1 /* SPDX-License-Identifier: LGPL-2.1+ */
3 This file is part of systemd.
5 Copyright 2010 Lennart Poettering
7 systemd is free software; you can redistribute it and/or modify it
8 under the terms of the GNU Lesser General Public License as published by
9 the Free Software Foundation; either version 2.1 of the License, or
10 (at your option) any later version.
12 systemd is distributed in the hope that it will be useful, but
13 WITHOUT ANY WARRANTY; without even the implied warranty of
14 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
15 Lesser General Public License for more details.
17 You should have received a copy of the GNU Lesser General Public License
18 along with systemd; If not, see <http://www.gnu.org/licenses/>.
27 #include <stdio_ext.h>
31 //#include <sys/statfs.h>
32 #include <sys/types.h>
33 #include <sys/xattr.h>
36 #include "alloc-util.h"
37 #include "cgroup-util.h"
39 #include "dirent-util.h"
40 #include "extract-word.h"
43 #include "format-util.h"
46 #include "login-util.h"
48 //#include "missing.h"
50 #include "parse-util.h"
51 #include "path-util.h"
52 #include "proc-cmdline.h"
53 #include "process-util.h"
55 //#include "special.h"
56 #include "stat-util.h"
57 #include "stdio-util.h"
58 #include "string-table.h"
59 #include "string-util.h"
61 #include "unit-name.h"
62 #include "user-util.h"
64 int cg_enumerate_processes(const char *controller, const char *path, FILE **_f) {
65 _cleanup_free_ char *fs = NULL;
71 r = cg_get_path(controller, path, "cgroup.procs", &fs);
83 int cg_read_pid(FILE *f, pid_t *_pid) {
86 /* Note that the cgroup.procs might contain duplicates! See
87 * cgroups.txt for details. */
93 if (fscanf(f, "%lu", &ul) != 1) {
98 return errno > 0 ? -errno : -EIO;
109 const char *controller,
114 _cleanup_free_ char *events = NULL, *content = NULL;
118 r = cg_get_path(controller, path, "cgroup.events", &events);
122 r = read_full_file(events, &content, NULL);
127 while ((line = strsep(&p, "\n"))) {
130 key = strsep(&line, " ");
134 if (strcmp(key, event))
144 #if 0 /// UNNEEDED by elogind
145 bool cg_ns_supported(void) {
146 static thread_local int enabled = -1;
151 if (access("/proc/self/ns/cgroup", F_OK) == 0)
160 int cg_enumerate_subgroups(const char *controller, const char *path, DIR **_d) {
161 _cleanup_free_ char *fs = NULL;
167 /* This is not recursive! */
169 r = cg_get_path(controller, path, NULL, &fs);
181 int cg_read_subgroup(DIR *d, char **fn) {
187 FOREACH_DIRENT_ALL(de, d, return -errno) {
190 if (de->d_type != DT_DIR)
193 if (dot_or_dot_dot(de->d_name))
196 b = strdup(de->d_name);
207 int cg_rmdir(const char *controller, const char *path) {
208 _cleanup_free_ char *p = NULL;
211 r = cg_get_path(controller, path, NULL, &p);
216 if (r < 0 && errno != ENOENT)
219 r = cg_hybrid_unified();
225 if (streq(controller, SYSTEMD_CGROUP_CONTROLLER)) {
226 r = cg_rmdir(SYSTEMD_CGROUP_CONTROLLER_LEGACY, path);
228 log_warning_errno(r, "Failed to remove compat systemd cgroup %s: %m", path);
235 const char *controller,
240 cg_kill_log_func_t log_kill,
243 _cleanup_set_free_ Set *allocated_set = NULL;
250 /* Don't send SIGCONT twice. Also, SIGKILL always works even when process is suspended, hence don't send
251 * SIGCONT on SIGKILL. */
252 if (IN_SET(sig, SIGCONT, SIGKILL))
253 flags &= ~CGROUP_SIGCONT;
255 /* This goes through the tasks list and kills them all. This
256 * is repeated until no further processes are added to the
257 * tasks list, to properly handle forking processes */
260 s = allocated_set = set_new(NULL);
265 my_pid = getpid_cached();
268 _cleanup_fclose_ FILE *f = NULL;
272 r = cg_enumerate_processes(controller, path, &f);
274 if (ret >= 0 && r != -ENOENT)
280 while ((r = cg_read_pid(f, &pid)) > 0) {
282 if ((flags & CGROUP_IGNORE_SELF) && pid == my_pid)
285 if (set_get(s, PID_TO_PTR(pid)) == PID_TO_PTR(pid))
289 log_kill(pid, sig, userdata);
291 /* If we haven't killed this process yet, kill
293 if (kill(pid, sig) < 0) {
294 if (ret >= 0 && errno != ESRCH)
297 if (flags & CGROUP_SIGCONT)
298 (void) kill(pid, SIGCONT);
306 r = set_put(s, PID_TO_PTR(pid));
322 /* To avoid racing against processes which fork
323 * quicker than we can kill them we repeat this until
324 * no new pids need to be killed. */
331 int cg_kill_recursive(
332 const char *controller,
337 cg_kill_log_func_t log_kill,
340 _cleanup_set_free_ Set *allocated_set = NULL;
341 _cleanup_closedir_ DIR *d = NULL;
349 s = allocated_set = set_new(NULL);
354 ret = cg_kill(controller, path, sig, flags, s, log_kill, userdata);
356 r = cg_enumerate_subgroups(controller, path, &d);
358 if (ret >= 0 && r != -ENOENT)
364 while ((r = cg_read_subgroup(d, &fn)) > 0) {
365 _cleanup_free_ char *p = NULL;
367 p = strjoin(path, "/", fn);
372 r = cg_kill_recursive(controller, p, sig, flags, s, log_kill, userdata);
373 if (r != 0 && ret >= 0)
376 if (ret >= 0 && r < 0)
379 if (flags & CGROUP_REMOVE) {
380 r = cg_rmdir(controller, path);
381 if (r < 0 && ret >= 0 && !IN_SET(r, -ENOENT, -EBUSY))
396 _cleanup_set_free_ Set *s = NULL;
409 my_pid = getpid_cached();
411 log_debug_elogind("Migrating \"%s\"/\"%s\" to \"%s\"/\"%s\" (%s)",
412 cfrom, pfrom, cto, pto,
413 (flags & CGROUP_IGNORE_SELF)
414 ? "ignoring self" : "watching self");
