-/*-*- Mode: C; c-basic-offset: 8 -*-*/
+/*-*- Mode: C; c-basic-offset: 8; indent-tabs-mode: nil -*-*/
/***
This file is part of systemd.
#include <sys/types.h>
#include <signal.h>
#include <sys/mount.h>
+#include <fcntl.h>
#include "cgroup.h"
+#include "cgroup-util.h"
#include "log.h"
-static int 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;
-}
-
int cgroup_bonding_realize(CGroupBonding *b) {
int r;
assert(b->path);
assert(b->controller);
- if (b->cgroup)
- return 0;
-
- if (!(b->cgroup = cgroup_new_cgroup(b->path)))
- return -ENOMEM;
-
- if (!cgroup_add_controller(b->cgroup, b->controller)) {
- r = -ENOMEM;
- goto fail;
+ r = cg_create(b->controller, b->path);
+ if (r < 0) {
+ log_warning("Failed to create cgroup %s:%s: %s", b->controller, b->path, strerror(-r));
+ return r;
}
- if (b->inherit)
- r = cgroup_create_cgroup_from_parent(b->cgroup, true);
- else
- r = cgroup_create_cgroup(b->cgroup, true);
-
- if (r != 0) {
- r = translate_error(r, errno);
- goto fail;
- }
+ b->realized = true;
return 0;
-
-fail:
- cgroup_free(&b->cgroup);
- b->cgroup = NULL;
- return r;
}
int cgroup_bonding_realize_list(CGroupBonding *first) {
CGroupBonding *b;
+ int r;
- LIST_FOREACH(by_unit, b, first) {
- int r;
-
- if ((r = cgroup_bonding_realize(b)) < 0)
+ LIST_FOREACH(by_unit, b, first)
+ if ((r = cgroup_bonding_realize(b)) < 0 && b->essential)
return r;
- }
return 0;
}
-void cgroup_bonding_free(CGroupBonding *b) {
+void cgroup_bonding_free(CGroupBonding *b, bool trim) {
assert(b);
if (b->unit) {
CGroupBonding *f;
- LIST_REMOVE(CGroupBonding, by_unit, b->unit->meta.cgroup_bondings, b);
+ LIST_REMOVE(CGroupBonding, by_unit, b->unit->cgroup_bondings, b);
- assert_se(f = hashmap_get(b->unit->meta.manager->cgroup_bondings, b->path));
- LIST_REMOVE(CGroupBonding, by_path, f, b);
+ if (streq(b->controller, SYSTEMD_CGROUP_CONTROLLER)) {
+ assert_se(f = hashmap_get(b->unit->manager->cgroup_bondings, b->path));
+ LIST_REMOVE(CGroupBonding, by_path, f, b);
- if (f)
- hashmap_replace(b->unit->meta.manager->cgroup_bondings, b->path, f);
- else
- hashmap_remove(b->unit->meta.manager->cgroup_bondings, b->path);
+ if (f)
+ hashmap_replace(b->unit->manager->cgroup_bondings, b->path, f);
+ else
+ hashmap_remove(b->unit->manager->cgroup_bondings, b->path);
+ }
}
- if (b->cgroup) {
- if (b->only_us && b->clean_up && cgroup_bonding_is_empty(b) > 0)
- cgroup_delete_cgroup_ext(b->cgroup, true);
-
- cgroup_free(&b->cgroup);
- }
+ if (b->realized && b->ours && trim)
+ cg_trim(b->controller, b->path, false);
free(b->controller);
free(b->path);
free(b);
}
-void cgroup_bonding_free_list(CGroupBonding *first) {
+void cgroup_bonding_free_list(CGroupBonding *first, bool remove_or_trim) {
CGroupBonding *b, *n;
LIST_FOREACH_SAFE(by_unit, b, n, first)
- cgroup_bonding_free(b);
+ cgroup_bonding_free(b, remove_or_trim);
+}
+
+void cgroup_bonding_trim(CGroupBonding *b, bool delete_root) {
