/* Generate children*/
static void fork_children(void)
{
while (shm->running_childs < max_children) {
int childno;
int pid = 0;
if (shm->spawn_no_more == TRUE)
return;
/* a new child means a new seed, or the new child
* will do the same syscalls as the one in the child it's replacing.
* (special case startup, or we reseed unnecessarily)
*/
if (shm->ready == TRUE)
reseed();
/* Find a space for it in the pid map */
childno = find_childno(EMPTY_PIDSLOT);
if (childno == CHILD_NOT_FOUND) {
outputerr("## Pid map was full!\n");
dump_childnos();
exit_main_fail();
}
fflush(stdout);
pid = fork();
if (pid == 0) {
/* Child process. */
init_child(childno);
child_process();
debugf("child %d %d exiting.\n", childno, getpid());
close_logfile(&this_child->logfile);
_exit(EXIT_SUCCESS);
} else {
if (pid == -1) {
/* We failed, wait for a child to exit before retrying. */
if (shm->running_childs > 0)
return;
output(0, "couldn't create child! (%s)\n", strerror(errno));
panic(EXIT_FORK_FAILURE);
exit_main_fail();
}
}
shm->children[childno]->pid = pid;
shm->running_childs++;
debugf("Created child %d (pid:%d) [total:%d/%d]\n",
childno, pid, shm->running_childs, max_children);
if (shm->exit_reason != STILL_RUNNING)
return;
}
shm->ready = TRUE;
debugf("created enough children\n");
}
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;
}
static FILE * find_child_logfile_handle(pid_t pid)
{
int i;
unsigned int j;
FILE *log = NULL;
i = find_childno(pid);
if (i != CHILD_NOT_FOUND) {
log = shm->children[i]->logfile;
} else {
/* This is pretty ugly, and should never happen,
* but try again a second later, in case we're racing setup/teardown.
* FIXME: We may not even need this now that we have proper locking; test it.
*/
sleep(1);
i = find_childno(pid);
if (i == CHILD_NOT_FOUND) {
outputerr("Couldn't find child for pid %d\n", pid);
return mainlogfile;
}
log = shm->children[i]->logfile;
}
if (log != NULL)
return log;
/* if the logfile hadn't been set, log to main. */
shm->children[i]->logfile = mainlogfile;
outputerr("## child %d logfile handle was null logging to main!\n", i);
outputerr("## Couldn't find logfile for pid %d\n", pid);
dump_childnos();
outputerr("## Logfiles for pids: ");
for_each_child(j)
outputerr("%p ", shm->children[j]->logfile);
outputerr("\n");
(void)fflush(stdout);
sleep(5);
return mainlogfile;
}
/*
* 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;
}
static FILE * find_logfile_handle(void)
{
pid_t pid;
int i;
pid = getpid();
if (pid == initpid)
return mainlogfile;
if (pid == shm->mainpid)
return mainlogfile;
if (pid == watchdog_pid)
return mainlogfile;
i = find_childno(pid);
if (i != CHILD_NOT_FOUND)
return shm->children[i]->logfile;
else {
/* try one more time. FIXME: This is awful. */
unsigned int j;
sleep(1);
i = find_childno(pid);
if (i != CHILD_NOT_FOUND)
return shm->children[i]->logfile;
outputerr("## Couldn't find logfile for pid %d\n", pid);
dump_childnos();
outputerr("## Logfiles for pids: ");
for_each_child(j)
outputerr("%p ", shm->children[j]->logfile);
outputerr("\n");
}
return NULL;
}
static void handle_child(pid_t childpid, int childstatus)
{
switch (childpid) {
case 0:
//debugf("Nothing changed. children:%d\n", shm->running_childs);
break;
case -1:
if (shm->exit_reason != STILL_RUNNING)
return;
if (errno == ECHILD) {
unsigned int i;
bool seen = FALSE;
debugf("All children exited!\n");
for_each_child(i) {
struct childdata *child;
child = shm->children[i];
if (child->pid != EMPTY_PIDSLOT) {
if (pid_alive(child->pid) == -1) {
debugf("Removing %d from pidmap\n", child->pid);
child->pid = EMPTY_PIDSLOT;
shm->running_childs--;
} else {
debugf("%d looks still alive! ignoring.\n", child->pid);
}
seen = TRUE;
}
}
if (seen == FALSE)
shm->running_childs = 0;
break;
}
output(0, "error! (%s)\n", strerror(errno));
break;
default:
debugf("Something happened to pid %d\n", childpid);
if (WIFEXITED(childstatus)) {
int childno;
childno = find_childno(childpid);
if (childno == CHILD_NOT_FOUND) {
/* If we reaped it, it wouldn't show up, so check that. */
if (shm->last_reaped != childpid) {
outputerr("## Couldn't find %d in list of pids.\n", childpid);
panic(EXIT_LOST_CHILD);
dump_childnos();
}
} else {
debugf("Child %d exited after %ld operations.\n",
childpid, shm->children[childno]->syscall.op_nr);
reap_child(childpid);
}
break;
} else if (WIFSIGNALED(childstatus)) {
handle_childsig(childpid, childstatus, FALSE);
} else if (WIFSTOPPED(childstatus)) {
handle_childsig(childpid, childstatus, TRUE);
} else if (WIFCONTINUED(childstatus)) {
break;
} else {
output(0, "erk, wtf\n");
}
}