Shadow deadlocks when managed process aborts #1715
Description
test.c:
#include <stdlib.h>
int main() {
abort();
}shadow.yaml:
general:
stop_time: 2 min
network:
graph:
type: 1_gbit_switch
hosts:
server:
network_node_id: 0
processes:
- path: test
start_time: 1
When running with:
shadow -l debug shadow.yaml
Shadow gets stuck at:
00:00:00.047023 [22763:shadow-worker] 00:00:01.000000000 [INFO] [server:11.0.0.1] [thread_preload.c:200] [threadpreload_run] forking new thread with environment 'SHADOW_SPAWNED=TRUE SHADOW_LOG_START_TIME=29754246233 SHADOW_LOG_LEVEL=4 SHADOW_INTERPOSE_METHOD=PRELOAD LD_PRELOAD=[snip] SHADOW_LOG_FILE=[snip] SHADOW_IPC_BLK=8,134217728,648,/shadow_shmemfile_29754.293148005-22755 SHADOW_USE_SECCOMP= SHADOW_PID=22755 SHADOW_TSC_HZ=2592000000', arguments '/home/steve/deadlock/test', and working directory '/home/steve/deadlock/shadow.data/hosts/server'
00:00:00.047101 [22763:shadow-worker] 00:00:01.000000000 [DEBUG] [server:11.0.0.1] [thread_preload.c:148] [_threadpreload_fork_exec] started process /home/steve/deadlock/test with PID 22776
00:00:00.047104 [22763:shadow-worker] 00:00:01.000000000 [INFO] [server:11.0.0.1] [process.c:522] [_process_start] process 'server.test.1000' started
00:00:00.047106 [22763:shadow-worker] 00:00:01.000000000 [DEBUG] [server:11.0.0.1] [process.c:576] [process_continue] switching to thread controller to continue executing process 'server.test.1000'
This bug only exists in preload mode.