chiark / gitweb /
core: fix detection of dead processes
[elogind.git] / src / core / unit.c
index f4d60bcf347d9f361d773633f0ba78994f3e1c8a..0277675f60c5229ecba79c36ea9eca2e887eb3da 100644 (file)
 #include "fileio-label.h"
 #include "bus-errors.h"
 #include "dbus.h"
+#include "execute.h"
+#include "virt.h"
 
 const UnitVTable * const unit_vtable[_UNIT_TYPE_MAX] = {
         [UNIT_SERVICE] = &service_vtable,
-        [UNIT_TIMER] = &timer_vtable,
         [UNIT_SOCKET] = &socket_vtable,
+        [UNIT_BUSNAME] = &busname_vtable,
         [UNIT_TARGET] = &target_vtable,
+        [UNIT_SNAPSHOT] = &snapshot_vtable,
         [UNIT_DEVICE] = &device_vtable,
         [UNIT_MOUNT] = &mount_vtable,
         [UNIT_AUTOMOUNT] = &automount_vtable,
-        [UNIT_SNAPSHOT] = &snapshot_vtable,
         [UNIT_SWAP] = &swap_vtable,
+        [UNIT_TIMER] = &timer_vtable,
         [UNIT_PATH] = &path_vtable,
         [UNIT_SLICE] = &slice_vtable,
         [UNIT_SCOPE] = &scope_vtable
@@ -117,7 +120,7 @@ int unit_add_name(Unit *u, const char *text) {
         if (!s)
                 return -ENOMEM;
 
-        if (!unit_name_is_valid(s, false)) {
+        if (!unit_name_is_valid(s, TEMPLATE_INVALID)) {
                 r = -EINVAL;
                 goto fail;
         }
@@ -479,6 +482,8 @@ void unit_free(Unit *u) {
 
         set_free_free(u->names);
 
+        unit_unwatch_all_pids(u);
+
         condition_free_list(u->conditions);
 
         unit_ref_unset(&u->slice);
@@ -669,6 +674,18 @@ int unit_add_exec_dependencies(Unit *u, ExecContext *c) {
         assert(u);
         assert(c);
 
+        if (c->working_directory) {
+                r = unit_require_mounts_for(u, c->working_directory);
+                if (r < 0)
+                        return r;
+        }
+
+        if (c->root_directory) {
+                r = unit_require_mounts_for(u, c->root_directory);
+                if (r < 0)
+                        return r;
+        }
+
         if (c->std_output != EXEC_OUTPUT_KMSG &&
             c->std_output != EXEC_OUTPUT_SYSLOG &&
             c->std_output != EXEC_OUTPUT_JOURNAL &&
@@ -1123,6 +1140,8 @@ _pure_ static const char *unit_get_status_message_format_try_harder(Unit *u, Job
         return NULL;
 }
 
+#pragma GCC diagnostic push
+#pragma GCC diagnostic ignored "-Wformat-nonliteral"
 static void unit_status_print_starting_stopping(Unit *u, JobType t) {
         const char *format;
 
@@ -1137,6 +1156,7 @@ static void unit_status_print_starting_stopping(Unit *u, JobType t) {
 
         unit_status_printf(u, "", format);
 }
+#pragma GCC diagnostic pop
 
 #pragma GCC diagnostic push
 #pragma GCC diagnostic ignored "-Wformat-nonliteral"
@@ -1299,8 +1319,11 @@ int unit_reload(Unit *u) {
         if (state == UNIT_RELOADING)
                 return -EALREADY;
 
