chiark / gitweb /
Prep v232: Do not listen to SYSTEMD_* environment variables to override things.
[elogind.git] / src / basic / cgroup-util.c
index 5dc631e..9dd93d0 100644 (file)
@@ -1,5 +1,3 @@
-/*-*- Mode: C; c-basic-offset: 8; indent-tabs-mode: nil -*-*/
-
 /***
   This file is part of systemd.
 
   along with systemd; If not, see <http://www.gnu.org/licenses/>.
 ***/
 
+#include <dirent.h>
 #include <errno.h>
-#include <unistd.h>
+#include <ftw.h>
+//#include <limits.h>
 #include <signal.h>
-#include <string.h>
+//#include <stddef.h>
 #include <stdlib.h>
-#include <dirent.h>
+#include <string.h>
 #include <sys/stat.h>
+//#include <sys/statfs.h>
 #include <sys/types.h>
-#include <ftw.h>
+#include <unistd.h>
 
-#include "set.h"
-#include "macro.h"
-#include "util.h"
-#include "formats-util.h"
-#include "process-util.h"
-#include "path-util.h"
-// #include "unit-name.h"
+#include "alloc-util.h"
+#include "cgroup-util.h"
+//#include "def.h"
+#include "dirent-util.h"
+#include "extract-word.h"
+#include "fd-util.h"
 #include "fileio.h"
-// #include "special.h"
-#include "mkdir.h"
+#include "formats-util.h"
+#include "fs-util.h"
+//#include "log.h"
 #include "login-util.h"
-#include "cgroup-util.h"
+#include "macro.h"
+//#include "missing.h"
+#include "mkdir.h"
+#include "parse-util.h"
+#include "path-util.h"
+#include "proc-cmdline.h"
+#include "process-util.h"
+#include "set.h"
+//#include "special.h"
+#include "stat-util.h"
+#include "stdio-util.h"
+#include "string-table.h"
+#include "string-util.h"
+#include "unit-name.h"
+#include "user-util.h"
 
 int cg_enumerate_processes(const char *controller, const char *path, FILE **_f) {
         _cleanup_free_ char *fs = NULL;
@@ -76,7 +91,7 @@ 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 +101,55 @@ 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;
@@ -149,7 +213,15 @@ int cg_rmdir(const char *controller, const char *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 +229,11 @@ int cg_kill(const char *controller, const char *path, int sig, bool sigcont, boo
 
         assert(sig >= 0);
 
+         /* Don't send SIGCONT twice. Also, SIGKILL always works even when process is suspended, hence don't send
+          * SIGCONT on SIGKILL. */
+        if (IN_SET(sig, SIGCONT, SIGKILL))
+                flags &= ~CGROUP_SIGCONT;
+
         /* This goes through the tasks list and kills them all. This
          * is repeated until no further processes are added to the
          * tasks list, to properly handle forking processes */
@@ -184,19 +261,22 @@ int cg_kill(const char *controller, const char *path, int sig, bool sigcont, boo
 
                 while ((r = cg_read_pid(f, &pid)) > 0) {
 
-                        if (ignore_self && pid == my_pid)
+                        if ((flags & CGROUP_IGNORE_SELF) && pid == my_pid)
                                 continue;
 
                         if (set_get(s, PID_TO_PTR(pid)) == PID_TO_PTR(pid))
                                 continue;
 
+                        if (log_kill)
+                                log_kill(pid, sig, userdata);
+
                         /* If we haven't killed this process yet, kill
                          * it */
                         if (kill(pid, sig) < 0) {
                                 if (ret >= 0 && errno != ESRCH)
                                         ret = -errno;
                         } else {
-                                if (sigcont && sig != SIGKILL)
+                                if (flags & CGROUP_SIGCONT)
                                         (void) kill(pid, SIGCONT);
 
                                 if (ret == 0)
@@ -230,7 +310,15 @@ int cg_kill(const char *controller, const char *path, int sig, bool sigcont, boo
         return ret;
 }
 
