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) {
1088 log_debug_elogind("Found %s:%s", line, e+1);
1093 /* Truncate suffix indicating the process is a zombie */
1094 e = endswith(p, " (deleted)");
1105 #if 0 /// UNNEEDED by elogind
1106 int cg_install_release_agent(const char *controller, const char *agent) {
1107 _cleanup_free_ char *fs = NULL, *contents = NULL;
1113 r = cg_unified_controller(controller);
1116 if (r > 0) /* doesn't apply to unified hierarchy */
1119 r = cg_get_path(controller, NULL, "release_agent", &fs);
1123 r = read_one_line_file(fs, &contents);
1127 sc = strstrip(contents);
1129 r = write_string_file(fs, agent, 0);
1132 } else if (!path_equal(sc, agent))
1136 r = cg_get_path(controller, NULL, "notify_on_release", &fs);
1140 contents = mfree(contents);
1141 r = read_one_line_file(fs, &contents);
1145 sc = strstrip(contents);
1146 if (streq(sc, "0")) {
1147 r = write_string_file(fs, "1", 0);
1154 if (!streq(sc, "1"))
1160 int cg_uninstall_release_agent(const char *controller) {
1161 _cleanup_free_ char *fs = NULL;
1164 r = cg_unified_controller(controller);
1167 if (r > 0) /* Doesn't apply to unified hierarchy */
1170 r = cg_get_path(controller, NULL, "notify_on_release", &fs);
1174 r = write_string_file(fs, "0", 0);
1180 r = cg_get_path(controller, NULL, "release_agent", &fs);
1184 r = write_string_file(fs, "", 0);
1192 int cg_is_empty(const char *controller, const char *path) {
1193 _cleanup_fclose_ FILE *f = NULL;
1199 r = cg_enumerate_processes(controller, path, &f);
1205 r = cg_read_pid(f, &pid);
1212 int cg_is_empty_recursive(const char *controller, const char *path) {
1217 /* The root cgroup is always populated */
1218 if (controller && (isempty(path) || path_equal(path, "/")))
1221 r = cg_unified_controller(controller);
1225 _cleanup_free_ char *t = NULL;
1227 /* On the unified hierarchy we can check empty state
1228 * via the "populated" attribute of "cgroup.events". */
1230 r = cg_read_event(controller, path, "populated", &t);
1234 return streq(t, "0");
1236 _cleanup_closedir_ DIR *d = NULL;
1239 r = cg_is_empty(controller, path);
1243 r = cg_enumerate_subgroups(controller, path, &d);
1249 while ((r = cg_read_subgroup(d, &fn)) > 0) {
1250 _cleanup_free_ char *p = NULL;
1252 p = strjoin(path, "/", fn);
1257 r = cg_is_empty_recursive(controller, p);
1268 int cg_split_spec(const char *spec, char **controller, char **path) {
1269 char *t = NULL, *u = NULL;
1275 if (!path_is_normalized(spec))
1283 *path = path_kill_slashes(t);
1292 e = strchr(spec, ':');
1294 if (!cg_controller_is_valid(spec))
1311 t = strndup(spec, e-spec);
1314 if (!cg_controller_is_valid(t)) {
1328 if (!path_is_normalized(u) ||
1329 !path_is_absolute(u)) {
1335 path_kill_slashes(u);
1351 int cg_mangle_path(const char *path, char **result) {
1352 _cleanup_free_ char *c = NULL, *p = NULL;
1359 /* First, check if it already is a filesystem path */
1360 if (path_startswith(path, "/sys/fs/cgroup")) {
1366 *result = path_kill_slashes(t);
1370 /* Otherwise, treat it as cg spec */
1371 r = cg_split_spec(path, &c, &p);
1375 return cg_get_path(c ?: SYSTEMD_CGROUP_CONTROLLER, p ?: "/", NULL, result);
1378 int cg_get_root_path(char **path) {
1384 r = cg_pid_get_path(SYSTEMD_CGROUP_CONTROLLER, 1, &p);
1388 #if 0 /// elogind does not support systemd scopes and slices
1389 e = endswith(p, "/" SPECIAL_INIT_SCOPE);
1391 e = endswith(p, "/" SPECIAL_SYSTEM_SLICE); /* legacy */
1393 e = endswith(p, "/system"); /* even more legacy */
1395 e = endswith(p, "/elogind");
1404 int cg_shift_path(const char *cgroup, const char *root, const char **shifted) {
1405 _cleanup_free_ char *rt = NULL;
1413 /* If the root was specified let's use that, otherwise
1414 * let's determine it from PID 1 */
1416 r = cg_get_root_path(&rt);
1421 log_debug_elogind("Determined root path: \"%s\"", root);
1424 p = path_startswith(cgroup, root);
1425 #if 0 /// With other controllers, elogind might end up in /elogind, and *p is 0
1426 if (p && p > cgroup)
