chiark / gitweb /
job: timeout every job independently of the unit
[elogind.git] / src / job.c
index 9937ed735b6bffd107c3043559e67fc77f8da729..2b422b48773918235fee6e9f0a138099c9bc338b 100644 (file)
--- a/src/job.c
+++ b/src/job.c
@@ -21,6 +21,8 @@
 
 #include <assert.h>
 #include <errno.h>
+#include <sys/timerfd.h>
+#include <sys/epoll.h>
 
 #include "set.h"
 #include "unit.h"
@@ -46,6 +48,8 @@ Job* job_new(Manager *m, JobType type, Unit *unit) {
         j->type = type;
         j->unit = unit;
 
+        j->timer_watch.type = WATCH_INVALID;
+
         /* We don't link it here, that's what job_dependency() is for */
 
         return j;
@@ -56,7 +60,7 @@ void job_free(Job *j) {
 
         /* Detach from next 'bigger' objects */
         if (j->installed) {
-                bus_job_send_removed_signal(j);
+                bus_job_send_removed_signal(j, !j->failed);
 
                 if (j->unit->meta.job == j) {
                         j->unit->meta.job = NULL;
@@ -76,6 +80,16 @@ void job_free(Job *j) {
         if (j->in_dbus_queue)
                 LIST_REMOVE(Job, dbus_queue, j->manager->dbus_job_queue, j);
 
+        if (j->timer_watch.type != WATCH_INVALID) {
+                assert(j->timer_watch.type == WATCH_JOB_TIMER);
+                assert(j->timer_watch.data.job == j);
+                assert(j->timer_watch.fd >= 0);
+
+                assert_se(epoll_ctl(j->manager->epoll_fd, EPOLL_CTL_DEL, j->timer_watch.fd, NULL) >= 0);
+                close_nointr_nofail(j->timer_watch.fd);
+        }
+
+        free(j->bus_client);
         free(j);
 }
 
@@ -275,25 +289,26 @@ bool job_type_is_redundant(JobType a, UnitActiveState b) {
         case JOB_START:
                 return
                         b == UNIT_ACTIVE ||
-                        b == UNIT_ACTIVE_RELOADING;
+                        b == UNIT_RELOADING;
 
         case JOB_STOP:
                 return
-                        b == UNIT_INACTIVE;
+                        b == UNIT_INACTIVE ||
+                        b == UNIT_MAINTENANCE;
 
         case JOB_VERIFY_ACTIVE:
                 return
                         b == UNIT_ACTIVE ||
-                        b == UNIT_ACTIVE_RELOADING;
+                        b == UNIT_RELOADING;
 
         case JOB_RELOAD:
                 return
-                        b == UNIT_ACTIVE_RELOADING;
+                        b == UNIT_RELOADING;
 
         case JOB_RELOAD_OR_START:
                 return
                         b == UNIT_ACTIVATING ||
-                        b == UNIT_ACTIVE_RELOADING;
+                        b == UNIT_RELOADING;
 
         case JOB_RESTART:
                 return
@@ -415,7 +430,7 @@ int job_run_and_invalidate(Job *j) {
 
                 case JOB_RESTART: {
                         UnitActiveState t = unit_active_state(j->unit);
-                        if (t == UNIT_INACTIVE || t == UNIT_ACTIVATING) {
+                        if (t == UNIT_INACTIVE || t == UNIT_MAINTENANCE || t == UNIT_ACTIVATING) {
                                 j->type = JOB_START;
                                 r = unit_start(j->unit);
                         } else
@@ -425,7 +440,7 @@ int job_run_and_invalidate(Job *j) {
 
                 case JOB_TRY_RESTART: {
                         UnitActiveState t = unit_active_state(j->unit);
-                        if (t == UNIT_INACTIVE || t == UNIT_DEACTIVATING)
+                        if (t == UNIT_INACTIVE || t == UNIT_MAINTENANCE || t == UNIT_DEACTIVATING)
                                 r = -ENOEXEC;
                         else if (t == UNIT_ACTIVATING) {
                                 j->type = JOB_START;
@@ -459,7 +474,6 @@ int job_finish_and_invalidate(Job *j, bool success) {
         assert(j);
         assert(j->installed);
 
-        log_debug("Job %s/%s finished, success=%s", j->unit->meta.id, job_type_to_string(j->type), yes_no(success));
         job_add_to_dbus_queue(j);
 
         /* Patch restart jobs so that they become normal start jobs */
@@ -469,17 +483,21 @@ int job_finish_and_invalidate(Job *j, bool success) {
                           j->unit->meta.id, job_type_to_string(j->type),
                           j->unit->meta.id, job_type_to_string(JOB_START));
 
-                j->state = JOB_RUNNING;
+                j->state = JOB_WAITING;
                 j->type = JOB_START;
-
-                job_add_to_run_queue(j);
                 return 0;
         }
 
