X-Git-Url: http://www.chiark.greenend.org.uk/ucgi/~ianmdlvl/git?p=elogind.git;a=blobdiff_plain;f=src%2Fjob.c;h=e57286fe89171a01a080c3e277d37bfd2608a817;hp=31e9cfe8d6f937ddf7973548c599f2c0d64283cb;hb=ac155bb885f9ea8aac3979a6b2686f0c8a9cc6e3;hpb=032ff4afc953cae076ce0ee6a0b85020eeb1a75a diff --git a/src/job.c b/src/job.c index 31e9cfe8d..e57286fe8 100644 --- a/src/job.c +++ b/src/job.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. @@ -21,6 +21,8 @@ #include #include +#include +#include #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; @@ -56,10 +60,10 @@ void job_free(Job *j) { /* Detach from next 'bigger' objects */ if (j->installed) { - bus_job_send_removed_signal(j, !j->failed); + bus_job_send_removed_signal(j); - if (j->unit->meta.job == j) { - j->unit->meta.job = NULL; + if (j->unit->job == j) { + j->unit->job = NULL; unit_add_to_gc_queue(j->unit); } @@ -76,10 +80,20 @@ 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); } -JobDependency* job_dependency_new(Job *subject, Job *object, bool matters) { +JobDependency* job_dependency_new(Job *subject, Job *object, bool matters, bool conflicts) { JobDependency *l; assert(object); @@ -87,7 +101,7 @@ JobDependency* job_dependency_new(Job *subject, Job *object, bool matters) { /* Adds a new job link, which encodes that the 'subject' job * needs the 'object' job in some way. If 'subject' is NULL * this means the 'anchor' job (i.e. the one the user - * explcitily asked for) is the requester. */ + * explicitly asked for) is the requester. */ if (!(l = new0(JobDependency, 1))) return NULL; @@ -95,6 +109,7 @@ JobDependency* job_dependency_new(Job *subject, Job *object, bool matters) { l->subject = subject; l->object = object; l->matters = matters; + l->conflicts = conflicts; if (subject) LIST_PREPEND(JobDependency, subject, subject->subject_list, l); @@ -119,30 +134,6 @@ void job_dependency_free(JobDependency *l) { free(l); } -void job_dependency_delete(Job *subject, Job *object, bool *matters) { - JobDependency *l; - - assert(object); - - LIST_FOREACH(object, l, object->object_list) { - assert(l->object == object); - - if (l->subject == subject) - break; - } - - if (!l) { - if (matters) - *matters = false; - return; - } - - if (matters) - *matters = l->matters; - - job_dependency_free(l); -} - void job_dump(Job *j, FILE*f, const char *prefix) { assert(j); assert(f); @@ -156,7 +147,7 @@ void job_dump(Job *j, FILE*f, const char *prefix) { "%s\tState: %s\n" "%s\tForced: %s\n", prefix, j->id, - prefix, j->unit->meta.id, job_type_to_string(j->type), + prefix, j->unit->id, job_type_to_string(j->type), prefix, job_state_to_string(j->state), prefix, yes_no(j->override)); } @@ -280,7 +271,7 @@ bool job_type_is_redundant(JobType a, UnitActiveState b) { case JOB_STOP: return b == UNIT_INACTIVE || - b == UNIT_MAINTENANCE; + b == UNIT_FAILED; case JOB_VERIFY_ACTIVE: return @@ -318,8 +309,12 @@ bool job_is_runnable(Job *j) { /* Checks whether there is any job running for the units this * job needs to be running after (in the case of a 'positive' - * job type) or before (in the case of a 'negative' job type - * . */ + * job type) or before (in the case of a 'negative' job + * type. */ + + /* First check if there is an override */ + if (j->ignore_order) + return