1428 if (p && p[0] && (p > cgroup))
1437 int cg_pid_get_path_shifted(pid_t pid, const char *root, char **cgroup) {
1438 _cleanup_free_ char *raw = NULL;
1445 r = cg_pid_get_path(SYSTEMD_CGROUP_CONTROLLER, pid, &raw);
1449 log_debug_elogind("Shifting path: \"%s\" (PID %u, root: \"%s\")",
1450 raw, pid, root ? root : "NULL");
1451 r = cg_shift_path(raw, root, &c);
1467 log_debug_elogind("Resulting cgroup:\"%s\"", *cgroup);
1472 int cg_path_decode_unit(const char *cgroup, char **unit) {
1479 n = strcspn(cgroup, "/");
1483 c = strndupa(cgroup, n);
1486 if (!unit_name_is_valid(c, UNIT_NAME_PLAIN|UNIT_NAME_INSTANCE))
1497 static bool valid_slice_name(const char *p, size_t n) {
1502 if (n < STRLEN("x.slice"))
1505 if (memcmp(p + n - 6, ".slice", 6) == 0) {
1511 c = cg_unescape(buf);
1513 return unit_name_is_valid(c, UNIT_NAME_PLAIN);
1519 static const char *skip_slices(const char *p) {
1522 /* Skips over all slice assignments */
1527 p += strspn(p, "/");
1529 n = strcspn(p, "/");
1530 if (!valid_slice_name(p, n))
1537 int cg_path_get_unit(const char *path, char **ret) {
1545 e = skip_slices(path);
1547 r = cg_path_decode_unit(e, &unit);
1551 /* We skipped over the slices, don't accept any now */
1552 if (endswith(unit, ".slice")) {
1561 int cg_pid_get_unit(pid_t pid, char **unit) {
1562 _cleanup_free_ char *cgroup = NULL;
1567 r = cg_pid_get_path_shifted(pid, NULL, &cgroup);
1571 return cg_path_get_unit(cgroup, unit);
1574 #if 0 /// UNNEEDED by elogind
1576 * Skip session-*.scope, but require it to be there.
1578 static const char *skip_session(const char *p) {
1584 p += strspn(p, "/");
1586 n = strcspn(p, "/");
1587 if (n < STRLEN("session-x.scope"))
1590 if (memcmp(p, "session-", 8) == 0 && memcmp(p + n - 6, ".scope", 6) == 0) {
1591 char buf[n - 8 - 6 + 1];
1593 memcpy(buf, p + 8, n - 8 - 6);
1596 /* Note that session scopes never need unescaping,
1597 * since they cannot conflict with the kernel's own
1598 * names, hence we don't need to call cg_unescape()
1601 if (!session_id_valid(buf))
1605 p += strspn(p, "/");
1613 * Skip user@*.service, but require it to be there.
1615 static const char *skip_user_manager(const char *p) {
1621 p += strspn(p, "/");
1623 n = strcspn(p, "/");
1624 if (n < STRLEN("user@x.service"))
1627 if (memcmp(p, "user@", 5) == 0 && memcmp(p + n - 8, ".service", 8) == 0) {
1628 char buf[n - 5 - 8 + 1];
1630 memcpy(buf, p + 5, n - 5 - 8);
1633 /* Note that user manager services never need unescaping,
1634 * since they cannot conflict with the kernel's own
1635 * names, hence we don't need to call cg_unescape()
1638 if (parse_uid(buf, NULL) < 0)
1642 p += strspn(p, "/");
1650 static const char *skip_user_prefix(const char *path) {
1655 /* Skip slices, if there are any */
1656 e = skip_slices(path);
1658 /* Skip the user manager, if it's in the path now... */
1659 t = skip_user_manager(e);
1663 /* Alternatively skip the user session if it is in the path... */
1664 return skip_session(e);
1667 int cg_path_get_user_unit(const char *path, char **ret) {
1673 t = skip_user_prefix(path);
1677 /* And from here on it looks pretty much the same as for a
1678 * system unit, hence let's use the same parser from here
1680 return cg_path_get_unit(t, ret);
1683 int cg_pid_get_user_unit(pid_t pid, char **unit) {
1684 _cleanup_free_ char *cgroup = NULL;
1689 r = cg_pid_get_path_shifted(pid, NULL, &cgroup);
1693 return cg_path_get_user_unit(cgroup, unit);
1696 int cg_path_get_machine_name(const char *path, char **machine) {
1697 _cleanup_free_ char *u = NULL;
1701 r = cg_path_get_unit(path, &u);
1705 sl = strjoina("/run/systemd/machines/unit:", u);
1706 return readlink_malloc(sl, machine);
1709 int cg_pid_get_machine_name(pid_t pid, char **machine) {
1710 _cleanup_free_ char *cgroup = NULL;
1715 r = cg_pid_get_path_shifted(pid, NULL, &cgroup);
1719 return cg_path_get_machine_name(cgroup, machine);
1723 int cg_path_get_session(const char *path, char **session) {
1724 #if 0 /// UNNEEDED by elogind
1725 _cleanup_free_ char *unit = NULL;
