(path);
66 assert(_f);
67
68 if ((r = cg_get_path(controller, path, "tasks", &fs)) < 0)
69 return r;
70
71 f = fopen(fs, "re");
72 free(fs);
73
74 if (!f)
75 return -errno;
76
77 *_f = f;
78 return 0;
79 }
80
81 int cg_read_pid(FILE *f, pid_t *_pid) {
82 unsigned long ul;
83
84 /* Note that the cgroup.procs might contain duplicates! See
85 * cgroups.txt for details. */
86
87 errno = 0;
88 if (fscanf(f, "%lu", &ul) != 1) {
89
90 if (feof(f))
91 return 0;
92
93 return errno ? -errno : -EIO;
94 }
95
96 if (ul <= 0)
97 return -EIO;
98
99 *_pid = (pid_t) ul;
100 return 1;
101 }
102
103 int cg_enumerate_subgroups(const char *controller, const char *path, DIR **_d) {
104 char *fs;
105 int r;
106 DIR *d;
107
108 assert(controller);
109 assert(path);
110 assert(_d);
111
112 /* This is not recursive! */
113
114 if ((r = cg_get_path(controller, path, NULL, &fs)) < 0)
115 return r;
116
117 d = opendir(fs);
118 free(fs);
119
120 if (!d)
121 return -errno;
122
123 *_d = d;
124 return 0;
125 }
126
127 int cg_read_subgroup(DIR *d, char **fn) {
128 struct dirent *de;
129
130 assert(d);
131
132 errno = 0;
133 while ((de = readdir(d))) {
134 char *b;
135
136 if (de->d_type != DT_DIR)
137 continue;
138
139 if (streq(de->d_name, ".") ||
140 streq(de->d_name, ".."))
141 continue;
142
143 if (!(b = strdup(de->d_name)))
144 return -ENOMEM;
145
146 *fn = b;
147 return 1;
148 }
149
150 if (errno)
151 return -errno;
152
153 return 0;
154 }
155
156 int cg_rmdir(const char *controller, const char *path) {
157 char *p;
158 int r;
159
160 if ((r = cg_get_path(controller, path, NULL, &p)) < 0)
161 return r;
162
163 r = rmdir(p);
164 free(p);
165
166 return r < 0 ? -errno : 0;
167 }
168
169 int cg_kill(const char *controller, const char *path, int sig, bool ignore_self, Set *s) {
170 bool done = false;
171 int r, ret = 0;
172 pid_t my_pid;
173 FILE *f = NULL;
174 Set *allocated_set = NULL;
175
176 assert(controller);
177 assert(path);
178 assert(sig >= 0);
179
180 /* This goes through the tasks list and kills them all. This
181 * is repeated until no further processes are added to the
182 * tasks list, to properly handle forking processes */
183
184 if (!s)
185 if (!(s = allocated_set = set_new(trivial_hash_func, trivial_compare_func)))
186 return -ENOMEM;
187
188 my_pid = getpid();
189
190 do {
191 pid_t pid = 0;
192 done = true;
193
194 if ((r = cg_enumerate_processes(controller, path, &f)) < 0) {
195 if (ret >= 0 && r != -ENOENT)
196 ret = r;
197
198 goto finish;
199 }
200
201 while ((r = cg_read_pid(f, &pid)) > 0) {
202
203 if (pid == my_pid && ignore_self)
204 continue;
205
206 if (set_get(s, LONG_TO_PTR(pid)) == LONG_TO_PTR(pid))
207 continue;
208
209 /* If we haven't killed this process yet, kill
210 * it */
211 if (kill(pid, sig) < 0) {
212 if (ret >= 0 && errno != ESRCH)
213 ret = -errno;
214 } else if (ret == 0)
215 ret = 1;
216
217 done = false;
218
219 if ((r = set_put(s, LONG_TO_PTR(pid))) < 0) {
220 if (ret >= 0)
221 ret = r;
222
223 goto finish;
224 }
225 }
226
227 if (r < 0) {
228 if (ret >= 0)
229 ret = r;
230
231 goto finish;
232 }
233
234 fclose(f);
235 f = NULL;
236
237 /* To avoid racing against processes which fork
238 * quicker than we can kill them we repeat this until
239 * no new pids need to be killed. */
240
241 } while (!done);
242
243 finish:
244 if (allocated_set)
245 set_free(allocated_set);
246
247 if (f)
248 fclose(f);
249
250 return ret;
251 }
252
