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, bool honour_sticky) {
160 r = cg_get_path(controller, path, NULL, &p);
167 /* If the sticky bit is set don't remove the directory */
169 tasks = strappend(p, "/tasks");
175 r = file_is_sticky(tasks);
187 return (r < 0 && errno != ENOENT) ? -errno : 0;
190 int cg_kill(const char *controller, const char *path, int sig, bool sigcont, bool ignore_self, Set *s) {
195 Set *allocated_set = NULL;
201 /* This goes through the tasks list and kills them all. This
202 * is repeated until no further processes are added to the
203 * tasks list, to properly handle forking processes */
206 if (!(s = allocated_set = set_new(trivial_hash_func, trivial_compare_func)))
215 if ((r = cg_enumerate_processes(controller, path, &f)) < 0) {
216 if (ret >= 0 && r != -ENOENT)
222 while ((r = cg_read_pid(f, &pid)) > 0) {
224 if (pid == my_pid && ignore_self)
227 if (set_get(s, LONG_TO_PTR(pid)) == LONG_TO_PTR(pid))
230 /* If we haven't killed this process yet, kill
232 if (kill(pid, sig) < 0) {
233 if (ret >= 0 && errno != ESRCH)
235 } else if (ret == 0) {
245 if ((r = set_put(s, LONG_TO_PTR(pid))) < 0) {
263 /* To avoid racing against processes which fork
264 * quicker than we can kill them we repeat this until
265 * no new pids need to be killed. */
271 set_free(allocated_set);
279 int cg_kill_recursive(const char *controller, const char *path, int sig, bool sigcont, bool ignore_self, bool rem, Set *s) {
283 Set *allocated_set = NULL;
290 if (!(s = allocated_set = set_new(trivial_hash_func, trivial_compare_func)))
293 ret = cg_kill(controller, path, sig, sigcont, ignore_self, s);
295 if ((r = cg_enumerate_subgroups(controller, path, &d)) < 0) {
296 if (ret >= 0 && r != -ENOENT)
302 while ((r = cg_read_subgroup(d, &fn)) > 0) {
305 r = asprintf(&p, "%s/%s", path, fn);
315 r = cg_kill_recursive(controller, p, sig, sigcont, ignore_self, rem, s);
318 if (r != 0 && ret >= 0)
322 if (r < 0 && ret >= 0)
326 if ((r = cg_rmdir(controller, path, true)) < 0) {
338 set_free(allocated_set);
343 int cg_kill_recursive_and_wait(const char *controller, const char *path, bool rem) {
349 /* This safely kills all processes; first it sends a SIGTERM,
350 * then checks 8 times after 200ms whether the group is now
351 * empty, then kills everything that is left with SIGKILL and
352 * finally checks 5 times after 200ms each whether the group
353 * is finally empty. */
355 for (i = 0; i < 15; i++) {
365 if ((r = cg_kill_recursive(controller, path, sig, true, true, rem, NULL)) <= 0)
368 usleep(200 * USEC_PER_MSEC);
374 int cg_migrate(const char *controller, const char *from, const char *to, bool ignore_self) {
385 if (!(s = set_new(trivial_hash_func, trivial_compare_func)))
394 if ((r = cg_enumerate_tasks(controller, from, &f)) < 0) {
395 if (ret >= 0 && r != -ENOENT)
401 while ((r = cg_read_pid(f, &pid)) > 0) {
403 /* This might do weird stuff if we aren't a
404 * single-threaded program. However, we
405 * luckily know we are not */
406 if (pid == my_pid && ignore_self)
409 if (set_get(s, LONG_TO_PTR(pid)) == LONG_TO_PTR(pid))
412 if ((r = cg_attach(controller, to, pid)) < 0) {
