Serge E. Hallyn wrote:
> Here is a small patchset I've been sitting on for awhile
> to make signaling mostly subject to user namespaces.  In
> particular,
>
> 	1. store user_namespace in user struct
> 	2. introduce CAP_NS_OVERRIDE
> 	3. require CAP_NS_OVERRIDE to signal another user namespace
>
> The first step should have been done all along.  Else wouldn't
> a hash collision on (ns1, uid) and (ns2, uid), however unlikely,
> give us wrong results at uid_hash_find()?

Unless I've completely misunderstood the code, each namespace has a
separate hash. Please correct me if I'm wrong.

> The main remaining signaling+userns issue is of course the
> siginfo.  Tacking a userns onto siginfo is a pain due to
> lifetime mgmt issues.  I haven't decided whether to just
> catch all the callers and fake uid=0 if user namespaces
> aren't the same, introduce some unique non-refcounted id to
> represent (user,user_ns), or find some other way to deal with
> it.
>
> thanks,
> -serge

-- 
Daniel Hokka Zakrisson
_______________________________________________
Containers mailing list
Containers@lists.linux-foundation.org
https://lists.linux-foundation.org/mailman/listinfo/containers