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"
50 #include "parse-util.h"
51 #include "path-util.h"
52 #include "proc-cmdline.h"
53 #include "process-util.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 bool cg_ns_supported(void) {
145 static thread_local
int enabled
= -1;
150 if (access("/proc/self/ns/cgroup", F_OK
) == 0)
158 int cg_enumerate_subgroups(const char *controller
, const char *path
, DIR **_d
) {
159 _cleanup_free_
char *fs
= NULL
;
165 /* This is not recursive! */
167 r
= cg_get_path(controller
, path
, NULL
, &fs
);
179 int cg_read_subgroup(DIR *d
, char **fn
) {
185 FOREACH_DIRENT_ALL(de
, d
, return -errno
) {
188 if (de
->d_type
!= DT_DIR
)
191 if (dot_or_dot_dot(de
->d_name
))
194 b
= strdup(de
->d_name
);
205 int cg_rmdir(const char *controller
, const char *path
) {
206 _cleanup_free_
char *p
= NULL
;
209 r
= cg_get_path(controller
, path
, NULL
, &p
);
214 if (r
< 0 && errno
!= ENOENT
)
217 r
= cg_hybrid_unified();
223 if (streq(controller
, SYSTEMD_CGROUP_CONTROLLER
)) {
224 r
= cg_rmdir(SYSTEMD_CGROUP_CONTROLLER_LEGACY
, path
);
226 log_warning_errno(r
, "Failed to remove compat systemd cgroup %s: %m", path
);
233 const char *controller
,
238 cg_kill_log_func_t log_kill
,
241 _cleanup_set_free_ Set
*allocated_set
= NULL
;
248 /* Don't send SIGCONT twice. Also, SIGKILL always works even when process is suspended, hence don't send
249 * SIGCONT on SIGKILL. */
250 if (IN_SET(sig
, SIGCONT
, SIGKILL
))
251 flags
&= ~CGROUP_SIGCONT
;
253 /* This goes through the tasks list and kills them all. This
254 * is repeated until no further processes are added to the
255 * tasks list, to properly handle forking processes */
258 s
= allocated_set
= set_new(NULL
);
263 my_pid
= getpid_cached();
266 _cleanup_fclose_
FILE *f
= NULL
;
270 r
= cg_enumerate_processes(controller
, path
, &f
);
272 if (ret
>= 0 && r
!= -ENOENT
)
278 while ((r
= cg_read_pid(f
, &pid
)) > 0) {
280 if ((flags
& CGROUP_IGNORE_SELF
) && pid
== my_pid
)
283 if (set_get(s
, PID_TO_PTR(pid
)) == PID_TO_PTR(pid
))
287 log_kill(pid
, sig
, userdata
);
289 /* If we haven't killed this process yet, kill
291 if (kill(pid
, sig
) < 0) {
292 if (ret
>= 0 && errno
!= ESRCH
)
295 if (flags
& CGROUP_SIGCONT
)
296 (void) kill(pid
, SIGCONT
);
304 r
= set_put(s
, PID_TO_PTR(pid
));
320 /* To avoid racing against processes which fork
321 * quicker than we can kill them we repeat this until
322 * no new pids need to be killed. */
329 int cg_kill_recursive(
330 const char *controller
,
335 cg_kill_log_func_t log_kill
,
338 _cleanup_set_free_ Set
*allocated_set
= NULL
;
339 _cleanup_closedir_
DIR *d
= NULL
;
347 s
= allocated_set
= set_new(NULL
);
352 ret
= cg_kill(controller
, path
, sig
, flags
, s
, log_kill
, userdata
);
354 r
= cg_enumerate_subgroups(controller
, path
, &d
);
356 if (ret
>= 0 && r
!= -ENOENT
)
362 while ((r
= cg_read_subgroup(d
, &fn
)) > 0) {
363 _cleanup_free_
char *p
= NULL
;
365 p
= strjoin(path
, "/", fn
);
370 r
= cg_kill_recursive(controller
, p
, sig
, flags
, s
, log_kill
, userdata
);
371 if (r
!= 0 && ret
>= 0)
374 if (ret
>= 0 && r
< 0)
377 if (flags
& CGROUP_REMOVE
) {
378 r
= cg_rmdir(controller
, path
);
379 if (r
< 0 && ret
>= 0 && !IN_SET(r
, -ENOENT
, -EBUSY
))
394 _cleanup_set_free_ Set
*s
= NULL
;
407 my_pid
= getpid_cached();
410 _cleanup_fclose_
FILE *f
= NULL
;
414 r
= cg_enumerate_processes(cfrom
, pfrom
, &f
);
416 if (ret
>= 0 && r
!= -ENOENT
)
422 while ((r
= cg_read_pid(f
, &pid
)) > 0) {
424 /* This might do weird stuff if we aren't a
425 * single-threaded program. However, we
426 * luckily know we are not */
427 if ((flags
& CGROUP_IGNORE_SELF
) && pid
== my_pid
)
430 if (set_get(s
, PID_TO_PTR(pid
)) == PID_TO_PTR(pid
))
433 /* Ignore kernel threads. Since they can only
434 * exist in the root cgroup, we only check for
437 (isempty(pfrom
) || path_equal(pfrom
, "/")) &&
438 is_kernel_thread(pid
) > 0)
441 r
= cg_attach(cto
, pto
, pid
);
443 if (ret
>= 0 && r
!= -ESRCH
)
450 r
= set_put(s
, PID_TO_PTR(pid
));
470 int cg_migrate_recursive(
477 _cleanup_closedir_
DIR *d
= NULL
;
486 ret
= cg_migrate(cfrom
, pfrom
, cto
, pto
, flags
);
488 r
= cg_enumerate_subgroups(cfrom
, pfrom
, &d
);
490 if (ret
>= 0 && r
!= -ENOENT
)
496 while ((r
= cg_read_subgroup(d
, &fn
)) > 0) {
497 _cleanup_free_
char *p
= NULL
;
499 p
= strjoin(pfrom
, "/", fn
);
504 r
= cg_migrate_recursive(cfrom
, p
, cto
, pto
, flags
);
505 if (r
!= 0 && ret
>= 0)
509 if (r
< 0 && ret
>= 0)
512 if (flags
& CGROUP_REMOVE
) {
513 r
= cg_rmdir(cfrom
, pfrom
);
514 if (r
< 0 && ret
>= 0 && !IN_SET(r
, -ENOENT
, -EBUSY
))
521 int cg_migrate_recursive_fallback(
535 r
= cg_migrate_recursive(cfrom
, pfrom
, cto
, pto
, flags
);
537 char prefix
[strlen(pto
) + 1];
539 /* This didn't work? Then let's try all prefixes of the destination */
541 PATH_FOREACH_PREFIX(prefix
, pto
) {
544 q
= cg_migrate_recursive(cfrom
, pfrom
, cto
, prefix
, flags
);
553 static const char *controller_to_dirname(const char *controller
) {
558 /* Converts a controller name to the directory name below
559 * /sys/fs/cgroup/ we want to mount it to. Effectively, this
