X-Git-Url: http://www.chiark.greenend.org.uk/ucgi/~ianmdlvl/git?p=elogind.git;a=blobdiff_plain;f=src%2Fbasic%2Fcgroup-util.c;h=0f4c955a01491e9e3ac0054dcb6ff36c1362ca07;hp=ec42e258ab5728b07e08203482c957d534af4c8f;hb=2f4b99eec00077890bc6b247aa8630fb42eb698e;hpb=4ebadef69ff96e92122a133cfecadf60816c4fa3
diff --git a/src/basic/cgroup-util.c b/src/basic/cgroup-util.c
index ec42e258a..0f4c955a0 100644
--- a/src/basic/cgroup-util.c
+++ b/src/basic/cgroup-util.c
@@ -1,5 +1,3 @@
-/*-*- Mode: C; c-basic-offset: 8; indent-tabs-mode: nil -*-*/
-
/***
This file is part of systemd.
@@ -19,28 +17,46 @@
along with systemd; If not, see .
***/
+#include
#include
-#include
+#include
+//#include
#include
-#include
+//#include
#include
-#include
+#include
#include
+//#include
#include
-#include
+#include
+#include
-#include "set.h"
+#include "alloc-util.h"
+#include "cgroup-util.h"
+//#include "def.h"
+#include "dirent-util.h"
+#include "extract-word.h"
+#include "fd-util.h"
+#include "fileio.h"
+#include "format-util.h"
+#include "fs-util.h"
+//#include "log.h"
+#include "login-util.h"
#include "macro.h"
-#include "util.h"
-#include "formats-util.h"
-#include "process-util.h"
+//#include "missing.h"
+#include "mkdir.h"
+#include "parse-util.h"
#include "path-util.h"
+#include "proc-cmdline.h"
+#include "process-util.h"
+#include "set.h"
+//#include "special.h"
+#include "stat-util.h"
+#include "stdio-util.h"
+#include "string-table.h"
+#include "string-util.h"
#include "unit-name.h"
-#include "fileio.h"
-#include "special.h"
-#include "mkdir.h"
-#include "login-util.h"
-#include "cgroup-util.h"
+#include "user-util.h"
int cg_enumerate_processes(const char *controller, const char *path, FILE **_f) {
_cleanup_free_ char *fs = NULL;
@@ -76,7 +92,7 @@ int cg_read_pid(FILE *f, pid_t *_pid) {
if (feof(f))
return 0;
- return errno ? -errno : -EIO;
+ return errno > 0 ? -errno : -EIO;
}
if (ul <= 0)
@@ -86,6 +102,55 @@ int cg_read_pid(FILE *f, pid_t *_pid) {
return 1;
}
+int cg_read_event(const char *controller, const char *path, const char *event,
+ char **val)
+{
+ _cleanup_free_ char *events = NULL, *content = NULL;
+ char *p, *line;
+ int r;
+
+ r = cg_get_path(controller, path, "cgroup.events", &events);
+ if (r < 0)
+ return r;
+
+ r = read_full_file(events, &content, NULL);
+ if (r < 0)
+ return r;
+
+ p = content;
+ while ((line = strsep(&p, "\n"))) {
+ char *key;
+
+ key = strsep(&line, " ");
+ if (!key || !line)
+ return -EINVAL;
+
+ if (strcmp(key, event))
+ continue;
+
+ *val = strdup(line);
+ return 0;
+ }
+
+ return -ENOENT;
+}
+
+#if 0 /// UNNEEDED by elogind
+bool cg_ns_supported(void) {
+ static thread_local int enabled = -1;
+
+ if (enabled >= 0)
+ return enabled;
+
+ if (access("/proc/self/ns/cgroup", F_OK) == 0)
+ enabled = 1;
+ else
+ enabled = 0;
+
+ return enabled;
+}
+#endif //0
+
int cg_enumerate_subgroups(const char *controller, const char *path, DIR **_d) {
_cleanup_free_ char *fs = NULL;
int r;
@@ -119,8 +184,7 @@ int cg_read_subgroup(DIR *d, char **fn) {
if (de->d_type != DT_DIR)
continue;
- if (streq(de->d_name, ".") ||
- streq(de->d_name, ".."))
+ if (dot_or_dot_dot(de->d_name))
continue;
b = strdup(de->d_name);
@@ -146,10 +210,30 @@ int cg_rmdir(const char *controller, const char *path) {
if (r < 0 && errno != ENOENT)
return -errno;
+ r = cg_hybrid_unified();
+ if (r < 0)
+ return r;
+ if (r == 0)
+ return 0;
+
+ if (streq(controller, SYSTEMD_CGROUP_CONTROLLER)) {
+ r = cg_rmdir(SYSTEMD_CGROUP_CONTROLLER_LEGACY, path);
+ if (r < 0)
+ log_warning_errno(r, "Failed to remove compat systemd cgroup %s: %m", path);
+ }
+
return 0;
}
-int cg_kill(const char *controller, const char *path, int sig, bool sigcont, bool ignore_self, Set *s) {
+int cg_kill(
+ const char *controller,
+ const char *path,
+ int sig,
+ CGroupFlags flags,
+ Set *s,
+ cg_kill_log_func_t log_kill,
+ void *userdata) {
+
_cleanup_set_free_ Set *allocated_set = NULL;
bool done = false;
int r, ret = 0;
@@ -157,6 +241,11 @@ int cg_kill(const char *controller, const char *path, int sig, bool sigcont, boo
assert(sig >= 0);
+ /* Don't send SIGCONT twice. Also, SIGKILL always works even when process is suspended, hence don't send
+ * SIGCONT on SIGKILL. */
+ if (IN_SET(sig, SIGCONT, SIGKILL))
+ flags &= ~CGROUP_SIGCONT;
+
/* This goes through the tasks list and kills them all. This
* is repeated until no further processes are added to the
* tasks list, to properly handle forking processes */
@@ -184,19 +273,22 @@ int cg_kill(const char *controller, const char *path, int sig, bool sigcont, boo
while ((r = cg_read_pid(f, &pid)) > 0) {
- if (ignore_self && pid == my_pid)
+ if ((flags & CGROUP_IGNORE_SELF) && pid == my_pid)
continue;
if (set_get(s, PID_TO_PTR(pid)) == PID_TO_PTR(pid))
continue;
+ if (log_kill)
+ log_kill(pid, sig, userdata);
+
/* If we haven't killed this process yet, kill
* it */
if (kill(pid, sig) < 0) {
if (ret >= 0 && errno != ESRCH)
ret = -errno;
} else {
- if (sigcont && sig != SIGKILL)
+ if (flags & CGROUP_SIGCONT)
(void) kill(pid, SIGCONT);
if (ret == 0)
@@ -230,7 +322,15 @@ int cg_kill(const char *controller, const char *path, int sig, bool sigcont, boo
return ret;
}
-int cg_kill_recursive(const char *controller, const char *path, int sig, bool sigcont, bool ignore_self, bool rem, Set *s) {
+int cg_kill_recursive(
+ const char *controller,
+ const char *path,
+ int sig,
+ CGroupFlags flags,
+ Set *s,
+ cg_kill_log_func_t log_kill,
+ void *userdata) {
+
_cleanup_set_free_ Set *allocated_set = NULL;
_cleanup_closedir_ DIR *d = NULL;
int r, ret;
@@ -245,7 +345,7 @@ int cg_kill_recursive(const char *controller, const char *path, int sig, bool si
return -ENOMEM;
}
- ret = cg_kill(controller, path, sig, sigcont, ignore_self, s);
+ ret = cg_kill(controller, path, sig, flags, s, log_kill, userdata);
r = cg_enumerate_subgroups(controller, path, &d);
if (r < 0) {
@@ -258,20 +358,19 @@ int cg_kill_recursive(const char *controller, const char *path, int sig, bool si
while ((r = cg_read_subgroup(d, &fn)) > 0) {