-        if (state != UNIT_ACTIVE)
+        if (state != UNIT_ACTIVE) {
+                log_warning_unit(u->id, "Unit %s cannot be reloaded because it is inactive.",
+                                 u->id);
                 return -ENOEXEC;
+        }
 
         following = unit_following(u);
         if (following) {
@@ -1511,7 +1534,9 @@ void unit_notify(Unit *u, UnitActiveState os, UnitActiveState ns, bool reload_su
          * sucessfully, since there's no change of state in that case. Which is
          * why it is handled in service_set_state() */
         if (UNIT_IS_INACTIVE_OR_FAILED(os) != UNIT_IS_INACTIVE_OR_FAILED(ns)) {
-                ExecContext *ec = unit_get_exec_context(u);
+                ExecContext *ec;
+
+                ec = unit_get_exec_context(u);
                 if (ec && exec_context_may_touch_console(ec)) {
                         if (UNIT_IS_INACTIVE_OR_FAILED(ns)) {
                                 m->n_on_console --;
@@ -1607,12 +1632,11 @@ void unit_notify(Unit *u, UnitActiveState os, UnitActiveState ns, bool reload_su
                 }
 
                 /* stop unneeded units regardless if going down was expected or not */
-                if (UNIT_IS_ACTIVE_OR_ACTIVATING(os) && UNIT_IS_INACTIVE_OR_DEACTIVATING(ns))
+                if (UNIT_IS_INACTIVE_OR_DEACTIVATING(ns))
                         check_unneeded_dependencies(u);
 
                 if (ns != os && ns == UNIT_FAILED) {
-                        log_notice_unit(u->id,
-                                        "Unit %s entered failed state.", u->id);
+                        log_notice_unit(u->id, "Unit %s entered failed state.", u->id);
                         unit_start_on_failure(u);
                 }
         }
@@ -1676,20 +1700,142 @@ void unit_notify(Unit *u, UnitActiveState os, UnitActiveState ns, bool reload_su
 }
 
 int unit_watch_pid(Unit *u, pid_t pid) {
+        int q, r;
+
         assert(u);
         assert(pid >= 1);
 
-        /* Watch a specific PID. We only support one unit watching
-         * each PID for now. */
+        /* Watch a specific PID. We only support one or two units
+         * watching each PID for now, not more. */
 
-        return hashmap_put(u->manager->watch_pids, LONG_TO_PTR(pid), u);
+        r = hashmap_ensure_allocated(&u->manager->watch_pids1, trivial_hash_func, trivial_compare_func);
+        if (r < 0)
+                return r;
+
+        r = set_ensure_allocated(&u->pids, trivial_hash_func, trivial_compare_func);
+        if (r < 0)
+                return r;
+
+        r = hashmap_put(u->manager->watch_pids1, LONG_TO_PTR(pid), u);
+        if (r == -EEXIST) {
+                r = hashmap_ensure_allocated(&u->manager->watch_pids2, trivial_hash_func, trivial_compare_func);
+                if (r < 0)
+                        return r;
+
+                r = hashmap_put(u->manager->watch_pids2, LONG_TO_PTR(pid), u);
+        }
+
+        q = set_put(u->pids, LONG_TO_PTR(pid));
+        if (q < 0)
+                return q;
+
+        return r;
 }
 
 void unit_unwatch_pid(Unit *u, pid_t pid) {
         assert(u);
         assert(pid >= 1);
 
