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=315c449656aa073ce6aadad84e0506b82e246a6f;hp=b1063e9db9991819e40f8ab252edc5d5860ef5a7;hb=d93247127eb2e073a6d3b5bcc67bcc4048d674fe;hpb=21ecddbbb2c9e06d080b2526a870898a8b90f52a
diff --git a/src/basic/cgroup-util.c b/src/basic/cgroup-util.c
index b1063e9db..315c44965 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 +17,47 @@
along with systemd; If not, see .
***/
+#include
#include
-#include
+#include
+//#include
#include
-#include
+//#include
#include
-#include
+#include
#include
+//#include
#include
-#include
+#include
+#include
+#include "alloc-util.h"
#include "cgroup-util.h"
-#include "set.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 "strv.h"
#include "unit-name.h"
-#include "fileio.h"
-#include "special.h"
-#include "mkdir.h"
-#include "login-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 +93,7 @@ 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 +103,55 @@ 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;
@@ -113,14 +179,13 @@ int cg_read_subgroup(DIR *d, char **fn) {
assert(d);
assert(fn);
- FOREACH_DIRENT(de, d, return -errno) {
+ FOREACH_DIRENT_ALL(de, d, return -errno) {
char *b;
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 +211,30 @@ 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 +242,11 @@ 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 */
@@ -184,20 +274,23 @@ 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, LONG_TO_PTR(pid)) == LONG_TO_PTR(pid))
+ 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)
- kill(pid, SIGCONT);
+ if (flags & CGROUP_SIGCONT)
+ (void) kill(pid, SIGCONT);
if (ret == 0)
ret = 1;
@@ -205,7 +298,7 @@ int cg_kill(const char *controller, const char *path, int sig, bool sigcont, boo
done = false;
- r = set_put(s, LONG_TO_PTR(pid));
+ r = set_put(s, PID_TO_PTR(pid));
if (r < 0) {
if (ret >= 0)
return r;
@@ -230,10 +323,18 @@ 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 = 0;
+ int r, ret;
char *fn;
assert(path);
@@ -245,7 +346,7 @@ 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 +359,19 @@ 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);
- if (ret >= 0 && r != 0)
+ 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 +380,13 @@ 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 +403,10 @@ 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;
@@ -315,10 +425,18 @@ 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))
continue;
- if (set_get(s, LONG_TO_PTR(pid)) == LONG_TO_PTR(pid))
+ /* Ignore kernel threads. Since they can only
+ * exist in the root cgroup, we only check for
+ * them there. */
+ if (cfrom &&
+ (isempty(pfrom) || path_equal(pfrom, "/")) &&
+ is_kernel_thread(pid) > 0)
continue;
r = cg_attach(cto, pto, pid);
@@ -330,7 +448,7 @@ int cg_migrate(const char *cfrom, const char *pfrom, const char *cto, const char
done = false;
- r = set_put(s, LONG_TO_PTR(pid));
+ r = set_put(s, PID_TO_PTR(pid));
if (r < 0) {
if (ret >= 0)
return r;
@@ -355,8 +473,7 @@ 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;
@@ -367,7 +484,7 @@ 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) {
@@ -380,16 +497,12 @@ 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) {
- if (ret >= 0)
- return -ENOMEM;
-
- return ret;
- }
+ if (!p)
+ 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;
}
@@ -397,7 +510,7 @@ 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;
@@ -411,8 +524,7 @@ int cg_migrate_recursive_fallback(
const char *pfrom,
const char *cto,
const char *pto,
- bool ignore_self,
- bool rem) {
+ CGroupFlags flags) {
int r;
@@ -421,121 +533,187 @@ 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];
/* This didn't work? Then let's try all prefixes of the destination */
PATH_FOREACH_PREFIX(prefix, pto) {
- r = cg_migrate_recursive(cfrom, pfrom, cto, prefix, ignore_self, rem);
- if (r >= 0)
- break;
+ int q;
+
+ q = cg_migrate_recursive(cfrom, pfrom, cto, prefix, flags);
+ if (q >= 0)
+ return q;
}
}
- return 0;
