Re: [kernel-hardening] Re: [PATCH v3 04/13] crypto/rng: ensure that the RNG is ready before using

[Eric Biggers]

Hi Jason,

On Tue, Jun 06, 2017 at 05:23:04PM +0200, Jason A. Donenfeld wrote:
> Hey again Eric,
> 
> One thing led to another and I wound up just rewriting all the crypto
> in big_keys.c. I'll include this for v4:
> 
> https://git.zx2c4.com/linux-dev/commit/?h=jd/rng-blocker&id=886ff283b9808aecb14aa8e397da8496a9635aed
> 
> Not only was the use of crypto/rng inappropriate, but the decision to
> go with aes-ecb is shocking. Seeing that this author had no other
> commits in the tree, and that all subsequent commits that mentioned
> his name were cleaning up his mess, I just went ahead and removed both
> the crypto/rng misusage and changed from aes-ecb to aes-gcm.
> 
> Anyway, I'll wait for some more reviews on v3, and then this can be
> reviewed for v4.
> 
> Regards,
> Jason

I agree that the use of ECB mode in big_key is broken, and thanks for trying to
fix it!  I think using GCM is good, but please leave a very conspicuous comment
where the nonce is being set to 0, noting that it's safe only because a unique
key is used to encrypt every big_key *and* the big_keys are not updatable (via
an .update method in the key_type), resulting in each GCM key being used for
only a single encryption.

Also, I think you should send this to the keyrings mailing list and maintainer
so it can be discussed and merged separately from your RNG changes.

Eric