_cleanup_free_ char *p = NULL;
- p = strjoin(path, "/", fn, NULL);
+ p = strjoin(path, "/", fn);
free(fn);
if (!p)
return -ENOMEM;
- r = cg_kill_recursive(controller, p, sig, sigcont, ignore_self, rem, s);
+ r = cg_kill_recursive(controller, p, sig, flags, s, log_kill, userdata);
if (r != 0 && ret >= 0)
ret = r;
}
-
if (ret >= 0 && r < 0)
ret = r;
- if (rem) {
+ if (flags & CGROUP_REMOVE) {
r = cg_rmdir(controller, path);
if (r < 0 && ret >= 0 && r != -ENOENT && r != -EBUSY)
return r;
@@ -280,7 +379,13 @@ int cg_kill_recursive(const char *controller, const char *path, int sig, bool si
return ret;
}
-int cg_migrate(const char *cfrom, const char *pfrom, const char *cto, const char *pto, bool ignore_self) {
+int cg_migrate(
+ const char *cfrom,
+ const char *pfrom,
+ const char *cto,
+ const char *pto,
+ CGroupFlags flags) {
+
bool done = false;
_cleanup_set_free_ Set *s = NULL;
int r, ret = 0;
@@ -297,6 +402,10 @@ int cg_migrate(const char *cfrom, const char *pfrom, const char *cto, const char
my_pid = getpid();
+ log_debug_elogind("Migrating \"%s\"/\"%s\" to \"%s\"/\"%s\" (%s)",
+ cfrom, pfrom, cto, pto,
+ (flags & CGROUP_IGNORE_SELF)
+ ? "ignoring self" : "watching self");
do {
_cleanup_fclose_ FILE *f = NULL;
pid_t pid = 0;
@@ -315,7 +424,7 @@ int cg_migrate(const char *cfrom, const char *pfrom, const char *cto, const char
/* This might do weird stuff if we aren't a
* single-threaded program. However, we
* luckily know we are not */
- if (ignore_self && pid == my_pid)
+ if ((flags & CGROUP_IGNORE_SELF) && pid == my_pid)
continue;
if (set_get(s, PID_TO_PTR(pid)) == PID_TO_PTR(pid))
@@ -363,8 +472,7 @@ int cg_migrate_recursive(
const char *pfrom,
const char *cto,
const char *pto,
- bool ignore_self,
- bool rem) {
+ CGroupFlags flags) {
_cleanup_closedir_ DIR *d = NULL;
int r, ret = 0;
@@ -375,7 +483,7 @@ int cg_migrate_recursive(
assert(cto);
assert(pto);
- ret = cg_migrate(cfrom, pfrom, cto, pto, ignore_self);
+ ret = cg_migrate(cfrom, pfrom, cto, pto, flags);
r = cg_enumerate_subgroups(cfrom, pfrom, &d);
if (r < 0) {
@@ -388,12 +496,12 @@ int cg_migrate_recursive(
while ((r = cg_read_subgroup(d, &fn)) > 0) {
_cleanup_free_ char *p = NULL;
- p = strjoin(pfrom, "/", fn, NULL);
+ p = strjoin(pfrom, "/", fn);
free(fn);
if (!p)
- return -ENOMEM;
+ return -ENOMEM;
- r = cg_migrate_recursive(cfrom, p, cto, pto, ignore_self, rem);
+ r = cg_migrate_recursive(cfrom, p, cto, pto, flags);
if (r != 0 && ret >= 0)
ret = r;
}
@@ -401,7 +509,7 @@ int cg_migrate_recursive(
if (r < 0 && ret >= 0)
ret = r;
- if (rem) {
+ if (flags & CGROUP_REMOVE) {
r = cg_rmdir(cfrom, pfrom);
if (r < 0 && ret >= 0 && r != -ENOENT && r != -EBUSY)
return r;
@@ -415,8 +523,7 @@ int cg_migrate_recursive_fallback(
const char *pfrom,
const char *cto,
const char *pto,
- bool ignore_self,
- bool rem) {
+ CGroupFlags flags) {
int r;
@@ -425,7 +532,7 @@ int cg_migrate_recursive_fallback(
assert(cto);
assert(pto);
- r = cg_migrate_recursive(cfrom, pfrom, cto, pto, ignore_self, rem);
+ r = cg_migrate_recursive(cfrom, pfrom, cto, pto, flags);
if (r < 0) {
char prefix[strlen(pto) + 1];
@@ -434,7 +541,7 @@ int cg_migrate_recursive_fallback(
PATH_FOREACH_PREFIX(prefix, pto) {
int q;
- q = cg_migrate_recursive(cfrom, pfrom, cto, prefix, ignore_self, rem);
+ q = cg_migrate_recursive(cfrom, pfrom, cto, prefix, flags);
if (q >= 0)
return q;
}
@@ -453,11 +560,18 @@ static const char *controller_to_dirname(const char *controller) {
* just cuts off the name= prefixed used for named
* hierarchies, if it is specified. */
+ if (streq(controller, SYSTEMD_CGROUP_CONTROLLER)) {
+ if (cg_hybrid_unified() > 0)
+ controller = SYSTEMD_CGROUP_CONTROLLER_HYBRID;
+ else
+ controller = SYSTEMD_CGROUP_CONTROLLER_LEGACY;
+ }
+
e = startswith(controller, "name=");
if (e)
return e;
- return controller;
+ return controller;
}
static int join_path_legacy(const char *controller, const char *path, const char *suffix, char **fs) {
@@ -472,17 +586,17 @@ static int join_path_legacy(const char *controller, const char *path, const char
if (isempty(path) && isempty(suffix))
t = strappend("/sys/fs/cgroup/", dn);
else if (isempty(path))
- t = strjoin("/sys/fs/cgroup/", dn, "/", suffix, NULL);
+ t = strjoin("/sys/fs/cgroup/", dn, "/", suffix);
else if (isempty(suffix))
- t = strjoin("/sys/fs/cgroup/", dn, "/", path, NULL);
- else
- t = strjoin("/sys/fs/cgroup/", dn, "/", path, "/", suffix, NULL);
+ t = strjoin("/sys/fs/cgroup/", dn, "/", path);
+ else
+ t = strjoin("/sys/fs/cgroup/", dn, "/", path, "/", suffix);
if (!t)
return -ENOMEM;
*fs = t;
return 0;
- }
+}
static int join_path_unified(const char *path, const char *suffix, char **fs) {
char *t;
@@ -496,7 +610,7 @@ static int join_path_unified(const char *path, const char *suffix, char **fs) {
else if (isempty(suffix))
t = strappend("/sys/fs/cgroup/", path);
else
- t = strjoin("/sys/fs/cgroup/", path, "/", suffix, NULL);
+ t = strjoin("/sys/fs/cgroup/", path, "/", suffix);
if (!t)
return -ENOMEM;
@@ -505,7 +619,7 @@ static int join_path_unified(const char *path, const char *suffix, char **fs) {
}
int cg_get_path(const char *controller, const char *path, const char *suffix, char **fs) {
- int unified, r;
+ int r;
assert(fs);
@@ -516,14 +630,14 @@ int cg_get_path(const char *controller, const char *path, const char *suffix, ch
* *below* the controllers, without any prefix. */
if (!path && !suffix)
- return -EINVAL;
+ return -EINVAL;
if (!suffix)
t = strdup(path);
else if (!path)
t = strdup(suffix);
else
- t = strjoin(path, "/", suffix, NULL);
+ t = strjoin(path, "/", suffix);
if (!t)
return -ENOMEM;
@@ -534,23 +648,22 @@ int cg_get_path(const char *controller, const char *path, const char *suffix, ch
if (!cg_controller_is_valid(controller))
return -EINVAL;
- unified = cg_unified();
- if (unified < 0)
- return unified;
-
- if (unified > 0)
+ r = cg_all_unified();
+ if (r < 0)
+ return r;
+ if (r > 0)
r = join_path_unified(path, suffix, fs);
else
r = join_path_legacy(controller, path, suffix, fs);
- if (r < 0)
- return r;
