Re: Need your help: different connection IDs in the same datagram

[David Schinazi <[e-mail]>]

On Wed, Jul 15, 2020 at 3:49 PM Martin Thomson <[e-mail]> wrote:

> On Thu, Jul 16, 2020, at 07:17, Ian Swett wrote:
> > I forgot there's a contradiction, because I thought we disallowed
> > sending mixed-CID packets.  So I'd prefer changing to MUST NOT coalesce
> > packets with different CIDs.  Even if you generate them and then
> > coalesce them(which we do), that's not that hard to enforce in the
> > coalescing code.
>
> OK, this is good information.  I had understood that you were concerned
> that you might violate a requirement to have consistent connection IDs.
>
> Here's one example of how you might:
>
> Client receives server's first flight, which comprises Initial, Handshake,
> 1-RTT (which includes NEW_CONNECTION_ID, because this is a good time to
> send that in a lot of cases).  Assume this all comes at once, in a single
> datagram.
>
> 1. Client processes Initial and Handshake.
> 2. Client generates a Handshake packet (that includes TLS Finished) and
> sets that aside.  This uses the connection ID from the servers Initial.
> 3. Client processes 1-RTT.  Sees NEW_CONNECTION_ID, decides to use it.
> 4. Client generates first application data, and makes a 1-RTT packet using
> the new connection ID.
> 5. Client gathers all generated packets and sends them.
>
> Does anyone care about doing this, or something like it?


I think we should disallow this unless someone can show it provides real
benefit.
If an implementation supports coalescing, it has to handle cases where the
last
packet doesn't fit so it needs to have code to for example send the
handshake
packet by itself if the 1-RTT packet is too big. Once one has such code,
it's pretty
simple to add a check that also triggers this if the connection IDs differ.
If someone
really cares about not losing the benefits of coalescing, then they can
make sure
that they don't act on NEW_CONNECTION_ID until the handshake is confirmed.


> Because if not, I make a new proposal that changes the requirements to:
>
> * Senders MUST ensure that all coalesced packets include the same value
> for the Destination Connection ID field.
> * Receivers SHOULD discard coalesced packets if they are for a different
> connection than the first; senders MAY discard coalesced packets that
> contain a different value in the Destination Connection ID field than the
> first.
>

I support this proposal.