416 _cleanup_fclose_ FILE *f = NULL;
420 r = cg_enumerate_processes(cfrom, pfrom, &f);
422 if (ret >= 0 && r != -ENOENT)
428 while ((r = cg_read_pid(f, &pid)) > 0) {
430 /* This might do weird stuff if we aren't a
431 * single-threaded program. However, we
432 * luckily know we are not */
433 if ((flags & CGROUP_IGNORE_SELF) && pid == my_pid)
436 if (set_get(s, PID_TO_PTR(pid)) == PID_TO_PTR(pid))
439 /* Ignore kernel threads. Since they can only
440 * exist in the root cgroup, we only check for
443 (isempty(pfrom) || path_equal(pfrom, "/")) &&
444 is_kernel_thread(pid) > 0)
447 r = cg_attach(cto, pto, pid);
449 if (ret >= 0 && r != -ESRCH)
456 r = set_put(s, PID_TO_PTR(pid));
476 int cg_migrate_recursive(
483 _cleanup_closedir_ DIR *d = NULL;
492 ret = cg_migrate(cfrom, pfrom, cto, pto, flags);
494 r = cg_enumerate_subgroups(cfrom, pfrom, &d);
496 if (ret >= 0 && r != -ENOENT)
502 while ((r = cg_read_subgroup(d, &fn)) > 0) {
503 _cleanup_free_ char *p = NULL;
505 p = strjoin(pfrom, "/", fn);
510 r = cg_migrate_recursive(cfrom, p, cto, pto, flags);
511 if (r != 0 && ret >= 0)
515 if (r < 0 && ret >= 0)
518 if (flags & CGROUP_REMOVE) {
519 r = cg_rmdir(cfrom, pfrom);
520 if (r < 0 && ret >= 0 && !IN_SET(r, -ENOENT, -EBUSY))
527 int cg_migrate_recursive_fallback(
541 r = cg_migrate_recursive(cfrom, pfrom, cto, pto, flags);
543 char prefix[strlen(pto) + 1];
545 /* This didn't work? Then let's try all prefixes of the destination */
547 PATH_FOREACH_PREFIX(prefix, pto) {
550 q = cg_migrate_recursive(cfrom, pfrom, cto, prefix, flags);
559 static const char *controller_to_dirname(const char *controller) {
564 /* Converts a controller name to the directory name below
565 * /sys/fs/cgroup/ we want to mount it to. Effectively, this
566 * just cuts off the name= prefixed used for named
567 * hierarchies, if it is specified. */
569 if (streq(controller, SYSTEMD_CGROUP_CONTROLLER)) {
570 if (cg_hybrid_unified() > 0)
571 controller = SYSTEMD_CGROUP_CONTROLLER_HYBRID;
573 controller = SYSTEMD_CGROUP_CONTROLLER_LEGACY;
576 e = startswith(controller, "name=");
583 static int join_path_legacy(const char *controller, const char *path, const char *suffix, char **fs) {
590 dn = controller_to_dirname(controller);
592 if (isempty(path) && isempty(suffix))
593 t = strappend("/sys/fs/cgroup/", dn);
594 else if (isempty(path))
595 t = strjoin("/sys/fs/cgroup/", dn, "/", suffix);
596 else if (isempty(suffix))
597 t = strjoin("/sys/fs/cgroup/", dn, "/", path);
599 t = strjoin("/sys/fs/cgroup/", dn, "/", path, "/", suffix);
607 static int join_path_unified(const char *path, const char *suffix, char **fs) {
612 if (isempty(path) && isempty(suffix))
613 t = strdup("/sys/fs/cgroup");
614 else if (isempty(path))
615 t = strappend("/sys/fs/cgroup/", suffix);
616 else if (isempty(suffix))
617 t = strappend("/sys/fs/cgroup/", path);
619 t = strjoin("/sys/fs/cgroup/", path, "/", suffix);
627 int cg_get_path(const char *controller, const char *path, const char *suffix, char **fs) {
635 /* If no controller is specified, we return the path
636 * *below* the controllers, without any prefix. */
638 if (!path && !suffix)
646 t = strjoin(path, "/", suffix);
650 *fs = path_kill_slashes(t);
654 if (!cg_controller_is_valid(controller))
657 r = cg_all_unified();
661 r = join_path_unified(path, suffix, fs);
663 r = join_path_legacy(controller, path, suffix, fs);
667 path_kill_slashes(*fs);
671 static int controller_is_accessible(const char *controller) {
676 /* Checks whether a specific controller is accessible,
677 * i.e. its hierarchy mounted. In the unified hierarchy all
678 * controllers are considered accessible, except for the named
681 if (!cg_controller_is_valid(controller))
684 r = cg_all_unified();
688 /* We don't support named hierarchies if we are using
689 * the unified hierarchy. */
691 if (streq(controller, SYSTEMD_CGROUP_CONTROLLER))
694 if (startswith(controller, "name="))
700 dn = controller_to_dirname(controller);
701 cc = strjoina("/sys/fs/cgroup/", dn);
703 if (laccess(cc, F_OK) < 0)
710 int cg_get_path_and_check(const char *controller, const char *path, const char *suffix, char **fs) {
716 /* Check if the specified controller is actually accessible */
717 r = controller_is_accessible(controller);
721 return cg_get_path(controller, path, suffix, fs);
724 static int trim_cb(const char *path, const struct stat *sb, int typeflag, struct FTW *ftwbuf) {
729 if (typeflag != FTW_DP)
732 if (ftwbuf->level < 1)
739 int cg_trim(const char *controller, const char *path, bool delete_root) {
740 _cleanup_free_ char *fs = NULL;
745 r = cg_get_path(controller, path, NULL, &fs);
750 if (nftw(fs, trim_cb, 64, FTW_DEPTH|FTW_MOUNT|FTW_PHYS) != 0) {
760 if (rmdir(fs) < 0 && errno != ENOENT)
764 q = cg_hybrid_unified();
767 if (q > 0 && streq(controller, SYSTEMD_CGROUP_CONTROLLER)) {
768 q = cg_trim(SYSTEMD_CGROUP_CONTROLLER_LEGACY, path, delete_root);
770 log_warning_errno(q, "Failed to trim compat systemd cgroup %s: %m", path);
776 int cg_create(const char *controller, const char *path) {
777 _cleanup_free_ char *fs = NULL;
780 r = cg_get_path_and_check(controller, path, NULL, &fs);
784 r = mkdir_parents(fs, 0755);
788 r = mkdir_errno_wrapper(fs, 0755);
794 r = cg_hybrid_unified();
