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 if (mkdir(fs
, 0755) < 0) {
790 r
= cg_hybrid_unified();
794 if (r
> 0 && streq(controller
, SYSTEMD_CGROUP_CONTROLLER
)) {
795 r
= cg_create(SYSTEMD_CGROUP_CONTROLLER_LEGACY
, path
);
797 log_warning_errno(r
, "Failed to create compat systemd cgroup %s: %m", path
);
803 int cg_create_and_attach(const char *controller
, const char *path
, pid_t pid
) {
808 r
= cg_create(controller
, path
);
812 q
= cg_attach(controller
, path
, pid
);
816 /* This does not remove the cgroup on failure */
820 int cg_attach(const char *controller
, const char *path
, pid_t pid
) {
821 _cleanup_free_
char *fs
= NULL
;
822 char c
[DECIMAL_STR_MAX(pid_t
) + 2];
828 r
= cg_get_path_and_check(controller
, path
, "cgroup.procs", &fs
);
833 pid
= getpid_cached();
835 xsprintf(c
, PID_FMT
"\n", pid
);
837 r
= write_string_file(fs
, c
, 0);
841 r
= cg_hybrid_unified();
845 if (r
> 0 && streq(controller
, SYSTEMD_CGROUP_CONTROLLER
)) {
846 r
= cg_attach(SYSTEMD_CGROUP_CONTROLLER_LEGACY
, path
, pid
);
848 log_warning_errno(r
, "Failed to attach "PID_FMT
" to compat systemd cgroup %s: %m", pid
, path
);
854 int cg_attach_fallback(const char *controller
, const char *path
, pid_t pid
) {
861 r
= cg_attach(controller
, path
, pid
);
863 char prefix
[strlen(path
) + 1];
865 /* This didn't work? Then let's try all prefixes of
868 PATH_FOREACH_PREFIX(prefix
, path
) {
871 q
= cg_attach(controller
, prefix
, pid
);
881 const char *controller
,
891 /* cgroupsv1, aka legacy/non-unified */
892 static const struct Attribute legacy_attributes
[] = {
893 { "cgroup.procs", true },
895 { "cgroup.clone_children", false },
899 /* cgroupsv2, aka unified */
900 static const struct Attribute unified_attributes
[] = {
901 { "cgroup.procs", true },
902 { "cgroup.subtree_control", true },
903 { "cgroup.threads", false },
907 static const struct Attribute
* const attributes
[] = {
908 [false] = legacy_attributes
,
909 [true] = unified_attributes
,
912 _cleanup_free_
char *fs
= NULL
;
913 const struct Attribute
*i
;
918 if (uid
== UID_INVALID
&& gid
== GID_INVALID
)
921 unified
= cg_unified_controller(controller
);
925 /* Configure access to the cgroup itself */
926 r
= cg_get_path(controller
, path
, NULL
, &fs
);
930 r
= chmod_and_chown(fs
, 0755, uid
, gid
);
934 /* Configure access to the cgroup's attributes */
935 for (i
= attributes
[unified
]; i
->name
; i
++) {
938 r
= cg_get_path(controller
, path
, i
->name
, &fs
);
942 r
= chmod_and_chown(fs
, 0644, uid
, gid
);
947 log_debug_errno(r
, "Failed to set access on cgroup %s, ignoring: %m", fs
);
951 if (streq(controller
, SYSTEMD_CGROUP_CONTROLLER
)) {
952 r
= cg_hybrid_unified();
956 /* Always propagate access mode from unified to legacy controller */
957 r
= cg_set_access(SYSTEMD_CGROUP_CONTROLLER_LEGACY
, path
, uid
, gid
);
959 log_debug_errno(r
, "Failed to set access on compatibility systemd cgroup %s, ignoring: %m", path
);
966 int cg_set_xattr(const char *controller
, const char *path
, const char *name
, const void *value
, size_t size
, int flags
) {
967 _cleanup_free_
char *fs
= NULL
;
972 assert(value
|| size
<= 0);
974 r
= cg_get_path(controller
, path
, NULL
, &fs
);
978 if (setxattr(fs
, name
, value
, size
, flags
) < 0)
984 int cg_get_xattr(const char *controller
, const char *path
, const char *name
, void *value
, size_t size
) {
985 _cleanup_free_
char *fs
= NULL
;
992 r
= cg_get_path(controller
, path
, NULL
, &fs
);
996 n
= getxattr(fs
, name
, value
, size
);
1003 int cg_pid_get_path(const char *controller
, pid_t pid
, char **path
) {
1004 _cleanup_fclose_
FILE *f
= NULL
;
1005 char line
[LINE_MAX
];
1006 const char *fs
, *controller_str
;
1014 if (!cg_controller_is_valid(controller
))
1017 controller
= SYSTEMD_CGROUP_CONTROLLER
;
1019 unified
= cg_unified_controller(controller
);
1023 if (streq(controller
