cgroup: reimplement the last bit of libcgroup functionality natively

12 files changed, 538 insertions, 382 deletions

diff --git a/src/cgroup-show.c b/src/cgroup-show.c
index 17183ce554..24d2ba398d 100644
--- a/src/cgroup-show.c
+++ b/src/cgroup-show.c
@@ -39,20 +39,6 @@ static int compare(const void *a, const void *b) {
         return 0;
 }
 
-static char *get_cgroup_path(const char *name) {
-
-        if (!name)
-                return strdup("/cgroup/systemd");
-
-        if (startswith(name, SYSTEMD_CGROUP_CONTROLLER ":"))
-                name += sizeof(SYSTEMD_CGROUP_CONTROLLER);
-
-        if (path_startswith(name, "/cgroup"))
-                return strdup(name);
-
-        return strappend("/cgroup/systemd/", name);
-}
-
 static unsigned ilog10(unsigned long ul) {
         int n = 0;
 
@@ -64,7 +50,7 @@ static unsigned ilog10(unsigned long ul) {
         return n;
 }
 
-static int show_cgroup_full(const char *path, const char *prefix, unsigned n_columns, bool more) {
+static int show_cgroup_one_by_path(const char *path, const char *prefix, unsigned n_columns, bool more) {
         char *fn;
         FILE *f;
         size_t n = 0, n_allocated = 0;
@@ -79,8 +65,8 @@ static int show_cgroup_full(const char *path, const char *prefix, unsigned n_col
         if (!prefix)
                 prefix = "";
 
-        if (!(p = get_cgroup_path(path)))
-                return -ENOMEM;
+        if ((r = cg_fix_path(path, &p)) < 0)
+                return r;
 
         r = asprintf(&fn, "%s/cgroup.procs", p);
         free(p);
@@ -171,15 +157,10 @@ finish:
         return r;
 }
 
-int show_cgroup(const char *path, const char *prefix, unsigned n_columns) {
-        return show_cgroup_full(path, prefix, n_columns, false);
-}
-
-int show_cgroup_recursive(const char *path, const char *prefix, unsigned n_columns) {
+int show_cgroup_by_path(const char *path, const char *prefix, unsigned n_columns) {
         DIR *d;
         char *last = NULL;
-        char *p1 = NULL, *p2 = NULL, *fn = NULL;
-        struct dirent *de;
+        char *p1 = NULL, *p2 = NULL, *fn = NULL, *gn = NULL;
         bool shown_pids = false;
         int r;
 
@@ -189,24 +170,18 @@ int show_cgroup_recursive(const char *path, const char *prefix, unsigned n_colum
         if (!prefix)
                 prefix = "";
 
-        if (!(fn = get_cgroup_path(path)))
-                return -ENOMEM;
+        if ((r = cg_fix_path(path, &fn)) < 0)
+                return r;
 
         if (!(d = opendir(fn))) {
                 free(fn);
                 return -errno;
         }
 
-        while ((de = readdir(d))) {
-
-                if (de->d_type != DT_DIR)
-                        continue;
-
-                if (ignore_file(de->d_name))
-                        continue;
+        while ((r = cg_read_subgroup(d, &gn)) > 0) {
 
                 if (!shown_pids) {
-                        show_cgroup_full(path, prefix, n_columns, true);
+                        show_cgroup_one_by_path(path, prefix, n_columns, true);
                         shown_pids = true;
                 }
 
@@ -219,20 +194,26 @@ int show_cgroup_recursive(const char *path, const char *prefix, unsigned n_colum
                                         goto finish;
                                 }
 
-                        show_cgroup_recursive(last, p1, n_columns-2);
+                        show_cgroup_by_path(last, p1, n_columns-2);
 
                         free(last);
                         last = NULL;
                 }
 
-                if (asprintf(&last, "%s/%s", strempty(path), de->d_name) < 0) {
+                r = asprintf(&last, "%s/%s", fn, gn);
+                free(gn);
+
+                if (r < 0) {
                         r = -ENOMEM;
                         goto finish;
                 }
         }
 
+        if (r < 0)
+                goto finish;
+
         if (!shown_pids)
-                show_cgroup_full(path, prefix, n_columns, !!last);
+                show_cgroup_one_by_path(path, prefix, n_columns, !!last);
 
         if (last) {
                 printf("%s\342\224\224 %s\n", prefix, file_name_from_path(last));
@@ -243,7 +224,7 @@ int show_cgroup_recursive(const char *path, const char *prefix, unsigned n_colum
                                 goto finish;
                         }
 
-                show_cgroup_recursive(last, p2, n_columns-2);
+                show_cgroup_by_path(last, p2, n_columns-2);
         }
 
         r = 0;
@@ -258,3 +239,19 @@ finish:
 
         return r;
 }
+
+int show_cgroup(const char *controller, const char *path, const char *prefix, unsigned n_columns) {
+        char *p;
+        int r;
+
+        assert(controller);
+        assert(path);
+
+        if ((r = cg_get_path(controller, path, NULL, &p)) < 0)
+                return r;
+
+        r = show_cgroup_by_path(p, prefix, n_columns);
+        free(p);
+
+        return r;
+}
diff --git a/src/cgroup-show.h b/src/cgroup-show.h
index 09a436a3c0..da794dee18 100644
--- a/src/cgroup-show.h
+++ b/src/cgroup-show.h
@@ -22,7 +22,7 @@
   along with systemd; If not, see <http://www.gnu.org/licenses/>.
 ***/
 
-int show_cgroup(const char *path, const char *prefix, unsigned columns);
-int show_cgroup_recursive(const char *path, const char *prefix, unsigned columns);
+int show_cgroup_by_path(const char *path, const char *prefix, unsigned columns);
+int show_cgroup(const char *controller, const char *path, const char *prefix, unsigned columns);
 
 #endif
diff --git a/src/cgroup-util.c b/src/cgroup-util.c
index e38d6f5c10..26e8f691d3 100644
--- a/src/cgroup-util.c
+++ b/src/cgroup-util.c
@@ -24,8 +24,7 @@
 #include <signal.h>
 #include <string.h>
 #include <stdlib.h>
-
-#include <libcgroup.h>
+#include <dirent.h>
 