798 if (r > 0 && streq(controller, SYSTEMD_CGROUP_CONTROLLER)) {
799 r = cg_create(SYSTEMD_CGROUP_CONTROLLER_LEGACY, path);
801 log_warning_errno(r, "Failed to create compat systemd cgroup %s: %m", path);
807 int cg_create_and_attach(const char *controller, const char *path, pid_t pid) {
812 r = cg_create(controller, path);
816 q = cg_attach(controller, path, pid);
820 /* This does not remove the cgroup on failure */
824 int cg_attach(const char *controller, const char *path, pid_t pid) {
825 _cleanup_free_ char *fs = NULL;
826 char c[DECIMAL_STR_MAX(pid_t) + 2];
832 r = cg_get_path_and_check(controller, path, "cgroup.procs", &fs);
837 pid = getpid_cached();
839 xsprintf(c, PID_FMT "\n", pid);
841 r = write_string_file(fs, c, 0);
845 r = cg_hybrid_unified();
849 if (r > 0 && streq(controller, SYSTEMD_CGROUP_CONTROLLER)) {
850 r = cg_attach(SYSTEMD_CGROUP_CONTROLLER_LEGACY, path, pid);
852 log_warning_errno(r, "Failed to attach "PID_FMT" to compat systemd cgroup %s: %m", pid, path);
858 int cg_attach_fallback(const char *controller, const char *path, pid_t pid) {
865 r = cg_attach(controller, path, pid);
867 char prefix[strlen(path) + 1];
869 /* This didn't work? Then let's try all prefixes of
872 PATH_FOREACH_PREFIX(prefix, path) {
875 q = cg_attach(controller, prefix, pid);
884 #if 0 /// UNNEEDED by elogind
886 const char *controller,
896 /* cgroupsv1, aka legacy/non-unified */
897 static const struct Attribute legacy_attributes[] = {
898 { "cgroup.procs", true },
900 { "cgroup.clone_children", false },
904 /* cgroupsv2, aka unified */
905 static const struct Attribute unified_attributes[] = {
906 { "cgroup.procs", true },
907 { "cgroup.subtree_control", true },
908 { "cgroup.threads", false },
912 static const struct Attribute* const attributes[] = {
913 [false] = legacy_attributes,
914 [true] = unified_attributes,
917 _cleanup_free_ char *fs = NULL;
918 const struct Attribute *i;
923 if (uid == UID_INVALID && gid == GID_INVALID)
926 unified = cg_unified_controller(controller);
930 /* Configure access to the cgroup itself */
931 r = cg_get_path(controller, path, NULL, &fs);
935 r = chmod_and_chown(fs, 0755, uid, gid);
939 /* Configure access to the cgroup's attributes */
940 for (i = attributes[unified]; i->name; i++) {
943 r = cg_get_path(controller, path, i->name, &fs);
947 r = chmod_and_chown(fs, 0644, uid, gid);
952 log_debug_errno(r, "Failed to set access on cgroup %s, ignoring: %m", fs);
956 if (streq(controller, SYSTEMD_CGROUP_CONTROLLER)) {
957 r = cg_hybrid_unified();
961 /* Always propagate access mode from unified to legacy controller */
962 r = cg_set_access(SYSTEMD_CGROUP_CONTROLLER_LEGACY, path, uid, gid);
964 log_debug_errno(r, "Failed to set access on compatibility elogind cgroup %s, ignoring: %m", path);
971 int cg_set_xattr(const char *controller, const char *path, const char *name, const void *value, size_t size, int flags) {
972 _cleanup_free_ char *fs = NULL;
977 assert(value || size <= 0);
979 r = cg_get_path(controller, path, NULL, &fs);
983 if (setxattr(fs, name, value, size, flags) < 0)
989 int cg_get_xattr(const char *controller, const char *path, const char *name, void *value, size_t size) {
990 _cleanup_free_ char *fs = NULL;
997 r = cg_get_path(controller, path, NULL, &fs);
1001 n = getxattr(fs, name, value, size);
1009 int cg_pid_get_path(const char *controller, pid_t pid, char **path) {
1010 _cleanup_fclose_ FILE *f = NULL;
1011 char line[LINE_MAX];
1012 #if 0 /// At elogind we do not want that (false alarm) "maybe uninitialized" warning
1013 const char *fs, *controller_str;
1015 const char *fs, *controller_str = NULL;
1024 if (!cg_controller_is_valid(controller))
1027 controller = SYSTEMD_CGROUP_CONTROLLER;
1029 unified = cg_unified_controller(controller);
1033 if (streq(controller, SYSTEMD_CGROUP_CONTROLLER))
1034 controller_str = SYSTEMD_CGROUP_CONTROLLER_LEGACY;
1036 controller_str = controller;
1038 cs = strlen(controller_str);
1041 fs = procfs_file_alloca(pid, "cgroup");
1042 log_debug_elogind("Searching for PID %u in \"%s\" (controller \"%s\")",
1043 pid, fs, controller);
1044 f = fopen(fs, "re");
1046 return errno == ENOENT ? -ESRCH : -errno;
1048 (void) __fsetlocking(f, FSETLOCKING_BYCALLER);
1050 FOREACH_LINE(line, f, return -errno) {
1056 e = startswith(line, "0:");
1066 const char *word, *state;
1069 l = strchr(line, ':');
1079 FOREACH_WORD_SEPARATOR(word, k, l, ",", state) {
1080 if (k == cs && memcmp(word, controller_str, cs) == 0) {
1090 log_debug_elogind("Found %s:%s", line, e+1);
1095 /* Truncate suffix indicating the process is a zombie */
1096 e = endswith(p, " (deleted)");
1107 #if 0 /// UNNEEDED by elogind
1108 int cg_install_release_agent(const char *controller, const char *agent) {
1109 _cleanup_free_ char *fs = NULL, *contents = NULL;
1115 r = cg_unified_controller(controller);
1118 if (r > 0) /* doesn't apply to unified hierarchy */
1121 r = cg_get_path(controller, NULL, "release_agent", &fs);
1125 r = read_one_line_file(fs, &contents);
1129 sc = strstrip(contents);
1131 r = write_string_file(fs, agent, 0);
1134 } else if (!path_equal(sc, agent))
1138 r = cg_get_path(controller, NULL, "notify_on_release", &fs);
1142 contents = mfree(contents);
1143 r = read_one_line_file(fs, &contents);
