X-Git-Url: https://www.chiark.greenend.org.uk/ucgi/~ianmdlvl/git?p=elogind.git;a=blobdiff_plain;f=src%2Fbasic%2Fcgroup-util.c;h=f7a892380a7e4975dd439ca22e0470e251ca6c70;hp=479254159e54c6b05f2e021fbe89adfacac47592;hb=34b8612104beeb72bf7e1deec998b66dbded5c73;hpb=5db0e7adf018c82dd63cd21d31dd313dff5561af

diff --git a/src/basic/cgroup-util.c b/src/basic/cgroup-util.c
index 479254159..f7a892380 100644
--- a/src/basic/cgroup-util.c
+++ b/src/basic/cgroup-util.c
@@ -29,18 +29,18 @@
 #include <sys/types.h>
 #include <ftw.h>
 
-#include "cgroup-util.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 "unit-name.h"
 #include "fileio.h"
-#include "special.h"
+// #include "special.h"
 #include "mkdir.h"
 #include "login-util.h"
+#include "cgroup-util.h"
 
 int cg_enumerate_processes(const char *controller, const char *path, FILE **_f) {
         _cleanup_free_ char *fs = NULL;
@@ -113,7 +113,7 @@ 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)
@@ -187,7 +187,7 @@ int cg_kill(const char *controller, const char *path, int sig, bool sigcont, boo
                         if (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 we haven't killed this process yet, kill
@@ -197,7 +197,7 @@ int cg_kill(const char *controller, const char *path, int sig, bool sigcont, boo
                                         ret = -errno;
                         } else {
                                 if (sigcont && sig != SIGKILL)
-                                        kill(pid, SIGCONT);
+                                        (void) kill(pid, SIGCONT);
 
                                 if (ret == 0)
                                         ret = 1;
@@ -205,7 +205,7 @@ 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;
@@ -233,7 +233,7 @@ int cg_kill(const char *controller, const char *path, int sig, bool sigcont, boo
 int cg_kill_recursive(const char *controller, const char *path, int sig, bool sigcont, bool ignore_self, bool rem, Set *s) {
         _cleanup_set_free_ Set *allocated_set = NULL;
         _cleanup_closedir_ DIR *d = NULL;
-        int r, ret = 0;
+        int r, ret;
         char *fn;
 
         assert(path);
@@ -264,7 +264,7 @@ int cg_kill_recursive(const char *controller, const char *path, int sig, bool si
                         return -ENOMEM;
 
                 r = cg_kill_recursive(controller, p, sig, sigcont, ignore_self, rem, s);
-                if (ret >= 0 && r != 0)
+                if (r != 0 && ret >= 0)
                         ret = r;
         }
 
@@ -297,6 +297,10 @@ 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,
+                          ignore_self ? "ignoring self" : "watching self");
+
         do {
                 _cleanup_fclose_ FILE *f = NULL;
                 pid_t pid = 0;
@@ -318,7 +322,15 @@ int cg_migrate(const char *cfrom, const char *pfrom, const char *cto, const char
                         if (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;
+
+                        /* 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 +342,7 @@ 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;
@@ -382,13 +394,9 @@ int cg_migrate_recursive(
 
                 p = strjoin(pfrom, "/", fn, NULL);
                 free(fn);
-                if (!p) {
-                        if (ret >= 0)
+                if (!p)
                                 return -ENOMEM;
 
-                        return ret;
-                }
-
                 r = cg_migrate_recursive(cfrom, p, cto, pto, ignore_self, rem);
                 if (r != 0 && ret >= 0)
                         ret = r;
@@ -428,114 +436,174 @@ int cg_migrate_recursive_fallback(
                 /* 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, ignore_self, rem);
+                        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
+        /* 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. */
+
+        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);
+        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, NULL);
+        else if (isempty(suffix))
+                t = strjoin("/sys/fs/cgroup/", dn, "/", path, NULL);
                 else
-                        return -EINVAL;
+                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;
+
+        assert(fs);
+
+        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, NULL);
         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;
+        int unified, r;
 
         assert(fs);
 
-        if (controller && !cg_controller_is_valid(controller))
+        if (!controller) {
+                char *t;
+
+                /* If no controller is specified, we return the path
+                 * *below* the controllers, without any prefix. */
+
+                if (!path && !suffix)
+                return -EINVAL;
+
+                if (!suffix)
+                        t = strdup(path);
+                else if (!path)
+                        t = strdup(suffix);
+                else
+                        t = strjoin(path, "/", suffix, NULL);
+                if (!t)
+                        return -ENOMEM;
+
+                *fs = path_kill_slashes(t);
+                return 0;
+        }
+
+        if (!cg_controller_is_valid(controller))
                 return -EINVAL;
 
