linux-kernel - Re: Intel GCM: __driver-gcm-aes-aesni setkey missing

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [day] [month] [year] [list]

Date:	Sun, 18 Jan 2015 19:15:47 +0100
From:	Stephan Mueller <smueller@...onox.de>
To:	Tadeusz Struk <tadeusz.struk@...el.com>
Cc:	aidan.o.mahony@...el.com, gabriele.paoloni@...el.com,
	adrian.hoban@...el.com, linux-crypto@...r.kernel.org,
	herbert@...dor.apana.org.au, 'LKML' <linux-kernel@...r.kernel.org>
Subject: Re: Intel GCM: __driver-gcm-aes-aesni setkey missing

Am Samstag, 17. Januar 2015, 17:37:06 schrieb Tadeusz Struk:

Hi Tadeusz,

> Hi Stephan,
> 
> On 01/17/2015 10:23 AM, Stephan Mueller wrote:
> > during testing of my algif_aead patch with the different GCM
> > implementations I am able to trigger a kernel crash from user space using
> > __driver-gcm-aes- aesni.
> > 
> > As I hope that algif_aead is going to be included, unprivileged userspace
> > would then reliably crash the kernel -- with the current kernel code,
> > userspace has no interface to trigger the issue.
> 
> Yes, that's a problem.
> 
> > As I am not sure what the purpose of __driver-gcm-aes-aesni is (only a
> > backend for RFC4106 GCM or a regular cipher), I did not yet create a
> > patch. IMHO there are two solutions:
> > 
> > - either create a valid setkey callback so that a key is set
> > 
> > - or create a noop setkey that returns -EOPNOTSUPP which effectively
> > disables that cipher for regular consumption.
> 
> __driver-gcm-aes-aesni is only a helper for rfc4106-gcm-aesni and it
> never supposed to be used on it's own. I think implementing a setkey
> function that only returns an error would be a good solution for this.

Ok, I will send a patch shortly.

> Another question is what if someone will ignore the error or skip the
> setsockopt(ALG_SET_KEY) altogether and still call the sendmsg() and
> read() to trigger encrypt()?

Using my libkcapi [1] test bench, I disabled key and IV submission for 
symmetric ciphers (tested cbc(aes) which invokes your AESNI code path on my 
box -- and gcm(aes) and ccm(aes) which again both use the AESNI core and the C 
implementation of GCM and CCM).

All tests with missing keys and IVs:

- showed a successful encryption / decryption with the CBC mode

- returned the error code of either ENOKEY or EINVAL for GCM / CCM 
encryption/decryption

There is no crash/BUG/WARN observed.

> 
> > Note, if it is only a backend for the RFC4106 implementation, may I ask
> > why
> > __driver-gcm-aes-aesni is implemented as a separate cipher that is
> > registered with the kernel crypto API?
> 
> This is because we need to have one instance of the helper tfm with its
> context per each of the rfc4106-gcm-aesni tfm instance and that was one
> convenient way to do this.

Then I concur with you that having a setkey function returning an error is the 
right way.
> 
> Tadeusz
> --
> To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
> the body of a message to majordomo@...r.kernel.org
> More majordomo info at  http://vger.kernel.org/majordomo-info.html
> Please read the FAQ at  http://www.tux.org/lkml/


-- 
Ciao
Stephan
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@...r.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/