1147 sc = strstrip(contents);
1148 if (streq(sc, "0")) {
1149 r = write_string_file(fs, "1", 0);
1156 if (!streq(sc, "1"))
1162 int cg_uninstall_release_agent(const char *controller) {
1163 _cleanup_free_ char *fs = NULL;
1166 r = cg_unified_controller(controller);
1169 if (r > 0) /* Doesn't apply to unified hierarchy */
1172 r = cg_get_path(controller, NULL, "notify_on_release", &fs);
1176 r = write_string_file(fs, "0", 0);
1182 r = cg_get_path(controller, NULL, "release_agent", &fs);
1186 r = write_string_file(fs, "", 0);
1194 int cg_is_empty(const char *controller, const char *path) {
1195 _cleanup_fclose_ FILE *f = NULL;
1201 r = cg_enumerate_processes(controller, path, &f);
1207 r = cg_read_pid(f, &pid);
1214 int cg_is_empty_recursive(const char *controller, const char *path) {
1219 /* The root cgroup is always populated */
1220 if (controller && (isempty(path) || path_equal(path, "/")))
1223 r = cg_unified_controller(controller);
1227 _cleanup_free_ char *t = NULL;
1229 /* On the unified hierarchy we can check empty state
1230 * via the "populated" attribute of "cgroup.events". */
1232 r = cg_read_event(controller, path, "populated", &t);
1236 return streq(t, "0");
1238 _cleanup_closedir_ DIR *d = NULL;
1241 r = cg_is_empty(controller, path);
1245 r = cg_enumerate_subgroups(controller, path, &d);
1251 while ((r = cg_read_subgroup(d, &fn)) > 0) {
1252 _cleanup_free_ char *p = NULL;
1254 p = strjoin(path, "/", fn);
1259 r = cg_is_empty_recursive(controller, p);
1270 int cg_split_spec(const char *spec, char **controller, char **path) {
1271 char *t = NULL, *u = NULL;
1277 if (!path_is_normalized(spec))
1285 *path = path_kill_slashes(t);
1294 e = strchr(spec, ':');
1296 if (!cg_controller_is_valid(spec))
1313 t = strndup(spec, e-spec);
1316 if (!cg_controller_is_valid(t)) {
1330 if (!path_is_normalized(u) ||
1331 !path_is_absolute(u)) {
1337 path_kill_slashes(u);
1353 int cg_mangle_path(const char *path, char **result) {
1354 _cleanup_free_ char *c = NULL, *p = NULL;
1361 /* First, check if it already is a filesystem path */
1362 if (path_startswith(path, "/sys/fs/cgroup")) {
1368 *result = path_kill_slashes(t);
1372 /* Otherwise, treat it as cg spec */
1373 r = cg_split_spec(path, &c, &p);
1377 return cg_get_path(c ?: SYSTEMD_CGROUP_CONTROLLER, p ?: "/", NULL, result);
1380 int cg_get_root_path(char **path) {
1386 r = cg_pid_get_path(SYSTEMD_CGROUP_CONTROLLER, 1, &p);
1390 #if 0 /// elogind does not support systemd scopes and slices
1391 e = endswith(p, "/" SPECIAL_INIT_SCOPE);
1393 e = endswith(p, "/" SPECIAL_SYSTEM_SLICE); /* legacy */
1395 e = endswith(p, "/system"); /* even more legacy */
1397 e = endswith(p, "/elogind");
1406 int cg_shift_path(const char *cgroup, const char *root, const char **shifted) {
1407 _cleanup_free_ char *rt = NULL;
1415 /* If the root was specified let's use that, otherwise
1416 * let's determine it from PID 1 */
1418 r = cg_get_root_path(&rt);
1423 log_debug_elogind("Determined root path: \"%s\"", root);
1426 p = path_startswith(cgroup, root);
1427 #if 0 /// With other controllers, elogind might end up in /elogind, and *p is 0
1428 if (p && p > cgroup)
1430 if (p && p[0] && (p > cgroup))
1439 int cg_pid_get_path_shifted(pid_t pid, const char *root, char **cgroup) {
1440 _cleanup_free_ char *raw = NULL;
1447 r = cg_pid_get_path(SYSTEMD_CGROUP_CONTROLLER, pid, &raw);
1451 log_debug_elogind("Shifting path: \"%s\" (PID %u, root: \"%s\")",
1452 raw, pid, root ? root : "NULL");
1453 r = cg_shift_path(raw, root, &c);
1469 log_debug_elogind("Resulting cgroup:\"%s\"", *cgroup);
1474 #if 0 /// UNNEEDED by elogind
1475 int cg_path_decode_unit(const char *cgroup, char **unit) {
1482 n = strcspn(cgroup, "/");
1486 c = strndupa(cgroup, n);
1489 if (!unit_name_is_valid(c, UNIT_NAME_PLAIN|UNIT_NAME_INSTANCE))
1500 static bool valid_slice_name(const char *p, size_t n) {
1505 if (n < STRLEN("x.slice"))
1508 if (memcmp(p + n - 6, ".slice", 6) == 0) {
1514 c = cg_unescape(buf);
1516 return unit_name_is_valid(c, UNIT_NAME_PLAIN);
1522 static const char *skip_slices(const char *p) {
1525 /* Skips over all slice assignments */
1530 p += strspn(p, "/");
1532 n = strcspn(p, "/");
1533 if (!valid_slice_name(p, n))
1540 int cg_path_get_unit(const char *path, char **ret) {
1548 e = skip_slices(path);
1550 r = cg_path_decode_unit(e, &unit);
1554 /* We skipped over the slices, don't accept any now */
1555 if (endswith(unit, ".slice")) {
1564 int cg_pid_get_unit(pid_t pid, char **unit) {
1565 _cleanup_free_ char *cgroup = NULL;
1570 r = cg_pid_get_path_shifted(pid, NULL, &cgroup);
1574 return cg_path_get_unit(cgroup, unit);
1578 * Skip session-*.scope, but require it to be there.
1580 static const char *skip_session(const char *p) {
1586 p += strspn(p, "/");
1588 n = strcspn(p, "/");
1589 if (n < STRLEN("session-x.scope"))
1592 if (memcmp(p, "session-", 8) == 0 && memcmp(p + n - 6, ".scope", 6) == 0) {
1593 char buf[n - 8 - 6 + 1];
1595 memcpy(buf, p + 8, n - 8 - 6);
1598 /* Note that session scopes never need unescaping,
1599 * since they cannot conflict with the kernel's own
1600 * names, hence we don't need to call cg_unescape()
1603 if (!session_id_valid(buf))
1607 p += strspn(p, "/");
1615 * Skip user@*.service, but require it to be there.