 #include "cgroup-util.h"
 #include "log.h"
@@ -33,54 +32,6 @@
 #include "macro.h"
 #include "util.h"
 
-/*
-   Currently, the only remaining functionality from libcgroup we call
-   here is:
-
-   - cgroup_walk_tree_begin()/cgroup_walk_tree_next()
-   - cgroup_delete_cgroup_ext()
- */
-
-int cg_translate_error(int error, int _errno) {
-
-        switch (error) {
-
-        case ECGROUPNOTCOMPILED:
-        case ECGROUPNOTMOUNTED:
-        case ECGROUPNOTEXIST:
-        case ECGROUPNOTCREATED:
-                return -ENOENT;
-
-        case ECGINVAL:
-                return -EINVAL;
-
-        case ECGROUPNOTALLOWED:
-                return -EPERM;
-
-        case ECGOTHER:
-                return -_errno;
-        }
-
-        return -EIO;
-}
-
-static struct cgroup* cg_new(const char *controller, const char *path) {
-        struct cgroup *cgroup;
-
-        assert(path);
-        assert(controller);
-
-        if (!(cgroup = cgroup_new_cgroup(path)))
-                return NULL;
-
-        if (!cgroup_add_controller(cgroup, controller)) {
-                cgroup_free(&cgroup);
-                return NULL;
-        }
-
-        return cgroup;
-}
-
 int cg_enumerate_processes(const char *controller, const char *path, FILE **_f) {
         char *fs;
         int r;
@@ -147,11 +98,77 @@ int cg_read_pid(FILE *f, pid_t *_pid) {
         return 1;
 }
 
+int cg_enumerate_subgroups(const char *controller, const char *path, DIR **_d) {
+        char *fs;
+        int r;
+        DIR *d;
+
+        assert(controller);
+        assert(path);
+        assert(_d);
+
+        /* This is not recursive! */
+
+        if ((r = cg_get_path(controller, path, NULL, &fs)) < 0)
+                return r;
+
+        d = opendir(fs);
+        free(fs);
+
+        if (!d)
+                return -errno;
+
+        *_d = d;
+        return 0;
+}
+
+int cg_read_subgroup(DIR *d, char **fn) {
+        struct dirent *de;
+
+        assert(d);
+
+        errno = 0;
+        while ((de = readdir(d))) {
+                char *b;
+
+                if (de->d_type != DT_DIR)
+                        continue;
+
+                if (streq(de->d_name, ".") ||
+                    streq(de->d_name, ".."))
+                        continue;
+
+                if (!(b = strdup(de->d_name)))
+                        return -ENOMEM;
+
+                *fn = b;
+                return 1;
+        }
+
+        if (errno)
+                return -errno;
+
+        return 0;
+}
+
+int cg_rmdir(const char *controller, const char *path) {
+        char *p;
+        int r;
+
+        if ((r = cg_get_path(controller, path, NULL, &p)) < 0)
+                return r;
+
+        r = rmdir(p);
+        free(p);
+
+        return r < 0 ? -errno : 0;
+}
+
 int cg_kill(const char *controller, const char *path, int sig, bool ignore_self) {
-        bool killed = false, done = false;
+        bool done = false;
         Set *s;
-        pid_t my_pid;
         int r, ret = 0;
+        pid_t my_pid;
         FILE *f = NULL;
 
         assert(controller);
@@ -171,8 +188,12 @@ int cg_kill(const char *controller, const char *path, int sig, bool ignore_self)
                 pid_t pid;
                 done = true;
 
-                if ((r = cg_enumerate_processes(controller, path, &f)) < 0)
+                if ((r = cg_enumerate_processes(controller, path, &f)) < 0) {
+                        if (ret >= 0)
+                                ret = r;
+
                         goto finish;
+                }
 
                 while ((r = cg_read_pid(f, &pid)) > 0) {
 
@@ -185,15 +206,26 @@ int cg_kill(const char *controller, const char *path, int sig, bool ignore_self)
                         /* If we haven't killed this process yet, kill
                          * it */
                         if (kill(pid, sig) < 0 && errno != ESRCH) {
-                                if (ret == 0)
+                                if (ret >= 0)
                                         ret = -errno;
-                        }
+                        } else if (ret == 0)
+                                ret = 1;
 
-                        killed = true;
                         done = false;
 
-                        if ((r = set_put(s, LONG_TO_PTR(pid))) < 0)
-                                break;
+                        if ((r = set_put(s, LONG_TO_PTR(pid))) < 0) {
+                                if (ret >= 0)
+                                        ret = r;
+
+                                goto finish;
+                        }
+                }
+
+                if (r < 0) {
+                        if (ret >= 0)
+                                ret = r;
+
+                        goto finish;
                 }
 
                 fclose(f);
@@ -203,7 +235,7 @@ int cg_kill(const char *controller, const char *path, int sig, bool ignore_self)
                  * quicker than we can kill them we repeat this until
                  * no new pids need to be killed. */
 
-        } while (!done && r >= 0);
+        } while (!done);
 
 finish:
         set_free(s);
@@ -211,69 +243,64 @@ finish:
         if (f)
                 fclose(f);
 
-        if (r < 0)
-                return r;
-
-        if (ret < 0)
-                return ret;
-
-        return !!killed;
+        return ret;
 }
 
-int cg_kill_recursive(const char *controller, const char *path, int sig, bool ignore_self) {
-        struct cgroup_file_info info;
-        int level = 0, r, ret = 0;
-        void *iterator = NULL;
-        bool killed = false;
+int cg_kill_recursive(const char *controller, const char *path, int sig, bool ignore_self, bool rem) {
+        int r, ret = 0;
+        DIR *d = NULL;
+        char *fn;
 
         assert(path);
         assert(controller);
         assert(sig >= 0);
 
-        zero(info);
+        ret = cg_kill(controller, path, sig, ignore_self);
 
-        r = cgroup_walk_tree_begin(controller, path, 0, &iterator, &info, &level);
-        while (r == 0) {
-                int k;
-                char *p;
+        if ((r = cg_enumerate_subgroups(controller, path, &d)) < 0) {
+                if (ret >= 0)
+                        ret = r;
 
-                if (info.type != CGROUP_FILE_TYPE_DIR)
-                        goto next;
+                goto finish;
+        }
 