-        if (_unlikely_(!good)) {
-                int r;
+        unified = cg_unified();
+        if (unified < 0)
+                return unified;
 
-                r = path_is_mount_point("/sys/fs/cgroup", 0);
+        if (unified > 0)
+                r = join_path_unified(path, suffix, fs);
+        else
+                r = join_path_legacy(controller, path, suffix, fs);
                 if (r < 0)
                         return r;
-                if (r == 0)
-                        return -ENOENT;
 
-                /* Cache this to save a few stat()s */
-                good = true;
+        path_kill_slashes(*fs);
+        return 0;
         }
 
-        p = controller ? normalize_controller(controller) : NULL;
+static int controller_is_accessible(const char *controller) {
+        int unified;
 
-        return join_path(p, path, suffix, fs);
-}
+        assert(controller);
 
-static int check_hierarchy(const char *p) {
-        const char *cc;
+        /* 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 */
 
-        assert(p);
+        if (!cg_controller_is_valid(controller))
+                return -EINVAL;
 
-        if (!filename_is_valid(p))
-                return 0;
+        unified = cg_unified();
+        if (unified < 0)
+                return unified;
+        if (unified > 0) {
+                /* We don't support named hierarchies if we are using
+                 * the unified hierarchy. */
 
-        /* Check if this controller actually really exists */
-        cc = strjoina("/sys/fs/cgroup/", p);
-        if (laccess(cc, F_OK) < 0)
-                return -errno;
+                if (streq(controller, ELOGIND_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);
+
+                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) {
-        const char *p;
         int r;
 
+        assert(controller);
         assert(fs);
 
-        if (!cg_controller_is_valid(controller))
-                return -EINVAL;
-
-        /* Normalize the controller syntax */
-        p = normalize_controller(controller);
-
-        /* Check if this controller actually really exists */
-        r = check_hierarchy(p);
+        /* 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) {
@@ -549,7 +617,7 @@ 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;
 }
 
@@ -564,8 +632,14 @@ 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)
@@ -575,23 +649,6 @@ int cg_trim(const char *controller, const char *path, bool delete_root) {
         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;
-
-        r = cg_migrate_recursive(controller, path, controller, parent, false, true);
-        return r == -ENOENT ? 0 : r;
-}
-#endif // 0
-
 int cg_create(const char *controller, const char *path) {
         _cleanup_free_ char *fs = NULL;
         int r;
@@ -615,8 +672,6 @@ int cg_create(const char *controller, const char *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 +688,6 @@ 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;
@@ -670,13 +724,15 @@ 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
@@ -691,7 +747,8 @@ 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;
@@ -711,7 +768,7 @@ int cg_set_task_access(
                 gid_t gid) {
 
         _cleanup_free_ char *fs = NULL, *procs = NULL;
-        int r;
+        int r, unified;
 
         assert(path);
 
@@ -729,13 +786,18 @@ int cg_set_task_access(
         if (r < 0)
                 return r;
 
+        unified = cg_unified();
+        if (unified < 0)
+                return unified;
+        if (unified)
+                return 0;
+
         /* Compatibility, Always keep values for "tasks" in sync with
          * "cgroup.procs" */
-        r = cg_get_path(controller, path, "tasks", &procs);
-        if (r < 0)
-                return r;
+        if (cg_get_path(controller, path, "tasks", &procs) >= 0)
+                (void) chmod_and_chown(procs, mode, uid, gid);
 
-        return chmod_and_chown(procs, mode, uid, gid);
+        return 0;
 }
 #endif // 0
 
@@ -743,64 +805,70 @@ 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;
+        size_t cs = 0;
+        int unified;
 
         assert(path);
         assert(pid >= 0);
 
-        if (controller) {
-                if (!cg_controller_is_valid(controller))
-                        return -EINVAL;
+        unified = cg_unified();
+        if (unified < 0)
+                return unified;
+        if (unified == 0) {
+                if (controller) {
+                        if (!cg_controller_is_valid(controller))
+                                return -EINVAL;
+                } else
+                        controller = ELOGIND_CGROUP_CONTROLLER;
 
