static void kill_all_kids(void)
{
unsigned int i;
shm->spawn_no_more = TRUE;
/* Wait for all the children to exit. */
while (shm->running_childs > 0) {
unsigned int alive;
/* Make sure there's no dead kids lying around.
* We need to do this in case the oom killer has been killing them,
* otherwise we end up stuck here with no child processes.
*/
alive = reap_dead_kids();
if (alive == 0)
return;
/* Ok, some kids are still alive. 'help' them along with a SIGKILL */
for_each_child(i) {
pid_t pid;
pid = shm->children[i]->pid;
if (pid == EMPTY_PIDSLOT)
continue;
kill(pid, SIGKILL);
}
/* wait a second to give kids a chance to exit. */
sleep(1);
if (shm_is_corrupt() == TRUE)
return;
}
/* Just to be sure, clear out the pid slots. */
for_each_child(i) {
shm->children[i]->pid = EMPTY_PIDSLOT;
}
}
static void watchdog(void)
{
static const char watchdogname[17]="trinity-watchdog";
static unsigned long lastcount = 0;
bool watchdog_exit = FALSE;
while (shm->ready == FALSE) {
usleep(1);
if (shm->exit_reason != STILL_RUNNING)
return;
}
output(0, "Watchdog is alive. (pid:%d)\n", watchdog_pid);
prctl(PR_SET_NAME, (unsigned long) &watchdogname);
(void)signal(SIGSEGV, SIG_DFL);
while (watchdog_exit == FALSE) {
int ret = 0;
unsigned int i;
if (shm_is_corrupt() == TRUE)
goto corrupt;
if (check_main_alive() == FALSE)
goto main_dead;
reap_dead_kids();
check_all_locks();
if (syscalls_todo && (shm->stats.total_syscalls_done >= syscalls_todo)) {
output(0, "Reached limit %d. Telling children to exit.\n", syscalls_todo);
panic(EXIT_REACHED_COUNT);
}
for_each_child(i) {
struct childdata *child = shm->children[i];
struct syscallrecord *rec = &child->syscall;
check_child_progress(child);
if (rec->op_nr > hiscore)
hiscore = rec->op_nr;
}
if (shm->stats.total_syscalls_done > 1) {
if (shm->stats.total_syscalls_done - lastcount > 10000) {
output(0, "%ld iterations. [F:%ld S:%ld HI:%ld]\n",
shm->stats.total_syscalls_done,
shm->stats.failures, shm->stats.successes,
hiscore);
lastcount = shm->stats.total_syscalls_done;
}
}
/* Only check taint if the mask allows it */
if (kernel_taint_mask != 0) {
ret = check_tainted();
if (((ret & kernel_taint_mask) & (~kernel_taint_initial)) != 0)
tainted_postmortem(ret);
}
main_dead:
/* Are we done ? */
if (shm->exit_reason != STILL_RUNNING) {
/* Give children a chance to exit. */
sleep(1);
/* Are there still children running ? */
if (pidmap_empty() == TRUE)
watchdog_exit = TRUE;
else {
static unsigned int last = 0;
if (last != shm->running_childs) {
last = shm->running_childs;
output(0, "exit_reason=%d, but %d children still running.\n",
shm->exit_reason, shm->running_childs);
}
kill_all_kids();
}
}
sleep(1);
}
corrupt:
kill_all_kids();
}
static void watchdog(void)
{
static unsigned long lastcount = 0;
bool watchdog_exit = FALSE;
while (watchdog_exit == FALSE) {
int ret = 0;
unsigned int i;
unsigned int stall_count = 0;
if (shm_is_corrupt() == TRUE)
goto corrupt;
if (check_main_alive() == FALSE)
goto main_dead;
reap_dead_kids();
check_all_locks();
if (syscalls_todo && (shm->stats.total_syscalls_done >= syscalls_todo)) {
output(0, "Reached limit %d. Telling children to exit.\n", syscalls_todo);
panic(EXIT_REACHED_COUNT);
}
for_each_child(i) {
struct childdata *child = shm->children[i];
struct syscallrecord *rec = &child->syscall;
if (is_child_making_progress(child) == FALSE)
stall_count++;
if (rec->op_nr > hiscore)
hiscore = rec->op_nr;
}
if (stall_count == shm->running_childs)
stall_genocide();
if (shm->stats.total_syscalls_done > 1) {
if (shm->stats.total_syscalls_done - lastcount > 10000) {
char stalltxt[]=" STALLED:XXXX";
if (stall_count > 0)
sprintf(stalltxt, " STALLED:%u", stall_count);
output(0, "%ld iterations. [F:%ld S:%ld HI:%ld%s]\n",
shm->stats.total_syscalls_done,
shm->stats.failures, shm->stats.successes,
hiscore,
stall_count ? stalltxt : "");
lastcount = shm->stats.total_syscalls_done;
}
}
/* Only check taint if the mask allows it */
if (kernel_taint_mask != 0) {
ret = check_tainted();
if (((ret & kernel_taint_mask) & (~kernel_taint_initial)) != 0)
tainted_postmortem(ret);
}
main_dead:
/* Are we done ? */
if (shm->exit_reason != STILL_RUNNING) {
/* Give children a chance to exit. */
sleep(1);
/* Are there still children running ? */
if (pidmap_empty() == TRUE)
watchdog_exit = TRUE;
else {
static unsigned int last = 0;
if (last != shm->running_childs) {
last = shm->running_childs;
output(0, "exit_reason=%d, but %d children still running.\n",
shm->exit_reason, shm->running_childs);
}
kill_all_kids();
}
}
sleep(1);
}
corrupt:
kill_all_kids();
}
