- From: Poul-Henning Kamp <phk@phk.freebsd.dk>
- Date: Fri, 15 Aug 2014 18:54:01 +0000
- To: Martin Thomson <martin.thomson@gmail.com>
- cc: Erik Nygren <erik@nygren.org>, Mark Nottingham <mnot@mnot.net>, HTTP Working Group <ietf-http-wg@w3.org>
-------- In message <CABkgnnXc9Di3-eLSrbhFDTwGkPmmnig67x-3-t0fCTM3c2YTFQ@mail.gmail.com> , Martin Thomson writes: >On 15 August 2014 07:50, Poul-Henning Kamp <phk@phk.freebsd.dk> wrote: >> I was talking only conceptually here, the actual protocol mechanics >> will be at least as tricky as HTTP/1 -> HTTP/2 upgrade. > >I've talked with several people about HTTP/1.1 and the mechanisms >we've defined. They are portable - in theory. ...and in theory that means that they will work in practice :-) >Incidentally, the difference between good crypto and bad crypto here >is that good crypto is fast and secure and bad crypto is just bad. What is good and what is bad crypto, depends a LOT on what your ephemeral lifetime requirement is,. And incidentally, cryptographers have a very strong tendency to forget that all crypto is ephemeral in the first place, because the usually unstated requirement is that the ephemeral lifetime be longer than the probable duration of the universe. For the isolated problem of defeating PM, the ephemeral lifetime requirement is a few seconds, reducing the customary 128+ bits to something on the order of 32. Good crypto which only lasts a couple of seconds would be terribly bad crypto in almost any other context. But conversely, a lot of crypto deemed bad or even "horrible" in normal contexts will be perfectly good crypto for the purpose of defeating PM. I'm not taking a position on what specific algorithms would or wouldn't be suitable here, but I will point out that it is much more important that it does 10Gbit/s on commodity hardware, so loadbalancers and such can keep up, than it be able to survive systematic attack for more than few seconds. If it wasn't because of the almost-always-known-plaintest property of HTTP headers, a simple XOR with a 32bit random number would be enough. -- Poul-Henning Kamp | UNIX since Zilog Zeus 3.20 phk@FreeBSD.ORG | TCP/IP since RFC 956 FreeBSD committer | BSD since 4.3-tahoe Never attribute to malice what can adequately be explained by incompetence.
Received on Friday, 15 August 2014 18:54:27 UTC