Re: [PATCH 1/4] netfilter: xtables: inclusion of xt_TEE

From: Patrick McHardy <kaber@trash.net>
To: Jan Engelhardt <jengelh@medozas.de>
Cc: netfilter-devel@vger.kernel.org
Subject: Re: [PATCH 1/4] netfilter: xtables: inclusion of xt_TEE
Date: Mon, 19 Apr 2010 14:42:02 +0200	[thread overview]
Message-ID: <4BCC4F9A.90709@trash.net> (raw)
In-Reply-To: <alpine.LSU.2.01.1004191434530.9177@obet.zrqbmnf.qr>

Jan Engelhardt wrote:
> On Monday 2010-04-19 14:20, Patrick McHardy wrote:
>>> +static bool
>>> +tee_tg_route4(struct sk_buff *skb, const struct xt_tee_tginfo *info)
>>> +{
>>> +	const struct iphdr *iph = ip_hdr(skb);
>>> +	struct net *net = pick_net(skb);
>>> +	struct rtable *rt;
>>> +	struct flowi fl;
>>> +
>>> +	memset(&fl, 0, sizeof(fl));
>>> +	if (!tee_tg_route_oif(&fl, net, info))
>>> +		return false;
>> Redoing the interface lookup once per packet is really highly
>> suboptimal. I'll change that to do the lookup once per new
>> rule or simply in userspace.
> 
> I thought about that too. But if you grab a reference to the dev on
> rule insertion, the rule would stop working when you down and up an
> interface, which does not match the regular iptables behavior (-i and
> -o options) at all.

Not down and up, unregister and register. But we don't need a
reference, just the ifindex. That can be updated on netdev events.