1731 r = cg_path_get_unit(path, &unit);
1735 start = startswith(unit, "session-");
1738 end = endswith(start, ".scope");
1743 if (!session_id_valid(start))
1746 /* Elogind uses a flat hierarchy, just "/SESSION". The only
1747 wrinkle is that SESSION might be escaped. */
1748 const char *e, *n, *start;
1751 log_debug_elogind("path is \"%s\"", path);
1752 assert(path[0] == '/');
1755 n = strchrnul(e, '/');
1759 start = strndupa(e, n - e);
1760 start = cg_unescape(start);
1769 log_debug_elogind("found session: \"%s\"", start);
1780 int cg_pid_get_session(pid_t pid, char **session) {
1781 _cleanup_free_ char *cgroup = NULL;
1784 r = cg_pid_get_path_shifted(pid, NULL, &cgroup);
1788 return cg_path_get_session(cgroup, session);
1791 int cg_path_get_owner_uid(const char *path, uid_t *uid) {
1792 #if 0 /// elogind needs one more value
1793 _cleanup_free_ char *slice = NULL;
1796 _cleanup_free_ char *slice = NULL, *p = NULL, *s = NULL;
1802 r = cg_path_get_slice(path, &slice);
1806 #if 0 /// elogind does not support systemd slices
1807 start = startswith(slice, "user-");
1810 end = endswith(start, ".slice");
1815 if (parse_uid(start, uid) < 0)
1818 p = strappend("/run/systemd/sessions/", slice);
1820 r = parse_env_file(p, NEWLINE, "UID", &s, NULL);
1828 if (parse_uid(s, uid) < 0)
1835 int cg_pid_get_owner_uid(pid_t pid, uid_t *uid) {
1836 _cleanup_free_ char *cgroup = NULL;
1839 r = cg_pid_get_path_shifted(pid, NULL, &cgroup);
1843 return cg_path_get_owner_uid(cgroup, uid);
1846 int cg_path_get_slice(const char *p, char **slice) {
1847 const char *e = NULL;
1852 #if 0 /// elogind does not support systemd slices
1853 /* Finds the right-most slice unit from the beginning, but
1854 * stops before we come to the first non-slice unit. */
1859 p += strspn(p, "/");
1861 n = strcspn(p, "/");
1862 if (!valid_slice_name(p, n)) {
1867 s = strdup(SPECIAL_ROOT_SLICE);
1875 return cg_path_decode_unit(e, slice);
1882 /* In elogind, what is reported here, is the location of
1883 * the session. This is derived from /proc/<self|PID>/cgroup.
1884 * In there we look at the controller, which will look something
1885 * like "1:name=openrc:/3".
1886 * The last part gets extracted (and is now p), which is "/3" in
1887 * this case. The three is the session id, and that can be mapped.
1889 e = startswith(p, "/");
1900 int cg_pid_get_slice(pid_t pid, char **slice) {
1901 _cleanup_free_ char *cgroup = NULL;
1906 r = cg_pid_get_path_shifted(pid, NULL, &cgroup);
1907 log_debug_elogind("Found cgroup %s for pid %u (result %d)",
1912 return cg_path_get_slice(cgroup, slice);
1915 int cg_path_get_user_slice(const char *p, char **slice) {
1916 #if 0 /// UNNEEDED by elogind
1922 #if 0 /// nothing to skip in elogind
1923 t = skip_user_prefix(p);
1928 #if 0 /// UNNEEDED by elogind
1929 /* And now it looks pretty much the same as for a system
1930 * slice, so let's just use the same parser from here on. */
1931 return cg_path_get_slice(t, slice);
1933 /* In elogind there is nothing to skip, we can use the path
1934 * directly. Generally speaking this is always a session id
1935 * to user mapping. */
1936 return cg_path_get_slice(p, slice);
1940 int cg_pid_get_user_slice(pid_t pid, char **slice) {
1941 _cleanup_free_ char *cgroup = NULL;
1946 r = cg_pid_get_path_shifted(pid, NULL, &cgroup);
1950 return cg_path_get_user_slice(cgroup, slice);
1953 char *cg_escape(const char *p) {
1954 bool need_prefix = false;
1956 /* This implements very minimal escaping for names to be used
1957 * as file names in the cgroup tree: any name which might
1958 * conflict with a kernel name or is prefixed with '_' is
1959 * prefixed with a '_'. That way, when reading cgroup names it
1960 * is sufficient to remove a single prefixing underscore if
1963 /* The return value of this function (unlike cg_unescape())
1966 if (IN_SET(p[0], 0, '_', '.') ||
1967 streq(p, "notify_on_release") ||
1968 streq(p, "release_agent") ||
1969 streq(p, "tasks") ||
1970 startswith(p, "cgroup."))