560 * just cuts off the name= prefixed used for named
561 * hierarchies, if it is specified. */
563 if (streq(controller
, SYSTEMD_CGROUP_CONTROLLER
)) {
564 if (cg_hybrid_unified() > 0)
565 controller
= SYSTEMD_CGROUP_CONTROLLER_HYBRID
;
567 controller
= SYSTEMD_CGROUP_CONTROLLER_LEGACY
;
570 e
= startswith(controller
, "name=");
577 static int join_path_legacy(const char *controller
, const char *path
, const char *suffix
, char **fs
) {
584 dn
= controller_to_dirname(controller
);
586 if (isempty(path
) && isempty(suffix
))
587 t
= strappend("/sys/fs/cgroup/", dn
);
588 else if (isempty(path
))
589 t
= strjoin("/sys/fs/cgroup/", dn
, "/", suffix
);
590 else if (isempty(suffix
))
591 t
= strjoin("/sys/fs/cgroup/", dn
, "/", path
);
593 t
= strjoin("/sys/fs/cgroup/", dn
, "/", path
, "/", suffix
);
601 static int join_path_unified(const char *path
, const char *suffix
, char **fs
) {
606 if (isempty(path
) && isempty(suffix
))
607 t
= strdup("/sys/fs/cgroup");
608 else if (isempty(path
))
609 t
= strappend("/sys/fs/cgroup/", suffix
);
610 else if (isempty(suffix
))
611 t
= strappend("/sys/fs/cgroup/", path
);
613 t
= strjoin("/sys/fs/cgroup/", path
, "/", suffix
);
621 int cg_get_path(const char *controller
, const char *path
, const char *suffix
, char **fs
) {
629 /* If no controller is specified, we return the path
630 * *below* the controllers, without any prefix. */
632 if (!path
&& !suffix
)
640 t
= strjoin(path
, "/", suffix
);
644 *fs
= path_kill_slashes(t
);
648 if (!cg_controller_is_valid(controller
))
651 r
= cg_all_unified();
655 r
= join_path_unified(path
, suffix
, fs
);
657 r
= join_path_legacy(controller
, path
, suffix
, fs
);
661 path_kill_slashes(*fs
);
665 static int controller_is_accessible(const char *controller
) {
670 /* Checks whether a specific controller is accessible,
671 * i.e. its hierarchy mounted. In the unified hierarchy all
672 * controllers are considered accessible, except for the named
675 if (!cg_controller_is_valid(controller
))
678 r
= cg_all_unified();
682 /* We don't support named hierarchies if we are using
683 * the unified hierarchy. */
685 if (streq(controller
, SYSTEMD_CGROUP_CONTROLLER
))
688 if (startswith(controller
, "name="))
694 dn
= controller_to_dirname(controller
);
695 cc
= strjoina("/sys/fs/cgroup/", dn
);
697 if (laccess(cc
, F_OK
) < 0)
704 int cg_get_path_and_check(const char *controller
, const char *path
, const char *suffix
, char **fs
) {
710 /* Check if the specified controller is actually accessible */
711 r
= controller_is_accessible(controller
);
715 return cg_get_path(controller
, path
, suffix
, fs
);
718 static int trim_cb(const char *path
, const struct stat
*sb
, int typeflag
, struct FTW
*ftwbuf
) {
723 if (typeflag
!= FTW_DP
)
726 if (ftwbuf
->level
< 1)
733 int cg_trim(const char *controller
, const char *path
, bool delete_root
) {
734 _cleanup_free_
char *fs
= NULL
;
739 r
= cg_get_path(controller
, path
, NULL
, &fs
);
744 if (nftw(fs
, trim_cb
, 64, FTW_DEPTH
|FTW_MOUNT
|FTW_PHYS
) != 0) {
754 if (rmdir(fs
) < 0 && errno
!= ENOENT
)
758 q
= cg_hybrid_unified();
761 if (q
> 0 && streq(controller
, SYSTEMD_CGROUP_CONTROLLER
)) {
762 q
= cg_trim(SYSTEMD_CGROUP_CONTROLLER_LEGACY
, path
, delete_root
);
764 log_warning_errno(q
, "Failed to trim compat systemd cgroup %s: %m", path
);
770 int cg_create(const char *controller
, const char *path
) {
771 _cleanup_free_
char *fs
= NULL
;
774 r
= cg_get_path_and_check(controller
, path
, NULL
, &fs
);
778 r
= mkdir_parents(fs
, 0755);
782 r
= mkdir_errno_wrapper(fs
, 0755);
788 r
= cg_hybrid_unified();
792 if (r
> 0 && streq(controller
, SYSTEMD_CGROUP_CONTROLLER
)) {
793 r
= cg_create(SYSTEMD_CGROUP_CONTROLLER_LEGACY
, path
);
795 log_warning_errno(r
, "Failed to create compat systemd cgroup %s: %m", path
);
801 int cg_create_and_attach(const char *controller
, const char *path
, pid_t pid
) {
806 r
= cg_create(controller
, path
);
810 q
= cg_attach(controller
, path
, pid
);
814 /* This does not remove the cgroup on failure */
818 int cg_attach(const char *controller
, const char *path
, pid_t pid
) {
819 _cleanup_free_
char *fs
= NULL
;
820 char c
[DECIMAL_STR_MAX(pid_t
) + 2];
826 r
= cg_get_path_and_check(controller
, path
, "cgroup.procs", &fs
);
831 pid
= getpid_cached();
833 xsprintf(c
, PID_FMT
"\n", pid
);
835 r
= write_string_file(fs
, c
, 0);
839 r
= cg_hybrid_unified();
843 if (r
> 0 && streq(controller
, SYSTEMD_CGROUP_CONTROLLER
)) {
844 r
= cg_attach(SYSTEMD_CGROUP_CONTROLLER_LEGACY
, path
, pid
);
846 log_warning_errno(r
, "Failed to attach "PID_FMT
" to compat systemd cgroup %s: %m", pid
, path
);
852 int cg_attach_fallback(const char *controller
, const char *path
, pid_t pid
) {
859 r
= cg_attach(controller
, path
, pid
);
861 char prefix
[strlen(path
) + 1];
863 /* This didn't work? Then let's try all prefixes of
866 PATH_FOREACH_PREFIX(prefix
, path
) {
869 q
= cg_attach(controller
, prefix
, pid
);
879 const char *controller
,
889 /* cgroupsv1, aka legacy/non-unified */
890 static const struct Attribute legacy_attributes
[] = {
891 { "cgroup.procs", true },
893 { "cgroup.clone_children", false },
