Linux Ethernet Bridging - Feb 2015 - [Bridge] [PATCH] bridge: make it possible for packets to traverse the bridge withour hitting netfilter

From: Imre Palik <imrep.amz at gmail.com>
Date: Tue, 10 Feb 2015 10:32:24 +0100
> From: "Palik, Imre" <imrep at amazon.de>
> 
> The netfilter code is made with flexibility instead of performance in mind.
> So when all we want is to pass packets between different interfaces, the
> performance penalty of hitting netfilter code can be considerable, even
when
> all the firewalling is disabled for the bridge.
> 
> This change makes it possible to disable netfilter both on a per bridge
basis,
> or for the whole bridging subsystem.  In the case interesting to us, this
can
> lead to more than 10% speedup compared to the case when only
bridge-iptables
> are disabled.
> 
> Cc: Anthony Liguori <aliguori at amazon.com>
> Signed-off-by: Imre Palik <imrep at amazon.de>
Sorry, no.

If I apply this, someone is going to try to submit a patch for every
damn protocol layer to add a stupid hack like this.

Makw NF_HOOK() faster instead.

On 02/11/15 23:29, David Miller wrote:
> From: Imre Palik <imrep.amz at gmail.com>
> Date: Tue, 10 Feb 2015 10:32:24 +0100
> 
>> From: "Palik, Imre" <imrep at amazon.de>
>>
>> The netfilter code is made with flexibility instead of performance in
mind.
>> So when all we want is to pass packets between different interfaces,
the
>> performance penalty of hitting netfilter code can be considerable, even
when
>> all the firewalling is disabled for the bridge.
>>
>> This change makes it possible to disable netfilter both on a per bridge
basis,
>> or for the whole bridging subsystem.  In the case interesting to us,
this can
>> lead to more than 10% speedup compared to the case when only
bridge-iptables
>> are disabled.
>>
>> Cc: Anthony Liguori <aliguori at amazon.com>
>> Signed-off-by: Imre Palik <imrep at amazon.de>
> 
> Sorry, no.
> 
> If I apply this, someone is going to try to submit a patch for every
> damn protocol layer to add a stupid hack like this.
Actually this is one of those patches.  There is already a "stupid hack
like this" for iptables and arptables.  (Implemented before git history,
and giving me 10% speedup.  Many thanks, whoever did it.)

I also searched various LKML archives, and it seems the existing "stupid
hacks" for iptables and arptables haven't resulted in any related patch
submission in the last ten years.  (Or my google-fu is weak.)

Moreover, I cannot imagine any other reasonable on/off switch for
bridge-netfilter than these three.  Of course, my imagination might be lacking
there.
> Makw NF_HOOK() faster instead.
As far as I see, flexibility implies trashing the cache/TLB.  So it is
impossible to have flexibility and performance at the same time.
(Except possibly with self modifying code, but that has its own set of
problems.)

Imre Palik <imrep.amz at gmail.com> wrote:
> On 02/11/15 23:29, David Miller wrote:
> > If I apply this, someone is going to try to submit a patch for every
> > damn protocol layer to add a stupid hack like this.
> 
> Actually this is one of those patches.  There is already a "stupid
hack like this" for iptables and arptables.  (Implemented before git
history, and giving me 10% speedup.  Many thanks, whoever did it.)
> 
> I also searched various LKML archives, and it seems the existing
"stupid hacks" for iptables and arptables haven't resulted in any
related patch submission in the last ten years.  (Or my google-fu is weak.)
> 
> Moreover, I cannot imagine any other reasonable on/off switch for
bridge-netfilter than these three.  Of course, my imagination might be lacking
there.
Why do you load the bridge netfilter module if you don't want it?
Loading it registers the internal hooks for the call-ip(6)tables and
sabotage hooks with NF_BRIDGE protocol so most of the NF_HOOK(NF_BRIDGE, ...
calls become active.