+ assert(b);
+
+ if (b->realized && b->ours)
+ cg_trim(b->controller, b->path, delete_root);
+}
+
+void cgroup_bonding_trim_list(CGroupBonding *first, bool delete_root) {
+ CGroupBonding *b;
+
+ LIST_FOREACH(by_unit, b, first)
+ cgroup_bonding_trim(b, delete_root);
}
int cgroup_bonding_install(CGroupBonding *b, pid_t pid) {
assert(b);
assert(pid >= 0);
- if (pid == 0)
- pid = getpid();
-
- if (!b->cgroup)
- return -ENOENT;
-
- if ((r = cgroup_attach_task_pid(b->cgroup, pid)))
- return translate_error(r, errno);
+ if ((r = cg_create_and_attach(b->controller, b->path, pid)) < 0)
+ return r;
+ b->realized = true;
return 0;
}
int cgroup_bonding_install_list(CGroupBonding *first, pid_t pid) {
CGroupBonding *b;
+ int r;
- LIST_FOREACH(by_unit, b, first) {
- int r;
-
- if ((r = cgroup_bonding_install(b, pid)) < 0)
+ LIST_FOREACH(by_unit, b, first)
+ if ((r = cgroup_bonding_install(b, pid)) < 0 && b->essential)
return r;
- }
return 0;
}
-int cgroup_bonding_kill(CGroupBonding *b, int sig) {
- int r;
- Set *s;
- bool done;
- bool killed = false;
-
+int cgroup_bonding_set_group_access(CGroupBonding *b, mode_t mode, uid_t uid, gid_t gid) {
assert(b);
- assert(sig > 0);
-
- if (!b->only_us)
- return -EAGAIN;
-
- if (!(s = set_new(trivial_hash_func, trivial_compare_func)))
- return -ENOMEM;
-
- do {
- void *iterator;
- pid_t pid;
-
- done = true;
- if ((r = cgroup_get_task_begin(b->path, b->controller, &iterator, &pid)) != 0) {
- if (r == ECGEOF) {
- r = 0;
- goto kill_done;
- } else {
- r = translate_error(r, errno);
- break;
- }
- }
-
- for (;;) {
- if (set_get(s, INT_TO_PTR(pid)) != INT_TO_PTR(pid)) {
+ if (!b->realized)
+ return -EINVAL;
- /* If we haven't killed this process
- * yet, kill it */
+ return cg_set_group_access(b->controller, b->path, mode, uid, gid);
+}
- if (kill(pid, sig) < 0 && errno != ESRCH) {
- r = -errno;
- break;
- }
+int cgroup_bonding_set_group_access_list(CGroupBonding *first, mode_t mode, uid_t uid, gid_t gid) {
+ CGroupBonding *b;
+ int r;
- killed = true;
- done = false;
+ LIST_FOREACH(by_unit, b, first) {
+ r = cgroup_bonding_set_group_access(b, mode, uid, gid);
+ if (r < 0)
+ return r;
+ }
- if ((r = set_put(s, INT_TO_PTR(pid))) < 0)
- break;
- }
+ return 0;
+}
- if ((r = cgroup_get_task_next(&iterator, &pid)) != 0) {
+int cgroup_bonding_set_task_access(CGroupBonding *b, mode_t mode, uid_t uid, gid_t gid, int sticky) {
+ assert(b);
- if (r == ECGEOF)
- r = 0;
- else
- r = translate_error(r, errno);
+ if (!b->realized)
+ return -EINVAL;
- break;
- }
- }
+ return cg_set_task_access(b->controller, b->path, mode, uid, gid, sticky);
+}
- kill_done:
- assert_se(cgroup_get_task_end(&iterator) == 0);
+int cgroup_bonding_set_task_access_list(CGroupBonding *first, mode_t mode, uid_t uid, gid_t gid, int sticky) {
+ CGroupBonding *b;
+ int r;
- /* To avoid racing against processes which fork
- * quicker than we can kill them we repeat this until