+ return r;
}
-static const char *normalize_controller(const char *controller) {
+static const char *controller_to_dirname(const char *controller) {
+ const char *e;
assert(controller);
- if (startswith(controller, "name="))
- return controller + 5;
- else
- return controller;
+ /* Converts a controller name to the directory name below
+ * /sys/fs/cgroup/ we want to mount it to. Effectively, this
+ * 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;
}
-static int join_path(const char *controller, const char *path, const char *suffix, char **fs) {
+static int join_path_legacy(const char *controller, const char *path, const char *suffix, char **fs) {
+ const char *dn;
char *t = NULL;
- if (!isempty(controller)) {
- if (!isempty(path) && !isempty(suffix))
- t = strjoin("/sys/fs/cgroup/", controller, "/", path, "/", suffix, NULL);
- else if (!isempty(path))
- t = strjoin("/sys/fs/cgroup/", controller, "/", path, NULL);
- else if (!isempty(suffix))
- t = strjoin("/sys/fs/cgroup/", controller, "/", suffix, NULL);
- else
- t = strappend("/sys/fs/cgroup/", controller);
- } else {
- if (!isempty(path) && !isempty(suffix))
- t = strjoin(path, "/", suffix, NULL);
- else if (!isempty(path))
- t = strdup(path);
- else
- return -EINVAL;
- }
+ assert(fs);
+ assert(controller);
+ dn = controller_to_dirname(controller);
+
+ if (isempty(path) && isempty(suffix))
+ t = strappend("/sys/fs/cgroup/", dn);
+ else if (isempty(path))
+ t = strjoin("/sys/fs/cgroup/", dn, "/", suffix);
+ else if (isempty(suffix))
+ t = strjoin("/sys/fs/cgroup/", dn, "/", path);
+ else
+ t = strjoin("/sys/fs/cgroup/", dn, "/", path, "/", suffix);
if (!t)
return -ENOMEM;
- *fs = path_kill_slashes(t);
+ *fs = t;
return 0;
}
-int cg_get_path(const char *controller, const char *path, const char *suffix, char **fs) {
- const char *p;
- static thread_local bool good = false;
+static int join_path_unified(const char *path, const char *suffix, char **fs) {
+ char *t;
assert(fs);
- if (controller && !cg_controller_is_valid(controller))
- return -EINVAL;
+ if (isempty(path) && isempty(suffix))
+ t = strdup("/sys/fs/cgroup");
+ else if (isempty(path))
+ t = strappend("/sys/fs/cgroup/", suffix);
+ else if (isempty(suffix))
+ t = strappend("/sys/fs/cgroup/", path);
+ else
+ t = strjoin("/sys/fs/cgroup/", path, "/", suffix);
+ if (!t)
+ return -ENOMEM;
- if (_unlikely_(!good)) {
- int r;
+ *fs = t;
+ return 0;
+}
- r = path_is_mount_point("/sys/fs/cgroup", 0);
- if (r < 0)
- return r;
- if (r == 0)
- return -ENOENT;
+int cg_get_path(const char *controller, const char *path, const char *suffix, char **fs) {
+ int r;
- /* Cache this to save a few stat()s */
- good = true;
- }
+ assert(fs);
- p = controller ? normalize_controller(controller) : NULL;
+ if (!controller) {
+ char *t;
- return join_path(p, path, suffix, fs);
-}
+ /* If no controller is specified, we return the path
+ * *below* the controllers, without any prefix. */
-static int check_hierarchy(const char *p) {
- const char *cc;
+ if (!path && !suffix)
+ return -EINVAL;
- assert(p);
+ if (!suffix)
+ t = strdup(path);
+ else if (!path)
+ t = strdup(suffix);
+ else
+ t = strjoin(path, "/", suffix);
+ if (!t)
+ return -ENOMEM;
- if (!filename_is_valid(p))
+ *fs = path_kill_slashes(t);
return 0;
+ }
- /* Check if this controller actually really exists */
- cc = strjoina("/sys/fs/cgroup/", p);
- if (laccess(cc, F_OK) < 0)
- return -errno;
+ if (!cg_controller_is_valid(controller))
+ return -EINVAL;
+
+ 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);
return 0;
}
-int cg_get_path_and_check(const char *controller, const char *path, const char *suffix, char **fs) {
- const char *p;
+static int controller_is_accessible(const char *controller) {
int r;
- assert(fs);
+ assert(controller);
+
+ /* Checks whether a specific controller is accessible,
+ * i.e. its hierarchy mounted. In the unified hierarchy all
+ * controllers are considered accessible, except for the named
+ * hierarchies */
