[PATCH 2/4] signals: send_signal: use si_fromuser() to detect from_ancestor_ns

Roland McGrath roland at redhat.com
Tue Oct 6 11:01:04 PDT 2009


> Yes sure, init can't kill itself with or without these changes. But,
> I think this is supposed behaviour which we do not want to change?

I agree.  (Actually, I think init shouldn't be "protected" that way at all.
You shoot downwards, you get your foot.  But we are not talking about
changing the global init behavior, and I do think that the container-init
behavior should be as consistent as possible with what global init sees.)

It's possible to meaningfully swallow a kill() signal because then nothing
happens at all.  The exec failure cases are special because all the damage
is already done so there is no way to avoid dying, and SIGKILL is just
making it more formal and graceful.


Thanks,
Roland


More information about the Containers mailing list