, SYSTEMD_CGROUP_CONTROLLER
))
1024 controller_str
= SYSTEMD_CGROUP_CONTROLLER_LEGACY
;
1026 controller_str
= controller
;
1028 cs
= strlen(controller_str
);
1031 fs
= procfs_file_alloca(pid
, "cgroup");
1032 f
= fopen(fs
, "re");
1034 return errno
== ENOENT
? -ESRCH
: -errno
;
1036 (void) __fsetlocking(f
, FSETLOCKING_BYCALLER
);
1038 FOREACH_LINE(line
, f
, return -errno
) {
1044 e
= startswith(line
, "0:");
1054 const char *word
, *state
;
1057 l
= strchr(line
, ':');
1067 FOREACH_WORD_SEPARATOR(word
, k
, l
, ",", state
) {
1068 if (k
== cs
&& memcmp(word
, controller_str
, cs
) == 0) {
1082 /* Truncate suffix indicating the process is a zombie */
1083 e
= endswith(p
, " (deleted)");
1094 int cg_install_release_agent(const char *controller
, const char *agent
) {
1095 _cleanup_free_
char *fs
= NULL
, *contents
= NULL
;
1101 r
= cg_unified_controller(controller
);
1104 if (r
> 0) /* doesn't apply to unified hierarchy */
1107 r
= cg_get_path(controller
, NULL
, "release_agent", &fs
);
1111 r
= read_one_line_file(fs
, &contents
);
1115 sc
= strstrip(contents
);
1117 r
= write_string_file(fs
, agent
, 0);
1120 } else if (!path_equal(sc
, agent
))
1124 r
= cg_get_path(controller
, NULL
, "notify_on_release", &fs
);
1128 contents
= mfree(contents
);
1129 r
= read_one_line_file(fs
, &contents
);
1133 sc
= strstrip(contents
);
1134 if (streq(sc
, "0")) {
1135 r
= write_string_file(fs
, "1", 0);
1142 if (!streq(sc
, "1"))
1148 int cg_uninstall_release_agent(const char *controller
) {
1149 _cleanup_free_
char *fs
= NULL
;
1152 r
= cg_unified_controller(controller
);
1155 if (r
> 0) /* Doesn't apply to unified hierarchy */
1158 r
= cg_get_path(controller
, NULL
, "notify_on_release", &fs
);
1162 r
= write_string_file(fs
, "0", 0);
1168 r
= cg_get_path(controller
, NULL
, "release_agent", &fs
);
1172 r
= write_string_file(fs
, "", 0);
1179 int cg_is_empty(const char *controller
, const char *path
) {
1180 _cleanup_fclose_
FILE *f
= NULL
;
1186 r
= cg_enumerate_processes(controller
, path
, &f
);
1192 r
= cg_read_pid(f
, &pid
);
1199 int cg_is_empty_recursive(const char *controller
, const char *path
) {
1204 /* The root cgroup is always populated */
1205 if (controller
&& (isempty(path
) || path_equal(path
, "/")))
1208 r
= cg_unified_controller(controller
);
1212 _cleanup_free_
char *t
= NULL
;
1214 /* On the unified hierarchy we can check empty state
1215 * via the "populated" attribute of "cgroup.events". */
1217 r
= cg_read_event(controller
, path
, "populated", &t
);
1221 return streq(t
, "0");
1223 _cleanup_closedir_
DIR *d
= NULL
;
1226 r
= cg_is_empty(controller
, path
);
1230 r
= cg_enumerate_subgroups(controller
, path
, &d
);
1236 while ((r
= cg_read_subgroup(d
, &fn
)) > 0) {
1237 _cleanup_free_
char *p
= NULL
;
1239 p
= strjoin(path
, "/", fn
);
1244 r
= cg_is_empty_recursive(controller
, p
);
1255 int cg_split_spec(const char *spec
, char **controller
, char **path
) {
1256 char *t
= NULL
, *u
= NULL
;
1262 if (!path_is_normalized(spec
))
1270 *path
= path_kill_slashes(t
);
1279 e
= strchr(spec
, ':');
1281 if (!cg_controller_is_valid(spec
))
1298 t
= strndup(spec
, e
-spec
);
1301 if (!cg_controller_is_valid(t
)) {
1315 if (!path_is_normalized(u
) ||
1316 !path_is_absolute(u
)) {
1322 path_kill_slashes(u
);
1338 int cg_mangle_path(const char *path
, char **result
) {
1339 _cleanup_free_
char *c
= NULL
, *p
= NULL
;
1346 /* First, check if it already is a filesystem path */
1347 if (path_startswith(path
, "/sys/fs/cgroup")) {
1353 *result
= path_kill_slashes(t
);
1357 /* Otherwise, treat it as cg spec */
1358 r
= cg_split_spec(path
, &c
, &p
);
1362 return cg_get_path(c
?: SYSTEMD_CGROUP_CONTROLLER
, p
?: "/", NULL
, result
);
1365 int cg_get_root_path(char **path
) {
1371 r
= cg_pid_get_path(SYSTEMD_CGROUP_CONTROLLER
, 1, &p
);
1375 e
= endswith(p
, "/" SPECIAL_INIT_SCOPE
);
1377 e
= endswith(p
