+/* SPDX-License-Identifier: LGPL-2.1+ */
/***
- This file is part of systemd.
-
Copyright 2010 Lennart Poettering
-
- systemd is free software; you can redistribute it and/or modify it
- under the terms of the GNU Lesser General Public License as published by
- the Free Software Foundation; either version 2.1 of the License, or
- (at your option) any later version.
-
- systemd is distributed in the hope that it will be useful, but
- WITHOUT ANY WARRANTY; without even the implied warranty of
- MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
- Lesser General Public License for more details.
-
- You should have received a copy of the GNU Lesser General Public License
- along with systemd; If not, see <http://www.gnu.org/licenses/>.
***/
#include <dirent.h>
//#include <limits.h>
#include <signal.h>
//#include <stddef.h>
+#include <stdio_ext.h>
#include <stdlib.h>
#include <string.h>
#include <sys/stat.h>
//#include <sys/statfs.h>
#include <sys/types.h>
+#include <sys/xattr.h>
#include <unistd.h>
#include "alloc-util.h"
#include "extract-word.h"
#include "fd-util.h"
#include "fileio.h"
-#include "formats-util.h"
+#include "format-util.h"
#include "fs-util.h"
//#include "log.h"
#include "login-util.h"
#include "stdio-util.h"
#include "string-table.h"
#include "string-util.h"
+#include "strv.h"
#include "unit-name.h"
#include "user-util.h"
return 1;
}
-int cg_read_event(const char *controller, const char *path, const char *event,
- char **val)
-{
+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;
return enabled;
}
-#endif //0
+#endif // 0
int cg_enumerate_subgroups(const char *controller, const char *path, DIR **_d) {
_cleanup_free_ char *fs = NULL;
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);
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;
}
return -ENOMEM;
}
- my_pid = getpid();
+ my_pid = getpid_cached();
do {
_cleanup_fclose_ FILE *f = NULL;
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;
if (flags & CGROUP_REMOVE) {
r = cg_rmdir(controller, path);
- if (r < 0 && ret >= 0 && r != -ENOENT && r != -EBUSY)
+ if (r < 0 && ret >= 0 && !IN_SET(r, -ENOENT, -EBUSY))
return r;
}
if (!s)
return -ENOMEM;
- my_pid = getpid();
+ my_pid = getpid_cached();
log_debug_elogind("Migrating \"%s\"/\"%s\" to \"%s\"/\"%s\" (%s)",
cfrom, pfrom, cto, pto,
* exist in the root cgroup, we only check for
* them there. */
if (cfrom &&
- (isempty(pfrom) || path_equal(pfrom, "/")) &&
+ empty_or_root(pfrom) &&
is_kernel_thread(pid) > 0)
continue;
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;
if (flags & CGROUP_REMOVE) {
r = cg_rmdir(cfrom, pfrom);
- if (r < 0 && ret >= 0 && r != -ENOENT && r != -EBUSY)
+ if (r < 0 && ret >= 0 && !IN_SET(r, -ENOENT, -EBUSY))
return r;
}
* 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;
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);
+ t = strjoin("/sys/fs/cgroup/", dn, "/", path);
else
- t = strjoin("/sys/fs/cgroup/", dn, "/", path, "/", suffix, NULL);
+ t = strjoin("/sys/fs/cgroup/", dn, "/", path, "/", suffix);
if (!t)
return -ENOMEM;
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;
}
int cg_get_path(const char *controller, const char *path, const char *suffix, char **fs) {
- int unified, r;
+ int r;
assert(fs);
else if (!path)
t = strdup(suffix);
else
- t = strjoin(path, "/", suffix, NULL);
+ t = strjoin(path, "/", suffix);
if (!t)
return -ENOMEM;
- *fs = path_kill_slashes(t);
+ *fs = path_simplify(t, false);
return 0;
}
if (!cg_controller_is_valid(controller))
return -EINVAL;
- unified = cg_all_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;
- path_kill_slashes(*fs);
+ path_simplify(*fs, false);
return 0;
}
static int controller_is_accessible(const char *controller) {
- int unified;
+ int r;
assert(controller);
if (!cg_controller_is_valid(controller))
return -EINVAL;
- unified = cg_all_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. */
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);
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;
}
+/* Create a cgroup in the hierarchy of controller.
