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 Lesser General Public License as published by
10 the Free Software Foundation; either version 2.1 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 Lesser General Public License for more details.
18 You should have received a copy of the GNU Lesser General Public License
19 along with systemd; If not, see <http://www.gnu.org/licenses/>.
29 #include <sys/types.h>
35 #include "formats-util.h"
36 #include "process-util.h"
37 #include "path-util.h"
38 #include "unit-name.h"
40 // #include "special.h"
42 #include "login-util.h"
43 #include "cgroup-util.h"
45 int cg_enumerate_processes(const char *controller, const char *path, FILE **_f) {
46 _cleanup_free_ char *fs = NULL;
52 r = cg_get_path(controller, path, "cgroup.procs", &fs);
64 int cg_read_pid(FILE *f, pid_t *_pid) {
67 /* Note that the cgroup.procs might contain duplicates! See
68 * cgroups.txt for details. */
74 if (fscanf(f, "%lu", &ul) != 1) {
79 return errno ? -errno : -EIO;
89 int cg_enumerate_subgroups(const char *controller, const char *path, DIR **_d) {
90 _cleanup_free_ char *fs = NULL;
96 /* This is not recursive! */
98 r = cg_get_path(controller, path, NULL, &fs);
110 int cg_read_subgroup(DIR *d, char **fn) {
116 FOREACH_DIRENT_ALL(de, d, return -errno) {
119 if (de->d_type != DT_DIR)
122 if (streq(de->d_name, ".") ||
123 streq(de->d_name, ".."))
126 b = strdup(de->d_name);
137 int cg_rmdir(const char *controller, const char *path) {
138 _cleanup_free_ char *p = NULL;
141 r = cg_get_path(controller, path, NULL, &p);
146 if (r < 0 && errno != ENOENT)
152 int cg_kill(const char *controller, const char *path, int sig, bool sigcont, bool ignore_self, Set *s) {
153 _cleanup_set_free_ Set *allocated_set = NULL;
160 /* This goes through the tasks list and kills them all. This
161 * is repeated until no further processes are added to the
162 * tasks list, to properly handle forking processes */
165 s = allocated_set = set_new(NULL);
173 _cleanup_fclose_ FILE *f = NULL;
177 r = cg_enumerate_processes(controller, path, &f);
179 if (ret >= 0 && r != -ENOENT)
185 while ((r = cg_read_pid(f, &pid)) > 0) {
187 if (ignore_self && pid == my_pid)
190 if (set_get(s, PID_TO_PTR(pid)) == PID_TO_PTR(pid))
193 /* If we haven't killed this process yet, kill
195 if (kill(pid, sig) < 0) {
196 if (ret >= 0 && errno != ESRCH)
199 if (sigcont && sig != SIGKILL)
200 (void) kill(pid, SIGCONT);
208 r = set_put(s, PID_TO_PTR(pid));
224 /* To avoid racing against processes which fork
225 * quicker than we can kill them we repeat this until
226 * no new pids need to be killed. */
233 int cg_kill_recursive(const char *controller, const char *path, int sig, bool sigcont, bool ignore_self, bool rem, Set *s) {
234 _cleanup_set_free_ Set *allocated_set = NULL;
235 _cleanup_closedir_ DIR *d = NULL;
243 s = allocated_set = set_new(NULL);
248 ret = cg_kill(controller, path, sig, sigcont, ignore_self, s);
250 r = cg_enumerate_subgroups(controller, path, &d);
252 if (ret >= 0 && r != -ENOENT)
258 while ((r = cg_read_subgroup(d, &fn)) > 0) {
259 _cleanup_free_ char *p = NULL;
261 p = strjoin(path, "/", fn, NULL);
266 r = cg_kill_recursive(controller, p, sig, sigcont, ignore_self, rem, s);
267 if (r != 0 && ret >= 0)
271 if (ret >= 0 && r < 0)
275 r = cg_rmdir(controller, path);
276 if (r < 0 && ret >= 0 && r != -ENOENT && r != -EBUSY)
283 int cg_migrate(const char *cfrom, const char *pfrom, const char *cto, const char *pto, bool ignore_self) {
285 _cleanup_set_free_ Set *s = NULL;
301 _cleanup_fclose_ FILE *f = NULL;
305 r = cg_enumerate_processes(cfrom, pfrom, &f);
307 if (ret >= 0 && r != -ENOENT)
313 while ((r = cg_read_pid(f, &pid)) > 0) {
315 /* This might do weird stuff if we aren't a
316 * single-threaded program. However, we
317 * luckily know we are not */
318 if (ignore_self && pid == my_pid)