if (!cg_controller_is_valid(controller))
return -EINVAL;
- /* Normalize the controller syntax */
- p = normalize_controller(controller);
+ 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, SYSTEMD_CGROUP_CONTROLLER))
+ return 0;
+
+ if (startswith(controller, "name="))
+ return -EOPNOTSUPP;
+
+ } else {
+ const char *cc, *dn;
+
+ dn = controller_to_dirname(controller);
+ cc = strjoina("/sys/fs/cgroup/", dn);
- /* Check if this controller actually really exists */
- r = check_hierarchy(p);
+ if (laccess(cc, F_OK) < 0)
+ return -errno;
+ }
+
+ return 0;
+}
+
+int cg_get_path_and_check(const char *controller, const char *path, const char *suffix, char **fs) {
+ int r;
+
+ assert(controller);
+ assert(fs);
+
+ /* Check if the specified controller is actually accessible */
+ r = controller_is_accessible(controller);
if (r < 0)
return r;
- return join_path(p, path, suffix, fs);
+ return cg_get_path(controller, path, suffix, fs);
}
static int trim_cb(const char *path, const struct stat *sb, int typeflag, struct FTW *ftwbuf) {
@@ -549,13 +727,13 @@ static int trim_cb(const char *path, const struct stat *sb, int typeflag, struct
if (ftwbuf->level < 1)
return 0;
- rmdir(path);
+ (void) rmdir(path);
return 0;
}
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);
@@ -564,33 +742,31 @@ int cg_trim(const char *controller, const char *path, bool delete_root) {
return r;
errno = 0;
- if (nftw(fs, trim_cb, 64, FTW_DEPTH|FTW_MOUNT|FTW_PHYS) != 0)
- r = errno ? -errno : -EIO;
+ if (nftw(fs, trim_cb, 64, FTW_DEPTH|FTW_MOUNT|FTW_PHYS) != 0) {
+ if (errno == ENOENT)
+ r = 0;
+ else if (errno > 0)
+ r = -errno;
+ else
+ r = -EIO;
+ }
if (delete_root) {
if (rmdir(fs) < 0 && errno != ENOENT)
return -errno;
}
- return r;
-}
-
-/// UNNEDED by elogind
-#if 0
-int cg_delete(const char *controller, const char *path) {
- _cleanup_free_ char *parent = NULL;
- int r;
-
- assert(path);
-
- r = path_get_parent(path, &parent);
- if (r < 0)
- return r;
+ 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);
+ }
- r = cg_migrate_recursive(controller, path, controller, parent, false, true);
- return r == -ENOENT ? 0 : r;
+ return r;
}
-#endif // 0
int cg_create(const char *controller, const char *path) {
_cleanup_free_ char *fs = NULL;
@@ -612,11 +788,19 @@ 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;
@@ -633,7 +817,6 @@ 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;
@@ -650,9 +833,23 @@ 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;
+
+ 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 write_string_file_no_create(fs, c);
+ return 0;
}
int cg_attach_fallback(const char *controller, const char *path, pid_t pid) {
@@ -670,17 +867,18 @@ int cg_attach_fallback(const char *controller, const char *path, pid_t pid) {
* the destination */
PATH_FOREACH_PREFIX(prefix, path) {
- r = cg_attach(controller, prefix, pid);
- if (r >= 0)
- break;
+ int q;
+
+ q = cg_attach(controller, prefix, pid);
+ if (q >= 0)
+ return q;
}
}
- return 0;
+ return r;
}
-/// UNNEEDED by elogind
-#if 0
+#if 0 /// UNNEEDED by elogind
int cg_set_group_access(
const char *controller,
const char *path,
@@ -691,7 +889,8 @@ int cg_set_group_access(
_cleanup_free_ char *fs = NULL;
int r;
- assert(path);
+ if (mode == MODE_INVALID && uid == UID_INVALID && gid == GID_INVALID)
+ return 0;
if (mode != MODE_INVALID)
mode &= 0777;
@@ -700,7 +899,20 @@ 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(
@@ -729,21 +941,72 @@ int cg_set_task_access(
if (r < 0)
return r;
- /* Compatibility, Always keep values for "tasks" in sync with
- * "cgroup.procs" */
- r = cg_get_path(controller, path, "tasks", &procs);
+ 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);
+ }
+
+ 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;
- return chmod_and_chown(procs, mode, uid, gid);
+ 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;
- size_t cs;
+ const char *fs, *controller_str = NULL;
+ size_t cs = 0;
+ int unified;