-int cg_kill_recursive(const char *controller, const char *path, int sig, bool sigcont, bool ignore_self, bool rem, Set *s) {
+int cg_kill_recursive(
+                const char *controller,
+                const char *path,
+                int sig,
+                CGroupFlags flags,
+                Set *s,
+                cg_kill_log_func_t log_kill,
+                void *userdata) {
+
         _cleanup_set_free_ Set *allocated_set = NULL;
         _cleanup_closedir_ DIR *d = NULL;
         int r, ret;
@@ -245,7 +333,7 @@ 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) {
@@ -263,15 +351,14 @@ int cg_kill_recursive(const char *controller, const char *path, int sig, bool si
                 if (!p)
                         return -ENOMEM;
 
-                r = cg_kill_recursive(controller, p, sig, sigcont, ignore_self, rem, s);
+                r = cg_kill_recursive(controller, p, sig, flags, s, log_kill, userdata);
                 if (r != 0 && ret >= 0)
                         ret = r;
         }
-
         if (ret >= 0 && r < 0)
                 ret = r;
 
-        if (rem) {
+        if (flags & CGROUP_REMOVE) {
                 r = cg_rmdir(controller, path);
                 if (r < 0 && ret >= 0 && r != -ENOENT && r != -EBUSY)
                         return r;
@@ -280,7 +367,13 @@ 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;
@@ -299,8 +392,8 @@ int cg_migrate(const char *cfrom, const char *pfrom, const char *cto, const char
 
         log_debug_elogind("Migrating \"%s\"/\"%s\" to \"%s\"/\"%s\" (%s)",
                           cfrom, pfrom, cto, pto,
-                          ignore_self ? "ignoring self" : "watching self");
-
+                          (flags & CGROUP_IGNORE_SELF)
+                          ? "ignoring self" : "watching self");
         do {
                 _cleanup_fclose_ FILE *f = NULL;
                 pid_t pid = 0;
@@ -319,7 +412,7 @@ int cg_migrate(const char *cfrom, const char *pfrom, const char *cto, const char
                         /* This might do weird stuff if we aren't a
                          * single-threaded program. However, we
                          * luckily know we are not */
-                        if (ignore_self && pid == my_pid)
+                        if ((flags & CGROUP_IGNORE_SELF) && pid == my_pid)
                                 continue;
 
                         if (set_get(s, PID_TO_PTR(pid)) == PID_TO_PTR(pid))
@@ -367,8 +460,7 @@ 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;
@@ -379,7 +471,7 @@ 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) {
@@ -395,9 +487,9 @@ int cg_migrate_recursive(
                 p = strjoin(pfrom, "/", fn, NULL);
                 free(fn);
                 if (!p)
-                                return -ENOMEM;
+                        return -ENOMEM;
 
-                r = cg_migrate_recursive(cfrom, p, cto, pto, ignore_self, rem);
+                r = cg_migrate_recursive(cfrom, p, cto, pto, flags);
                 if (r != 0 && ret >= 0)
                         ret = r;
         }
@@ -405,7 +497,7 @@ 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;
@@ -419,8 +511,7 @@ int cg_migrate_recursive_fallback(
                 const char *pfrom,
                 const char *cto,
                 const char *pto,
-                bool ignore_self,
-                bool rem) {
+                CGroupFlags flags) {
 
         int r;
 
@@ -429,7 +520,7 @@ int cg_migrate_recursive_fallback(
         assert(cto);
         assert(pto);
 
-        r = cg_migrate_recursive(cfrom, pfrom, cto, pto, ignore_self, rem);
+        r = cg_migrate_recursive(cfrom, pfrom, cto, pto, flags);
         if (r < 0) {
                 char prefix[strlen(pto) + 1];
 
@@ -438,7 +529,7 @@ int cg_migrate_recursive_fallback(
                 PATH_FOREACH_PREFIX(prefix, pto) {
                         int q;
 
-                        q = cg_migrate_recursive(cfrom, pfrom, cto, prefix, ignore_self, rem);
+                        q = cg_migrate_recursive(cfrom, pfrom, cto, prefix, flags);
                         if (q >= 0)
                                 return q;
                 }
@@ -461,7 +552,7 @@ static const char *controller_to_dirname(const char *controller) {
         if (e)
                 return e;
 
