X-Git-Url: https://www.chiark.greenend.org.uk/ucgi/~ianmdlvl/git?p=elogind.git;a=blobdiff_plain;f=src%2Fcgroup.c;h=1f6139e25f752176b2d19781c1253c8e608f8dc9;hp=83fe2ef207e1ddb72b0111a844d407722c3f4abb;hb=170dcb7bd575f61aa75ce55d7fc0183ace6b8b76;hpb=55096547212928b0ba83fca2595cae0d66d3c0b0

diff --git a/src/cgroup.c b/src/cgroup.c
index 83fe2ef20..1f6139e25 100644
--- a/src/cgroup.c
+++ b/src/cgroup.c
@@ -1,4 +1,4 @@
-/*-*- Mode: C; c-basic-offset: 8 -*-*/
+/*-*- Mode: C; c-basic-offset: 8; indent-tabs-mode: nil -*-*/
 
 /***
   This file is part of systemd.
@@ -25,8 +25,7 @@
 #include <sys/types.h>
 #include <signal.h>
 #include <sys/mount.h>
-
-#include <libcgroup.h>
+#include <fcntl.h>
 
 #include "cgroup.h"
 #include "cgroup-util.h"
@@ -39,17 +38,14 @@ int cgroup_bonding_realize(CGroupBonding *b) {
         assert(b->path);
         assert(b->controller);
 
-        if (b->realized)
-                return 0;
-
-        if ((r = cg_create(b->controller, b->path)) < 0)
+        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;
+        }
 
         b->realized = true;
 
-        if (b->only_us && b->clean_up)
-                cg_trim(b->controller, b->path, false);
-
         return 0;
 }
 
@@ -58,53 +54,50 @@ int cgroup_bonding_realize_list(CGroupBonding *first) {
         int r;
 
         LIST_FOREACH(by_unit, b, first)
-                if ((r = cgroup_bonding_realize(b)) < 0)
+                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->realized && b->only_us && b->clean_up) {
-
-                if (cgroup_bonding_is_empty(b) > 0)
-                        cg_delete(b->controller, b->path);
-                else
-                        cg_trim(b->controller, b->path, false);
-        }
+        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->only_us && b->clean_up)
+        if (b->realized && b->ours)
                 cg_trim(b->controller, b->path, delete_root);
 }
 
@@ -133,42 +126,97 @@ int cgroup_bonding_install_list(CGroupBonding *first, pid_t pid) {
         int r;
 
         LIST_FOREACH(by_unit, b, first)
-                if ((r = cgroup_bonding_install(b, pid)) < 0)
+                if ((r = cgroup_bonding_install(b, pid)) < 0 && b->essential)
                         return r;
 
         return 0;
 }
 
-int cgroup_bonding_kill(CGroupBonding *b, int sig) {
+int cgroup_bonding_set_group_access(CGroupBonding *b, mode_t mode, uid_t uid, gid_t gid) {
+        assert(b);
+
+        if (!b->realized)
+                return -EINVAL;
+
+        return cg_set_group_access(b->controller, b->path, mode, uid, gid);
+}
+
+int cgroup_bonding_set_group_access_list(CGroupBonding *first, mode_t mode, uid_t uid, gid_t gid) {
+        CGroupBonding *b;
         int r;
 
+        LIST_FOREACH(by_unit, b, first) {
+                r = cgroup_bonding_set_group_access(b, mode, uid, gid);
+                if (r < 0)
+                        return r;
+        }
+
+        return 0;
+}
+
+int cgroup_bonding_set_task_access(CGroupBonding *b, mode_t mode, uid_t uid, gid_t gid, int sticky) {
         assert(b);
-        assert(sig >= 0);
 
-        if ((r = cgroup_bonding_realize(b)) < 0)
-                return r;
+        if (!b->realized)
+                return -EINVAL;
+
+        return cg_set_task_access(b->controller, b->path, mode, uid, gid, sticky);
+}
+
+int cgroup_bonding_set_task_access_list(CGroupBonding *first, mode_t mode, uid_t uid, gid_t gid, int sticky) {
+        CGroupBonding *b;
+        int r;
+
+        LIST_FOREACH(by_unit, b, first) {
+                r = cgroup_bonding_set_task_access(b, mode, uid, gid, sticky);
+                if (r < 0)
+                        return r;
+        }
+
+        return 0;
+}
 
-        assert(b->realized);
+int cgroup_bonding_kill(CGroupBonding *b, int sig, bool sigcont, Set *s) {
+        assert(b);
+        assert(sig >= 0);
+
+        /* Don't kill cgroups that aren't ours */
+        if (!b->ours)
+                return 0;
 
-        return cg_kill_recursive(b->controller, b->path, sig, true);
+        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 = 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
@@ -186,7 +234,7 @@ int cgroup_bonding_is_empty(CGroupBonding *b) {
                 return 1;
 
         /* It's not only us using this cgroup, so we just don't know */
-        return b->only_us ? 0 : -EAGAIN;
+        return b->ours ? 0 : -EAGAIN;
 }
 
 int cgroup_bonding_is_empty_list(CGroupBonding *first) {
@@ -210,81 +258,152 @@ int cgroup_bonding_is_empty_list(CGroupBonding *first) {
 }
 
 int manager_setup_cgroup(Manager *m) {
-        char *cp;
+        char *current = NULL, *path = NULL;
         int r;
-        pid_t pid;
         char suffix[32];
 
         assert(m);
 