253 int cg_kill_recursive(const char *controller, const char *path, int sig, bool ignore_self, bool rem, Set *s) {
254 int r, ret = 0;
255 DIR *d = NULL;
256 char *fn;
257 Set *allocated_set = NULL;
258
259 assert(path);
260 assert(controller);
261 assert(sig >= 0);
262
263 if (!s)
264 if (!(s = allocated_set = set_new(trivial_hash_func, trivial_compare_func)))
265 return -ENOMEM;
266
267 ret = cg_kill(controller, path, sig, ignore_self, s);
268
269 if ((r = cg_enumerate_subgroups(controller, path, &d)) < 0) {
270 if (ret >= 0 && r != -ENOENT)
271 ret = r;
272
273 goto finish;
274 }
275
276 while ((r = cg_read_subgroup(d, &fn)) > 0) {
277 char *p = NULL;
278
279 r = asprintf(&p, "%s/%s", path, fn);
280 free(fn);
281
282 if (r < 0) {
283 if (ret >= 0)
284 ret = -ENOMEM;
285
286 goto finish;
287 }
288
289 r = cg_kill_recursive(controller, p, sig, ignore_self, rem, s);
290 free(p);
291
292 if (r != 0 && ret >= 0)
293 ret = r;
294 }
295
296 if (r < 0 && ret >= 0)
297 ret = r;
298
299 if (rem)
300 if ((r = cg_rmdir(controller, path)) < 0) {
301 if (ret >= 0 &&
302 r != -ENOENT &&
303 r != -EBUSY)
304 ret = r;
305 }
306
307 finish:
308 if (d)
309 closedir(d);
310
311 if (allocated_set)
312 set_free(allocated_set);
313
314 return ret;
315 }
316
317 int cg_kill_recursive_and_wait(const char *controller, const char *path, bool rem) {
318 unsigned i;
319
320 assert(path);
321 assert(controller);
322
323 /* This safely kills all processes; first it sends a SIGTERM,
324 * then checks 8 times after 200ms whether the group is now
325 * empty, then kills everything that is left with SIGKILL and
326 * finally checks 5 times after 200ms each whether the group
327 * is finally empty. */
328
329 for (i = 0; i < 15; i++) {
330 int sig, r;
331
332 if (i <= 0)
333 sig = SIGTERM;
334 else if (i == 9)
335 sig = SIGKILL;
336 else
337 sig = 0;
338
339 if ((r = cg_kill_recursive(controller, path, sig, true, rem, NULL)) <= 0)
340 return r;
341
342 usleep(200 * USEC_PER_MSEC);
343 }
344
345 return 0;
346 }
347
348 int cg_migrate(const char *controller, const char *from, const char *to, bool ignore_self) {
349 bool done = false;
350 Set *s;
351 int r, ret = 0;
352 pid_t my_pid;
353 FILE *f = NULL;
354
355 assert(controller);
356 assert(from);
357 assert(to);
358
359 if (!(s = set_new(trivial_hash_func, trivial_compare_func)))
360 return -ENOMEM;
361
362 my_pid = getpid();
363
364 do {
365 pid_t pid = 0;
366 done = true;
367
368 if ((r = cg_enumerate_tasks(controller, from, &f)) < 0) {
369 if (ret >= 0 && r != -ENOENT)
370 ret = r;
371
372 goto finish;
373 }
374
375 while ((r = cg_read_pid(f, &pid)) > 0) {
376
377 /* This might do weird stuff if we aren't a
378 * single-threaded program. However, we
379 * luckily know we are not */
380 if (pid == my_pid && ignore_self)
381 continue;
382
383 if (set_get(s, LONG_TO_PTR(pid)) == LONG_TO_PTR(pid))
384 continue;
385
386 if ((r = cg_attach(controller, to, pid)) < 0) {
387 if (ret >= 0 && r != -ESRCH)
388 ret = r;
389 } else if (ret == 0)
390 ret = 1;
391
392 done = false;
393
394 if ((r = set_put(s, LONG_TO_PTR(pid))) < 0) {
395 if (ret >= 0)
396 ret = r;
397
398 goto finish;
399 }
400 }
401
402 if (r < 0) {
403 if (ret >= 0)
404 ret = r;
405
406 goto finish;
407 }
408
409 fclose(f);
410 f = NULL;
411
412 } while (!done);
413
414 finish:
415 set_free(s);
416
417 if (f)
418 fclose(f);
419
420 return ret;
421 }
422
423 int cg_migrate_recursive(const char *controller, const char *from, const char *to, bool ignore_self, bool rem) {
424 int r, ret = 0;