-                return controller;
+        return controller;
 }
 
 static int join_path_legacy(const char *controller, const char *path, const char *suffix, char **fs) {
@@ -479,14 +570,14 @@ static int join_path_legacy(const char *controller, const char *path, const char
                 t = strjoin("/sys/fs/cgroup/", dn, "/", suffix, NULL);
         else if (isempty(suffix))
                 t = strjoin("/sys/fs/cgroup/", dn, "/", path, NULL);
-                else
+        else
                 t = strjoin("/sys/fs/cgroup/", dn, "/", path, "/", suffix, NULL);
         if (!t)
                 return -ENOMEM;
 
         *fs = t;
         return 0;
-        }
+}
 
 static int join_path_unified(const char *path, const char *suffix, char **fs) {
         char *t;
@@ -520,7 +611,7 @@ int cg_get_path(const char *controller, const char *path, const char *suffix, ch
                  * *below* the controllers, without any prefix. */
 
                 if (!path && !suffix)
-                return -EINVAL;
+                        return -EINVAL;
 
                 if (!suffix)
                         t = strdup(path);
@@ -538,7 +629,7 @@ int cg_get_path(const char *controller, const char *path, const char *suffix, ch
         if (!cg_controller_is_valid(controller))
                 return -EINVAL;
 
-        unified = cg_unified();
+        unified = cg_all_unified();
         if (unified < 0)
                 return unified;
 
@@ -546,12 +637,12 @@ int cg_get_path(const char *controller, const char *path, const char *suffix, ch
                 r = join_path_unified(path, suffix, fs);
         else
                 r = join_path_legacy(controller, path, suffix, fs);
-                if (r < 0)
-                        return r;
+        if (r < 0)
+                return r;
 
         path_kill_slashes(*fs);
         return 0;
-        }
+}
 
 static int controller_is_accessible(const char *controller) {
         int unified;
@@ -566,14 +657,14 @@ static int controller_is_accessible(const char *controller) {
         if (!cg_controller_is_valid(controller))
                 return -EINVAL;
 
-        unified = cg_unified();
+        unified = cg_all_unified();
         if (unified < 0)
                 return unified;
         if (unified > 0) {
                 /* We don't support named hierarchies if we are using
                  * the unified hierarchy. */
 
-                if (streq(controller, ELOGIND_CGROUP_CONTROLLER))
+                if (streq(controller, SYSTEMD_CGROUP_CONTROLLER))
                         return 0;
 
                 if (startswith(controller, "name="))
@@ -635,7 +726,7 @@ int cg_trim(const char *controller, const char *path, bool delete_root) {
         if (nftw(fs, trim_cb, 64, FTW_DEPTH|FTW_MOUNT|FTW_PHYS) != 0) {
                 if (errno == ENOENT)
                         r = 0;
-                else if (errno != 0)
+                else if (errno > 0)
                         r = -errno;
                 else
                         r = -EIO;
@@ -704,9 +795,9 @@ 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);
 
-        return write_string_file_no_create(fs, c);
+        return write_string_file(fs, c, 0);
 }
 
 int cg_attach_fallback(const char *controller, const char *path, pid_t pid) {
@@ -735,8 +826,7 @@ int cg_attach_fallback(const char *controller, const char *path, pid_t pid) {
         return r;
 }
 
-/// UNNEEDED by elogind
-#if 0
+#if 0 /// UNNEEDED by elogind
 int cg_set_group_access(
                 const char *controller,
                 const char *path,
@@ -786,7 +876,7 @@ int cg_set_task_access(
         if (r < 0)
                 return r;
 
-        unified = cg_unified();
+        unified = cg_unified(controller);
         if (unified < 0)
                 return unified;
         if (unified)
@@ -799,6 +889,43 @@ int cg_set_task_access(
 
         return 0;
 }
+
+int cg_set_xattr(const char *controller, const char *path, const char *name, const void *value, size_t size, int flags) {
+        _cleanup_free_ char *fs = NULL;
+        int r;
+
+        assert(path);
+        assert(name);
+        assert(value || size <= 0);
+
+        r = cg_get_path(controller, path, NULL, &fs);
+        if (r < 0)
+                return r;
+
+        if (setxattr(fs, name, value, size, flags) < 0)
+                return -errno;
+
+        return 0;
+}
+
+int cg_get_xattr(const char *controller, const char *path, const char *name, void *value, size_t size) {
+        _cleanup_free_ char *fs = NULL;
+        ssize_t n;
+        int r;
+
+        assert(path);
+        assert(name);
+
+        r = cg_get_path(controller, path, NULL, &fs);
+        if (r < 0)
+                return r;
+
+        n = getxattr(fs, name, value, size);
+        if (n < 0)
+                return -errno;
+
+        return (int) n;
+}
 #endif // 0
 
 int cg_pid_get_path(const char *controller, pid_t pid, char **path) {
@@ -811,20 +938,21 @@ int cg_pid_get_path(const char *controller, pid_t pid, char **path) {
         assert(path);
         assert(pid >= 0);
 