, "/" SPECIAL_SYSTEM_SLICE
); /* legacy */
1379 e
= endswith(p
, "/system"); /* even more legacy */
1387 int cg_shift_path(const char *cgroup
, const char *root
, const char **shifted
) {
1388 _cleanup_free_
char *rt
= NULL
;
1396 /* If the root was specified let's use that, otherwise
1397 * let's determine it from PID 1 */
1399 r
= cg_get_root_path(&rt
);
1406 p
= path_startswith(cgroup
, root
);
1407 if (p
&& p
> cgroup
)
1415 int cg_pid_get_path_shifted(pid_t pid
, const char *root
, char **cgroup
) {
1416 _cleanup_free_
char *raw
= NULL
;
1423 r
= cg_pid_get_path(SYSTEMD_CGROUP_CONTROLLER
, pid
, &raw
);
1427 r
= cg_shift_path(raw
, root
, &c
);
1447 int cg_path_decode_unit(const char *cgroup
, char **unit
) {
1454 n
= strcspn(cgroup
, "/");
1458 c
= strndupa(cgroup
, n
);
1461 if (!unit_name_is_valid(c
, UNIT_NAME_PLAIN
|UNIT_NAME_INSTANCE
))
1472 static bool valid_slice_name(const char *p
, size_t n
) {
1477 if (n
< STRLEN("x.slice"))
1480 if (memcmp(p
+ n
- 6, ".slice", 6) == 0) {
1486 c
= cg_unescape(buf
);
1488 return unit_name_is_valid(c
, UNIT_NAME_PLAIN
);
1494 static const char *skip_slices(const char *p
) {
1497 /* Skips over all slice assignments */
1502 p
+= strspn(p
, "/");
1504 n
= strcspn(p
, "/");
1505 if (!valid_slice_name(p
, n
))
1512 int cg_path_get_unit(const char *path
, char **ret
) {
1520 e
= skip_slices(path
);
1522 r
= cg_path_decode_unit(e
, &unit
);
1526 /* We skipped over the slices, don't accept any now */
1527 if (endswith(unit
, ".slice")) {
1536 int cg_pid_get_unit(pid_t pid
, char **unit
) {
1537 _cleanup_free_
char *cgroup
= NULL
;
1542 r
= cg_pid_get_path_shifted(pid
, NULL
, &cgroup
);
1546 return cg_path_get_unit(cgroup
, unit
);
1550 * Skip session-*.scope, but require it to be there.
1552 static const char *skip_session(const char *p
) {
1558 p
+= strspn(p
, "/");
1560 n
= strcspn(p
, "/");
1561 if (n
< STRLEN("session-x.scope"))
1564 if (memcmp(p
, "session-", 8) == 0 && memcmp(p
+ n
- 6, ".scope", 6) == 0) {
1565 char buf
[n
- 8 - 6 + 1];
1567 memcpy(buf
, p
+ 8, n
- 8 - 6);
1570 /* Note that session scopes never need unescaping,
1571 * since they cannot conflict with the kernel's own
1572 * names, hence we don't need to call cg_unescape()
1575 if (!session_id_valid(buf
))
1579 p
+= strspn(p
, "/");
1587 * Skip user@*.service, but require it to be there.
1589 static const char *skip_user_manager(const char *p
) {
1595 p
+= strspn(p
, "/");
1597 n
= strcspn(p
, "/");
1598 if (n
< STRLEN("user@x.service"))
1601 if (memcmp(p
, "user@", 5) == 0 && memcmp(p
+ n
- 8, ".service", 8) == 0) {
1602 char buf
[n
- 5 - 8 + 1];
1604 memcpy(buf
, p
+ 5, n
- 5 - 8);
1607 /* Note that user manager services never need unescaping,
1608 * since they cannot conflict with the kernel's own
1609 * names, hence we don't need to call cg_unescape()
1612 if (parse_uid(buf
, NULL
) < 0)
1616 p
+= strspn(p
, "/");
1624 static const char *skip_user_prefix(const char *path
) {
1629 /* Skip slices, if there are any */
1630 e
= skip_slices(path
);
1632 /* Skip the user manager, if it's in the path now... */
1633 t
= skip_user_manager(e
);
1637 /* Alternatively skip the user session if it is in the path... */
1638 return skip_session(e
);
1641 int cg_path_get_user_unit(const char *path
, char **ret
) {
1647 t
= skip_user_prefix(path
);
1651 /* And from here on it looks pretty much the same as for a
1652 * system unit, hence let's use the same parser from here
1654 return cg_path_get_unit(t
, ret
);
1657 int cg_pid_get_user_unit(pid_t pid
, char **unit
) {
1658 _cleanup_free_
char *cgroup
= NULL
;
1663 r
= cg_pid_get_path_shifted(pid
, NULL
, &cgroup
);
1667 return cg_path_get_user_unit(cgroup
, unit
);
1670 int cg_path_get_machine_name(const char *path
, char **machine
) {
1671 _cleanup_free_
char *u
= NULL
;
1675 r
= cg_path_get_unit(path
, &u
);
1679 sl
= strjoina("/run/systemd/machines/unit:", u