321 if (set_get(s, PID_TO_PTR(pid)) == PID_TO_PTR(pid))
324 /* Ignore kernel threads. Since they can only
325 * exist in the root cgroup, we only check for
328 (isempty(pfrom) || path_equal(pfrom, "/")) &&
329 is_kernel_thread(pid) > 0)
332 r = cg_attach(cto, pto, pid);
334 if (ret >= 0 && r != -ESRCH)
341 r = set_put(s, PID_TO_PTR(pid));
361 int cg_migrate_recursive(
369 _cleanup_closedir_ DIR *d = NULL;
378 ret = cg_migrate(cfrom, pfrom, cto, pto, ignore_self);
380 r = cg_enumerate_subgroups(cfrom, pfrom, &d);
382 if (ret >= 0 && r != -ENOENT)
388 while ((r = cg_read_subgroup(d, &fn)) > 0) {
389 _cleanup_free_ char *p = NULL;
391 p = strjoin(pfrom, "/", fn, NULL);
396 r = cg_migrate_recursive(cfrom, p, cto, pto, ignore_self, rem);
397 if (r != 0 && ret >= 0)
401 if (r < 0 && ret >= 0)
405 r = cg_rmdir(cfrom, pfrom);
406 if (r < 0 && ret >= 0 && r != -ENOENT && r != -EBUSY)
413 int cg_migrate_recursive_fallback(
428 r = cg_migrate_recursive(cfrom, pfrom, cto, pto, ignore_self, rem);
430 char prefix[strlen(pto) + 1];
432 /* This didn't work? Then let's try all prefixes of the destination */
434 PATH_FOREACH_PREFIX(prefix, pto) {
437 q = cg_migrate_recursive(cfrom, pfrom, cto, prefix, ignore_self, rem);
446 static const char *controller_to_dirname(const char *controller) {
451 /* Converts a controller name to the directory name below
452 * /sys/fs/cgroup/ we want to mount it to. Effectively, this
453 * just cuts off the name= prefixed used for named
454 * hierarchies, if it is specified. */
456 e = startswith(controller, "name=");
463 static int join_path_legacy(const char *controller, const char *path, const char *suffix, char **fs) {
470 dn = controller_to_dirname(controller);
472 if (isempty(path) && isempty(suffix))
473 t = strappend("/sys/fs/cgroup/", dn);
474 else if (isempty(path))
475 t = strjoin("/sys/fs/cgroup/", dn, "/", suffix, NULL);
476 else if (isempty(suffix))
477 t = strjoin("/sys/fs/cgroup/", dn, "/", path, NULL);
479 t = strjoin("/sys/fs/cgroup/", dn, "/", path, "/", suffix, NULL);
487 static int join_path_unified(const char *path, const char *suffix, char **fs) {
492 if (isempty(path) && isempty(suffix))
493 t = strdup("/sys/fs/cgroup");
494 else if (isempty(path))
495 t = strappend("/sys/fs/cgroup/", suffix);
496 else if (isempty(suffix))
497 t = strappend("/sys/fs/cgroup/", path);
499 t = strjoin("/sys/fs/cgroup/", path, "/", suffix, NULL);
507 int cg_get_path(const char *controller, const char *path, const char *suffix, char **fs) {
515 /* If no controller is specified, we return the path
516 * *below* the controllers, without any prefix. */
518 if (!path && !suffix)
526 t = strjoin(path, "/", suffix, NULL);
530 *fs = path_kill_slashes(t);
534 if (!cg_controller_is_valid(controller))
537 unified = cg_unified();
542 r = join_path_unified(path, suffix, fs);
544 r = join_path_legacy(controller, path, suffix, fs);
548 path_kill_slashes(*fs);
552 static int controller_is_accessible(const char *controller) {
557 /* Checks whether a specific controller is accessible,
558 * i.e. its hierarchy mounted. In the unified hierarchy all
559 * controllers are considered accessible, except for the named
562 if (!cg_controller_is_valid(controller))
565 unified = cg_unified();
569 /* We don't support named hierarchies if we are using
570 * the unified hierarchy. */
572 if (streq(controller, ELOGIND_CGROUP_CONTROLLER))
575 if (startswith(controller, "name="))
581 dn = controller_to_dirname(controller);
582 cc = strjoina("/sys/fs/cgroup/", dn);
584 if (laccess(cc, F_OK) < 0)
591 int cg_get_path_and_check(const char *controller, const char *path, const char *suffix, char **fs) {
597 /* Check if the specified controller is actually accessible */
598 r = controller_is_accessible(controller);
602 return cg_get_path(controller, path, suffix, fs);
