Re: CVE-2024-40761: Apache Answer: Avatar URL leaked user email addresses

[Jeffrey Walton <noloader () gmail com>]

On Wed, Sep 25, 2024 at 5:45 PM Goldberg, Adam <Adam.Goldberg () sony com> wrote:
> > On Wed, Sep 25, 2024 at 06:28:16AM +0000, Enxin Xie wrote:
> > > Using the MD5 value of a user's email to access Gravatar is insecure and can lead to the leakage of user email. 
> > > The official recommendation is to use SHA256 instead.
>
> > For practical purposes, this sounds like almost no change to me.  I've
> > just checked and 
> > https://urldefense.com/v3/__https://docs.gravatar.com/api/avatars/hash/__;!!JmoZiZGBv3RvKRSx!6zoU_J4wgUshOcGT7WCRwWgz0hjESorDYcuCX8cOARG6zrVpuLHmeayYJmf2ZnIO1QaQVFfeopQ2u6GQ6g$
> >  does say:
>
> > > All URLs on Gravatar are based on the use of the hashed value of an
> > > email address. Images and profiles are both accessed via the hash of an
> > > email, and it is considered the primary way of identifying an identity
> > > within the system. To ensure a consistent and accurate hash, the
> > > following steps should be taken to create a hash:
> > >
> > > 1. Trim leading and trailing whitespace from an email address
> > > 2. Force all characters to lower-case
> > > 3. hash the final string with SHA256
>
> Note that this is a recommendation, "the following steps *should* ...", which doesn't require that those three steps 
> be taken.
>
> > So Gravatar URLs by design allow for quick checking of email addresses
> > against them, and thus allow to infer not-too-cryptic addresses.  Both
> > MD5 and SHA-256 are very fast, with speeds in many billion per second
> > per GPU, with SHA-256 being only a few times slower than MD5.  MD5's
> > cryptographic weaknesses are irrelevant to this use case.
>
> > So I think this CVE should either be rejected (as the issue is with
> > Gravatar, not with implementations) or considered unfixable (within
> > spec) and thus not fixed.
>
> See above, it seems to be an implementation issue (at least in part -- an application must take specific actions in 
> order to create the hash in a secure way).

I believe this is a use case for Aumasson and Bernstein's SipHash,
<https://eprint.iacr.org/2012/351>. Wikipedia has a nice description
of how SIpHash differs from a hash like SHA; see
<https://en.wikipedia.org/wiki/SipHash>.

Jeff