1975 dot = strrchr(p, '.');
1980 for (c = 0; c < _CGROUP_CONTROLLER_MAX; c++) {
1983 n = cgroup_controller_to_string(c);
1988 if (memcmp(p, n, l) != 0)
1998 return strappend("_", p);
2003 char *cg_unescape(const char *p) {
2006 /* The return value of this function (unlike cg_escape())
2007 * doesn't need free()! */
2015 #define CONTROLLER_VALID \
2019 bool cg_controller_is_valid(const char *p) {
2025 if (streq(p, SYSTEMD_CGROUP_CONTROLLER))
2028 s = startswith(p, "name=");
2032 if (IN_SET(*p, 0, '_'))
2035 for (t = p; *t; t++)
2036 if (!strchr(CONTROLLER_VALID, *t))
2039 if (t - p > FILENAME_MAX)
2045 #if 0 /// UNNEEDED by elogind
2046 int cg_slice_to_path(const char *unit, char **ret) {
2047 _cleanup_free_ char *p = NULL, *s = NULL, *e = NULL;
2054 if (streq(unit, SPECIAL_ROOT_SLICE)) {
2064 if (!unit_name_is_valid(unit, UNIT_NAME_PLAIN))
2067 if (!endswith(unit, ".slice"))
2070 r = unit_name_to_prefix(unit, &p);
2074 dash = strchr(p, '-');
2076 /* Don't allow initial dashes */
2081 _cleanup_free_ char *escaped = NULL;
2082 char n[dash - p + sizeof(".slice")];
2084 /* Don't allow trailing or double dashes */
2085 if (IN_SET(dash[1], 0, '-'))
2088 strcpy(stpncpy(n, p, dash - p), ".slice");
2089 if (!unit_name_is_valid(n, UNIT_NAME_PLAIN))
2092 escaped = cg_escape(n);
2096 if (!strextend(&s, escaped, "/", NULL))
2099 dash = strchr(dash+1, '-');
2102 e = cg_escape(unit);
2106 if (!strextend(&s, e, NULL))
2116 int cg_set_attribute(const char *controller, const char *path, const char *attribute, const char *value) {
2117 _cleanup_free_ char *p = NULL;
2120 r = cg_get_path(controller, path, attribute, &p);
2124 return write_string_file(p, value, 0);
2127 int cg_get_attribute(const char *controller, const char *path, const char *attribute, char **ret) {
2128 _cleanup_free_ char *p = NULL;
2131 r = cg_get_path(controller, path, attribute, &p);
2135 return read_one_line_file(p, ret);
2138 #if 0 /// UNNEEDED by elogind
2139 int cg_get_keyed_attribute(
2140 const char *controller,
2142 const char *attribute,
2144 char **ret_values) {
2146 _cleanup_free_ char *filename = NULL, *contents = NULL;
2147 _cleanup_fclose_ FILE *f = NULL;
2149 size_t n, i, n_done = 0;
2153 /* Reads one or more fields of a cgroupsv2 keyed attribute file. The 'keys' parameter should be an strv with
2154 * all keys to retrieve. The 'ret_values' parameter should be passed as string size with the same number of
2155 * entries as 'keys'. On success each entry will be set to the value of the matching key.