+        log_debug("Job %s/%s finished, success=%s", j->unit->meta.id, job_type_to_string(j->type), yes_no(success));
+
+        j->failed = !success;
         u = j->unit;
         t = j->type;
         job_free(j);
 
+        if (!success)
+                unit_status_printf(u, "Starting %s " ANSI_HIGHLIGHT_ON "failed" ANSI_HIGHLIGHT_OFF ".\n", unit_description(u));
+
         /* Fail depending jobs on failure */
         if (!success) {
 
@@ -488,14 +506,16 @@ int job_finish_and_invalidate(Job *j, bool success) {
                     t == JOB_RELOAD_OR_START) {
 
                         SET_FOREACH(other, u->meta.dependencies[UNIT_REQUIRED_BY], i)
-                                if (other->meta.job &&
+                                if (!other->meta.ignore_dependency_failure &&
+                                    other->meta.job &&
                                     (other->meta.job->type == JOB_START ||
                                      other->meta.job->type == JOB_VERIFY_ACTIVE ||
                                      other->meta.job->type == JOB_RELOAD_OR_START))
                                         job_finish_and_invalidate(other->meta.job, false);
 
                         SET_FOREACH(other, u->meta.dependencies[UNIT_REQUIRED_BY_OVERRIDABLE], i)
-                                if (other->meta.job &&
+                                if (!other->meta.ignore_dependency_failure &&
+                                    other->meta.job &&
                                     !other->meta.job->override &&
                                     (other->meta.job->type == JOB_START ||
                                      other->meta.job->type == JOB_VERIFY_ACTIVE ||
@@ -505,7 +525,8 @@ int job_finish_and_invalidate(Job *j, bool success) {
                 } else if (t == JOB_STOP) {
 
                         SET_FOREACH(other, u->meta.dependencies[UNIT_CONFLICTS], i)
-                                if (other->meta.job &&
+                                if (!other->meta.ignore_dependency_failure &&
+                                    other->meta.job &&
                                     (other->meta.job->type == JOB_START ||
                                      other->meta.job->type == JOB_VERIFY_ACTIVE ||
                                      other->meta.job->type == JOB_RELOAD_OR_START))
@@ -524,6 +545,53 @@ int job_finish_and_invalidate(Job *j, bool success) {
         return 0;
 }
 
+int job_start_timer(Job *j) {
+        struct itimerspec its;
+        struct epoll_event ev;
+        int fd, r;
+        assert(j);
+
+        if (j->unit->meta.job_timeout <= 0 ||
+            j->timer_watch.type == WATCH_JOB_TIMER)
+                return 0;
+
+        assert(j->timer_watch.type == WATCH_INVALID);
+
+        if ((fd = timerfd_create(CLOCK_MONOTONIC, TFD_NONBLOCK|TFD_CLOEXEC)) < 0) {
+                r = -errno;
+                goto fail;
+        }
+
+        zero(its);
+        timespec_store(&its.it_value, j->unit->meta.job_timeout);
+
+        if (timerfd_settime(fd, 0, &its, NULL) < 0) {
+                r = -errno;
+                goto fail;
+        }
+
+        zero(ev);
+        ev.data.ptr = &j->timer_watch;
+        ev.events = EPOLLIN;
+
+        if (epoll_ctl(j->manager->epoll_fd, EPOLL_CTL_ADD, fd, &ev) < 0) {
+                r = -errno;
+                goto fail;
+        }
+
+        j->timer_watch.type = WATCH_JOB_TIMER;
+        j->timer_watch.fd = fd;
+        j->timer_watch.data.job = j;
+
+        return 0;
+
+fail:
+        if (fd >= 0)
+                close_nointr_nofail(fd);
+
+        return r;
+}
+
 void job_add_to_run_queue(Job *j) {
         assert(j);
         assert(j->installed);
@@ -542,10 +610,9 @@ void job_add_to_dbus_queue(Job *j) {
         if (j->in_dbus_queue)
                 return;
 
-        if (set_isempty(j->manager->subscribed)) {
-                j->sent_dbus_new_signal = true;
-                return;
-        }
+        /* We don't check if anybody is subscribed here, since this
+         * job might just have been created and not yet assigned to a
+         * connection/client. */
 
         LIST_PREPEND(Job, dbus_queue, j->manager->dbus_job_queue, j);
         j->in_dbus_queue = true;
@@ -562,6 +629,14 @@ char *job_dbus_path(Job *j) {
         return p;
 }
 
+void job_timer_event(Job *j, uint64_t n_elapsed, Watch *w) {
+        assert(j);
+        assert(w == &j->timer_watch);
+
+        log_warning("Job %s/%s timed out.", j->unit->meta.id, job_type_to_string(j->type));
+        job_finish_and_invalidate(j, false);
+}
+
 static const char* const job_state_table[_JOB_STATE_MAX] = {
         [JOB_WAITING] = "waiting",
         [JOB_RUNNING] = "running"