605 static int trim_cb(const char *path, const struct stat *sb, int typeflag, struct FTW *ftwbuf) {
610 if (typeflag != FTW_DP)
613 if (ftwbuf->level < 1)
620 int cg_trim(const char *controller, const char *path, bool delete_root) {
621 _cleanup_free_ char *fs = NULL;
626 r = cg_get_path(controller, path, NULL, &fs);
631 if (nftw(fs, trim_cb, 64, FTW_DEPTH|FTW_MOUNT|FTW_PHYS) != 0) {
641 if (rmdir(fs) < 0 && errno != ENOENT)
648 int cg_create(const char *controller, const char *path) {
649 _cleanup_free_ char *fs = NULL;
652 r = cg_get_path_and_check(controller, path, NULL, &fs);
656 r = mkdir_parents(fs, 0755);
660 if (mkdir(fs, 0755) < 0) {
671 int cg_create_and_attach(const char *controller, const char *path, pid_t pid) {
676 r = cg_create(controller, path);
680 q = cg_attach(controller, path, pid);
684 /* This does not remove the cgroup on failure */
688 int cg_attach(const char *controller, const char *path, pid_t pid) {
689 _cleanup_free_ char *fs = NULL;
690 char c[DECIMAL_STR_MAX(pid_t) + 2];
696 r = cg_get_path_and_check(controller, path, "cgroup.procs", &fs);
703 snprintf(c, sizeof(c), PID_FMT"\n", pid);
705 return write_string_file_no_create(fs, c);
708 int cg_attach_fallback(const char *controller, const char *path, pid_t pid) {
715 r = cg_attach(controller, path, pid);
717 char prefix[strlen(path) + 1];
719 /* This didn't work? Then let's try all prefixes of
722 PATH_FOREACH_PREFIX(prefix, path) {
725 q = cg_attach(controller, prefix, pid);
734 /// UNNEEDED by elogind
736 int cg_set_group_access(
737 const char *controller,
743 _cleanup_free_ char *fs = NULL;
746 if (mode == MODE_INVALID && uid == UID_INVALID && gid == GID_INVALID)
749 if (mode != MODE_INVALID)
752 r = cg_get_path(controller, path, NULL, &fs);
756 return chmod_and_chown(fs, mode, uid, gid);
759 int cg_set_task_access(
760 const char *controller,
766 _cleanup_free_ char *fs = NULL, *procs = NULL;
771 if (mode == MODE_INVALID && uid == UID_INVALID && gid == GID_INVALID)
774 if (mode != MODE_INVALID)
777 r = cg_get_path(controller, path, "cgroup.procs", &fs);
781 r = chmod_and_chown(fs, mode, uid, gid);
785 unified = cg_unified();
791 /* Compatibility, Always keep values for "tasks" in sync with
793 if (cg_get_path(controller, path, "tasks", &procs) >= 0)
794 (void) chmod_and_chown(procs, mode, uid, gid);
800 int cg_pid_get_path(const char *controller, pid_t pid, char **path) {
801 _cleanup_fclose_ FILE *f = NULL;
810 unified = cg_unified();
815 if (!cg_controller_is_valid(controller))
818 controller = ELOGIND_CGROUP_CONTROLLER;
820 cs = strlen(controller);
823 fs = procfs_file_alloca(pid, "cgroup");
826 return errno == ENOENT ? -ESRCH : -errno;
828 FOREACH_LINE(line, f, return -errno) {
834 e = startswith(line, "0:");
844 const char *word, *state;
847 l = strchr(line, ':');
857 FOREACH_WORD_SEPARATOR(word, k, l, ",", state) {
858 if (k == cs && memcmp(word, controller, cs) == 0) {
879 int cg_install_release_agent(const char *controller, const char *agent) {
880 _cleanup_free_ char *fs = NULL, *contents = NULL;
886 unified = cg_unified();
889 if (unified) /* doesn't apply to unified hierarchy */
892 r = cg_get_path(controller, NULL, "release_agent", &fs);
896 r = read_one_line_file(fs, &contents);
900 sc = strstrip(contents);
902 r = write_string_file_no_create(fs, agent);
905 } else if (!path_equal(sc, agent))
909 r = cg_get_path(controller, NULL, "notify_on_release", &fs);
913 contents = mfree(contents);
914 r = read_one_line_file(fs, &contents);
918 sc = strstrip(contents);
919 if (streq(sc, "0")) {
920 r = write_string_file_no_create(fs, "1");
933 int cg_uninstall_release_agent(const char *controller) {
934 _cleanup_free_ char *fs = NULL;
937 unified = cg_unified();
940 if (unified) /* Doesn't apply to unified hierarchy */
943 r = cg_get_path(controller, NULL, "notify_on_release", &fs);
947 r = write_string_file_no_create(fs, "0");
