build_library/qemu_template.sh: use POSIX kill -0 for swtpm cleanup watcher for macOS+linux portability

[shipitdev]

Description

(reopening (#4231) since i deleted the head of the last fork)
Use POSIX kill -0 for swtpm cleanup watcher to ensure cross-platform compatibility
In build_library/qemu_template.sh, the background monitor responsible for cleaning up the swtpm daemon process checks for parent termination by probing the /proc filesystem:

(while [ -e "/proc/${PARENT}" ]; do sleep 1; done; kill "${SWTPM_PROC}" 2>/dev/null; exit 0) &

On macOS (Darwin), the /proc filesystem does not exist. Because of this, [ -e "/proc/${PARENT}" ] immediately returns false in 0 seconds, which causes the watcher to kill swtpm before QEMU can connect to the socket. Running with software TPM (-T) on macOS immediately crashes with Failed to connect to TPM socket: Connection refused.

Solution

Replace /proc/${PARENT} with the POSIX standard kill -0 "${PARENT}".

kill -0 checks if the parent process ID is alive without sending a terminating signal:

While the launcher script is running ➡️ kill -0 returns 0 (true) and keeps swtpm alive.
When the launcher script exits ➡️ kill -0 returns non-zero (false) and cleanly kills swtpm to avoid orphaned processes.
This makes the cleanup logic fully portable across Linux, macOS, and BSD.

How to use

Run the QEMU launcher with a TPM directory (-T):

./flatcar_production_qemu_uefi.sh -T /tmp/swtpm-state ...

Testing done

1. macOS: Verified that kill -0 "${PARENT}" keeps the daemon process alive while the parent script runs, and cleanly terminates it within 1 second after the parent exits.

# 1. Start mock background daemon
$ sleep 30 & DAEMON_PID=$!

# 2. Run parent script with the POSIX kill -0 watcher
$ (
    PARENT=$$
    (while kill -0 "${PARENT}" 2>/dev/null; do sleep 1; done; kill "${DAEMON_PID}" 2>/dev/null) &
    echo "Parent running (PID $PARENT)... Daemon alive? $(kill -0 $DAEMON_PID 2>/dev/null && echo YES || echo NO)"
    sleep 2
    echo "Parent exiting..."
  )

Parent running (PID 8524)... Daemon alive? YES
Parent exiting...

# 3. Verify daemon was cleanly terminated by watcher after parent exit
$ sleep 1
$ kill -0 $DAEMON_PID 2>/dev/null && echo "Daemon alive" || echo "Daemon killed successfully"
Daemon killed successfully

2. Linux / POSIX: Verified that kill -0 is POSIX compliant (kill(2)) and functions identically across Linux distros.

• Changelog entries added in the respective changelog/ directory (user-facing change, bug fix, security fix, update)
• Inspected CI output for image differences: /boot and /usr size, packages, list files for any missing binaries, kernel modules, config files, kernel modules, etc.

[chewi]

The fix looks fine, although when I try it on a Mac myself, QEMU dies immediately with this:

qemu-system-aarch64: -device tpm-tis-device,tpmdev=tpm0: Error: ret = HV_BAD_ARGUMENT (0xfae94003, at ../accel/hvf/hvf-all.c:123)

I guess you didn't hit that.

[shipitdev]

The fix looks fine, although when I try it on a Mac myself, QEMU dies immediately with this:

qemu-system-aarch64: -device tpm-tis-device,tpmdev=tpm0: Error: ret = HV_BAD_ARGUMENT (0xfae94003, at ../accel/hvf/hvf-all.c:123)

I guess you didn't hit that.

Ah yeah, I ran into that! With all the different Apple Silicon HVF quirks popping up, I didn't want to flood you with a whole laundry list of issues on day one lol, so I kept the scope of this PR strictly focused on fixing the swtpm daemon watcher and /proc portability.

i'm not sure where to proceed from here though Would you like me to add a quick note to the PR description mentioning the limitation or something ?

Although it looks like another upstream QEMU issue with tpm-tis-device under HVF. i'm gonna have a look at it later today

[chewi]

It's fine, we can just take this now, especially if it's just an upstream issue. Most users don't need a TPM anyway.