413 if (ret >= 0 && r != -ESRCH)
420 if ((r = set_put(s, LONG_TO_PTR(pid))) < 0) {
449 int cg_migrate_recursive(const char *controller, const char *from, const char *to, bool ignore_self, bool rem) {
458 ret = cg_migrate(controller, from, to, ignore_self);
460 if ((r = cg_enumerate_subgroups(controller, from, &d)) < 0) {
461 if (ret >= 0 && r != -ENOENT)
466 while ((r = cg_read_subgroup(d, &fn)) > 0) {
469 r = asprintf(&p, "%s/%s", from, fn);
479 r = cg_migrate_recursive(controller, p, to, ignore_self, rem);
482 if (r != 0 && ret >= 0)
486 if (r < 0 && ret >= 0)
490 if ((r = cg_rmdir(controller, from, true)) < 0) {
504 int cg_get_path(const char *controller, const char *path, const char *suffix, char **fs) {
507 static __thread bool good = false;
512 if (_unlikely_(!good)) {
515 r = path_is_mount_point("/sys/fs/cgroup");
517 return r < 0 ? r : -ENOENT;
519 /* Cache this to save a few stat()s */
523 if (isempty(controller))
526 /* This is a very minimal lookup from controller names to
527 * paths. Since we have mounted most hierarchies ourselves
528 * should be kinda safe, but eventually we might want to
529 * extend this to have a fallback to actually check
530 * /proc/mounts. Might need caching then. */
532 if (streq(controller, SYSTEMD_CGROUP_CONTROLLER))
534 else if (startswith(controller, "name="))
540 t = join("/sys/fs/cgroup/", p, "/", path, "/", suffix, NULL);
542 t = join("/sys/fs/cgroup/", p, "/", path, NULL);
544 t = join("/sys/fs/cgroup/", p, "/", suffix, NULL);
546 t = join("/sys/fs/cgroup/", p, NULL);
551 path_kill_slashes(t);
557 int cg_trim(const char *controller, const char *path, bool delete_root) {
564 if ((r = cg_get_path(controller, path, NULL, &fs)) < 0)
567 r = rm_rf(fs, true, delete_root, true);
570 return r == -ENOENT ? 0 : r;
573 int cg_delete(const char *controller, const char *path) {
580 if ((r = parent_of_path(path, &parent)) < 0)
583 r = cg_migrate_recursive(controller, path, parent, false, true);
586 return r == -ENOENT ? 0 : r;
589 int cg_create(const char *controller, const char *path) {
596 if ((r = cg_get_path(controller, path, NULL, &fs)) < 0)
599 r = mkdir_parents(fs, 0755);
602 if (mkdir(fs, 0755) >= 0)
604 else if (errno == EEXIST)
615 int cg_attach(const char *controller, const char *path, pid_t pid) {
624 if ((r = cg_get_path(controller, path, "tasks", &fs)) < 0)
630 snprintf(c, sizeof(c), "%lu\n", (unsigned long) pid);
633 r = write_one_line_file(fs, c);
639 int cg_create_and_attach(const char *controller, const char *path, pid_t pid) {
646 if ((r = cg_create(controller, path)) < 0)
649 if ((q = cg_attach(controller, path, pid)) < 0)
652 /* This does not remove the cgroup on failure */
657 int cg_set_group_access(const char *controller, const char *path, mode_t mode, uid_t uid, gid_t gid) {
664 if ((r = cg_get_path(controller, path, NULL, &fs)) < 0)
667 r = chmod_and_chown(fs, mode, uid, gid);
673 int cg_set_task_access(const char *controller, const char *path, mode_t mode, uid_t uid, gid_t gid) {
680 if ((r = cg_get_path(controller, path, "tasks", &fs)) < 0)
683 r = chmod_and_chown(fs, mode, uid, gid);
689 int cg_get_by_pid(const char *controller, pid_t pid, char **path) {