+ * Returns 0 if the group already existed, 1 on success, negative otherwise.
+ */
int cg_create(const char *controller, const char *path) {
_cleanup_free_ char *fs = NULL;
int r;
if (r < 0)
return r;
- if (mkdir(fs, 0755) < 0) {
+ r = mkdir_errno_wrapper(fs, 0755);
+ if (r == -EEXIST)
+ return 0;
+ if (r < 0)
+ return r;
- if (errno == EEXIST)
- return 0;
+ r = cg_hybrid_unified();
+ if (r < 0)
+ return r;
- return -errno;
+ 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;
return r;
if (pid == 0)
- pid = getpid();
+ pid = getpid_cached();
xsprintf(c, PID_FMT "\n", pid);
- return write_string_file(fs, c, 0);
+ r = write_string_file(fs, c, 0);
+ if (r < 0)
+ return r;
+
+ r = cg_hybrid_unified();
+ if (r < 0)
+ return r;
+
+ 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) {
}
#if 0 /// UNNEEDED by elogind
-int cg_set_group_access(
+int cg_set_access(
const char *controller,
const char *path,
- mode_t mode,
uid_t uid,
gid_t gid) {
+ struct Attribute {
+ const char *name;
+ bool fatal;
+ };
+
+ /* cgroupsv1, aka legacy/non-unified */
+ static const struct Attribute legacy_attributes[] = {
+ { "cgroup.procs", true },
+ { "tasks", false },
+ { "cgroup.clone_children", false },
+ {},
+ };
+
+ /* cgroupsv2, aka unified */
+ static const struct Attribute unified_attributes[] = {
+ { "cgroup.procs", true },
+ { "cgroup.subtree_control", true },
+ { "cgroup.threads", false },
+ {},
+ };
+
+ static const struct Attribute* const attributes[] = {
+ [false] = legacy_attributes,
+ [true] = unified_attributes,
+ };
+
_cleanup_free_ char *fs = NULL;
- int r;
+ const struct Attribute *i;
+ int r, unified;
+
+ assert(path);
- if (mode == MODE_INVALID && uid == UID_INVALID && gid == GID_INVALID)
+ if (uid == UID_INVALID && gid == GID_INVALID)
return 0;
- if (mode != MODE_INVALID)
- mode &= 0777;
+ unified = cg_unified_controller(controller);
+ if (unified < 0)
+ return unified;
+ /* Configure access to the cgroup itself */
r = cg_get_path(controller, path, NULL, &fs);
if (r < 0)
return r;
- return chmod_and_chown(fs, mode, uid, gid);
-}
+ r = chmod_and_chown(fs, 0755, uid, gid);
+ if (r < 0)
+ return r;
-int cg_set_task_access(
- const char *controller,
- const char *path,
- mode_t mode,
- uid_t uid,
- gid_t gid) {
+ /* Configure access to the cgroup's attributes */
+ for (i = attributes[unified]; i->name; i++) {
+ fs = mfree(fs);
- _cleanup_free_ char *fs = NULL, *procs = NULL;
- int r, unified;
+ r = cg_get_path(controller, path, i->name, &fs);
+ if (r < 0)
+ return r;
- assert(path);
+ r = chmod_and_chown(fs, 0644, uid, gid);
+ if (r < 0) {
+ if (i->fatal)
+ return r;
- if (mode == MODE_INVALID && uid == UID_INVALID && gid == GID_INVALID)
- return 0;
+ log_debug_errno(r, "Failed to set access on cgroup %s, ignoring: %m", fs);
+ }
+ }
- if (mode != MODE_INVALID)
- mode &= 0666;
+ if (streq(controller, SYSTEMD_CGROUP_CONTROLLER)) {
+ r = cg_hybrid_unified();
+ if (r < 0)
+ return r;
+ if (r > 0) {
+ /* Always propagate access mode from unified to legacy controller */
+ r = cg_set_access(SYSTEMD_CGROUP_CONTROLLER_LEGACY, path, uid, gid);
+ if (r < 0)
+ log_debug_errno(r, "Failed to set access on compatibility elogind cgroup %s, ignoring: %m", path);
+ }
+ }
- r = cg_get_path(controller, path, "cgroup.procs", &fs);
+ 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;
- r = chmod_and_chown(fs, mode, uid, gid);