897 /* cgroupsv2, aka unified */
898 static const struct Attribute unified_attributes
[] = {
899 { "cgroup.procs", true },
900 { "cgroup.subtree_control", true },
901 { "cgroup.threads", false },
905 static const struct Attribute
* const attributes
[] = {
906 [false] = legacy_attributes
,
907 [true] = unified_attributes
,
910 _cleanup_free_
char *fs
= NULL
;
911 const struct Attribute
*i
;
916 if (uid
== UID_INVALID
&& gid
== GID_INVALID
)
919 unified
= cg_unified_controller(controller
);
923 /* Configure access to the cgroup itself */
924 r
= cg_get_path(controller
, path
, NULL
, &fs
);
928 r
= chmod_and_chown(fs
, 0755, uid
, gid
);
932 /* Configure access to the cgroup's attributes */
933 for (i
= attributes
[unified
]; i
->name
; i
++) {
936 r
= cg_get_path(controller
, path
, i
->name
, &fs
);
940 r
= chmod_and_chown(fs
, 0644, uid
, gid
);
945 log_debug_errno(r
, "Failed to set access on cgroup %s, ignoring: %m", fs
);
949 if (streq(controller
, SYSTEMD_CGROUP_CONTROLLER
)) {
950 r
= cg_hybrid_unified();
954 /* Always propagate access mode from unified to legacy controller */
955 r
= cg_set_access(SYSTEMD_CGROUP_CONTROLLER_LEGACY
, path
, uid
, gid
);
957 log_debug_errno(r
, "Failed to set access on compatibility systemd cgroup %s, ignoring: %m", path
);
964 int cg_set_xattr(const char *controller
, const char *path
, const char *name
, const void *value
, size_t size
, int flags
) {
965 _cleanup_free_
char *fs
= NULL
;
970 assert(value
|| size
<= 0);
972 r
= cg_get_path(controller
, path
, NULL
, &fs
);
976 if (setxattr(fs
, name
, value
, size
, flags
) < 0)
982 int cg_get_xattr(const char *controller
, const char *path
, const char *name
, void *value
, size_t size
) {
983 _cleanup_free_
char *fs
= NULL
;
990 r
= cg_get_path(controller
, path
, NULL
, &fs
);
994 n
= getxattr(fs
, name
, value
, size
);
1001 int cg_pid_get_path(const char *controller
, pid_t pid
, char **path
) {
1002 _cleanup_fclose_
FILE *f
= NULL
;
1003 char line
[LINE_MAX
];
1004 const char *fs
, *controller_str
;
1012 if (!cg_controller_is_valid(controller
))
1015 controller
= SYSTEMD_CGROUP_CONTROLLER
;
1017 unified
= cg_unified_controller(controller
);
1021 if (streq(controller
, SYSTEMD_CGROUP_CONTROLLER
))
1022 controller_str
= SYSTEMD_CGROUP_CONTROLLER_LEGACY
;
1024 controller_str
= controller
;
1026 cs
= strlen(controller_str
);
1029 fs
= procfs_file_alloca(pid
, "cgroup");
1030 f
= fopen(fs
, "re");
1032 return errno
== ENOENT
? -ESRCH
: -errno
;
1034 (void) __fsetlocking(f
, FSETLOCKING_BYCALLER
);
1036 FOREACH_LINE(line
, f
, return -errno
) {
1042 e
= startswith(line
, "0:");
1052 const char *word
, *state
;
1055 l
= strchr(line
, ':');
1065 FOREACH_WORD_SEPARATOR(word
, k
, l
, ",", state
)
1066 if (k
== cs
&& memcmp(word
, controller_str
, cs
) == 0) {
1078 /* Truncate suffix indicating the process is a zombie */
1079 e
= endswith(p
, " (deleted)");
1090 int cg_install_release_agent(const char *controller
, const char *agent
) {
1091 _cleanup_free_
char *fs
= NULL
, *contents
= NULL
;
1097 r
= cg_unified_controller(controller
);
1100 if (r
> 0) /* doesn't apply to unified hierarchy */
1103 r
= cg_get_path(controller
, NULL
, "release_agent", &fs
);
1107 r
= read_one_line_file(fs
, &contents
);
1111 sc
= strstrip(contents
);
1113 r
= write_string_file(fs
, agent
, 0);
1116 } else if (!path_equal(sc
, agent
))
1120 r
= cg_get_path(controller
, NULL
, "notify_on_release", &fs
);
1124 contents
= mfree(contents
);
1125 r
= read_one_line_file(fs
, &contents
);
1129 sc
= strstrip(contents
);
1130 if (streq(sc
, "0")) {
1131 r
= write_string_file(fs
, "1", 0);
1138 if (!streq(sc
, "1"))
1144 int cg_uninstall_release_agent(const char *controller
) {
1145 _cleanup_free_
char *fs
= NULL
;
1148 r
= cg_unified_controller(controller
);
1151 if (r
> 0) /* Doesn't apply to unified hierarchy */
1154 r
= cg_get_path(controller
, NULL
, "notify_on_release", &fs
);
1158 r
= write_string_file(fs
, "0", 0);
1164 r
= cg_get_path(controller
, NULL
, "release_agent", &fs
);
1168 r
= write_string_file(fs
, "", 0);
1175 int cg_is_empty(const char *controller
, const char *path
) {
1176 _cleanup_fclose_
FILE *f
= NULL
;
1182 r
= cg_enumerate_processes(controller
, path
, &f
);
1188 r
= cg_read_pid(f
, &pid
);
1195 int cg_is_empty_recursive(const char *controller
, const char *path
) {
1200 /* The root cgroup is always populated */
1201 if (controller
&& (isempty(path
) || path_equal(path
, "/")))
1204 r
= cg_unified_controller(controller
);
1208 _cleanup_free_
char *t
= NULL
;
1210 /* On the unified hierarchy we can check empty state
1211 * via the "populated" attribute of "cgroup.events". */
1213 r
= cg_read_event(controller
, path
, "populated", &t
);
1217 return streq(t
, "0");
1219 _cleanup_closedir_
DIR *d
= NULL
;
1222 r
= cg_is_empty(controller
, path
);
1226 r
= cg_enumerate_subgroups(controller
, path
, &d
);
1232 while ((r
= cg_read_subgroup(d
, &fn
)) > 0) {
1233 _cleanup_free_
char *p
= NULL
;
1235 p
= strjoin(path
, "/", fn
);
1240 r
= cg_is_empty_recursive(controller
, p
);
1251 int cg_split_spec(const char *spec
, char **controller
, char **path
) {