2157 * If the attribute file doesn't exist at all returns ENOENT, if any key is not found returns ENXIO. */
2159 r = cg_get_path(controller, path, attribute, &filename);
2163 r = read_full_file(filename, &contents, NULL);
2167 n = strv_length(keys);
2168 if (n == 0) /* No keys to retrieve? That's easy, we are done then */
2171 /* Let's build this up in a temporary array for now in order not to clobber the return parameter on failure */
2172 v = newa0(char*, n);
2174 for (p = contents; *p;) {
2175 const char *w = NULL;
2177 for (i = 0; i < n; i++)
2179 w = first_word(p, keys[i]);
2187 l = strcspn(w, NEWLINE);
2188 v[i] = strndup(w, l);
2200 p += strcspn(p, NEWLINE);
2202 p += strspn(p, NEWLINE);
2208 for (i = 0; i < n; i++)
2214 memcpy(ret_values, v, sizeof(char*) * n);
2219 int cg_create_everywhere(CGroupMask supported, CGroupMask mask, const char *path) {
2223 /* This one will create a cgroup in our private tree, but also
2224 * duplicate it in the trees specified in mask, and remove it
2227 /* First create the cgroup in our own hierarchy. */
2228 r = cg_create(SYSTEMD_CGROUP_CONTROLLER, path);
2232 /* If we are in the unified hierarchy, we are done now */
2233 r = cg_all_unified();
2239 /* Otherwise, do the same in the other hierarchies */
2240 for (c = 0; c < _CGROUP_CONTROLLER_MAX; c++) {
2241 CGroupMask bit = CGROUP_CONTROLLER_TO_MASK(c);
2244 n = cgroup_controller_to_string(c);
2247 (void) cg_create(n, path);
2248 else if (supported & bit)
2249 (void) cg_trim(n, path, true);
2255 int cg_attach_everywhere(CGroupMask supported, const char *path, pid_t pid, cg_migrate_callback_t path_callback, void *userdata) {
2259 r = cg_attach(SYSTEMD_CGROUP_CONTROLLER, path, pid);
2263 r = cg_all_unified();
2269 for (c = 0; c < _CGROUP_CONTROLLER_MAX; c++) {
2270 CGroupMask bit = CGROUP_CONTROLLER_TO_MASK(c);
2271 const char *p = NULL;
2273 if (!(supported & bit))
2277 p = path_callback(bit, userdata);
2282 (void) cg_attach_fallback(cgroup_controller_to_string(c), p, pid);
2288 int cg_attach_many_everywhere(CGroupMask supported, const char *path, Set* pids, cg_migrate_callback_t path_callback, void *userdata) {
2293 SET_FOREACH(pidp, pids, i) {
2294 pid_t pid = PTR_TO_PID(pidp);
2297 q = cg_attach_everywhere(supported, path, pid, path_callback, userdata);
2298 if (q < 0 && r >= 0)
2305 int cg_migrate_everywhere(CGroupMask supported, const char *from, const char *to, cg_migrate_callback_t to_callback, void *userdata) {
2309 if (!path_equal(from, to)) {
2310 r = cg_migrate_recursive(SYSTEMD_CGROUP_CONTROLLER, from, SYSTEMD_CGROUP_CONTROLLER, to, CGROUP_REMOVE);
2315 q = cg_all_unified();
2321 for (c = 0; c < _CGROUP_CONTROLLER_MAX; c++) {
2322 CGroupMask bit = CGROUP_CONTROLLER_TO_MASK(c);
2323 const char *p = NULL;
2325 if (!(supported & bit))
2329 p = to_callback(bit, userdata);
2334 (void) cg_migrate_recursive_fallback(SYSTEMD_CGROUP_CONTROLLER, to, cgroup_controller_to_string(c), p, 0);
2340 int cg_trim_everywhere(CGroupMask supported, const char *path, bool delete_root) {
2344 r = cg_trim(SYSTEMD_CGROUP_CONTROLLER, path, delete_root);
2348 q = cg_all_unified();
2354 for (c = 0; c < _CGROUP_CONTROLLER_MAX; c++) {
2355 CGroupMask bit = CGROUP_CONTROLLER_TO_MASK(c);
2357 if (!(supported & bit))
2360 (void) cg_trim(cgroup_controller_to_string(c), path, delete_root);
2367 int cg_mask_to_string(CGroupMask mask, char **ret) {
2368 _cleanup_free_ char *s = NULL;
2369 size_t n = 0, allocated = 0;
2380 for (c = 0; c < _CGROUP_CONTROLLER_MAX; c++) {
2384 if (!(mask & CGROUP_CONTROLLER_TO_MASK(c)))
2387 k = cgroup_controller_to_string(c);
2390 if (!GREEDY_REALLOC(s, allocated, n + space + l + 1))
2395 memcpy(s + n + space, k, l);