assert(path);
assert(pid >= 0);
@@ -751,56 +1014,69 @@ int cg_pid_get_path(const char *controller, pid_t pid, char **path) {
if (controller) {
if (!cg_controller_is_valid(controller))
return -EINVAL;
-
- controller = normalize_controller(controller);
} else
- controller = ELOGIND_CGROUP_CONTROLLER;
+ controller = SYSTEMD_CGROUP_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;
- fs = procfs_file_alloca(pid, "cgroup");
+ 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;
- cs = strlen(controller);
-
FOREACH_LINE(line, f, return -errno) {
- char *l, *p, *e;
- size_t k;
- const char *word, *state;
- bool found = false;
+ char *e, *p;
truncate_nl(line);
- l = strchr(line, ':');
- if (!l)
- continue;
+ if (unified) {
+ e = startswith(line, "0:");
+ if (!e)
+ continue;
- l++;
- e = strchr(l, ':');
- if (!e)
- continue;
+ e = strchr(e, ':');
+ if (!e)
+ continue;
+ } else {
+ char *l;
+ size_t k;
+ const char *word, *state;
+ bool found = false;
+
+ l = strchr(line, ':');
+ if (!l)
+ continue;
- *e = 0;
+ l++;
+ e = strchr(l, ':');
+ if (!e)
+ continue;
- 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 (k == 5 + cs &&
- memcmp(word, "name=", 5) == 0 &&
- memcmp(word+5, controller, 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;
@@ -809,18 +1085,22 @@ int cg_pid_get_path(const char *controller, pid_t pid, char **path) {
return 0;
}
- return -ENOENT;
+ 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;
- char *sc;
+ const char *sc;
int r;
assert(agent);
+ 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);
if (r < 0)
return r;
@@ -830,11 +1110,11 @@ int cg_install_release_agent(const char *controller, const char *agent) {
return r;
sc = strstrip(contents);
- if (sc[0] == 0) {
- r = write_string_file_no_create(fs, agent);
+ if (isempty(sc)) {
+ r = write_string_file(fs, agent, 0);
if (r < 0)
return r;
- } else if (!streq(sc, agent))
+ } else if (!path_equal(sc, agent))
return -EEXIST;
fs = mfree(fs);
@@ -849,7 +1129,7 @@ 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;
@@ -866,11 +1146,17 @@ int cg_uninstall_release_agent(const char *controller) {
_cleanup_free_ char *fs = NULL;
int r;
+ 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;
@@ -880,81 +1166,92 @@ 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, bool ignore_self) {
+int cg_is_empty(const char *controller, const char *path) {
_cleanup_fclose_ FILE *f = NULL;
- pid_t pid = 0, self_pid;
- bool found = false;
+ pid_t pid;
int r;
assert(path);
r = cg_enumerate_processes(controller, path, &f);
+ if (r == -ENOENT)
+ return 1;
if (r < 0)
- return r == -ENOENT ? 1 : r;
-
- self_pid = getpid();
-
- while ((r = cg_read_pid(f, &pid)) > 0) {
-
- if (ignore_self && pid == self_pid)
- continue;
-
- found = true;
- break;
- }
+ return r;
+ r = cg_read_pid(f, &pid);
if (r < 0)
return r;
- return !found;
+ return r == 0;
}
-int cg_is_empty_recursive(const char *controller, const char *path, bool ignore_self) {
- _cleanup_closedir_ DIR *d = NULL;
- char *fn;
+int cg_is_empty_recursive(const char *controller, const char *path) {
int r;
assert(path);
- r = cg_is_empty(controller, path, ignore_self);
- if (r <= 0)
- return r;
+ /* The root cgroup is always populated */
+ if (controller && (isempty(path) || path_equal(path, "/")))
+ return false;
- r = cg_enumerate_subgroups(controller, path, &d);
+ r = cg_unified_controller(controller);
if (r < 0)
- return r == -ENOENT ? 1 : r;
+ return r;
+ if (r > 0) {
+ _cleanup_free_ char *t = NULL;
- while ((r = cg_read_subgroup(d, &fn)) > 0) {
- _cleanup_free_ char *p = NULL;
+ /* On the unified hierarchy we can check empty state
+ * via the "populated" attribute of "cgroup.events". */
- p = strjoin(path, "/", fn, NULL);
- free(fn);
- if (!p)
- return -ENOMEM;
+ 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;
- r = cg_is_empty_recursive(controller, p, ignore_self);
+ r = cg_is_empty(controller, path);
if (r <= 0)
return r;
- }
- if (r < 0)
- return r;
