[Devel] Re: Q: How complete is the pid namespace in mainline

[Eric W. Biederman]

sukadev at us.ibm.com writes:

> Eric W. Biederman [ebiederm at xmission.com] wrote:
> | sukadev at us.ibm.com writes:
> | >
> | > Dave had suggested we print a warning the first time a container-init
> forks()
> | > without a handler for a fatal signal. I was planning on adding that as
> | > patch 4 of the signal patch set and get some feedback.
> | 
> | Yes.  How to cleanly handle signalling of container init is
> | a tricky one.  It does sound like you have made a reasonable start
> | there.
> | 
> | Suka it is a lot more then that.  How much more I'm not certain
> | of.  I suspect the only way to find the rest of the cases is
> | just go through the code with a fine tooth come and read and look.
>
> I agree. I did not mean to ignore the kthread conversions and was only
> referring to the core pid namespace clone stuff.

Sure, and that make sense.

> | So far doing that it has not at all hard for me to find either
> | bugs or places where the implementation can be improved.
> | 
> | Currently we have little things like kill(-1,...) signalling the
> | wrong set of processes, and a couple of proc bugs.
>
> I just realized the fix for this is in the signal patchset I was
> referring to.
>
> https://lists.linux-foundation.org/pipermail/containers/2007-August/006987.html
>
> I notice that you have sent a patch for the kill -1.

Yes. I'm trying to get out as many simple little bug fixes
as I can.

Sorry for missing the fact you guys had generated some patches
to address this.  Still I think mine is a little more comprehensive
and shorter ;)

That bug is on my list of really nasty bugs I want to avoid.

> The proc_mnt bug Linus found seems to have slipped through when
> merging Pavel's and my patches.

I really don't mind a handful of little bugs, it would be
surprising if something hadn't slipped through at this point.

As long as everyone is aware that it is going to take a bit
to find everything and stabilizing it all and everyone keeps
looking we should be fine.

Oh.  Do you know if there was a good reason for forcing
the tty, session, and process group of a the first process
in a pid namespace?

Eric
_______________________________________________
Containers mailing list
Containers at lists.linux-foundation.org
https://lists.linux-foundation.org/mailman/listinfo/containers