2410 int cg_mask_from_string(const char *value, CGroupMask *mask) {
2415 _cleanup_free_ char *n = NULL;
2419 r = extract_first_word(&value, &n, NULL, 0);
2425 v = cgroup_controller_from_string(n);
2429 *mask |= CGROUP_CONTROLLER_TO_MASK(v);
2434 int cg_mask_supported(CGroupMask *ret) {
2435 CGroupMask mask = 0;
2438 /* Determines the mask of supported cgroup controllers. Only
2439 * includes controllers we can make sense of and that are
2440 * actually accessible. */
2442 r = cg_all_unified();
2446 _cleanup_free_ char *root = NULL, *controllers = NULL, *path = NULL;
2448 /* In the unified hierarchy we can read the supported
2449 * and accessible controllers from a the top-level
2450 * cgroup attribute */
2452 r = cg_get_root_path(&root);
2456 r = cg_get_path(SYSTEMD_CGROUP_CONTROLLER, root, "cgroup.controllers", &path);
2460 r = read_one_line_file(path, &controllers);
2464 r = cg_mask_from_string(controllers, &mask);
2468 /* Currently, we support the cpu, memory, io and pids
2469 * controller in the unified hierarchy, mask
2470 * everything else off. */
2471 mask &= CGROUP_MASK_CPU | CGROUP_MASK_MEMORY | CGROUP_MASK_IO | CGROUP_MASK_PIDS;
2476 /* In the legacy hierarchy, we check whether which
2477 * hierarchies are mounted. */
2479 for (c = 0; c < _CGROUP_CONTROLLER_MAX; c++) {
2482 n = cgroup_controller_to_string(c);
2483 if (controller_is_accessible(n) >= 0)
2484 mask |= CGROUP_CONTROLLER_TO_MASK(c);
2492 #if 0 /// UNNEEDED by elogind
2493 int cg_kernel_controllers(Set **ret) {
2494 _cleanup_set_free_free_ Set *controllers = NULL;
2495 _cleanup_fclose_ FILE *f = NULL;
2500 /* Determines the full list of kernel-known controllers. Might
2501 * include controllers we don't actually support, arbitrary
2502 * named hierarchies and controllers that aren't currently
2503 * accessible (because not mounted). */
2505 controllers = set_new(&string_hash_ops);
2509 f = fopen("/proc/cgroups", "re");
2511 if (errno == ENOENT) {
2519 (void) __fsetlocking(f, FSETLOCKING_BYCALLER);
2521 /* Ignore the header line */
2522 (void) read_line(f, (size_t) -1, NULL);
2529 if (fscanf(f, "%ms %*i %*i %i", &controller, &enabled) != 2) {
2534 if (ferror(f) && errno > 0)
2545 if (!cg_controller_is_valid(controller)) {
2550 r = set_consume(controllers, controller);
2562 static thread_local CGroupUnified unified_cache = CGROUP_UNIFIED_UNKNOWN;
2564 /* The hybrid mode was initially implemented in v232 and simply mounted cgroup v2 on /sys/fs/cgroup/systemd. This
2565 * unfortunately broke other tools (such as docker) which expected the v1 "name=systemd" hierarchy on
2566 * /sys/fs/cgroup/systemd. From v233 and on, the hybrid mode mountnbs v2 on /sys/fs/cgroup/unified and maintains
2567 * "name=systemd" hierarchy on /sys/fs/cgroup/systemd for compatibility with other tools.
2569 * To keep live upgrade working, we detect and support v232 layout. When v232 layout is detected, to keep cgroup v2
2570 * process management but disable the compat dual layout, we return %true on
2571 * cg_unified_controller(SYSTEMD_CGROUP_CONTROLLER) and %false on cg_hybrid_unified().
2573 static thread_local bool unified_systemd_v232;
2575 static int cg_unified_update(void) {
2579 /* Checks if we support the unified hierarchy. Returns an
2580 * error when the cgroup hierarchies aren't mounted yet or we
2581 * have any other trouble determining if the unified hierarchy
2584 if (unified_cache >= CGROUP_UNIFIED_NONE)
2587 if (statfs("/sys/fs/cgroup/", &fs) < 0)
2588 return log_debug_errno(errno, "statfs(\"/sys/fs/cgroup/\" failed: %m");
2590 if (F_TYPE_EQUAL(fs.f_type, CGROUP2_SUPER_MAGIC)) {
2591 log_debug("Found cgroup2 on /sys/fs/cgroup/, full unified hierarchy");
2592 unified_cache = CGROUP_UNIFIED_ALL;