);
1680 return readlink_malloc(sl
, machine
);
1683 int cg_pid_get_machine_name(pid_t pid
, char **machine
) {
1684 _cleanup_free_
char *cgroup
= NULL
;
1689 r
= cg_pid_get_path_shifted(pid
, NULL
, &cgroup
);
1693 return cg_path_get_machine_name(cgroup
, machine
);
1696 int cg_path_get_session(const char *path
, char **session
) {
1697 _cleanup_free_
char *unit
= NULL
;
1703 r
= cg_path_get_unit(path
, &unit
);
1707 start
= startswith(unit
, "session-");
1710 end
= endswith(start
, ".scope");
1715 if (!session_id_valid(start
))
1731 int cg_pid_get_session(pid_t pid
, char **session
) {
1732 _cleanup_free_
char *cgroup
= NULL
;
1735 r
= cg_pid_get_path_shifted(pid
, NULL
, &cgroup
);
1739 return cg_path_get_session(cgroup
, session
);
1742 int cg_path_get_owner_uid(const char *path
, uid_t
*uid
) {
1743 _cleanup_free_
char *slice
= NULL
;
1749 r
= cg_path_get_slice(path
, &slice
);
1753 start
= startswith(slice
, "user-");
1756 end
= endswith(start
, ".slice");
1761 if (parse_uid(start
, uid
) < 0)
1767 int cg_pid_get_owner_uid(pid_t pid
, uid_t
*uid
) {
1768 _cleanup_free_
char *cgroup
= NULL
;
1771 r
= cg_pid_get_path_shifted(pid
, NULL
, &cgroup
);
1775 return cg_path_get_owner_uid(cgroup
, uid
);
1778 int cg_path_get_slice(const char *p
, char **slice
) {
1779 const char *e
= NULL
;
1784 /* Finds the right-most slice unit from the beginning, but
1785 * stops before we come to the first non-slice unit. */
1790 p
+= strspn(p
, "/");
1792 n
= strcspn(p
, "/");
1793 if (!valid_slice_name(p
, n
)) {
1798 s
= strdup(SPECIAL_ROOT_SLICE
);
1806 return cg_path_decode_unit(e
, slice
);
1814 int cg_pid_get_slice(pid_t pid
, char **slice
) {
1815 _cleanup_free_
char *cgroup
= NULL
;
1820 r
= cg_pid_get_path_shifted(pid
, NULL
, &cgroup
);
1824 return cg_path_get_slice(cgroup
, slice
);
1827 int cg_path_get_user_slice(const char *p
, char **slice
) {
1832 t
= skip_user_prefix(p
);
1836 /* And now it looks pretty much the same as for a system
1837 * slice, so let's just use the same parser from here on. */
1838 return cg_path_get_slice(t
, slice
);
1841 int cg_pid_get_user_slice(pid_t pid
, char **slice
) {
1842 _cleanup_free_
char *cgroup
= NULL
;
1847 r
= cg_pid_get_path_shifted(pid
, NULL
, &cgroup
);
1851 return cg_path_get_user_slice(cgroup
, slice
);
1854 char *cg_escape(const char *p
) {
1855 bool need_prefix
= false;
1857 /* This implements very minimal escaping for names to be used
1858 * as file names in the cgroup tree: any name which might
1859 * conflict with a kernel name or is prefixed with '_' is
1860 * prefixed with a '_'. That way, when reading cgroup names it
1861 * is sufficient to remove a single prefixing underscore if
1864 /* The return value of this function (unlike cg_unescape())
1867 if (IN_SET(p
[0], 0, '_', '.') ||
1868 streq(p
, "notify_on_release") ||
1869 streq(p
, "release_agent") ||
1870 streq(p
, "tasks") ||
1871 startswith(p
, "cgroup."))
1876 dot
= strrchr(p
, '.');
1881 for (c
= 0; c
< _CGROUP_CONTROLLER_MAX
; c
++) {
1884 n
= cgroup_controller_to_string(c
);
1889 if (memcmp(p
, n
, l
) != 0)
1899 return strappend("_", p
);
1904 char *cg_unescape(const char *p
) {
1907 /* The return value of this function (unlike cg_escape())
1908 * doesn't need free()! */
1916 #define CONTROLLER_VALID \
1920 bool cg_controller_is_valid(const char *p
) {
1926 if (streq(p
, SYSTEMD_CGROUP_CONTROLLER
))
1929 s
= startswith(p
, "name=");
1933 if (IN_SET(*p
, 0, '_'))
1936 for (t
= p
; *t
; t
++)
1937 if (!strchr(CONTROLLER_VALID
, *t
))
1940 if (t
- p
> FILENAME_MAX
)
1946 int cg_slice_to_path(const char *unit
, char **ret
) {
1947 _cleanup_free_
char *p
= NULL
, *s
= NULL
, *e
= NULL
;
1954 if (streq(unit
, SPECIAL_ROOT_SLICE
)) {
1964 if (!unit_name_is_valid(unit
, UNIT_NAME_PLAIN
))
1967 if (!endswith(unit
, ".slice"))
1970 r
= unit_name_to_prefix(unit
, &p
);
1974 dash
= strchr(p