+ r = cg_enumerate_subgroups(controller, path, &d);
+ if (r == -ENOENT)
+ return 1;
+ if (r < 0)
+ return r;
- return 1;
+ while ((r = cg_read_subgroup(d, &fn)) > 0) {
+ _cleanup_free_ char *p = NULL;
+
+ p = strjoin(path, "/", fn);
+ free(fn);
+ if (!p)
+ return -ENOMEM;
+
+ r = cg_is_empty_recursive(controller, p);
+ if (r <= 0)
+ return r;
+ }
+ if (r < 0)
+ return r;
+
+ return true;
+ }
}
int cg_split_spec(const char *spec, char **controller, char **path) {
- const char *e;
char *t = NULL, *u = NULL;
- _cleanup_free_ char *v = NULL;
+ const char *e;
assert(spec);
@@ -982,7 +1279,7 @@ int cg_split_spec(const char *spec, char **controller, char **path) {
return -EINVAL;
if (controller) {
- t = strdup(normalize_controller(spec));
+ t = strdup(spec);
if (!t)
return -ENOMEM;
@@ -995,10 +1292,7 @@ int cg_split_spec(const char *spec, char **controller, char **path) {
return 0;
}
- v = strndup(spec, e-spec);
- if (!v)
- return -ENOMEM;
- t = strdup(normalize_controller(v));
+ t = strndup(spec, e-spec);
if (!t)
return -ENOMEM;
if (!cg_controller_is_valid(t)) {
@@ -1006,13 +1300,9 @@ int cg_split_spec(const char *spec, char **controller, char **path) {
return -EINVAL;
}
- if (streq(e+1, "")) {
- u = strdup("/");
- if (!u) {
- free(t);
- return -ENOMEM;
- }
- } else {
+ if (isempty(e+1))
+ u = NULL;
+ else {
u = strdup(e+1);
if (!u) {
free(t);
@@ -1066,25 +1356,34 @@ 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;
- e = endswith(p, "/" SPECIAL_SYSTEM_SLICE);
+ e = endswith(p, "/" SPECIAL_INIT_SCOPE);
+ if (!e)
+ e = endswith(p, "/" SPECIAL_SYSTEM_SLICE); /* legacy */
+ if (!e)
+ e = endswith(p, "/system"); /* even more legacy */
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) {
@@ -1104,10 +1403,15 @@ 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 (p)
+#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;
@@ -1123,10 +1427,12 @@ 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;
@@ -1143,11 +1449,13 @@ 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;
@@ -1357,8 +1665,6 @@ int cg_path_get_user_unit(const char *path, char **ret) {
return cg_path_get_unit(t, ret);
}
-/// UNNEDED by elogind
-#if 0
int cg_pid_get_user_unit(pid_t pid, char **unit) {
_cleanup_free_ char *cgroup = NULL;
int r;
@@ -1371,25 +1677,20 @@ int cg_pid_get_user_unit(pid_t pid, char **unit) {
return cg_path_get_user_unit(cgroup, unit);
}
-#endif // 0
int cg_path_get_machine_name(const char *path, char **machine) {
- _cleanup_free_ char *u = NULL, *sl = NULL;
+ _cleanup_free_ char *u = NULL;
+ const char *sl;
int r;
r = cg_path_get_unit(path, &u);
if (r < 0)
return r;
- sl = strjoin("/run/systemd/machines/unit:", u, NULL);
- if (!sl)
- return -ENOMEM;
-
+ sl = strjoina("/run/systemd/machines/unit:", u);
return readlink_malloc(sl, machine);
}
-/// UNNEDED by elogind
-#if 0
int cg_pid_get_machine_name(pid_t pid, char **machine) {
_cleanup_free_ char *cgroup = NULL;
int r;
@@ -1405,6 +1706,7 @@ int cg_pid_get_machine_name(pid_t pid, char **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;
@@ -1425,10 +1727,31 @@ 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;
@@ -1450,6 +1773,7 @@ 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;
@@ -1475,8 +1799,6 @@ int cg_path_get_owner_uid(const char *path, uid_t *uid) {
return 0;
}
-/// UNNEDED by elogind
-#if 0
int cg_pid_get_owner_uid(pid_t pid, uid_t *uid) {
_cleanup_free_ char *cgroup = NULL;
int r;
@@ -1487,7 +1809,6 @@ int cg_pid_get_owner_uid(pid_t pid, uid_t *uid) {
return cg_path_get_owner_uid(cgroup, uid);
}
-#endif // 0
int cg_path_get_slice(const char *p, char **slice) {
const char *e = NULL;
@@ -1509,7 +1830,7 @@ 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;
@@ -1564,6 +1885,7 @@ 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;
@@ -1583,31 +1905,38 @@ char *cg_escape(const char *p) {
p[0] == '.' ||
streq(p, "notify_on_release") ||
streq(p, "release_agent") ||
- streq(p, "tasks"))
+ streq(p, "tasks") ||
+ startswith(p, "cgroup."))