-                if (asprintf(&p, "%s/%s", path, info.path) < 0) {
-                        ret = -ENOMEM;
-                        break;
-                }
+        while ((r = cg_read_subgroup(d, &fn)) > 0) {
+                char *p = NULL;
 
-                k = cg_kill(controller, p, sig, ignore_self);
-                free(p);
+                r = asprintf(&p, "%s/%s", path, fn);
+                free(fn);
+
+                if (r < 0) {
+                        if (ret >= 0)
+                                ret = -ENOMEM;
 
-                if (k < 0) {
-                        if (ret == 0)
-                                ret = k;
-                } else if (k > 0)
-                        killed = true;
+                        goto finish;
+                }
 
-        next:
+                r = cg_kill_recursive(controller, p, sig, ignore_self, rem);
+                free(p);
 
-                r = cgroup_walk_tree_next(0, &iterator, &info, level);
+                if (r != 0 && ret >= 0)
+                        ret = r;
         }
 
-        if (ret == 0) {
-                if (r == 0 || r == ECGEOF)
-                        ret = !!killed;
-                else if (r == ECGOTHER && errno == ENOENT)
-                        ret = -ESRCH;
-                else
-                        ret = cg_translate_error(r, errno);
-        }
+        if (r < 0 && ret >= 0)
+                ret = r;
+
+        if (rem)
+                if ((r = cg_rmdir(controller, path)) < 0) {
+                        if (ret >= 0)
+                                ret = r;
+                }
 
-        assert_se(cgroup_walk_tree_end(&iterator) == 0);
+finish:
+        if (d)
+                closedir(d);
 
         return ret;
 }
 
-int cg_kill_recursive_and_wait(const char *controller, const char *path) {
+int cg_kill_recursive_and_wait(const char *controller, const char *path, bool rem) {
         unsigned i;
 
         assert(path);
@@ -294,7 +321,7 @@ int cg_kill_recursive_and_wait(const char *controller, const char *path) {
                 else
                         sig = 0;
 
-                if ((r = cg_kill_recursive(controller, path, sig, true)) <= 0)
+                if ((r = cg_kill_recursive(controller, path, sig, true, rem)) <= 0)
                         return r;
 
                 usleep(50 * USEC_PER_MSEC);
@@ -304,7 +331,8 @@ int cg_kill_recursive_and_wait(const char *controller, const char *path) {
 }
 
 int cg_migrate(const char *controller, const char *from, const char *to, bool ignore_self) {
-        bool migrated = false, done = false;
+        bool done = false;
+        Set *s;
         int r, ret = 0;
         pid_t my_pid;
         FILE *f = NULL;
@@ -313,96 +341,119 @@ int cg_migrate(const char *controller, const char *from, const char *to, bool ig
         assert(from);
         assert(to);
 
+        if (!(s = set_new(trivial_hash_func, trivial_compare_func)))
+                return -ENOMEM;
+
         my_pid = getpid();
 
         do {
                 pid_t pid;
                 done = true;
 
-                if ((r = cg_enumerate_tasks(controller, from, &f)) < 0)
+                if ((r = cg_enumerate_tasks(controller, from, &f)) < 0) {
+                        if (ret >= 0)
+                                ret = r;
+
                         goto finish;
+                }
 
                 while ((r = cg_read_pid(f, &pid)) > 0) {
 
+                        /* This might do weird stuff if we aren't a
+                         * single-threaded program. However, we
+                         * luckily know we are not */
                         if (pid == my_pid && ignore_self)
                                 continue;
 
+                        if (set_get(s, LONG_TO_PTR(pid)) == LONG_TO_PTR(pid))
+                                continue;
+
                         if ((r = cg_attach(controller, to, pid)) < 0) {
-                                if (ret == 0)
-                                        ret = -r;
-                        }
+                                if (ret >= 0)
+                                        ret = r;
+                        } else if (ret == 0)
+                                ret = 1;
 
-                        migrated = true;
                         done = false;
+
+                        if ((r = set_put(s, LONG_TO_PTR(pid))) < 0) {
+                                if (ret >= 0)
+                                        ret = r;
+
+                                goto finish;
+                        }
+                }
+
+                if (r < 0) {
+                        if (ret >= 0)
+                                ret = r;
+
+                        goto finish;
                 }
 
                 fclose(f);
                 f = NULL;
 
-        } while (!done && r >= 0);
+        } while (!done);
 
 finish:
+        set_free(s);
 
         if (f)
                 fclose(f);
 
-        if (r < 0)
-                return r;
-
-        if (ret < 0)
-                return ret;
-
-        return !!migrated;
+        return ret;
 }
 
-int cg_migrate_recursive(const char *controller, const char *from, const char *to, bool ignore_self) {
-        struct cgroup_file_info info;
-        int level = 0, r, ret = 0;
-        void *iterator = NULL;
-        bool migrated = false;
+int cg_migrate_recursive(const char *controller, const char *from, const char *to, bool ignore_self, bool rem) {
+        int r, ret = 0;
+        DIR *d = NULL;
+        char *fn;
 
         assert(controller);
         assert(from);
         assert(to);
 
-        zero(info);
+        ret = cg_migrate(controller, from, to, ignore_self);
 
-        r = cgroup_walk_tree_begin(controller, from, 0, &iterator, &info, &level);
-        while (r == 0) {
-                int k;
-                char *p;
+        if ((r = cg_enumerate_subgroups(controller, from, &d)) < 0) {
+                if (ret >= 0)
+                        ret = r;
+                goto finish;
+        }
+
+        while ((r = cg_read_subgroup(d, &fn)) > 0) {
+                char *p = NULL;
 
-                if (info.type != CGROUP_FILE_TYPE_DIR)
-                        goto next;
+                r = asprintf(&p, "%s/%s", from, fn);
+                free(fn);
 
-                if (asprintf(&p, "%s/%s", from, info.path) < 0) {
-                        ret = -ENOMEM;
-                        break;
+                if (r < 0) {
+                        if (ret >= 0)
+                                ret = -ENOMEM;
+
+                        goto finish;
                 }
 
-                k = cg_migrate(controller, p, to, ignore_self);
+                r = cg_migrate_recursive(controller, p, to, ignore_self, rem);
                 free(p);
 
