Re: [TLS] Consensus call on Implicit IV for AEAD

> On Apr 17, 2015, at 5:47 AM, Ilari Liusvaara <ilari.liusvaara@elisanet.fi> wrote:
> 
> On Fri, Apr 03, 2015 at 01:34:15PM -0700, Joseph Salowey wrote:
>> In the interim meeting we had consensus to use an implicit IV for AEAD.
>> The proposal was to use the record sequence number and pad with zeros as
>> described in pull request 155 (
>> https://github.com/tlswg/tls13-spec/pull/155/files).  This was also
>> discussed in the IETF-92 meeting in Dallas along with options to change the
>> offset.  The consensus was to stay with the original proposal.  We are
>> posting to the mailing list to confirm this consensus.
> 
> I really do not like padding with zeroes, since it would weaken
> encryption relative to TLS v1.2. And this is not just if the
> symmetric encryption is already broken, but applies to all[1].
> 
> This is especially prominent with AES-128[2]. And "Just use AES-256"
> does not solve the issue, since it will probably not be available
> in important environments[3].
> 
> None of TLS v1.2, IPSec nor SSH use AES-GCM this way, all have
> secret[4] fixed nonce-parts (SSH does use Chacha20 without per-
> connection nonce[5]).
> 
> 
> OTOH, the explicit per-record nonce field for AES-GCM seems really
> unnecressary. Everybody uses it sequentially anyway, so it just
> wastes 8 bytes per record.
> 
> 
> 
> [1] If the algorithm is broken, if the zeroes or even just known
> nonce further help breaking depends on type of attack (but of
> course, one rather not use such broken algorithms).[6]
> 
> [2] Substantial weakening happens far before collisions become
> likely. So much less than AGL's "2^60" mentioned in the meeting.

Really?  Can you provide a citation?  Because if I’m reading it correctly, this is a claim that AES-128 is not a secure PRP, with significantly less than 2^60 known/chosen plaintexts, keys and compute time.

> [3] And I suppose nobody really likes AES-192 (despite the fact
> that seemingly just about every CPU with "AES acceleration" is able
> to accelerate that too). :->

… and this is a claim that AES-192 and AES-256 are somehow substantially more secure than AES-128 as PRPs.

> [4] Secret meaning those are derived from the main key block.

… and this is a claim that adding an extra key, which is in the nonce instead of an actual key to the cipher, makes it more secure.

All this said, I agree that it’s wise to reduce the amount of cribtext you give an attacker, so long as it doesn’t cost too much complexity.  If we are really providing a fatal amount of cribtext, then we need to fix that, but then again, we should probably toss the cipher out at that point.

Cheers,
— Mike

> 
> [5] The encrypted stream looks like random noise, which makes it
> harder to tell where packet boundaries are (and Chacha20 is
> 256-bit keys anyway).
> 
> [6] SSLv3 had known nonces for EXPORT ciphers, but secret ones
> for strong ones. Hmm...
> 
> 
> 
> -Ilari
> 
> _______________________________________________
> TLS mailing list
> TLS@ietf.org
> https://www.ietf.org/mailman/listinfo/tls