Re: [openpgp] Disabling compression in OpenPGP

Jon Callas <jon@callas.org> Tue, 18 March 2014 18:19 UTC

Return-Path: <jon@callas.org>
X-Original-To: openpgp@ietfa.amsl.com
Delivered-To: openpgp@ietfa.amsl.com
Received: from localhost (ietfa.amsl.com [127.0.0.1]) by ietfa.amsl.com (Postfix) with ESMTP id BF38C1A044F for <openpgp@ietfa.amsl.com>; Tue, 18 Mar 2014 11:19:47 -0700 (PDT)
X-Virus-Scanned: amavisd-new at amsl.com
X-Spam-Flag: NO
X-Spam-Score: -1.901
X-Spam-Level:
X-Spam-Status: No, score=-1.901 tagged_above=-999 required=5 tests=[BAYES_00=-1.9, SPF_PASS=-0.001] autolearn=ham
Received: from mail.ietf.org ([4.31.198.44]) by localhost (ietfa.amsl.com [127.0.0.1]) (amavisd-new, port 10024) with ESMTP id ooGreNcuRXew for <openpgp@ietfa.amsl.com>; Tue, 18 Mar 2014 11:19:40 -0700 (PDT)
Received: from mail.merrymeet.com (merrymeet.com [173.164.244.100]) by ietfa.amsl.com (Postfix) with ESMTP id 6994B1A0453 for <openpgp@ietf.org>; Tue, 18 Mar 2014 11:15:32 -0700 (PDT)
Received: from localhost (localhost [127.0.0.1]) by mail.merrymeet.com (Postfix) with ESMTP id E97044F84D50; Tue, 18 Mar 2014 11:15:23 -0700 (PDT)
X-Virus-Scanned: amavisd-new at merrymeet.com
Received: from mail.merrymeet.com ([127.0.0.1]) by localhost (merrymeet.com [127.0.0.1]) (amavisd-new, port 10024) with ESMTP id Ct4JlXyTHqSF; Tue, 18 Mar 2014 11:15:14 -0700 (PDT)
Received: from keys.merrymeet.com (keys.merrymeet.com [173.164.244.97]) by mail.merrymeet.com (Postfix) with ESMTPSA id D5FB64F84D3E; Tue, 18 Mar 2014 11:15:12 -0700 (PDT)
Received: from [10.0.23.30] ([173.164.244.98]) by keys.merrymeet.com (PGP Universal service); Tue, 18 Mar 2014 11:15:14 -0700
X-PGP-Universal: processed; by keys.merrymeet.com on Tue, 18 Mar 2014 11:15:14 -0700
Mime-Version: 1.0 (Mac OS X Mail 7.2 \(1874\))
From: Jon Callas <jon@callas.org>
In-Reply-To: <CALR0uiJG6GcngWMUkg6NrP7_4uwf8+QDn6aMF-qonOpRMLdo3w@mail.gmail.com>
Date: Tue, 18 Mar 2014 11:15:11 -0700
Message-Id: <0AB5C684-6F51-48BD-9FF3-8810981F8C31@callas.org>
References: <CALR0uiJG6GcngWMUkg6NrP7_4uwf8+QDn6aMF-qonOpRMLdo3w@mail.gmail.com>
To: Alfredo Pironti <alfredo.pironti@inria.fr>
X-Mailer: Apple Mail (2.1874)
Content-Type: text/plain; charset=windows-1252
Content-Transfer-Encoding: quoted-printable
Archived-At: http://mailarchive.ietf.org/arch/msg/openpgp/Vf-Pl9XQ5V_pSPujfNyVWFaymsU
Cc: openpgp@ietf.org, Jon Callas <jon@callas.org>
Subject: Re: [openpgp] Disabling compression in OpenPGP
X-BeenThere: openpgp@ietf.org
X-Mailman-Version: 2.1.15
Precedence: list
List-Id: "Ongoing discussion of OpenPGP issues." <openpgp.ietf.org>
List-Unsubscribe: <https://www.ietf.org/mailman/options/openpgp>, <mailto:openpgp-request@ietf.org?subject=unsubscribe>
List-Archive: <http://www.ietf.org/mail-archive/web/openpgp/>
List-Post: <mailto:openpgp@ietf.org>
List-Help: <mailto:openpgp-request@ietf.org?subject=help>
List-Subscribe: <https://www.ietf.org/mailman/listinfo/openpgp>, <mailto:openpgp-request@ietf.org?subject=subscribe>
X-List-Received-Date: Tue, 18 Mar 2014 18:19:48 -0000