-                controller = normalize_controller(controller);
-        } else
-                controller = ELOGIND_CGROUP_CONTROLLER;
+                cs = strlen(controller);
+        }
 
         fs = procfs_file_alloca(pid, "cgroup");
-
         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;
-
-                l++;
-                e = strchr(l, ':');
-                if (!e)
-                        continue;
+                if (unified) {
+                        e = startswith(line, "0:");
+                        if (!e)
+                                continue;
 
-                *e = 0;
+                        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;
 
-                FOREACH_WORD_SEPARATOR(word, k, l, ",", state) {
+                        l++;
+                        e = strchr(l, ':');
+                        if (!e)
+                                continue;
 
-                        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, 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;
-
                 p = strdup(e + 1);
                 if (!p)
                         return -ENOMEM;
@@ -809,18 +877,22 @@ 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;
-        int r;
+        const char *sc;
+        int r, unified;
 
         assert(agent);
 
+        unified = cg_unified();
+        if (unified < 0)
+                return unified;
+        if (unified) /* doesn't apply to unified hierarchy */
+                return -EOPNOTSUPP;
+
         r = cg_get_path(controller, NULL, "release_agent", &fs);
         if (r < 0)
                 return r;
@@ -830,21 +902,19 @@ int cg_install_release_agent(const char *controller, const char *agent) {
                 return r;
 
         sc = strstrip(contents);
-        if (sc[0] == 0) {
+        if (isempty(sc)) {
                 r = write_string_file_no_create(fs, agent);
                 if (r < 0)
                         return r;
-        } else if (!streq(sc, agent))
+        } else if (!path_equal(sc, agent))
                 return -EEXIST;
 
-        free(fs);
-        fs = NULL;
+        fs = mfree(fs);
         r = cg_get_path(controller, NULL, "notify_on_release", &fs);
         if (r < 0)
                 return r;
 
-        free(contents);
-        contents = NULL;
+        contents = mfree(contents);
         r = read_one_line_file(fs, &contents);
         if (r < 0)
                 return r;
@@ -866,7 +936,13 @@ int cg_install_release_agent(const char *controller, const char *agent) {
 
 int cg_uninstall_release_agent(const char *controller) {
         _cleanup_free_ char *fs = NULL;
-        int r;
+        int r, unified;
+
+        unified = cg_unified();
+        if (unified < 0)
+                return unified;
+        if (unified) /* Doesn't apply to unified hierarchy */
+                return -EOPNOTSUPP;
 
         r = cg_get_path(controller, NULL, "notify_on_release", &fs);
         if (r < 0)
@@ -876,8 +952,7 @@ int cg_uninstall_release_agent(const char *controller) {
         if (r < 0)
                 return r;
 
-        free(fs);
-        fs = NULL;
+        fs = mfree(fs);
 
         r = cg_get_path(controller, NULL, "release_agent", &fs);
         if (r < 0)
@@ -889,51 +964,70 @@ int cg_uninstall_release_agent(const char *controller) {
 
         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;
+
+        r = cg_read_pid(f, &pid);
         if (r < 0)
-                return r == -ENOENT ? 1 : r;
+                return r;
 
-        self_pid = getpid();
+        return r == 0;
+}
 
-        while ((r = cg_read_pid(f, &pid)) > 0) {
+int cg_is_empty_recursive(const char *controller, const char *path) {
+        int unified, r;
 
-                if (ignore_self && pid == self_pid)
-                        continue;
+        assert(path);
 
-                found = true;
-                break;
-        }
+        /* The root cgroup is always populated */
+        if (controller && (isempty(path) || path_equal(path, "/")))
+                return false;
+
+        unified = cg_unified();
+        if (unified < 0)
+                return unified;
+
+        if (unified > 0) {
+                _cleanup_free_ char *populated = NULL, *t = NULL;
 
+                /* On the unified hierarchy we can check empty state
+                 * via the "cgroup.populated" attribute. */
+
+                r = cg_get_path(controller, path, "cgroup.populated", &populated);
         if (r < 0)
                 return r;
 
-        return !found;
-}
+                r = read_one_line_file(populated, &t);
+                if (r == -ENOENT)
+                        return 1;
+                if (r < 0)
+                        return r;
 
-int cg_is_empty_recursive(const char *controller, const char *path, bool ignore_self) {
+                return streq(t, "0");
+        } else {
         _cleanup_closedir_ DIR *d = NULL;
         char *fn;
-        int r;
 