+ if (r < 0)
+ return r;
path_kill_slashes(*fs);
return 0;
- }
+}
static int controller_is_accessible(const char *controller) {
- int unified;
+ int r;
assert(controller);
@@ -562,15 +675,15 @@ static int controller_is_accessible(const char *controller) {
if (!cg_controller_is_valid(controller))
return -EINVAL;
- unified = cg_unified();
- if (unified < 0)
- return unified;
- if (unified > 0) {
+ r = cg_all_unified();
+ if (r < 0)
+ return r;
+ if (r > 0) {
/* We don't support named hierarchies if we are using
* the unified hierarchy. */
- if (streq(controller, ELOGIND_CGROUP_CONTROLLER))
- return 0;
+ if (streq(controller, SYSTEMD_CGROUP_CONTROLLER))
+ return 0;
if (startswith(controller, "name="))
return -EOPNOTSUPP;
@@ -581,8 +694,8 @@ static int controller_is_accessible(const char *controller) {
dn = controller_to_dirname(controller);
cc = strjoina("/sys/fs/cgroup/", dn);
- if (laccess(cc, F_OK) < 0)
- return -errno;
+ if (laccess(cc, F_OK) < 0)
+ return -errno;
}
return 0;
@@ -619,7 +732,7 @@ static int trim_cb(const char *path, const struct stat *sb, int typeflag, struct
int cg_trim(const char *controller, const char *path, bool delete_root) {
_cleanup_free_ char *fs = NULL;
- int r = 0;
+ int r = 0, q;
assert(path);
@@ -631,7 +744,7 @@ int cg_trim(const char *controller, const char *path, bool delete_root) {
if (nftw(fs, trim_cb, 64, FTW_DEPTH|FTW_MOUNT|FTW_PHYS) != 0) {
if (errno == ENOENT)
r = 0;
- else if (errno != 0)
+ else if (errno > 0)
r = -errno;
else
r = -EIO;
@@ -642,6 +755,15 @@ int cg_trim(const char *controller, const char *path, bool delete_root) {
return -errno;
}
+ q = cg_hybrid_unified();
+ if (q < 0)
+ return q;
+ if (q > 0 && streq(controller, SYSTEMD_CGROUP_CONTROLLER)) {
+ q = cg_trim(SYSTEMD_CGROUP_CONTROLLER_LEGACY, path, delete_root);
+ if (q < 0)
+ log_warning_errno(q, "Failed to trim compat systemd cgroup %s: %m", path);
+ }
+
return r;
}
@@ -665,11 +787,19 @@ int cg_create(const char *controller, const char *path) {
return -errno;
}
+ r = cg_hybrid_unified();
+ if (r < 0)
+ return r;
+
+ if (r > 0 && streq(controller, SYSTEMD_CGROUP_CONTROLLER)) {
+ r = cg_create(SYSTEMD_CGROUP_CONTROLLER_LEGACY, path);
+ if (r < 0)
+ log_warning_errno(r, "Failed to create compat systemd cgroup %s: %m", path);
+ }
+
return 1;
}
-/// UNNEEDED by elogind
-#if 0
int cg_create_and_attach(const char *controller, const char *path, pid_t pid) {
int r, q;
@@ -686,7 +816,6 @@ int cg_create_and_attach(const char *controller, const char *path, pid_t pid) {
/* This does not remove the cgroup on failure */
return r;
}
-#endif // 0
int cg_attach(const char *controller, const char *path, pid_t pid) {
_cleanup_free_ char *fs = NULL;
@@ -703,9 +832,23 @@ int cg_attach(const char *controller, const char *path, pid_t pid) {
if (pid == 0)
pid = getpid();
- snprintf(c, sizeof(c), PID_FMT"\n", pid);
+ xsprintf(c, PID_FMT "\n", pid);
+
+ r = write_string_file(fs, c, 0);
+ if (r < 0)
+ return r;
+
+ r = cg_hybrid_unified();
+ if (r < 0)
+ return r;
- return write_string_file_no_create(fs, c);
+ if (r > 0 && streq(controller, SYSTEMD_CGROUP_CONTROLLER)) {
+ r = cg_attach(SYSTEMD_CGROUP_CONTROLLER_LEGACY, path, pid);
+ if (r < 0)
+ log_warning_errno(r, "Failed to attach "PID_FMT" to compat systemd cgroup %s: %m", pid, path);
+ }
+
+ return 0;
}
int cg_attach_fallback(const char *controller, const char *path, pid_t pid) {
@@ -734,8 +877,7 @@ int cg_attach_fallback(const char *controller, const char *path, pid_t pid) {
return r;
}
-/// UNNEEDED by elogind
-#if 0
+#if 0 /// UNNEEDED by elogind
int cg_set_group_access(
const char *controller,
const char *path,
@@ -756,7 +898,20 @@ int cg_set_group_access(
if (r < 0)
return r;
- return chmod_and_chown(fs, mode, uid, gid);
+ r = chmod_and_chown(fs, mode, uid, gid);
+ if (r < 0)
+ return r;
+
+ r = cg_hybrid_unified();
+ if (r < 0)
+ return r;
+ if (r > 0 && streq(controller, SYSTEMD_CGROUP_CONTROLLER)) {
+ r = cg_set_group_access(SYSTEMD_CGROUP_CONTROLLER_LEGACY, path, mode, uid, gid);
+ if (r < 0)
+ log_warning_errno(r, "Failed to set group access on compat systemd cgroup %s: %m", path);
+ }
+
+ return 0;
}
int cg_set_task_access(
@@ -767,7 +922,7 @@ int cg_set_task_access(
gid_t gid) {
_cleanup_free_ char *fs = NULL, *procs = NULL;
- int r, unified;
+ int r;
assert(path);
@@ -785,45 +940,97 @@ int cg_set_task_access(
if (r < 0)
return r;
- unified = cg_unified();
- if (unified < 0)
- return unified;
- if (unified)
- return 0;
+ r = cg_unified_controller(controller);
+ if (r < 0)
+ return r;
+ if (r == 0) {
+ /* Compatibility, Always keep values for "tasks" in sync with
+ * "cgroup.procs" */
+ if (cg_get_path(controller, path, "tasks", &procs) >= 0)
+ (void) chmod_and_chown(procs, mode, uid, gid);
+ }
- /* Compatibility, Always keep values for "tasks" in sync with
- * "cgroup.procs" */
- if (cg_get_path(controller, path, "tasks", &procs) >= 0)
- (void) chmod_and_chown(procs, mode, uid, gid);
+ r = cg_hybrid_unified();
+ if (r < 0)
+ return r;
+ if (r > 0 && streq(controller, SYSTEMD_CGROUP_CONTROLLER)) {
+ r = cg_set_task_access(SYSTEMD_CGROUP_CONTROLLER_LEGACY, path, mode, uid, gid);
+ if (r < 0)
+ log_warning_errno(r, "Failed to set task access on compat systemd cgroup %s: %m", path);
+ }
return 0;
}
+
+int cg_set_xattr(const char *controller, const char *path, const char *name, const void *value, size_t size, int flags) {
+ _cleanup_free_ char *fs = NULL;
+ int r;
+
+ assert(path);
+ assert(name);
+ assert(value || size <= 0);
+
+ r = cg_get_path(controller, path, NULL, &fs);
+ if (r < 0)
+ return r;
+
+ if (setxattr(fs, name, value, size, flags) < 0)
+ return -errno;
+
+ return 0;
+}
+
+int cg_get_xattr(const char *controller, const char *path, const char *name, void *value, size_t size) {
+ _cleanup_free_ char *fs = NULL;
+ ssize_t n;
+ int r;
+
+ assert(path);
+ assert(name);
+
+ r = cg_get_path(controller, path, NULL, &fs);
+ if (r < 0)
+ return r;
+
+ n = getxattr(fs, name, value, size);
+ if (n < 0)
+ return -errno;
+
+ return (int) n;
