There is an ongoing discussion on the QUIC issue tracker that might
benefit from some attention here.

https://github.com/quicwg/base-drafts/issues/1834

Of most interest here is that there is no mechanism by which a TLS
endpoint can limit the size of handshake messages.  This wasn't a
problem prior to TLS 1.3, renegotiation notwithstanding, but if an
endpoint needs to buffer a 16Mb message it will probably be unhappy.
And many TLS implementations buffer.  Knowing where short of 16Mb is
safe to send is tricky.  Right now, all we do is rely on the natural
pressure to keep tickets small and the fact that tickets are pure
overhead to keep things under control, but it's a hole that is a
little irritating to leave unpatched.

Of secondary concern for QUIC and DTLS, but not TLS, is the total
amount of data that can be sent in handshake messages.  If that data
arrives out of order, it would be nice to bound the amount of memory
that is apportioned for receiving and reassembling handshake messages.
I mention this only because the same solution might be used to fix
both problems.

(I was motivated to post this after reading the ticket requests draft
from Chris, which we should publish.  That draft doesn't run really
cause this problem, but it made me think that 255 tickets might make a
bit of a mess if sent all at once.)

_______________________________________________
TLS mailing list
[email protected]
https://www.ietf.org/mailman/listinfo/tls

Reply via email to