1617 static const char *skip_user_manager(const char *p) {
1623 p += strspn(p, "/");
1625 n = strcspn(p, "/");
1626 if (n < STRLEN("user@x.service"))
1629 if (memcmp(p, "user@", 5) == 0 && memcmp(p + n - 8, ".service", 8) == 0) {
1630 char buf[n - 5 - 8 + 1];
1632 memcpy(buf, p + 5, n - 5 - 8);
1635 /* Note that user manager services never need unescaping,
1636 * since they cannot conflict with the kernel's own
1637 * names, hence we don't need to call cg_unescape()
1640 if (parse_uid(buf, NULL) < 0)
1644 p += strspn(p, "/");
1652 static const char *skip_user_prefix(const char *path) {
1657 /* Skip slices, if there are any */
1658 e = skip_slices(path);
1660 /* Skip the user manager, if it's in the path now... */
1661 t = skip_user_manager(e);
1665 /* Alternatively skip the user session if it is in the path... */
1666 return skip_session(e);
1669 int cg_path_get_user_unit(const char *path, char **ret) {
1675 t = skip_user_prefix(path);
1679 /* And from here on it looks pretty much the same as for a
1680 * system unit, hence let's use the same parser from here
1682 return cg_path_get_unit(t, ret);
1685 int cg_pid_get_user_unit(pid_t pid, char **unit) {
1686 _cleanup_free_ char *cgroup = NULL;
1691 r = cg_pid_get_path_shifted(pid, NULL, &cgroup);
1695 return cg_path_get_user_unit(cgroup, unit);
1698 int cg_path_get_machine_name(const char *path, char **machine) {
1699 _cleanup_free_ char *u = NULL;
1703 r = cg_path_get_unit(path, &u);
1707 sl = strjoina("/run/systemd/machines/unit:", u);
1708 return readlink_malloc(sl, machine);
1711 int cg_pid_get_machine_name(pid_t pid, char **machine) {
1712 _cleanup_free_ char *cgroup = NULL;
1717 r = cg_pid_get_path_shifted(pid, NULL, &cgroup);
1721 return cg_path_get_machine_name(cgroup, machine);
1725 int cg_path_get_session(const char *path, char **session) {
1726 #if 0 /// UNNEEDED by elogind
1727 _cleanup_free_ char *unit = NULL;
1733 r = cg_path_get_unit(path, &unit);
1737 start = startswith(unit, "session-");
1740 end = endswith(start, ".scope");
1745 if (!session_id_valid(start))
1748 /* Elogind uses a flat hierarchy, just "/SESSION". The only
1749 wrinkle is that SESSION might be escaped. */
1750 const char *e, *n, *start;
1753 log_debug_elogind("path is \"%s\"", path);
1754 assert(path[0] == '/');
1757 n = strchrnul(e, '/');
1761 start = strndupa(e, n - e);
1762 start = cg_unescape(start);
1771 log_debug_elogind("found session: \"%s\"", start);
1782 int cg_pid_get_session(pid_t pid, char **session) {
1783 _cleanup_free_ char *cgroup = NULL;
1786 r = cg_pid_get_path_shifted(pid, NULL, &cgroup);
1790 return cg_path_get_session(cgroup, session);
1793 int cg_path_get_owner_uid(const char *path, uid_t *uid) {
1794 #if 0 /// elogind needs one more value
1795 _cleanup_free_ char *slice = NULL;
1798 _cleanup_free_ char *slice = NULL, *p = NULL, *s = NULL;
1804 r = cg_path_get_slice(path, &slice);
1808 #if 0 /// elogind does not support systemd slices
1809 start = startswith(slice, "user-");
1812 end = endswith(start, ".slice");
1817 if (parse_uid(start, uid) < 0)
1820 p = strappend("/run/systemd/sessions/", slice);
1822 r = parse_env_file(p, NEWLINE, "UID", &s, NULL);
1830 if (parse_uid(s, uid) < 0)
1837 int cg_pid_get_owner_uid(pid_t pid, uid_t *uid) {
1838 _cleanup_free_ char *cgroup = NULL;
1841 r = cg_pid_get_path_shifted(pid, NULL, &cgroup);
1845 return cg_path_get_owner_uid(cgroup, uid);
1848 int cg_path_get_slice(const char *p, char **slice) {
1849 const char *e = NULL;
1854 #if 0 /// elogind does not support systemd slices
1855 /* Finds the right-most slice unit from the beginning, but
1856 * stops before we come to the first non-slice unit. */
1861 p += strspn(p, "/");
1863 n = strcspn(p, "/");
1864 if (!valid_slice_name(p, n)) {
1869 s = strdup(SPECIAL_ROOT_SLICE);
1877 return cg_path_decode_unit(e, slice);
1884 /* In elogind, what is reported here, is the location of
1885 * the session. This is derived from /proc/<self|PID>/cgroup.
1886 * In there we look at the controller, which will look something
1887 * like "1:name=openrc:/3".
1888 * The last part gets extracted (and is now p), which is "/3" in
1889 * this case. The three is the session id, and that can be mapped.
1891 e = startswith(p, "/");
1902 int cg_pid_get_slice(pid_t pid, char **slice) {
1903 _cleanup_free_ char *cgroup = NULL;
1908 r = cg_pid_get_path_shifted(pid, NULL, &cgroup);
1909 log_debug_elogind("Found cgroup %s for pid %u (result %d)",
1914 return cg_path_get_slice(cgroup, slice);
1917 int cg_path_get_user_slice(const char *p, char **slice) {
1918 #if 0 /// UNNEEDED by elogind
1924 #if 0 /// nothing to skip in elogind
1925 t = skip_user_prefix(p);
1930 #if 0 /// UNNEEDED by elogind
1931 /* And now it looks pretty much the same as for a system
1932 * slice, so let's just use the same parser from here on. */
1933 return cg_path_get_slice(t, slice);
1935 /* In elogind there is nothing to skip, we can use the path
1936 * directly. Generally speaking this is always a session id
1937 * to user mapping. */
1938 return cg_path_get_slice(p, slice);
1942 int cg_pid_get_user_slice(pid_t pid, char **slice) {
1943 _cleanup_free_ char *cgroup = NULL;
1948 r = cg_pid_get_path_shifted(pid, NULL, &cgroup);
1952 return cg_path_get_user_slice(cgroup, slice);
1955 char *cg_escape(const char *p) {
1956 bool need_prefix = false;
1958 /* This implements very minimal escaping for names to be used
1959 * as file names in the cgroup tree: any name which might
1960 * conflict with a kernel name or is prefixed with '_' is
1961 * prefixed with a '_'. That way, when reading cgroup names it
1962 * is sufficient to remove a single prefixing underscore if
1965 /* The return value of this function (unlike cg_unescape())
1968 if (IN_SET(p[0], 0, '_', '.') ||
1969 streq(p, "notify_on_release") ||
1970 streq(p, "release_agent") ||
1971 streq(p, "tasks") ||
1972 startswith(p, "cgroup."))
1977 dot = strrchr(p, '.');
1982 for (c = 0; c < _CGROUP_CONTROLLER_MAX; c++) {
1985 n = cgroup_controller_to_string(c);
1990 if (memcmp(p, n, l) != 0)
2000 return strappend("_", p);
2005 char *cg_unescape(const char *p) {
2008 /* The return value of this function (unlike cg_escape())
2009 * doesn't need free()! */
2017 #define CONTROLLER_VALID \
2021 bool cg_controller_is_valid(const char *p) {
2027 if (streq(p, SYSTEMD_CGROUP_CONTROLLER))
2030 s = startswith(p, "name=");
2034 if (IN_SET(*p, 0, '_'))
2037 for (t = p; *t; t++)
2038 if (!strchr(CONTROLLER_VALID, *t))
2041 if (t - p > FILENAME_MAX)
2047 #if 0 /// UNNEEDED by elogind
2048 int cg_slice_to_path(const char *unit, char **ret) {
2049 _cleanup_free_ char *p = NULL, *s = NULL, *e = NULL;
2056 if (streq(unit, SPECIAL_ROOT_SLICE)) {
2066 if (!unit_name_is_valid(unit, UNIT_NAME_PLAIN))
2069 if (!endswith(unit, ".slice"))
2072 r = unit_name_to_prefix(unit, &p);
2076 dash = strchr(p, '-');
2078 /* Don't allow initial dashes */
2083 _cleanup_free_ char *escaped = NULL;
2084 char n[dash - p + sizeof(".slice")];
2086 /* Don't allow trailing or double dashes */
2087 if (IN_SET(dash[1], 0, '-'))
2090 strcpy(stpncpy(n, p, dash - p), ".slice");
2091 if (!unit_name_is_valid(n, UNIT_NAME_PLAIN))
2094 escaped = cg_escape(n);
2098 if (!strextend(&s, escaped, "/", NULL))
2101 dash = strchr(dash+1, '-');
2104 e = cg_escape(unit);
2108 if (!strextend(&s, e, NULL))
2118 int cg_set_attribute(const char *controller, const char *path, const char *attribute, const char *value) {
2119 _cleanup_free_ char *p = NULL;
2122 r = cg_get_path(controller, path, attribute, &p);
2126 return write_string_file(p, value, 0);
2129 int cg_get_attribute(const char *controller, const char *path, const char *attribute, char **ret) {
2130 _cleanup_free_ char *p = NULL;
2133 r = cg_get_path(controller, path, attribute, &p);
2137 return read_one_line_file(p, ret);
2140 #if 0 /// UNNEEDED by elogind
2141 int cg_get_keyed_attribute(
2142 const char *controller,
2144 const char *attribute,
2146 char **ret_values) {
2148 _cleanup_free_ char *filename = NULL, *contents = NULL;
2149 _cleanup_fclose_ FILE *f = NULL;
2151 size_t n, i, n_done = 0;
2155 /* Reads one or more fields of a cgroupsv2 keyed attribute file. The 'keys' parameter should be an strv with
2156 * all keys to retrieve. The 'ret_values' parameter should be passed as string size with the same number of
2157 * entries as 'keys'. On success each entry will be set to the value of the matching key.