On Mar 18, 2014, at 9:00 AM, Alfredo Pironti <alfredo.pironti@inria.fr> wrote:

> Dear list,
> 
> It is well known that compressing data before encrypting them leaks much about the plaintext [1]. Recently, this has been exploited against the TLS protocol in the so-called CRIME attack [2].
> 
> Looking at RFC 4880, section 2.3, I read
> “OpenPGP implementations SHOULD compress the message after applying the signature but before encryption.”
> And indeed, gpg faithfully follows the spec by enabling compression by default.
> 
> I have done some preliminary work on password managers that rely on OpenPGP (gpg, in fact) to encrypt the passwords. Unsurprisingly, it turns out that compressing the password before encrypting it leaks much of the password entropy, making dictionary attacks significantly easier to mount. (In my preliminary experiments I used a password dictionary containing about 4 million passwords. If the attacker knows the original password length and its compressed length, then for some combinations of the two the candidate dictionary entries can reduce to as few as some hundreds.)
> 
> I believe similar attacks can be mounted in different contexts where OpenPGP is used. Hence, I propose to start discussion to amend RFC 4880 to at least discourage (if not forbid) the use of compression.
> 
> I welcome comments and suggestions.
> Alfredo Pironti

I think you're confusing a number of things, as well as cherry picking your data.

On the other side, there's the Katz-Schneier attack (which is more of a CFB attack than OpenPGP per se) that is essentially thwarted by compression. There are other attacks where compressing helps because it obscures plaintext. I'm eliding much, so go look at it.

OpenPGP is in general not an interactive or on-line protocol, and there are all sorts of attacks that are easy against interactive or on-line attacks and not against off-line. CRIME is one of them -- it exploits the interactive nature of TLS and makes repeated connections to learn something about the plaintext. It isn't the *compression* that does it, it's deltas between similar requests. You don't get to do that with OpenPGP. Note also that there are two easy mitigations against CRIME -- one is to disable compression and other other is to restrict the interactivity of the attack.

Similarly, most side-channel attacks aren't possible when all you have is ciphertext and you're attacking it yourself. TLS is an on-line protocol and there are many, many attacks against it that simply don't apply to blob encryption, which is what OpenPGP is.

There are a number of compression oracles that exist because of carelessness. Kelsey's paper that you quote is a fantastic paper because it flew in the face of the naive belief that compression was always good (on the grounds that it makes the plaintext less predictable). Compression is *often* good for that very reason. Compression is sometimes bad. Asserting that because compression is sometimes bad, it must always be bad is stretching the point, in my opinion.

If you believe that there are attacks that can be mounted OpenPGP because of, find one. You'll have a great paper.

Even better, come up with a generalized attack. Here's a framework that you can use:

OpenPGP is mostly bulk CFB encryption of a data blob. There are headers and stuff (which could make it easier or harder depending on data leaks), but at the core of the protocol, it's just CFB. OpenPGP compression is mostly ZIP or GZIP compression. Again, there are details with headers. In this case, the header issue might be significant, as OpenPGP often deletes some normal ZIP headers which could be used as known plaintext.

Nonetheless, here's where you should put your efforts:

* Take a file or other data blob. Call it P (for plaintext). 

* Compress P with ZIP or GZIP or even BZIP. Call that Z (for zip).

* Encrypt that with a simple CFB encryptor using some key K and some suitable cipher; I'd pick AES. Call that C (for ciphertext).

* Use comparisons between P, Z, and C to mount an attack on K -- giving a key-recovery attack. Or bypass the key and cipher and use knowledge about fragments of P and Z to undo pieces of C -- which would be some form of known plaintext attack. etc.

This gives a simplified model of what's going on inside OpenPGP, and might be easier to work with. Note, however, that OpenPGP typically throws away Z -- it's computed as an intermediate value and discarded. So it's conceivable that there's an attack that's possible but hard to mount.

Alternatively, you could use compression options in the gpg command line to get ciphertext of compressed and uncompressed outputs and do the same thing.

You describe an interesting case where there's a structured file of passwords. You could look at cases where the attacker would have knowledge of the plaintext or parts of it in varying ways. For example, it could be a sorted file, or it could contain all N-character passwords in a group (that might be itself sorted) and so on. 

You could also devise interactive and non-interactive versions of your attack. Note that the non-interactive ones are more interesting, of course. 

I hope this helps.

	Jon