Re: [TLS] Collisions (Re: Consensus Call: FNV vs SHA1)

["Kemp, David P." <[e-mail]>]

In general, protocols and standardized data structures may make use of
non-cryptographic hash functions in cases where collisions are
anticipated and well-handled: i.e. as optimizations (optional) rather
than as integrity (required) mechanisms.  PKIX AKI/SKI is an example:

   "The authority key identifier extension provides a means of
   identifying the public key corresponding to the private key 
   used to sign a CRL. ... This extension is especially useful
   where an issuer has more than one signing key, either due to
   multiple concurrent key pairs or due to changeover."

PKIX suggests two SHA-1 -based methods of generating SKIs (including the
4 bit constant 0100 followed by the 60 least significant bits of the
SHA-1 hash of a public key), and says "Other methods of generating
unique numbers are also acceptable."

The same reflexive "broken algorithm" evaluation concerns may exist for
this use of SHA-1, and FNV would perform the function just as well
without raising any eyebrows.

If a collision did occur, an implementation would have to do an extra
signature validation if it first picked the wrong certificate to
validate a signature.  Not much of a penalty, and one which would have
to be handled anyway for the case where AKI/SKI is not populated at all.

Dave



----- Nico wrote: ----

Aside from that, I think FNV is a great hash function.  Its place will
typically not be in Internet protocols, but in implementations.

[*] I _can_ think of non-cryptographic uses of hash functions in
    Internet protocols, starting with the use we made in GS2.  But in
    general I think we'll want to review such uses carefully.  In the
    case of GS2 it saved us the bother of creating a registry of SASL
    mechanism names for GSS-API mechanisms.  Here the hash is used to
    identify cached items, which seems fine provided that collisions are
    well-handled.  In the GS2 case we decided that collisions were not a
    problem for us (after all, we could always assign new OIDs to GSS
    mechanisms that collide).