+}
#endif // 0
int cg_pid_get_path(const char *controller, pid_t pid, char **path) {
_cleanup_fclose_ FILE *f = NULL;
char line[LINE_MAX];
- const char *fs;
+ const char *fs, *controller_str;
size_t cs = 0;
int unified;
assert(path);
assert(pid >= 0);
- unified = cg_unified();
- if (unified < 0)
- return unified;
- if (unified == 0) {
if (controller) {
if (!cg_controller_is_valid(controller))
return -EINVAL;
} else
- controller = ELOGIND_CGROUP_CONTROLLER;
+ controller = SYSTEMD_CGROUP_CONTROLLER;
- cs = strlen(controller);
+ unified = cg_unified_controller(controller);
+ if (unified < 0)
+ return unified;
+ if (unified == 0) {
+ if (streq(controller, SYSTEMD_CGROUP_CONTROLLER))
+ controller_str = SYSTEMD_CGROUP_CONTROLLER_LEGACY;
+ else
+ controller_str = controller;
+
+ cs = strlen(controller_str);
}
fs = procfs_file_alloca(pid, "cgroup");
+ log_debug_elogind("Searching for PID %u in \"%s\" (controller \"%s\")",
+ pid, fs, controller);
f = fopen(fs, "re");
if (!f)
return errno == ENOENT ? -ESRCH : -errno;
@@ -843,31 +1050,32 @@ int cg_pid_get_path(const char *controller, pid_t pid, char **path) {
continue;
} else {
char *l;
- size_t k;
- const char *word, *state;
- bool found = false;
+ size_t k;
+ const char *word, *state;
+ bool found = false;
- l = strchr(line, ':');
- if (!l)
- continue;
+ l = strchr(line, ':');
+ if (!l)
+ continue;
- l++;
- e = strchr(l, ':');
- if (!e)
- continue;
+ l++;
+ e = strchr(l, ':');
+ if (!e)
+ continue;
- *e = 0;
- FOREACH_WORD_SEPARATOR(word, k, l, ",", state) {
- if (k == cs && memcmp(word, controller, cs) == 0) {
- found = true;
- break;
+ *e = 0;
+ FOREACH_WORD_SEPARATOR(word, k, l, ",", state) {
+ if (k == cs && memcmp(word, controller_str, cs) == 0) {
+ found = true;
+ break;
+ }
}
- }
- if (!found)
- continue;
+ if (!found)
+ continue;
}
+ log_debug_elogind("Found %s:%s", line, e+1);
p = strdup(e + 1);
if (!p)
return -ENOMEM;
@@ -879,19 +1087,17 @@ int cg_pid_get_path(const char *controller, pid_t pid, char **path) {
return -ENODATA;
}
-/// UNNEEDED by elogind
-#if 0
int cg_install_release_agent(const char *controller, const char *agent) {
_cleanup_free_ char *fs = NULL, *contents = NULL;
const char *sc;
- int r, unified;
+ int r;
assert(agent);
- unified = cg_unified();
- if (unified < 0)
- return unified;
- if (unified) /* doesn't apply to unified hierarchy */
+ r = cg_unified_controller(controller);
+ if (r < 0)
+ return r;
+ if (r > 0) /* doesn't apply to unified hierarchy */
return -EOPNOTSUPP;
r = cg_get_path(controller, NULL, "release_agent", &fs);
@@ -904,7 +1110,7 @@ int cg_install_release_agent(const char *controller, const char *agent) {
sc = strstrip(contents);
if (isempty(sc)) {
- r = write_string_file_no_create(fs, agent);
+ r = write_string_file(fs, agent, 0);
if (r < 0)
return r;
} else if (!path_equal(sc, agent))
@@ -922,7 +1128,7 @@ int cg_install_release_agent(const char *controller, const char *agent) {
sc = strstrip(contents);
if (streq(sc, "0")) {
- r = write_string_file_no_create(fs, "1");
+ r = write_string_file(fs, "1", 0);
if (r < 0)
return r;
@@ -937,19 +1143,19 @@ int cg_install_release_agent(const char *controller, const char *agent) {
int cg_uninstall_release_agent(const char *controller) {
_cleanup_free_ char *fs = NULL;
- int r, unified;
+ int r;
- unified = cg_unified();
- if (unified < 0)
- return unified;
- if (unified) /* Doesn't apply to unified hierarchy */
+ r = cg_unified_controller(controller);
+ if (r < 0)
+ return r;
+ if (r > 0) /* Doesn't apply to unified hierarchy */
return -EOPNOTSUPP;
r = cg_get_path(controller, NULL, "notify_on_release", &fs);
if (r < 0)
return r;
- r = write_string_file_no_create(fs, "0");
+ r = write_string_file(fs, "0", 0);
if (r < 0)
return r;
@@ -959,13 +1165,12 @@ int cg_uninstall_release_agent(const char *controller) {
if (r < 0)
return r;
- r = write_string_file_no_create(fs, "");
+ r = write_string_file(fs, "", 0);
if (r < 0)
return r;
return 0;
}
-#endif // 0
int cg_is_empty(const char *controller, const char *path) {
_cleanup_fclose_ FILE *f = NULL;
@@ -988,7 +1193,7 @@ int cg_is_empty(const char *controller, const char *path) {
}
int cg_is_empty_recursive(const char *controller, const char *path) {
- int unified, r;
+ int r;
assert(path);
@@ -996,55 +1201,48 @@ int cg_is_empty_recursive(const char *controller, const char *path) {
if (controller && (isempty(path) || path_equal(path, "/")))
return false;
- unified = cg_unified();
- if (unified < 0)
- return unified;
-
- if (unified > 0) {
- _cleanup_free_ char *populated = NULL, *t = NULL;
-
- /* On the unified hierarchy we can check empty state
- * via the "cgroup.populated" attribute. */
-
- r = cg_get_path(controller, path, "cgroup.populated", &populated);
+ r = cg_unified_controller(controller);
if (r < 0)
return r;
+ if (r > 0) {
+ _cleanup_free_ char *t = NULL;
- r = read_one_line_file(populated, &t);
- if (r == -ENOENT)
- return 1;
+ /* On the unified hierarchy we can check empty state
+ * via the "populated" attribute of "cgroup.events". */
+
+ r = cg_read_event(controller, path, "populated", &t);
if (r < 0)
return r;
return streq(t, "0");
} else {
- _cleanup_closedir_ DIR *d = NULL;
- char *fn;
+ _cleanup_closedir_ DIR *d = NULL;
+ char *fn;
r = cg_is_empty(controller, path);
- if (r <= 0)
- return r;
+ if (r <= 0)
+ return r;
- r = cg_enumerate_subgroups(controller, path, &d);
+ r = cg_enumerate_subgroups(controller, path, &d);
if (r == -ENOENT)
return 1;
- if (r < 0)
+ if (r < 0)
return r;
- while ((r = cg_read_subgroup(d, &fn)) > 0) {
- _cleanup_free_ char *p = NULL;
+ while ((r = cg_read_subgroup(d, &fn)) > 0) {
+ _cleanup_free_ char *p = NULL;
- p = strjoin(path, "/", fn, NULL);
- free(fn);
- if (!p)
- return -ENOMEM;
+ p = strjoin(path, "/", fn);
+ free(fn);
+ if (!p)
+ return -ENOMEM;
r = cg_is_empty_recursive(controller, p);
- if (r <= 0)
+ if (r <= 0)
+ return r;
+ }
+ if (r < 0)
return r;
- }
- if (r < 0)
- return r;
return true;
}
@@ -1157,16 +1355,17 @@ int cg_mangle_path(const char *path, char **result) {