703 if (asprintf(&fs, "/proc/%lu/cgroup", (unsigned long) pid) < 0)
710 return errno == ENOENT ? -ESRCH : -errno;
712 cs = strlen(controller);
719 if (!(fgets(line, sizeof(line), f))) {
723 r = errno ? -errno : -EIO;
729 if (!(l = strchr(line, ':')))
733 if (strncmp(l, controller, cs) != 0)
739 if (!(p = strdup(l + cs + 1))) {
757 int cg_install_release_agent(const char *controller, const char *agent) {
758 char *fs = NULL, *contents = NULL, *line = NULL, *sc;
764 if ((r = cg_get_path(controller, NULL, "release_agent", &fs)) < 0)
767 if ((r = read_one_line_file(fs, &contents)) < 0)
770 sc = strstrip(contents);
773 if (asprintf(&line, "%s\n", agent) < 0) {
778 if ((r = write_one_line_file(fs, line)) < 0)
781 } else if (!streq(sc, agent)) {
788 if ((r = cg_get_path(controller, NULL, "notify_on_release", &fs)) < 0)
793 if ((r = read_one_line_file(fs, &contents)) < 0)
796 sc = strstrip(contents);
798 if (streq(sc, "0")) {
799 if ((r = write_one_line_file(fs, "1\n")) < 0)
803 } else if (!streq(sc, "1")) {
817 int cg_is_empty(const char *controller, const char *path, bool ignore_self) {
826 if ((r = cg_enumerate_tasks(controller, path, &f)) < 0)
827 return r == -ENOENT ? 1 : r;
829 while ((r = cg_read_pid(f, &pid)) > 0) {
831 if (ignore_self && pid == getpid())
846 int cg_is_empty_recursive(const char *controller, const char *path, bool ignore_self) {
854 if ((r = cg_is_empty(controller, path, ignore_self)) <= 0)
857 if ((r = cg_enumerate_subgroups(controller, path, &d)) < 0)
858 return r == -ENOENT ? 1 : r;
860 while ((r = cg_read_subgroup(d, &fn)) > 0) {
863 r = asprintf(&p, "%s/%s", path, fn);
871 r = cg_is_empty_recursive(controller, p, ignore_self);
889 int cg_split_spec(const char *spec, char **controller, char **path) {
891 char *t = NULL, *u = NULL;
894 assert(controller || path);
899 if (!(t = strdup(spec)))
911 if (!(e = strchr(spec, ':'))) {
913 if (strchr(spec, '/') || spec[0] == 0)
917 if (!(t = strdup(spec)))
931 memchr(spec, '/', e-spec))
935 if (!(t = strndup(spec, e-spec)))
939 if (!(u = strdup(e+1))) {
953 int cg_join_spec(const char *controller, const char *path, char **spec) {
957 if (!path_is_absolute(path) ||
958 controller[0] == 0 ||
959 strchr(controller, ':') ||
960 strchr(controller, '/'))
963 if (asprintf(spec, "%s:%s", controller, path) < 0)
969 int cg_fix_path(const char *path, char **result) {
976 /* First check if it already is a filesystem path */
977 if (path_is_absolute(path) &&
978 path_startswith(path, "/sys/fs/cgroup") &&
979 access(path, F_OK) >= 0) {
981 if (!(t = strdup(path)))
988 /* Otherwise treat it as cg spec */
989 if ((r = cg_split_spec(path, &c, &p)) < 0)
992 r = cg_get_path(c ? c : SYSTEMD_CGROUP_CONTROLLER, p ? p : "/", NULL, result);
999 int cg_get_user_path(char **path) {
1004 /* Figure out the place to put user cgroups below. We use the
1005 * same as PID 1 has but with the "/system" suffix replaced by
1008 if (cg_get_by_pid(SYSTEMD_CGROUP_CONTROLLER, 1, &root) < 0)
1009 p = strdup("/user");
1011 if (endswith(root, "/system"))
1012 root[strlen(root) - 7] = 0;
1013 else if (streq(root, "/"))
1016 p = strappend(root, "/user");