-        hashmap_remove_value(u->manager->watch_pids, LONG_TO_PTR(pid), u);
+        hashmap_remove_value(u->manager->watch_pids1, LONG_TO_PTR(pid), u);
+        hashmap_remove_value(u->manager->watch_pids2, LONG_TO_PTR(pid), u);
+        set_remove(u->pids, LONG_TO_PTR(pid));
+}
+
+static int watch_pids_in_path(Unit *u, const char *path) {
+        _cleanup_closedir_ DIR *d = NULL;
+        _cleanup_fclose_ FILE *f = NULL;
+        int ret = 0, r;
+
+        assert(u);
+        assert(path);
+
+        /* Adds all PIDs from a specific cgroup path to the set of PIDs we watch. */
+
+        r = cg_enumerate_processes(SYSTEMD_CGROUP_CONTROLLER, path, &f);
+        if (r >= 0) {
+                pid_t pid;
+
+                while ((r = cg_read_pid(f, &pid)) > 0) {
+                        r = unit_watch_pid(u, pid);
+                        if (r < 0 && ret >= 0)
+                                ret = r;
+                }
+                if (r < 0 && ret >= 0)
+                        ret = r;
+
+        } else if (ret >= 0)
+                ret = r;
+
+        r = cg_enumerate_subgroups(SYSTEMD_CGROUP_CONTROLLER, path, &d);
+        if (r >= 0) {
+                char *fn;
+
+                while ((r = cg_read_subgroup(d, &fn)) > 0) {
+                        _cleanup_free_ char *p = NULL;
+
+                        p = strjoin(path, "/", fn, NULL);
+                        free(fn);
+
+                        if (!p)
+                                return -ENOMEM;
+
+                        r = watch_pids_in_path(u, p);
+                        if (r < 0 && ret >= 0)
+                                ret = r;
+                }
+                if (r < 0 && ret >= 0)
+                        ret = r;
+
+        } else if (ret >= 0)
+                ret = r;
+
+        return ret;
+}
+
+
+int unit_watch_all_pids(Unit *u) {
+        assert(u);
+
+        if (!u->cgroup_path)
+                return -ENOENT;
+
+        /* Adds all PIDs from our cgroup to the set of PIDs we watch */
+
+        return watch_pids_in_path(u, u->cgroup_path);
+}
+
+void unit_unwatch_all_pids(Unit *u) {
+        Iterator i;
+        void *e;
+
+        assert(u);
+
+        SET_FOREACH(e, u->pids, i) {
+                hashmap_remove_value(u->manager->watch_pids1, e, u);
+                hashmap_remove_value(u->manager->watch_pids2, e, u);
+        }
+
+        set_free(u->pids);
+        u->pids = NULL;
+}
+
+void unit_tidy_watch_pids(Unit *u, pid_t except1, pid_t except2) {
+        Iterator i;
+        void *e;
+
+        assert(u);
+
+        /* Cleans dead PIDs from our list */
+
+        SET_FOREACH(e, u->pids, i) {
+                pid_t pid = PTR_TO_LONG(e);
+
+                if (pid == except1 || pid == except2)
+                        continue;
+
+                if (kill(pid, 0) < 0 && errno == ESRCH)
+                        set_remove(u->pids, e);
+        }
 }
 
 bool unit_job_is_applicable(Unit *u, JobType j) {
@@ -1745,6 +1891,7 @@ int unit_add_dependency(Unit *u, UnitDependency d, Unit *other, bool add_referen
                 [UNIT_TRIGGERED_BY] = UNIT_TRIGGERS,
                 [UNIT_PROPAGATES_RELOAD_TO] = UNIT_RELOAD_PROPAGATED_FROM,
                 [UNIT_RELOAD_PROPAGATED_FROM] = UNIT_PROPAGATES_RELOAD_TO,
+                [UNIT_JOINS_NAMESPACE_OF] = UNIT_JOINS_NAMESPACE_OF,
         };
         int r, q = 0, v = 0, w = 0;
 
@@ -1760,34 +1907,47 @@ int unit_add_dependency(Unit *u, UnitDependency d, Unit *other, bool add_referen
         if (u == other)
                 return 0;
 
-        if ((r = set_ensure_allocated(&u->dependencies[d], trivial_hash_func, trivial_compare_func)) < 0)
+        r = set_ensure_allocated(&u->dependencies[d], trivial_hash_func, trivial_compare_func);
+        if (r < 0)
                 return r;
 
-        if (inverse_table[d] != _UNIT_DEPENDENCY_INVALID)
-                if ((r = set_ensure_allocated(&other->dependencies[inverse_table[d]], trivial_hash_func, trivial_compare_func)) < 0)
+        if (inverse_table[d] != _UNIT_DEPENDENCY_INVALID) {
+                r = set_ensure_allocated(&other->dependencies[inverse_table[d]], trivial_hash_func, trivial_compare_func);
+                if (r < 0)
+                        return r;
+        }
+
+        if (add_reference) {
+                r = set_ensure_allocated(&u->dependencies[UNIT_REFERENCES], trivial_hash_func, trivial_compare_func);
+                if (r < 0)
                         return r;
 