-        assert(path);
-
-        r = cg_is_empty(controller, path, ignore_self);
+                r = cg_is_empty(controller, path);
         if (r <= 0)
                 return r;
 
         r = cg_enumerate_subgroups(controller, path, &d);
+                if (r == -ENOENT)
+                        return 1;
         if (r < 0)
-                return r == -ENOENT ? 1 : r;
+                        return r;
 
         while ((r = cg_read_subgroup(d, &fn)) > 0) {
                 _cleanup_free_ char *p = NULL;
@@ -943,21 +1037,20 @@ int cg_is_empty_recursive(const char *controller, const char *path, bool ignore_
                 if (!p)
                         return -ENOMEM;
 
-                r = cg_is_empty_recursive(controller, p, ignore_self);
+                        r = cg_is_empty_recursive(controller, p);
                 if (r <= 0)
                         return r;
         }
-
         if (r < 0)
                 return r;
 
-        return 1;
+                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);
 
@@ -985,7 +1078,7 @@ 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;
 
@@ -998,10 +1091,7 @@ 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)) {
@@ -1009,13 +1099,9 @@ 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);
@@ -1073,6 +1159,8 @@ int cg_mangle_path(const char *path, char **result) {
 }
 
 int cg_get_root_path(char **path) {
+/// elogind does not support systemd scopes and slices
+#if 0
         char *p, *e;
         int r;
 
@@ -1082,12 +1170,20 @@ int cg_get_root_path(char **path) {
         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(ELOGIND_CGROUP_CONTROLLER, 1, path);
+#endif // 0
 }
 
 int cg_shift_path(const char *cgroup, const char *root, const char **shifted) {
@@ -1107,10 +1203,11 @@ 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 (p && p[0] && (p > cgroup))
                 *shifted = p - 1;
         else
                 *shifted = cgroup;
@@ -1130,6 +1227,8 @@ int cg_pid_get_path_shifted(pid_t pid, const char *root, char **cgroup) {
         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;
@@ -1146,10 +1245,13 @@ 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;
 }
 
+/// UNNEEDED by elogind
+#if 0
 int cg_path_decode_unit(const char *cgroup, char **unit){
         char *c, *s;
         size_t n;
@@ -1360,8 +1462,6 @@ 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;
@@ -1374,25 +1474,20 @@ 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;
@@ -1408,6 +1503,9 @@ 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
         _cleanup_free_ char *unit = NULL;
         char *start, *end;
         int r;
@@ -1428,10 +1526,29 @@ int cg_path_get_session(const char *path, char **session) {
         *end = 0;
         if (!session_id_valid(start))
                 return -ENXIO;
+#else
+        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;
@@ -1453,6 +1570,8 @@ int cg_pid_get_session(pid_t pid, char **session) {
         return cg_path_get_session(cgroup, session);
 }
 
+/// UNNEEDED by elogind
+#if 0
 int cg_path_get_owner_uid(const char *path, uid_t *uid) {
         _cleanup_free_ char *slice = NULL;
         char *start, *end;
@@ -1478,8 +1597,6 @@ 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;
@@ -1490,7 +1607,6 @@ 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;
@@ -1567,6 +1683,7 @@ 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;
@@ -1586,30 +1703,37 @@ 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 (l != strlen(n))
+                                        continue;
+
+                                if (memcmp(p, n, l) != 0)
+                                        continue;
 
-                                if (check_hierarchy(n) >= 0)
                                         need_prefix = true;
+                                break;
                         }
                 }
         }
 
         if (need_prefix)
                 return strappend("_", p);
-        else
+
                 return strdup(p);
 }
 
@@ -1722,6 +1846,7 @@ 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;
@@ -1734,6 +1859,8 @@ int cg_set_attribute(const char *controller, const char *path, const char *attri
         return write_string_file_no_create(p, value);
 }
 
+/// 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;
@@ -1744,21 +1871,10 @@ 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;
-        int r;
+int cg_create_everywhere(CGroupMask supported, CGroupMask mask, const char *path) {
+        CGroupController c;
+        int r, unified;
 
         /* This one will create a cgroup in our private tree, but also
          * duplicate it in the trees specified in mask, and remove it
@@ -1769,44 +1885,58 @@ int cg_create_everywhere(CGroupControllerMask supported, CGroupControllerMask ma
         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 */
+        unified = cg_unified();
+        if (unified < 0)
+                return unified;
+        if (unified > 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 r;
+int cg_attach_everywhere(CGroupMask supported, const char *path, pid_t pid, cg_migrate_callback_t path_callback, void *userdata) {
+        CGroupController c;
+        int r, unified;
 
         r = cg_attach(ELOGIND_CGROUP_CONTROLLER, path, pid);
         if (r < 0)
                 return r;
 
