Hi Mat,

 >>
>> Should we go in the direction of doing something like:
>>
>> acipher = l_asymmetric_cipher_new(type);
>>
>> l_asymmetric_cipher_set_pubkey(acipher, key, key_size);
>> l_asymmetric_cipher_set_privkey(acipher, key, key_size);
>>
>> l_asymmetric_cipher_get_key_size(acipher);
>
> Last week I pushed back on adding a public/private flag to
> l_asymmetric_cipher_new, but as I keep working on this the flag is
> becoming the most appealing. If there are separate key setting
> functions, those key set function end up being thin wrappers around a
> common internal function anyway, and makes the asym cipher api more
> complicated to use and implement.
>
> Here's what I'd like to do:
>
>   * Add public/private flag to l_asymmetric_cipher_new
>   * Move TLS to keyctl-based crypto
>   * Remove l_asymmetric_cipher_get_key_size (TLS is the only user) and
> the ASN.1 parsing code.
>

Do we still need to know the key size in order to provide a proper 
buffer for certain implementations?

>>> -    cipher->cipher.decrypt_sk = create_alg("akcipher", alg_name,
>>> -                        key, key_length);
>>> -    if (cipher->cipher.decrypt_sk < 0)
>>> -        goto error_close;
>>> +    if (!cipher->public_key) {
>>> +        cipher->cipher.decrypt_sk = create_alg("akcipher", alg_name,
>>> +                            key, key_length,
>>> +                            cipher->public_key);
>>> +        if (cipher->cipher.decrypt_sk < 0)
>>> +            goto error_close;
>>> +    }

By my reading decrypt_sk is only initialized if cipher->public_key is 
false...

>> This should probably check for validity of decrypt_sk
>
> There should never be a struct l_asymmetric_cipher with a bad decrypt_sk
> - validity is checked in l_asymmetric_cipher_new and it never returns a
> struct with a bad socket handle. If the socket gets corrupted somehow,
> operate_cipher will let us know. Is this only important if the set key
> functions are split out?
>

Why bother operating on invalid sockets.  If we were provided a public 
key, then we can't decrypt, right?

> Do you know why we have separate encrypt_sk and decrypt_sk sockets? I
> tried using the same socket for all operations and it works fine.
>

This is another weirdness in the userspace API.  I think it is an issue 
in certain skcipher implementations.  They don't keep a separate encrypt 
/ decrypt state.  So if you don't keep two separate sockets, you get 
screwy results.  I think ARC4 is our biggest culprit.  Look in the 
history of cipher.c for details.  git show 
632dfc8fc9d53718e5596cad25e0a71c373cf964

Regards,
-Denis