Re: PMTU discovery is broken on kernel 3.7.1 for UDP sockets

[Steffen Klassert <steffen.klassert@secunet.com>]

On Thu, Dec 20, 2012 at 01:35:35PM +0100, Steffen Klassert wrote:
> On Thu, Dec 20, 2012 at 03:22:13PM +0400, Yurij M. Plotnikov wrote:
> > On 12/20/12 11:34, Steffen Klassert wrote:
> > >
> > >diff --git a/net/ipv4/ip_sockglue.c b/net/ipv4/ip_sockglue.c
> > >index 3c9d208..1049ce0 100644
> > >--- a/net/ipv4/ip_sockglue.c
> > >+++ b/net/ipv4/ip_sockglue.c
> > >@@ -1198,7 +1198,7 @@ static int do_ip_getsockopt(struct sock *sk, int level, int optname,
> > >  	{
> > >  		struct dst_entry *dst;
> > >  		val = 0;
> > >-		dst = sk_dst_get(sk);
> > >+		dst = sk_dst_check(sk, 0);
> > >  		if (dst) {
> > >  			val = dst_mtu(dst);
> > >  			dst_release(dst);
> > With this patch kernel 3.7.1 works perfect. All described problems
> > are fixed.
> 
> Thanks for testing!
> 
> I'm not sure if we can't use this as a fix. I think with this patch it
> could happen that we return -ENOTCONN instead of a pmtu value on a
> connected socket. Perhaps it is better to update the cached dst_entry in
> ipv4_sk_update_pmtu() when we receive the -EMSGSIZE. I'll do some
> investigation.
> 

It turned out that updating the cached dst_entry in ipv4_sk_update_pmtu()
is not trivial. We need to implement proper socket locking and we need
socket release calback functions for all protocols that use
ipv4_sk_update_pmtu(), similar to tcp.

Today is my last office day for this year, so we probably have to defer
a solution to the next year.

Thanks.