Re: [Cfrg] [TLS] Closing out tls1.3 "Limits on key usage" PRs (#765/#769).

[Andrey Jivsov <[e-mail]>]

I also think that counting in blocks is cleaner. Counting in bytes is a
close alternative.

I have a few nits below:

On 03/02/2017 01:44 PM, Brian Smith wrote:
> Aaron Zauner <[e-mail]> wrote:
>> I'm not sure that text on key-usage limits in blocks in a spec
>> that fundamentally deals in records is less confusing, quite
>> the opposite (at least to me).
> 
> 1. Consider an implementation that negotiates with another
> implementation to use a very large record size such as 1MB records. If
> the limit is specified in terms of records then the limit would need
> to be readjusted to the new max record size, or else the new extension
> is potentially unsafe to use. This shows that specifying the limits in
> terms of records is brittle.

Hm, while I agree that counting in wrong units is the main problem here,
we over-count with records, which means that we re-key faster than needed.

> 2. If it is only safe to use an AES-GCM key for a certain number of
> blocks, where in the code is the best place to enforce the limit on
> the number of blocks? IMO, it is better to enforce it in the AES-GCM
> implementation itself, underneath the TLS layer. In that case the
> limit is best expressed in terms of the number of blocks. Specifying
> the limit in terms of records would be optimizing for implementations
> that enforce the limit at the wrong layer of abstraction.

Given that the event of reaching the limit requires TLS-level action,
checking the limit at the TLS level seems reasonable. The lower-level
block-cipher code should still have the logic to error out (using a more
relaxed limit).

> 
>> As I pointed out earlier: I strongly recommend that any changes
>> to the spec are as clear als possible to engineers
>> (non-crypto/math people) -- e.g. why the spec is suddenly
>> dealing in blocks instead of records et cetera. Again; I really
>> don't see any reason to change text here - to me all suggested
>> changes are even more confusing.
> 
> Given a limit in blocks, the arithmetic to keep track of the number of
> blocks is trivial, and very similar to the arithmetic that's already
> needed to split up a large byte stream into records and keep track of
> the record sequence number.

Also please consider counting in bytes. This may be more natural,
especially with AES-GCM that doesn't expose higher levels to the block
size granularity (as much as CBC does, anyway).

> 
> Cheers,
> Brian
>