- * no new pids need to be killed. */
+ LIST_FOREACH(by_unit, b, first) {
+ r = cgroup_bonding_set_task_access(b, mode, uid, gid, sticky);
+ if (r < 0)
+ return r;
+ }
- } while (!done && r >= 0);
+ return 0;
+}
- set_free(s);
+int cgroup_bonding_kill(CGroupBonding *b, int sig, bool sigcont, Set *s) {
+ assert(b);
+ assert(sig >= 0);
- if (r < 0)
- return r;
+ /* Don't kill cgroups that aren't ours */
+ if (!b->ours)
+ return 0;
- return killed ? 0 : -ESRCH;
+ return cg_kill_recursive(b->controller, b->path, sig, sigcont, true, false, s);
}
-int cgroup_bonding_kill_list(CGroupBonding *first, int sig) {
+int cgroup_bonding_kill_list(CGroupBonding *first, int sig, bool sigcont, Set *s) {
CGroupBonding *b;
- int r = -EAGAIN;
+ Set *allocated_set = NULL;
+ int ret = -EAGAIN, r;
+
+ if (!first)
+ return 0;
+
+ if (!s)
+ if (!(s = allocated_set = set_new(trivial_hash_func, trivial_compare_func)))
+ return -ENOMEM;
LIST_FOREACH(by_unit, b, first) {
- if ((r = cgroup_bonding_kill(b, sig)) < 0) {
- if (r == -EAGAIN || -ESRCH)
+ if ((r = cgroup_bonding_kill(b, sig, sigcont, s)) < 0) {
+ if (r == -EAGAIN || r == -ESRCH)
continue;
- return r;
+ ret = r;
+ goto finish;
}
- return 0;
+ if (ret < 0 || r > 0)
+ ret = r;
}
- return r;
+finish:
+ if (allocated_set)
+ set_free(allocated_set);
+
+ return ret;
}
/* Returns 1 if the group is empty, 0 if it is not, -EAGAIN if we
* cannot know */
int cgroup_bonding_is_empty(CGroupBonding *b) {
- void *iterator;
- pid_t pid;
int r;
assert(b);
- r = cgroup_get_task_begin(b->path, b->controller, &iterator, &pid);
-
- if (r == 0 || r == ECGEOF)
- cgroup_get_task_end(&iterator);
+ if ((r = cg_is_empty_recursive(b->controller, b->path, true)) < 0)
+ return r;
- /* Hmm, no PID in this group? Then it is definitely empty */
- if (r == ECGEOF)
+ /* If it is empty it is empty */
+ if (r > 0)
return 1;
- /* Some error? Let's return it */
- if (r != 0)
- return translate_error(r, errno);
-
- /* It's not empty, and we are the only user, then it is
- * definitely not empty */
- if (b->only_us)
- return 0;
-
- /* There are PIDs in the group but we aren't the only users,
- * hence we cannot say */
- return -EAGAIN;
+ /* It's not only us using this cgroup, so we just don't know */
+ return b->ours ? 0 : -EAGAIN;
}
int cgroup_bonding_is_empty_list(CGroupBonding *first) {
return -EAGAIN;
}
-static int install_release_agent(Manager *m, const char *mount_point) {
- char *p, *c, *sc;
+int manager_setup_cgroup(Manager *m) {
+ char *current = NULL, *path = NULL;
int r;
+ char suffix[32];
assert(m);
- assert(mount_point);
- if (asprintf(&p, "%s/release_agent", mount_point) < 0)
- return -ENOMEM;
-
- if ((r = read_one_line_file(p, &c)) < 0) {
- free(p);
- return r;
+ /* 0. Be nice to Ingo Molnar #628004 */
+ if (path_is_mount_point("/sys/fs/cgroup/systemd", false) <= 0) {
+ log_warning("No control group support available, not creating root group.");