-                if (k < 0) {
-                        if (ret == 0)
-                                ret = k;
-                } else if (k > 0)
-                        migrated = true;
-
-        next:
-                r = cgroup_walk_tree_next(0, &iterator, &info, level);
+                if (r != 0 && ret >= 0)
+                        ret = r;
         }
 
-        if (ret == 0) {
-                if (r == 0 || r == ECGEOF)
-                        r = !!migrated;
-                else if (r == ECGOTHER && errno == ENOENT)
-                        r = -ESRCH;
-                else
-                        r = cg_translate_error(r, errno);
-        }
+        if (r < 0 && ret >= 0)
+                ret = r;
+
+        if (rem)
+                if ((r = cg_rmdir(controller, from)) < 0) {
+                        if (ret >= 0)
+                                ret = r;
+                }
 
-        assert_se(cgroup_walk_tree_end(&iterator) == 0);
+finish:
+        if (d)
+                closedir(d);
 
         return ret;
 }
@@ -470,24 +521,17 @@ int cg_trim(const char *controller, const char *path, bool delete_root) {
 }
 
 int cg_delete(const char *controller, const char *path) {
-        struct cgroup *cg;
+        char *parent;
         int r;
 
         assert(controller);
         assert(path);
 
-        if (!(cg = cg_new(controller, path)))
-                return -ENOMEM;
-
-        if ((r = cgroup_delete_cgroup_ext(cg, CGFLAG_DELETE_RECURSIVE|CGFLAG_DELETE_IGNORE_MIGRATION)) != 0) {
-                r = cg_translate_error(r, errno);
-                goto finish;
-        }
-
-        r = 0;
+        if ((r = parent_of_path(path, &parent)) < 0)
+                return r;
 
-finish:
-        cgroup_free(&cg);
+        r = cg_migrate_recursive(controller, path, parent, false, true);
+        free(parent);
 
         return r;
 }
@@ -739,63 +783,154 @@ int cg_is_empty(const char *controller, const char *path, bool ignore_self) {
 }
 
 int cg_is_empty_recursive(const char *controller, const char *path, bool ignore_self) {
-        struct cgroup_file_info info;
-        int level = 0, r, ret = 0;
-        void *iterator = NULL;
-        bool empty = true;
+        int r;
+        DIR *d = NULL;
+        char *fn;
 
         assert(controller);
         assert(path);
 
-        zero(info);
+        if ((r = cg_is_empty(controller, path, ignore_self)) <= 0)
+                return r;
+
+        if ((r = cg_enumerate_subgroups(controller, path, &d)) < 0)
+                return r;
 
-        r = cgroup_walk_tree_begin(controller, path, 0, &iterator, &info, &level);
-        while (r == 0) {
-                int k;
-                char *p;
+        while ((r = cg_read_subgroup(d, &fn)) > 0) {
+                char *p = NULL;
 
-                if (info.type != CGROUP_FILE_TYPE_DIR)
-                        goto next;
+                r = asprintf(&p, "%s/%s", path, fn);
+                free(fn);
 
-                if (asprintf(&p, "%s/%s", path, info.path) < 0) {
-                        ret = -ENOMEM;
-                        break;
+                if (r < 0) {
+                        r = -ENOMEM;
+                        goto finish;
                 }
 
-                k = cg_is_empty(controller, p, ignore_self);
+                r = cg_is_empty_recursive(controller, p, ignore_self);
                 free(p);
 
-                if (k < 0) {
-                        ret = k;
-                        break;
-                } else if (k == 0) {
-                        empty = false;
-                        break;
+                if (r <= 0)
+                        goto finish;
+        }
+
+        if (r >= 0)
+                r = 1;
+
+finish:
+
+        if (d)
+                closedir(d);
+
+        return r;
+}
+
+int cg_split_spec(const char *spec, char **controller, char **path) {
+        const char *e;
+        char *t = NULL, *u = NULL;
+
+        assert(spec);
+        assert(controller || path);
+
+        if (*spec == '/') {
+
+                if (path) {
+                        if (!(t = strdup(spec)))
+                                return -ENOMEM;
+
+                        *path = t;
                 }
 
-        next:
-                r = cgroup_walk_tree_next(0, &iterator, &info, level);
+                if (controller)
+                        *controller = NULL;
+
+                return 0;
         }
 
-        if (ret == 0) {
-                if (r == 0 || r == ECGEOF)
-                        ret = !!empty;
-                else if (r == ECGOTHER && errno == ENOENT)
-                        ret = -ESRCH;
-                else
-                        ret = cg_translate_error(r, errno);
+        if (!(e = strchr(spec, ':'))) {
+
+                if (strchr(spec, '/') || spec[0] == 0)
+                        return -EINVAL;
+
+                if (controller) {
+                        if (!(t = strdup(spec)))
+                                return -ENOMEM;
+
+                        *controller = t;
+                }
+
+                if (path)
+                        *path = NULL;
+
+                return 0;
         }
 
-        assert_se(cgroup_walk_tree_end(&iterator) == 0);
+        if (e[1] != '/' ||
+            e == spec ||
+            memchr(spec, '/', e-spec))
+                return -EINVAL;
 
-        return ret;
+        if (controller)
+                if (!(t = strndup(spec, e-spec)))
+                        return -ENOMEM;
+
+        if (path)
+                if (!(u = strdup(e+1))) {
+                        free(t);
+                        return -ENOMEM;
+                }
+
+        if (controller)
+                *controller = t;
+
+        if (path)
+                *path = u;
+
+        return 0;
 }
 
-int cg_init(void) {
-        int r;
+int cg_join_spec(const char *controller, const char *path, char **spec) {
+        assert(controller);
+        assert(path);
 
