On Sun, Oct 13, 2013 at 09:51:03AM -0700, Eric Dumazet wrote:

On Sun, 2013-10-13 at 12:40 +0200, Steinar H. Gunderson wrote:

quoted

[resending with proper quoting and cc]

On Mon, Oct 07, 2013 at 05:09:10AM +0200, Hannes Frederic Sowa wrote:

quoted

Could you try to check (with e.g. nstat) if any of the following counters
change if the icmp messages hit the host?

TcpExtOutOfWindowIcmps
Icmp6InErrors
TcpExtTCPMinTTLDrop
TcpExtListenDrops

Hi,

I am now seeing the same problem against a 3.9.0 machine (that I've never had
problems with before). I looked through the four nstat counters, and none of
them are increasing. Here's /proc/net/ipv6_route from the server while it's
going on; the client is 2001:67c:a4:1:71f9:c84b:8b2a:c65b (second line) and
the server is 2001:67c:29f4::2007.

I am not sure, but the skb_is_gso() checks seem too lazy.

Why don't we check that gso_size+headers is below mtu, I dont know.

Was this intended for the UFO gso thread? ;)