+ return 0;
}
- sc = strstrip(c);
-
- if (sc[0] == 0) {
- if ((r = write_one_line_file(p, CGROUP_AGENT_PATH "\n" )) < 0) {
- free(p);
- free(c);
- return r;
- }
- } else if (!streq(sc, CGROUP_AGENT_PATH)) {
- free(p);
- free(c);
- return -EEXIST;
+ /* 1. Determine hierarchy */
+ if ((r = cg_get_by_pid(SYSTEMD_CGROUP_CONTROLLER, 0, ¤t)) < 0) {
+ log_error("Cannot determine cgroup we are running in: %s", strerror(-r));
+ goto finish;
}
- free(c);
- free(p);
-
- if (asprintf(&p, "%s/notify_on_release", mount_point) < 0)
- return -ENOMEM;
-
- if ((r = read_one_line_file(p, &c)) < 0) {
- free(p);
- return r;
+ if (m->running_as == MANAGER_SYSTEM)
+ strcpy(suffix, "/system");
+ else {
+ snprintf(suffix, sizeof(suffix), "/systemd-%lu", (unsigned long) getpid());
+ char_array_0(suffix);
}
- sc = strstrip(c);
+ free(m->cgroup_hierarchy);
+ if (endswith(current, suffix)) {
+ /* We probably got reexecuted and can continue to use our root cgroup */
+ m->cgroup_hierarchy = current;
+ current = NULL;
- if (streq(sc, "0")) {
- if ((r = write_one_line_file(p, "1\n")) < 0) {
- free(p);
- free(c);
- return r;
+ } else {
+ /* We need a new root cgroup */
+ m->cgroup_hierarchy = NULL;
+ if (asprintf(&m->cgroup_hierarchy, "%s%s", streq(current, "/") ? "" : current, suffix) < 0) {
+ log_error("Out of memory");
+ r = -ENOMEM;
+ goto finish;
}
- } else if (!streq(sc, "1")) {
- free(p);
- free(c);
- return -EIO;
}
- free(p);
- free(c);
-
- return 0;
-}
-
-static int create_hierarchy_cgroup(Manager *m) {
- struct cgroup *cg;
- int r;
+ /* 2. Show data */
+ if ((r = cg_get_path(SYSTEMD_CGROUP_CONTROLLER, m->cgroup_hierarchy, NULL, &path)) < 0) {
+ log_error("Cannot find cgroup mount point: %s", strerror(-r));
+ goto finish;
+ }
- assert(m);
+ log_debug("Using cgroup controller " SYSTEMD_CGROUP_CONTROLLER ". File system hierarchy is at %s.", path);
- if (!(cg = cgroup_new_cgroup(m->cgroup_hierarchy)))
- return -ENOMEM;
+ /* 3. Install agent */
+ if ((r = cg_install_release_agent(SYSTEMD_CGROUP_CONTROLLER, SYSTEMD_CGROUP_AGENT_PATH)) < 0)
+ log_warning("Failed to install release agent, ignoring: %s", strerror(-r));
+ else if (r > 0)
+ log_debug("Installed release agent.");
+ else
+ log_debug("Release agent already installed.");
- if (!(cgroup_add_controller(cg, m->cgroup_controller))) {
- r = -ENOMEM;
+ /* 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;
}
- if ((r = cgroup_create_cgroup(cg, true)) != 0) {
- log_error("Failed to create cgroup hierarchy group: %s", cgroup_strerror(r));
- r = translate_error(r, errno);
- goto finish;
- }
+ /* 5. And pin it, so that it cannot be unmounted */
+ if (m->pin_cgroupfs_fd >= 0)
+ close_nointr_nofail(m->pin_cgroupfs_fd);
- if ((r = cgroup_attach_task(cg)) != 0) {
- log_error("Failed to add ourselves to hierarchy group: %s", cgroup_strerror(r));
- r = translate_error(r, errno);
