Google filtering ICMPv6 type 2?

[Erik Kline]

2010/1/4 Mark Kamichoff <prox at prolixium.com>:
> Hi -
>
> I noticed today that HTTP requests to ipv6.google.com (and
> www.google.com with Google-over-IPv6) seemed to be unusually laggy and
> frequently timing out.  I looked a little deeper and saw that Google may
> be inadvertently dropping ICMPv6 type 2 (packet too big) messages.  I'm
> seeing things like this for TCP packets that are attempting to enter a
> tunnel (MTU 1280) I have setup between me and a box of mine with native
> IPv6 connectivity:
>
> 22:21:44.513137 IP6 2001:4860:800e::69.80 >
> 2001:48c8:1:105:219:d1ff:fe21:d70d.49530: . 1:1429(1428) ack 119 win 90
> <nop,nop,timestamp 1528748730 1102342799>
> 22:21:44.513197 IP6 2001:48c8:1:2::2 > 2001:4860:800e::69: ICMP6, packet
> too big, mtu 1280, length 1240
> 22:21:44.513263 IP6 2001:4860:800e::69.80 >
> 2001:48c8:1:105:219:d1ff:fe21:d70d.49530: . 1429:2857(1428) ack 119 win
> 90 <nop,nop,timestamp 1528748730 1102342799>
> 22:21:44.513298 IP6 2001:48c8:1:2::2 > 2001:4860:800e::69: ICMP6, packet
> too big, mtu 1280, length 1240
> 22:21:44.513388 IP6 2001:4860:800e::69.80 >
> 2001:48c8:1:105:219:d1ff:fe21:d70d.49530: P 2857:4280(1423) ack 119 win
> 90 <nop,nop,timestamp 1528748730 1102342799>
>
> Sorry if that's hard to read, the full packet capture is here:
>
> http://www.prolixium.com/files/mynews/googleipv6.cap.txt
>
> I've tested this from the above setup (in NYC) and from an HE tunnel to
> Ashburn, VA.  Both experience similar issues.  However, I did another
> test from a friend's box who has an HE tunnel to Chicago, IL, and didn't
> see a problem.
>
> Anyone else seeing this?  Perhaps Google put an erroneous filter on some
> of their peering routers, but not others?
>
> Hopefully it'll get yanked.  I'd hate to go back to IPv4 for searching
> the web!
>
> - Mark
>
> --
> Mark Kamichoff
> prox at prolixium.com
> http://www.prolixium.com/
>

This is an issue that's been reported and we're looking in to right
now.  Sometimes pMTUd works, sometimes not, as you note.  Hopefully
we'll have some useful fix soon.

-Erik