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

[sukadev at us.ibm.com]

Eric W. Biederman [ebiederm at xmission.com] wrote:
| 
| Guys how complete do you fee the pid namespace support is that
| has been merged into Linus's tree?
| 
| My impression until I started reading through code earlier today
| was that the support was just about done except for a couple of
| tricky details.

The only thing that I know is pending is the issue of signalling
container-init. We have not been able to find a clean fix for it.

The problem now is that a process in a child namespace can terminate
its container-init and thereby the entire container. We have a 3-patch
set (Oleg's and mine) that kind of addresses this.  The scenario where
the patchset fails is :

	- the container-init has a blockable, fatal signal blocked 

	- a descendant of the container-init posts the fatal signal to
	  container-init.

	- container-init then unblocks the signal without ignoring or
	  handling the signal.

In this case again the container-init can be terminated. 

(by fatal I mean a signal whose default action is to terminate the process
SIGKILL is of couse not blockable and is not a problem)

This issue can be addressed in user-space by the container-init - which
should just ignore the fatal signal or setup a handler for it.

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.

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