-        if ((r = cgroup_init()) != 0)
-                return cg_translate_error(r, errno);
+        if (!path_is_absolute(path) ||
+            controller[0] == 0 ||
+            strchr(controller, ':') ||
+            strchr(controller, '/'))
+                return -EINVAL;
+
+        if (asprintf(spec, "%s:%s", controller, path) < 0)
+                return -ENOMEM;
 
         return 0;
 }
+
+int cg_fix_path(const char *path, char **result) {
+        char *t, *c, *p;
+        int r;
+
+        assert(path);
+        assert(result);
+
+        /* First check if it already is a filesystem path */
+        if (path_is_absolute(path) &&
+            path_startswith(path, "/cgroup") &&
+            access(path, F_OK) >= 0) {
+
+                if (!(t = strdup(path)))
+                        return -ENOMEM;
+
+                *result = t;
+                return 0;
+        }
+
+        /* Otherwise treat it as cg spec */
+        if ((r = cg_split_spec(path, &c, &p)) < 0)
+                return r;
+
+        r = cg_get_path(c ? c : SYSTEMD_CGROUP_CONTROLLER, p ? p : "/", NULL, result);
+        free(c);
+        free(p);
+
+        return r;
+}
diff --git a/src/cgroup-util.h b/src/cgroup-util.h
index 716c635a95..8f96571243 100644
--- a/src/cgroup-util.h
+++ b/src/cgroup-util.h
@@ -24,29 +24,36 @@
 
 #include <sys/types.h>
 #include <stdio.h>
+#include <dirent.h>
 
 #include "set.h"
 
 #define SYSTEMD_CGROUP_CONTROLLER "name=systemd"
 
-int cg_translate_error(int error, int _errno);
-
 int cg_enumerate_processes(const char *controller, const char *path, FILE **_f);
 int cg_enumerate_tasks(const char *controller, const char *path, FILE **_f);
-
 int cg_read_pid(FILE *f, pid_t *_pid);
 
+int cg_enumerate_subgroups(const char *controller, const char *path, DIR **_d);
+int cg_read_subgroup(DIR *d, char **fn);
+
 int cg_kill(const char *controller, const char *path, int sig, bool ignore_self);
-int cg_kill_recursive(const char *controller, const char *path, int sig, bool ignore_self);
-int cg_kill_recursive_and_wait(const char *controller, const char *path);
+int cg_kill_recursive(const char *controller, const char *path, int sig, bool ignore_self, bool remove);
+int cg_kill_recursive_and_wait(const char *controller, const char *path, bool remove);
 
 int cg_migrate(const char *controller, const char *from, const char *to, bool ignore_self);
-int cg_migrate_recursive(const char *controller, const char *from, const char *to, bool ignore_self);
+int cg_migrate_recursive(const char *controller, const char *from, const char *to, bool ignore_self, bool remove);
+
+int cg_split_spec(const char *spec, char **controller, char **path);
+int cg_join_spec(const char *controller, const char *path, char **spec);
+int cg_fix_path(const char *path, char **result);
 
 int cg_get_path(const char *controller, const char *path, const char *suffix, char **fs);
 int cg_get_by_pid(const char *controller, pid_t pid, char **path);
 
 int cg_trim(const char *controller, const char *path, bool delete_root);
+
+int cg_rmdir(const char *controller, const char *path);
 int cg_delete(const char *controller, const char *path);
 
 int cg_create(const char *controller, const char *path);
@@ -61,6 +68,4 @@ int cg_install_release_agent(const char *controller, const char *agent);
 int cg_is_empty(const char *controller, const char *path, bool ignore_self);
 int cg_is_empty_recursive(const char *controller, const char *path, bool ignore_self);
 
-int cg_init(void);
-
 #endif
diff --git a/src/cgroup.c b/src/cgroup.c
index 7397b52981..02c2919187 100644
--- a/src/cgroup.c
+++ b/src/cgroup.c
@@ -149,7 +149,7 @@ int cgroup_bonding_kill(CGroupBonding *b, int sig) {
 
         assert(b->realized);
 
-        return cg_kill_recursive(b->controller, b->path, sig, true);
+        return cg_kill_recursive(b->controller, b->path, sig, true, false);
 }
 
 int cgroup_bonding_kill_list(CGroupBonding *first, int sig) {
@@ -215,13 +215,7 @@ int manager_setup_cgroup(Manager *m) {
 
         assert(m);
 
-        /* 1. Initialize libcg */
-        if ((r = cg_init()) < 0) {
-                log_error("Failed to initialize libcg: %s", strerror(-r));
-                goto finish;
-        }
-
-        /* 2. Determine hierarchy */
+        /* 1. Determine hierarchy */
         if ((r = cg_get_by_pid(SYSTEMD_CGROUP_CONTROLLER, 0, &current)) < 0)
                 goto finish;
 
@@ -243,13 +237,13 @@ int manager_setup_cgroup(Manager *m) {
                 }
         }
 
-        /* 3. Show data */
+        /* 2. Show data */
         if ((r = cg_get_path(SYSTEMD_CGROUP_CONTROLLER, m->cgroup_hierarchy, NULL, &path)) < 0)
                 goto finish;
 
         log_debug("Using cgroup controller " SYSTEMD_CGROUP_CONTROLLER ". File system hierarchy is at %s.", path);
 
-        /* 4. Install agent */
+        /* 3. Install agent */
         if ((r = cg_install_release_agent(SYSTEMD_CGROUP_CONTROLLER, CGROUP_AGENT_PATH)) < 0)
                 log_warning("Failed to install release agent, ignoring: %s", strerror(-r));
         else if (r > 0)
@@ -257,13 +251,13 @@ int manager_setup_cgroup(Manager *m) {
         else
                 log_debug("Release agent already installed.");
 
-        /* 5. Realize the group */
+        /* 4. Realize the group */
         if ((r = cg_create_and_attach(SYSTEMD_CGROUP_CONTROLLER, m->cgroup_hierarchy, 0)) < 0) {
                 log_error("Failed to create root cgroup hierarchy: %s", strerror(-r));
                 goto finish;
         }
 
-        /* 6. And pin it, so that it cannot be unmounted */
+        /* 5. And pin it, so that it cannot be unmounted */
         if (m->pin_cgroupfs_fd >= 0)
                 close_nointr_nofail(m->pin_cgroupfs_fd);
 
diff --git a/src/pam-module.c b/src/pam-module.c
index 1441987fdc..69b9b0ce78 100644
--- a/src/pam-module.c
+++ b/src/pam-module.c
@@ -258,12 +258,6 @@ _public_ PAM_EXTERN int pam_sm_open_session(
         if (sd_booted() <= 0)
                 return PAM_SUCCESS;
 
-        if ((r = cg_init()) < 0) {
-                pam_syslog(handle, LOG_ERR, "libcgroup initialization failed: %s", strerror(-r));
-                r = PAM_SESSION_ERR;
-                goto finish;
-        }
-
         if ((r = get_user_data(handle, &username, &pw)) != PAM_SUCCESS)
                 goto finish;
 