if (r < 0)
return r;
- return cg_get_path(c ? c : ELOGIND_CGROUP_CONTROLLER, p ? p : "/", NULL, result);
+ return cg_get_path(c ?: SYSTEMD_CGROUP_CONTROLLER, p ?: "/", NULL, result);
}
int cg_get_root_path(char **path) {
+#if 0 /// elogind does not support systemd scopes and slices
char *p, *e;
int r;
assert(path);
- r = cg_pid_get_path(ELOGIND_CGROUP_CONTROLLER, 1, &p);
+ r = cg_pid_get_path(SYSTEMD_CGROUP_CONTROLLER, 1, &p);
if (r < 0)
return r;
@@ -1180,6 +1379,10 @@ int cg_get_root_path(char **path) {
*path = p;
return 0;
+#else
+ assert(path);
+ return cg_pid_get_path(SYSTEMD_CGROUP_CONTROLLER, 1, path);
+#endif // 0
}
int cg_shift_path(const char *cgroup, const char *root, const char **shifted) {
@@ -1199,10 +1402,15 @@ int cg_shift_path(const char *cgroup, const char *root, const char **shifted) {
return r;
root = rt;
+ log_debug_elogind("Determined root path: \"%s\"", root);
}
p = path_startswith(cgroup, root);
+#if 0 /// With other controllers, elogind might end up in /elogind, and *p is 0
if (p && p > cgroup)
+#else
+ if (p && p[0] && (p > cgroup))
+#endif // 0
*shifted = p - 1;
else
*shifted = cgroup;
@@ -1218,10 +1426,12 @@ int cg_pid_get_path_shifted(pid_t pid, const char *root, char **cgroup) {
assert(pid >= 0);
assert(cgroup);
- r = cg_pid_get_path(ELOGIND_CGROUP_CONTROLLER, pid, &raw);
+ r = cg_pid_get_path(SYSTEMD_CGROUP_CONTROLLER, pid, &raw);
if (r < 0)
return r;
+ log_debug_elogind("Shifting path: \"%s\" (PID %u, root: \"%s\")",
+ raw, pid, root ? root : "NULL");
r = cg_shift_path(raw, root, &c);
if (r < 0)
return r;
@@ -1238,11 +1448,13 @@ int cg_pid_get_path_shifted(pid_t pid, const char *root, char **cgroup) {
*cgroup = n;
}
+ log_debug_elogind("Resulting cgroup:\"%s\"", *cgroup);
return 0;
}
-int cg_path_decode_unit(const char *cgroup, char **unit){
+#if 0 /// UNNEEDED by elogind
+int cg_path_decode_unit(const char *cgroup, char **unit) {
char *c, *s;
size_t n;
@@ -1490,8 +1702,10 @@ int cg_pid_get_machine_name(pid_t pid, char **machine) {
return cg_path_get_machine_name(cgroup, machine);
}
+#endif // 0
int cg_path_get_session(const char *path, char **session) {
+#if 0 /// UNNEEDED by elogind
_cleanup_free_ char *unit = NULL;
char *start, *end;
int r;
@@ -1512,10 +1726,31 @@ int cg_path_get_session(const char *path, char **session) {
*end = 0;
if (!session_id_valid(start))
return -ENXIO;
+#else
+ /* Elogind uses a flat hierarchy, just "/SESSION". The only
+ wrinkle is that SESSION might be escaped. */
+ const char *e, *n, *start;
+
+ assert(path);
+ log_debug_elogind("path is \"%s\"", path);
+ assert(path[0] == '/');
+
+ e = path + 1;
+ n = strchrnul(e, '/');
+ if (e == n)
+ return -ENOENT;
+
+ start = strndupa(e, n - e);
+ start = cg_unescape(start);
+
+ if (!start[0])
+ return -ENOENT;
+#endif // 0
if (session) {
char *rr;
+ log_debug_elogind("found session: \"%s\"", start);
rr = strdup(start);
if (!rr)
return -ENOMEM;
@@ -1537,6 +1772,7 @@ int cg_pid_get_session(pid_t pid, char **session) {
return cg_path_get_session(cgroup, session);
}
+#if 0 /// UNNEEDED by elogind
int cg_path_get_owner_uid(const char *path, uid_t *uid) {
_cleanup_free_ char *slice = NULL;
char *start, *end;
@@ -1593,7 +1829,7 @@ int cg_path_get_slice(const char *p, char **slice) {
if (!e) {
char *s;
- s = strdup("-.slice");
+ s = strdup(SPECIAL_ROOT_SLICE);
if (!s)
return -ENOMEM;
@@ -1648,6 +1884,7 @@ int cg_pid_get_user_slice(pid_t pid, char **slice) {
return cg_path_get_user_slice(cgroup, slice);
}
+#endif // 0
char *cg_escape(const char *p) {
bool need_prefix = false;
@@ -1689,7 +1926,7 @@ char *cg_escape(const char *p) {
if (memcmp(p, n, l) != 0)
continue;
- need_prefix = true;
+ need_prefix = true;
break;
}
}
@@ -1698,7 +1935,7 @@ char *cg_escape(const char *p) {
if (need_prefix)
return strappend("_", p);
- return strdup(p);
+ return strdup(p);
}
char *cg_unescape(const char *p) {
@@ -1723,6 +1960,9 @@ bool cg_controller_is_valid(const char *p) {
if (!p)
return false;
+ if (streq(p, SYSTEMD_CGROUP_CONTROLLER))
+ return true;
+
s = startswith(p, "name=");
if (s)
p = s;
@@ -1740,8 +1980,7 @@ bool cg_controller_is_valid(const char *p) {
return true;
}
-/// UNNEEDED by elogind
-#if 0
+#if 0 /// UNNEEDED by elogind
int cg_slice_to_path(const char *unit, char **ret) {
_cleanup_free_ char *p = NULL, *s = NULL, *e = NULL;
const char *dash;
@@ -1750,7 +1989,7 @@ int cg_slice_to_path(const char *unit, char **ret) {
assert(unit);
assert(ret);
- if (streq(unit, "-.slice")) {
+ if (streq(unit, SPECIAL_ROOT_SLICE)) {
char *x;
x = strdup("");
@@ -1810,6 +2049,7 @@ int cg_slice_to_path(const char *unit, char **ret) {
return 0;
}
+#endif // 0
int cg_set_attribute(const char *controller, const char *path, const char *attribute, const char *value) {
_cleanup_free_ char *p = NULL;
@@ -1819,7 +2059,7 @@ int cg_set_attribute(const char *controller, const char *path, const char *attri
if (r < 0)
return r;
- return write_string_file_no_create(p, value);
+ return write_string_file(p, value, 0);
}
int cg_get_attribute(const char *controller, const char *path, const char *attribute, char **ret) {
@@ -1832,28 +2072,69 @@ int cg_get_attribute(const char *controller, const char *path, const char *attri
return read_one_line_file(p, ret);
}
-#endif // 0
-/// UNNEEDED by elogind
-#if 0
+#if 0 /// UNNEEDED by elogind
+int cg_get_keyed_attribute(const char *controller, const char *path, const char *attribute, const char **keys, char **values) {
+ _cleanup_free_ char *filename = NULL, *content = NULL;
+ char *line, *p;
+ int i, r;
+
+ for (i = 0; keys[i]; i++)
+ values[i] = NULL;
+
+ r = cg_get_path(controller, path, attribute, &filename);
+ if (r < 0)
+ return r;
+
+ r = read_full_file(filename, &content, NULL);
+ if (r < 0)
+ return r;
+
+ p = content;
+ while ((line = strsep(&p, "\n"))) {
+ char *key;
+
+ key = strsep(&line, " ");
+
+ for (i = 0; keys[i]; i++) {
+ if (streq(key, keys[i])) {
+ values[i] = strdup(line);
+ break;
+ }
+ }
+ }
+
+ for (i = 0; keys[i]; i++) {
+ if (!values[i]) {