-        if (add_reference)
-                if ((r = set_ensure_allocated(&u->dependencies[UNIT_REFERENCES], trivial_hash_func, trivial_compare_func)) < 0 ||
-                    (r = set_ensure_allocated(&other->dependencies[UNIT_REFERENCED_BY], trivial_hash_func, trivial_compare_func)) < 0)
+                r = set_ensure_allocated(&other->dependencies[UNIT_REFERENCED_BY], trivial_hash_func, trivial_compare_func);
+                if (r < 0)
                         return r;
+        }
 
-        if ((q = set_put(u->dependencies[d], other)) < 0)
+        q = set_put(u->dependencies[d], other);
+        if (q < 0)
                 return q;
 
-        if (inverse_table[d] != _UNIT_DEPENDENCY_INVALID)
-                if ((v = set_put(other->dependencies[inverse_table[d]], u)) < 0) {
+        if (inverse_table[d] != _UNIT_DEPENDENCY_INVALID && inverse_table[d] != d) {
+                v = set_put(other->dependencies[inverse_table[d]], u);
+                if (v < 0) {
                         r = v;
                         goto fail;
                 }
+        }
 
         if (add_reference) {
-                if ((w = set_put(u->dependencies[UNIT_REFERENCES], other)) < 0) {
+                w = set_put(u->dependencies[UNIT_REFERENCES], other);
+                if (w < 0) {
                         r = w;
                         goto fail;
                 }
 
-                if ((r = set_put(other->dependencies[UNIT_REFERENCED_BY], u)) < 0)
+                r = set_put(other->dependencies[UNIT_REFERENCED_BY], u);
+                if (r < 0)
                         goto fail;
         }
 
@@ -1829,7 +1989,7 @@ static const char *resolve_template(Unit *u, const char *name, const char*path,
         assert(p);
 
         if (!name)
-                name = path_get_file_name(path);
+                name = basename(path);
 
         if (!unit_name_is_template(name)) {
                 *p = NULL;
@@ -2082,6 +2242,7 @@ bool unit_can_serialize(Unit *u) {
 }
 
 int unit_serialize(Unit *u, FILE *f, FDSet *fds, bool serialize_jobs) {
+        ExecRuntime *rt;
         int r;
 
         assert(u);
@@ -2095,17 +2256,11 @@ int unit_serialize(Unit *u, FILE *f, FDSet *fds, bool serialize_jobs) {
         if (r < 0)
                 return r;
 
-
-        if (serialize_jobs) {
-                if (u->job) {
-                        fprintf(f, "job\n");
-                        job_serialize(u->job, f, fds);
-                }
-
-                if (u->nop_job) {
-                        fprintf(f, "job\n");
-                        job_serialize(u->nop_job, f, fds);
-                }
+        rt = unit_get_exec_runtime(u);
+        if (rt) {
+                r = exec_runtime_serialize(rt, u, f, fds);
+                if (r < 0)
+                        return r;
         }
 
         dual_timestamp_serialize(f, "inactive-exit-timestamp", &u->inactive_exit_timestamp);
@@ -2122,6 +2277,18 @@ int unit_serialize(Unit *u, FILE *f, FDSet *fds, bool serialize_jobs) {
         if (u->cgroup_path)
                 unit_serialize_item(u, f, "cgroup", u->cgroup_path);
 
+        if (serialize_jobs) {
+                if (u->job) {
+                        fprintf(f, "job\n");
+                        job_serialize(u->job, f, fds);
+                }
+
+                if (u->nop_job) {
+                        fprintf(f, "job\n");
+                        job_serialize(u->nop_job, f, fds);
+                }
+        }
+
         /* End marker */
         fputc('\n', f);
         return 0;
@@ -2155,6 +2322,8 @@ void unit_serialize_item(Unit *u, FILE *f, const char *key, const char *value) {
 }
 
 int unit_deserialize(Unit *u, FILE *f, FDSet *fds) {
+        size_t offset;
+        ExecRuntime **rt = NULL;
         int r;
 
         assert(u);
@@ -2164,6 +2333,10 @@ int unit_deserialize(Unit *u, FILE *f, FDSet *fds) {
         if (!unit_can_serialize(u))
                 return 0;
 
