On 15 nov 2013, at 12:14, Hannes Tschofenig
<hannes(_dot_)tschofenig(_at_)gmx(_dot_)net> wrote:
We mandate other things in protocol specification as well (that aim to take
performance, for example, to a specific level) then why not also certain
security features.
I'm all for mandating certain security features. A big part of that is not
having insecure stuff in version 1 of protocols, because once the can is open,
the worms never (completely) crawl back in. However, mandates in the form "if
you do X you may only do it in manner Y" don't mean much, because the IETF has
no real-world power beyond the text of its specifications.
That aside, just saying "you MUST do TLS with HTTP/2.0" doesn't buy much
security in a world where CAs are not trustworthy, people still use RC4/MD5,
use woefully short keys for otherwise strong algorithms, browsers have
effectively trained people to always click "visit anyway" and so on.
Also, there are cases where encryption isn't needed, and cases where it's not
desired. An example close to home: a while back, some IETF meeting related
page, the agenda perhaps, would only be available over HTTPS. And it was
extremely slow. Not sure if this was because the server was overloaded, session
keys were negotiated every time (which shouldn't be necessary) or the page
wouldn't be cached (which should be possible), but the end result was that a
static file that is available to everyone without credentials was much harder
to access for no benefit. Unless you pad all your pages as well as their URLs
to the same length, it's trivial for snoops to see who got which page simply by
observing the length of the exchange.
With security, the perfect tends to be the enemy of the good. Let's focus our
efforts on getting everything encrypted that needs to be encrypted, and do so
according to the state of the art rather than sloppily as often happens today,
rather than fight over whether people get to serve unencrypted stuff over
HTTP/1.1 or HTTP/2.0. "Encrypt everything" makes for a good soundbite, but it's
a terrible policy.