+ for (i = 0; keys[i]; i++) {
+ free(values[i]);
+ values[i] = NULL;
+ }
+ return -ENOENT;
+ }
+ }
+
+ return 0;
+}
+
int cg_create_everywhere(CGroupMask supported, CGroupMask mask, const char *path) {
CGroupController c;
- int r, unified;
+ int r;
/* This one will create a cgroup in our private tree, but also
* duplicate it in the trees specified in mask, and remove it
* in all others */
/* First create the cgroup in our own hierarchy. */
- r = cg_create(ELOGIND_CGROUP_CONTROLLER, path);
+ r = cg_create(SYSTEMD_CGROUP_CONTROLLER, path);
if (r < 0)
return r;
/* If we are in the unified hierarchy, we are done now */
- unified = cg_unified();
- if (unified < 0)
- return unified;
- if (unified > 0)
+ r = cg_all_unified();
+ if (r < 0)
+ return r;
+ if (r > 0)
return 0;
/* Otherwise, do the same in the other hierarchies */
@@ -1871,34 +2152,33 @@ int cg_create_everywhere(CGroupMask supported, CGroupMask mask, const char *path
return 0;
}
-#endif // 0
int cg_attach_everywhere(CGroupMask supported, const char *path, pid_t pid, cg_migrate_callback_t path_callback, void *userdata) {
CGroupController c;
- int r, unified;
+ int r;
- r = cg_attach(ELOGIND_CGROUP_CONTROLLER, path, pid);
+ r = cg_attach(SYSTEMD_CGROUP_CONTROLLER, path, pid);
if (r < 0)
return r;
- unified = cg_unified();
- if (unified < 0)
- return unified;
- if (unified > 0)
+ r = cg_all_unified();
+ if (r < 0)
+ return r;
+ if (r > 0)
return 0;
for (c = 0; c < _CGROUP_CONTROLLER_MAX; c++) {
CGroupMask bit = CGROUP_CONTROLLER_TO_MASK(c);
- const char *p = NULL;
+ const char *p = NULL;
if (!(supported & bit))
continue;
- if (path_callback)
- p = path_callback(bit, userdata);
+ if (path_callback)
+ p = path_callback(bit, userdata);
- if (!p)
- p = path;
+ if (!p)
+ p = path;
(void) cg_attach_fallback(cgroup_controller_to_string(c), p, pid);
}
@@ -1906,8 +2186,6 @@ int cg_attach_everywhere(CGroupMask supported, const char *path, pid_t pid, cg_m
return 0;
}
-/// UNNEEDED by elogind
-#if 0
int cg_attach_many_everywhere(CGroupMask supported, const char *path, Set* pids, cg_migrate_callback_t path_callback, void *userdata) {
Iterator i;
void *pidp;
@@ -1927,34 +2205,34 @@ int cg_attach_many_everywhere(CGroupMask supported, const char *path, Set* pids,
int cg_migrate_everywhere(CGroupMask supported, const char *from, const char *to, cg_migrate_callback_t to_callback, void *userdata) {
CGroupController c;
- int r = 0, unified;
+ int r = 0, q;
if (!path_equal(from, to)) {
- r = cg_migrate_recursive(ELOGIND_CGROUP_CONTROLLER, from, ELOGIND_CGROUP_CONTROLLER, to, false, true);
+ r = cg_migrate_recursive(SYSTEMD_CGROUP_CONTROLLER, from, SYSTEMD_CGROUP_CONTROLLER, to, CGROUP_REMOVE);
if (r < 0)
return r;
}
- unified = cg_unified();
- if (unified < 0)
- return unified;
- if (unified > 0)
+ q = cg_all_unified();
+ if (q < 0)
+ return q;
+ if (q > 0)
return r;
for (c = 0; c < _CGROUP_CONTROLLER_MAX; c++) {
CGroupMask bit = CGROUP_CONTROLLER_TO_MASK(c);
- const char *p = NULL;
+ const char *p = NULL;
if (!(supported & bit))
continue;
- if (to_callback)
- p = to_callback(bit, userdata);
+ if (to_callback)
+ p = to_callback(bit, userdata);
- if (!p)
- p = to;
+ if (!p)
+ p = to;
- (void) cg_migrate_recursive_fallback(ELOGIND_CGROUP_CONTROLLER, to, cgroup_controller_to_string(c), p, false, false);
+ (void) cg_migrate_recursive_fallback(SYSTEMD_CGROUP_CONTROLLER, to, cgroup_controller_to_string(c), p, 0);
}
return 0;
@@ -1962,16 +2240,16 @@ int cg_migrate_everywhere(CGroupMask supported, const char *from, const char *to
int cg_trim_everywhere(CGroupMask supported, const char *path, bool delete_root) {
CGroupController c;
- int r, unified;
+ int r, q;
- r = cg_trim(ELOGIND_CGROUP_CONTROLLER, path, delete_root);
+ r = cg_trim(SYSTEMD_CGROUP_CONTROLLER, path, delete_root);
if (r < 0)
return r;
- unified = cg_unified();
- if (unified < 0)
- return unified;
- if (unified > 0)
+ q = cg_all_unified();
+ if (q < 0)
+ return q;
+ if (q > 0)
return r;
for (c = 0; c < _CGROUP_CONTROLLER_MAX; c++) {
@@ -1985,21 +2263,75 @@ int cg_trim_everywhere(CGroupMask supported, const char *path, bool delete_root)
return 0;
}
+#endif // 0
+
+int cg_mask_to_string(CGroupMask mask, char **ret) {
+ const char *controllers[_CGROUP_CONTROLLER_MAX + 1];
+ CGroupController c;
+ int i = 0;
+ char *s;
+
+ assert(ret);
+
+ if (mask == 0) {
+ *ret = NULL;
+ return 0;
+ }
+
+ for (c = 0; c < _CGROUP_CONTROLLER_MAX; c++) {
+
+ if (!(mask & CGROUP_CONTROLLER_TO_MASK(c)))
+ continue;
+
+ controllers[i++] = cgroup_controller_to_string(c);
+ controllers[i] = NULL;
+ }
+
+ s = strv_join((char **)controllers, NULL);
+ if (!s)
+ return -ENOMEM;
+
+ *ret = s;
+ return 0;
+}
+
+int cg_mask_from_string(const char *value, CGroupMask *mask) {
+ assert(mask);
+ assert(value);
+
+ for (;;) {
+ _cleanup_free_ char *n = NULL;
+ CGroupController v;
+ int r;
+
+ r = extract_first_word(&value, &n, NULL, 0);
+ if (r < 0)
+ return r;
+ if (r == 0)
+ break;
+
+ v = cgroup_controller_from_string(n);
+ if (v < 0)
+ continue;
+
+ *mask |= CGROUP_CONTROLLER_TO_MASK(v);
+ }
+ return 0;
+}
int cg_mask_supported(CGroupMask *ret) {
CGroupMask mask = 0;
- int r, unified;
+ int r;
/* Determines the mask of supported cgroup controllers. Only
* includes controllers we can make sense of and that are
* actually accessible. */
- unified = cg_unified();
- if (unified < 0)
- return unified;
- if (unified > 0) {
+ r = cg_all_unified();
+ if (r < 0)
+ return r;
+ if (r > 0) {
_cleanup_free_ char *root = NULL, *controllers = NULL, *path = NULL;
- const char *c;
/* In the unified hierarchy we can read the supported
* and accessible controllers from a the top-level
@@ -2009,7 +2341,7 @@ int cg_mask_supported(CGroupMask *ret) {
if (r < 0)
return r;
- r = cg_get_path(ELOGIND_CGROUP_CONTROLLER, root, "cgroup.controllers", &path);
+ r = cg_get_path(SYSTEMD_CGROUP_CONTROLLER, root, "cgroup.controllers", &path);
if (r < 0)
return r;
@@ -2017,27 +2349,14 @@ int cg_mask_supported(CGroupMask *ret) {
if (r < 0)
return r;
- c = controllers;
