PTU causes process to exit prematurely

[charmoniumQ]

This issue can be reproduced in a Docker container running Ubuntu 22.04 (e.g., docker run -it --rm ubuntu:22.04). Run the following script (or copy/paste each step):

set -ex

apt update
apt install -y curl git time proftpd gcc cmake make nano proftpd proftpd-mod-vroot

# Build provenance-to-use
git clone --recursive https://github.com/depaul-dice/provenance-to-use.git
cd provenance-to-use
# This patch is needed to compile provenance-to-use!
curl https://patch-diff.githubusercontent.com/raw/depaul-dice/provenance-to-use/pull/4.patch | patch --strip=1
./run.sh -r
export PATH=$PWD:$PATH
cd ..

# Set up Profptd srv directory
mkdir --parents proftpd/srv
echo Hello World > proftpd/srv/test

# Write Proftpd config file
USER=$(id --user --name)
GROUP=$(id --group --name)
echo $USER $GROUP
cat <<EOF > proftpd/conf
DefaultAddress   0.0.0.0
Port             54123
User             $USER
Group            $GROUP
DelayTable       "$PWD/proftpd/delay"
ScoreboardFile   "$PWD/proftpd/scoreboard"
PidFile          "$PWD/proftpd/proftpd.pid"
TransferLog      "$PWD/proftpd/xferlog"
DefaultRoot      "$PWD/proftpd/srv"
LoadFile         "/lib/proftpd/mod_vroot.so"
LoadModule       mod_vroot.c
VRootEngine       on
WtmpLog           off
RequireValidShell off
<Anonymous $PWD/proftpd/srv/>
    User $USER
    Group $GROUP
    UserAlias anonymous $USER
    <Directory *>
        AllowAll
    </Directory>
</Anonymous>
EOF

Make sure the script completes successfully. Then we will test the behavior of a script with and without PTU. The script is

proftpd --nodaemon --config $PWD/proftpd/conf &
pid=$! 
sleep 5
kill $pid
wait $pid

The script starts ProFTPD in the background, records its PID, waits for 5 seconds, kills ProFTPD, and waits for it to finish shutting down In a realistic setting, sleep 5 would be replaced with some FTP client activity (e.g., a load test). This script should take a little more than five seconds of "real" time, which should be shown on the third field of the last line. Indeed it is when we run natively (no PTU):

root@af7254134e60:/# cd PROBE/benchmark
root@af7254134e60:/PROBE/benchmark# \time sh -c 'proftpd --nodaemon --config $PWD/proftpd/conf & pid=$! && sleep 5 && kill $pid && wait $pid'
2024-07-04 22:06:24,378 af7254134e60 proftpd[22159] 0.0.0.0: ProFTPD 1.3.8 (stable) (built Tue Jan 1 1980 00:00:00 UTC) standalone mode STARTUP
2024-07-04 22:06:29,776 af7254134e60 proftpd[22159] 0.0.0.0: ProFTPD killed (signal 15)
2024-07-04 22:06:29,776 af7254134e60 proftpd[22159] 0.0.0.0: ProFTPD 1.3.8 standalone mode SHUTDOWN
0.00user 0.00system 0:05.40elapsed 0%CPU (0avgtext+0avgdata 3200maxresident)k
0inputs+144outputs (0major+640minor)pagefaults 0swaps
root@af7254134e60:/PROBE/benchmark#

The nix shell <pkgs>... --command <cmd>... simply runs <cmd>... in a reproducible environment containing <pkgs>....

However if we do the same script with ptu, the script does not wait 5 seconds. The whole script completes in a fraction of a second. This means that the client load, represented by sleep 5 would not get a chance to connect to the FTP server.

root@af7254134e60:/PROBE/benchmark# \time ptu sh -c 'proftpd --nodaemon --config $PWD/proftpd/conf & pid=$! && sleep 5 && kill $pid && wait $pid'
Provenance log file: /PROBE/benchmark/cde-package/provenance.cde-root.5.log
Command terminated by signal 11
0.00user 0.01system 0:00.07elapsed 17%CPU (0avgtext+0avgdata 2688maxresident)k
0inputs+1416outputs (0major+523minor)pagefaults 0swaps
root@af7254134e60:/PROBE/benchmark# 2024-07-04 22:09:33,001 af7254134e60 proftpd[22204] 0.0.0.0: ProFTPD 1.3.8 (stable) (built Tue Jan 1 1980 00:00:00 UTC) standalone mode STARTUP
root@af7254134e60:/PROBE/benchmark# 
root@af7254134e60:/PROBE/benchmark# 
root@af7254134e60:/PROBE/benchmark# 2024-07-04 22:09:38,337 af7254134e60 proftpd[22204] 0.0.0.0: ProFTPD killed (signal 15)
2024-07-04 22:09:38,338 af7254134e60 proftpd[22204] 0.0.0.0: ProFTPD 1.3.8 standalone mode SHUTDOWN
root@af7254134e60:/PROBE/benchmark# 

Note that the prompt returns to the user as soon as the output of time is finished (line beginning with 0inputs), at which point, I pressed <enter> to get new blank prompts. It seems that the tracee process (sh -c '...') continues to execute, since we see its output but perhaps gets orphaned by the ptu process, which already terminated.