From: Andy Lutomirski Subject: Re: BPF hash algo (Re: [kernel-hardening] Re: [PATCH v7 3/6] random: use SipHash in place of MD5) Date: Thu, 22 Dec 2016 08:53:54 -0800 Message-ID: References: Mime-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Cc: Hannes Frederic Sowa , Daniel Borkmann , Alexei Starovoitov , "kernel-hardening@lists.openwall.com" , "Theodore Ts'o" , Netdev , LKML , Linux Crypto Mailing List , David Laight , Eric Dumazet , Linus Torvalds , Eric Biggers , Tom Herbert , Andi Kleen , "David S. Miller" , Jean-Philippe Aumasson To: "Jason A. Donenfeld" Return-path: Received: from mail-ua0-f174.google.com ([209.85.217.174]:35135 "EHLO mail-ua0-f174.google.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S941458AbcLVQyQ (ORCPT ); Thu, 22 Dec 2016 11:54:16 -0500 Received: by mail-ua0-f174.google.com with SMTP id 2so112579746uax.2 for ; Thu, 22 Dec 2016 08:54:15 -0800 (PST) In-Reply-To: Sender: linux-crypto-owner@vger.kernel.org List-ID: On Thu, Dec 22, 2016 at 8:28 AM, Jason A. Donenfeld wrote: > Hi all, > > I don't know what your design requirements are for this. It looks like > you're generating some kind of crypto digest of a program, and you > need to avoid collisions. If you'd like to go with a PRF (keyed hash > function) that uses some kernel secret key, then I'd strongly suggest > using Keyed-Blake2. Alternatively, if you need for userspace to be > able to calculate the same hash, and don't want to use some kernel > secret, then I'd still suggest using Blake2, which will be fast and > secure. > > If you can wait until January, I'll work on a commit adding the > primitive to the tree. I've already written it and I just need to get > things cleaned up. > >> Blake2 is both less stable (didn't they slightly change it recently?) > > No, Blake2 is very stable. It's also extremely secure and has been > extensively studied. Not to mention it's faster than SHA2. And if you > want to use it as a PRF, it's obvious better suited and faster to use > Blake2's keyed PRF mode than HMAC-SHA2. > > If you don't care about performance, and you don't want to use a PRF, > then just use SHA2-256. If you're particularly concerned about certain > types of attacks, you could go with SHA2-512 truncated to 256 bytes, > but somehow I doubt you need this. I don't think this cares about performance. (Well, it cares about performance, but the verifier will likely dominiate the cost by such a large margin that the hash algo doesn't matter.) And staying FIPS-compliant-ish is worth a little bit, so I'd advocate for something in the SHA2 family. > If userspace hasn't landed, can we get away with changing this code > after 4.10? Or should we just fix it before 4.10? Or should we revert > it before 4.10? Development-policy-things like this I have zero clue > about, so I heed to your guidance. I think it should be fixed or reverted before 4.10. --Andy