425 DIR *d = NULL;
426 char *fn;
427
428 assert(controller);
429 assert(from);
430 assert(to);
431
432 ret = cg_migrate(controller, from, to, ignore_self);
433
434 if ((r = cg_enumerate_subgroups(controller, from, &d)) < 0) {
435 if (ret >= 0 && r != -ENOENT)
436 ret = r;
437 goto finish;
438 }
439
440 while ((r = cg_read_subgroup(d, &fn)) > 0) {
441 char *p = NULL;
442
443 r = asprintf(&p, "%s/%s", from, fn);
444 free(fn);
445
446 if (r < 0) {
447 if (ret >= 0)
448 ret = -ENOMEM;
449
450 goto finish;
451 }
452
453 r = cg_migrate_recursive(controller, p, to, ignore_self, rem);
454 free(p);
455
456 if (r != 0 && ret >= 0)
457 ret = r;
458 }
459
460 if (r < 0 && ret >= 0)
461 ret = r;
462
463 if (rem)
464 if ((r = cg_rmdir(controller, from)) < 0) {
465 if (ret >= 0 &&
466 r != -ENOENT &&
467 r != -EBUSY)
468 ret = r;
469 }
470
471 finish:
472 if (d)
473 closedir(d);
474
475 return ret;
476 }
477
478 int cg_get_path(const char *controller, const char *path, const char *suffix, char **fs) {
479 const char *p;
480 char *mp;
481 int r;
482
483 assert(controller);
484 assert(fs);
485
486 /* This is a very minimal lookup from controller names to
487 * paths. Since we have mounted most hierarchies ourselves
488 * should be kinda safe, but eventually we might want to
489 * extend this to have a fallback to actually check
490 * /proc/mounts. Might need caching then. */
491
492 if (streq(controller, SYSTEMD_CGROUP_CONTROLLER))
493 p = "systemd";
494 else if (startswith(controller, "name="))
495 p = controller + 5;
496 else
497 p = controller;
498
499 if (asprintf(&mp, "/sys/fs/cgroup/%s", p) < 0)
500 return -ENOMEM;
501
502 if ((r = path_is_mount_point(mp)) <= 0) {
503 free(mp);
504 return r < 0 ? r : -ENOENT;
505 }
506
507 if (path && suffix)
508 r = asprintf(fs, "%s/%s/%s", mp, path, suffix);
509 else if (path)
510 r = asprintf(fs, "%s/%s", mp, path);
511 else if (suffix)
512 r = asprintf(fs, "%s/%s", mp, suffix);
513 else {
514 path_kill_slashes(mp);
515 *fs = mp;
516 return 0;
517 }
518
519 free(mp);
520 path_kill_slashes(*fs);
521 return r < 0 ? -ENOMEM : 0;
522 }
523
524 int cg_trim(const char *controller, const char *path, bool delete_root) {
525 char *fs;
526 int r;
527
528 assert(controller);
529 assert(path);
530
531 if ((r = cg_get_path(controller, path, NULL, &fs)) < 0)
532 return r;
533
534 r = rm_rf(fs, true, delete_root);
535 free(fs);
536
537 return r == -ENOENT ? 0 : r;
538 }
539
540 int cg_delete(const char *controller, const char *path) {
541 char *parent;
542 int r;
543
544 assert(controller);
545 assert(path);
546
547 if ((r = parent_of_path(path, &parent)) < 0)
548 return r;
549
550 r = cg_migrate_recursive(controller, path, parent, false, true);
551 free(parent);
552
553 return r == -ENOENT ? 0 : r;
554 }
555
556 int cg_create(const char *controller, const char *path) {
557 char *fs;
558 int r;
559
560 assert(controller);
561 assert(path);
562
563 if ((r = cg_get_path(controller, path, NULL, &fs)) < 0)
564 return r;
565
566 r = mkdir_parents(fs, 0755);
567
568 if (r >= 0) {
569 if (mkdir(fs, 0755) >= 0)
570 r = 1;
571 else if (errno == EEXIST)
572 r = 0;
573 else
574 r = -errno;
575 }
576
577 free(fs);
578
579 return r;
580 }
581
582 int cg_attach(const char *controller, const char *path, pid_t pid) {
583 char *fs;
584 int r;
585 char c[32];
586
587 assert(controller);
588 assert(path);
589 assert(pid >= 0);
590
591 if ((r = cg_get_path(controller, path, "tasks", &fs)) < 0)
592 return r;
593
