Re: deprecating Postel's principle- considered harmful

[Paul Wouters <[e-mail]>]

On Wed, 8 May 2019, Dave Cridland wrote:

>       Many UDP encapsulations of IP packets do not recalculate the outer UDP
>       checksum. It's a good thing we accept these datagrams with technical
>       errors.
> 
> There's two observations to be made here, if I understand correctly:
> 
> a) The lack of properly checking the outer UDP checksum means that implementations could avoid recalculating it.
> 
> b) We could not enforce such checking now, because of such implementations.

It is generally signaled by setting the checksum to 0.  I guess by now,
packet mangling has become so rare, that router implementers prefered
not to check at all and leave it up to the endpoint. We only know what
some of the well known endpoints do.

> I appreciate what you're saying, but it's unclear if either is a good thing.

To me, it makes sense that the endpoints are the only ones checking
these, but I'm not a router vendor and not really qualfied to speak
on these issues. So like you, I'm on the fence on whether this is a
good thing or not. But if it _is_ a good thing, than we only got there
via the Postel principle.

Paul