-        NULSTR_FOREACH(n, mask_names) {
+        unified = cg_unified();
+        if (unified < 0)
+                return unified;
+        if (unified > 0)
+                return 0;
 
-                if (supported & bit) {
+        for (c = 0; c < _CGROUP_CONTROLLER_MAX; c++) {
+                CGroupMask bit = CGROUP_CONTROLLER_TO_MASK(c);
                         const char *p = NULL;
 
+                if (!(supported & bit))
+                        continue;
+
                         if (path_callback)
                                 p = path_callback(bit, userdata);
 
                         if (!p)
                                 p = path;
 
-                        cg_attach_fallback(n, p, pid);
-                }
-
-                bit <<= 1;
+                (void) cg_attach_fallback(cgroup_controller_to_string(c), p, pid);
         }
 
         return 0;
@@ -1814,27 +1944,26 @@ int cg_attach_everywhere(CGroupControllerMask supported, const char *path, pid_t
 
 /// 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, unified;
 
         if (!path_equal(from, to))  {
                 r = cg_migrate_recursive(ELOGIND_CGROUP_CONTROLLER, from, ELOGIND_CGROUP_CONTROLLER, to, false, true);
@@ -1842,58 +1971,132 @@ int cg_migrate_everywhere(CGroupControllerMask supported, const char *from, cons
                         return r;
         }
 
-        NULSTR_FOREACH(n, mask_names) {
-                if (supported & bit) {
+        unified = cg_unified();
+        if (unified < 0)
+                return unified;
+        if (unified > 0)
+                return r;
+
+        for (c = 0; c < _CGROUP_CONTROLLER_MAX; c++) {
+                CGroupMask bit = CGROUP_CONTROLLER_TO_MASK(c);
                         const char *p = NULL;
 
+                if (!(supported & bit))
+                        continue;
+
                         if (to_callback)
                                 p = to_callback(bit, userdata);
 
                         if (!p)
                                 p = to;
 
-                        cg_migrate_recursive_fallback(ELOGIND_CGROUP_CONTROLLER, to, n, p, false, false);
-                }
-
-                bit <<= 1;
+                (void) cg_migrate_recursive_fallback(ELOGIND_CGROUP_CONTROLLER, to, cgroup_controller_to_string(c), p, false, false);
         }
 
         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, unified;
 
         r = cg_trim(ELOGIND_CGROUP_CONTROLLER, path, delete_root);
         if (r < 0)
                 return r;
 
-        NULSTR_FOREACH(n, mask_names) {
-                if (supported & bit)
-                        cg_trim(n, path, delete_root);
+        unified = cg_unified();
+        if (unified < 0)
+                return unified;
+        if (unified > 0)
+                return r;
+
+        for (c = 0; c < _CGROUP_CONTROLLER_MAX; c++) {
+                CGroupMask bit = CGROUP_CONTROLLER_TO_MASK(c);
 
-                bit <<= 1;
+                if (!(supported & bit))
+                        continue;
+
+                (void) cg_trim(cgroup_controller_to_string(c), path, delete_root);
         }
 
         return 0;
 }
+#endif // 0
+
+int cg_mask_supported(CGroupMask *ret) {
+        CGroupMask mask = 0;
+        int r, unified;
+
+        /* Determines the mask of supported cgroup controllers. Only
+         * includes controllers we can make sense of and that are
+         * actually accessible. */
 
-CGroupControllerMask cg_mask_supported(void) {
-        CGroupControllerMask bit = 1, mask = 0;
-        const char *n;
+        unified = cg_unified();
+        if (unified < 0)
+                return unified;
+        if (unified > 0) {
+                _cleanup_free_ char *root = NULL, *controllers = NULL, *path = NULL;
+                const char *c;
 
-        NULSTR_FOREACH(n, mask_names) {
-                if (check_hierarchy(n) >= 0)
-                        mask |= bit;
+                /* In the unified hierarchy we can read the supported
+                 * and accessible controllers from a the top-level
+                 * cgroup attribute */
 