1252 char *t
= NULL
, *u
= NULL
;
1258 if (!path_is_normalized(spec
))
1266 *path
= path_kill_slashes(t
);
1275 e
= strchr(spec
, ':');
1277 if (!cg_controller_is_valid(spec
))
1294 t
= strndup(spec
, e
-spec
);
1297 if (!cg_controller_is_valid(t
)) {
1311 if (!path_is_normalized(u
) ||
1312 !path_is_absolute(u
)) {
1318 path_kill_slashes(u
);
1334 int cg_mangle_path(const char *path
, char **result
) {
1335 _cleanup_free_
char *c
= NULL
, *p
= NULL
;
1342 /* First, check if it already is a filesystem path */
1343 if (path_startswith(path
, "/sys/fs/cgroup")) {
1349 *result
= path_kill_slashes(t
);
1353 /* Otherwise, treat it as cg spec */
1354 r
= cg_split_spec(path
, &c
, &p
);
1358 return cg_get_path(c
?: SYSTEMD_CGROUP_CONTROLLER
, p
?: "/", NULL
, result
);
1361 int cg_get_root_path(char **path
) {
1367 r
= cg_pid_get_path(SYSTEMD_CGROUP_CONTROLLER
, 1, &p
);
1371 e
= endswith(p
, "/" SPECIAL_INIT_SCOPE
);
1373 e
= endswith(p
, "/" SPECIAL_SYSTEM_SLICE
); /* legacy */
1375 e
= endswith(p
, "/system"); /* even more legacy */
1383 int cg_shift_path(const char *cgroup
, const char *root
, const char **shifted
) {
1384 _cleanup_free_
char *rt
= NULL
;
1392 /* If the root was specified let's use that, otherwise
1393 * let's determine it from PID 1 */
1395 r
= cg_get_root_path(&rt
);
1402 p
= path_startswith(cgroup
, root
);
1403 if (p
&& p
> cgroup
)
1411 int cg_pid_get_path_shifted(pid_t pid
, const char *root
, char **cgroup
) {
1412 _cleanup_free_
char *raw
= NULL
;
1419 r
= cg_pid_get_path(SYSTEMD_CGROUP_CONTROLLER
, pid
, &raw
);
1423 r
= cg_shift_path(raw
, root
, &c
);
1443 int cg_path_decode_unit(const char *cgroup
, char **unit
) {
1450 n
= strcspn(cgroup
, "/");
1454 c
= strndupa(cgroup
, n
);
1457 if (!unit_name_is_valid(c
, UNIT_NAME_PLAIN
|UNIT_NAME_INSTANCE
))
1468 static bool valid_slice_name(const char *p
, size_t n
) {
1473 if (n
< STRLEN("x.slice"))
1476 if (memcmp(p
+ n
- 6, ".slice", 6) == 0) {
1482 c
= cg_unescape(buf
);
1484 return unit_name_is_valid(c
, UNIT_NAME_PLAIN
);
1490 static const char *skip_slices(const char *p
) {
1493 /* Skips over all slice assignments */
1498 p
+= strspn(p
, "/");
1500 n
= strcspn(p
, "/");
1501 if (!valid_slice_name(p
, n
))
1508 int cg_path_get_unit(const char *path
, char **ret
) {
1516 e
= skip_slices(path
);
1518 r
= cg_path_decode_unit(e
, &unit
);
1522 /* We skipped over the slices, don't accept any now */
1523 if (endswith(unit
, ".slice")) {
1532 int cg_pid_get_unit(pid_t pid
, char **unit
) {
1533 _cleanup_free_
char *cgroup
= NULL
;
1538 r
= cg_pid_get_path_shifted(pid
, NULL
, &cgroup
);
1542 return cg_path_get_unit(cgroup
, unit
);
1546 * Skip session-*.scope, but require it to be there.
1548 static const char *skip_session(const char *p
) {
1554 p
+= strspn(p
, "/");
1556 n
= strcspn(p
, "/");
1557 if (n
< STRLEN("session-x.scope"))
1560 if (memcmp(p
, "session-", 8) == 0 && memcmp(p
+ n
- 6, ".scope", 6) == 0) {
1561 char buf
[n
- 8 - 6 + 1];
1563 memcpy(buf
, p
+ 8, n
- 8 - 6);
1566 /* Note that session scopes never need unescaping,
1567 * since they cannot conflict with the kernel's own
1568 * names, hence we don't need to call cg_unescape()
1571 if (!session_id_valid(buf
))
1575 p
+= strspn(p
, "/");
1583 * Skip user@*.service, but require it to be there.
1585 static const char *skip_user_manager(const char *p
) {
1591 p
+= strspn(p
, "/");
1593 n
= strcspn(p
, "/");
1594 if (n
< STRLEN("user@x.service"))
1597 if (memcmp(p
, "user@", 5) == 0 && memcmp(p
+ n
- 8, ".service", 8) == 0) {
1598 char buf
[n
- 5 - 8 + 1];
1600 memcpy(buf
, p
+ 5, n
- 5 - 8);
1603 /* Note that user manager services never need unescaping,
1604 * since they cannot conflict with the kernel's own
1605 * names, hence we don't need to call cg_unescape()
1608 if (parse_uid(buf
, NULL
) < 0)
1612 p
+= strspn(p
, "/");
1620 static const char *skip_user_prefix(const char *path
) {
1625 /* Skip slices, if there are any */
1626 e
= skip_slices(path
);
1628 /* Skip the user manager, if it's in the path now... */
1629 t
= skip_user_manager(e
);
1633 /* Alternatively skip the user session if it is in the path... */
1634 return skip_session(e
);
1637 int cg_path_get_user_unit(const char *path
, char **ret
) {
1643 t
= skip_user_prefix(path
);
1647 /* And from here on it looks pretty much the same as for a
1648 * system unit, hence let's use the same parser from here
1650 return cg_path_get_unit(t
, ret
);
1653 int cg_pid_get_user_unit(pid_t pid
, char **unit
) {
1654 _cleanup_free_
char *cgroup
= NULL
;
1659 r
= cg_pid_get_path_shifted(pid
, NULL
, &cgroup
);
1663 return cg_path_get_user_unit(cgroup
, unit
);
1666 int cg_path_get_machine_name(const char *path
, char **machine
) {
1667 _cleanup_free_
char *u
= NULL
;
1671 r
= cg_path_get_unit(path
, &u
);
1675 sl
= strjoina("/run/systemd/machines/unit:", u
);
1676 return readlink_malloc(sl
, machine
);
1679 int cg_pid_get_machine_name(pid_t pid
, char **machine
) {
1680 _cleanup_free_
char *cgroup
= NULL
;
1685 r
= cg_pid_get_path_shifted(pid
, NULL
, &cgroup
);
1689 return cg_path_get_machine_name(cgroup
, machine
);
1692 int cg_path_get_session(const char *path