2593 #if 0 /// The handling of cgroups is a bit different with elogind
2594 } else if (F_TYPE_EQUAL(fs.f_type, TMPFS_MAGIC)) {
2595 log_debug("Found cgroup2 on /sys/fs/cgroup/unified, unified hierarchy for systemd controller");
2597 } else if (F_TYPE_EQUAL(fs.f_type, CGROUP_SUPER_MAGIC)
2598 || F_TYPE_EQUAL(fs.f_type, TMPFS_MAGIC)) {
2600 if (statfs("/sys/fs/cgroup/unified/", &fs) == 0 &&
2601 F_TYPE_EQUAL(fs.f_type, CGROUP2_SUPER_MAGIC)) {
2602 unified_cache = CGROUP_UNIFIED_SYSTEMD;
2603 unified_systemd_v232 = false;
2605 #if 0 /// There is no sub-grouping within elogind
2606 if (statfs("/sys/fs/cgroup/systemd/", &fs) < 0)
2607 return log_debug_errno(errno, "statfs(\"/sys/fs/cgroup/systemd\" failed: %m");
2609 if (F_TYPE_EQUAL(fs.f_type, CGROUP2_SUPER_MAGIC)) {
2610 log_debug("Found cgroup2 on /sys/fs/cgroup/systemd, unified hierarchy for systemd controller (v232 variant)");
2611 unified_cache = CGROUP_UNIFIED_SYSTEMD;
2612 unified_systemd_v232 = true;
2613 } else if (F_TYPE_EQUAL(fs.f_type, CGROUP_SUPER_MAGIC)) {
2614 log_debug("Found cgroup on /sys/fs/cgroup/systemd, legacy hierarchy");
2615 unified_cache = CGROUP_UNIFIED_NONE;
2617 log_debug("Unexpected filesystem type %llx mounted on /sys/fs/cgroup/systemd, assuming legacy hierarchy",
2618 (unsigned long long) fs.f_type);
2619 unified_cache = CGROUP_UNIFIED_NONE;
2622 unified_cache = CGROUP_UNIFIED_NONE;
2626 log_debug("Unknown filesystem type %llx mounted on /sys/fs/cgroup.",
2627 (unsigned long long) fs.f_type);
2634 int cg_unified_controller(const char *controller) {
2637 r = cg_unified_update();
2641 if (unified_cache == CGROUP_UNIFIED_NONE)
2644 if (unified_cache >= CGROUP_UNIFIED_ALL)
2647 #if 0 /// only if elogind is the controller we can use cgroups2 in hybrid mode
2648 return streq_ptr(controller, SYSTEMD_CGROUP_CONTROLLER);
2650 return streq_ptr(controller, SYSTEMD_CGROUP_CONTROLLER_HYBRID);
2654 int cg_all_unified(void) {
2657 r = cg_unified_update();
2661 return unified_cache >= CGROUP_UNIFIED_ALL;
2664 int cg_hybrid_unified(void) {
2667 r = cg_unified_update();
2671 return unified_cache == CGROUP_UNIFIED_SYSTEMD && !unified_systemd_v232;
2674 int cg_unified_flush(void) {
2675 unified_cache = CGROUP_UNIFIED_UNKNOWN;
2677 return cg_unified_update();
2680 #if 0 /// UNNEEDED by elogind
2681 int cg_enable_everywhere(CGroupMask supported, CGroupMask mask, const char *p) {
2682 _cleanup_fclose_ FILE *f = NULL;
2683 _cleanup_free_ char *fs = NULL;
2692 r = cg_all_unified();
2695 if (r == 0) /* on the legacy hiearchy there's no joining of controllers defined */
2698 r = cg_get_path(SYSTEMD_CGROUP_CONTROLLER, p, "cgroup.subtree_control", &fs);
2702 for (c = 0; c < _CGROUP_CONTROLLER_MAX; c++) {
2703 CGroupMask bit = CGROUP_CONTROLLER_TO_MASK(c);
2706 if (!(supported & bit))
2709 n = cgroup_controller_to_string(c);
2711 char s[1 + strlen(n) + 1];
2713 s[0] = mask & bit ? '+' : '-';
2717 f = fopen(fs, "we");
2719 log_debug_errno(errno, "Failed to open cgroup.subtree_control file of %s: %m", p);
2724 r = write_string_stream(f, s, 0);
2726 log_debug_errno(r, "Failed to enable controller %s for %s (%s): %m", n, p, fs);
2734 bool cg_is_unified_wanted(void) {
2735 static thread_local int wanted = -1;
2738 const bool is_default = DEFAULT_HIERARCHY == CGROUP_UNIFIED_ALL;
2740 /* If we have a cached value, return that. */
2744 /* If the hierarchy is already mounted, then follow whatever
2745 * was chosen for it. */
2746 if (cg_unified_flush() >= 0)
2747 return (wanted = unified_cache >= CGROUP_UNIFIED_ALL);
2749 #if 0 /// elogind is not init and has no business with kernel command line
2750 /* Otherwise, let's see what the kernel command line has to say.