need_prefix = true;
else {
const char *dot;
dot = strrchr(p, '.');
if (dot) {
+ CGroupController c;
+ size_t l = dot - p;
- if (dot - p == 6 && memcmp(p, "cgroup", 6) == 0)
- need_prefix = true;
- else {
- char *n;
+ for (c = 0; c < _CGROUP_CONTROLLER_MAX; c++) {
+ const char *n;
- n = strndupa(p, dot - p);
+ n = cgroup_controller_to_string(c);
- if (check_hierarchy(n) >= 0)
- need_prefix = true;
+ if (l != strlen(n))
+ continue;
+
+ if (memcmp(p, n, l) != 0)
+ continue;
+
+ need_prefix = true;
+ break;
}
}
}
if (need_prefix)
return strappend("_", p);
- else
- return strdup(p);
+
+ return strdup(p);
}
char *cg_unescape(const char *p) {
@@ -1632,6 +1961,9 @@ 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;
@@ -1649,8 +1981,7 @@ 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;
@@ -1659,7 +1990,7 @@ 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("");
@@ -1719,6 +2050,7 @@ 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;
@@ -1728,7 +2060,7 @@ 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) {
@@ -1741,20 +2073,53 @@ int cg_get_attribute(const char *controller, const char *path, const char *attri
return read_one_line_file(p, ret);
}
-#endif // 0
-static const char mask_names[] =
- "cpu\0"
- "cpuacct\0"
- "blkio\0"
- "memory\0"
- "devices\0";
-
-/// UNNEEDED by elogind
-#if 0
-int cg_create_everywhere(CGroupControllerMask supported, CGroupControllerMask mask, const char *path) {
- CGroupControllerMask bit = 1;
- const char *n;
+#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;
/* This one will create a cgroup in our private tree, but also
@@ -1762,135 +2127,258 @@ int cg_create_everywhere(CGroupControllerMask supported, CGroupControllerMask ma
* 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;
- /* Then, do the same in the other hierarchies */
- NULSTR_FOREACH(n, mask_names) {
+ /* If we are in the unified hierarchy, we are done now */
+ r = cg_all_unified();
+ if (r < 0)
+ return r;
+ if (r > 0)
+ return 0;
+
+ /* Otherwise, do the same in the other hierarchies */
+ for (c = 0; c < _CGROUP_CONTROLLER_MAX; c++) {
+ CGroupMask bit = CGROUP_CONTROLLER_TO_MASK(c);
+ const char *n;
+
+ n = cgroup_controller_to_string(c);
+
if (mask & bit)
- cg_create(n, path);
+ (void) cg_create(n, path);
else if (supported & bit)
- cg_trim(n, path, true);
-
- bit <<= 1;
+ (void) cg_trim(n, path, true);
}
return 0;
}
-#endif // 0
-int cg_attach_everywhere(CGroupControllerMask supported, const char *path, pid_t pid, cg_migrate_callback_t path_callback, void *userdata) {
- CGroupControllerMask bit = 1;
- const char *n;
+int cg_attach_everywhere(CGroupMask supported, const char *path, pid_t pid, cg_migrate_callback_t path_callback, void *userdata) {
+ CGroupController c;
int r;
- r = cg_attach(ELOGIND_CGROUP_CONTROLLER, path, pid);
+ r = cg_attach(SYSTEMD_CGROUP_CONTROLLER, path, pid);
if (r < 0)
return r;
- NULSTR_FOREACH(n, mask_names) {
+ r = cg_all_unified();
+ if (r < 0)
+ return r;
+ if (r > 0)
+ return 0;
- if (supported & bit) {
- const char *p = NULL;
+ for (c = 0; c < _CGROUP_CONTROLLER_MAX; c++) {
+ CGroupMask bit = CGROUP_CONTROLLER_TO_MASK(c);
+ const char *p = NULL;
- if (path_callback)
- p = path_callback(bit, userdata);
+ if (!(supported & bit))