, char **session
) {
1693 _cleanup_free_
char *unit
= NULL
;
1699 r
= cg_path_get_unit(path
, &unit
);
1703 start
= startswith(unit
, "session-");
1706 end
= endswith(start
, ".scope");
1711 if (!session_id_valid(start
))
1727 int cg_pid_get_session(pid_t pid
, char **session
) {
1728 _cleanup_free_
char *cgroup
= NULL
;
1731 r
= cg_pid_get_path_shifted(pid
, NULL
, &cgroup
);
1735 return cg_path_get_session(cgroup
, session
);
1738 int cg_path_get_owner_uid(const char *path
, uid_t
*uid
) {
1739 _cleanup_free_
char *slice
= NULL
;
1745 r
= cg_path_get_slice(path
, &slice
);
1749 start
= startswith(slice
, "user-");
1752 end
= endswith(start
, ".slice");
1757 if (parse_uid(start
, uid
) < 0)
1763 int cg_pid_get_owner_uid(pid_t pid
, uid_t
*uid
) {
1764 _cleanup_free_
char *cgroup
= NULL
;
1767 r
= cg_pid_get_path_shifted(pid
, NULL
, &cgroup
);
1771 return cg_path_get_owner_uid(cgroup
, uid
);
1774 int cg_path_get_slice(const char *p
, char **slice
) {
1775 const char *e
= NULL
;
1780 /* Finds the right-most slice unit from the beginning, but
1781 * stops before we come to the first non-slice unit. */
1786 p
+= strspn(p
, "/");
1788 n
= strcspn(p
, "/");
1789 if (!valid_slice_name(p
, n
)) {
1794 s
= strdup(SPECIAL_ROOT_SLICE
);
1802 return cg_path_decode_unit(e
, slice
);
1810 int cg_pid_get_slice(pid_t pid
, char **slice
) {
1811 _cleanup_free_
char *cgroup
= NULL
;
1816 r
= cg_pid_get_path_shifted(pid
, NULL
, &cgroup
);
1820 return cg_path_get_slice(cgroup
, slice
);
1823 int cg_path_get_user_slice(const char *p
, char **slice
) {
1828 t
= skip_user_prefix(p
);
1832 /* And now it looks pretty much the same as for a system
1833 * slice, so let's just use the same parser from here on. */
1834 return cg_path_get_slice(t
, slice
);
1837 int cg_pid_get_user_slice(pid_t pid
, char **slice
) {
1838 _cleanup_free_
char *cgroup
= NULL
;
1843 r
= cg_pid_get_path_shifted(pid
, NULL
, &cgroup
);
1847 return cg_path_get_user_slice(cgroup
, slice
);
1850 char *cg_escape(const char *p
) {
1851 bool need_prefix
= false;
1853 /* This implements very minimal escaping for names to be used
1854 * as file names in the cgroup tree: any name which might
1855 * conflict with a kernel name or is prefixed with '_' is
1856 * prefixed with a '_'. That way, when reading cgroup names it
1857 * is sufficient to remove a single prefixing underscore if
1860 /* The return value of this function (unlike cg_unescape())
1863 if (IN_SET(p
[0], 0, '_', '.') ||
1864 streq(p
, "notify_on_release") ||
1865 streq(p
, "release_agent") ||
1866 streq(p
, "tasks") ||
1867 startswith(p
, "cgroup."))
1872 dot
= strrchr(p
, '.');
1877 for (c
= 0; c
< _CGROUP_CONTROLLER_MAX
; c
++) {
1880 n
= cgroup_controller_to_string(c
);
1885 if (memcmp(p
, n
, l
) != 0)
1895 return strappend("_", p
);
1900 char *cg_unescape(const char *p
) {
1903 /* The return value of this function (unlike cg_escape())
1904 * doesn't need free()! */
1912 #define CONTROLLER_VALID \
1916 bool cg_controller_is_valid(const char *p
) {
1922 if (streq(p
, SYSTEMD_CGROUP_CONTROLLER
))
1925 s
= startswith(p
, "name=");
1929 if (IN_SET(*p
, 0, '_'))
1932 for (t
= p
; *t
; t
++)
1933 if (!strchr(CONTROLLER_VALID
, *t
))
1936 if (t
- p
> FILENAME_MAX
)
1942 int cg_slice_to_path(const char *unit
, char **ret
) {
1943 _cleanup_free_
char *p
= NULL
, *s
= NULL
, *e
= NULL
;
1950 if (streq(unit
, SPECIAL_ROOT_SLICE
)) {
1960 if (!unit_name_is_valid(unit
, UNIT_NAME_PLAIN
))
1963 if (!endswith(unit
, ".slice"))
1966 r
= unit_name_to_prefix(unit
, &p
);
1970 dash
= strchr(p
, '-');
1972 /* Don't allow initial dashes */
1977 _cleanup_free_
char *escaped
= NULL
;
1978 char n
[dash
- p
+ sizeof(".slice")];
1980 /* Don't allow trailing or double dashes */
1981 if (IN_SET(dash
[1], 0, '-'))
1984 strcpy(stpncpy(n
, p
, dash
- p
), ".slice");
1985 if (!unit_name_is_valid(n
, UNIT_NAME_PLAIN
))
1988 escaped
= cg_escape(n
);
1992 if (!strextend(&s
, escaped
, "/", NULL
))
1995 dash
= strchr(dash
+1, '-');
1998 e
= cg_escape(unit
);
2002 if (!strextend(&s
, e
, NULL
))
2011 int cg_set_attribute(const char *controller
, const char *path
, const char *attribute
, const char *value
) {
2012 _cleanup_free_
char *p
= NULL
;
2015 r
= cg_get_path(controller
, path
, attribute
, &p
);
2019 return write_string_file(p
, value
, 0);
2022 int cg_get_attribute(const char *controller
, const char *path
, const char *attribute
, char **ret
) {
2023 _cleanup_free_
char *p
= NULL
;
2026 r
= cg_get_path(controller
, path
, attribute
, &p
);
2030 return read_one_line_file(p
, ret
);
2033 int cg_get_keyed_attribute(
2034 const char *controller
,
2036 const char *attribute
,
2038 char **ret_values
) {
2040 _cleanup_free_
char *filename
= NULL
, *contents
= NULL
;
2041 _cleanup_fclose_
FILE *f
= NULL
;
2043 size_t n
, i
, n_done
= 0;
2047 /* Reads one or more fields of a cgroupsv2 keyed attribute file. The 'keys' parameter should be an strv with
2048 * all keys to retrieve. The 'ret_values' parameter should be passed as string size with the same number of
2049 * entries as 'keys'. On success each entry will be set to the value of the matching key.