+ 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;
- unified = cg_all_unified();
- if (unified < 0)
- return unified;
- if (unified)
- return 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);
+ n = getxattr(fs, name, value, size);
+ if (n < 0)
+ return -errno;
- return 0;
+ 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;
+#if 0 /// At elogind we do not want that (false alarm) "maybe uninitialized" warning
+ const char *fs, *controller_str;
+#else
+ const char *fs, *controller_str = NULL;
+#endif // 0
size_t cs = 0;
int unified;
assert(path);
assert(pid >= 0);
- unified = cg_all_unified();
+ if (controller) {
+ if (!cg_controller_is_valid(controller))
+ return -EINVAL;
+ } else
+ controller = SYSTEMD_CGROUP_CONTROLLER;
+
+ unified = cg_unified_controller(controller);
if (unified < 0)
return unified;
if (unified == 0) {
- if (controller) {
- if (!cg_controller_is_valid(controller))
- return -EINVAL;
- } else
- controller = SYSTEMD_CGROUP_CONTROLLER;
+ if (streq(controller, SYSTEMD_CGROUP_CONTROLLER))
+ controller_str = SYSTEMD_CGROUP_CONTROLLER_LEGACY;
+ else
+ controller_str = controller;
- cs = strlen(controller);
+ cs = strlen(controller_str);
}
fs = procfs_file_alloca(pid, "cgroup");
if (!f)
return errno == ENOENT ? -ESRCH : -errno;
+ (void) __fsetlocking(f, FSETLOCKING_BYCALLER);
+
FOREACH_LINE(line, f, return -errno) {
char *e, *p;
continue;
*e = 0;
- FOREACH_WORD_SEPARATOR(word, k, l, ",", state) {
- if (k == cs && memcmp(word, controller, cs) == 0) {
+ FOREACH_WORD_SEPARATOR(word, k, l, ",", state)
+ if (k == cs && memcmp(word, controller_str, cs) == 0) {
found = true;
break;
}
- }
-
if (!found)
continue;
}
if (!p)
return -ENOMEM;
+ /* Truncate suffix indicating the process is a zombie */
+ e = endswith(p, " (deleted)");
+ if (e)
+ *e = 0;
+
*path = p;
return 0;
}
return -ENODATA;
}
+#if 0 /// UNNEEDED by elogind
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_all_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);
int cg_uninstall_release_agent(const char *controller) {
_cleanup_free_ char *fs = NULL;
- int r, unified;
+ int r;
- unified = cg_all_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);
return 0;
}
+#endif // 0
int cg_is_empty(const char *controller, const char *path) {
_cleanup_fclose_ FILE *f = NULL;
}
int cg_is_empty_recursive(const char *controller, const char *path) {
- int unified, r;
+ int r;
assert(path);
/* The root cgroup is always populated */
- if (controller && (isempty(path) || path_equal(path, "/")))
+ if (controller && empty_or_root(path))
return false;
- unified = cg_all_unified();
- if (unified < 0)
- return unified;
-
- if (unified > 0) {
+ r = cg_unified_controller(controller);
+ if (r < 0)
+ return r;
+ if (r > 0) {
_cleanup_free_ char *t = NULL;
/* On the unified hierarchy we can check empty state
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;
assert(spec);
if (*spec == '/') {
- if (!path_is_safe(spec))
+ if (!path_is_normalized(spec))
return -EINVAL;
if (path) {
if (!t)
return -ENOMEM;
- *path = path_kill_slashes(t);
+ *path = path_simplify(t, false);
}
if (controller)
return -ENOMEM;
}
- if (!path_is_safe(u) ||
+ if (!path_is_normalized(u) ||
!path_is_absolute(u)) {
free(t);
free(u);
return -EINVAL;
}
- path_kill_slashes(u);
+ path_simplify(u, false);
}
if (controller)
if (!t)
return -ENOMEM;
- *result = path_kill_slashes(t);