953 r = cg_get_path(controller, NULL, "release_agent", &fs);
957 r = write_string_file_no_create(fs, "");
964 int cg_is_empty(const char *controller, const char *path) {
965 _cleanup_fclose_ FILE *f = NULL;
971 r = cg_enumerate_processes(controller, path, &f);
977 r = cg_read_pid(f, &pid);
984 int cg_is_empty_recursive(const char *controller, const char *path) {
989 /* The root cgroup is always populated */
990 if (controller && (isempty(path) || path_equal(path, "/")))
993 unified = cg_unified();
998 _cleanup_free_ char *populated = NULL, *t = NULL;
1000 /* On the unified hierarchy we can check empty state
1001 * via the "cgroup.populated" attribute. */
1003 r = cg_get_path(controller, path, "cgroup.populated", &populated);
1007 r = read_one_line_file(populated, &t);
1013 return streq(t, "0");
1015 _cleanup_closedir_ DIR *d = NULL;
1018 r = cg_is_empty(controller, path);
1022 r = cg_enumerate_subgroups(controller, path, &d);
1028 while ((r = cg_read_subgroup(d, &fn)) > 0) {
1029 _cleanup_free_ char *p = NULL;
1031 p = strjoin(path, "/", fn, NULL);
1036 r = cg_is_empty_recursive(controller, p);
1047 int cg_split_spec(const char *spec, char **controller, char **path) {
1048 char *t = NULL, *u = NULL;
1054 if (!path_is_safe(spec))
1062 *path = path_kill_slashes(t);
1071 e = strchr(spec, ':');
1073 if (!cg_controller_is_valid(spec))
1090 t = strndup(spec, e-spec);
1093 if (!cg_controller_is_valid(t)) {
1107 if (!path_is_safe(u) ||
1108 !path_is_absolute(u)) {
1114 path_kill_slashes(u);
1130 int cg_mangle_path(const char *path, char **result) {
1131 _cleanup_free_ char *c = NULL, *p = NULL;
1138 /* First, check if it already is a filesystem path */
1139 if (path_startswith(path, "/sys/fs/cgroup")) {
1145 *result = path_kill_slashes(t);
1149 /* Otherwise, treat it as cg spec */
1150 r = cg_split_spec(path, &c, &p);
1154 return cg_get_path(c ? c : ELOGIND_CGROUP_CONTROLLER, p ? p : "/", NULL, result);
1157 int cg_get_root_path(char **path) {
1163 r = cg_pid_get_path(ELOGIND_CGROUP_CONTROLLER, 1, &p);
1167 /// elogind does not support systemd scopes and slices
1169 e = endswith(p, "/" SPECIAL_INIT_SCOPE);
1171 e = endswith(p, "/" SPECIAL_SYSTEM_SLICE); /* legacy */
1173 e = endswith(p, "/system"); /* even more legacy */
1182 int cg_shift_path(const char *cgroup, const char *root, const char **shifted) {
1183 _cleanup_free_ char *rt = NULL;
1191 /* If the root was specified let's use that, otherwise
1192 * let's determine it from PID 1 */
1194 r = cg_get_root_path(&rt);
1201 p = path_startswith(cgroup, root);
1202 if (p && p > cgroup)
1210 int cg_pid_get_path_shifted(pid_t pid, const char *root, char **cgroup) {
1211 _cleanup_free_ char *raw = NULL;
1218 r = cg_pid_get_path(ELOGIND_CGROUP_CONTROLLER, pid, &raw);
1222 r = cg_shift_path(raw, root, &c);
1242 int cg_path_decode_unit(const char *cgroup, char **unit){
1249 n = strcspn(cgroup, "/");
1253 c = strndupa(cgroup, n);
1256 if (!unit_name_is_valid(c, UNIT_NAME_PLAIN|UNIT_NAME_INSTANCE))
1267 static bool valid_slice_name(const char *p, size_t n) {
1272 if (n < strlen("x.slice"))
1275 if (memcmp(p + n - 6, ".slice", 6) == 0) {
1281 c = cg_unescape(buf);
1283 return unit_name_is_valid(c, UNIT_NAME_PLAIN);
1289 static const char *skip_slices(const char *p) {
1292 /* Skips over all slice assignments */
1297 p += strspn(p, "/");
1299 n = strcspn(p, "/");
1300 if (!valid_slice_name(p, n))
1307 int cg_path_get_unit(const char *path, char **ret) {
1315 e = skip_slices(path);
1317 r = cg_path_decode_unit(e, &unit);
1321 /* We skipped over the slices, don't accept any now */
1322 if (endswith(unit, ".slice")) {
1331 int cg_pid_get_unit(pid_t pid, char **unit) {
1332 _cleanup_free_ char *cgroup = NULL;
1337 r = cg_pid_get_path_shifted(pid, NULL, &cgroup);
1341 return cg_path_get_unit(cgroup, unit);
1345 * Skip session-*.scope, but require it to be there.
