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 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)
219 if ((r = set_put(s, LONG_TO_PTR(pid))) < 0) {
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. */
245 set_free(allocated_set);
253 int cg_kill_recursive(const char *controller, const char *path, int sig, bool ignore_self, bool rem, Set *s) {
257 Set *allocated_set = NULL;
264 if (!(s = allocated_set = set_new(trivial_hash_func, trivial_compare_func)))
267 ret = cg_kill(controller, path, sig, ignore_self, s);
269 if ((r = cg_enumerate_subgroups(controller, path, &d)) < 0) {
270 if (ret >= 0 && r != -ENOENT)
276 while ((r = cg_read_subgroup(d, &fn)) > 0) {
279 r = asprintf(&p, "%s/%s", path, fn);
289 r = cg_kill_recursive(controller, p, sig, ignore_self, rem, s);
292 if (r != 0 && ret >= 0)
296 if (r < 0 && ret >= 0)
300 if ((r = cg_rmdir(controller, path)) < 0) {
312 set_free(allocated_set);
317 int cg_kill_recursive_and_wait(const char *controller, const char *path, bool rem) {
323 /* This safely kills all processes; first it sends a SIGTERM,
324 * then checks 8 times after 50ms whether the group is
325 * now empty, and finally kills everything that is left with
328 for (i = 0; i < 10; i++) {
338 if ((r = cg_kill_recursive(controller, path, sig, true, rem, NULL)) <= 0)
341 usleep(50 * USEC_PER_MSEC);
347 int cg_migrate(const char *controller, const char *from, const char *to, bool ignore_self) {
358 if (!(s = set_new(trivial_hash_func, trivial_compare_func)))
367 if ((r = cg_enumerate_tasks(controller, from, &f)) < 0) {
368 if (ret >= 0 && r != -ENOENT)
374 while ((r = cg_read_pid(f, &pid)) > 0) {
376 /* This might do weird stuff if we aren't a
377 * single-threaded program. However, we
378 * luckily know we are not */
379 if (pid == my_pid && ignore_self)
382 if (set_get(s, LONG_TO_PTR(pid)) == LONG_TO_PTR(pid))
385 if ((r = cg_attach(controller, to, pid)) < 0) {
386 if (ret >= 0 && r != -ESRCH)
393 if ((r = set_put(s, LONG_TO_PTR(pid))) < 0) {
422 int cg_migrate_recursive(const char *controller, const char *from, const char *to, bool ignore_self, bool rem) {
431 ret = cg_migrate(controller, from, to, ignore_self);
433 if ((r = cg_enumerate_subgroups(controller, from, &d)) < 0) {
434 if (ret >= 0 && r != -ENOENT)
439 while ((r = cg_read_subgroup(d, &fn)) > 0) {
442 r = asprintf(&p, "%s/%s", from, fn);
452 r = cg_migrate_recursive(controller, p, to, ignore_self, rem);
455 if (r != 0 && ret >= 0)
459 if (r < 0 && ret >= 0)
463 if ((r = cg_rmdir(controller, from)) < 0) {
477 int cg_get_path(const char *controller, const char *path, const char *suffix, char **fs) {
485 /* This is a very minimal lookup from controller names to
486 * paths. Since we have mounted most hierarchies ourselves
487 * should be kinda safe, but eventually we might want to
488 * extend this to have a fallback to actually check
489 * /proc/mounts. Might need caching then. */
491 if (streq(controller, SYSTEMD_CGROUP_CONTROLLER))
493 else if (startswith(controller, "name="))
498 if (asprintf(&mp, "/sys/fs/cgroup/%s", p) < 0)
501 if ((r = path_is_mount_point(mp)) <= 0) {
503 return r < 0 ? r : -ENOENT;
507 r = asprintf(fs, "%s/%s/%s", mp, path, suffix);
509 r = asprintf(fs, "%s/%s", mp, path);
511 r = asprintf(fs, "%s/%s", mp, suffix);
513 path_kill_slashes(mp);
519 path_kill_slashes(*fs);
520 return r < 0 ? -ENOMEM : 0;
523 int cg_trim(const char *controller, const char *path, bool delete_root) {
530 if ((r = cg_get_path(controller, path, NULL, &fs)) < 0)
533 r = rm_rf(fs, true, delete_root);
536 return r == -ENOENT ? 0 : r;
539 int cg_delete(const char *controller, const char *path) {
546 if ((r = parent_of_path(path, &parent)) < 0)
549 r = cg_migrate_recursive(controller, path, parent, false, true);
552 return r == -ENOENT ? 0 : r;
555 int cg_create(const char *controller, const char *path) {
