Lock thread_pool.wait_mutex before forking to avoid a race condition …

…between rad_fork, rad_waitpid and reap_children.

There is a race condition that can occur under high load where a child is reaped before being added to the waiters list.

src/main/threads.c

@@ -1142,6 +1142,7 @@ pid_t rad_fork(void)
 	/*
 	 *	Fork & save the PID for later reaping.
 	 */
+	pthread_mutex_lock(&thread_pool.wait_mutex);
 	child_pid = fork();
 	if (child_pid > 0) {
 		int rcode;
@@ -1152,16 +1153,20 @@ pid_t rad_fork(void)
 
 		tf->pid = child_pid;
 
-		pthread_mutex_lock(&thread_pool.wait_mutex);
 		rcode = fr_hash_table_insert(thread_pool.waiters, tf);
-		pthread_mutex_unlock(&thread_pool.wait_mutex);
 
 		if (!rcode) {
 			radlog(L_ERR, "Failed to store PID, creating what will be a zombie process %d",
 			       (int) child_pid);
 			free(tf);
 		}
 	}
+	/*
+	 *	Do not unlock in child process
+	 */
+	if(child_pid != 0 ) {
+		pthread_mutex_unlock(&thread_pool.wait_mutex);
+	}
 
 	/*
 	 *	Return whatever we were told.