+ if ((m->pin_cgroupfs_fd = open(path, O_RDONLY|O_CLOEXEC|O_DIRECTORY|O_NOCTTY|O_NONBLOCK)) < 0) {
+ log_error("Failed to open pin file: %m");
+ r = -errno;
goto finish;
}
- r = 0;
+ log_debug("Created root group.");
finish:
- cgroup_free(&cg);
+ free(current);
+ free(path);
+
return r;
}
-int manager_setup_cgroup(Manager *m) {
- char *mp, *cp;
- int r;
- pid_t pid;
- char suffix[32];
-
+void manager_shutdown_cgroup(Manager *m, bool delete) {
assert(m);
- if ((r = cgroup_init()) != 0) {
- log_error("Failed to initialize libcg: %s", cgroup_strerror(r));
- return translate_error(r, errno);
- }
-
- free(m->cgroup_controller);
- if (!(m->cgroup_controller = strdup("debug")))
- return -ENOMEM;
-
- if ((r = cgroup_get_subsys_mount_point(m->cgroup_controller, &mp)))
- return translate_error(r, errno);
+ if (delete && m->cgroup_hierarchy)
+ cg_delete(SYSTEMD_CGROUP_CONTROLLER, m->cgroup_hierarchy);
- pid = getpid();
-
- if ((r = cgroup_get_current_controller_path(pid, m->cgroup_controller, &cp))) {
- free(mp);
- return translate_error(r, errno);
+ if (m->pin_cgroupfs_fd >= 0) {
+ close_nointr_nofail(m->pin_cgroupfs_fd);
+ m->pin_cgroupfs_fd = -1;
}
- snprintf(suffix, sizeof(suffix), "/systemd-%u", (unsigned) pid);
- char_array_0(suffix);
-
free(m->cgroup_hierarchy);
-
- if (endswith(cp, suffix))
- /* We probably got reexecuted and can continue to use our root cgroup */
- m->cgroup_hierarchy = cp;
- else {
- /* We need a new root cgroup */
-
- m->cgroup_hierarchy = NULL;
- r = asprintf(&m->cgroup_hierarchy, "%s%s", streq(cp, "/") ? "" : cp, suffix);
- free(cp);
-
- if (r < 0) {
- free(mp);
- return -ENOMEM;
- }
- }
-
- log_debug("Using cgroup controller <%s>, hierarchy mounted at <%s>, using root group <%s>.",
- m->cgroup_controller,
- mp,
- m->cgroup_hierarchy);
-
- if ((r = install_release_agent(m, mp)) < 0)
- log_warning("Failed to install release agent, ignoring: %s", strerror(-r));
- else
- log_debug("Installed release agent, or already installed.");
-
- free(mp);
-
- if ((r = create_hierarchy_cgroup(m)) < 0)
- log_error("Failed to create root cgroup hierarchy: %s", strerror(-r));
- else
- log_debug("Created root group.");
-
- return r;
+ m->cgroup_hierarchy = NULL;
}
-int manager_shutdown_cgroup(Manager *m, bool delete) {
- struct cgroup *cg;
- int r;
+int cgroup_bonding_get(Manager *m, const char *cgroup, CGroupBonding **bonding) {
+ CGroupBonding *b;
+ char *p;
assert(m);
+ assert(cgroup);
+ assert(bonding);
- if (!m->cgroup_hierarchy)
- return 0;
+ b = hashmap_get(m->cgroup_bondings, cgroup);
+ if (b) {
+ *bonding = b;
+ return 1;
+ }
- if (!(cg = cgroup_new_cgroup(m->cgroup_hierarchy)))
+ p = strdup(cgroup);
+ if (!p)
return -ENOMEM;
- if (!(cgroup_add_controller(cg, m->cgroup_controller))) {
- r = -ENOMEM;
- goto finish;
- }
+ for (;;) {
+ char *e;
- /* Often enough we won't be able to delete the cgroup we
- * ourselves are in, hence ignore all errors here */
- if (delete)