-        if ((r = cgroup_init()) != 0) {
-                log_error("Failed to initialize libcg: %s", cgroup_strerror(r));
-                return cg_translate_error(r, errno);
+        /* 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;
         }
 
-        free(m->cgroup_mount_point);
-        m->cgroup_mount_point = NULL;
-        if ((r = cgroup_get_subsys_mount_point(SYSTEMD_CGROUP_CONTROLLER, &m->cgroup_mount_point)))
-                return cg_translate_error(r, errno);
-
-        pid = getpid();
-
-        if ((r = cgroup_get_current_controller_path(pid, SYSTEMD_CGROUP_CONTROLLER, &cp)))
-                return cg_translate_error(r, errno);
+        /* 1. Determine hierarchy */
+        if ((r = cg_get_by_pid(SYSTEMD_CGROUP_CONTROLLER, 0, &current)) < 0) {
+                log_error("Cannot determine cgroup we are running in: %s", strerror(-r));
+                goto finish;
+        }
 
-        snprintf(suffix, sizeof(suffix), "/systemd-%u", (unsigned) pid);
-        char_array_0(suffix);
+        if (m->running_as == MANAGER_SYSTEM)
+                strcpy(suffix, "/system");
+        else {
+                snprintf(suffix, sizeof(suffix), "/systemd-%lu", (unsigned long) getpid());
+                char_array_0(suffix);
+        }
 
         free(m->cgroup_hierarchy);
-
-        if (endswith(cp, suffix))
+        if (endswith(current, 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 = current;
+                current = NULL;
 
+        } 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 (asprintf(&m->cgroup_hierarchy, "%s%s", streq(current, "/") ? "" : current, suffix) < 0) {
+                        log_error("Out of memory");
+                        r = -ENOMEM;
+                        goto finish;
+                }
+        }
 
-                if (r < 0)
-                        return -ENOMEM;
+        /* 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;
         }
 
-        log_debug("Using cgroup controller <" SYSTEMD_CGROUP_CONTROLLER ">, hierarchy mounted at <%s>, using root group <%s>.",
-                  m->cgroup_mount_point,
-                  m->cgroup_hierarchy);
+        log_debug("Using cgroup controller " SYSTEMD_CGROUP_CONTROLLER ". File system hierarchy is at %s.", path);
 
-        if ((r = cg_install_release_agent(SYSTEMD_CGROUP_CONTROLLER, CGROUP_AGENT_PATH)) < 0)
+        /* 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("Installed release agent, or already installed.");
+                log_debug("Release agent already installed.");
 
-        if ((r = cg_create_and_attach(SYSTEMD_CGROUP_CONTROLLER, m->cgroup_hierarchy, 0)) < 0)
+        /* 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));
-        else
-                log_debug("Created root group.");
+                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 ((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;
+        }
+
+        log_debug("Created root group.");
+
+finish:
+        free(current);
+        free(path);
 
         return r;
 }
 
-int manager_shutdown_cgroup(Manager *m) {
+void manager_shutdown_cgroup(Manager *m, bool delete) {
         assert(m);
 
-        if (!m->cgroup_hierarchy)
-                return 0;
+        if (delete && m->cgroup_hierarchy)
+                cg_delete(SYSTEMD_CGROUP_CONTROLLER, m->cgroup_hierarchy);
+
+        if (m->pin_cgroupfs_fd >= 0) {
+                close_nointr_nofail(m->pin_cgroupfs_fd);
+                m->pin_cgroupfs_fd = -1;
+        }
+
+        free(m->cgroup_hierarchy);
+        m->cgroup_hierarchy = NULL;
+}
+
+int cgroup_bonding_get(Manager *m, const char *cgroup, CGroupBonding **bonding) {
+        CGroupBonding *b;
+        char *p;
+
+        assert(m);
+        assert(cgroup);
+        assert(bonding);
+
+        b = hashmap_get(m->cgroup_bondings, cgroup);
+        if (b) {
+                *bonding = b;
+                return 1;
+        }
+
+        p = strdup(cgroup);
+        if (!p)
+                return -ENOMEM;
+
+        for (;;) {
+                char *e;
+
+                e = strrchr(p, '/');
+                if (!e || e == p) {
+                        free(p);
+                        *bonding = NULL;
+                        return 0;
+                }
 
-        return cg_delete(SYSTEMD_CGROUP_CONTROLLER, m->cgroup_hierarchy);
+                *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;
@@ -292,7 +411,8 @@ int cgroup_notify_empty(Manager *m, const char *group) {
                 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)
@@ -301,9 +421,13 @@ int cgroup_notify_empty(Manager *m, const char *group) {
                         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;
@@ -312,17 +436,31 @@ int cgroup_notify_empty(Manager *m, const char *group) {
 Unit* cgroup_unit_by_pid(Manager *m, pid_t pid) {
         CGroupBonding *l, *b;
         char *group = NULL;
-        int r;
 
         assert(m);
 
         if (pid <= 1)
                 return NULL;
 
-        if ((r = cg_get_by_pid(SYSTEMD_CGROUP_CONTROLLER, pid, &group)))
+        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) {
@@ -330,7 +468,7 @@ Unit* cgroup_unit_by_pid(Manager *m, pid_t pid) {
                 if (!b->unit)
                         continue;
 
-                if (b->only_us)
+                if (b->ours)
                         return b->unit;
         }
 
@@ -359,3 +497,60 @@ char *cgroup_bonding_to_string(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;
+
+}