static unsigned int reap_dead_kids(void)
{
unsigned int i;
unsigned int alive = 0;
unsigned int reaped = 0;
for_each_child(i) {
struct childdata *child;
pid_t pid;
int ret;
child = shm->children[i];
pid = child->pid;
if (pid == EMPTY_PIDSLOT)
continue;
if (pid_is_valid(pid) == FALSE) {
static bool once = FALSE;
if (once != FALSE)
return 0;
output(0, "Sanity check failed! Found pid %u at pidslot %u!\n", pid, i);
dump_childnos();
if (shm->exit_reason == STILL_RUNNING)
panic(EXIT_PID_OUT_OF_RANGE);
dump_childdata(child);
once = TRUE;
return 0;
}
ret = kill(pid, 0);
/* If it disappeared, reap it. */
if (ret == -1) {
if (errno == ESRCH) {
output(0, "pid %u has disappeared. Reaping.\n", pid);
reap_child(pid);
reaped++;
} else {
output(0, "problem checking on pid %u (%d:%s)\n", pid, errno, strerror(errno));
}
} else {
alive++;
}
if (shm->running_childs == 0)
return 0;
}
if (reaped != 0)
output(0, "Reaped %d dead children\n", reaped);
return alive;
}
/*
* Make sure various entries in the shm look sensible.
* We use this to make sure that random syscalls haven't corrupted it.
*
* also check the pids for sanity.
*/
static int shm_is_corrupt(void)
{
unsigned int i;
// FIXME: The '500000' is magic, and should be dynamically calculated.
// On startup, we should figure out how many getpid()'s per second we can do,
// and use that.
if (shm->stats.total_syscalls_done - shm->stats.previous_op_count > 500000) {
output(0, "Execcount increased dramatically! (old:%ld new:%ld):\n",
shm->stats.previous_op_count, shm->stats.total_syscalls_done);
panic(EXIT_SHM_CORRUPTION);
return TRUE;
}
shm->stats.previous_op_count = shm->stats.total_syscalls_done;
for_each_child(i) {
struct childdata *child;
pid_t pid;
child = shm->children[i];
pid = child->pid;
if (pid == EMPTY_PIDSLOT)
continue;
if (pid_is_valid(pid) == FALSE) {
static bool once = FALSE;
if (once != FALSE)
return TRUE;
output(0, "Sanity check failed! Found pid %u at pidslot %u!\n", pid, i);
dump_childnos();
if (shm->exit_reason == STILL_RUNNING)
panic(EXIT_PID_OUT_OF_RANGE);
dump_childdata(child);
once = TRUE;
return TRUE;
}
}
return FALSE;
}
/*
* Make sure various entries in the shm look sensible.
* We use this to make sure that random syscalls haven't corrupted it.
*
* also check the pids for sanity.
*/
static int shm_is_corrupt(void)
{
unsigned int i;
if (shm->stats.op_count < shm->stats.previous_op_count) {
output(0, "Execcount went backwards! (old:%ld new:%ld):\n",
shm->stats.previous_op_count, shm->stats.op_count);
panic(EXIT_SHM_CORRUPTION);
return TRUE;
}
shm->stats.previous_op_count = shm->stats.op_count;
for_each_child(i) {
struct childdata *child;
pid_t pid;
child = shm->children[i];
pid = pids[i];
if (pid == EMPTY_PIDSLOT)
continue;
if (pid_is_valid(pid) == FALSE) {
static bool once = FALSE;
if (once != FALSE)
return TRUE;
output(0, "Sanity check failed! Found pid %u at pidslot %u!\n", pid, i);
dump_childnos();
if (shm->exit_reason == STILL_RUNNING)
panic(EXIT_PID_OUT_OF_RANGE);
dump_childdata(child);
once = TRUE;
return TRUE;
}
}
return FALSE;
}