, '-');
1976 /* Don't allow initial dashes */
1981 _cleanup_free_
char *escaped
= NULL
;
1982 char n
[dash
- p
+ sizeof(".slice")];
1984 /* Don't allow trailing or double dashes */
1985 if (IN_SET(dash
[1], 0, '-'))
1988 strcpy(stpncpy(n
, p
, dash
- p
), ".slice");
1989 if (!unit_name_is_valid(n
, UNIT_NAME_PLAIN
))
1992 escaped
= cg_escape(n
);
1996 if (!strextend(&s
, escaped
, "/", NULL
))
1999 dash
= strchr(dash
+1, '-');
2002 e
= cg_escape(unit
);
2006 if (!strextend(&s
, e
, NULL
))
2015 int cg_set_attribute(const char *controller
, const char *path
, const char *attribute
, const char *value
) {
2016 _cleanup_free_
char *p
= NULL
;
2019 r
= cg_get_path(controller
, path
, attribute
, &p
);
2023 return write_string_file(p
, value
, 0);
2026 int cg_get_attribute(const char *controller
, const char *path
, const char *attribute
, char **ret
) {
2027 _cleanup_free_
char *p
= NULL
;
2030 r
= cg_get_path(controller
, path
, attribute
, &p
);
2034 return read_one_line_file(p
, ret
);
2037 int cg_get_keyed_attribute(const char *controller
, const char *path
, const char *attribute
, const char **keys
, char **values
) {
2038 _cleanup_free_
char *filename
= NULL
, *content
= NULL
;
2042 for (i
= 0; keys
[i
]; i
++)
2045 r
= cg_get_path(controller
, path
, attribute
, &filename
);
2049 r
= read_full_file(filename
, &content
, NULL
);
2054 while ((line
= strsep(&p
, "\n"))) {
2057 key
= strsep(&line
, " ");
2059 for (i
= 0; keys
[i
]; i
++) {
2060 if (streq(key
, keys
[i
])) {
2061 values
[i
] = strdup(line
);
2067 for (i
= 0; keys
[i
]; i
++) {
2069 for (i
= 0; keys
[i
]; i
++) {
2070 values
[i
] = mfree(values
[i
]);
2079 int cg_create_everywhere(CGroupMask supported
, CGroupMask mask
, const char *path
) {
2083 /* This one will create a cgroup in our private tree, but also
2084 * duplicate it in the trees specified in mask, and remove it
2087 /* First create the cgroup in our own hierarchy. */
2088 r
= cg_create(SYSTEMD_CGROUP_CONTROLLER
, path
);
2092 /* If we are in the unified hierarchy, we are done now */
2093 r
= cg_all_unified();
2099 /* Otherwise, do the same in the other hierarchies */
2100 for (c
= 0; c
< _CGROUP_CONTROLLER_MAX
; c
++) {
2101 CGroupMask bit
= CGROUP_CONTROLLER_TO_MASK(c
);
2104 n
= cgroup_controller_to_string(c
);
2107 (void) cg_create(n
, path
);
2108 else if (supported
& bit
)
2109 (void) cg_trim(n
, path
, true);
2115 int cg_attach_everywhere(CGroupMask supported
, const char *path
, pid_t pid
, cg_migrate_callback_t path_callback
, void *userdata
) {
2119 r
= cg_attach(SYSTEMD_CGROUP_CONTROLLER
, path
, pid
);
2123 r
= cg_all_unified();
2129 for (c
= 0; c
< _CGROUP_CONTROLLER_MAX
; c
++) {
2130 CGroupMask bit
= CGROUP_CONTROLLER_TO_MASK(c
);
2131 const char *p
= NULL
;
2133 if (!(supported
& bit
))
2137 p
= path_callback(bit
, userdata
);
2142 (void) cg_attach_fallback(cgroup_controller_to_string(c
), p
, pid
);
2148 int cg_attach_many_everywhere(CGroupMask supported
, const char *path
, Set
* pids
, cg_migrate_callback_t path_callback
, void *userdata
) {
2153 SET_FOREACH(pidp
, pids
, i
) {
2154 pid_t pid
= PTR_TO_PID(pidp
);
2157 q
= cg_attach_everywhere(supported
, path
, pid
, path_callback
, userdata
);
2158 if (q
< 0 && r
>= 0)
2165 int cg_migrate_everywhere(CGroupMask supported
, const char *from
, const char *to
, cg_migrate_callback_t to_callback
, void *userdata
) {
2169 if (!path_equal(from
, to
)) {
2170 r
= cg_migrate_recursive(SYSTEMD_CGROUP_CONTROLLER
, from
, SYSTEMD_CGROUP_CONTROLLER
, to
, CGROUP_REMOVE
);
2175 q
= cg_all_unified();
2181 for (c
= 0; c
< _CGROUP_CONTROLLER_MAX
; c
++) {
2182 CGroupMask bit
= CGROUP_CONTROLLER_TO_MASK(c
);
2183 const char *p
= NULL
;
2185 if (!(supported
& bit
))
2189 p
= to_callback(bit
, userdata
);
2194 (void) cg_migrate_recursive_fallback(SYSTEMD_CGROUP_CONTROLLER
, to
, cgroup_controller_to_string(c
), p