- for (;;) {
- _cleanup_free_ char *n = NULL;
- CGroupController v;
-
- r = extract_first_word(&c, &n, NULL, 0);
- if (r < 0)
- return r;
- if (r == 0)
- break;
-
- v = cgroup_controller_from_string(n);
- if (v < 0)
- continue;
-
- mask |= CGROUP_CONTROLLER_TO_MASK(v);
- }
+ r = cg_mask_from_string(controllers, &mask);
+ if (r < 0)
+ return r;
- /* Currently, we only support the memory controller in
- * the unified hierarchy, mask everything else off. */
- mask &= CGROUP_MASK_MEMORY;
+ /* Currently, we support the cpu, memory, io and pids
+ * controller in the unified hierarchy, mask
+ * everything else off. */
+ mask &= CGROUP_MASK_CPU | CGROUP_MASK_MEMORY | CGROUP_MASK_IO | CGROUP_MASK_PIDS;
} else {
CGroupController c;
@@ -2058,6 +2377,7 @@ int cg_mask_supported(CGroupMask *ret) {
return 0;
}
+#if 0 /// UNNEEDED by elogind
int cg_kernel_controllers(Set *controllers) {
_cleanup_fclose_ FILE *f = NULL;
char buf[LINE_MAX];
@@ -2090,7 +2410,7 @@ int cg_kernel_controllers(Set *controllers) {
if (feof(f))
break;
- if (ferror(f) && errno != 0)
+ if (ferror(f) && errno > 0)
return -errno;
return -EBADMSG;
@@ -2115,9 +2435,21 @@ int cg_kernel_controllers(Set *controllers) {
}
#endif // 0
-static thread_local int unified_cache = -1;
+static thread_local CGroupUnified unified_cache = CGROUP_UNIFIED_UNKNOWN;
+
+/* The hybrid mode was initially implemented in v232 and simply mounted cgroup v2 on /sys/fs/cgroup/systemd. This
+ * unfortunately broke other tools (such as docker) which expected the v1 "name=systemd" hierarchy on
+ * /sys/fs/cgroup/systemd. From v233 and on, the hybrid mode mountnbs v2 on /sys/fs/cgroup/unified and maintains
+ * "name=systemd" hierarchy on /sys/fs/cgroup/systemd for compatibility with other tools.
+ *
+ * To keep live upgrade working, we detect and support v232 layout. When v232 layout is detected, to keep cgroup v2
+ * process management but disable the compat dual layout, we return %true on
+ * cg_unified_controller(SYSTEMD_CGROUP_CONTROLLER) and %false on cg_hybrid_unified().
+ */
+static thread_local bool unified_systemd_v232;
+
+static int cg_unified_update(void) {
-int cg_unified(void) {
struct statfs fs;
/* Checks if we support the unified hierarchy. Returns an
@@ -2125,43 +2457,105 @@ int cg_unified(void) {
* have any other trouble determining if the unified hierarchy
* is supported. */
- if (unified_cache >= 0)
- return unified_cache;
+ if (unified_cache >= CGROUP_UNIFIED_NONE)
+ return 0;
if (statfs("/sys/fs/cgroup/", &fs) < 0)
return -errno;
- if (F_TYPE_EQUAL(fs.f_type, CGROUP_SUPER_MAGIC))
- unified_cache = true;
- else if (F_TYPE_EQUAL(fs.f_type, TMPFS_MAGIC))
- unified_cache = false;
- else
- return -ENOEXEC;
+#if 0 /// UNNEEDED by elogind
+ if (F_TYPE_EQUAL(fs.f_type, CGROUP2_SUPER_MAGIC))
+ unified_cache = CGROUP_UNIFIED_ALL;
+ else if (F_TYPE_EQUAL(fs.f_type, TMPFS_MAGIC)) {
+ if (statfs("/sys/fs/cgroup/unified/", &fs) == 0 &&
+ F_TYPE_EQUAL(fs.f_type, CGROUP2_SUPER_MAGIC)) {
+ unified_cache = CGROUP_UNIFIED_SYSTEMD;
+ unified_systemd_v232 = false;
+ } else if (statfs("/sys/fs/cgroup/systemd/", &fs) == 0 &&
+ F_TYPE_EQUAL(fs.f_type, CGROUP2_SUPER_MAGIC)) {
+ unified_cache = CGROUP_UNIFIED_SYSTEMD;
+ unified_systemd_v232 = true;
+ } else {
+ if (statfs("/sys/fs/cgroup/systemd/", &fs) < 0)
+ return -errno;
+ if (!F_TYPE_EQUAL(fs.f_type, CGROUP_SUPER_MAGIC))
+ return -ENOMEDIUM;
+ unified_cache = CGROUP_UNIFIED_NONE;
+ }
+ } else
+ return -ENOMEDIUM;
+#else
+ /* elogind can not support the unified hierarchy as a controller,
+ * so always assume a classical hierarchy.
+ * If, and only *if*, someone really wants to substitute systemd-login
+ * in an environment managed by systemd with elogind, we might have to
+ * add such a support. */
+ unified_cache = CGROUP_UNIFIED_NONE;
+#endif // 0
+
+ return 0;
+}
+
+int cg_unified_controller(const char *controller) {
+ int r;
+
+ r = cg_unified_update();
+ if (r < 0)
+ return r;
+
+ if (unified_cache == CGROUP_UNIFIED_NONE)
+ return false;
+
+ if (unified_cache >= CGROUP_UNIFIED_ALL)
+ return true;
+
+ return streq_ptr(controller, SYSTEMD_CGROUP_CONTROLLER);
+}
+
+int cg_all_unified(void) {
+ int r;
+
+ r = cg_unified_update();
+ if (r < 0)
+ return r;
- return unified_cache;
+ return unified_cache >= CGROUP_UNIFIED_ALL;
}
-void cg_unified_flush(void) {
- unified_cache = -1;
+int cg_hybrid_unified(void) {
+ int r;
+
+ r = cg_unified_update();
+ if (r < 0)
+ return r;
+
+ return unified_cache == CGROUP_UNIFIED_SYSTEMD && !unified_systemd_v232;
}
+int cg_unified_flush(void) {
+ unified_cache = CGROUP_UNIFIED_UNKNOWN;
+
+ return cg_unified_update();
+}
+
+#if 0 /// UNNEEDED by elogind
int cg_enable_everywhere(CGroupMask supported, CGroupMask mask, const char *p) {
_cleanup_free_ char *fs = NULL;
CGroupController c;
- int r, unified;
+ int r;
assert(p);
if (supported == 0)
return 0;
- unified = cg_unified();
- if (unified < 0)
- return unified;
- if (!unified) /* on the legacy hiearchy there's no joining of controllers defined */
+ r = cg_all_unified();
+ if (r < 0)
+ return r;
+ if (r == 0) /* on the legacy hiearchy there's no joining of controllers defined */
return 0;
- r = cg_get_path(ELOGIND_CGROUP_CONTROLLER, p, "cgroup.subtree_control", &fs);
+ r = cg_get_path(SYSTEMD_CGROUP_CONTROLLER, p, "cgroup.subtree_control", &fs);
if (r < 0)
return r;
@@ -2190,46 +2584,182 @@ int cg_enable_everywhere(CGroupMask supported, CGroupMask mask, const char *p) {
bool cg_is_unified_wanted(void) {
static thread_local int wanted = -1;
- int r, unified;
+ int r;
+ bool b;
+ const bool is_default = DEFAULT_HIERARCHY == CGROUP_UNIFIED_ALL;
+
+ /* If we have a cached value, return that. */
+ if (wanted >= 0)
+ return wanted;
/* If the hierarchy is already mounted, then follow whatever
* was chosen for it. */
- unified = cg_unified();
- if (unified >= 0)
- return unified;
+ if (cg_unified_flush() >= 0)
+ return (wanted = unified_cache >= CGROUP_UNIFIED_ALL);
+
+ /* Otherwise, let's see what the kernel command line has to say.