+ continue;
- if (!p)
- p = path;
+ if (path_callback)
+ p = path_callback(bit, userdata);
- cg_attach_fallback(n, p, pid);
- }
+ if (!p)
+ p = path;
- bit <<= 1;
+ (void) cg_attach_fallback(cgroup_controller_to_string(c), p, pid);
}
return 0;
}
-/// UNNEEDED by elogind
-#if 0
-int cg_attach_many_everywhere(CGroupControllerMask supported, const char *path, Set* pids, cg_migrate_callback_t path_callback, void *userdata) {
+int cg_attach_many_everywhere(CGroupMask supported, const char *path, Set* pids, cg_migrate_callback_t path_callback, void *userdata) {
Iterator i;
void *pidp;
int r = 0;
SET_FOREACH(pidp, pids, i) {
- pid_t pid = PTR_TO_LONG(pidp);
+ pid_t pid = PTR_TO_PID(pidp);
int q;
q = cg_attach_everywhere(supported, path, pid, path_callback, userdata);
- if (q < 0)
+ if (q < 0 && r >= 0)
r = q;
}
return r;
}
-int cg_migrate_everywhere(CGroupControllerMask supported, const char *from, const char *to, cg_migrate_callback_t to_callback, void *userdata) {
- CGroupControllerMask bit = 1;
- const char *n;
- int r;
+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, 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;
}
- NULSTR_FOREACH(n, mask_names) {
- if (supported & bit) {
- const char *p = NULL;
+ q = cg_all_unified();
+ if (q < 0)
+ return q;
+ if (q > 0)
+ return r;
- if (to_callback)
- p = to_callback(bit, userdata);
+ for (c = 0; c < _CGROUP_CONTROLLER_MAX; c++) {
+ CGroupMask bit = CGROUP_CONTROLLER_TO_MASK(c);
+ const char *p = NULL;
- if (!p)
- p = to;
+ if (!(supported & bit))
+ continue;
- cg_migrate_recursive_fallback(ELOGIND_CGROUP_CONTROLLER, to, n, p, false, false);
- }
+ if (to_callback)
+ p = to_callback(bit, userdata);
+
+ if (!p)
+ p = to;
- bit <<= 1;
+ (void) cg_migrate_recursive_fallback(SYSTEMD_CGROUP_CONTROLLER, to, cgroup_controller_to_string(c), p, 0);
}
return 0;
}
-int cg_trim_everywhere(CGroupControllerMask supported, const char *path, bool delete_root) {
- CGroupControllerMask bit = 1;
- const char *n;
- int r;
+int cg_trim_everywhere(CGroupMask supported, const char *path, bool delete_root) {
+ CGroupController c;
+ 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;
- NULSTR_FOREACH(n, mask_names) {
- if (supported & bit)
- cg_trim(n, path, delete_root);
+ 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);
+
+ if (!(supported & bit))
+ continue;
- bit <<= 1;
+ (void) cg_trim(cgroup_controller_to_string(c), path, 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;
+
+ /* Determines the mask of supported cgroup controllers. Only
+ * includes controllers we can make sense of and that are
+ * actually accessible. */
+
+ r = cg_all_unified();
+ if (r < 0)
+ return r;
+ if (r > 0) {
+ _cleanup_free_ char *root = NULL, *controllers = NULL, *path = NULL;
-CGroupControllerMask cg_mask_supported(void) {
- CGroupControllerMask bit = 1, mask = 0;
- const char *n;
+ /* In the unified hierarchy we can read the supported
+ * and accessible controllers from a the top-level
+ * cgroup attribute */
- NULSTR_FOREACH(n, mask_names) {
- if (check_hierarchy(n) >= 0)
- mask |= bit;
+ r = cg_get_root_path(&root);
+ if (r < 0)
+ return r;
- bit <<= 1;
+ r = cg_get_path(SYSTEMD_CGROUP_CONTROLLER, root, "cgroup.controllers", &path);
+ if (r < 0)
+ return r;