+        offset = UNIT_VTABLE(u)->exec_runtime_offset;
+        if (offset > 0)
+                rt = (ExecRuntime**) ((uint8_t*) u + offset);
+
         for (;;) {
                 char line[LINE_MAX], *l, *v;
                 size_t k;
@@ -2276,6 +2449,14 @@ int unit_deserialize(Unit *u, FILE *f, FDSet *fds) {
                         continue;
                 }
 
+                if (rt) {
+                        r = exec_runtime_deserialize_item(rt, u, l, v, fds);
+                        if (r < 0)
+                                return r;
+                        if (r > 0)
+                                continue;
+                }
+
                 r = UNIT_VTABLE(u)->deserialize_item(u, l, v, fds);
                 if (r < 0)
                         return r;
@@ -2575,7 +2756,7 @@ UnitFileState unit_get_unit_file_state(Unit *u) {
         if (u->unit_file_state < 0 && u->fragment_path)
                 u->unit_file_state = unit_file_get_state(
                                 u->manager->running_as == SYSTEMD_SYSTEM ? UNIT_FILE_SYSTEM : UNIT_FILE_USER,
-                                NULL, path_get_file_name(u->fragment_path));
+                                NULL, basename(u->fragment_path));
 
         return u->unit_file_state;
 }
@@ -2660,6 +2841,16 @@ CGroupContext *unit_get_cgroup_context(Unit *u) {
         return (CGroupContext*) ((uint8_t*) u + offset);
 }
 
+ExecRuntime *unit_get_exec_runtime(Unit *u) {
+        size_t offset;
+
+        offset = UNIT_VTABLE(u)->exec_runtime_offset;
+        if (offset <= 0)
+                return NULL;
+
+        return *(ExecRuntime**) ((uint8_t*) u + offset);
+}
+
 static int drop_in_file(Unit *u, UnitSetPropertiesMode mode, const char *name, char **_p, char **_q) {
         _cleanup_free_ char *b = NULL;
         char *p, *q;
@@ -2856,7 +3047,7 @@ int unit_kill_context(
                 pid_t control_pid,
                 bool main_pid_alien) {
 
-        int sig, wait_for_exit = 0, r;
+        int sig, wait_for_exit = false, r;
 
         assert(u);
         assert(c);
@@ -2873,12 +3064,12 @@ int unit_kill_context(
                         _cleanup_free_ char *comm = NULL;
                         get_process_comm(main_pid, &comm);
 
-                        log_warning_unit(u->id, "Failed to kill main process %li (%s): %s",
-                                         (long) main_pid, strna(comm), strerror(-r));
+                        log_warning_unit(u->id, "Failed to kill main process " PID_FMT " (%s): %s", main_pid, strna(comm), strerror(-r));
                 } else {
-                        wait_for_exit = !main_pid_alien;
+                        if (!main_pid_alien)
+                                wait_for_exit = true;
 
-                        if (c->send_sighup)
+                        if (c->send_sighup && !sigkill)
                                 kill(main_pid, SIGHUP);
                 }
         }
@@ -2890,18 +3081,16 @@ int unit_kill_context(
                         _cleanup_free_ char *comm = NULL;
                         get_process_comm(control_pid, &comm);
 
-                        log_warning_unit(u->id,
-                                         "Failed to kill control process %li (%s): %s",
-                                         (long) control_pid, strna(comm), strerror(-r));
+                        log_warning_unit(u->id, "Failed to kill control process " PID_FMT " (%s): %s", control_pid, strna(comm), strerror(-r));
                 } else {
                         wait_for_exit = true;
 
-                        if (c->send_sighup)
+                        if (c->send_sighup && !sigkill)
                                 kill(control_pid, SIGHUP);
                 }
         }
 
