1 /*-*- Mode: C; c-basic-offset: 8; indent-tabs-mode: nil -*-*/
4 This file is part of systemd.
6 Copyright 2010 Lennart Poettering
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.
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.
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/>.
29 #include <sys/types.h>
31 #include "cgroup-util.h"
37 int cg_enumerate_processes(const char *controller, const char *path, FILE **_f) {
46 if ((r = cg_get_path(controller, path, "cgroup.procs", &fs)) < 0)
59 int cg_enumerate_tasks(const char *controller, const char *path, FILE **_f) {
68 if ((r = cg_get_path(controller, path, "tasks", &fs)) < 0)
81 int cg_read_pid(FILE *f, pid_t *_pid) {
84 /* Note that the cgroup.procs might contain duplicates! See
85 * cgroups.txt for details. */
88 if (fscanf(f, "%lu", &ul) != 1) {
93 return errno ? -errno : -EIO;
103 int cg_enumerate_subgroups(const char *controller, const char *path, DIR **_d) {
112 /* This is not recursive! */
114 if ((r = cg_get_path(controller, path, NULL, &fs)) < 0)
127 int cg_read_subgroup(DIR *d, char **fn) {
133 while ((de = readdir(d))) {
136 if (de->d_type != DT_DIR)
139 if (streq(de->d_name, ".") ||
140 streq(de->d_name, ".."))
143 if (!(b = strdup(de->d_name)))
156 int cg_rmdir(const char *controller, const char *path) {
160 if ((r = cg_get_path(controller, path, NULL, &p)) < 0)
166 return r < 0 ? -errno : 0;
169 int cg_kill(const char *controller, const char *path, int sig, bool sigcont, bool ignore_self, Set *s) {
174 Set *allocated_set = NULL;
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 */
185 if (!(s = allocated_set = set_new(trivial_hash_func, trivial_compare_func)))
194 if ((r = cg_enumerate_processes(controller, path, &f)) < 0) {
195 if (ret >= 0 && r != -ENOENT)
201 while ((r = cg_read_pid(f, &pid)) > 0) {
203 if (pid == my_pid && ignore_self)
206 if (set_get(s, LONG_TO_PTR(pid)) == LONG_TO_PTR(pid))
209 /* If we haven't killed this process yet, kill
211 if (kill(pid, sig) < 0) {
212 if (ret >= 0 && errno != ESRCH)
214 } else if (ret == 0) {
224 if ((r = set_put(s, LONG_TO_PTR(pid))) < 0) {
242 /* To avoid racing against processes which fork
243 * quicker than we can kill them we repeat this until
244 * no new pids need to be killed. */
250 set_free(allocated_set);
258 int cg_kill_recursive(const char *controller, const char *path, int sig, bool sigcont, bool ignore_self, bool rem, Set *s) {
262 Set *allocated_set = NULL;
269 if (!(s = allocated_set = set_new(trivial_hash_func, trivial_compare_func)))
272 ret = cg_kill(controller, path, sig, sigcont, ignore_self, s);
274 if ((r = cg_enumerate_subgroups(controller, path, &d)) < 0) {
275 if (ret >= 0 && r != -ENOENT)
281 while ((r = cg_read_subgroup(d, &fn)) > 0) {
284 r = asprintf(&p, "%s/%s", path, fn);
294 r = cg_kill_recursive(controller, p, sig, sigcont, ignore_self, rem, s);
297 if (r != 0 && ret >= 0)
301 if (r < 0 && ret >= 0)
305 if ((r = cg_rmdir(controller, path)) < 0) {
317 set_free(allocated_set);
322 int cg_kill_recursive_and_wait(const char *controller, const char *path, bool rem) {
328 /* This safely kills all processes; first it sends a SIGTERM,
329 * then checks 8 times after 200ms whether the group is now
330 * empty, then kills everything that is left with SIGKILL and
331 * finally checks 5 times after 200ms each whether the group
332 * is finally empty. */
334 for (i = 0; i < 15; i++) {
