Re: [PATCH RFC net-next] net: ipvs: Adjust gso_size for IPPROTO_TCP

[Julian Anastasov <ja@ssi.bg>]

	Hello,

On Mon, 7 May 2018, Martin KaFai Lau wrote:

> On Sat, May 05, 2018 at 03:58:25PM +0300, Julian Anastasov wrote:
> > 
> > 	So, except the RTF_LOCAL check in __ip6_rt_update_pmtu
> > we should have no other issues. Only one minor bit is strange to me,
> > why rt6_insert_exception warns for RTF_PCPU if rt6_cache_allowed_for_pmtu
> > allows it when returning true...
> hmm...I am not sure I follow this bits.  Where is the warn?

        if (ort->rt6i_flags & (RTF_CACHE | RTF_PCPU))
                ort = ort->from;

	Sorry, my fault, I missed above re-assignment...

        WARN_ON_ONCE(ort->rt6i_flags & (RTF_CACHE | RTF_PCPU));

> Note that "nrt6" and "from" are passed to rt6_insert_exception()
> instead of "rt6".
> 
> > 
> > 	Also, commit 0d3f6d297bfb allows rt6_do_update_pmtu() for
> > routes without RTF_CACHE, RTF_PCPU and rt6i_node. Should we
> > restrict rt6_do_update_pmtu only to RTF_CACHE routes?
> > 
> >  	if (!rt6_cache_allowed_for_pmtu(rt6)) {
> > -		rt6_do_update_pmtu(rt6, mtu);
> The existing rt6_do_update_pmtu() looks correct.
> The mtu of the dst created by icmp6_dst_alloc()
> needs to be udpated and this dst does not have
> the RTF_CACHE.

	Aha, ok. I thought, only RTF_CACHE routes can
hold PMTU.

Regards