Linux Ethernet Bridging - Dec 2017 - [Bridge] [PATCH net-next] bridge: ebtables: Avoid resetting limit rule state

Hi Linus,

On Mon, Dec 04, 2017 at 05:53:35AM +0100, Linus L?ssing
wrote:
> Hi Pablo,
> 
> Thanks for your reply!
> 
> On Tue, Nov 28, 2017 at 12:30:08AM +0100, Pablo Neira Ayuso wrote:
> > [...]
> > > diff --git a/net/bridge/netfilter/ebt_limit.c
b/net/bridge/netfilter/ebt_limit.c
> > > index 61a9f1be1263..f74b48633feb 100644
> > > --- a/net/bridge/netfilter/ebt_limit.c
> > > +++ b/net/bridge/netfilter/ebt_limit.c
> > > @@ -69,6 +69,10 @@ static int ebt_limit_mt_check(const struct
xt_mtchk_param *par)
> > >  {
> > >  	struct ebt_limit_info *info = par->matchinfo;
> > >  
> > > +	/* Do not reset state on unrelated table changes */
> > > +	if (info->prev)
> > > +		return 0;
> > 
> > What kernel version are you using? I suspect you don't have this
> > applied?
> 
> I'm indeed using a 4.4.102 kernel, as LEDE is still in the process
> of updating to 4.14. So 4.4 with LEDE is where I got the measurement
> results from.
> 
> > 
> > commit ec23189049651b16dc2ffab35a4371dc1f491aca
> > Author: Willem de Bruijn <willemb at google.com>
> > Date:   Mon Jan 2 17:19:46 2017 -0500
> > 
> >     xtables: extend matches and targets with .usersize
> 
> And so, no I do not have this patch. I looked at it now, but it
> does not seem to have any relation with .matchinfo, does it?
> 
> I also had a quick look at a 4.15-rc1 kernel in a VM now. I still
> end up in ebt_limit_mt_check() with the variables being reset
> when editing the table somewhere.
My question is if your fix would work with 4.15-rc1.

On Thu, Dec 07, 2017 at 01:26:19AM +0100, Pablo Neira Ayuso
wrote:
> > I also had a quick look at a 4.15-rc1 kernel in a VM now. I still
> > end up in ebt_limit_mt_check() with the variables being reset
> > when editing the table somewhere.
> 
> My question is if your fix would work with 4.15-rc1.
You are absoluetly right, it's not working anymore since the
commit you mentioned initially :-(
("xtables: extend matches and targets with .usersize").
info->prev is always 0 since exactly this commit.

That means, trying tricks in ebt_limit_mt_check() is too
late now, the old values are already overwritten? (or is there
some commit scheme which installs the ebt_limit_info provided
by ebt_limit_check() some time after its call?)

Extending the usersize to include info->prev would probably be too
hackish/ugly, right?