Shim logging seems to cause memory corruption #2457
Description
When running the following simulation, shadow gets stuck (0% CPU usage, memory and disk usage don't noticeably increase). Shadow is built in debug mode in an Ubuntu 22.04 container.
apt install npm
npm install http-server -g
rm -rf shadow.data && shadow --strace-logging-mode standard --log-level trace shadow.yaml > shadow.log
general:
stop_time: 10s
model_unblocked_syscall_latency: true
network:
graph:
type: 1_gbit_switch
hosts:
server:
network_node_id: 0
processes:
- path: /usr/bin/node
args: /usr/local/bin/http-server -p 80 -d
start_time: 3s
client:
network_node_id: 0
processes:
- path: /usr/bin/curl
args: server
start_time: 5sThe simulation completes successfully if the --log-level trace is changed to --log-level debug, or if an early return is added to shimlogger_log.
If you debug the node plugin in gdb:
break _managedthread_fork_exec
handle SIGSYS noprint
run
set follow-fork-mode child
cont
the backtrace after shadow becomes stuck shows:
(gdb) bt
#0 0x00007ffff541c9ca in shim_native_syscallv () from /tmp/shadow/build/src/lib/shim/libshadow-shim.so
Backtrace stopped: Cannot access memory at address 0x7ffff2a8c9f8