@@ -354,44 +348,29 @@ finish:
 }
 
 static int session_remains(pam_handle_t *handle, const char *user_path) {
-        struct cgroup_file_info info;
-        int level = 0, r;
-        void *iterator = NULL;
+        int r;
         bool remains = false;
+        DIR *d;
+        char *subgroup;
 
-        zero(info);
-
-        r = cgroup_walk_tree_begin(SYSTEMD_CGROUP_CONTROLLER, user_path, 0, &iterator, &info, &level);
-        while (r == 0) {
-
-                if (info.type != CGROUP_FILE_TYPE_DIR)
-                        goto next;
-
-                if (streq(info.path, ""))
-                        goto next;
-
-                if (streq(info.path, "no-session"))
-                        goto next;
+        if ((r = cg_enumerate_subgroups(SYSTEMD_CGROUP_CONTROLLER, user_path, &d)) < 0)
+                return r;
 
-                remains = true;
-                break;
+        while ((r = cg_read_subgroup(d, &subgroup)) > 0) {
 
-        next:
+                remains = !streq(subgroup, "no-session");
+                free(subgroup);
 
-                r = cgroup_walk_tree_next(0, &iterator, &info, level);
+                if (remains)
+                        break;
         }
 
+        closedir(d);
 
-        if (remains)
-                r = 1;
-        else if (r == 0 || r == ECGEOF)
-                r = 0;
-        else
-                r = cg_translate_error(r, errno);
-
-        assert_se(cgroup_walk_tree_end(&iterator) == 0);
+        if (r < 0)
+                return r;
 
-        return r;
+        return !!remains;
 }
 
 _public_ PAM_EXTERN int pam_sm_close_session(
@@ -425,6 +404,10 @@ _public_ PAM_EXTERN int pam_sm_close_session(
                 goto finish;
         }
 
+        /* We are probably still in some session/no-session dir. Move ourselves out of the way as first step */
+        if ((r = cg_attach(SYSTEMD_CGROUP_CONTROLLER, "/user", 0)) < 0)
+                pam_syslog(handle, LOG_ERR, "Failed to move us away: %s", strerror(-r));
+
         if (asprintf(&user_path, "/user/%s", username) < 0) {
                 r = PAM_BUF_ERR;
                 goto finish;
@@ -439,28 +422,19 @@ _public_ PAM_EXTERN int pam_sm_close_session(
                 }
 
                 if (kill_session)  {
-                        /* Kill processes in session cgroup */
-                        if ((r = cg_kill_recursive_and_wait(SYSTEMD_CGROUP_CONTROLLER, session_path)) < 0)
+                        /* Kill processes in session cgroup, and delete it */
+                        if ((r = cg_kill_recursive_and_wait(SYSTEMD_CGROUP_CONTROLLER, session_path, true)) < 0)
                                 pam_syslog(handle, LOG_ERR, "Failed to kill session cgroup: %s", strerror(-r));
-
-                } else  {
+                } else {
                         /* Migrate processes from session to
                          * no-session cgroup. First, try to create the
                          * no-session group in case it doesn't exist
-                         * yet. */
+                         * yet. Also, delete the session group. */
                         create_user_group(handle, nosession_path, pw, 0);
 
-                        if ((r = cg_migrate_recursive(SYSTEMD_CGROUP_CONTROLLER, session_path, nosession_path, false)) < 0)
+                        if ((r = cg_migrate_recursive(SYSTEMD_CGROUP_CONTROLLER, session_path, nosession_path, false, true)) < 0)
                                 pam_syslog(handle, LOG_ERR, "Failed to migrate session cgroup: %s", strerror(-r));
                 }
-
-                /* Delete session cgroup */
-                if (r < 0)
-                        pam_syslog(handle, LOG_INFO, "Couldn't empty session cgroup, not deleting.");
-                else {
-                        if ((r = cg_delete(SYSTEMD_CGROUP_CONTROLLER, session_path)) < 0)
-                                pam_syslog(handle, LOG_ERR, "Failed to delete session cgroup: %s", strerror(-r));
-                }
         }
 
         /* GC user tree */
@@ -473,25 +447,26 @@ _public_ PAM_EXTERN int pam_sm_close_session(
         if (kill_user && r == 0) {
 
                 /* Kill no-session cgroup */
-                if ((r = cg_kill_recursive_and_wait(SYSTEMD_CGROUP_CONTROLLER, user_path)) < 0)
+                if ((r = cg_kill_recursive_and_wait(SYSTEMD_CGROUP_CONTROLLER, user_path, true)) < 0)
                         pam_syslog(handle, LOG_ERR, "Failed to kill user cgroup: %s", strerror(-r));
         } else {
 
                 if ((r = cg_is_empty_recursive(SYSTEMD_CGROUP_CONTROLLER, user_path, true)) < 0)
                         pam_syslog(handle, LOG_ERR, "Failed to check user cgroup: %s", strerror(-r));
 
-                /* If we managed to kill somebody, don't cleanup the cgroup. */
-                if (r == 0)
+                /* Remove user cgroup */
+                if (r > 0) {
+                        if ((r = cg_delete(SYSTEMD_CGROUP_CONTROLLER, user_path)) < 0)
+                                pam_syslog(handle, LOG_ERR, "Failed to delete user cgroup: %s", strerror(-r));
+
+                /* If we managed to find somebody, don't cleanup the cgroup. */
+                } else if (r == 0)
                         r = -EBUSY;
         }
 
         if (r >= 0) {
                 const char *runtime_dir;
 
-                /* Remove user cgroup */
-                if ((r = cg_delete(SYSTEMD_CGROUP_CONTROLLER, user_path)) < 0)
-                        pam_syslog(handle, LOG_ERR, "Failed to delete user cgroup: %s", strerror(-r));
-
                 /* This will migrate us to the /user cgroup. */
 
                 if ((runtime_dir = pam_getenv(handle, "XDG_RUNTIME_DIR")))
diff --git a/src/systemctl.c b/src/systemctl.c
index 45e900c5fe..32974aecce 100644
--- a/src/systemctl.c
+++ b/src/systemctl.c
@@ -1064,7 +1064,6 @@ typedef struct UnitStatusInfo {
 
 static void print_status_info(UnitStatusInfo *i) {
         ExecStatusInfo *p;
-        int r;
 
         assert(i);
 
