[RFC][PATCH 2/4] sunrpc: Use utsnamespaces

Serge E. Hallyn serue at us.ibm.com
Tue Jan 6 16:01:32 PST 2009


Quoting Chuck Lever (chuck.lever at oracle.com):
> On Jan 6, 2009, at Jan 6, 2009, 3:02 PM, Serge E. Hallyn wrote:
>> Quoting Matt Helsley (matthltc at us.ibm.com):
>>> We can often specify the UTS namespace to use when starting an RPC  
>>> client.
>>> However sometimes no UTS namespace is available (specifically during 
>>> system
>>> shutdown as the last NFS mount in a container is unmounted) so fall
>>> back to the initial UTS namespace.
>>
>> So what happens if we take this patch and do nothing else?
>
> I thought the point of this was to prevent incorrect container nodenames 
> from leaking onto the network.

But define incorrect.  If container B does an nfs mount, container c
is launched with a tree in that mount, container B dies, and container C
umounts it.  Should the umount belong to container B (for having
mounted it), container C (for having umount it), or the init_utsname
(for being the physical host/kernel)?

I get the feeling that consensus on this thread is that init_utsname
is actually the best choice, but OTOH if I have 3 containers on my
host, for apache, mysql, and postfix servers, and each is doing
nfs mounts from a physically remote machine, maybe I care about
having them report separate nodenames?

(that's a question, I really don't know...)

-serge


More information about the Containers mailing list