job: timeout every job independently of the unit

This commit is contained in:
Lennart Poettering 2010-07-17 04:09:28 +02:00
parent 064f51fa29
commit faf919f1eb
10 changed files with 121 additions and 18 deletions

2
fixme
View File

@ -35,8 +35,6 @@
* systemctl status $PID, systemctl stop $PID!
* timeout waiting for mount devices?
* place /etc/inittab with explaining blurb.
* In command lines, support both "$FOO" and $FOO

View File

@ -90,6 +90,7 @@
" <property name=\"DefaultControlGroup\" type=\"s\" access=\"read\"/>\n" \
" <property name=\"ControlGroups\" type=\"as\" access=\"read\"/>\n" \
" <property name=\"NeedDaemonReload\" type=\"b\" access=\"read\"/>\n" \
" <property name=\"JobTimeoutUSec\" type=\"t\" access=\"read\"/>\n" \
" </interface>\n"
#define BUS_UNIT_PROPERTIES \
@ -125,7 +126,8 @@
{ "org.freedesktop.systemd1.Unit", "DefaultDependencies", bus_property_append_bool, "b", &u->meta.default_dependencies }, \
{ "org.freedesktop.systemd1.Unit", "DefaultControlGroup", bus_unit_append_default_cgroup, "s", u }, \
{ "org.freedesktop.systemd1.Unit", "ControlGroups", bus_unit_append_cgroups, "as", u }, \
{ "org.freedesktop.systemd1.Unit", "NeedDaemonReload", bus_unit_append_need_daemon_reload, "b", u }
{ "org.freedesktop.systemd1.Unit", "NeedDaemonReload", bus_unit_append_need_daemon_reload, "b", u }, \
{ "org.freedesktop.systemd1.Unit", "JobTimeoutUSec", bus_property_append_usec, "t", &u->meta.job_timeout }
int bus_unit_append_names(Manager *m, DBusMessageIter *i, const char *property, void *data);
int bus_unit_append_dependencies(Manager *m, DBusMessageIter *i, const char *property, void *data);

View File

@ -35,6 +35,15 @@ static const UnitActiveState state_translation_table[_DEVICE_STATE_MAX] = {
[DEVICE_PLUGGED] = UNIT_ACTIVE
};
static void device_init(Unit *u) {
Device *d = DEVICE(u);
assert(d);
assert(d->meta.load_state == UNIT_STUB);
d->meta.job_timeout = DEFAULT_TIMEOUT_USEC;
}
static void device_done(Unit *u) {
Device *d = DEVICE(u);
@ -456,6 +465,8 @@ const UnitVTable device_vtable = {
.no_snapshots = true,
.no_isolate = true,
.init = device_init,
.load = unit_load_fragment_and_dropin_optional,
.done = device_done,
.coldplug = device_coldplug,

View File

@ -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;
@ -76,6 +80,15 @@ 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);
}
@ -472,8 +485,6 @@ int job_finish_and_invalidate(Job *j, bool success) {
j->state = JOB_WAITING;
j->type = JOB_START;
job_add_to_run_queue(j);
return 0;
}
@ -534,6 +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);
@ -571,6 +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"

View File

@ -102,6 +102,8 @@ struct Job {
JobType type;
JobState state;
Watch timer_watch;
/* Note that this bus object is not ref counted here. */
DBusConnection *bus;
char *bus_client;
@ -138,9 +140,14 @@ bool job_is_runnable(Job *j);
void job_add_to_run_queue(Job *j);
void job_add_to_dbus_queue(Job *j);
int job_start_timer(Job *j);
void job_timer_event(Job *j, uint64_t n_elapsed, Watch *w);
int job_run_and_invalidate(Job *j);
int job_finish_and_invalidate(Job *j, bool success);
char *job_dbus_path(Job *j);
const char* job_type_to_string(JobType t);
JobType job_type_from_string(const char *s);
@ -150,6 +157,4 @@ JobState job_state_from_string(const char *s);
const char* job_mode_to_string(JobMode t);
JobMode job_mode_from_string(const char *s);
char *job_dbus_path(Job *j);
#endif

View File

@ -1565,6 +1565,7 @@ static int load_from_path(Unit *u, const char *path) {
{ "OnlyByDependency", config_parse_bool, &u->meta.only_by_dependency, "Unit" },
{ "DefaultDependencies", config_parse_bool, &u->meta.default_dependencies, "Unit" },
{ "IgnoreDependencyFailure",config_parse_bool, &u->meta.ignore_dependency_failure, "Unit" },
{ "JobTimeoutSec", config_parse_usec, &u->meta.job_timeout, "Unit" },
{ "PIDFile", config_parse_path, &u->service.pid_file, "Service" },
{ "ExecStartPre", config_parse_exec, u->service.exec_command+SERVICE_EXEC_START_PRE, "Service" },

