[Edm] fuzzy grease

["Brian Trammell (IETF)" <[e-mail]>]

Greetings, all,

Expanding on the point I improved at the microphone during Monday’s meeting: IMO we should lean into "Variation of protocol extension points with positive use in mind ... can be thought of, to some extent, as protocol fuzzing.” 

It occurred to me that greasing *is* a form of fuzzing, with a somewhat different target and a different system under test. A sender’s greasing fuzzes, inline, the distributed system made up of the receiver and anything on path. Greasing encryption-protected codepoints, such that the wire image is protected from on-path inspection, results in more conventional fuzzing of the remote endpoint.

(Indeed one could say that all protocol exchanges, even without greasing, are a degenerate form of fuzzing, given that all implementations are at least a little buggy. But that’s a philosophical point.)

Okay, cool. Is this insight useful? 

I’d thought that there might be some possibility to take existing tools and techniques for fuzzing "off the shelf”, but on some review it’s not clear that there’s a “unified practice of fuzzing” beyond various implementations of “put a bunch of random bits (for some definition of random) here (for various definitions of here)”. So greasing with assumed on-path observers is basically just distributed-system fuzzing with a definition of “random" and "here" that is restricted to either greasing-reserved codepoints, or codepoints we don’t think are reserved.

So maybe the advice coming from this insight is “where greasing is designed to be less random than fuzzing, we should have a good reason why” (which is itself a special case of “where the wire image contains observable structure beyond pure entropy, we should have a good reason why”, which the IAB has more or less already said).

Cheers,

Brian