-                bit <<= 1;
+                r = cg_get_root_path(&root);
+                if (r < 0)
+                        return r;
+
+                r = cg_get_path(ELOGIND_CGROUP_CONTROLLER, root, "cgroup.controllers", &path);
+                if (r < 0)
+                        return r;
+
+                r = read_one_line_file(path, &controllers);
+                if (r < 0)
+                        return r;
+
+                c = controllers;
+                for (;;) {
+                        _cleanup_free_ char *n = NULL;
+                        CGroupController v;
+
+                        r = extract_first_word(&c, &n, NULL, 0);
+                        if (r < 0)
+                                return r;
+                        if (r == 0)
+                                break;
+
+                        v = cgroup_controller_from_string(n);
+                        if (v < 0)
+                                continue;
+
+                        mask |= CGROUP_CONTROLLER_TO_MASK(v);
+        }
+
+                /* Currently, we only support the memory controller in
+                 * the unified hierarchy, mask everything else off. */
+                mask &= CGROUP_MASK_MEMORY;
+
+        } 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;
 }
 
+/// UNNEEDED by elogind
+#if 0
 int cg_kernel_controllers(Set *controllers) {
         _cleanup_fclose_ FILE *f = NULL;
         char buf[LINE_MAX];
@@ -1901,6 +2104,11 @@ 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)
@@ -1921,7 +2129,7 @@ int cg_kernel_controllers(Set *controllers) {
                         if (feof(f))
                                 break;
 
-                        if (ferror(f) && errno)
+                        if (ferror(f) && errno != 0)
                                 return -errno;
 
                         return -EBADMSG;
@@ -1932,7 +2140,7 @@ int cg_kernel_controllers(Set *controllers) {
                         continue;
                 }
 
-                if (!filename_is_valid(controller)) {
+                if (!cg_controller_is_valid(controller)) {
                         free(controller);
                         return -EBADMSG;
                 }
@@ -1945,3 +2153,137 @@ int cg_kernel_controllers(Set *controllers) {
         return 0;
 }
 #endif // 0
+
+static thread_local int unified_cache = -1;
+
+int cg_unified(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 >= 0)
+                return unified_cache;
+
+        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))
+#else
+        if (F_TYPE_EQUAL(fs.f_type, TMPFS_MAGIC))
+#endif // elogind
+                unified_cache = false;
+        else
+                return -ENOEXEC;
+
+        return unified_cache;
+}
+
+/// UNNEEDED by elogind
+#if 0
+void cg_unified_flush(void) {
+        unified_cache = -1;
+}
+
+int cg_enable_everywhere(CGroupMask supported, CGroupMask mask, const char *p) {
+        _cleanup_free_ char *fs = NULL;
+        CGroupController c;
+        int r, unified;
+
+        assert(p);
+
+        if (supported == 0)
+                return 0;
+
+        unified = cg_unified();
+        if (unified < 0)
+                return unified;
+        if (!unified) /* on the legacy hiearchy there's no joining of controllers defined */
+                return 0;
+
+        r = cg_get_path(ELOGIND_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;
+}
+#endif // 0
+
+bool cg_is_unified_wanted(void) {
+        static thread_local int wanted = -1;
+        int r, unified;
+
+        /* If the hierarchy is already mounted, then follow whatever
+         * was chosen for it. */
+        unified = cg_unified();
+        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.unified_cgroup_hierarchy", NULL);
+        if (r > 0)
+                return (wanted = true);
+        else {
+                _cleanup_free_ char *value = NULL;
+
+                r = get_proc_cmdline_key("systemd.unified_cgroup_hierarchy=", &value);
+                if (r < 0)
+                        return false;
+                if (r == 0)
+                        return (wanted = false);
+
+                return (wanted = parse_boolean(value) > 0);
+        }
+}
+
+/// UNNEEDED by elogind
+#if 0
+bool cg_is_legacy_wanted(void) {
+        return !cg_is_unified_wanted();
+}
+#endif // 0
+
+static const char *cgroup_controller_table[_CGROUP_CONTROLLER_MAX] = {
+        [CGROUP_CONTROLLER_CPU] = "cpu",
+        [CGROUP_CONTROLLER_CPUACCT] = "cpuacct",
+        [CGROUP_CONTROLLER_BLKIO] = "blkio",
+        [CGROUP_CONTROLLER_MEMORY] = "memory",
+        [CGROUP_CONTROLLER_DEVICE] = "devices",
+};
+
+DEFINE_STRING_TABLE_LOOKUP(cgroup_controller, CGroupController);