-        unified = cg_unified();
+        if (controller) {
+                if (!cg_controller_is_valid(controller))
+                        return -EINVAL;
+        } else
+                controller = SYSTEMD_CGROUP_CONTROLLER;
+
+        unified = cg_unified(controller);
         if (unified < 0)
                 return unified;
-        if (unified == 0) {
-                if (controller) {
-                        if (!cg_controller_is_valid(controller))
-                                return -EINVAL;
-                } else
-                        controller = ELOGIND_CGROUP_CONTROLLER;
-
+        if (unified == 0)
                 cs = strlen(controller);
-        }
 
         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;
@@ -869,6 +997,7 @@ int cg_pid_get_path(const char *controller, pid_t pid, char **path) {
                                 continue;
                 }
 
+                log_debug_elogind("Found %s:%s", line, e+1);
                 p = strdup(e + 1);
                 if (!p)
                         return -ENOMEM;
@@ -887,7 +1016,7 @@ int cg_install_release_agent(const char *controller, const char *agent) {
 
         assert(agent);
 
-        unified = cg_unified();
+        unified = cg_unified(controller);
         if (unified < 0)
                 return unified;
         if (unified) /* doesn't apply to unified hierarchy */
@@ -903,7 +1032,7 @@ int cg_install_release_agent(const char *controller, const char *agent) {
 
         sc = strstrip(contents);
         if (isempty(sc)) {
-                r = write_string_file_no_create(fs, agent);
+                r = write_string_file(fs, agent, 0);
                 if (r < 0)
                         return r;
         } else if (!path_equal(sc, agent))
@@ -921,7 +1050,7 @@ 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;
 
@@ -938,7 +1067,7 @@ int cg_uninstall_release_agent(const char *controller) {
         _cleanup_free_ char *fs = NULL;
         int r, unified;
 
-        unified = cg_unified();
+        unified = cg_unified(controller);
         if (unified < 0)
                 return unified;
         if (unified) /* Doesn't apply to unified hierarchy */
@@ -948,7 +1077,7 @@ int cg_uninstall_release_agent(const char *controller) {
         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;
 
@@ -958,7 +1087,7 @@ 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;
 
@@ -994,55 +1123,49 @@ int cg_is_empty_recursive(const char *controller, const char *path) {
         if (controller && (isempty(path) || path_equal(path, "/")))
                 return false;
 
-        unified = cg_unified();
+        unified = cg_unified(controller);
         if (unified < 0)
                 return unified;
 
         if (unified > 0) {
-                _cleanup_free_ char *populated = NULL, *t = NULL;
+                _cleanup_free_ char *t = NULL;
 
                 /* On the unified hierarchy we can check empty state
-                 * via the "cgroup.populated" attribute. */
+                 * via the "populated" attribute of "cgroup.events". */
 
-                r = cg_get_path(controller, path, "cgroup.populated", &populated);
-        if (r < 0)
-                return r;
-
-                r = read_one_line_file(populated, &t);
-                if (r == -ENOENT)
-                        return 1;
+                r = cg_read_event(controller, path, "populated", &t);
                 if (r < 0)
                         return r;
 
                 return streq(t, "0");
         } else {
-        _cleanup_closedir_ DIR *d = NULL;
-        char *fn;
+                _cleanup_closedir_ DIR *d = NULL;
+                char *fn;
 
                 r = cg_is_empty(controller, path);
-        if (r <= 0)
-                return r;
+                if (r <= 0)
+                        return r;
 
-        r = cg_enumerate_subgroups(controller, path, &d);
+                r = cg_enumerate_subgroups(controller, path, &d);
                 if (r == -ENOENT)
                         return 1;
-        if (r < 0)
+                if (r < 0)
                         return r;
 