2159 * If the attribute file doesn't exist at all returns ENOENT, if any key is not found returns ENXIO. */
2161 r = cg_get_path(controller, path, attribute, &filename);
2165 r = read_full_file(filename, &contents, NULL);
2169 n = strv_length(keys);
2170 if (n == 0) /* No keys to retrieve? That's easy, we are done then */
2173 /* Let's build this up in a temporary array for now in order not to clobber the return parameter on failure */
2174 v = newa0(char*, n);
2176 for (p = contents; *p;) {
2177 const char *w = NULL;
2179 for (i = 0; i < n; i++)
2181 w = first_word(p, keys[i]);
2189 l = strcspn(w, NEWLINE);
2190 v[i] = strndup(w, l);
2202 p += strcspn(p, NEWLINE);
2204 p += strspn(p, NEWLINE);
2210 for (i = 0; i < n; i++)
2216 memcpy(ret_values, v, sizeof(char*) * n);
2221 int cg_create_everywhere(CGroupMask supported, CGroupMask mask, const char *path) {
2225 /* This one will create a cgroup in our private tree, but also
2226 * duplicate it in the trees specified in mask, and remove it
2229 /* First create the cgroup in our own hierarchy. */
2230 r = cg_create(SYSTEMD_CGROUP_CONTROLLER, path);
2234 /* If we are in the unified hierarchy, we are done now */
2235 r = cg_all_unified();
2241 /* Otherwise, do the same in the other hierarchies */
2242 for (c = 0; c < _CGROUP_CONTROLLER_MAX; c++) {
2243 CGroupMask bit = CGROUP_CONTROLLER_TO_MASK(c);
2246 n = cgroup_controller_to_string(c);
2249 (void) cg_create(n, path);
2250 else if (supported & bit)
2251 (void) cg_trim(n, path, true);
2257 int cg_attach_everywhere(CGroupMask supported, const char *path, pid_t pid, cg_migrate_callback_t path_callback, void *userdata) {
2261 r = cg_attach(SYSTEMD_CGROUP_CONTROLLER, path, pid);
2265 r = cg_all_unified();
2271 for (c = 0; c < _CGROUP_CONTROLLER_MAX; c++) {
2272 CGroupMask bit = CGROUP_CONTROLLER_TO_MASK(c);
2273 const char *p = NULL;
2275 if (!(supported & bit))
2279 p = path_callback(bit, userdata);
2284 (void) cg_attach_fallback(cgroup_controller_to_string(c), p, pid);
2290 int cg_attach_many_everywhere(CGroupMask supported, const char *path, Set* pids, cg_migrate_callback_t path_callback, void *userdata) {
2295 SET_FOREACH(pidp, pids, i) {
2296 pid_t pid = PTR_TO_PID(pidp);
2299 q = cg_attach_everywhere(supported, path, pid, path_callback, userdata);
2300 if (q < 0 && r >= 0)
2307 int cg_migrate_everywhere(CGroupMask supported, const char *from, const char *to, cg_migrate_callback_t to_callback, void *userdata) {
2311 if (!path_equal(from, to)) {
2312 r = cg_migrate_recursive(SYSTEMD_CGROUP_CONTROLLER, from, SYSTEMD_CGROUP_CONTROLLER, to, CGROUP_REMOVE);
2317 q = cg_all_unified();
2323 for (c = 0; c < _CGROUP_CONTROLLER_MAX; c++) {
2324 CGroupMask bit = CGROUP_CONTROLLER_TO_MASK(c);
2325 const char *p = NULL;
2327 if (!(supported & bit))
2331 p = to_callback(bit, userdata);
2336 (void) cg_migrate_recursive_fallback(SYSTEMD_CGROUP_CONTROLLER, to, cgroup_controller_to_string(c), p, 0);
2342 int cg_trim_everywhere(CGroupMask supported, const char *path, bool delete_root) {
2346 r = cg_trim(SYSTEMD_CGROUP_CONTROLLER, path, delete_root);
2350 q = cg_all_unified();
2356 for (c = 0; c < _CGROUP_CONTROLLER_MAX; c++) {
2357 CGroupMask bit = CGROUP_CONTROLLER_TO_MASK(c);
2359 if (!(supported & bit))
2362 (void) cg_trim(cgroup_controller_to_string(c), path, delete_root);
2369 int cg_mask_to_string(CGroupMask mask, char **ret) {
2370 _cleanup_free_ char *s = NULL;
2371 size_t n = 0, allocated = 0;
2382 for (c = 0; c < _CGROUP_CONTROLLER_MAX; c++) {
2386 if (!(mask & CGROUP_CONTROLLER_TO_MASK(c)))
2389 k = cgroup_controller_to_string(c);
2392 if (!GREEDY_REALLOC(s, allocated, n + space + l + 1))
2397 memcpy(s + n + space, k, l);
2412 int cg_mask_from_string(const char *value, CGroupMask *mask) {
2417 _cleanup_free_ char *n = NULL;
2421 r = extract_first_word(&value, &n, NULL, 0);
2427 v = cgroup_controller_from_string(n);
2431 *mask |= CGROUP_CONTROLLER_TO_MASK(v);
2436 int cg_mask_supported(CGroupMask *ret) {
2437 CGroupMask mask = 0;
2440 /* Determines the mask of supported cgroup controllers. Only
2441 * includes controllers we can make sense of and that are
2442 * actually accessible. */
2444 r = cg_all_unified();
2448 _cleanup_free_ char *root = NULL, *controllers = NULL, *path = NULL;
2450 /* In the unified hierarchy we can read the supported
2451 * and accessible controllers from a the top-level
2452 * cgroup attribute */
2454 r = cg_get_root_path(&root);
2458 r = cg_get_path(SYSTEMD_CGROUP_CONTROLLER, root, "cgroup.controllers", &path);
2462 r = read_one_line_file(path, &controllers);
2466 r = cg_mask_from_string(controllers, &mask);
2470 /* Currently, we support the cpu, memory, io and pids
2471 * controller in the unified hierarchy, mask
2472 * everything else off. */
2473 mask &= CGROUP_MASK_CPU | CGROUP_MASK_MEMORY | CGROUP_MASK_IO | CGROUP_MASK_PIDS;
2478 /* In the legacy hierarchy, we check whether which
2479 * hierarchies are mounted. */
2481 for (c = 0; c < _CGROUP_CONTROLLER_MAX; c++) {
2484 n = cgroup_controller_to_string(c);
2485 if (controller_is_accessible(n) >= 0)
2486 mask |= CGROUP_CONTROLLER_TO_MASK(c);
2494 #if 0 /// UNNEEDED by elogind
2495 int cg_kernel_controllers(Set **ret) {
2496 _cleanup_set_free_free_ Set *controllers = NULL;
2497 _cleanup_fclose_ FILE *f = NULL;
2502 /* Determines the full list of kernel-known controllers. Might
2503 * include controllers we don't actually support, arbitrary
2504 * named hierarchies and controllers that aren't currently
2505 * accessible (because not mounted). */
