/* remove an event from the queue */
void remove_event(squeue_t *sq, timed_event *event)
{
#ifdef USE_EVENT_BROKER
/* send event data to broker */
broker_timed_event(NEBTYPE_TIMEDEVENT_REMOVE, NEBFLAG_NONE, NEBATTR_NONE, event, NULL);
#endif
if (!event || !event->sq_event)
return;
if (sq)
squeue_remove(sq, event->sq_event);
else
logit(NSLOG_RUNTIME_ERROR, TRUE,
"Error: remove_event() called for %s event with NULL sq parameter\n",
EVENT_TYPE_STR(event->event_type));
if (sq == nagios_squeue)
track_events(event->event_type, -1);
event->sq_event = NULL; /* mark this event as unscheduled */
/*
* if we catch an event from the queue which gets removed when
* we go polling for input (as might happen with f.e. downtime
* events that we get "cancel" commands for just as they are
* about to start or expire), we must make sure we mark the
* current event as no longer scheduled, or we'll run into
* segfaults and memory corruptions for sure.
*/
if (event == current_event) {
current_event = NULL;
}
}
static void destroy_job(child_process *cp)
{
/*
* we must remove the job's timeout ticker,
* or we'll end up accessing an already free()'d
* pointer, or the pointer to a different child.
*/
squeue_remove(sq, cp->ei->sq_event);
running_jobs--;
fanout_remove(ptab, cp->ei->pid);
if (cp->outstd.buf) {
free(cp->outstd.buf);
cp->outstd.buf = NULL;
}
if (cp->outerr.buf) {
free(cp->outerr.buf);
cp->outerr.buf = NULL;
}
kvvec_destroy(cp->request, KVVEC_FREE_ALL);
free(cp->cmd);
free(cp->ei);
free(cp);
}
static int finish_job(child_process *cp, int reason)
{
static struct kvvec resp = KVVEC_INITIALIZER;
struct rusage *ru = &cp->rusage;
int i, ret;
/* how many key/value pairs do we need? */
if (kvvec_init(&resp, 12 + cp->request->kv_pairs) == NULL) {
/* what the hell do we do now? */
exit_worker();
}
gettimeofday(&cp->stop, NULL);
if (running_jobs != squeue_size(sq)) {
wlog("running_jobs(%d) != squeue_size(sq) (%d)\n",
running_jobs, squeue_size(sq));
wlog("started: %d; running: %d; finished: %d\n",
started, running_jobs, started - running_jobs);
}
/*
* we must remove the job's timeout ticker,
* or we'll end up accessing an already free()'d
* pointer, or the pointer to a different child.
*/
squeue_remove(sq, cp->sq_event);
/* get rid of still open filedescriptors */
if (cp->outstd.fd != -1)
iobroker_close(iobs, cp->outstd.fd);
if (cp->outerr.fd != -1)
iobroker_close(iobs, cp->outerr.fd);
cp->runtime = tv_delta_f(&cp->start, &cp->stop);
/*
* Now build the return message.
* First comes the request, minus environment variables
*/
for (i = 0; i < cp->request->kv_pairs; i++) {
struct key_value *kv = &cp->request->kv[i];
/* skip environment macros */
if (kv->key_len == 3 && !strcmp(kv->key, "env")) {
continue;
}
kvvec_addkv_wlen(&resp, kv->key, kv->key_len, kv->value, kv->value_len);
}
kvvec_addkv(&resp, "wait_status", (char *)mkstr("%d", cp->ret));
kvvec_addkv_wlen(&resp, "outstd", 6, cp->outstd.buf, cp->outstd.len);
kvvec_addkv_wlen(&resp, "outerr", 6, cp->outerr.buf, cp->outerr.len);
kvvec_add_tv(&resp, "start", cp->start);
kvvec_add_tv(&resp, "stop", cp->stop);
kvvec_addkv(&resp, "runtime", (char *)mkstr("%f", cp->runtime));
if (!reason) {
/* child exited nicely */
kvvec_addkv(&resp, "exited_ok", "1");
kvvec_add_tv(&resp, "ru_utime", ru->ru_utime);
kvvec_add_tv(&resp, "ru_stime", ru->ru_stime);
kvvec_add_long(&resp, "ru_minflt", ru->ru_minflt);
kvvec_add_long(&resp, "ru_majflt", ru->ru_majflt);
kvvec_add_long(&resp, "ru_nswap", ru->ru_nswap);
kvvec_add_long(&resp, "ru_inblock", ru->ru_inblock);
kvvec_add_long(&resp, "ru_oublock", ru->ru_oublock);
kvvec_add_long(&resp, "ru_nsignals", ru->ru_nsignals);
} else {
/* some error happened */
kvvec_addkv(&resp, "exited_ok", "0");
kvvec_addkv(&resp, "error_code", (char *)mkstr("%d", reason));
}
ret = send_kvvec(master_sd, &resp);
if (ret < 0 && errno == EPIPE)
exit_worker();
running_jobs--;
if (cp->outstd.buf) {
free(cp->outstd.buf);
cp->outstd.buf = NULL;
}
if (cp->outerr.buf) {
free(cp->outerr.buf);
cp->outerr.buf = NULL;
}
kvvec_destroy(cp->request, KVVEC_FREE_ALL);
free(cp->cmd);
free(cp);
return 0;
}
/*
* "What can the harvest hope for, if not for the care
* of the Reaper Man?"
* -- Terry Pratchett, Reaper Man
*
* We end up here no matter if the job is stale (ie, the child is
* stuck in uninterruptable sleep) or if it's the first time we try
* to kill it.
* A job is considered reaped once we reap our direct child, in
* which case init will become parent of our grandchildren.