+ *result = path_simplify(t, false);
return 0;
}
}
int cg_get_root_path(char **path) {
-#if 0 /// elogind does not support systemd scopes and slices
char *p, *e;
int r;
if (r < 0)
return r;
+#if 0 /// elogind does not support systemd scopes and slices
e = endswith(p, "/" SPECIAL_INIT_SCOPE);
if (!e)
e = endswith(p, "/" SPECIAL_SYSTEM_SLICE); /* legacy */
if (!e)
e = endswith(p, "/system"); /* even more legacy */
+#else
+ e = endswith(p, "/elogind");
+#endif // 0
if (e)
*e = 0;
*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) {
}
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;
if (r < 0)
return r;
- if (c == raw) {
- *cgroup = raw;
- raw = NULL;
- } else {
+ if (c == raw)
+ *cgroup = TAKE_PTR(raw);
+ else {
char *n;
n = strdup(c);
return 0;
}
-#if 0 /// UNNEEDED by elogind
int cg_path_decode_unit(const char *cgroup, char **unit) {
char *c, *s;
size_t n;
assert(cgroup);
assert(unit);
+#if 0 /// elogind has a different naming: <controller>:/<session id>. So prefix is always len < 3
n = strcspn(cgroup, "/");
if (n < 3)
return -ENXIO;
+#else
+ n = strspn(cgroup, "/") + 1;
+#endif // 0
c = strndupa(cgroup, n);
c = cg_unescape(c);
+#if 0 /// elogind session ids are never valid unit names.
if (!unit_name_is_valid(c, UNIT_NAME_PLAIN|UNIT_NAME_INSTANCE))
return -ENXIO;
+#endif // 0
s = strdup(c);
if (!s)
if (!p)
return false;
- if (n < strlen("x.slice"))
+ if (n < STRLEN("x.slice"))
return false;
if (memcmp(p + n - 6, ".slice", 6) == 0) {
return cg_path_get_unit(cgroup, unit);
}
+#if 0 /// UNNEEDED by elogind
/**
* Skip session-*.scope, but require it to be there.
*/
p += strspn(p, "/");
n = strcspn(p, "/");
- if (n < strlen("session-x.scope"))
+ if (n < STRLEN("session-x.scope"))
return NULL;
if (memcmp(p, "session-", 8) == 0 && memcmp(p + n - 6, ".scope", 6) == 0) {
p += strspn(p, "/");
n = strcspn(p, "/");
- if (n < strlen("user@x.service"))
+ if (n < STRLEN("user@x.service"))
return NULL;
if (memcmp(p, "user@", 5) == 0 && memcmp(p + n - 8, ".service", 8) == 0) {
return cg_path_get_session(cgroup, session);
}
-#if 0 /// UNNEEDED by elogind
int cg_path_get_owner_uid(const char *path, uid_t *uid) {
+#if 0 /// elogind needs one more value
_cleanup_free_ char *slice = NULL;
char *start, *end;
+#else
+ _cleanup_free_ char *slice = NULL, *p = NULL, *s = NULL;
+#endif // 0
int r;
assert(path);
if (r < 0)
return r;
+#if 0 /// elogind does not support systemd slices
start = startswith(slice, "user-");
if (!start)
return -ENXIO;
*end = 0;
if (parse_uid(start, uid) < 0)
return -ENXIO;
+#else
+ p = strappend("/run/systemd/sessions/", slice);
+
+ r = parse_env_file(p, NEWLINE, "UID", &s, NULL);
+ if (r == -ENOENT)
+ return -ENXIO;
+ if (r < 0)
+ return r;
+ if (isempty(s))
+ return -EIO;
+
+ if (parse_uid(s, uid) < 0)
+ return -ENXIO;
+#endif // 0
return 0;
}
assert(p);
assert(slice);
+#if 0 /// elogind does not support systemd slices
/* Finds the right-most slice unit from the beginning, but
* stops before we come to the first non-slice unit. */
e = p;
p += n;
}
+#else
+ /* In elogind, what is reported here, is the location of
+ * the session. This is derived from /proc/<self|PID>/cgroup.
+ * In there we look at the controller, which will look something
+ * like "1:name=openrc:/3".
+ * The last part gets extracted (and is now p), which is "/3" in
+ * this case. The three is the session id, and that can be mapped.