-        if (c->kill_mode == KILL_CONTROL_GROUP && u->cgroup_path) {
+        if ((c->kill_mode == KILL_CONTROL_GROUP || (c->kill_mode == KILL_MIXED && sigkill)) && u->cgroup_path) {
                 _cleanup_set_free_ Set *pid_set = NULL;
 
                 /* Exclude the main/control pids from being killed via the cgroup */
@@ -2914,15 +3103,25 @@ int unit_kill_context(
                         if (r != -EAGAIN && r != -ESRCH && r != -ENOENT)
                                 log_warning_unit(u->id, "Failed to kill control group: %s", strerror(-r));
                 } else if (r > 0) {
-                        wait_for_exit = true;
-                        if (c->send_sighup) {
+
+                        /* FIXME: For now, we will not wait for the
+                         * cgroup members to die, simply because
+                         * cgroup notification is unreliable. It
+                         * doesn't work at all in containers, and
+                         * outside of containers it can be confused
+                         * easily by leaving directories in the
+                         * cgroup. */
+
+                        /* wait_for_exit = true; */
+
+                        if (c->send_sighup && !sigkill) {
                                 set_free(pid_set);
 
                                 pid_set = unit_pid_set(main_pid, control_pid);
                                 if (!pid_set)
                                         return -ENOMEM;
 
-                                cg_kill_recursive(SYSTEMD_CGROUP_CONTROLLER, u->cgroup_path, SIGHUP, true, true, false, pid_set);
+                                cg_kill_recursive(SYSTEMD_CGROUP_CONTROLLER, u->cgroup_path, SIGHUP, false, true, false, pid_set);
                         }
                 }
         }
@@ -2944,17 +3143,15 @@ int unit_require_mounts_for(Unit *u, const char *path) {
          * units can easily determine which units to make themselves a
          * dependency of. */
 
+        if (!path_is_absolute(path))
+                return -EINVAL;
+
         p = strdup(path);
         if (!p)
                 return -ENOMEM;
 
         path_kill_slashes(p);
 
-        if (!path_is_absolute(p)) {
-                free(p);
-                return -EINVAL;
-        }
-
         if (!path_is_safe(p)) {
                 free(p);
                 return -EPERM;
@@ -3010,6 +3207,33 @@ int unit_require_mounts_for(Unit *u, const char *path) {
         return 0;
 }
 
+int unit_setup_exec_runtime(Unit *u) {
+        ExecRuntime **rt;
+        size_t offset;
+        Iterator i;
+        Unit *other;
+
+        offset = UNIT_VTABLE(u)->exec_runtime_offset;
+        assert(offset > 0);
+
+        /* Check if ther already is an ExecRuntime for this unit? */
+        rt = (ExecRuntime**) ((uint8_t*) u + offset);
+        if (*rt)
+                return 0;
+
+        /* Try to get it from somebody else */
+        SET_FOREACH(other, u->dependencies[UNIT_JOINS_NAMESPACE_OF], i) {
+
+                *rt = unit_get_exec_runtime(other);
+                if (*rt) {
+                        exec_runtime_ref(*rt);
+                        return 0;
+                }
+        }
+
+        return exec_runtime_make(rt, unit_get_exec_context(u), u->id);
+}
+
 static const char* const unit_active_state_table[_UNIT_ACTIVE_STATE_MAX] = {
         [UNIT_ACTIVE] = "active",
         [UNIT_RELOADING] = "reloading",
@@ -3045,6 +3269,7 @@ static const char* const unit_dependency_table[_UNIT_DEPENDENCY_MAX] = {
         [UNIT_RELOAD_PROPAGATED_FROM] = "ReloadPropagatedFrom",
         [UNIT_REFERENCES] = "References",
         [UNIT_REFERENCED_BY] = "ReferencedBy",
+        [UNIT_JOINS_NAMESPACE_OF] = "JoinsNamespaceOf",
 };
 
 DEFINE_STRING_TABLE_LOOKUP(unit_dependency, UnitDependency);