Re: [PATCH v4 3/6] cipher: Return result length from asymmetric cipher

[Denis Kenzior <denkenz@gmail.com>]

[-- Attachment #1: Type: text/plain, Size: 2178 bytes --]

Hi Andrew,

On 06/16/2016 08:30 PM, Andrzej Zaborowski wrote:
> Hi,
>
> On 17 June 2016 at 02:17, Mat Martineau
> <mathew.j.martineau@linux.intel.com> wrote:
>> On Thu, 16 Jun 2016, Denis Kenzior wrote:
>>> On 06/15/2016 12:18 PM, Mat Martineau wrote:
>>>> @@ -1790,9 +1813,12 @@ static void tls_handle_rsa_client_key_xchg(struct
>>>> l_tls *tls,
>>>>                  return;
>>>>          }
>>>>
>>>> -       result = l_asymmetric_cipher_decrypt(rsa_server_privkey, buf + 2,
>>>> -                                               pre_master_secret,
>>>> -                                               key_size, 48);
>>>> +       pre_master_secret = l_malloc(key_size);
>>>> +
>>>> +       bytes_decrypted = l_asymmetric_cipher_decrypt(rsa_server_privkey,
>>>> +                                                       buf + 2,
>>>> +
>>>> pre_master_secret,
>>>> +                                                       key_size,
>>>> key_size);
>>>
>>>
>>> So what's the verdict, I thought we decided to accept short reads?
>>
>>
>> Short reads work with software RSA in the kernel, but we're still sorting
>> out expectations for crypto hardware drivers. For now, long reads are still
>> the way to ensure compatibility.
>
> I guess it depends on what kernel you're aiming for but your test tree
> has only the software pkcs1pad driver.  Upstream can in theory add a
> new driver at any time but AF_ALG is still under review there.
>
> By the way I wouldn't call these short reads as we're reading all the
> data that the socket has to read ;)  In the commit message I think Mat
> may be referring to the other short reads.

I think this is what Mat means.  That the AF_ALG socket is expecting the 
buffer to recvmsg to be key_size bytes.  E.g. a 'short-read'.

>
> I don't think these kernel api changes should spill into tls.c though.
>

Anyway, the bottom line is for pkcs1pad we should be able to pass a 
buffer < key_size for recvmsg.  Nothing else makes sense.  For sendmsg, 
the output buffer should always be key_size.

For rsa primitive operation, the recvmsg buffer 'should' be key_size bytes.

Regards,
-Denis