View File

@ -1124,6 +1124,7 @@ static int transaction_apply(Manager *m) {
job_add_to_run_queue(j);
job_add_to_dbus_queue(j);
job_start_timer(j);
}
/* As last step, kill all remaining job dependencies. */
@ -2022,7 +2023,8 @@ static int process_event(Manager *m, struct epoll_event *ev) {
UNIT_VTABLE(w->data.unit)->fd_event(w->data.unit, w->fd, ev->events, w);
break;
case WATCH_TIMER: {
case WATCH_UNIT_TIMER:
case WATCH_JOB_TIMER: {
uint64_t v;
ssize_t k;
@ -2035,7 +2037,10 @@ static int process_event(Manager *m, struct epoll_event *ev) {
return k < 0 ? -errno : -EIO;
}
UNIT_VTABLE(w->data.unit)->timer_event(w->data.unit, v, w);
if (w->type == WATCH_UNIT_TIMER)
UNIT_VTABLE(w->data.unit)->timer_event(w->data.unit, v, w);
else
job_timer_event(w->data.job, v, w);
break;
}

View File

@ -57,7 +57,8 @@ enum WatchType {
WATCH_SIGNAL,
WATCH_NOTIFY,
WATCH_FD,
WATCH_TIMER,
WATCH_UNIT_TIMER,
WATCH_JOB_TIMER,
WATCH_MOUNT,
WATCH_UDEV,
WATCH_DBUS_WATCH,
@ -69,6 +70,7 @@ struct Watch {
WatchType type;
union {
union Unit *unit;
struct Job *job;
DBusWatch *bus_watch;
DBusTimeout *bus_timeout;
} data;

View File

@ -587,7 +587,8 @@ void unit_dump(Unit *u, FILE *f, const char *prefix) {
timestamp1[FORMAT_TIMESTAMP_MAX],
timestamp2[FORMAT_TIMESTAMP_MAX],
timestamp3[FORMAT_TIMESTAMP_MAX],
timestamp4[FORMAT_TIMESTAMP_MAX];
timestamp4[FORMAT_TIMESTAMP_MAX],
timespan[FORMAT_TIMESPAN_MAX];
assert(u);
assert(u->meta.type >= 0);
@ -627,6 +628,9 @@ void unit_dump(Unit *u, FILE *f, const char *prefix) {
if (u->meta.fragment_path)
fprintf(f, "%s\tFragment Path: %s\n", prefix, u->meta.fragment_path);
if (u->meta.job_timeout > 0)
fprintf(f, "%s\tJob Timeout: %s\n", prefix, format_timespan(timespan, sizeof(timespan), u->meta.job_timeout));
for (d = 0; d < _UNIT_DEPENDENCY_MAX; d++) {
Unit *other;
@ -1003,7 +1007,6 @@ void unit_notify(Unit *u, UnitActiveState os, UnitActiveState ns) {
* failed previously due to EAGAIN. */
job_add_to_run_queue(u->meta.job);
/* Let's check whether this state change constitutes a
* finished job, or maybe cotradicts a running job and
* hence needs to invalidate jobs. */
@ -1189,18 +1192,23 @@ int unit_watch_timer(Unit *u, usec_t delay, Watch *w) {
assert(u);
assert(w);
assert(w->type == WATCH_INVALID || (w->type == WATCH_TIMER && w->data.unit == u));
assert(w->type == WATCH_INVALID || (w->type == WATCH_UNIT_TIMER && w->data.unit == u));
/* This will try to reuse the old timer if there is one */
if (w->type == WATCH_TIMER) {
if (w->type == WATCH_UNIT_TIMER) {
assert(w->data.unit == u);
assert(w->fd >= 0);
ours = false;
fd = w->fd;
} else {
} else if (w->type == WATCH_INVALID) {
ours = true;
if ((fd = timerfd_create(CLOCK_MONOTONIC, TFD_NONBLOCK|TFD_CLOEXEC)) < 0)
return -errno;
}
} else
assert_not_reached("Invalid watch type");
zero(its);
@ -1231,8 +1239,8 @@ int unit_watch_timer(Unit *u, usec_t delay, Watch *w) {
goto fail;
}
w->type = WATCH_UNIT_TIMER;
w->fd = fd;
w->type = WATCH_TIMER;
w->data.unit = u;
return 0;
@ -1251,7 +1259,9 @@ void unit_unwatch_timer(Unit *u, Watch *w) {
if (w->type == WATCH_INVALID)
return;
assert(w->type == WATCH_TIMER && w->data.unit == u);
assert(w->type == WATCH_UNIT_TIMER);
assert(w->data.unit == u);
assert(w->fd >= 0);
assert_se(epoll_ctl(u->meta.manager->epoll_fd, EPOLL_CTL_DEL, w->fd, NULL) >= 0);
close_nointr_nofail(w->fd);

View File

@ -143,6 +143,7 @@ struct Meta {
Set *dependencies[_UNIT_DEPENDENCY_MAX];
char *description;
char *fragment_path; /* if loaded from a config file this is the primary path to it */
usec_t fragment_mtime;
@ -150,6 +151,8 @@ struct Meta {
* the job for it */
Job *job;
usec_t job_timeout;
dual_timestamp inactive_exit_timestamp;
dual_timestamp active_enter_timestamp;
dual_timestamp active_exit_timestamp;