Hello,
The AES-CBC implementation in the s5p-sss crypto driver is failing the improved
crypto self-tests I currently have out for review. The improved tests check
that all CBC implementations update the IV buffer to be the last ciphertext
block. This has always been required so that requests can be chained, but
unfortunately it wasn't tested for by the self-tests until now.
See the boot logs for Exynos platforms from the KernelCI job here:
https://kernelci.org/boot/all/job/ardb/branch/for-kernelci/kernel/v5.0-rc1-149-g64c945c018af/
alg: skcipher: cbc-aes-s5p encryption test failed (wrong output IV) on test vector 0, cfg=\"in-place\"
00000000: 3d af ba 42 9d 9e b4 30 b4 22 da 80 2c 9f ac 41
You can reproduce by pulling from
https://git.kernel.org/pub/scm/linux/kernel/git/ebiggers/linux.git
branch "iv-out-testing", unsetting CONFIG_CRYPTO_MANAGER_DISABLE_TESTS,
setting CONFIG_CRYPTO_MANAGER_EXTRA_TESTS=y, rebooting and checking dmesg.
(CRYPTO_MANAGER_EXTRA_TESTS is probably unneeded for this, but you might as well
use it. Also you can optionally revert the last patch, which makes crypto
self-test failures cause a kernel panic for testing purposes. It's possible
that ctr-aes-s5p is failing too but it was just never gotten to...)
The patch series is also available on the linux-crypto mailing list:
https://patchwork.kernel.org/cover/10811951/
Note that I don't have this hardware myself, so if it turns out that no one is
interested in fixing this anytime soon I'll instead have to propose disabling
these algorithm(s) until they can be fixed.
Thanks,
- Eric
On Fri, 15 Feb 2019 at 19:51, Eric Biggers <[email protected]> wrote:
>
> Hello,
>
> The AES-CBC implementation in the s5p-sss crypto driver is failing the improved
> crypto self-tests I currently have out for review. The improved tests check
> that all CBC implementations update the IV buffer to be the last ciphertext
> block. This has always been required so that requests can be chained, but
> unfortunately it wasn't tested for by the self-tests until now.
>
> See the boot logs for Exynos platforms from the KernelCI job here:
> https://kernelci.org/boot/all/job/ardb/branch/for-kernelci/kernel/v5.0-rc1-149-g64c945c018af/
>
> alg: skcipher: cbc-aes-s5p encryption test failed (wrong output IV) on test vector 0, cfg=\"in-place\"
> 00000000: 3d af ba 42 9d 9e b4 30 b4 22 da 80 2c 9f ac 41
>
> You can reproduce by pulling from
> https://git.kernel.org/pub/scm/linux/kernel/git/ebiggers/linux.git
> branch "iv-out-testing", unsetting CONFIG_CRYPTO_MANAGER_DISABLE_TESTS,
> setting CONFIG_CRYPTO_MANAGER_EXTRA_TESTS=y, rebooting and checking dmesg.
>
> (CRYPTO_MANAGER_EXTRA_TESTS is probably unneeded for this, but you might as well
> use it. Also you can optionally revert the last patch, which makes crypto
> self-test failures cause a kernel panic for testing purposes. It's possible
> that ctr-aes-s5p is failing too but it was just never gotten to...)
>
> The patch series is also available on the linux-crypto mailing list:
> https://patchwork.kernel.org/cover/10811951/
>
> Note that I don't have this hardware myself, so if it turns out that no one is
> interested in fixing this anytime soon I'll instead have to propose disabling
> these algorithm(s) until they can be fixed.
Thanks for the report. I'll take a look. I have the HW so I think I
will be able to reproduce it easily.
Best regards,
Krzysztof
Hi,
On 15.02.2019 19:51, Eric Biggers wrote:
> Hello,
>
> The AES-CBC implementation in the s5p-sss crypto driver is failing the improved
> crypto self-tests I currently have out for review. The improved tests check
> that all CBC implementations update the IV buffer to be the last ciphertext
> block. This has always been required so that requests can be chained, but
> unfortunately it wasn't tested for by the self-tests until now.
>
> See the boot logs for Exynos platforms from the KernelCI job here:
> https://kernelci.org/boot/all/job/ardb/branch/for-kernelci/kernel/v5.0-rc1-149-g64c945c018af/
>
> alg: skcipher: cbc-aes-s5p encryption test failed (wrong output IV) on test vector 0, cfg=\"in-place\"
> 00000000: 3d af ba 42 9d 9e b4 30 b4 22 da 80 2c 9f ac 41
> [...]
I checked dmesg logs after my patch with:
root@target:~# dmesg |grep alg: |grep aes
[ 4.087168] crypto: alg: self-tests for aes-generic (aes) passed
[ 6.305202] crypto: alg: self-tests for ecb-aes-s5p (ecb(aes)) passed
[ 6.405489] crypto: alg: self-tests for cbc-aes-s5p (cbc(aes)) passed
[ 6.505000] crypto: alg: self-tests for ctr-aes-s5p (ctr(aes)) passed
and I noticed there are no cbc-aes-generic nor ctr-aes-generic
Did you test them in separate tree ? Why they are not registered in /proc/crypto ?
--
Best regards,
Kamil Konieczny
Samsung R&D Institute Poland
On Tue, 19 Feb 2019 at 10:58, Kamil Konieczny
<[email protected]> wrote:
>
> Hi,
>
> On 15.02.2019 19:51, Eric Biggers wrote:
> > Hello,
> >
> > The AES-CBC implementation in the s5p-sss crypto driver is failing the improved
> > crypto self-tests I currently have out for review. The improved tests check
> > that all CBC implementations update the IV buffer to be the last ciphertext
> > block. This has always been required so that requests can be chained, but
> > unfortunately it wasn't tested for by the self-tests until now.
> >
> > See the boot logs for Exynos platforms from the KernelCI job here:
> > https://kernelci.org/boot/all/job/ardb/branch/for-kernelci/kernel/v5.0-rc1-149-g64c945c018af/
> >
> > alg: skcipher: cbc-aes-s5p encryption test failed (wrong output IV) on test vector 0, cfg=\"in-place\"
> > 00000000: 3d af ba 42 9d 9e b4 30 b4 22 da 80 2c 9f ac 41
> > [...]
>
> I checked dmesg logs after my patch with:
>
> root@target:~# dmesg |grep alg: |grep aes
> [ 4.087168] crypto: alg: self-tests for aes-generic (aes) passed
> [ 6.305202] crypto: alg: self-tests for ecb-aes-s5p (ecb(aes)) passed
> [ 6.405489] crypto: alg: self-tests for cbc-aes-s5p (cbc(aes)) passed
> [ 6.505000] crypto: alg: self-tests for ctr-aes-s5p (ctr(aes)) passed
>
> and I noticed there are no cbc-aes-generic nor ctr-aes-generic
>
> Did you test them in separate tree ? Why they are not registered in /proc/crypto ?
>
Generic CBC and CTR are template based, so they are only instantiated on demand.