1347 static const char *skip_session(const char *p) {
1353 p += strspn(p, "/");
1355 n = strcspn(p, "/");
1356 if (n < strlen("session-x.scope"))
1359 if (memcmp(p, "session-", 8) == 0 && memcmp(p + n - 6, ".scope", 6) == 0) {
1360 char buf[n - 8 - 6 + 1];
1362 memcpy(buf, p + 8, n - 8 - 6);
1365 /* Note that session scopes never need unescaping,
1366 * since they cannot conflict with the kernel's own
1367 * names, hence we don't need to call cg_unescape()
1370 if (!session_id_valid(buf))
1374 p += strspn(p, "/");
1382 * Skip user@*.service, but require it to be there.
1384 static const char *skip_user_manager(const char *p) {
1390 p += strspn(p, "/");
1392 n = strcspn(p, "/");
1393 if (n < strlen("user@x.service"))
1396 if (memcmp(p, "user@", 5) == 0 && memcmp(p + n - 8, ".service", 8) == 0) {
1397 char buf[n - 5 - 8 + 1];
1399 memcpy(buf, p + 5, n - 5 - 8);
1402 /* Note that user manager services never need unescaping,
1403 * since they cannot conflict with the kernel's own
1404 * names, hence we don't need to call cg_unescape()
1407 if (parse_uid(buf, NULL) < 0)
1411 p += strspn(p, "/");
1419 static const char *skip_user_prefix(const char *path) {
1424 /* Skip slices, if there are any */
1425 e = skip_slices(path);
1427 /* Skip the user manager, if it's in the path now... */
1428 t = skip_user_manager(e);
1432 /* Alternatively skip the user session if it is in the path... */
1433 return skip_session(e);
1436 int cg_path_get_user_unit(const char *path, char **ret) {
1442 t = skip_user_prefix(path);
1446 /* And from here on it looks pretty much the same as for a
1447 * system unit, hence let's use the same parser from here
1449 return cg_path_get_unit(t, ret);
1452 int cg_pid_get_user_unit(pid_t pid, char **unit) {
1453 _cleanup_free_ char *cgroup = NULL;
1458 r = cg_pid_get_path_shifted(pid, NULL, &cgroup);
1462 return cg_path_get_user_unit(cgroup, unit);
1465 int cg_path_get_machine_name(const char *path, char **machine) {
1466 _cleanup_free_ char *u = NULL;
1470 r = cg_path_get_unit(path, &u);
1474 sl = strjoina("/run/systemd/machines/unit:", u);
1475 return readlink_malloc(sl, machine);
1478 int cg_pid_get_machine_name(pid_t pid, char **machine) {
1479 _cleanup_free_ char *cgroup = NULL;
1484 r = cg_pid_get_path_shifted(pid, NULL, &cgroup);
1488 return cg_path_get_machine_name(cgroup, machine);
1491 int cg_path_get_session(const char *path, char **session) {
1492 _cleanup_free_ char *unit = NULL;
1498 r = cg_path_get_unit(path, &unit);
1502 start = startswith(unit, "session-");
1505 end = endswith(start, ".scope");
1510 if (!session_id_valid(start))
1526 int cg_pid_get_session(pid_t pid, char **session) {
1527 _cleanup_free_ char *cgroup = NULL;
1530 r = cg_pid_get_path_shifted(pid, NULL, &cgroup);
1534 return cg_path_get_session(cgroup, session);
1537 int cg_path_get_owner_uid(const char *path, uid_t *uid) {
1538 _cleanup_free_ char *slice = NULL;
1544 r = cg_path_get_slice(path, &slice);
1548 start = startswith(slice, "user-");
1551 end = endswith(start, ".slice");
1556 if (parse_uid(start, uid) < 0)
1562 int cg_pid_get_owner_uid(pid_t pid, uid_t *uid) {
1563 _cleanup_free_ char *cgroup = NULL;
1566 r = cg_pid_get_path_shifted(pid, NULL, &cgroup);
1570 return cg_path_get_owner_uid(cgroup, uid);
1573 int cg_path_get_slice(const char *p, char **slice) {
1574 const char *e = NULL;
1579 /* Finds the right-most slice unit from the beginning, but
1580 * stops before we come to the first non-slice unit. */
1585 p += strspn(p, "/");
1587 n = strcspn(p, "/");