-        while ((r = cg_read_subgroup(d, &fn)) > 0) {
-                _cleanup_free_ char *p = NULL;
+                while ((r = cg_read_subgroup(d, &fn)) > 0) {
+                        _cleanup_free_ char *p = NULL;
 
-                p = strjoin(path, "/", fn, NULL);
-                free(fn);
-                if (!p)
-                        return -ENOMEM;
+                        p = strjoin(path, "/", fn, NULL);
+                        free(fn);
+                        if (!p)
+                                return -ENOMEM;
 
                         r = cg_is_empty_recursive(controller, p);
-                if (r <= 0)
+                        if (r <= 0)
+                                return r;
+                }
+                if (r < 0)
                         return r;
-        }
-        if (r < 0)
-                return r;
 
                 return true;
         }
@@ -1155,18 +1278,17 @@ int cg_mangle_path(const char *path, char **result) {
         if (r < 0)
                 return r;
 
-        return cg_get_path(c ? c : ELOGIND_CGROUP_CONTROLLER, p ? p : "/", NULL, result);
+        return cg_get_path(c ?: SYSTEMD_CGROUP_CONTROLLER, p ?: "/", NULL, result);
 }
 
 int cg_get_root_path(char **path) {
-/// elogind does not support systemd scopes and slices
-#if 0
+#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;
 
@@ -1182,7 +1304,7 @@ int cg_get_root_path(char **path) {
         return 0;
 #else
         assert(path);
-        return cg_pid_get_path(ELOGIND_CGROUP_CONTROLLER, 1, path);
+        return cg_pid_get_path(SYSTEMD_CGROUP_CONTROLLER, 1, path);
 #endif // 0
 }
 
@@ -1207,7 +1329,7 @@ int cg_shift_path(const char *cgroup, const char *root, const char **shifted) {
         }
 
         p = path_startswith(cgroup, root);
-        if (p && p[0] && (p > cgroup))
+        if (p && p > cgroup)
                 *shifted = p - 1;
         else
                 *shifted = cgroup;
@@ -1223,7 +1345,7 @@ 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;
 
@@ -1250,9 +1372,8 @@ int cg_pid_get_path_shifted(pid_t pid, const char *root, char **cgroup) {
         return 0;
 }
 
-/// UNNEEDED by elogind
-#if 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;
 
@@ -1503,9 +1624,7 @@ int cg_pid_get_machine_name(pid_t pid, char **machine) {
 #endif // 0
 
 int cg_path_get_session(const char *path, char **session) {
-        /* Elogind uses a flat hierarchy, just "/SESSION".  The only
-           wrinkle is that SESSION might be escaped.  */
-#if 0
+#if 0 /// UNNEEDED by elogind
         _cleanup_free_ char *unit = NULL;
         char *start, *end;
         int r;
@@ -1527,6 +1646,8 @@ int cg_path_get_session(const char *path, char **session) {
         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);
@@ -1570,8 +1691,7 @@ int cg_pid_get_session(pid_t pid, char **session) {
         return cg_path_get_session(cgroup, session);
 }
 
-/// UNNEEDED by elogind
-#if 0
+#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;
@@ -1628,7 +1748,7 @@ 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;
 
@@ -1725,7 +1845,7 @@ char *cg_escape(const char *p) {
                                 if (memcmp(p, n, l) != 0)
                                         continue;
 
-                                        need_prefix = true;
+                                need_prefix = true;
                                 break;
                         }
                 }
@@ -1734,7 +1854,7 @@ char *cg_escape(const char *p) {
         if (need_prefix)
                 return strappend("_", p);
 
-                return strdup(p);
+        return strdup(p);
 }
 
 char *cg_unescape(const char *p) {
@@ -1776,8 +1896,7 @@ 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;
@@ -1786,7 +1905,7 @@ 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("");
@@ -1856,11 +1975,9 @@ 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);
 }
 
-/// UNNEEDED by elogind
-#if 0
 int cg_get_attribute(const char *controller, const char *path, const char *attribute, char **ret) {
         _cleanup_free_ char *p = NULL;
         int r;
@@ -1872,6 +1989,50 @@ int cg_get_attribute(const char *controller, const char *path, const char *attri
         return read_one_line_file(p, ret);
 }
 