, 0);
2200 int cg_trim_everywhere(CGroupMask supported
, const char *path
, bool delete_root
) {
2204 r
= cg_trim(SYSTEMD_CGROUP_CONTROLLER
, path
, delete_root
);
2208 q
= cg_all_unified();
2214 for (c
= 0; c
< _CGROUP_CONTROLLER_MAX
; c
++) {
2215 CGroupMask bit
= CGROUP_CONTROLLER_TO_MASK(c
);
2217 if (!(supported
& bit
))
2220 (void) cg_trim(cgroup_controller_to_string(c
), path
, delete_root
);
2226 int cg_mask_to_string(CGroupMask mask
, char **ret
) {
2227 _cleanup_free_
char *s
= NULL
;
2228 size_t n
= 0, allocated
= 0;
2239 for (c
= 0; c
< _CGROUP_CONTROLLER_MAX
; c
++) {
2243 if (!(mask
& CGROUP_CONTROLLER_TO_MASK(c
)))
2246 k
= cgroup_controller_to_string(c
);
2249 if (!GREEDY_REALLOC(s
, allocated
, n
+ space
+ l
+ 1))
2254 memcpy(s
+ n
+ space
, k
, l
);
2269 int cg_mask_from_string(const char *value
, CGroupMask
*mask
) {
2274 _cleanup_free_
char *n
= NULL
;
2278 r
= extract_first_word(&value
, &n
, NULL
, 0);
2284 v
= cgroup_controller_from_string(n
);
2288 *mask
|= CGROUP_CONTROLLER_TO_MASK(v
);
2293 int cg_mask_supported(CGroupMask
*ret
) {
2294 CGroupMask mask
= 0;
2297 /* Determines the mask of supported cgroup controllers. Only
2298 * includes controllers we can make sense of and that are
2299 * actually accessible. */
2301 r
= cg_all_unified();
2305 _cleanup_free_
char *root
= NULL
, *controllers
= NULL
, *path
= NULL
;
2307 /* In the unified hierarchy we can read the supported
2308 * and accessible controllers from a the top-level
2309 * cgroup attribute */
2311 r
= cg_get_root_path(&root
);
2315 r
= cg_get_path(SYSTEMD_CGROUP_CONTROLLER
, root
, "cgroup.controllers", &path
);
2319 r
= read_one_line_file(path
, &controllers
);
2323 r
= cg_mask_from_string(controllers
, &mask
);
2327 /* Currently, we support the cpu, memory, io and pids
2328 * controller in the unified hierarchy, mask
2329 * everything else off. */
2330 mask
&= CGROUP_MASK_CPU
| CGROUP_MASK_MEMORY
| CGROUP_MASK_IO
| CGROUP_MASK_PIDS
;
2335 /* In the legacy hierarchy, we check whether which
2336 * hierarchies are mounted. */
2338 for (c
= 0; c
< _CGROUP_CONTROLLER_MAX
; c
++) {
2341 n
= cgroup_controller_to_string(c
);
2342 if (controller_is_accessible(n
) >= 0)
2343 mask
|= CGROUP_CONTROLLER_TO_MASK(c
);
2351 int cg_kernel_controllers(Set
**ret
) {
2352 _cleanup_set_free_free_ Set
*controllers
= NULL
;
2353 _cleanup_fclose_
FILE *f
= NULL
;
2358 /* Determines the full list of kernel-known controllers. Might
2359 * include controllers we don't actually support, arbitrary
2360 * named hierarchies and controllers that aren't currently
2361 * accessible (because not mounted). */
2363 controllers
= set_new(&string_hash_ops
);
2367 f
= fopen("/proc/cgroups", "re");
2369 if (errno
== ENOENT
) {
2377 (void) __fsetlocking(f
, FSETLOCKING_BYCALLER
);
2379 /* Ignore the header line */
2380 (void) read_line(f
, (size_t) -1, NULL
);
2387 if (fscanf(f
, "%ms %*i %*i %i", &controller
, &enabled
) != 2) {
2392 if (ferror(f
) && errno
> 0)
2403 if (!cg_controller_is_valid(controller
)) {
2408 r
= set_consume(controllers
, controller
);
2419 static thread_local CGroupUnified unified_cache
= CGROUP_UNIFIED_UNKNOWN
;
2421 /* The hybrid mode was initially implemented in v232 and simply mounted cgroup v2 on /sys/fs/cgroup/systemd. This
2422 * unfortunately broke other tools (such as docker) which expected the v1 "name=systemd" hierarchy on
2423 * /sys/fs/cgroup/systemd. From v233 and on, the hybrid mode mountnbs v2 on /sys/fs/cgroup/unified and maintains
2424 * "name=systemd" hierarchy on /sys/fs/cgroup/systemd for compatibility with other tools.
2426 * To keep live upgrade working, we detect and support v232 layout. When v232 layout is detected, to keep cgroup v2
2427 * process management but disable the compat dual layout, we return %true on
2428 * cg_unified_controller(SYSTEMD_CGROUP_CONTROLLER) and %false on cg_hybrid_unified().