1588 if (!valid_slice_name(p, n)) {
1593 s = strdup("-.slice");
1601 return cg_path_decode_unit(e, slice);
1609 int cg_pid_get_slice(pid_t pid, char **slice) {
1610 _cleanup_free_ char *cgroup = NULL;
1615 r = cg_pid_get_path_shifted(pid, NULL, &cgroup);
1619 return cg_path_get_slice(cgroup, slice);
1622 int cg_path_get_user_slice(const char *p, char **slice) {
1627 t = skip_user_prefix(p);
1631 /* And now it looks pretty much the same as for a system
1632 * slice, so let's just use the same parser from here on. */
1633 return cg_path_get_slice(t, slice);
1636 int cg_pid_get_user_slice(pid_t pid, char **slice) {
1637 _cleanup_free_ char *cgroup = NULL;
1642 r = cg_pid_get_path_shifted(pid, NULL, &cgroup);
1646 return cg_path_get_user_slice(cgroup, slice);
1649 char *cg_escape(const char *p) {
1650 bool need_prefix = false;
1652 /* This implements very minimal escaping for names to be used
1653 * as file names in the cgroup tree: any name which might
1654 * conflict with a kernel name or is prefixed with '_' is
1655 * prefixed with a '_'. That way, when reading cgroup names it
1656 * is sufficient to remove a single prefixing underscore if
1659 /* The return value of this function (unlike cg_unescape())
1665 streq(p, "notify_on_release") ||
1666 streq(p, "release_agent") ||
1667 streq(p, "tasks") ||
1668 startswith(p, "cgroup."))
1673 dot = strrchr(p, '.');
1678 for (c = 0; c < _CGROUP_CONTROLLER_MAX; c++) {
1681 n = cgroup_controller_to_string(c);
1686 if (memcmp(p, n, l) != 0)
1696 return strappend("_", p);
1701 char *cg_unescape(const char *p) {
1704 /* The return value of this function (unlike cg_escape())
1705 * doesn't need free()! */
1713 #define CONTROLLER_VALID \
1717 bool cg_controller_is_valid(const char *p) {
1723 s = startswith(p, "name=");
1727 if (*p == 0 || *p == '_')
1730 for (t = p; *t; t++)
1731 if (!strchr(CONTROLLER_VALID, *t))
1734 if (t - p > FILENAME_MAX)
1740 /// UNNEEDED by elogind
1742 int cg_slice_to_path(const char *unit, char **ret) {
1743 _cleanup_free_ char *p = NULL, *s = NULL, *e = NULL;
1750 if (streq(unit, "-.slice")) {
1760 if (!unit_name_is_valid(unit, UNIT_NAME_PLAIN))
1763 if (!endswith(unit, ".slice"))
1766 r = unit_name_to_prefix(unit, &p);
1770 dash = strchr(p, '-');
1772 /* Don't allow initial dashes */
1777 _cleanup_free_ char *escaped = NULL;
1778 char n[dash - p + sizeof(".slice")];
1780 /* Don't allow trailing or double dashes */
1781 if (dash[1] == 0 || dash[1] == '-')
1784 strcpy(stpncpy(n, p, dash - p), ".slice");
1785 if (!unit_name_is_valid(n, UNIT_NAME_PLAIN))
1788 escaped = cg_escape(n);
1792 if (!strextend(&s, escaped, "/", NULL))
1795 dash = strchr(dash+1, '-');
1798 e = cg_escape(unit);
1802 if (!strextend(&s, e, NULL))
1812 int cg_set_attribute(const char *controller, const char *path, const char *attribute, const char *value) {
1813 _cleanup_free_ char *p = NULL;
1816 r = cg_get_path(controller, path, attribute, &p);
1820 return write_string_file_no_create(p, value);
1823 /// UNNEEDED by elogind
1825 int cg_get_attribute(const char *controller, const char *path, const char *attribute, char **ret) {
1826 _cleanup_free_ char *p = NULL;
1829 r = cg_get_path(controller, path, attribute, &p);
1833 return read_one_line_file(p, ret);
1836 int cg_create_everywhere(CGroupMask supported, CGroupMask mask, const char *path) {
1840 /* This one will create a cgroup in our private tree, but also
1841 * duplicate it in the trees specified in mask, and remove it
1844 /* First create the cgroup in our own hierarchy. */