2051 * If the attribute file doesn't exist at all returns ENOENT, if any key is not found returns ENXIO. */
2053 r
= cg_get_path(controller
, path
, attribute
, &filename
);
2057 r
= read_full_file(filename
, &contents
, NULL
);
2061 n
= strv_length(keys
);
2062 if (n
== 0) /* No keys to retrieve? That's easy, we are done then */
2065 /* Let's build this up in a temporary array for now in order not to clobber the return parameter on failure */
2066 v
= newa0(char*, n
);
2068 for (p
= contents
; *p
;) {
2069 const char *w
= NULL
;
2071 for (i
= 0; i
< n
; i
++)
2073 w
= first_word(p
, keys
[i
]);
2081 l
= strcspn(w
, NEWLINE
);
2082 v
[i
] = strndup(w
, l
);
2094 p
+= strcspn(p
, NEWLINE
);
2096 p
+= strspn(p
, NEWLINE
);
2102 for (i
= 0; i
< n
; i
++)
2108 memcpy(ret_values
, v
, sizeof(char*) * n
);
2113 int cg_create_everywhere(CGroupMask supported
, CGroupMask mask
, const char *path
) {
2117 /* This one will create a cgroup in our private tree, but also
2118 * duplicate it in the trees specified in mask, and remove it
2121 /* First create the cgroup in our own hierarchy. */
2122 r
= cg_create(SYSTEMD_CGROUP_CONTROLLER
, path
);
2126 /* If we are in the unified hierarchy, we are done now */
2127 r
= cg_all_unified();
2133 /* Otherwise, do the same in the other hierarchies */
2134 for (c
= 0; c
< _CGROUP_CONTROLLER_MAX
; c
++) {
2135 CGroupMask bit
= CGROUP_CONTROLLER_TO_MASK(c
);
2138 n
= cgroup_controller_to_string(c
);
2141 (void) cg_create(n
, path
);
2142 else if (supported
& bit
)
2143 (void) cg_trim(n
, path
, true);
2149 int cg_attach_everywhere(CGroupMask supported
, const char *path
, pid_t pid
, cg_migrate_callback_t path_callback
, void *userdata
) {
2153 r
= cg_attach(SYSTEMD_CGROUP_CONTROLLER
, path
, pid
);
2157 r
= cg_all_unified();
2163 for (c
= 0; c
< _CGROUP_CONTROLLER_MAX
; c
++) {
2164 CGroupMask bit
= CGROUP_CONTROLLER_TO_MASK(c
);
2165 const char *p
= NULL
;
2167 if (!(supported
& bit
))
2171 p
= path_callback(bit
, userdata
);
2176 (void) cg_attach_fallback(cgroup_controller_to_string(c
), p
, pid
);
2182 int cg_attach_many_everywhere(CGroupMask supported
, const char *path
, Set
* pids
, cg_migrate_callback_t path_callback
, void *userdata
) {
2187 SET_FOREACH(pidp
, pids
, i
) {
2188 pid_t pid
= PTR_TO_PID(pidp
);
2191 q
= cg_attach_everywhere(supported
, path
, pid
, path_callback
, userdata
);
2192 if (q
< 0 && r
>= 0)
2199 int cg_migrate_everywhere(CGroupMask supported
, const char *from
, const char *to
, cg_migrate_callback_t to_callback
, void *userdata
) {
2203 if (!path_equal(from
, to
)) {
2204 r
= cg_migrate_recursive(SYSTEMD_CGROUP_CONTROLLER
, from
, SYSTEMD_CGROUP_CONTROLLER
, to
, CGROUP_REMOVE
);
2209 q
= cg_all_unified();
2215 for (c
= 0; c
< _CGROUP_CONTROLLER_MAX
; c
++) {
2216 CGroupMask bit
= CGROUP_CONTROLLER_TO_MASK(c
);
2217 const char *p
= NULL
;
2219 if (!(supported
& bit
))
2223 p
= to_callback(bit
, userdata
);
2228 (void) cg_migrate_recursive_fallback(SYSTEMD_CGROUP_CONTROLLER
, to
, cgroup_controller_to_string(c
), p
, 0);
2234 int cg_trim_everywhere(CGroupMask supported
, const char *path
, bool delete_root
) {
2238 r
= cg_trim(SYSTEMD_CGROUP_CONTROLLER
, path
, delete_root
);
2242 q
= cg_all_unified();
2248 for (c
= 0; c
< _CGROUP_CONTROLLER_MAX
; c
++) {
2249 CGroupMask bit
= CGROUP_CONTROLLER_TO_MASK(c
);
2251 if (!(supported
& bit
))
2254 (void) cg_trim(cgroup_controller_to_string(c
), path
, delete_root
);
2260 int cg_mask_to_string(CGroupMask mask
, char **ret
) {
2261 _cleanup_free_
char *s
= NULL
;
2262 size_t n
= 0, allocated
= 0;
2273 for (c
= 0; c
< _CGROUP_CONTROLLER_MAX
; c
++) {
2277 if (!(mask
& CGROUP_CONTROLLER_TO_MASK(c
)))
2280 k
= cgroup_controller_to_string(c
);
2283 if (!GREEDY_REALLOC(s
, allocated
, n
+ space
+ l
+ 1))
2288 memcpy(s
+ n
+ space
, k
, l
);
2303 int cg_mask_from_string(const char *value
, CGroupMask
*mask
) {
2308 _cleanup_free_
char *n
= NULL
;
2312 r
= extract_first_word(&value
, &n
, NULL
, 0);
2318 v
= cgroup_controller_from_string(n
);
2322 *mask
|= CGROUP_CONTROLLER_TO_MASK(v
);
2327 int cg_mask_supported(CGroupMask
*ret
) {
2328 CGroupMask mask
= 0;
2331 /* Determines the mask of supported cgroup controllers. Only
2332 * includes controllers we can make sense of and that are
2333 * actually accessible. */
2335 r
= cg_all_unified();
2339 _cleanup_free_
char *root
= NULL
, *controllers
= NULL
, *path
= NULL
;
2341 /* In the unified hierarchy we can read the supported
2342 * and accessible controllers from a the top-level
2343 * cgroup attribute */
2345 r
= cg_get_root_path(&root
);
2349 r
= cg_get_path(SYSTEMD_CGROUP_CONTROLLER
, root
, "cgroup.controllers", &path
);
2353 r
= read_one_line_file(path
, &controllers
);
2357 r
= cg_mask_from_string(controllers
, &mask
);
2361 /* Currently, we support the cpu, memory, io and pids
2362 * controller in the unified hierarchy, mask
2363 * everything else off. */
2364 mask
&= CGROUP_MASK_CPU
| CGROUP_MASK_MEMORY
| CGROUP_MASK_IO
| CGROUP_MASK_PIDS
;
2369 /* In the legacy hierarchy, we check whether which
2370 * hierarchies are mounted. */
2372 for (c
= 0; c
< _CGROUP_CONTROLLER_MAX
; c
++) {
2375 n
= cgroup_controller_to_string(c
);
2376 if (controller_is_accessible(n
) >= 0)
2377 mask
|= CGROUP_CONTROLLER_TO_MASK(c
);
2385 int cg_kernel_controllers(Set
**ret
) {
2386 _cleanup_set_free_free_ Set
*controllers
= NULL
;