2507 controllers = set_new(&string_hash_ops);
2511 f = fopen("/proc/cgroups", "re");
2513 if (errno == ENOENT) {
2521 (void) __fsetlocking(f, FSETLOCKING_BYCALLER);
2523 /* Ignore the header line */
2524 (void) read_line(f, (size_t) -1, NULL);
2531 if (fscanf(f, "%ms %*i %*i %i", &controller, &enabled) != 2) {
2536 if (ferror(f) && errno > 0)
2547 if (!cg_controller_is_valid(controller)) {
2552 r = set_consume(controllers, controller);
2564 static thread_local CGroupUnified unified_cache = CGROUP_UNIFIED_UNKNOWN;
2566 /* The hybrid mode was initially implemented in v232 and simply mounted cgroup v2 on /sys/fs/cgroup/systemd. This
2567 * unfortunately broke other tools (such as docker) which expected the v1 "name=systemd" hierarchy on
2568 * /sys/fs/cgroup/systemd. From v233 and on, the hybrid mode mountnbs v2 on /sys/fs/cgroup/unified and maintains
2569 * "name=systemd" hierarchy on /sys/fs/cgroup/systemd for compatibility with other tools.
2571 * To keep live upgrade working, we detect and support v232 layout. When v232 layout is detected, to keep cgroup v2
2572 * process management but disable the compat dual layout, we return %true on
2573 * cg_unified_controller(SYSTEMD_CGROUP_CONTROLLER) and %false on cg_hybrid_unified().
2575 static thread_local bool unified_systemd_v232;
2577 static int cg_unified_update(void) {
2581 /* Checks if we support the unified hierarchy. Returns an
2582 * error when the cgroup hierarchies aren't mounted yet or we
2583 * have any other trouble determining if the unified hierarchy
2586 if (unified_cache >= CGROUP_UNIFIED_NONE)
2589 if (statfs("/sys/fs/cgroup/", &fs) < 0)
2590 return log_debug_errno(errno, "statfs(\"/sys/fs/cgroup/\" failed: %m");
2592 if (F_TYPE_EQUAL(fs.f_type, CGROUP2_SUPER_MAGIC)) {
2593 log_debug("Found cgroup2 on /sys/fs/cgroup/, full unified hierarchy");
2594 unified_cache = CGROUP_UNIFIED_ALL;
2595 #if 0 /// The handling of cgroups is a bit different with elogind
2596 } else if (F_TYPE_EQUAL(fs.f_type, TMPFS_MAGIC)) {
2597 log_debug("Found cgroup2 on /sys/fs/cgroup/unified, unified hierarchy for systemd controller");
2599 } else if (F_TYPE_EQUAL(fs.f_type, CGROUP_SUPER_MAGIC)
2600 || F_TYPE_EQUAL(fs.f_type, TMPFS_MAGIC)) {
2602 if (statfs("/sys/fs/cgroup/unified/", &fs) == 0 &&
2603 F_TYPE_EQUAL(fs.f_type, CGROUP2_SUPER_MAGIC)) {
2604 unified_cache = CGROUP_UNIFIED_SYSTEMD;
2605 unified_systemd_v232 = false;
2607 #if 0 /// There is no sub-grouping within elogind
2608 if (statfs("/sys/fs/cgroup/systemd/", &fs) < 0)
2609 return log_debug_errno(errno, "statfs(\"/sys/fs/cgroup/systemd\" failed: %m");
2611 if (F_TYPE_EQUAL(fs.f_type, CGROUP2_SUPER_MAGIC)) {
2612 log_debug("Found cgroup2 on /sys/fs/cgroup/systemd, unified hierarchy for systemd controller (v232 variant)");
2613 unified_cache = CGROUP_UNIFIED_SYSTEMD;
2614 unified_systemd_v232 = true;
2615 } else if (F_TYPE_EQUAL(fs.f_type, CGROUP_SUPER_MAGIC)) {
2616 log_debug("Found cgroup on /sys/fs/cgroup/systemd, legacy hierarchy");
2617 unified_cache = CGROUP_UNIFIED_NONE;
2619 log_debug("Unexpected filesystem type %llx mounted on /sys/fs/cgroup/systemd, assuming legacy hierarchy",
2620 (unsigned long long) fs.f_type);
2621 unified_cache = CGROUP_UNIFIED_NONE;
2624 unified_cache = CGROUP_UNIFIED_NONE;
2628 log_debug("Unknown filesystem type %llx mounted on /sys/fs/cgroup.",
2629 (unsigned long long) fs.f_type);
2636 int cg_unified_controller(const char *controller) {
2639 r = cg_unified_update();
2643 if (unified_cache == CGROUP_UNIFIED_NONE)
2646 if (unified_cache >= CGROUP_UNIFIED_ALL)
2649 #if 0 /// only if elogind is the controller we can use cgroups2 in hybrid mode
2650 return streq_ptr(controller, SYSTEMD_CGROUP_CONTROLLER);
2652 return streq_ptr(controller, SYSTEMD_CGROUP_CONTROLLER_HYBRID);
2656 int cg_all_unified(void) {
2659 r = cg_unified_update();
2663 return unified_cache >= CGROUP_UNIFIED_ALL;
2666 int cg_hybrid_unified(void) {
2669 r = cg_unified_update();
2673 return unified_cache == CGROUP_UNIFIED_SYSTEMD && !unified_systemd_v232;
2676 int cg_unified_flush(void) {
2677 unified_cache = CGROUP_UNIFIED_UNKNOWN;
2679 return cg_unified_update();
2682 #if 0 /// UNNEEDED by elogind
2683 int cg_enable_everywhere(CGroupMask supported, CGroupMask mask, const char *p) {
2684 _cleanup_fclose_ FILE *f = NULL;
2685 _cleanup_free_ char *fs = NULL;
2694 r = cg_all_unified();
2697 if (r == 0) /* on the legacy hiearchy there's no joining of controllers defined */
2700 r = cg_get_path(SYSTEMD_CGROUP_CONTROLLER, p, "cgroup.subtree_control", &fs);
2704 for (c = 0; c < _CGROUP_CONTROLLER_MAX; c++) {
2705 CGroupMask bit = CGROUP_CONTROLLER_TO_MASK(c);
2708 if (!(supported & bit))
2711 n = cgroup_controller_to_string(c);
2713 char s[1 + strlen(n) + 1];
2715 s[0] = mask & bit ? '+' : '-';
2719 f = fopen(fs, "we");
2721 log_debug_errno(errno, "Failed to open cgroup.subtree_control file of %s: %m", p);
2726 r = write_string_stream(f, s, 0);
2728 log_debug_errno(r, "Failed to enable controller %s for %s (%s): %m", n, p, fs);
2736 bool cg_is_unified_wanted(void) {
2737 static thread_local int wanted = -1;
2740 const bool is_default = DEFAULT_HIERARCHY == CGROUP_UNIFIED_ALL;
2742 /* If we have a cached value, return that. */
2746 /* If the hierarchy is already mounted, then follow whatever
2747 * was chosen for it. */
2748 if (cg_unified_flush() >= 0)
2749 return (wanted = unified_cache >= CGROUP_UNIFIED_ALL);
2751 #if 0 /// elogind is not init and has no business with kernel command line
2752 /* Otherwise, let's see what the kernel command line has to say.