+ */
+ e = startswith(p, "/");
+
+ if (e)
+ *slice = strdup(e);
+ else
+ *slice = strdup(p);
+
+ return 0;
+#endif // 0
}
int cg_pid_get_slice(pid_t pid, char **slice) {
assert(slice);
r = cg_pid_get_path_shifted(pid, NULL, &cgroup);
+ log_debug_elogind("Found cgroup %s for pid %u (result %d)",
+ cgroup, pid, r);
if (r < 0)
return r;
}
int cg_path_get_user_slice(const char *p, char **slice) {
+#if 0 /// UNNEEDED by elogind
const char *t;
+#endif // 0
assert(p);
assert(slice);
+#if 0 /// nothing to skip in elogind
t = skip_user_prefix(p);
if (!t)
return -ENXIO;
+#endif // 0
+#if 0 /// UNNEEDED by elogind
/* And now it looks pretty much the same as for a system
* slice, so let's just use the same parser from here on. */
return cg_path_get_slice(t, slice);
+#else
+ /* In elogind there is nothing to skip, we can use the path
+ * directly. Generally speaking this is always a session id
+ * to user mapping. */
+ return cg_path_get_slice(p, slice);
+#endif // 0
}
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;
/* The return value of this function (unlike cg_unescape())
* needs free()! */
- if (p[0] == 0 ||
- p[0] == '_' ||
- p[0] == '.' ||
+ if (IN_SET(p[0], 0, '_', '.') ||
streq(p, "notify_on_release") ||
streq(p, "release_agent") ||
streq(p, "tasks") ||
if (!p)
return false;
+ if (streq(p, SYSTEMD_CGROUP_CONTROLLER))
+ return true;
+
s = startswith(p, "name=");
if (s)
p = s;
- if (*p == 0 || *p == '_')
+ if (IN_SET(*p, 0, '_'))
return false;
for (t = p; *t; t++)
_cleanup_free_ char *escaped = NULL;
char n[dash - p + sizeof(".slice")];
+#if HAS_FEATURE_MEMORY_SANITIZER
+ /* msan doesn't instrument stpncpy, so it thinks
+ * n is later used unitialized:
+ * https://github.com/google/sanitizers/issues/926
+ */
+ zero(n);
+#endif
+
/* Don't allow trailing or double dashes */
- if (dash[1] == 0 || dash[1] == '-')
+ if (IN_SET(dash[1], 0, '-'))
return -EINVAL;
strcpy(stpncpy(n, p, dash - p), ".slice");
if (!strextend(&s, e, NULL))
return -ENOMEM;
- *ret = s;
- s = NULL;
+ *ret = TAKE_PTR(s);
return 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;
+int cg_get_keyed_attribute(
+ const char *controller,
+ const char *path,
+ const char *attribute,
+ char **keys,
+ char **ret_values) {
+
+ _cleanup_free_ char *filename = NULL, *contents = NULL;
+ const char *p;
+ size_t n, i, n_done = 0;
+ char **v;
+ int r;
- for (i = 0; keys[i]; i++)
- values[i] = NULL;
+ /* Reads one or more fields of a cgroupsv2 keyed attribute file. The 'keys' parameter should be an strv with
+ * all keys to retrieve. The 'ret_values' parameter should be passed as string size with the same number of
+ * entries as 'keys'. On success each entry will be set to the value of the matching key.