2387 _cleanup_fclose_
FILE *f
= NULL
;
2392 /* Determines the full list of kernel-known controllers. Might
2393 * include controllers we don't actually support, arbitrary
2394 * named hierarchies and controllers that aren't currently
2395 * accessible (because not mounted). */
2397 controllers
= set_new(&string_hash_ops
);
2401 f
= fopen("/proc/cgroups", "re");
2403 if (errno
== ENOENT
) {
2411 (void) __fsetlocking(f
, FSETLOCKING_BYCALLER
);
2413 /* Ignore the header line */
2414 (void) read_line(f
, (size_t) -1, NULL
);
2421 if (fscanf(f
, "%ms %*i %*i %i", &controller
, &enabled
) != 2) {
2426 if (ferror(f
) && errno
> 0)
2437 if (!cg_controller_is_valid(controller
)) {
2442 r
= set_consume(controllers
, controller
);
2453 static thread_local CGroupUnified unified_cache
= CGROUP_UNIFIED_UNKNOWN
;
2455 /* The hybrid mode was initially implemented in v232 and simply mounted cgroup v2 on /sys/fs/cgroup/systemd. This
2456 * unfortunately broke other tools (such as docker) which expected the v1 "name=systemd" hierarchy on
2457 * /sys/fs/cgroup/systemd. From v233 and on, the hybrid mode mountnbs v2 on /sys/fs/cgroup/unified and maintains
2458 * "name=systemd" hierarchy on /sys/fs/cgroup/systemd for compatibility with other tools.
2460 * To keep live upgrade working, we detect and support v232 layout. When v232 layout is detected, to keep cgroup v2
2461 * process management but disable the compat dual layout, we return %true on
2462 * cg_unified_controller(SYSTEMD_CGROUP_CONTROLLER) and %false on cg_hybrid_unified().
2464 static thread_local
bool unified_systemd_v232
;
2466 static int cg_unified_update(void) {
2470 /* Checks if we support the unified hierarchy. Returns an
2471 * error when the cgroup hierarchies aren't mounted yet or we
2472 * have any other trouble determining if the unified hierarchy
2475 if (unified_cache
>= CGROUP_UNIFIED_NONE
)
2478 if (statfs("/sys/fs/cgroup/", &fs
) < 0)
2479 return log_debug_errno(errno
, "statfs(\"/sys/fs/cgroup/\" failed: %m");
2481 if (F_TYPE_EQUAL(fs
.f_type
, CGROUP2_SUPER_MAGIC
)) {
2482 log_debug("Found cgroup2 on /sys/fs/cgroup/, full unified hierarchy");
2483 unified_cache
= CGROUP_UNIFIED_ALL
;
2484 } else if (F_TYPE_EQUAL(fs
.f_type
, TMPFS_MAGIC
)) {
2485 if (statfs("/sys/fs/cgroup/unified/", &fs
) == 0 &&
2486 F_TYPE_EQUAL(fs
.f_type
, CGROUP2_SUPER_MAGIC
)) {
2487 log_debug("Found cgroup2 on /sys/fs/cgroup/unified, unified hierarchy for systemd controller");
2488 unified_cache
= CGROUP_UNIFIED_SYSTEMD
;
2489 unified_systemd_v232
= false;
2491 if (statfs("/sys/fs/cgroup/systemd/", &fs
) < 0)
2492 return log_debug_errno(errno
, "statfs(\"/sys/fs/cgroup/systemd\" failed: %m");
2494 if (F_TYPE_EQUAL(fs
.f_type
, CGROUP2_SUPER_MAGIC
)) {
2495 log_debug("Found cgroup2 on /sys/fs/cgroup/systemd, unified hierarchy for systemd controller (v232 variant)");
2496 unified_cache
= CGROUP_UNIFIED_SYSTEMD
;
2497 unified_systemd_v232
= true;
2498 } else if (F_TYPE_EQUAL(fs
.f_type
, CGROUP_SUPER_MAGIC
)) {
2499 log_debug("Found cgroup on /sys/fs/cgroup/systemd, legacy hierarchy");
2500 unified_cache
= CGROUP_UNIFIED_NONE
;
2502 log_debug("Unexpected filesystem type %llx mounted on /sys/fs/cgroup/systemd, assuming legacy hierarchy",
2503 (unsigned long long) fs
.f_type
);
2504 unified_cache
= CGROUP_UNIFIED_NONE
;
2508 log_debug("Unknown filesystem type %llx mounted on /sys/fs/cgroup.",
2509 (unsigned long long) fs
.f_type
);
2516 int cg_unified_controller(const char *controller
) {
2519 r
= cg_unified_update();
2523 if (unified_cache
== CGROUP_UNIFIED_NONE
)
2526 if (unified_cache
>= CGROUP_UNIFIED_ALL
)
2529 return streq_ptr(controller
, SYSTEMD_CGROUP_CONTROLLER
);
2532 int cg_all_unified(void) {
2535 r
= cg_unified_update();
2539 return unified_cache
>= CGROUP_UNIFIED_ALL
;
2542 int cg_hybrid_unified(void) {
2545 r
= cg_unified_update();
2549 return unified_cache
== CGROUP_UNIFIED_SYSTEMD
&& !unified_systemd_v232
;
2552 int cg_unified_flush(void) {
2553 unified_cache
= CGROUP_UNIFIED_UNKNOWN
;
2555 return cg_unified_update();
2558 int cg_enable_everywhere(CGroupMask supported
, CGroupMask mask
, const char *p
) {
2559 _cleanup_fclose_
FILE *f
= NULL
;
2560 _cleanup_free_
char *fs
= NULL
;
2569 r
= cg_all_unified();
2572 if (r
== 0) /* on the legacy hiearchy there's no joining of controllers defined */
2575 r
= cg_get_path(SYSTEMD_CGROUP_CONTROLLER
, p
, "cgroup.subtree_control", &fs
);
2579 for (c
= 0; c
< _CGROUP_CONTROLLER_MAX
; c
++) {
2580 CGroupMask bit
= CGROUP_CONTROLLER_TO_MASK(c
);
2583 if (!(supported
& bit
))
2586 n
= cgroup_controller_to_string(c
);
2588 char s
[1 + strlen(n
) + 1];
2590 s
[0] = mask
& bit
? '+' : '-';
2594 f
= fopen(fs
, "we");
2596 log_debug_errno(errno
, "Failed to open cgroup.subtree_control file of %s: %m", p
);
2601 r
= write_string_stream(f
, s
, 0);
2603 log_debug_errno(r
, "Failed to enable controller %s for %s (%s): %m", n
, p
, fs
);
2610 bool cg_is_unified_wanted(void) {
2611 static thread_local
int wanted
= -1;
2614 const bool is_default
= DEFAULT_HIERARCHY
== CGROUP_UNIFIED_ALL
;
2616 /* If we have a cached value, return that. */
2620 /* If the hierarchy is already mounted, then follow whatever
2621 * was chosen for it. */
2622 if (cg_unified_flush() >= 0)
2623 return (wanted
= unified_cache
>= CGROUP_UNIFIED_ALL
);
2625 /* Otherwise, let's see what the kernel command line has to say.