true; if (j->type == JOB_START || j->type == JOB_VERIFY_ACTIVE || @@ -331,19 +326,19 @@ bool job_is_runnable(Job *j) { * dependencies, regardless whether they are * starting or stopping something. */ - SET_FOREACH(other, j->unit->meta.dependencies[UNIT_AFTER], i) - if (other->meta.job) + SET_FOREACH(other, j->unit->dependencies[UNIT_AFTER], i) + if (other->job) return false; } /* Also, if something else is being stopped and we should * change state after it, then lets wait. */ - SET_FOREACH(other, j->unit->meta.dependencies[UNIT_BEFORE], i) - if (other->meta.job && - (other->meta.job->type == JOB_STOP || - other->meta.job->type == JOB_RESTART || - other->meta.job->type == JOB_TRY_RESTART)) + SET_FOREACH(other, j->unit->dependencies[UNIT_BEFORE], i) + if (other->job && + (other->job->type == JOB_STOP || + other->job->type == JOB_RESTART || + other->job->type == JOB_TRY_RESTART)) return false; /* This means that for a service a and a service b where b @@ -362,6 +357,8 @@ bool job_is_runnable(Job *j) { int job_run_and_invalidate(Job *j) { int r; + uint32_t id; + Manager *m; assert(j); assert(j->installed); @@ -380,12 +377,24 @@ int job_run_and_invalidate(Job *j) { j->state = JOB_RUNNING; job_add_to_dbus_queue(j); + /* While we execute this operation the job might go away (for + * example: because it is replaced by a new, conflicting + * job.) To make sure we don't access a freed job later on we + * store the id here, so that we can verify the job is still + * valid. */ + id = j->id; + m = j->manager; + switch (j->type) { case JOB_START: r = unit_start(j->unit); + + /* If this unit cannot be started, then simply + * wait */ if (r == -EBADR) r = 0; + break; case JOB_VERIFY_ACTIVE: { @@ -401,6 +410,11 @@ int job_run_and_invalidate(Job *j) { case JOB_STOP: r = unit_stop(j->unit); + + /* If this unit cannot stopped, then simply + * wait. */ + if (r == -EBADR) + r = 0; break; case JOB_RELOAD: @@ -408,15 +422,21 @@ int job_run_and_invalidate(Job *j) { break; case JOB_RELOAD_OR_START: - if (unit_active_state(j->unit) == UNIT_ACTIVE) + if (unit_active_state(j->unit) == UNIT_ACTIVE) { + j->type = JOB_RELOAD; r = unit_reload(j->unit); - else + } else { + j->type = JOB_START; r = unit_start(j->unit); + + if (r == -EBADR) + r = 0; + } break; case JOB_RESTART: { UnitActiveState t = unit_active_state(j->unit); - if (t == UNIT_INACTIVE || t == UNIT_MAINTENANCE || t == UNIT_ACTIVATING) { + if (t == UNIT_INACTIVE || t == UNIT_FAILED || t == UNIT_ACTIVATING) { j->type = JOB_START; r = unit_start(j->unit); } else @@ -426,13 +446,15 @@ int job_run_and_invalidate(Job *j) { case JOB_TRY_RESTART: { UnitActiveState t = unit_active_state(j->unit); - if (t == UNIT_INACTIVE || t == UNIT_MAINTENANCE || t == UNIT_DEACTIVATING) + if (t == UNIT_INACTIVE || t == UNIT_FAILED || t == UNIT_DEACTIVATING) r = -ENOEXEC; else if (t == UNIT_ACTIVATING) { j->type = JOB_START; r = unit_start(j->unit); - } else + } else { + j->type = JOB_RESTART; r = unit_stop(j->unit); + } break; } @@ -440,90 +462,228 @@ int job_run_and_invalidate(Job *j) { assert_not_reached("Unknown job type"); } - if (r == -EALREADY) - r = job_finish_and_invalidate(j, true); - else if (r == -EAGAIN) { - j->state = JOB_WAITING; - return -EAGAIN; - } else if (r < 0) - r = job_finish_and_invalidate(j, false); + if ((j = manager_get_job(m, id))) { + if (r == -EALREADY) + r = job_finish_and_invalidate(j, JOB_DONE); + else if (r == -ENOEXEC) + r = job_finish_and_invalidate(j, JOB_SKIPPED); + else if (r == -EAGAIN) + j->state = JOB_WAITING; + else if (r < 0) + r = job_finish_and_invalidate(j, JOB_FAILED); + } return r; } -int job_finish_and_invalidate(Job *j, bool success) { +static void job_print_status_message(Unit *u, JobType t, JobResult result) { + assert(u); + + if (t == JOB_START) { + + switch (result) { + + case JOB_DONE: + unit_status_printf(u, ANSI_HIGHLIGHT_GREEN_ON " OK " ANSI_HIGHLIGHT_OFF, "Started %s", unit_description(u)); + break; + + case JOB_FAILED: + unit_status_printf(u, ANSI_HIGHLIGHT_RED_ON "FAILED" ANSI_HIGHLIGHT_OFF, "Failed to start %s", unit_description(u)); + unit_status_printf(u, NULL, "See 'systemctl status %s' for details.", u->id); + break; + + case JOB_DEPENDENCY: + unit_status_printf(u, ANSI_HIGHLIGHT_RED_ON " ABORT" ANSI_HIGHLIGHT_OFF, "Dependency failed. Aborted start of %s", unit_description(u)); + break; + + case JOB_TIMEOUT: + unit_status_printf(u, ANSI_HIGHLIGHT_RED_ON " TIME " ANSI_HIGHLIGHT_OFF, "Timed out starting %s", unit_description(u)); + break; + + default: + ; + } + + } else if (t == JOB_STOP) { + + switch (result) { + + case JOB_TIMEOUT: + unit_status_printf(u, ANSI_HIGHLIGHT_RED_ON " TIME " ANSI_HIGHLIGHT_OFF, "Timed out stopping %s", unit_description(u)); + break; + + case JOB_DONE: + case JOB_FAILED: + unit_status_printf(u, ANSI_HIGHLIGHT_GREEN_ON " OK " ANSI_HIGHLIGHT_OFF, "Stopped %s", unit_description(u)); + break; + + default: + ; + } + } +} + +int job_finish_and_invalidate(Job *j, JobResult result) { Unit *u; Unit *other; JobType t; Iterator i; + bool recursed = false; 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 */ - if (success && (j->type == JOB_RESTART || j->type == JOB_TRY_RESTART)) { + if (result == JOB_DONE && (j->type == JOB_RESTART || j->type == JOB_TRY_RESTART)) { log_debug("Converting job %s/%s -> %s/%s", - j->unit->meta.id, job_type_to_string(j->type), - j->unit->meta.id, job_type_to_string(JOB_START)); + j->unit->id, job_type_to_string(j->type), + j->unit->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; + + u = j->unit; + goto finish; } - j->failed = !success; + j->result = result; + + log_debug("Job %s/%s finished, result=%s", j->unit->id, job_type_to_string(j->type), job_result_to_string(result)); + + if (result == JOB_FAILED) + j->manager->n_failed_jobs ++; + u = j->unit; t = j->type; job_free(j); + job_print_status_message(u, t, result); + /* Fail depending jobs on failure */ - if (!success) { + if (result != JOB_DONE) { if (t == JOB_START || t == JOB_VERIFY_ACTIVE || t == JOB_RELOAD_OR_START) { - SET_FOREACH(other, u->meta.dependencies[UNIT_REQUIRED_BY], i) - if (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 && - !other->meta.job->override && - (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->dependencies[UNIT_REQUIRED_BY], i) + if (other->job && + (other->job->type == JOB_START || + other->job->type == JOB_VERIFY_ACTIVE || + other->job->type == JOB_RELOAD_OR_START)) { + job_finish_and_invalidate(other->job, JOB_DEPENDENCY); + recursed = true; + } + + SET_FOREACH(other, u->dependencies[UNIT_BOUND_BY], i) + if (other->job && + (other->job->type == JOB_START || + other->job->type == JOB_VERIFY_ACTIVE || + other->job->type == JOB_RELOAD_OR_START)) { + job_finish_and_invalidate(other->job, JOB_DEPENDENCY); + recursed = true; + } + + SET_FOREACH(other, u->dependencies[UNIT_REQUIRED_BY_OVERRIDABLE], i) + if (other->job && + !other->job->override && + (other->job->type == JOB_START || + other->job->type == JOB_VERIFY_ACTIVE || + other->job->type == JOB_RELOAD_OR_START)) { + job_finish_and_invalidate(other->job, JOB_DEPENDENCY); + recursed = true; + } } else if (t == JOB_STOP) { - SET_FOREACH(other, u->meta.dependencies[UNIT_CONFLICTS], i) - if (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->dependencies[UNIT_CONFLICTED_BY], i) + if (other->job && + (other->job->type == JOB_START || + other->job->type == JOB_VERIFY_ACTIVE || + other->job->type == JOB_RELOAD_OR_START)) { + job_finish_and_invalidate(other->job, JOB_DEPENDENCY); + recursed = true; + } } } + /* Trigger OnFailure dependencies that are not generated by + * the unit itself. We don't tread JOB_CANCELED as failure in + * this context. And JOB_FAILURE is already handled by the + * unit itself. */ + if (result == JOB_TIMEOUT || result == JOB_DEPENDENCY) { + log_notice("Job %s/%s failed with result '%s'.", + u->id, + job_type_to_string(t), + job_result_to_string(result)); + + unit_trigger_on_failure(u); + } + +finish: /* Try to start the next jobs that can be started */ - SET_FOREACH(other, u->meta.dependencies[UNIT_AFTER], i) - if (other->meta.job) - job_add_to_run_queue(other->meta.job); - SET_FOREACH(other, u->meta.dependencies[UNIT_BEFORE], i) - if (other->meta.job) - job_add_to_run_queue(other->meta.job); + SET_FOREACH(other, u->dependencies[UNIT_AFTER], i) + if (other->job) + job_add_to_run_queue(other->job); + SET_FOREACH(other, u->dependencies[UNIT_BEFORE], i) + if (other->job) + job_add_to_run_queue(other->job); + + manager_check_finished(u->manager); + + return recursed; +} + +int job_start_timer(Job *j) { + struct itimerspec its; + struct epoll_event ev; + int fd, r; + assert(j); + + if (j->unit->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->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) { @@ -544,10 +704,9 @@ 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; @@ -564,6 +723,14 @@ 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->id, job_type_to_string(j->type)); + job_finish_and_invalidate(j, JOB_TIMEOUT); +} + static const char* const job_state_table[_JOB_STATE_MAX] = { [JOB_WAITING] = "waiting", [JOB_RUNNING] = "running" @@ -586,7 +753,20 @@ DEFINE_STRING_TABLE_LOOKUP(job_type, JobType); static const char* const job_mode_table[_JOB_MODE_MAX] = { [JOB_FAIL] = "fail", [JOB_REPLACE] = "replace", - [JOB_ISOLATE] = "isolate" + [JOB_ISOLATE] = "isolate", + [JOB_IGNORE_DEPENDENCIES] = "ignore-dependencies", + [JOB_IGNORE_REQUIREMENTS] = "ignore-requirements" }; DEFINE_STRING_TABLE_LOOKUP(job_mode, JobMode); + +static const char* const job_result_table[_JOB_RESULT_MAX] = { + [JOB_DONE] = "done", + [JOB_CANCELED] = "canceled", + [JOB_TIMEOUT] = "timeout", + [JOB_FAILED] = "failed", + [JOB_DEPENDENCY] = "dependency", + [JOB_SKIPPED] = "skipped" +}; + +DEFINE_STRING_TABLE_LOOKUP(job_result, JobResult);