Re: [tcpm] PLPMTUD for all protocols

[William Herrin <[e-mail]>]

On Wed, Mar 28, 2018 at 4:43 AM, Mikael Abrahamsson <[e-mail]> wrote:
> In trying to advocate for all protocols implementing PLPMTUD (RFC4821) and
> shipping with it default on, I received pushback when I suggested/asked why
> modern TCP stacks don't come with PMTU blackhole detect turned on.

Hi Mikael,

My knee-jerk response reading the RFC is that the up-probe process
seems very brittle with layer violations all over the place.


> What's wrong with it, and how do we fix it?

Don't attempt a layer-3 general purpose solution that boils down to a
large set of dirty hacks? Spitballing here, but in TCP you could
create an option for the remote to respond with the largest received
fragment size when a fragmented packet is received and if the option
is negotiated then send the up-probes with the DF bit cleared. If the
remote doesn't support it, shrink the MSS but don't try to grow it
again. If the remote does respond with the largest fragment size then
you get the exact path MTU as a result of one probe without disrupting
the communication flow.

Sacrilege to suggest sending anything without the DF bit set, but if
the router's return has a black hole, maybe the answer is to let the
router communicate forward instead.

Regards,
Bill Herrin


-- 
William Herrin ................ [e-mail]  [e-mail]
Dirtside Systems ......... Web: <http://www.dirtside.com/>