Re: [PATCH] netfilter: per netns nf_conntrack_cachep

[Jon Masters]

On Wed, 2010-02-03 at 21:09 +0200, Alexey Dobriyan wrote:
> On Wed, Feb 03, 2010 at 01:38:09PM -0500, Jon Masters wrote:
> > *). Per namespace cacheing allocation (the cachep bits). We know it's
> > still possible for weirdness to happen in the SLAB cache here.
> 
> Tiny race, needs reproducer.

Maybe. I think it's worth fixing anyway.

> > *). Per namespace hashsize tracking. Existing code corrupts hashtables
> > if the global size is changed when there is more than one netns
> 
> I think, no.
> Changing hash size will change hashsize for all netns, current and future.

Nope. Look at the logic in nf_conntrack_set_hashsize where you iterate
over init_net.ct.hash but don't touch other namespaces. So then you go
setting nf_conntrack_htable_size and will deference that in accessing
other per-namespace hashtables using the wrong size information.

> > *). Per namespace expectations. This is for similar reasons to the need
> > for multiple hashtables, though I haven't poked at that.
> 
> Expectation cache is not SLAB_DESTROY_BY_RCU, so the logic doesn't
> apply, I hope.

I'm not sure, I didn't poke at that much yet. But hashtables certainly
need fixing unless you can point out where I'm wrong.

> > I also think it is necessary to expose net namespace layout
> 
> Not necessary. Why?

How am I as a sysadmin supposed to figure out which net namespaces exist
on my system, and as a developer, supposed to debug these situations?

Jon.


--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/