2430 static thread_local
bool unified_systemd_v232
;
2432 static int cg_unified_update(void) {
2436 /* Checks if we support the unified hierarchy. Returns an
2437 * error when the cgroup hierarchies aren't mounted yet or we
2438 * have any other trouble determining if the unified hierarchy
2441 if (unified_cache
>= CGROUP_UNIFIED_NONE
)
2444 if (statfs("/sys/fs/cgroup/", &fs
) < 0)
2445 return log_debug_errno(errno
, "statfs(\"/sys/fs/cgroup/\" failed: %m");
2447 if (F_TYPE_EQUAL(fs
.f_type
, CGROUP2_SUPER_MAGIC
)) {
2448 log_debug("Found cgroup2 on /sys/fs/cgroup/, full unified hierarchy");
2449 unified_cache
= CGROUP_UNIFIED_ALL
;
2450 } else if (F_TYPE_EQUAL(fs
.f_type
, TMPFS_MAGIC
)) {
2451 if (statfs("/sys/fs/cgroup/unified/", &fs
) == 0 &&
2452 F_TYPE_EQUAL(fs
.f_type
, CGROUP2_SUPER_MAGIC
)) {
2453 log_debug("Found cgroup2 on /sys/fs/cgroup/unified, unified hierarchy for systemd controller");
2454 unified_cache
= CGROUP_UNIFIED_SYSTEMD
;
2455 unified_systemd_v232
= false;
2457 if (statfs("/sys/fs/cgroup/systemd/", &fs
) < 0)
2458 return log_debug_errno(errno
, "statfs(\"/sys/fs/cgroup/systemd\" failed: %m");
2460 if (F_TYPE_EQUAL(fs
.f_type
, CGROUP2_SUPER_MAGIC
)) {
2461 log_debug("Found cgroup2 on /sys/fs/cgroup/systemd, unified hierarchy for systemd controller (v232 variant)");
2462 unified_cache
= CGROUP_UNIFIED_SYSTEMD
;
2463 unified_systemd_v232
= true;
2464 } else if (F_TYPE_EQUAL(fs
.f_type
, CGROUP_SUPER_MAGIC
)) {
2465 log_debug("Found cgroup on /sys/fs/cgroup/systemd, legacy hierarchy");
2466 unified_cache
= CGROUP_UNIFIED_NONE
;
2468 log_debug("Unexpected filesystem type %llx mounted on /sys/fs/cgroup/systemd, assuming legacy hierarchy",
2469 (unsigned long long) fs
.f_type
);
2470 unified_cache
= CGROUP_UNIFIED_NONE
;
2474 log_debug("Unknown filesystem type %llx mounted on /sys/fs/cgroup.",
2475 (unsigned long long) fs
.f_type
);
2482 int cg_unified_controller(const char *controller
) {
2485 r
= cg_unified_update();
2489 if (unified_cache
== CGROUP_UNIFIED_NONE
)
2492 if (unified_cache
>= CGROUP_UNIFIED_ALL
)
2495 return streq_ptr(controller
, SYSTEMD_CGROUP_CONTROLLER
);
2498 int cg_all_unified(void) {
2501 r
= cg_unified_update();
2505 return unified_cache
>= CGROUP_UNIFIED_ALL
;
2508 int cg_hybrid_unified(void) {
2511 r
= cg_unified_update();
2515 return unified_cache
== CGROUP_UNIFIED_SYSTEMD
&& !unified_systemd_v232
;
2518 int cg_unified_flush(void) {
2519 unified_cache
= CGROUP_UNIFIED_UNKNOWN
;
2521 return cg_unified_update();
2524 int cg_enable_everywhere(CGroupMask supported
, CGroupMask mask
, const char *p
) {
2525 _cleanup_fclose_
FILE *f
= NULL
;
2526 _cleanup_free_
char *fs
= NULL
;
2535 r
= cg_all_unified();
2538 if (r
== 0) /* on the legacy hiearchy there's no joining of controllers defined */
2541 r
= cg_get_path(SYSTEMD_CGROUP_CONTROLLER
, p
, "cgroup.subtree_control", &fs
);
2545 for (c
= 0; c
< _CGROUP_CONTROLLER_MAX
; c
++) {
2546 CGroupMask bit
= CGROUP_CONTROLLER_TO_MASK(c
);
2549 if (!(supported
& bit
))
2552 n
= cgroup_controller_to_string(c
);
2554 char s
[1 + strlen(n
) + 1];
2556 s
[0] = mask
& bit
? '+' : '-';
2560 f
= fopen(fs
, "we");
2562 log_debug_errno(errno
, "Failed to open cgroup.subtree_control file of %s: %m", p
);
2567 r
= write_string_stream(f
, s
, 0);
2569 log_debug_errno(r
, "Failed to enable controller %s for %s (%s): %m", n
, p
, fs
);
2576 bool cg_is_unified_wanted(void) {
2577 static thread_local
int wanted
= -1;
2580 const bool is_default
= DEFAULT_HIERARCHY
== CGROUP_UNIFIED_ALL
;
2582 /* If we have a cached value, return that. */
2586 /* If the hierarchy is already mounted, then follow whatever
2587 * was chosen for it. */
2588 if (cg_unified_flush() >= 0)
2589 return (wanted
= unified_cache
>= CGROUP_UNIFIED_ALL
);
2591 /* Otherwise, let's see what the kernel command line has to say.