594 if (pid == 0)
595 pid = getpid();
596
597 snprintf(c, sizeof(c), "%lu\n", (unsigned long) pid);
598 char_array_0(c);
599
600 r = write_one_line_file(fs, c);
601 free(fs);
602
603 return r;
604 }
605
606 int cg_create_and_attach(const char *controller, const char *path, pid_t pid) {
607 int r, q;
608
609 assert(controller);
610 assert(path);
611 assert(pid >= 0);
612
613 if ((r = cg_create(controller, path)) < 0)
614 return r;
615
616 if ((q = cg_attach(controller, path, pid)) < 0)
617 return q;
618
619 /* This does not remove the cgroup on failure */
620
621 return r;
622 }
623
624 int cg_set_group_access(const char *controller, const char *path, mode_t mode, uid_t uid, gid_t gid) {
625 char *fs;
626 int r;
627
628 assert(controller);
629 assert(path);
630
631 if ((r = cg_get_path(controller, path, NULL, &fs)) < 0)
632 return r;
633
634 r = chmod_and_chown(fs, mode, uid, gid);
635 free(fs);
636
637 return r;
638 }
639
640 int cg_set_task_access(const char *controller, const char *path, mode_t mode, uid_t uid, gid_t gid) {
641 char *fs;
642 int r;
643
644 assert(controller);
645 assert(path);
646
647 if ((r = cg_get_path(controller, path, "tasks", &fs)) < 0)
648 return r;
649
650 r = chmod_and_chown(fs, mode, uid, gid);
651 free(fs);
652
653 return r;
654 }
655
656 int cg_get_by_pid(const char *controller, pid_t pid, char **path) {
657 int r;
658 char *p = NULL;
659 FILE *f;
660 char *fs;
661 size_t cs;
662
663 assert(controller);
664 assert(path);
665 assert(pid >= 0);
666
667 if (pid == 0)
668 pid = getpid();
669
670 if (asprintf(&fs, "/proc/%lu/cgroup", (unsigned long) pid) < 0)
671 return -ENOMEM;
672
673 f = fopen(fs, "re");
674 free(fs);
675
676 if (!f)
677 return errno == ENOENT ? -ESRCH : -errno;
678
679 cs = strlen(controller);
680
681 while (!feof(f)) {
682 char line[LINE_MAX];
683 char *l;
684
685 errno = 0;
686 if (!(fgets(line, sizeof(line), f))) {
687 if (feof(f))
688 break;
689
690 r = errno ? -errno : -EIO;
691 goto finish;
692 }
693
694 truncate_nl(line);
695
696 if (!(l = strchr(line, ':')))
697 continue;
698
699 l++;
700 if (strncmp(l, controller, cs) != 0)
701 continue;
702
703 if (l[cs] != ':')
704 continue;
705
706 if (!(p = strdup(l + cs + 1))) {
707 r = -ENOMEM;
708 goto finish;
709 }
710
711 *path = p;
712 r = 0;
713 goto finish;
714 }
715
716 r = -ENOENT;
717
718 finish:
719 fclose(f);
720
721 return r;
722 }
723
724 int cg_install_release_agent(const char *controller, const char *agent) {
725 char *fs = NULL, *contents = NULL, *line = NULL, *sc;
726 int r;
727
728 assert(controller);
729 assert(agent);
730
731 if ((r = cg_get_path(controller, NULL, "release_agent", &fs)) < 0)
732 return r;
733
734 if ((r = read_one_line_file(fs, &contents)) < 0)
735 goto finish;
736
737 sc = strstrip(contents);
738 if (sc[0] == 0) {
739
740 if (asprintf(&line, "%s\n", agent) < 0) {
741 r = -ENOMEM;
742 goto finish;
743 }
744
745 if ((r = write_one_line_file(fs, line)) < 0)
746 goto finish;
747
748 } else if (!streq(sc, agent)) {
749 r = -EEXIST;
750 goto finish;
751 }
752
753 free(fs);
754 fs = NULL;
755 if ((r = cg_get_path(controller, NULL, "notify_on_release", &fs)) < 0)
756 goto finish;
757
758 free(contents);
759 contents = NULL;
760 if ((r = read_one_line_file(fs, &contents)) < 0)
761 goto finish;
762
763 sc = strstrip(contents);
764
765 if (streq(sc, "0")) {
766 if ((r = write_one_line_file(fs, "1\n")) < 0)
767 goto finish;
768
769 r = 1;
770 } else if (!streq(sc, "1")) {
771 r = -EIO;