+ *
+ * If the attribute file doesn't exist at all returns ENOENT, if any key is not found returns ENXIO. */
r = cg_get_path(controller, path, attribute, &filename);
if (r < 0)
return r;
- r = read_full_file(filename, &content, NULL);
+ r = read_full_file(filename, &contents, NULL);
if (r < 0)
return r;
- p = content;
- while ((line = strsep(&p, "\n"))) {
- char *key;
+ n = strv_length(keys);
+ if (n == 0) /* No keys to retrieve? That's easy, we are done then */
+ return 0;
- key = strsep(&line, " ");
+ /* Let's build this up in a temporary array for now in order not to clobber the return parameter on failure */
+ v = newa0(char*, n);
- for (i = 0; keys[i]; i++) {
- if (streq(key, keys[i])) {
- values[i] = strdup(line);
- break;
+ for (p = contents; *p;) {
+ const char *w = NULL;
+
+ for (i = 0; i < n; i++)
+ if (!v[i]) {
+ w = first_word(p, keys[i]);
+ if (w)
+ break;
}
- }
- }
- for (i = 0; keys[i]; i++) {
- if (!values[i]) {
- for (i = 0; keys[i]; i++) {
- free(values[i]);
- values[i] = NULL;
+ if (w) {
+ size_t l;
+
+ l = strcspn(w, NEWLINE);
+ v[i] = strndup(w, l);
+ if (!v[i]) {
+ r = -ENOMEM;
+ goto fail;
}
- return -ENOENT;
- }
+
+ n_done++;
+ if (n_done >= n)
+ goto done;
+
+ p = w + l;
+ } else
+ p += strcspn(p, NEWLINE);
+
+ p += strspn(p, NEWLINE);
}
+ r = -ENXIO;
+
+fail:
+ for (i = 0; i < n; i++)
+ free(v[i]);
+
+ return r;
+
+done:
+ memcpy(ret_values, v, sizeof(char*) * n);
return 0;
+
}
int cg_create_everywhere(CGroupMask supported, CGroupMask mask, const char *path) {
CGroupController c;
- int r, unified;
+ bool created;
+ 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 */
+ * in all others.
+ *
+ * Returns 0 if the group already existed in the systemd hierarchy,
+ * 1 on success, negative otherwise.
+ */
/* First create the cgroup in our own hierarchy. */
r = cg_create(SYSTEMD_CGROUP_CONTROLLER, path);
if (r < 0)
return r;
+ created = !!r;
/* If we are in the unified hierarchy, we are done now */
- unified = cg_all_unified();
- if (unified < 0)
- return unified;
- if (unified > 0)
- return 0;
+ r = cg_all_unified();
+ if (r < 0)
+ return r;
+ if (r > 0)
+ return created;
/* Otherwise, do the same in the other hierarchies */
for (c = 0; c < _CGROUP_CONTROLLER_MAX; c++) {
(void) cg_trim(n, path, true);
}
- return 0;
+ return created;
}
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(SYSTEMD_CGROUP_CONTROLLER, path, pid);
if (r < 0)
return r;
- unified = cg_all_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++) {
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(SYSTEMD_CGROUP_CONTROLLER, from, SYSTEMD_CGROUP_CONTROLLER, to, CGROUP_REMOVE);
return r;
}
- unified = cg_all_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++) {
int cg_trim_everywhere(CGroupMask supported, const char *path, bool delete_root) {
CGroupController c;
- int r, unified;
+ int r, q;
r = cg_trim(SYSTEMD_CGROUP_CONTROLLER, path, delete_root);
if (r < 0)
return r;
- unified = cg_all_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++) {
}
#endif // 0
+int cg_mask_to_string(CGroupMask mask, char **ret) {
+ _cleanup_free_ char *s = NULL;
+ size_t n = 0, allocated = 0;
+ bool space = false;
+ CGroupController c;
+
+ assert(ret);
+
+ if (mask == 0) {
+ *ret = NULL;
+ return 0;
+ }
+
+ for (c = 0; c < _CGROUP_CONTROLLER_MAX; c++) {
+ const char *k;
+ size_t l;
+
+ if (!(mask & CGROUP_CONTROLLER_TO_MASK(c)))
+ continue;
+
+ k = cgroup_controller_to_string(c);
+ l = strlen(k);
+
+ if (!GREEDY_REALLOC(s, allocated, n + space + l + 1))
+ return -ENOMEM;
+
+ if (space)
+ s[n] = ' ';
+ memcpy(s + n + space, k, l);
+ n += space + l;
+
+ space = true;
+ }
+
+ assert(s);
+
+ s[n] = 0;