2626 * Since checking is expensive, cache a non-error result. */
2627 r
= proc_cmdline_get_bool("systemd.unified_cgroup_hierarchy", &b
);
2629 return (wanted
= r
> 0 ? b
: is_default
);
2632 bool cg_is_legacy_wanted(void) {
2633 static thread_local
int wanted
= -1;
2635 /* If we have a cached value, return that. */
2639 /* Check if we have cgroups2 already mounted. */
2640 if (cg_unified_flush() >= 0 &&
2641 unified_cache
== CGROUP_UNIFIED_ALL
)
2642 return (wanted
= false);
2644 /* Otherwise, assume that at least partial legacy is wanted,
2645 * since cgroups2 should already be mounted at this point. */
2646 return (wanted
= true);
2649 bool cg_is_hybrid_wanted(void) {
2650 static thread_local
int wanted
= -1;
2653 const bool is_default
= DEFAULT_HIERARCHY
>= CGROUP_UNIFIED_SYSTEMD
;
2654 /* We default to true if the default is "hybrid", obviously,
2655 * but also when the default is "unified", because if we get
2656 * called, it means that unified hierarchy was not mounted. */
2658 /* If we have a cached value, return that. */
2662 /* If the hierarchy is already mounted, then follow whatever
2663 * was chosen for it. */
2664 if (cg_unified_flush() >= 0 &&
2665 unified_cache
== CGROUP_UNIFIED_ALL
)
2666 return (wanted
= false);
2668 /* Otherwise, let's see what the kernel command line has to say.
2669 * Since checking is expensive, cache a non-error result. */
2670 r
= proc_cmdline_get_bool("systemd.legacy_systemd_cgroup_controller", &b
);
2672 /* The meaning of the kernel option is reversed wrt. to the return value
2673 * of this function, hence the negation. */
2674 return (wanted
= r
> 0 ? !b
: is_default
);
2677 int cg_weight_parse(const char *s
, uint64_t *ret
) {
2682 *ret
= CGROUP_WEIGHT_INVALID
;
2686 r
= safe_atou64(s
, &u
);
2690 if (u
< CGROUP_WEIGHT_MIN
|| u
> CGROUP_WEIGHT_MAX
)
2697 const uint64_t cgroup_io_limit_defaults
[_CGROUP_IO_LIMIT_TYPE_MAX
] = {
2698 [CGROUP_IO_RBPS_MAX
] = CGROUP_LIMIT_MAX
,
2699 [CGROUP_IO_WBPS_MAX
] = CGROUP_LIMIT_MAX
,
2700 [CGROUP_IO_RIOPS_MAX
] = CGROUP_LIMIT_MAX
,
2701 [CGROUP_IO_WIOPS_MAX
] = CGROUP_LIMIT_MAX
,
2704 static const char* const cgroup_io_limit_type_table
[_CGROUP_IO_LIMIT_TYPE_MAX
] = {
2705 [CGROUP_IO_RBPS_MAX
] = "IOReadBandwidthMax",
2706 [CGROUP_IO_WBPS_MAX
] = "IOWriteBandwidthMax",
2707 [CGROUP_IO_RIOPS_MAX
] = "IOReadIOPSMax",
2708 [CGROUP_IO_WIOPS_MAX
] = "IOWriteIOPSMax",
2711 DEFINE_STRING_TABLE_LOOKUP(cgroup_io_limit_type
, CGroupIOLimitType
);
2713 int cg_cpu_shares_parse(const char *s
, uint64_t *ret
) {
2718 *ret
= CGROUP_CPU_SHARES_INVALID
;
2722 r
= safe_atou64(s
, &u
);
2726 if (u
< CGROUP_CPU_SHARES_MIN
|| u
> CGROUP_CPU_SHARES_MAX
)
2733 int cg_blkio_weight_parse(const char *s
, uint64_t *ret
) {
2738 *ret
= CGROUP_BLKIO_WEIGHT_INVALID
;
2742 r
= safe_atou64(s
, &u
);
2746 if (u
< CGROUP_BLKIO_WEIGHT_MIN
|| u
> CGROUP_BLKIO_WEIGHT_MAX
)
2753 bool is_cgroup_fs(const struct statfs
*s
) {
2754 return is_fs_type(s
, CGROUP_SUPER_MAGIC
) ||
2755 is_fs_type(s
, CGROUP2_SUPER_MAGIC
);
2758 bool fd_is_cgroup_fs(int fd
) {
2761 if (fstatfs(fd
, &s
) < 0)
2764 return is_cgroup_fs(&s
);
2767 static const char *cgroup_controller_table
[_CGROUP_CONTROLLER_MAX
] = {
2768 [CGROUP_CONTROLLER_CPU
] = "cpu",
2769 [CGROUP_CONTROLLER_CPUACCT
] = "cpuacct",
2770 [CGROUP_CONTROLLER_IO
] = "io",
2771 [CGROUP_CONTROLLER_BLKIO
] = "blkio",
2772 [CGROUP_CONTROLLER_MEMORY
] = "memory",
2773 [CGROUP_CONTROLLER_DEVICES
] = "devices",
2774 [CGROUP_CONTROLLER_PIDS
] = "pids",
2777 DEFINE_STRING_TABLE_LOOKUP(cgroup_controller
, CGroupController
);