2751 * Since checking is expensive, cache a non-error result. */
2752 r = proc_cmdline_get_bool("systemd.unified_cgroup_hierarchy", &b);
2755 return (wanted = r > 0 ? b : is_default);
2758 bool cg_is_legacy_wanted(void) {
2759 static thread_local int wanted = -1;
2761 /* If we have a cached value, return that. */
2765 /* Check if we have cgroups2 already mounted. */
2766 if (cg_unified_flush() >= 0 &&
2767 unified_cache == CGROUP_UNIFIED_ALL)
2768 return (wanted = false);
2770 /* Otherwise, assume that at least partial legacy is wanted,
2771 * since cgroups2 should already be mounted at this point. */
2772 return (wanted = true);
2775 bool cg_is_hybrid_wanted(void) {
2776 static thread_local int wanted = -1;
2779 const bool is_default = DEFAULT_HIERARCHY >= CGROUP_UNIFIED_SYSTEMD;
2780 /* We default to true if the default is "hybrid", obviously,
2781 * but also when the default is "unified", because if we get
2782 * called, it means that unified hierarchy was not mounted. */
2784 /* If we have a cached value, return that. */
2788 /* If the hierarchy is already mounted, then follow whatever
2789 * was chosen for it. */
2790 if (cg_unified_flush() >= 0 &&
2791 unified_cache == CGROUP_UNIFIED_ALL)
2792 return (wanted = false);
2794 #if 0 /// elogind is not init and has no business with kernel command line
2795 /* Otherwise, let's see what the kernel command line has to say.
2796 * Since checking is expensive, cache a non-error result. */
2797 r = proc_cmdline_get_bool("systemd.legacy_systemd_cgroup_controller", &b);
2800 /* The meaning of the kernel option is reversed wrt. to the return value
2801 * of this function, hence the negation. */
2802 return (wanted = r > 0 ? !b : is_default);
2805 #if 0 /// UNNEEDED by elogind
2806 int cg_weight_parse(const char *s, uint64_t *ret) {
2811 *ret = CGROUP_WEIGHT_INVALID;
2815 r = safe_atou64(s, &u);
2819 if (u < CGROUP_WEIGHT_MIN || u > CGROUP_WEIGHT_MAX)
2826 const uint64_t cgroup_io_limit_defaults[_CGROUP_IO_LIMIT_TYPE_MAX] = {
2827 [CGROUP_IO_RBPS_MAX] = CGROUP_LIMIT_MAX,
2828 [CGROUP_IO_WBPS_MAX] = CGROUP_LIMIT_MAX,
2829 [CGROUP_IO_RIOPS_MAX] = CGROUP_LIMIT_MAX,
2830 [CGROUP_IO_WIOPS_MAX] = CGROUP_LIMIT_MAX,
2833 static const char* const cgroup_io_limit_type_table[_CGROUP_IO_LIMIT_TYPE_MAX] = {
2834 [CGROUP_IO_RBPS_MAX] = "IOReadBandwidthMax",
2835 [CGROUP_IO_WBPS_MAX] = "IOWriteBandwidthMax",
2836 [CGROUP_IO_RIOPS_MAX] = "IOReadIOPSMax",
2837 [CGROUP_IO_WIOPS_MAX] = "IOWriteIOPSMax",
2840 DEFINE_STRING_TABLE_LOOKUP(cgroup_io_limit_type, CGroupIOLimitType);
2842 int cg_cpu_shares_parse(const char *s, uint64_t *ret) {
2847 *ret = CGROUP_CPU_SHARES_INVALID;
2851 r = safe_atou64(s, &u);
2855 if (u < CGROUP_CPU_SHARES_MIN || u > CGROUP_CPU_SHARES_MAX)
2862 int cg_blkio_weight_parse(const char *s, uint64_t *ret) {
2867 *ret = CGROUP_BLKIO_WEIGHT_INVALID;
2871 r = safe_atou64(s, &u);
2875 if (u < CGROUP_BLKIO_WEIGHT_MIN || u > CGROUP_BLKIO_WEIGHT_MAX)
2883 bool is_cgroup_fs(const struct statfs *s) {
2884 return is_fs_type(s, CGROUP_SUPER_MAGIC) ||
2885 is_fs_type(s, CGROUP2_SUPER_MAGIC);
2888 bool fd_is_cgroup_fs(int fd) {
2891 if (fstatfs(fd, &s) < 0)
2894 return is_cgroup_fs(&s);
2897 static const char *cgroup_controller_table[_CGROUP_CONTROLLER_MAX] = {
2898 [CGROUP_CONTROLLER_CPU] = "cpu",
2899 [CGROUP_CONTROLLER_CPUACCT] = "cpuacct",
2900 [CGROUP_CONTROLLER_IO] = "io",
2901 [CGROUP_CONTROLLER_BLKIO] = "blkio",
2902 [CGROUP_CONTROLLER_MEMORY] = "memory",
2903 [CGROUP_CONTROLLER_DEVICES] = "devices",
2904 [CGROUP_CONTROLLER_PIDS] = "pids",
2907 DEFINE_STRING_TABLE_LOOKUP(cgroup_controller, CGroupController);