2753 * Since checking is expensive, cache a non-error result. */
2754 r = proc_cmdline_get_bool("systemd.unified_cgroup_hierarchy", &b);
2757 return (wanted = r > 0 ? b : is_default);
2760 bool cg_is_legacy_wanted(void) {
2761 static thread_local int wanted = -1;
2763 /* If we have a cached value, return that. */
2767 /* Check if we have cgroups2 already mounted. */
2768 if (cg_unified_flush() >= 0 &&
2769 unified_cache == CGROUP_UNIFIED_ALL)
2770 return (wanted = false);
2772 /* Otherwise, assume that at least partial legacy is wanted,
2773 * since cgroups2 should already be mounted at this point. */
2774 return (wanted = true);
2777 bool cg_is_hybrid_wanted(void) {
2778 static thread_local int wanted = -1;
2781 const bool is_default = DEFAULT_HIERARCHY >= CGROUP_UNIFIED_SYSTEMD;
2782 /* We default to true if the default is "hybrid", obviously,
2783 * but also when the default is "unified", because if we get
2784 * called, it means that unified hierarchy was not mounted. */
2786 /* If we have a cached value, return that. */
2790 /* If the hierarchy is already mounted, then follow whatever
2791 * was chosen for it. */
2792 if (cg_unified_flush() >= 0 &&
2793 unified_cache == CGROUP_UNIFIED_ALL)
2794 return (wanted = false);
2796 #if 0 /// elogind is not init and has no business with kernel command line
2797 /* Otherwise, let's see what the kernel command line has to say.
2798 * Since checking is expensive, cache a non-error result. */
2799 r = proc_cmdline_get_bool("systemd.legacy_systemd_cgroup_controller", &b);
2802 /* The meaning of the kernel option is reversed wrt. to the return value
2803 * of this function, hence the negation. */
2804 return (wanted = r > 0 ? !b : is_default);
2807 #if 0 /// UNNEEDED by elogind
2808 int cg_weight_parse(const char *s, uint64_t *ret) {
2813 *ret = CGROUP_WEIGHT_INVALID;
2817 r = safe_atou64(s, &u);
2821 if (u < CGROUP_WEIGHT_MIN || u > CGROUP_WEIGHT_MAX)
2828 const uint64_t cgroup_io_limit_defaults[_CGROUP_IO_LIMIT_TYPE_MAX] = {
2829 [CGROUP_IO_RBPS_MAX] = CGROUP_LIMIT_MAX,
2830 [CGROUP_IO_WBPS_MAX] = CGROUP_LIMIT_MAX,
2831 [CGROUP_IO_RIOPS_MAX] = CGROUP_LIMIT_MAX,
2832 [CGROUP_IO_WIOPS_MAX] = CGROUP_LIMIT_MAX,
2835 static const char* const cgroup_io_limit_type_table[_CGROUP_IO_LIMIT_TYPE_MAX] = {
2836 [CGROUP_IO_RBPS_MAX] = "IOReadBandwidthMax",
2837 [CGROUP_IO_WBPS_MAX] = "IOWriteBandwidthMax",
2838 [CGROUP_IO_RIOPS_MAX] = "IOReadIOPSMax",
2839 [CGROUP_IO_WIOPS_MAX] = "IOWriteIOPSMax",
2842 DEFINE_STRING_TABLE_LOOKUP(cgroup_io_limit_type, CGroupIOLimitType);
2844 int cg_cpu_shares_parse(const char *s, uint64_t *ret) {
2849 *ret = CGROUP_CPU_SHARES_INVALID;
2853 r = safe_atou64(s, &u);
2857 if (u < CGROUP_CPU_SHARES_MIN || u > CGROUP_CPU_SHARES_MAX)
2864 int cg_blkio_weight_parse(const char *s, uint64_t *ret) {
2869 *ret = CGROUP_BLKIO_WEIGHT_INVALID;
2873 r = safe_atou64(s, &u);
2877 if (u < CGROUP_BLKIO_WEIGHT_MIN || u > CGROUP_BLKIO_WEIGHT_MAX)
2885 bool is_cgroup_fs(const struct statfs *s) {
2886 return is_fs_type(s, CGROUP_SUPER_MAGIC) ||
2887 is_fs_type(s, CGROUP2_SUPER_MAGIC);
2890 bool fd_is_cgroup_fs(int fd) {
2893 if (fstatfs(fd, &s) < 0)
2896 return is_cgroup_fs(&s);
2899 static const char *cgroup_controller_table[_CGROUP_CONTROLLER_MAX] = {
2900 [CGROUP_CONTROLLER_CPU] = "cpu",
2901 [CGROUP_CONTROLLER_CPUACCT] = "cpuacct",
2902 [CGROUP_CONTROLLER_IO] = "io",
2903 [CGROUP_CONTROLLER_BLKIO] = "blkio",
2904 [CGROUP_CONTROLLER_MEMORY] = "memory",
2905 [CGROUP_CONTROLLER_DEVICES] = "devices",
2906 [CGROUP_CONTROLLER_PIDS] = "pids",
2909 DEFINE_STRING_TABLE_LOOKUP(cgroup_controller, CGroupController);