child: refactor command execution to use goroutines with Pdeathsig

[fahedouch]

fixing #182

better controller for child Pdeathsig

[AkihiroSuda]

Thanks, how to test this?

[fahedouch]

you can test with the following script:

@#!/bin/bash

# Test script to verify Pdeathsig behavior using rootlesskit itself
# This script:
# 1. Uses rootlesskit to spawn a long-running process
# 2. Kills the rootlesskit parent process
# 3. Verifies that the child process is killed as expected

set -e

echo "Starting Pdeathsig test using rootlesskit..."

# Create a temporary directory for test artifacts
TEMP_DIR=$(mktemp -d)
echo "Created temporary directory: $TEMP_DIR"

# Create a marker file that will be touched by the child process if it's still alive
MARKER_FILE="$TEMP_DIR/child_still_alive"

# Create a script that will be executed by rootlesskit
CHILD_SCRIPT="$TEMP_DIR/child_script.sh"
cat > "$CHILD_SCRIPT" << 'EOF'
#!/bin/bash
echo "Child process started with PID: $$"
echo "Parent PID: $PPID"

# Register a trap to handle signals
trap 'echo "Child received signal, exiting"; exit 1' TERM INT

# Run for 30 seconds, checking if parent is still alive
for i in {1..30}; do
    echo "Child still running (iteration $i)..."
    
    # Check if parent has changed (died)
    CURRENT_PPID=$(ps -o ppid= -p $$)
    if [ "$CURRENT_PPID" != "$PPID" ]; then
        echo "Parent changed from $PPID to $CURRENT_PPID"
        if [ "$CURRENT_PPID" = "1" ]; then
            echo "Parent is now init (PID 1), parent has died"
            echo "Child should be killed by Pdeathsig, but if you see this message, it wasn't"
            touch MARKER_FILE_PLACEHOLDER
            exit 1
        fi
    fi
    
    sleep 1
done

# If we reach here, the child wasn't killed
echo "Child completed normally (this shouldn't happen if Pdeathsig is working)"
touch MARKER_FILE_PLACEHOLDER
EOF

# Replace the placeholder with the actual marker file path
sed -i "s|MARKER_FILE_PLACEHOLDER|$MARKER_FILE|g" "$CHILD_SCRIPT"
chmod +x "$CHILD_SCRIPT"

# Start rootlesskit with the child script
echo "Starting rootlesskit..."
rootlesskit "$CHILD_SCRIPT" &
ROOTLESSKIT_PID=$!
echo "Rootlesskit started with PID: $ROOTLESSKIT_PID"

# Wait a moment for the child to start
sleep 2

# Find the child process
CHILD_PID=$(pgrep -P $ROOTLESSKIT_PID)
if [ -z "$CHILD_PID" ]; then
    echo "Failed to find child process"
    exit 1
fi
echo "Found child process with PID: $CHILD_PID"

# Kill the rootlesskit process
echo "Killing rootlesskit process (PID: $ROOTLESSKIT_PID)..."
kill -9 $ROOTLESSKIT_PID

# Wait a moment for the child to be killed
sleep 2

# Check if the child process is still running
if ps -p $CHILD_PID > /dev/null; then
    echo "FAIL: Child process (PID: $CHILD_PID) is still running after parent was killed"
    kill -9 $CHILD_PID  # Clean up
    exit 1
else
    echo "PASS: Child process (PID: $CHILD_PID) was killed as expected"
fi

# Check if the marker file exists
if [ -f "$MARKER_FILE" ]; then
    echo "FAIL: Marker file exists, which means the child process wasn't killed by Pdeathsig"
    exit 1
else
    echo "PASS: Marker file doesn't exist, which means the child process was killed by Pdeathsig"
fi

echo "Test completed successfully!"
rm -rf "$TEMP_DIR"
exit 0

I can convert this script into integration test if needed.

[AkihiroSuda]

integration test

Yes, please

rootlesskit/.github/workflows/main.yaml

Lines 40 to 54 in e83d763

	- name: "Integration test: exit-code"
	run: docker run --rm --privileged rootlesskit:test-integration ./integration-exit-code.sh
	- name: "Integration test: propagation"
	run: docker run --rm --privileged rootlesskit:test-integration ./integration-propagation.sh
	- name: "Integration test: propagation (with `mount --make-rshared /`)"
	run: docker run --rm --privileged rootlesskit:test-integration sh -exc "sudo mount --make-rshared / && ./integration-propagation.sh"
	- name: "Integration test: restart"
	run: docker run --rm --privileged rootlesskit:test-integration ./integration-restart.sh
	- name: "Integration test: port"
	# NOTE: "--net=host" is a bad hack to enable IPv6
	run: docker run --rm --net=host --privileged rootlesskit:test-integration ./integration-port.sh
	- name: "Integration test: IPv6 routing"
	run: docker run --rm --privileged --sysctl net.ipv6.conf.all.disable_ipv6=0 rootlesskit:test-integration ./integration-ipv6.sh
	- name: "Integration test: systemd socket activation"
	run: docker run --rm --net=none --privileged rootlesskit:test-integration ./integration-systemd-socket.sh

[fahedouch]

The script works on the workstation but not in the CI environment; I suspect a timing issue.

[fahedouch]

@AkihiroSuda new changes are introduced:

1. Modified child.go to ensure consistent Pdeathsig behavior regardless of reaper setting:

   • Added a goroutine with runtime.LockOSThread() for the non-reaper case (opt.Reaper is false)
   • Explicitly set Pdeathsig with unix.Prctl(unix.PR_SET_PDEATHSIG, uintptr(unix.SIGKILL), 0, 0, 0) in both cases
   • Created a new runWithoutReap() function to handle command execution without reaping

2. Enhanced integration-pdeathsig.sh to:

   • Test both reaper settings (true and false) to verify consistent behavior

new changes ensure that child processes are properly terminated when the parent is killed with SIGKILL, regardless of whether the reaper functionality is enabled.

test fail not related to PR

[AkihiroSuda]

Thanks

[AkihiroSuda]

Regression:

• v3.0.0-alpha.0 regression: bash: cannot set terminal process group (-1): Inappropriate ioctl for devic #557