772 goto finish;
773 } else
774 r = 0;
775
776 finish:
777 free(fs);
778 free(contents);
779 free(line);
780
781 return r;
782 }
783
784 int cg_is_empty(const char *controller, const char *path, bool ignore_self) {
785 pid_t pid = 0;
786 int r;
787 FILE *f = NULL;
788 bool found = false;
789
790 assert(controller);
791 assert(path);
792
793 if ((r = cg_enumerate_tasks(controller, path, &f)) < 0)
794 return r == -ENOENT ? 1 : r;
795
796 while ((r = cg_read_pid(f, &pid)) > 0) {
797
798 if (ignore_self && pid == getpid())
799 continue;
800
801 found = true;
802 break;
803 }
804
805 fclose(f);
806
807 if (r < 0)
808 return r;
809
810 return !found;
811 }
812
813 int cg_is_empty_recursive(const char *controller, const char *path, bool ignore_self) {
814 int r;
815 DIR *d = NULL;
816 char *fn;
817
818 assert(controller);
819 assert(path);
820
821 if ((r = cg_is_empty(controller, path, ignore_self)) <= 0)
822 return r;
823
824 if ((r = cg_enumerate_subgroups(controller, path, &d)) < 0)
825 return r == -ENOENT ? 1 : r;
826
827 while ((r = cg_read_subgroup(d, &fn)) > 0) {
828 char *p = NULL;
829
830 r = asprintf(&p, "%s/%s", path, fn);
831 free(fn);
832
833 if (r < 0) {
834 r = -ENOMEM;
835 goto finish;
836 }
837
838 r = cg_is_empty_recursive(controller, p, ignore_self);
839 free(p);
840
841 if (r <= 0)
842 goto finish;
843 }
844
845 if (r >= 0)
846 r = 1;
847
848 finish:
849
850 if (d)
851 closedir(d);
852
853 return r;
854 }
855
856 int cg_split_spec(const char *spec, char **controller, char **path) {
857 const char *e;
858 char *t = NULL, *u = NULL;
859
860 assert(spec);
861 assert(controller || path);
862
863 if (*spec == '/') {
864
865 if (path) {
866 if (!(t = strdup(spec)))
867 return -ENOMEM;
868
869 *path = t;
870 }
871
872 if (controller)
873 *controller = NULL;
874
875 return 0;
876 }
877
878 if (!(e = strchr(spec, ':'))) {
879
880 if (strchr(spec, '/') || spec[0] == 0)
881 return -EINVAL;
882
883 if (controller) {
884 if (!(t = strdup(spec)))
885 return -ENOMEM;
886
887 *controller = t;
888 }
889
890 if (path)
891 *path = NULL;
892
893 return 0;
894 }
895
896 if (e[1] != '/' ||
897 e == spec ||
898 memchr(spec, '/', e-spec))
899 return -EINVAL;
900
901 if (controller)
902 if (!(t = strndup(spec, e-spec)))
903 return -ENOMEM;
904
905 if (path)
906 if (!(u = strdup(e+1))) {
907 free(t);
908 return -ENOMEM;
909 }
910
911 if (controller)
912 *controller = t;
913
914 if (path)
915 *path = u;
916
917 return 0;
918 }
919
920 int cg_join_spec(const char *controller, const char *path, char **spec) {
921 assert(controller);
922 assert(path);
923
924 if (!path_is_absolute(path) ||
925 controller[0] == 0 ||
926 strchr(controller, ':') ||
927 strchr(controller, '/'))
928 return -EINVAL;
929
930 if (asprintf(spec, "%s:%s", controller, path) < 0)
931 return -ENOMEM;
932
933 return 0;
934 }
935
936 int cg_fix_path(const char *path, char **result) {
937 char *t, *c, *p;
938 int r;
939
940 assert(path);
941 assert(result);
942
943 /* First check if it already is a filesystem path */
944 if (path_is_absolute(path) &&
945 path_startswith(path, "/sys/fs/cgroup") &&
946 access(path, F_OK) >= 0) {
947
948 if (!(t = strdup(path)))
949 return -ENOMEM;
950
951 *result = t;
952 return 0;
953 }
954
955 /* Otherwise treat it as cg spec */
956 if ((r = cg_split_spec(path, &c, &p)) < 0)
957 return r;
958
959 r = cg_get_path(c ? c : SYSTEMD_CGROUP_CONTROLLER, p ? p : "/", NULL, result);
960 free(c);
961 free(p);
962
963 return r;
964 }