Re: IQ Handling vulnerabilities

On Mon, Feb 10, 2014 at 7:33 PM, Joe Hildebrand (jhildebr) <
[email protected]> wrote:

> Or add a nonce to each round:
>
> nonce = crytpo_rand()
> start = sha1(nonce + crytpo_rand())
> start+1 = sha1(nonce + start)
>
> Regardless, there exist solutions.
>

At the risk of diving into what's quite clearly a bit of a bikeshed at this
point...

I wondered whether there's value in going a step further. What if an entity
constructed ids based on a random per-session key, the original "to" value,
and a suitable salt (which itself need not be cryptographically random),
such that the id was the salt followed by a (truncated, probably) HMAC
result as follows:

id = salt + hmac(salt + to, session-key)[0:10]

That way, on receipt of such an id in an iq response or error, the sender
could validate the id somewhat securely?

I suspect a single session-key is a little simplistic; but that's easy to
adapt.

The advantage here isn't really in the <iq/> case we've been mostly
discussing, but the other cases of a returned id - since we don't want to
track outbound directed presence, or message, ids for an arbitrary length
of time. (I assume).

Dave.

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