* It's also considered fully reaped if kill() results in ESRCH or
* EPERM, or if wait()ing for the process group results in ECHILD.
*/
static void kill_job(child_process *cp, int reason)
{
int ret, status, reaped = 0;
int pid = cp ? cp->ei->pid : 0;
/*
* first attempt at reaping, so see if we just failed to
* notice that things were going wrong her
*/
if (reason == ETIME && !check_completion(cp, WNOHANG)) {
timeouts++;
wlog("job %d with pid %d reaped at timeout. timeouts=%u; started=%u", cp->id, pid, timeouts, started);
return;
}
/* brutal but efficient */
if (kill(-cp->ei->pid, SIGKILL) < 0) {
if (errno == ESRCH) {
reaped = 1;
} else {
wlog("kill(-%d, SIGKILL) failed: %s\n", cp->ei->pid, strerror(errno));
}
}
/*
* we must iterate at least once, in case kill() returns
* ESRCH when there's zombies
*/
do {
ret = waitpid(cp->ei->pid, &status, WNOHANG);
if (ret < 0 && errno == EINTR)
continue;
if (ret == cp->ei->pid || (ret < 0 && errno == ECHILD)) {
reaped = 1;
break;
}
if (!ret) {
struct timeval tv;
gettimeofday(&tv, NULL);
/*
* stale process (signal may not have been delivered, or
* the child can be stuck in uninterruptible sleep). We
* can't hang around forever, so just reschedule a new
* reap attempt later.
*/
if (reason == ESTALE) {
tv.tv_sec += 5;
wlog("Failed to reap child with pid %d. Next attempt @ %lu.%lu", cp->ei->pid, tv.tv_sec, tv.tv_usec);
} else {
tv.tv_usec = 250000;
if (tv.tv_usec > 1000000) {
tv.tv_usec -= 1000000;
tv.tv_sec += 1;
}
cp->ei->state = ESTALE;
finish_job(cp, reason);
}
squeue_remove(sq, cp->ei->sq_event);
cp->ei->sq_event = squeue_add_tv(sq, &tv, cp);
return;
}
} while (!reaped);
if (cp->ei->state != ESTALE)
finish_job(cp, reason);
else
wlog("job %d (pid=%d): Dormant child reaped", cp->id, cp->ei->pid);
destroy_job(cp);
}
int main(int argc, char **argv)
{
squeue_t *sq;
struct timeval tv;
sq_test_event a, b, c, d, *x;
t_set_colors(0);
t_start("squeue tests");
a.id = 1;
b.id = 2;
c.id = 3;
d.id = 4;
gettimeofday(&tv, NULL);
/* Order in is a, b, c, d, but we should get b, c, d, a out. */
srand(tv.tv_usec ^ tv.tv_sec);
t((sq = squeue_create(1024)) != NULL);
t(squeue_size(sq) == 0);
/* we fill and empty the squeue completely once before testing */
sq_test_random(sq);
t(squeue_size(sq) == 0, "Size should be 0 after first sq_test_random");
t((a.evt = squeue_add(sq, time(NULL) + 9, &a)) != NULL);
t(squeue_size(sq) == 1);
t((b.evt = squeue_add(sq, time(NULL) + 3, &b)) != NULL);
t(squeue_size(sq) == 2);
t((c.evt = squeue_add_msec(sq, time(NULL) + 5, 0, &c)) != NULL);
t(squeue_size(sq) == 3);
t((d.evt = squeue_add_usec(sq, time(NULL) + 5, 1, &d)) != NULL);
t(squeue_size(sq) == 4);
/* add and remove lots. remainder should be what we have above */
sq_test_random(sq);
/* testing squeue_peek() */
t((x = (sq_test_event *)squeue_peek(sq)) != NULL);
t(x == &b, "x: %p; a: %p; b: %p; c: %p; d: %p\n", x, &a, &b, &c, &d);
t(x->id == b.id);
t(squeue_size(sq) == 4);
/* testing squeue_remove() and re-add */
t(squeue_remove(sq, b.evt) == 0);
t(squeue_size(sq) == 3);
t((x = squeue_peek(sq)) != NULL);
t(x == &c);
t((b.evt = squeue_add(sq, time(NULL) + 3, &b)) != NULL);
t(squeue_size(sq) == 4);
/* peek should now give us the &b event (again) */
t((x = squeue_peek(sq)) != NULL);
if (x != &b) {
printf("about to fail pretty f*****g hard...\n");
printf("ea: %p; &b: %p; &c: %p; ed: %p; x: %p\n",
&a, &b, &c, &d, x);
}
t(x == &b);
t(x->id == b.id);
t(squeue_size(sq) == 4);
/* testing squeue_pop(), lifo manner */
t((x = squeue_pop(sq)) != NULL);
t(squeue_size(sq) == 3,
"squeue_size(sq) = %d\n", squeue_size(sq));
t(x == &b, "x: %p; &b: %p\n", x, &b);
t(x->id == b.id, "x->id: %lu; d.id: %lu\n", x->id, d.id);
/* Test squeue_pop() */
t((x = squeue_pop(sq)) != NULL);
t(squeue_size(sq) == 2);
t(x == &c, "x->id: %lu; c.id: %lu\n", x->id, c.id);
t(x->id == c.id, "x->id: %lu; c.id: %lu\n", x->id, c.id);
/* this should fail gracefully (-1 return from squeue_remove()) */
t(squeue_remove(NULL, NULL) == -1);
t(squeue_remove(NULL, a.evt) == -1);
squeue_foreach(sq, sq_walker, NULL);
/* clean up to prevent false valgrind positives */
squeue_destroy(sq, 0);
return t_end();
}