+#if 0 /// UNNEEDED by elogind
+int cg_get_keyed_attribute(const char *controller, const char *path, const char *attribute, const char **keys, char **values) {
+        _cleanup_free_ char *filename = NULL, *content = NULL;
+        char *line, *p;
+        int i, r;
+
+        for (i = 0; keys[i]; i++)
+                values[i] = NULL;
+
+        r = cg_get_path(controller, path, attribute, &filename);
+        if (r < 0)
+                return r;
+
+        r = read_full_file(filename, &content, NULL);
+        if (r < 0)
+                return r;
+
+        p = content;
+        while ((line = strsep(&p, "\n"))) {
+                char *key;
+
+                key = strsep(&line, " ");
+
+                for (i = 0; keys[i]; i++) {
+                        if (streq(key, keys[i])) {
+                                values[i] = strdup(line);
+                                break;
+                        }
+                }
+        }
+
+        for (i = 0; keys[i]; i++) {
+                if (!values[i]) {
+                        for (i = 0; keys[i]; i++) {
+                                free(values[i]);
+                                values[i] = NULL;
+                        }
+                        return -ENOENT;
+                }
+        }
+
+        return 0;
+}
+
 int cg_create_everywhere(CGroupMask supported, CGroupMask mask, const char *path) {
         CGroupController c;
         int r, unified;
@@ -1881,12 +2042,12 @@ int cg_create_everywhere(CGroupMask supported, CGroupMask mask, const char *path
          * in all others */
 
         /* First create the cgroup in our own hierarchy. */
-        r = cg_create(ELOGIND_CGROUP_CONTROLLER, path);
+        r = cg_create(SYSTEMD_CGROUP_CONTROLLER, path);
         if (r < 0)
                 return r;
 
         /* If we are in the unified hierarchy, we are done now */
-        unified = cg_unified();
+        unified = cg_all_unified();
         if (unified < 0)
                 return unified;
         if (unified > 0)
@@ -1912,11 +2073,11 @@ int cg_attach_everywhere(CGroupMask supported, const char *path, pid_t pid, cg_m
         CGroupController c;
         int r, unified;
 
-        r = cg_attach(ELOGIND_CGROUP_CONTROLLER, path, pid);
+        r = cg_attach(SYSTEMD_CGROUP_CONTROLLER, path, pid);
         if (r < 0)
                 return r;
 
-        unified = cg_unified();
+        unified = cg_all_unified();
         if (unified < 0)
                 return unified;
         if (unified > 0)
@@ -1924,16 +2085,16 @@ int cg_attach_everywhere(CGroupMask supported, const char *path, pid_t pid, cg_m
 
         for (c = 0; c < _CGROUP_CONTROLLER_MAX; c++) {
                 CGroupMask bit = CGROUP_CONTROLLER_TO_MASK(c);
-                        const char *p = NULL;
+                const char *p = NULL;
 
                 if (!(supported & bit))
                         continue;
 
-                        if (path_callback)
-                                p = path_callback(bit, userdata);
+                if (path_callback)
+                        p = path_callback(bit, userdata);
 
-                        if (!p)
-                                p = path;
+                if (!p)
+                        p = path;
 
                 (void) cg_attach_fallback(cgroup_controller_to_string(c), p, pid);
         }
@@ -1963,12 +2124,12 @@ int cg_migrate_everywhere(CGroupMask supported, const char *from, const char *to
         int r = 0, unified;
 
         if (!path_equal(from, to))  {
-                r = cg_migrate_recursive(ELOGIND_CGROUP_CONTROLLER, from, ELOGIND_CGROUP_CONTROLLER, to, false, true);
+                r = cg_migrate_recursive(SYSTEMD_CGROUP_CONTROLLER, from, SYSTEMD_CGROUP_CONTROLLER, to, CGROUP_REMOVE);
                 if (r < 0)
                         return r;
         }
 
-        unified = cg_unified();
+        unified = cg_all_unified();
         if (unified < 0)
                 return unified;
         if (unified > 0)
@@ -1976,18 +2137,18 @@ int cg_migrate_everywhere(CGroupMask supported, const char *from, const char *to
 
         for (c = 0; c < _CGROUP_CONTROLLER_MAX; c++) {
                 CGroupMask bit = CGROUP_CONTROLLER_TO_MASK(c);
-                        const char *p = NULL;
+                const char *p = NULL;
 
                 if (!(supported & bit))
                         continue;
 
