Re: [PATCH 1/5] netfilter: ipv6: move POSTROUTING invocation before fragmentation

[Patrick McHardy <kaber@trash.net>]

Jan Engelhardt wrote:
> On Thursday 2010-04-01 13:17, Patrick McHardy wrote:
>>> (2010/03/31 19:31), Jan Engelhardt wrote:
>>>> Patrick McHardy notes: "We used to invoke IPv4 POST_ROUTING after
>>>> fragmentation as well just to defragment the packets in conntrack
>>>> immediately afterwards, but that got changed during the
>>>> netfilter-ipsec integration. Ideally IPv6 would behave like IPv4."
>>>>
>>>> This patch makes it so. Sending an oversized frame (e.g. `ping6
>>>> -s64000 -c1 ::1`) will now show up in POSTROUTING as a single skb
>>>> rather than multiple ones.
>>> I am not in favor doing this
>>> because we theoretically make fragments __before__ routing
>>> in output path (as we reassemble __after__ routing in input path).
>> That's true, but is symetry for fragment handling really something
>> worth keeping? Besides avoiding one refragmentation pass in conntrack,
> 
> I am not quite following where this extra refragmentation is
> happening.

This was supposed to read "one more *de*fragmentation pass. In
IPv6 we don't have to refragment, but simply output the original
fragments.

> Assuming [nf-packet-flow.png] as a base, there are two
> spots in which conntrack/defrag happens: PREROUTING and OUTPUT.
> 
> This translates to:
> 
>> its a lot easier to construct your ruleset when you don't have to
>> take care of fragments.
> 
> We never see fragments in the ruleset
> 
>  a) for netif_rx received packets, defrag will be run early
>     (yes, there's raw, but that's special anyway)
> 
>  b) locally-generated packets are fragmented only after all of
>     Netfilter is done.

You're assuming conntrack is used.