- cgroup_delete_cgroup_ext(cg, CGFLAG_DELETE_IGNORE_MIGRATION|CGFLAG_DELETE_RECURSIVE);
- r = 0;
+ e = strrchr(p, '/');
+ if (!e || e == p) {
+ free(p);
+ *bonding = NULL;
+ return 0;
+ }
-finish:
- cgroup_free(&cg);
- return r;
+ *e = 0;
+ b = hashmap_get(m->cgroup_bondings, p);
+ if (b) {
+ free(p);
+ *bonding = b;
+ return 1;
+ }
+ }
}
int cgroup_notify_empty(Manager *m, const char *group) {
CGroupBonding *l, *b;
+ int r;
assert(m);
assert(group);
- if (!(l = hashmap_get(m->cgroup_bondings, group)))
- return 0;
+ r = cgroup_bonding_get(m, group, &l);
+ if (r <= 0)
+ return r;
LIST_FOREACH(by_path, b, l) {
int t;
if (!b->unit)
continue;
- if ((t = cgroup_bonding_is_empty_list(b)) < 0) {
+ t = cgroup_bonding_is_empty_list(b);
+ if (t < 0) {
/* If we don't know, we don't know */
if (t != -EAGAIN)
continue;
}
- if (t > 0)
+ if (t > 0) {
+ /* If it is empty, let's delete it */
+ cgroup_bonding_trim_list(b->unit->cgroup_bondings, true);
+
if (UNIT_VTABLE(b->unit)->cgroup_notify_empty)
UNIT_VTABLE(b->unit)->cgroup_notify_empty(b->unit);
+ }
}
return 0;
}
+Unit* cgroup_unit_by_pid(Manager *m, pid_t pid) {
+ CGroupBonding *l, *b;
+ char *group = NULL;
+
+ assert(m);
+
+ if (pid <= 1)
+ return NULL;
+
+ if (cg_get_by_pid(SYSTEMD_CGROUP_CONTROLLER, pid, &group) < 0)
+ return NULL;
+
+ l = hashmap_get(m->cgroup_bondings, group);
+
+ if (!l) {
+ char *slash;
+
+ while ((slash = strrchr(group, '/'))) {
+ if (slash == group)
+ break;
+
+ *slash = 0;
+
+ if ((l = hashmap_get(m->cgroup_bondings, group)))
+ break;
+ }
+ }
+
+ free(group);
+
+ LIST_FOREACH(by_path, b, l) {
+
+ if (!b->unit)
+ continue;
+
+ if (b->ours)
+ return b->unit;
+ }
+
+ return NULL;
+}
+
CGroupBonding *cgroup_bonding_find_list(CGroupBonding *first, const char *controller) {
CGroupBonding *b;
return r;
}
+
+pid_t cgroup_bonding_search_main_pid(CGroupBonding *b) {
+ FILE *f;
+ pid_t pid = 0, npid, mypid;
+
+ assert(b);
+
+ if (!b->ours)
+ return 0;
+
+ if (cg_enumerate_processes(b->controller, b->path, &f) < 0)
+ return 0;
+
+ mypid = getpid();
+
+ while (cg_read_pid(f, &npid) > 0) {
+ pid_t ppid;
+
+ if (npid == pid)
+ continue;
+
+ /* Ignore processes that aren't our kids */
+ if (get_parent_of_pid(npid, &ppid) >= 0 && ppid != mypid)
+ continue;
+
+ if (pid != 0) {
+ /* Dang, there's more than one daemonized PID
+ in this group, so we don't know what process
+ is the main process. */
+ pid = 0;
+ break;
+ }
+
+ pid = npid;
+ }
+
+ fclose(f);
+
+ return pid;
+}
+
+pid_t cgroup_bonding_search_main_pid_list(CGroupBonding *first) {
+ CGroupBonding *b;
+ pid_t pid;
+
+ /* Try to find a main pid from this cgroup, but checking if
+ * there's only one PID in the cgroup and returning it. Later
+ * on we might want to add additional, smarter heuristics
+ * here. */
+
+ LIST_FOREACH(by_unit, b, first)
+ if ((pid = cgroup_bonding_search_main_pid(b)) != 0)
+ return pid;
+
+ return 0;
+
+}
~ianmdlvl / elogind.git / blobdiff