@@ -1195,10 +1194,7 @@ static void print_status_info(UnitStatusInfo *i) {
                 else
                         c = 0;
 
-                if ((r = cg_init()) < 0)
-                        log_error("Failed to initialize libcg: %s", strerror(-r));
-                else
-                        show_cgroup_recursive(i->default_control_group, "\t\t  ", c);
+                show_cgroup_by_path(i->default_control_group, "\t\t  ", c);
         }
 }
 
diff --git a/src/systemd-cgls.c b/src/systemd-cgls.c
index bd8d263d0e..cd03e91b83 100644
--- a/src/systemd-cgls.c
+++ b/src/systemd-cgls.c
@@ -83,11 +83,6 @@ int main(int argc, char *argv[]) {
                 goto finish;
         }
 
-        if (cg_init() < 0) {
-                log_error("Failed to initialize libcg: %s", strerror(-r));
-                goto finish;
-        }
-
         if (optind < argc) {
                 unsigned i;
 
@@ -95,7 +90,7 @@ int main(int argc, char *argv[]) {
                         int q;
                         printf("%s:\n", argv[i]);
 
-                        if ((q = show_cgroup_recursive(argv[i], NULL, 0)) < 0)
+                        if ((q = show_cgroup_by_path(argv[i], NULL, 0)) < 0)
                                 r = q;
                 }
 
@@ -109,9 +104,9 @@ int main(int argc, char *argv[]) {
 
                 if (path_startswith(p, "/cgroup")) {
                         printf("Working Directory %s:\n", p);
-                        r = show_cgroup_recursive(p, NULL, 0);
+                        r = show_cgroup_by_path(p, NULL, 0);
                 } else
-                        r = show_cgroup_recursive(NULL, NULL, 0);
+                        r = show_cgroup(SYSTEMD_CGROUP_CONTROLLER, "/", NULL, 0);
 
                 free(p);
         }
diff --git a/src/test-cgroup.c b/src/test-cgroup.c
index 45e7f0fe03..4a54b1bc8f 100644
--- a/src/test-cgroup.c
+++ b/src/test-cgroup.c
@@ -28,8 +28,7 @@
 
 int main(int argc, char*argv[]) {
         char *path;
-
-        assert_se(cg_init() >= 0);
+        char *c, *p;
 
         assert_se(cg_create(SYSTEMD_CGROUP_CONTROLLER, "/test-a") == 0);
         assert_se(cg_create(SYSTEMD_CGROUP_CONTROLLER, "/test-a") == 0);
@@ -62,21 +61,44 @@ int main(int argc, char*argv[]) {
         assert_se(cg_is_empty_recursive(SYSTEMD_CGROUP_CONTROLLER, "/test-a", false) > 0);
         assert_se(cg_is_empty_recursive(SYSTEMD_CGROUP_CONTROLLER, "/test-b", false) == 0);
 
-        assert_se(cg_kill_recursive(SYSTEMD_CGROUP_CONTROLLER, "/test-a", 0, false) == 0);
-        assert_se(cg_kill_recursive(SYSTEMD_CGROUP_CONTROLLER, "/test-b", 0, false) > 0);
+        assert_se(cg_kill_recursive(SYSTEMD_CGROUP_CONTROLLER, "/test-a", 0, false, false) == 0);
+        assert_se(cg_kill_recursive(SYSTEMD_CGROUP_CONTROLLER, "/test-b", 0, false, false) > 0);
 
-        assert_se(cg_migrate_recursive(SYSTEMD_CGROUP_CONTROLLER, "/test-b", "/test-a", false) == 0);
+        assert_se(cg_migrate_recursive(SYSTEMD_CGROUP_CONTROLLER, "/test-b", "/test-a", false, false) > 0);
 
         assert_se(cg_is_empty_recursive(SYSTEMD_CGROUP_CONTROLLER, "/test-a", false) == 0);
         assert_se(cg_is_empty_recursive(SYSTEMD_CGROUP_CONTROLLER, "/test-b", false) > 0);
 
-        assert_se(cg_kill_recursive(SYSTEMD_CGROUP_CONTROLLER, "/test-a", 0, false) > 0);
-        assert_se(cg_kill_recursive(SYSTEMD_CGROUP_CONTROLLER, "/test-b", 0, false) == 0);
+        assert_se(cg_kill_recursive(SYSTEMD_CGROUP_CONTROLLER, "/test-a", 0, false, false) > 0);
+        assert_se(cg_kill_recursive(SYSTEMD_CGROUP_CONTROLLER, "/test-b", 0, false, false) == 0);
 
         cg_trim(SYSTEMD_CGROUP_CONTROLLER, "/", false);
 
         assert_se(cg_delete(SYSTEMD_CGROUP_CONTROLLER, "/test-b") < 0);
-        assert_se(cg_delete(SYSTEMD_CGROUP_CONTROLLER, "/test-a") == 0);
+        assert_se(cg_delete(SYSTEMD_CGROUP_CONTROLLER, "/test-a") >= 0);
+
+        assert_se(cg_split_spec("foobar:/", &c, &p) == 0);
+        assert(streq(c, "foobar"));
+        assert(streq(p, "/"));
+        free(c);
+        free(p);
+
+        assert_se(cg_split_spec("foobar:", &c, &p) < 0);
+        assert_se(cg_split_spec("foobar:asdfd", &c, &p) < 0);
+        assert_se(cg_split_spec(":///", &c, &p) < 0);
+        assert_se(cg_split_spec(":", &c, &p) < 0);
+        assert_se(cg_split_spec("", &c, &p) < 0);
+        assert_se(cg_split_spec("fo/obar:/", &c, &p) < 0);
+
+        assert_se(cg_split_spec("/", &c, &p) >= 0);
+        assert(c == NULL);
+        assert(streq(p, "/"));
+        free(p);
+
+        assert_se(cg_split_spec("foo", &c, &p) >= 0);
+        assert(streq(c, "foo"));
+        assert(p == NULL);
+        free(c);
 
         return 0;
 }
diff --git a/src/unit.c b/src/unit.c
index 66372f2a92..f786359fac 100644
--- a/src/unit.c
+++ b/src/unit.c
@@ -1511,12 +1511,9 @@ char *unit_dbus_path(Unit *u) {
         if (!(e = bus_path_escape(u->meta.id)))
                 return NULL;
 