-                        if (to_callback)
-                                p = to_callback(bit, userdata);
+                if (to_callback)
+                        p = to_callback(bit, userdata);
 
-                        if (!p)
-                                p = to;
+                if (!p)
+                        p = to;
 
-                (void) cg_migrate_recursive_fallback(ELOGIND_CGROUP_CONTROLLER, to, cgroup_controller_to_string(c), p, false, false);
+                (void) cg_migrate_recursive_fallback(SYSTEMD_CGROUP_CONTROLLER, to, cgroup_controller_to_string(c), p, 0);
         }
 
         return 0;
@@ -1997,11 +2158,11 @@ int cg_trim_everywhere(CGroupMask supported, const char *path, bool delete_root)
         CGroupController c;
         int r, unified;
 
-        r = cg_trim(ELOGIND_CGROUP_CONTROLLER, path, delete_root);
+        r = cg_trim(SYSTEMD_CGROUP_CONTROLLER, path, delete_root);
         if (r < 0)
                 return r;
 
-        unified = cg_unified();
+        unified = cg_all_unified();
         if (unified < 0)
                 return unified;
         if (unified > 0)
@@ -2028,7 +2189,7 @@ int cg_mask_supported(CGroupMask *ret) {
          * includes controllers we can make sense of and that are
          * actually accessible. */
 
-        unified = cg_unified();
+        unified = cg_all_unified();
         if (unified < 0)
                 return unified;
         if (unified > 0) {
@@ -2043,7 +2204,7 @@ int cg_mask_supported(CGroupMask *ret) {
                 if (r < 0)
                         return r;
 
-                r = cg_get_path(ELOGIND_CGROUP_CONTROLLER, root, "cgroup.controllers", &path);
+                r = cg_get_path(SYSTEMD_CGROUP_CONTROLLER, root, "cgroup.controllers", &path);
                 if (r < 0)
                         return r;
 
@@ -2067,11 +2228,12 @@ int cg_mask_supported(CGroupMask *ret) {
                                 continue;
 
                         mask |= CGROUP_CONTROLLER_TO_MASK(v);
-        }
+                }
 
-                /* Currently, we only support the memory controller in
-                 * the unified hierarchy, mask everything else off. */
-                mask &= CGROUP_MASK_MEMORY;
+                /* Currently, we support the cpu, memory, io and pids
+                 * controller in the unified hierarchy, mask
+                 * everything else off. */
+                mask &= CGROUP_MASK_CPU | CGROUP_MASK_MEMORY | CGROUP_MASK_IO | CGROUP_MASK_PIDS;
 
         } else {
                 CGroupController c;
@@ -2092,8 +2254,7 @@ int cg_mask_supported(CGroupMask *ret) {
         return 0;
 }
 
-/// UNNEEDED by elogind
-#if 0
+#if 0 /// UNNEEDED by elogind
 int cg_kernel_controllers(Set *controllers) {
         _cleanup_fclose_ FILE *f = NULL;
         char buf[LINE_MAX];
@@ -2126,7 +2287,7 @@ int cg_kernel_controllers(Set *controllers) {
                         if (feof(f))
                                 break;
 
-                        if (ferror(f) && errno != 0)
+                        if (ferror(f) && errno > 0)
                                 return -errno;
 
                         return -EBADMSG;
@@ -2151,9 +2312,10 @@ int cg_kernel_controllers(Set *controllers) {
 }
 #endif // 0
 
-static thread_local int unified_cache = -1;
+static thread_local CGroupUnified unified_cache = CGROUP_UNIFIED_UNKNOWN;
+
+static int cg_update_unified(void) {
 
-int cg_unified(void) {
         struct statfs fs;
 
         /* Checks if we support the unified hierarchy. Returns an
@@ -2161,35 +2323,57 @@ int cg_unified(void) {
          * have any other trouble determining if the unified hierarchy
          * is supported. */
 
-        if (unified_cache >= 0)
-                return unified_cache;
+        if (unified_cache >= CGROUP_UNIFIED_NONE)
+                return 0;
 
         if (statfs("/sys/fs/cgroup/", &fs) < 0)
                 return -errno;
 
