Re: Ciphers and modes of operation

[Gregory BELLIER]

Dr. Stephen Henson a écrit :
On Wed, Feb 24, 2010, Gregory BELLIER wrote:
  
Hello Weidong and Stephen,

So please, if you have time, take a look at this code sample, maybe you'll 
notice something in just a glance.

#define BLOCK_CIPHER_def_dyna(cname, nmode, mode, MODE, kstruct, nid, 
block_size, \
             key_len, iv_len, flags, init_key, cleanup, \
             set_asn1, get_asn1, ctrl) \
static EVP_CIPHER cname##_##mode = { \
   nid##_##nmode, block_size, key_len, iv_len, \
   flags | EVP_CIPH_##MODE##_MODE, \
   init_key, \
   cname##_##mode##_cipher, \
   cleanup, \
   sizeof(kstruct), \
   set_asn1, get_asn1,\
   ctrl, \
   NULL \
}; \

#define IMPLEMENT_BLOCK_CIPHER_CBC(cname, ksched, cprefix, kstruct, nid, \
                    block_size, key_len, iv_len, cbits, \
                    flags, init_key, \
                   cleanup, set_asn1, get_asn1, ctrl) \
     BLOCK_CIPHER_func_cbc(cname, cprefix, kstruct, ksched) \
     BLOCK_CIPHER_def_dyna(cname, cbc, cbc, CBC, kstruct, nid, 0, 0, 0, \
               flags, init_key, cleanup, set_asn1, get_asn1, ctrl)

IMPLEMENT_BLOCK_CIPHER_CBC(myc, ks, MYC, EVP_MYC_KEY,
                       NID_myc, 16, MYC_KEY_LENGTH,16, 128,
                       EVP_CIPH_VARIABLE_LENGTH, myc_init_key, NULL,
                       EVP_CIPHER_set_asn1_iv, EVP_CIPHER_get_asn1_iv, 
NULL)

const EVP_CIPHER *EVP_myc_cbc(void)
{
   myc_cbc.block_size = 16;
   myc_cbc.iv_len         = 16;
   myc_cbc.key_len      = 32;
    return &myc_cbc;
}

    

Just a comment about that and a point of reference for anyone who might also
want to implement a cipher. All this IMPLEMENT/DECLARE business is done
internally so we can defined many cipher structures which are almost identical
and any changes to them need only a macro change and not having to manually
update over a hundred definitions as well.

If you define your own cipher you don't and arguably shouldn't do that because
it can be difficult to see what is going on. You can just define the
EVP_CIPHER structure manually. There is an example of this "manual cipher"
definition in crypto/evp/e_xcbc_d.c 

One thing to note is that OpenSSL expects a cipher to function correctly if
the input and output buffers are the same (i.e. to cipher "in place").

I'd suggest testing with something like the 'enc' utility at first to see if
you can encrypt/decrypt data sensibly. If you have any test vectors then try
those too.
In fact, I don't need to redefine it but I can't find a way to do 
otherwise. If I don't do so, I get an error about the modes of operation 
I didn't declare.

The testenc runs fine :
sh ./testenc
[...]
myc
myc base64
myc-cbc
myc-cbc base64


Therefore, because I don't implement/declare all the modes of operation, 
I guess I have to define the EVP_CIPHER structure manually.
But, I know I don't do it well but it works. At least, from the 
beginning of the make tests to the error.
I'll try however what you suggest.

Do you think what I've done could be the cause of the bug ?