From cac9023a361a6ad3e98134cdbb6ceeb68304d047 Mon Sep 17 00:00:00 2001
From: swenzel <sandro.wenzel@cern.ch>
Date: Wed, 28 May 2025 18:30:19 +0200
Subject: [PATCH] o2-sim: Harden sigkill

---
 run/o2sim_parallel.cxx | 22 ++++++++++++++++++----
 1 file changed, 18 insertions(+), 4 deletions(-)

diff --git a/run/o2sim_parallel.cxx b/run/o2sim_parallel.cxx
index 24be9743e6a03..3e28428938b20 100644
--- a/run/o2sim_parallel.cxx
+++ b/run/o2sim_parallel.cxx
@@ -352,7 +352,9 @@ void launchShutdownThread()
     }
     LOG(info) << "Shutdown timer expired ... force killing remaining children";
     for (auto p : gChildProcesses) {
-      killpg(p, SIGKILL);
+      if (p != 0 && killpg(p, 0) == 0) { // see if process still exists
+        killpg(p, SIGKILL);
+      }
     }
   };
   threads.push_back(std::thread(lambda));
@@ -440,6 +442,12 @@ int main(int argc, char* argv[])
   // we enable the forked version of the code by default
   setenv("ALICE_SIMFORKINTERNAL", "ON", 1);
 
+  // force execution as own process group
+  if (setpgid(0, 0) == -1) {
+    perror("setpgid");
+    exit(1);
+  }
+
   TStopwatch timer;
   timer.Start();
   auto o2env = getenv("O2_ROOT");
@@ -703,7 +711,9 @@ int main(int argc, char* argv[])
           if (!shutdown_initiated) {
             shutdown_initiated = true;
             for (auto p : gChildProcesses) {
-              killpg(p, SIGTERM);
+              if (killpg(p, 0) == 0) {
+                killpg(p, SIGTERM);
+              }
             }
           }
         } else {
@@ -733,7 +743,9 @@ int main(int argc, char* argv[])
         LOG(info) << "Problem detected (or child received termination signal) ... shutting down whole system ";
         for (auto p : gChildProcesses) {
           LOG(info) << "TERMINATING " << p;
-          killpg(p, SIGTERM); // <--- makes sure to shutdown "unknown" child pids via the group property
+          if (killpg(p, 0) == 0) {
+            killpg(p, SIGTERM); // <--- makes sure to shutdown "unknown" child pids via the group property
+          }
         }
         LOG(error) << "SHUTTING DOWN DUE TO SIGNALED EXIT IN COMPONENT " << cpid;
         o2::simpubsub::publishMessage(externalpublishchannel, o2::simpubsub::simStatusString("O2SIM", "STATE", "FAILURE"));
@@ -751,7 +763,9 @@ int main(int argc, char* argv[])
     for (auto p : gChildProcesses) {
       if (p != mergerpid) {
         LOG(info) << "SHUTTING DOWN CHILD PROCESS (normal thread)" << p;
-        killpg(p, SIGTERM);
+        if (killpg(p, 0) == 0) {
+          killpg(p, SIGTERM);
+        }
       }
     }
   }