G+: IPv6 connections to Google stallingThis has got me …

David Coles
IPv6 connections to Google stalling

This has got me totally flumoxed. I can't use most Google services since (especially GMail) since they hang during sign-in. Even Search has trouble.

Proto Recv-Q Send-Q Local Address           Foreign Address         State       PID/Program name
tcp6       0   5678 dcoles-t440s:56614      2404:6800:4004:80c::1002:https ESTABLISHED 29981/google-chrome

So far the solution has been to switch to Firefox and set 'network.dns.disableIPv6=true'.

Sadly this started happening after I returned to Tokyo from the US (after updating my system to Ubuntu 14.04beta), so not exactly sure what changed.
David Coles
Sounds a little similar to https://bugzilla.redhat.com/show_bug.cgi?id=1069633
Sphericus Spherium
IPv6 to google working well here. Plus and Hangouts at least.
posted via IPv6
David Coles
Yeah. I don't think it's a global issue otherwise there'd be much more complaints. So might be something very specific to work's DHCP6 based architecture.
Sphericus Spherium
Good luck, hope you guys get it fixed.
Jeremy Visser
90% of the time I experience random hanging with IPv6, it's an MTU or TCP MSS issue.
David Coles
Ah. Jackpot. Thanks +Jeremy Visser, that seems to have done it. Tested with tracepath6 and eventually set it to 1460 (odd number) and can browse again. Now just need to hunt down who's blocking ICMPv6 messages.
Sphericus Spherium
David, http://publib.boulder.ibm.com/infocenter/aix/v7r1/index.jsp?topic=%2Fcom.ibm.aix.prftungd%2Fdoc%2Fprftungd%2Ftcp_max_seg_size_tuning.htm explains the MSS/MTU issue.
Namely,  "Ethernet with a MTU of 1500 would result in a MSS of 1460 after subtracting 20 bytes for IPv4 header and 20 bytes for TCP header."
I suppose the follow on question is, did you set 1460 on the L2, L3 or L4 Layer?
David Coles
Device MTU (so on Ethernet/L2) layer. Hence my confusion since it's not a MTU I've seen before. That would give a MSS of 1420 for TCP over IPv4 (not sure about IPv6 due to the extension headers). Usually PPPoE gives a MTU of 1492, but I'd not encountered one of 1460  (though apparently happens on some PPPoE ISPs). Really I have to find what on earth I'm connected to and more curiously why did it break over the last month! Will have to try a 13.10 live USB to try and rule out the distro upgrade.
Sphericus Spherium
Well being on Internode, I know that their standard recommendation is L2 MTU of 1492, IP MTU of 1460 and TCP MSS of 1420.
Jeremy Visser
+David Coles The root cause is that the device with the MTU of 1460 is not doing TCP MSS clamping.

If it’s a Linux box, it’s a one-line change to iptables. (One-line change on Cisco, too.)