-        if (asprintf(&p, "/org/freedesktop/systemd1/unit/%s", e) < 0) {
-                free(e);
-                return NULL;
-        }
-
+        p = strappend("/org/freedesktop/systemd1/unit/", e);
         free(e);
+
         return p;
 }
 
@@ -1526,8 +1523,13 @@ int unit_add_cgroup(Unit *u, CGroupBonding *b) {
 
         assert(u);
         assert(b);
+
         assert(b->path);
 
+        if (!b->controller)
+                if (!(b->controller = strdup(SYSTEMD_CGROUP_CONTROLLER)))
+                        return -ENOMEM;
+
         /* Ensure this hasn't been added yet */
         assert(!b->unit);
 
@@ -1566,7 +1568,6 @@ static char *default_cgroup_path(Unit *u) {
 }
 
 int unit_add_cgroup_from_text(Unit *u, const char *name) {
-        size_t n;
         char *controller = NULL, *path = NULL;
         CGroupBonding *b = NULL;
         int r;
@@ -1574,38 +1575,20 @@ int unit_add_cgroup_from_text(Unit *u, const char *name) {
         assert(u);
         assert(name);
 
-        /* Detect controller name */
-        n = strcspn(name, ":");
-
-        if (name[n] == 0 ||
-            (name[n] == ':' && name[n+1] == 0)) {
-
-                /* Only controller name, no path? */
-
-                if (!(path = default_cgroup_path(u)))
-                        return -ENOMEM;
-
-        } else {
-                const char *p;
-
-                /* Controller name, and path. */
-                p = name+n+1;
+        if ((r = cg_split_spec(name, &controller, &path)) < 0)
+                return r;
 
-                if (!path_is_absolute(p))
-                        return -EINVAL;
+        if (!path)
+                path = default_cgroup_path(u);
 
-                if (!(path = strdup(p)))
-                        return -ENOMEM;
-        }
-
-        if (n > 0)
-                controller = strndup(name, n);
-        else
+        if (!controller)
                 controller = strdup(SYSTEMD_CGROUP_CONTROLLER);
 
-        if (!controller) {
-                r = -ENOMEM;
-                goto fail;
+        if (!path || !controller) {
+                free(path);
+                free(controller);
+
+                return -ENOMEM;
         }
 
         if (cgroup_bonding_find_list(u->meta.cgroup_bondings, controller)) {
@@ -1650,9 +1633,6 @@ int unit_add_default_cgroup(Unit *u) {
         if (!(b = new0(CGroupBonding, 1)))
                 return -ENOMEM;
 
-        if (!(b->controller = strdup(SYSTEMD_CGROUP_CONTROLLER)))
-                goto fail;
-
         if (!(b->path = default_cgroup_path(u)))
                 goto fail;
 
diff --git a/src/util.c b/src/util.c
index 69eb14be61..fa34137bdc 100644
--- a/src/util.c
+++ b/src/util.c
@@ -609,8 +609,22 @@ int get_process_cmdline(pid_t pid, size_t max_length, char **line) {
 
         fclose(f);
 
-        if (r[0] == 0)
-                return get_process_name(pid, line);
+        /* Kernel threads have no argv[] */
+        if (r[0] == 0) {
+                char *t;
+                int h;
+
+                free(r);
+
+                if ((h = get_process_name(pid, &t)) < 0)
+                        return h;
+
+                h = asprintf(&r, "[%s]", t);
+                free(t);
+
+                if (h < 0)
+                        return -ENOMEM;
+        }
 
         *line = r;
         return 0;
@@ -698,6 +712,48 @@ int readlink_and_make_absolute(const char *p, char **r) {
         return 0;
 }
 
+int parent_of_path(const char *path, char **_r) {
+        const char *e, *a = NULL, *b = NULL, *p;
+        char *r;
+        bool slash = false;
+
+        assert(path);
+        assert(_r);
+
+        if (!*path)
+                return -EINVAL;
+
+        for (e = path; *e; e++) {
+
+                if (!slash && *e == '/') {
+                        a = b;
+                        b = e;
+                        slash = true;
+                } else if (slash && *e != '/')
+                        slash = false;
+        }
+
+        if (*(e-1) == '/')
+                p = a;
+        else
+                p = b;
+
+        if (!p)
+                return -EINVAL;
+
+        if (p == path)
+                r = strdup("/");
+        else
+                r = strndup(path, p-path);
+
+        if (!r)
+                return -ENOMEM;
+
+        *_r = r;
+        return 0;
+}
+
+
 char *file_name_from_path(const char *p) {
         char *r;
 
@@ -2201,25 +2257,24 @@ ssize_t loop_write(int fd, const void *buf, size_t nbytes, bool do_poll) {
 
 int path_is_mount_point(const char *t) {
         struct stat a, b;
-        char *copy;
+        char *parent;
+        int r;
 
         if (lstat(t, &a) < 0) {
-
                 if (errno == ENOENT)
                         return 0;
 
                 return -errno;
         }
 
-        if (!(copy = strdup(t)))
-                return -ENOMEM;
+        if ((r = parent_of_path(t, &parent)) < 0)
+                return r;
 
-        if (lstat(dirname(copy), &b) < 0) {
-                free(copy);
-                return -errno;
-        }
+        r = lstat(parent, &b);
+        free(parent);
 
-        free(copy);
+        if (r < 0)
+                return -errno;
 
         return a.st_dev != b.st_dev;
 }
diff --git a/src/util.h b/src/util.h
index f1cd080e42..e4b1f81c06 100644
--- a/src/util.h
+++ b/src/util.h
@@ -212,6 +212,8 @@ int safe_mkdir(const char *path, mode_t mode, uid_t uid, gid_t gid);
 int mkdir_parents(const char *path, mode_t mode);
 int mkdir_p(const char *path, mode_t mode);
 
+int parent_of_path(const char *path, char **parent);
+
 int rmdir_parents(const char *path, const char *stop);
 
 int get_process_name(pid_t pid, char **name);