344 if ((r = cg_kill_recursive(controller, path, sig, true, true, rem, NULL)) <= 0)
347 usleep(200 * USEC_PER_MSEC);
353 int cg_migrate(const char *controller, const char *from, const char *to, bool ignore_self) {
364 if (!(s = set_new(trivial_hash_func, trivial_compare_func)))
373 if ((r = cg_enumerate_tasks(controller, from, &f)) < 0) {
374 if (ret >= 0 && r != -ENOENT)
380 while ((r = cg_read_pid(f, &pid)) > 0) {
382 /* This might do weird stuff if we aren't a
383 * single-threaded program. However, we
384 * luckily know we are not */
385 if (pid == my_pid && ignore_self)
388 if (set_get(s, LONG_TO_PTR(pid)) == LONG_TO_PTR(pid))
391 if ((r = cg_attach(controller, to, pid)) < 0) {
392 if (ret >= 0 && r != -ESRCH)
399 if ((r = set_put(s, LONG_TO_PTR(pid))) < 0) {
428 int cg_migrate_recursive(const char *controller, const char *from, const char *to, bool ignore_self, bool rem) {
437 ret = cg_migrate(controller, from, to, ignore_self);
439 if ((r = cg_enumerate_subgroups(controller, from, &d)) < 0) {
440 if (ret >= 0 && r != -ENOENT)
445 while ((r = cg_read_subgroup(d, &fn)) > 0) {
448 r = asprintf(&p, "%s/%s", from, fn);
458 r = cg_migrate_recursive(controller, p, to, ignore_self, rem);
461 if (r != 0 && ret >= 0)
465 if (r < 0 && ret >= 0)
469 if ((r = cg_rmdir(controller, from)) < 0) {
483 int cg_get_path(const char *controller, const char *path, const char *suffix, char **fs) {
487 static __thread bool good = false;
492 /* This is a very minimal lookup from controller names to
493 * paths. Since we have mounted most hierarchies ourselves
494 * should be kinda safe, but eventually we might want to
495 * extend this to have a fallback to actually check
496 * /proc/mounts. Might need caching then. */
498 if (streq(controller, SYSTEMD_CGROUP_CONTROLLER))
500 else if (startswith(controller, "name="))
505 if (asprintf(&mp, "/sys/fs/cgroup/%s", p) < 0)
509 if ((r = path_is_mount_point(mp)) <= 0) {
511 return r < 0 ? r : -ENOENT;
514 /* Cache this to save a few stat()s */
519 r = asprintf(fs, "%s/%s/%s", mp, path, suffix);
521 r = asprintf(fs, "%s/%s", mp, path);
523 r = asprintf(fs, "%s/%s", mp, suffix);
525 path_kill_slashes(mp);
531 path_kill_slashes(*fs);
532 return r < 0 ? -ENOMEM : 0;
535 int cg_trim(const char *controller, const char *path, bool delete_root) {
542 if ((r = cg_get_path(controller, path, NULL, &fs)) < 0)
545 r = rm_rf(fs, true, delete_root);
548 return r == -ENOENT ? 0 : r;
551 int cg_delete(const char *controller, const char *path) {
558 if ((r = parent_of_path(path, &parent)) < 0)
561 r = cg_migrate_recursive(controller, path, parent, false, true);
564 return r == -ENOENT ? 0 : r;
567 int cg_create(const char *controller, const char *path) {
574 if ((r = cg_get_path(controller, path, NULL, &fs)) < 0)
577 r = mkdir_parents(fs, 0755);
580 if (mkdir(fs, 0755) >= 0)
582 else if (errno == EEXIST)
593 int cg_attach(const char *controller, const char *path, pid_t pid) {
602 if ((r = cg_get_path(controller, path, "tasks", &fs)) < 0)
608 snprintf(c, sizeof(c), "%lu\n", (unsigned long) pid);
611 r = write_one_line_file(fs, c);
617 int cg_create_and_attach(const char *controller, const char *path, pid_t pid) {
624 if ((r = cg_create(controller, path)) < 0)
627 if ((q = cg_attach(controller, path, pid)) < 0)
630 /* This does not remove the cgroup on failure */
635 int cg_set_group_access(const char *controller, const char *path, mode_t mode, uid_t uid, gid_t gid) {
642 if ((r = cg_get_path(controller, path, NULL, &fs)) < 0)