562 if ((r = cg_get_path(controller, path, NULL, &fs)) < 0)
565 r = mkdir_parents(fs, 0755);
568 if (mkdir(fs, 0755) >= 0)
570 else if (errno == EEXIST)
581 int cg_attach(const char *controller, const char *path, pid_t pid) {
590 if ((r = cg_get_path(controller, path, "tasks", &fs)) < 0)
596 snprintf(c, sizeof(c), "%lu\n", (unsigned long) pid);
599 r = write_one_line_file(fs, c);
605 int cg_create_and_attach(const char *controller, const char *path, pid_t pid) {
612 if ((r = cg_create(controller, path)) < 0)
615 if ((q = cg_attach(controller, path, pid)) < 0)
618 /* This does not remove the cgroup on failure */
623 int cg_set_group_access(const char *controller, const char *path, mode_t mode, uid_t uid, gid_t gid) {
630 if ((r = cg_get_path(controller, path, NULL, &fs)) < 0)
633 r = chmod_and_chown(fs, mode, uid, gid);
639 int cg_set_task_access(const char *controller, const char *path, mode_t mode, uid_t uid, gid_t gid) {
646 if ((r = cg_get_path(controller, path, "tasks", &fs)) < 0)
649 r = chmod_and_chown(fs, mode, uid, gid);
655 int cg_get_by_pid(const char *controller, pid_t pid, char **path) {
669 if (asprintf(&fs, "/proc/%lu/cgroup", (unsigned long) pid) < 0)
676 return errno == ENOENT ? -ESRCH : -errno;
678 cs = strlen(controller);
685 if (!(fgets(line, sizeof(line), f))) {
689 r = errno ? -errno : -EIO;
695 if (!(l = strchr(line, ':')))
699 if (strncmp(l, controller, cs) != 0)
705 if (!(p = strdup(l + cs + 1))) {
723 int cg_install_release_agent(const char *controller, const char *agent) {
724 char *fs = NULL, *contents = NULL, *line = NULL, *sc;
730 if ((r = cg_get_path(controller, NULL, "release_agent", &fs)) < 0)
733 if ((r = read_one_line_file(fs, &contents)) < 0)
736 sc = strstrip(contents);
739 if (asprintf(&line, "%s\n", agent) < 0) {
744 if ((r = write_one_line_file(fs, line)) < 0)
747 } else if (!streq(sc, agent)) {
754 if ((r = cg_get_path(controller, NULL, "notify_on_release", &fs)) < 0)
759 if ((r = read_one_line_file(fs, &contents)) < 0)
762 sc = strstrip(contents);
764 if (streq(sc, "0")) {
765 if ((r = write_one_line_file(fs, "1\n")) < 0)
769 } else if (!streq(sc, "1")) {
783 int cg_is_empty(const char *controller, const char *path, bool ignore_self) {
792 if ((r = cg_enumerate_tasks(controller, path, &f)) < 0)
793 return r == -ENOENT ? 1 : r;
795 while ((r = cg_read_pid(f, &pid)) > 0) {
797 if (ignore_self && pid == getpid())
812 int cg_is_empty_recursive(const char *controller, const char *path, bool ignore_self) {
820 if ((r = cg_is_empty(controller, path, ignore_self)) <= 0)
823 if ((r = cg_enumerate_subgroups(controller, path, &d)) < 0)
824 return r == -ENOENT ? 1 : r;
826 while ((r = cg_read_subgroup(d, &fn)) > 0) {
829 r = asprintf(&p, "%s/%s", path, fn);
837 r = cg_is_empty_recursive(controller, p, ignore_self);
855 int cg_split_spec(const char *spec, char **controller, char **path) {
857 char *t = NULL, *u = NULL;
860 assert(controller || path);
865 if (!(t = strdup(spec)))
877 if (!(e = strchr(spec, ':'))) {
879 if (strchr(spec, '/') || spec[0] == 0)
883 if (!(t = strdup(spec)))
897 memchr(spec, '/', e-spec))
901 if (!(t = strndup(spec, e-spec)))
905 if (!(u = strdup(e+1))) {
919 int cg_join_spec(const char *controller, const char *path, char **spec) {
923 if (!path_is_absolute(path) ||
924 controller[0] == 0 ||
925 strchr(controller, ':') ||
926 strchr(controller, '/'))
929 if (asprintf(spec, "%s:%s", controller, path) < 0)
935 int cg_fix_path(const char *path, char **result) {
942 /* First check if it already is a filesystem path */
943 if (path_is_absolute(path) &&
944 path_startswith(path, "/sys/fs/cgroup") &&
945 access(path, F_OK) >= 0) {
947 if (!(t = strdup(path)))
954 /* Otherwise treat it as cg spec */
955 if ((r = cg_split_spec(path, &c, &p)) < 0)
958 r = cg_get_path(c ? c : SYSTEMD_CGROUP_CONTROLLER, p ? p : "/", NULL, result);