+ *ret = TAKE_PTR(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_all_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
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 support the cpu, memory, io and pids
* controller in the unified hierarchy, mask
}
#if 0 /// UNNEEDED by elogind
-int cg_kernel_controllers(Set *controllers) {
+int cg_kernel_controllers(Set **ret) {
+ _cleanup_set_free_free_ Set *controllers = NULL;
_cleanup_fclose_ FILE *f = NULL;
- char buf[LINE_MAX];
int r;
- assert(controllers);
+ assert(ret);
/* Determines the full list of kernel-known controllers. Might
* include controllers we don't actually support, arbitrary
* named hierarchies and controllers that aren't currently
* accessible (because not mounted). */
+ controllers = set_new(&string_hash_ops);
+ if (!controllers)
+ return -ENOMEM;
+
f = fopen("/proc/cgroups", "re");
if (!f) {
- if (errno == ENOENT)
+ if (errno == ENOENT) {
+ *ret = NULL;
return 0;
+ }
+
return -errno;
}
+ (void) __fsetlocking(f, FSETLOCKING_BYCALLER);
+
/* Ignore the header line */
- (void) fgets(buf, sizeof(buf), f);
+ (void) read_line(f, (size_t) -1, NULL);
for (;;) {
char *controller;
return r;
}
+ *ret = TAKE_PTR(controllers);
+
return 0;
}
#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_all_unified(void) {
struct statfs fs;
/* Checks if we support the unified hierarchy. Returns an
* 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 0 /// UNNEEDED by elogind
- if (F_TYPE_EQUAL(fs.f_type, CGROUP2_SUPER_MAGIC))
+ return log_debug_errno(errno, "statfs(\"/sys/fs/cgroup/\") failed: %m");
+
+ if (F_TYPE_EQUAL(fs.f_type, CGROUP2_SUPER_MAGIC)) {
+ log_debug("Found cgroup2 on /sys/fs/cgroup/, full unified hierarchy");
+ unified_cache = CGROUP_UNIFIED_ALL;
+#if 0 /// The handling of cgroups is a bit different with elogind
+ } else if (F_TYPE_EQUAL(fs.f_type, TMPFS_MAGIC)) {
+ log_debug("Found cgroup2 on /sys/fs/cgroup/unified, unified hierarchy for systemd controller");
#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. */
- if (F_TYPE_EQUAL(fs.f_type, TMPFS_MAGIC)) {
+ } else if (F_TYPE_EQUAL(fs.f_type, CGROUP_SUPER_MAGIC)
+ || F_TYPE_EQUAL(fs.f_type, TMPFS_MAGIC)) {
#endif // 0
- unified_cache = true;
- else if (F_TYPE_EQUAL(fs.f_type, TMPFS_MAGIC))
- unified_cache = false;
- else
+ 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 0 /// There is no sub-grouping within elogind
+ if (statfs("/sys/fs/cgroup/systemd/", &fs) < 0)
+ return log_debug_errno(errno, "statfs(\"/sys/fs/cgroup/systemd\" failed: %m");
+
+ if (F_TYPE_EQUAL(fs.f_type, CGROUP2_SUPER_MAGIC)) {
+ log_debug("Found cgroup2 on /sys/fs/cgroup/systemd, unified hierarchy for systemd controller (v232 variant)");
+ unified_cache = CGROUP_UNIFIED_SYSTEMD;
+ unified_systemd_v232 = true;
+ } else if (F_TYPE_EQUAL(fs.f_type, CGROUP_SUPER_MAGIC)) {
+ log_debug("Found cgroup on /sys/fs/cgroup/systemd, legacy hierarchy");
+ unified_cache = CGROUP_UNIFIED_NONE;
+ } else {
+ log_debug("Unexpected filesystem type %llx mounted on /sys/fs/cgroup/systemd, assuming legacy hierarchy",
+ (unsigned long long) fs.f_type);
+ unified_cache = CGROUP_UNIFIED_NONE;
+ }
+#else
+ unified_cache = CGROUP_UNIFIED_NONE;
+#endif // 0
+ }
+ } else {
+ log_debug("Unknown filesystem type %llx mounted on /sys/fs/cgroup.",
+ (unsigned long long) fs.f_type);
return -ENOMEDIUM;
+ }
- return unified_cache;
+ return 0;
}
-#if 0 /// UNNEEDED by elogind
-void cg_unified_flush(void) {
- unified_cache = -1;
+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;
+
+#if 0 /// only if elogind is the controller we can use cgroups2 in hybrid mode
+ return streq_ptr(controller, SYSTEMD_CGROUP_CONTROLLER);
+#else
+ return streq_ptr(controller, SYSTEMD_CGROUP_CONTROLLER_HYBRID);
+#endif // 0
}
+int cg_all_unified(void) {