+
+ r = read_one_line_file(path, &controllers);
+ if (r < 0)
+ return r;
+
+ 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
+ * everything else off. */
+ mask &= CGROUP_MASK_CPU | CGROUP_MASK_MEMORY | CGROUP_MASK_IO | CGROUP_MASK_PIDS;
+
+ } else {
+ CGroupController c;
+
+ /* In the legacy hierarchy, we check whether which
+ * hierarchies are mounted. */
+
+ for (c = 0; c < _CGROUP_CONTROLLER_MAX; c++) {
+ const char *n;
+
+ n = cgroup_controller_to_string(c);
+ if (controller_is_accessible(n) >= 0)
+ mask |= CGROUP_CONTROLLER_TO_MASK(c);
+ }
}
- return mask;
+ *ret = mask;
+ return 0;
}
+#if 0 /// UNNEEDED by elogind
int cg_kernel_controllers(Set *controllers) {
_cleanup_fclose_ FILE *f = NULL;
char buf[LINE_MAX];
@@ -1898,6 +2386,11 @@ int cg_kernel_controllers(Set *controllers) {
assert(controllers);
+ /* 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). */
+
f = fopen("/proc/cgroups", "re");
if (!f) {
if (errno == ENOENT)
@@ -1918,7 +2411,7 @@ int cg_kernel_controllers(Set *controllers) {
if (feof(f))
break;
- if (ferror(f) && errno)
+ if (ferror(f) && errno > 0)
return -errno;
return -EBADMSG;
@@ -1929,7 +2422,7 @@ int cg_kernel_controllers(Set *controllers) {
continue;
}
- if (!filename_is_valid(controller)) {
+ if (!cg_controller_is_valid(controller)) {
free(controller);
return -EBADMSG;
}
@@ -1942,3 +2435,332 @@ int cg_kernel_controllers(Set *controllers) {
return 0;
}
#endif // 0
+
+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) {
+
+ struct statfs fs;
+
+ /* Checks if we support the unified hierarchy. Returns an
+ * error when the cgroup hierarchies aren't mounted yet or we
+ * have any other trouble determining if the unified hierarchy
+ * is supported. */
+
+ 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))
+ 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 >= 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_free_ char *fs = NULL;
+ CGroupController c;
+ int r;
+
+ assert(p);
+
+ if (supported == 0)
+ return 0;
+
+ 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);
+ if (r < 0)
+ return r;
+
+ for (c = 0; c < _CGROUP_CONTROLLER_MAX; c++) {
+ CGroupMask bit = CGROUP_CONTROLLER_TO_MASK(c);
+ const char *n;
+
+ if (!(supported & bit))
+ continue;
+
+ n = cgroup_controller_to_string(c);
+ {
+ char s[1 + strlen(n) + 1];
+
+ s[0] = mask & bit ? '+' : '-';
+ strcpy(s + 1, n);
+
+ r = write_string_file(fs, s, 0);
+ if (r < 0)
+ log_debug_errno(r, "Failed to enable controller %s for %s (%s): %m", n, p, fs);
+ }
+ }
+
+ return 0;
+}
+
+bool cg_is_unified_wanted(void) {
+ static thread_local int wanted = -1;
+ 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. */
+ 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;
+
+ /* If we have a cached value, return that. */
+ if (wanted >= 0)
+ return wanted;
+
+ /* Check if we have cgroups2 already mounted. */
+ if (cg_unified_flush() >= 0 &&
+ unified_cache == CGROUP_UNIFIED_ALL)
+ 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. */
+
+ /* 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;
+}
+
+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_DEVICES] = "devices",
+ [CGROUP_CONTROLLER_PIDS] = "pids",
+};
+
+DEFINE_STRING_TABLE_LOOKUP(cgroup_controller, CGroupController);