Re: [BEHAVE] Fwd: IPv6 hosts sending <1280 byte packets

["Templin, Fred L" <[e-mail]>]

Hi Iljitsch,

> -----Original Message-----
> From: Iljitsch van Beijnum [mailto:[e-mail]]
> Sent: Tuesday, February 09, 2010 12:26 PM
> To: Templin, Fred L
> Cc: Dan Wing; [e-mail]
> Subject: Re: [BEHAVE] Fwd: IPv6 hosts sending <1280 byte packets
> 
> On 9 feb 2010, at 20:10, Templin, Fred L wrote:
> 
> > Another concern I should have called out earlier is what
> > if the IPv6 host is on a 1500 link and opens a TCP
> > connection to the IPv4 server with MSS=1440. The IPv4
> > server sends a 1480b IPv4 packet that gets translated
> > into a 1500b IPv6 packet. The packet makes its way
> > through the IPv6 network toward the IPv6 host, but
> > first it hits a 1480 MTU link (e.g., and IPv6 over IPv4
> > tunnel). The ICMPv6 PTB gets triggered, then translated
> > into ICMPv4 fragmentation needed, then dropped by
> > a filtering gateway in the IPv4 Internet.
> 
> > Black hole, right?
> 
> Indeed. Note however that the same black hole would exist if both ends are IPv6 or both ends are IPv4
> and there is no translator.

Eh? For IPv6<->IPv6, we have been assured that PMTUD
is fixed. Is that not the case?

Also, for IPv4-to-IPv4 we have the convenient fact that
links with MTU<1500 are rare in the Internet core so
tripping over a low MTU link is a rare event. Granted
that does not get us on the path to 9KB MTUs, but at
least it works in some fashion through good fortune.

> I don't think we can reasonably expect NAT64s to fix all possible black
> holes.

It's not a matter of fixing all black holes; it is
rather an analysis of what new sorts of black holes
are made possible when translators are used.

> In practice, I think NAT64 users would have to avoid reduced MTUs between the translator and the IPv6
> hosts (and I would assume that a deployment that needs NAT64 is far enough along that it doesn't need
> IPv6-in-IP tunneling)

AFAICT, it may still be some time before ISP networks can
enable ubiquitous native IPv6. For those that can't there
is IPv6-in-IPv4 tunneling for the near term. Does that
mean that those ISPs can't deploy translators?

> or artificially lower the IPv6 host MTU or the translator MTU / rewrite the MSS
> if a reduced MTU is present between the translator and the IPv6 hosts.

Artificially lowering the IPv6 host MTU may be unsatisfactory
if the host is connected to a network where it could otherwise
be using larger packet sizes. And, MSS clamping only works
for TCP.

Fred
[e-mail]