+ int r;
+
+ r = cg_unified_update();
+ if (r < 0)
+ return r;
+
+ return unified_cache >= CGROUP_UNIFIED_ALL;
+}
+
+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_fclose_ FILE *f = NULL;
_cleanup_free_ char *fs = NULL;
CGroupController c;
- int r, unified;
+ int r;
assert(p);
if (supported == 0)
return 0;
- unified = cg_all_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(SYSTEMD_CGROUP_CONTROLLER, p, "cgroup.subtree_control", &fs);
s[0] = mask & bit ? '+' : '-';
strcpy(s + 1, n);
- r = write_string_file(fs, s, 0);
- if (r < 0)
+ if (!f) {
+ f = fopen(fs, "we");
+ if (!f) {
+ log_debug_errno(errno, "Failed to open cgroup.subtree_control file of %s: %m", p);
+ break;
+ }
+ }
+
+ r = write_string_stream(f, s, 0);
+ if (r < 0) {
log_debug_errno(r, "Failed to enable controller %s for %s (%s): %m", n, p, fs);
+ clearerr(f);
+ }
}
}
return 0;
}
+#endif // 0
bool cg_is_unified_wanted(void) {
static thread_local int wanted = -1;
- int r, unified;
-
- /* If the hierarchy is already mounted, then follow whatever
- * was chosen for it. */
- unified = cg_all_unified();
- if (unified >= 0)
- return unified;
+#if 0 /// UNNEEDED by elogind
+ int r;
+ bool b;
+#endif // 0
+ const bool is_default = DEFAULT_HIERARCHY == CGROUP_UNIFIED_ALL;
- /* 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;
+ /* If the hierarchy is already mounted, then follow whatever
+ * was chosen for it. */
+ if (cg_unified_flush() >= 0)
+ return (wanted = unified_cache >= CGROUP_UNIFIED_ALL);
- r = get_proc_cmdline_key("systemd.unified_cgroup_hierarchy=", &value);
- if (r < 0)
- return false;
- if (r == 0)
- return (wanted = false);
+#if 0 /// elogind is not init and has no business with kernel command line
+ /* 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 = parse_boolean(value) > 0);
- }
+ return (wanted = r > 0 ? b : is_default);
+#else
+ return is_default;
+#endif // 0
}
bool cg_is_legacy_wanted(void) {
- return !cg_is_unified_wanted();
-}
-#else
-bool cg_is_legacy_wanted(void) {
- return true;
-
-bool cg_is_unified_systemd_controller_wanted(void) {
static thread_local int wanted = -1;
- int r, unified;
-
- /* If the unified hierarchy is requested in full, no need to
- * bother with this. */
- if (cg_is_unified_wanted())
- return 0;
-
- /* If the hierarchy is already mounted, then follow whatever
- * was chosen for it. */
- unified = cg_unified(SYSTEMD_CGROUP_CONTROLLER);
- if (unified >= 0)
- return unified;
- /* 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.legacy_systemd_cgroup_controller", NULL);
- if (r > 0) {
- wanted = false;
- } 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.legacy_systemd_cgroup_controller=", &value);
- if (r < 0)
- return true;
+ /* Otherwise, assume that at least partial legacy is wanted,
+ * since cgroups2 should already be mounted at this point. */
+ return (wanted = true);
+}
- if (r == 0)
- wanted = true;
- else
- wanted = parse_boolean(value) <= 0;
- }
+bool cg_is_hybrid_wanted(void) {
+ static thread_local int wanted = -1;
+#if 0 /// UNNEEDED by elogind
+ int r;
+ bool b;
+#endif // 0
+ 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;
-}
+ /* If we have a cached value, return that. */
+ if (wanted >= 0)
+ return wanted;
-bool cg_is_legacy_systemd_controller_wanted(void) {
- return cg_is_legacy_wanted() && !cg_is_unified_systemd_controller_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);
+
+#if 0 /// elogind is not init and has no business with kernel command line
+ /* 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
+ return is_default;
#endif // 0
+}
#if 0 /// UNNEEDED by elogind
int cg_weight_parse(const char *s, uint64_t *ret) {
}
#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",