-/// elogind can not support the unified hierarchy as a controller,
-/// so always assume a classical hierarchy.
-/// If, ond only *if*, someone really wants to substitute systemd-login
-/// in an environment managed by systemd with elogin, we might have to
-/// add such a support.
-#if 0
-        if (F_TYPE_EQUAL(fs.f_type, CGROUP_SUPER_MAGIC))
-                unified_cache = true;
-        else if (F_TYPE_EQUAL(fs.f_type, TMPFS_MAGIC))
+#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)) {
 #else
-        if (F_TYPE_EQUAL(fs.f_type, TMPFS_MAGIC))
+        /* 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)) {
 #endif // 0
-                unified_cache = false;
+                if (statfs("/sys/fs/cgroup/systemd/", &fs) < 0)
+                        return -errno;
+
+                unified_cache = F_TYPE_EQUAL(fs.f_type, CGROUP2_SUPER_MAGIC) ?
+                        CGROUP_UNIFIED_SYSTEMD : CGROUP_UNIFIED_NONE;
+        } else
+                return -ENOMEDIUM;
+
+        return 0;
+}
+
+int cg_unified(const char *controller) {
+
+        int r;
+
+        r = cg_update_unified();
+        if (r < 0)
+                return r;
+
+        if (streq_ptr(controller, SYSTEMD_CGROUP_CONTROLLER))
+                return unified_cache >= CGROUP_UNIFIED_SYSTEMD;
         else
-                return -ENOEXEC;
+                return unified_cache >= CGROUP_UNIFIED_ALL;
+}
 
-        return unified_cache;
+int cg_all_unified(void) {
+
+        return cg_unified(NULL);
 }
 
-/// UNNEEDED by elogind
-#if 0
+#if 0 /// UNNEEDED by elogind
 void cg_unified_flush(void) {
-        unified_cache = -1;
+        unified_cache = CGROUP_UNIFIED_UNKNOWN;
 }
 
 int cg_enable_everywhere(CGroupMask supported, CGroupMask mask, const char *p) {
@@ -2202,13 +2386,13 @@ int cg_enable_everywhere(CGroupMask supported, CGroupMask mask, const char *p) {
         if (supported == 0)
                 return 0;
 
-        unified = cg_unified();
+        unified = cg_all_unified();
         if (unified < 0)
                 return unified;
         if (!unified) /* on the legacy hiearchy there's no joining of controllers defined */
                 return 0;
 
-        r = cg_get_path(ELOGIND_CGROUP_CONTROLLER, p, "cgroup.subtree_control", &fs);
+        r = cg_get_path(SYSTEMD_CGROUP_CONTROLLER, p, "cgroup.subtree_control", &fs);
         if (r < 0)
                 return r;
 
@@ -2241,7 +2425,7 @@ bool cg_is_unified_wanted(void) {
 
         /* If the hierarchy is already mounted, then follow whatever
          * was chosen for it. */
-        unified = cg_unified();
+        unified = cg_all_unified();
         if (unified >= 0)
                 return unified;
 
@@ -2270,14 +2454,158 @@ bool cg_is_unified_wanted(void) {
 bool cg_is_legacy_wanted(void) {
         return !cg_is_unified_wanted();
 }
+
+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 (wanted >= 0)
+                return wanted;
+
+        r = get_proc_cmdline_key("systemd.legacy_systemd_cgroup_controller", NULL);
+        if (r > 0) {
+        if (r > 0)
+                wanted = false;
+        } else {
+        else {
+                _cleanup_free_ char *value = NULL;
+
+                r = get_proc_cmdline_key("systemd.legacy_systemd_cgroup_controller=", &value);
+                if (r < 0)
+                        return true;
+
+                if (r == 0)
+                        wanted = true;
+                else
+                        wanted = parse_boolean(value) <= 0;
+        }
+
+        return wanted;
+}
+
+bool cg_is_legacy_systemd_controller_wanted(void) {
+        return cg_is_legacy_wanted() && !cg_is_unified_systemd_controller_wanted();
+}
+#else
+bool cg_is_legacy_wanted(void) {
+        return true;
+}
 #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_DEVICE] = "devices",
+        [CGROUP_CONTROLLER_DEVICES] = "devices",
+        [CGROUP_CONTROLLER_PIDS] = "pids",
 };
 
 DEFINE_STRING_TABLE_LOOKUP(cgroup_controller, CGroupController);