1845 r = cg_create(ELOGIND_CGROUP_CONTROLLER, path);
1849 /* If we are in the unified hierarchy, we are done now */
1850 unified = cg_unified();
1856 /* Otherwise, do the same in the other hierarchies */
1857 for (c = 0; c < _CGROUP_CONTROLLER_MAX; c++) {
1858 CGroupMask bit = CGROUP_CONTROLLER_TO_MASK(c);
1861 n = cgroup_controller_to_string(c);
1864 (void) cg_create(n, path);
1865 else if (supported & bit)
1866 (void) cg_trim(n, path, true);
1873 int cg_attach_everywhere(CGroupMask supported, const char *path, pid_t pid, cg_migrate_callback_t path_callback, void *userdata) {
1877 r = cg_attach(ELOGIND_CGROUP_CONTROLLER, path, pid);
1881 unified = cg_unified();
1887 for (c = 0; c < _CGROUP_CONTROLLER_MAX; c++) {
1888 CGroupMask bit = CGROUP_CONTROLLER_TO_MASK(c);
1889 const char *p = NULL;
1891 if (!(supported & bit))
1895 p = path_callback(bit, userdata);
1900 (void) cg_attach_fallback(cgroup_controller_to_string(c), p, pid);
1906 /// UNNEEDED by elogind
1908 int cg_attach_many_everywhere(CGroupMask supported, const char *path, Set* pids, cg_migrate_callback_t path_callback, void *userdata) {
1913 SET_FOREACH(pidp, pids, i) {
1914 pid_t pid = PTR_TO_PID(pidp);
1917 q = cg_attach_everywhere(supported, path, pid, path_callback, userdata);
1918 if (q < 0 && r >= 0)
1925 int cg_migrate_everywhere(CGroupMask supported, const char *from, const char *to, cg_migrate_callback_t to_callback, void *userdata) {
1929 if (!path_equal(from, to)) {
1930 r = cg_migrate_recursive(ELOGIND_CGROUP_CONTROLLER, from, ELOGIND_CGROUP_CONTROLLER, to, false, true);
1935 unified = cg_unified();
1941 for (c = 0; c < _CGROUP_CONTROLLER_MAX; c++) {
1942 CGroupMask bit = CGROUP_CONTROLLER_TO_MASK(c);
1943 const char *p = NULL;
1945 if (!(supported & bit))
1949 p = to_callback(bit, userdata);
1954 (void) cg_migrate_recursive_fallback(ELOGIND_CGROUP_CONTROLLER, to, cgroup_controller_to_string(c), p, false, false);
1960 int cg_trim_everywhere(CGroupMask supported, const char *path, bool delete_root) {
1964 r = cg_trim(ELOGIND_CGROUP_CONTROLLER, path, delete_root);
1968 unified = cg_unified();
1974 for (c = 0; c < _CGROUP_CONTROLLER_MAX; c++) {
1975 CGroupMask bit = CGROUP_CONTROLLER_TO_MASK(c);
1977 if (!(supported & bit))
1980 (void) cg_trim(cgroup_controller_to_string(c), path, delete_root);
1987 int cg_mask_supported(CGroupMask *ret) {
1988 CGroupMask mask = 0;
1991 /* Determines the mask of supported cgroup controllers. Only
1992 * includes controllers we can make sense of and that are
1993 * actually accessible. */
1995 unified = cg_unified();
1999 _cleanup_free_ char *root = NULL, *controllers = NULL, *path = NULL;
2002 /* In the unified hierarchy we can read the supported
2003 * and accessible controllers from a the top-level
2004 * cgroup attribute */
2006 r = cg_get_root_path(&root);
2010 r = cg_get_path(ELOGIND_CGROUP_CONTROLLER, root, "cgroup.controllers", &path);
2014 r = read_one_line_file(path, &controllers);
2020 _cleanup_free_ char *n = NULL;
2023 r = extract_first_word(&c, &n, NULL, 0);
2029 v = cgroup_controller_from_string(n);
2033 mask |= CGROUP_CONTROLLER_TO_MASK(v);
2036 /* Currently, we only support the memory controller in
2037 * the unified hierarchy, mask everything else off. */
2038 mask &= CGROUP_MASK_MEMORY;
2043 /* In the legacy hierarchy, we check whether which
2044 * hierarchies are mounted. */
2046 for (c = 0; c < _CGROUP_CONTROLLER_MAX; c++) {
2049 n = cgroup_controller_to_string(c);
2050 if (controller_is_accessible(n) >= 0)