645 r = chmod_and_chown(fs, mode, uid, gid);
651 int cg_set_task_access(const char *controller, const char *path, mode_t mode, uid_t uid, gid_t gid) {
658 if ((r = cg_get_path(controller, path, "tasks", &fs)) < 0)
661 r = chmod_and_chown(fs, mode, uid, gid);
667 int cg_get_by_pid(const char *controller, pid_t pid, char **path) {
681 if (asprintf(&fs, "/proc/%lu/cgroup", (unsigned long) pid) < 0)
688 return errno == ENOENT ? -ESRCH : -errno;
690 cs = strlen(controller);
697 if (!(fgets(line, sizeof(line), f))) {
701 r = errno ? -errno : -EIO;
707 if (!(l = strchr(line, ':')))
711 if (strncmp(l, controller, cs) != 0)
717 if (!(p = strdup(l + cs + 1))) {
735 int cg_install_release_agent(const char *controller, const char *agent) {
736 char *fs = NULL, *contents = NULL, *line = NULL, *sc;
742 if ((r = cg_get_path(controller, NULL, "release_agent", &fs)) < 0)
745 if ((r = read_one_line_file(fs, &contents)) < 0)
748 sc = strstrip(contents);
751 if (asprintf(&line, "%s\n", agent) < 0) {
756 if ((r = write_one_line_file(fs, line)) < 0)
759 } else if (!streq(sc, agent)) {
766 if ((r = cg_get_path(controller, NULL, "notify_on_release", &fs)) < 0)
771 if ((r = read_one_line_file(fs, &contents)) < 0)
774 sc = strstrip(contents);
776 if (streq(sc, "0")) {
777 if ((r = write_one_line_file(fs, "1\n")) < 0)
781 } else if (!streq(sc, "1")) {
795 int cg_is_empty(const char *controller, const char *path, bool ignore_self) {
804 if ((r = cg_enumerate_tasks(controller, path, &f)) < 0)
805 return r == -ENOENT ? 1 : r;
807 while ((r = cg_read_pid(f, &pid)) > 0) {
809 if (ignore_self && pid == getpid())
824 int cg_is_empty_recursive(const char *controller, const char *path, bool ignore_self) {
832 if ((r = cg_is_empty(controller, path, ignore_self)) <= 0)
835 if ((r = cg_enumerate_subgroups(controller, path, &d)) < 0)
836 return r == -ENOENT ? 1 : r;
838 while ((r = cg_read_subgroup(d, &fn)) > 0) {
841 r = asprintf(&p, "%s/%s", path, fn);
849 r = cg_is_empty_recursive(controller, p, ignore_self);
867 int cg_split_spec(const char *spec, char **controller, char **path) {
869 char *t = NULL, *u = NULL;
872 assert(controller || path);
877 if (!(t = strdup(spec)))
889 if (!(e = strchr(spec, ':'))) {
891 if (strchr(spec, '/') || spec[0] == 0)
895 if (!(t = strdup(spec)))
909 memchr(spec, '/', e-spec))
913 if (!(t = strndup(spec, e-spec)))
917 if (!(u = strdup(e+1))) {
931 int cg_join_spec(const char *controller, const char *path, char **spec) {
935 if (!path_is_absolute(path) ||
936 controller[0] == 0 ||
937 strchr(controller, ':') ||
938 strchr(controller, '/'))
941 if (asprintf(spec, "%s:%s", controller, path) < 0)
947 int cg_fix_path(const char *path, char **result) {
954 /* First check if it already is a filesystem path */
955 if (path_is_absolute(path) &&
956 path_startswith(path, "/sys/fs/cgroup") &&
957 access(path, F_OK) >= 0) {
959 if (!(t = strdup(path)))
966 /* Otherwise treat it as cg spec */
967 if ((r = cg_split_spec(path, &c, &p)) < 0)
970 r = cg_get_path(c ? c : SYSTEMD_CGROUP_CONTROLLER, p ? p : "/", NULL, result);
977 int cg_get_user_path(char **path) {
982 /* Figure out the place to put user cgroups below. We use the
983 * same as PID 1 has but with the "/system" suffix replaced by
986 if (cg_get_by_pid(SYSTEMD_CGROUP_CONTROLLER, 1, &root) < 0)
989 if (endswith(root, "/system"))
990 root[strlen(root) - 7] = 0;
991 else if (streq(root, "/"))
994 p = strappend(root, "/user");