2592 * Since checking is expensive, cache a non-error result. */
2593 r
= proc_cmdline_get_bool("systemd.unified_cgroup_hierarchy", &b
);
2595 return (wanted
= r
> 0 ? b
: is_default
);
2598 bool cg_is_legacy_wanted(void) {
2599 static thread_local
int wanted
= -1;
2601 /* If we have a cached value, return that. */
2605 /* Check if we have cgroups2 already mounted. */
2606 if (cg_unified_flush() >= 0 &&
2607 unified_cache
== CGROUP_UNIFIED_ALL
)
2608 return (wanted
= false);
2610 /* Otherwise, assume that at least partial legacy is wanted,
2611 * since cgroups2 should already be mounted at this point. */
2612 return (wanted
= true);
2615 bool cg_is_hybrid_wanted(void) {
2616 static thread_local
int wanted
= -1;
2619 const bool is_default
= DEFAULT_HIERARCHY
>= CGROUP_UNIFIED_SYSTEMD
;
2620 /* We default to true if the default is "hybrid", obviously,
2621 * but also when the default is "unified", because if we get
2622 * called, it means that unified hierarchy was not mounted. */
2624 /* If we have a cached value, return that. */
2628 /* If the hierarchy is already mounted, then follow whatever
2629 * was chosen for it. */
2630 if (cg_unified_flush() >= 0 &&
2631 unified_cache
== CGROUP_UNIFIED_ALL
)
2632 return (wanted
= false);
2634 /* Otherwise, let's see what the kernel command line has to say.
2635 * Since checking is expensive, cache a non-error result. */
2636 r
= proc_cmdline_get_bool("systemd.legacy_systemd_cgroup_controller", &b
);
2638 /* The meaning of the kernel option is reversed wrt. to the return value
2639 * of this function, hence the negation. */
2640 return (wanted
= r
> 0 ? !b
: is_default
);
2643 int cg_weight_parse(const char *s
, uint64_t *ret
) {
2648 *ret
= CGROUP_WEIGHT_INVALID
;
2652 r
= safe_atou64(s
, &u
);
2656 if (u
< CGROUP_WEIGHT_MIN
|| u
> CGROUP_WEIGHT_MAX
)
2663 const uint64_t cgroup_io_limit_defaults
[_CGROUP_IO_LIMIT_TYPE_MAX
] = {
2664 [CGROUP_IO_RBPS_MAX
] = CGROUP_LIMIT_MAX
,
2665 [CGROUP_IO_WBPS_MAX
] = CGROUP_LIMIT_MAX
,
2666 [CGROUP_IO_RIOPS_MAX
] = CGROUP_LIMIT_MAX
,
2667 [CGROUP_IO_WIOPS_MAX
] = CGROUP_LIMIT_MAX
,
2670 static const char* const cgroup_io_limit_type_table
[_CGROUP_IO_LIMIT_TYPE_MAX
] = {
2671 [CGROUP_IO_RBPS_MAX
] = "IOReadBandwidthMax",
2672 [CGROUP_IO_WBPS_MAX
] = "IOWriteBandwidthMax",
2673 [CGROUP_IO_RIOPS_MAX
] = "IOReadIOPSMax",
2674 [CGROUP_IO_WIOPS_MAX
] = "IOWriteIOPSMax",
2677 DEFINE_STRING_TABLE_LOOKUP(cgroup_io_limit_type
, CGroupIOLimitType
);
2679 int cg_cpu_shares_parse(const char *s
, uint64_t *ret
) {
2684 *ret
= CGROUP_CPU_SHARES_INVALID
;
2688 r
= safe_atou64(s
, &u
);
2692 if (u
< CGROUP_CPU_SHARES_MIN
|| u
> CGROUP_CPU_SHARES_MAX
)
2699 int cg_blkio_weight_parse(const char *s
, uint64_t *ret
) {
2704 *ret
= CGROUP_BLKIO_WEIGHT_INVALID
;
2708 r
= safe_atou64(s
, &u
);
2712 if (u
< CGROUP_BLKIO_WEIGHT_MIN
|| u
> CGROUP_BLKIO_WEIGHT_MAX
)
2719 bool is_cgroup_fs(const struct statfs
*s
) {
2720 return is_fs_type(s
, CGROUP_SUPER_MAGIC
) ||
2721 is_fs_type(s
, CGROUP2_SUPER_MAGIC
);
2724 bool fd_is_cgroup_fs(int fd
) {
2727 if (fstatfs(fd
, &s
) < 0)
2730 return is_cgroup_fs(&s
);
2733 static const char *cgroup_controller_table
[_CGROUP_CONTROLLER_MAX
] = {
2734 [CGROUP_CONTROLLER_CPU
] = "cpu",
2735 [CGROUP_CONTROLLER_CPUACCT
] = "cpuacct",
2736 [CGROUP_CONTROLLER_IO
] = "io",
2737 [CGROUP_CONTROLLER_BLKIO
] = "blkio",
2738 [CGROUP_CONTROLLER_MEMORY
] = "memory",
2739 [CGROUP_CONTROLLER_DEVICES
] = "devices",
2740 [CGROUP_CONTROLLER_PIDS
] = "pids",
2743 DEFINE_STRING_TABLE_LOOKUP(cgroup_controller
, CGroupController
);