2051 mask |= CGROUP_CONTROLLER_TO_MASK(c);
2059 /// UNNEEDED by elogind
2061 int cg_kernel_controllers(Set *controllers) {
2062 _cleanup_fclose_ FILE *f = NULL;
2066 assert(controllers);
2068 /* Determines the full list of kernel-known controllers. Might
2069 * include controllers we don't actually support, arbitrary
2070 * named hierarchies and controllers that aren't currently
2071 * accessible (because not mounted). */
2073 f = fopen("/proc/cgroups", "re");
2075 if (errno == ENOENT)
2080 /* Ignore the header line */
2081 (void) fgets(buf, sizeof(buf), f);
2088 if (fscanf(f, "%ms %*i %*i %i", &controller, &enabled) != 2) {
2093 if (ferror(f) && errno != 0)
2104 if (!cg_controller_is_valid(controller)) {
2109 r = set_consume(controllers, controller);
2118 static thread_local int unified_cache = -1;
2120 int cg_unified(void) {
2123 /* Checks if we support the unified hierarchy. Returns an
2124 * error when the cgroup hierarchies aren't mounted yet or we
2125 * have any other trouble determining if the unified hierarchy
2128 if (unified_cache >= 0)
2129 return unified_cache;
2131 if (statfs("/sys/fs/cgroup/", &fs) < 0)
2134 /// elogind can not support the unified hierarchy as a controller,
2135 /// so always assume a classical hierarchy.
2136 /// If, ond only *if*, someone really wants to substitute systemd-login
2137 /// in an environment managed by systemd with elogin, we might have to
2138 /// add such a support.
2140 if (F_TYPE_EQUAL(fs.f_type, CGROUP_SUPER_MAGIC))
2141 unified_cache = true;
2142 else if (F_TYPE_EQUAL(fs.f_type, TMPFS_MAGIC))
2144 if (F_TYPE_EQUAL(fs.f_type, TMPFS_MAGIC))
2146 unified_cache = false;
2150 return unified_cache;
2153 /// UNNEEDED by elogind
2155 void cg_unified_flush(void) {
2159 int cg_enable_everywhere(CGroupMask supported, CGroupMask mask, const char *p) {
2160 _cleanup_free_ char *fs = NULL;
2169 unified = cg_unified();
2172 if (!unified) /* on the legacy hiearchy there's no joining of controllers defined */
2175 r = cg_get_path(ELOGIND_CGROUP_CONTROLLER, p, "cgroup.subtree_control", &fs);
2179 for (c = 0; c < _CGROUP_CONTROLLER_MAX; c++) {
2180 CGroupMask bit = CGROUP_CONTROLLER_TO_MASK(c);
2183 if (!(supported & bit))
2186 n = cgroup_controller_to_string(c);
2188 char s[1 + strlen(n) + 1];
2190 s[0] = mask & bit ? '+' : '-';
2193 r = write_string_file(fs, s, 0);
2195 log_debug_errno(r, "Failed to enable controller %s for %s (%s): %m", n, p, fs);
2203 bool cg_is_unified_wanted(void) {
2204 static thread_local int wanted = -1;
2207 /* If the hierarchy is already mounted, then follow whatever
2208 * was chosen for it. */
2209 unified = cg_unified();
2213 /* Otherwise, let's see what the kernel command line has to
2214 * say. Since checking that is expensive, let's cache the
2219 r = get_proc_cmdline_key("systemd.unified_cgroup_hierarchy", NULL);
2221 return (wanted = true);
2223 _cleanup_free_ char *value = NULL;
2225 r = get_proc_cmdline_key("systemd.unified_cgroup_hierarchy=", &value);
2229 return (wanted = false);
2231 return (wanted = parse_boolean(value) > 0);
2235 /// UNNEEDED by elogind
2237 bool cg_is_legacy_wanted(void) {
2238 return !cg_is_unified_wanted();
2242 static const char *cgroup_controller_table[_CGROUP_CONTROLLER_MAX] = {
2243 [CGROUP_CONTROLLER_CPU] = "cpu",
2244 [CGROUP_CONTROLLER_CPUACCT] = "cpuacct",
2245 [CGROUP_CONTROLLER_BLKIO] = "blkio",
2246 [CGROUP_CONTROLLER_MEMORY] = "memory",
2247 [CGROUP_CONTROLLER_DEVICE] = "devices",
2250 DEFINE_STRING_TABLE_LOOKUP(cgroup_controller, CGroupController);