The first one is trivial. The second is slightly trickier. Libvirt has this virCommand subsystem (found under src/util/vircommand.c) which it uses to spawn commands. The subsystem allows libvirt to both daemonize a process and learn its PID. The latter of course does not work if command daemonizes itself (hence libvirt will not use qemu-pr-helper --daemon). BUT, we still need qemu-pr-helper to lock the pidfile (so that when libvirt tries to lock it it is denied). NB, libvirt uses fnctl(cmd = F_SETLK) to lock pidfiles. So leaking locked FD into qemu-pr-helper is not an option because that doesn't survive fork().
Another, trivial reason might be that if I run: qemu-pr-helper --pidfile /tmp/pr.pid (without --daemon), I'd still expect qemu-pr-helper to write pidfile because I told it to. Michal Privoznik (2): qemu-pr-helper: Daemonize before dropping privileges qemu-pr-helper: Write pidfile more often scsi/qemu-pr-helper.c | 20 ++++++++++++-------- 1 file changed, 12 insertions(+), 8 deletions(-) -- 2.16.1