+ * Since checking is expensive, cache a non-error result. */
+ r = proc_cmdline_get_bool("systemd.unified_cgroup_hierarchy", &b);
+
+ return (wanted = r > 0 ? b : is_default);
+}
+
+bool cg_is_legacy_wanted(void) {
+ static thread_local int wanted = -1;
- /* Otherwise, let's see what the kernel command line has to
- * say. Since checking that is expensive, let's cache the
- * result. */
+ /* If we have a cached value, return that. */
if (wanted >= 0)
return wanted;
- r = get_proc_cmdline_key("systemd.unified_cgroup_hierarchy", NULL);
- if (r > 0)
- return (wanted = true);
- else {
- _cleanup_free_ char *value = NULL;
+ /* Check if we have cgroups2 already mounted. */
+ if (cg_unified_flush() >= 0 &&
+ unified_cache == CGROUP_UNIFIED_ALL)
+ return (wanted = false);
- r = get_proc_cmdline_key("systemd.unified_cgroup_hierarchy=", &value);
- if (r < 0)
- return false;
- if (r == 0)
- return (wanted = false);
+ /* Otherwise, assume that at least partial legacy is wanted,
+ * since cgroups2 should already be mounted at this point. */
+ return (wanted = true);
+}
+
+bool cg_is_hybrid_wanted(void) {
+ static thread_local int wanted = -1;
+ int r;
+ bool b;
+ const bool is_default = DEFAULT_HIERARCHY >= CGROUP_UNIFIED_SYSTEMD;
+ /* We default to true if the default is "hybrid", obviously,
+ * but also when the default is "unified", because if we get
+ * called, it means that unified hierarchy was not mounted. */
- return (wanted = parse_boolean(value) > 0);
+ /* If we have a cached value, return that. */
+ if (wanted >= 0)
+ return wanted;
+
+ /* If the hierarchy is already mounted, then follow whatever
+ * was chosen for it. */
+ if (cg_unified_flush() >= 0 &&
+ unified_cache == CGROUP_UNIFIED_ALL)
+ return (wanted = false);
+
+ /* Otherwise, let's see what the kernel command line has to say.
+ * Since checking is expensive, cache a non-error result. */
+ r = proc_cmdline_get_bool("systemd.legacy_systemd_cgroup_controller", &b);
+
+ /* The meaning of the kernel option is reversed wrt. to the return value
+ * of this function, hence the negation. */
+ return (wanted = r > 0 ? !b : is_default);
+}
+#else
+bool cg_is_unified_wanted(void) {
+ return false;
+}
+bool cg_is_legacy_wanted(void) {
+ return true;
+}
+bool cg_is_hybrid_wanted(void) {
+ return false;
+}
+#endif // 0
+
+#if 0 /// UNNEEDED by elogind
+int cg_weight_parse(const char *s, uint64_t *ret) {
+ uint64_t u;
+ int r;
+
+ if (isempty(s)) {
+ *ret = CGROUP_WEIGHT_INVALID;
+ return 0;
}
+
+ r = safe_atou64(s, &u);
+ if (r < 0)
+ return r;
+
+ if (u < CGROUP_WEIGHT_MIN || u > CGROUP_WEIGHT_MAX)
+ return -ERANGE;
+
+ *ret = u;
+ return 0;
}
-bool cg_is_legacy_wanted(void) {
- return !cg_is_unified_wanted();
+const uint64_t cgroup_io_limit_defaults[_CGROUP_IO_LIMIT_TYPE_MAX] = {
+ [CGROUP_IO_RBPS_MAX] = CGROUP_LIMIT_MAX,
+ [CGROUP_IO_WBPS_MAX] = CGROUP_LIMIT_MAX,
+ [CGROUP_IO_RIOPS_MAX] = CGROUP_LIMIT_MAX,
+ [CGROUP_IO_WIOPS_MAX] = CGROUP_LIMIT_MAX,
+};
+
+static const char* const cgroup_io_limit_type_table[_CGROUP_IO_LIMIT_TYPE_MAX] = {
+ [CGROUP_IO_RBPS_MAX] = "IOReadBandwidthMax",
+ [CGROUP_IO_WBPS_MAX] = "IOWriteBandwidthMax",
+ [CGROUP_IO_RIOPS_MAX] = "IOReadIOPSMax",
+ [CGROUP_IO_WIOPS_MAX] = "IOWriteIOPSMax",
+};
+
+DEFINE_STRING_TABLE_LOOKUP(cgroup_io_limit_type, CGroupIOLimitType);
+
+int cg_cpu_shares_parse(const char *s, uint64_t *ret) {
+ uint64_t u;
+ int r;
+
+ if (isempty(s)) {
+ *ret = CGROUP_CPU_SHARES_INVALID;
+ return 0;
+ }
+
+ r = safe_atou64(s, &u);
+ if (r < 0)
+ return r;
+
+ if (u < CGROUP_CPU_SHARES_MIN || u > CGROUP_CPU_SHARES_MAX)
+ return -ERANGE;
+
+ *ret = u;
+ return 0;
+}
+
+int cg_blkio_weight_parse(const char *s, uint64_t *ret) {
+ uint64_t u;
+ int r;
+
+ if (isempty(s)) {
+ *ret = CGROUP_BLKIO_WEIGHT_INVALID;
+ return 0;
+ }
+
+ r = safe_atou64(s, &u);
+ if (r < 0)
+ return r;
+
+ if (u < CGROUP_BLKIO_WEIGHT_MIN || u > CGROUP_BLKIO_WEIGHT_MAX)
+ return -ERANGE;
+
+ *ret = u;
+ return 0;
+}
+#endif // 0
+
+bool is_cgroup_fs(const struct statfs *s) {
+ return is_fs_type(s, CGROUP_SUPER_MAGIC) ||
+ is_fs_type(s, CGROUP2_SUPER_MAGIC);
+}
+
+bool fd_is_cgroup_fs(int fd) {
+ struct statfs s;
+
+ if (fstatfs(fd, &s) < 0)
+ return -errno;
+
+ return is_cgroup_fs(&s);
}
static const char *cgroup_controller_table[_CGROUP_CONTROLLER_MAX] = {
[CGROUP_CONTROLLER_CPU] = "cpu",
[CGROUP_CONTROLLER_CPUACCT] = "cpuacct",
+ [CGROUP_CONTROLLER_IO] = "io",
[CGROUP_CONTROLLER_BLKIO] = "blkio",
[CGROUP_CONTROLLER_MEMORY] = "memory",
- [CGROUP_CONTROLLER_DEVICE] = "devices",
+ [CGROUP_CONTROLLER_DEVICES] = "devices",
+ [CGROUP_CONTROLLER_PIDS] = "pids",
};
DEFINE_STRING_TABLE_LOOKUP(cgroup_controller, CGroupController);