Follow @Openwall on Twitter for new release announcements and other news
[<prev] [next>] [<thread-prev] [day] [month] [year] [list]
Date: Sat, 8 Jun 2013 14:09:58 -0400
From:  <jfoug@....net>
To: john-dev@...ts.openwall.com
Cc: Sayantan Datta <std2048@...il.com>
Subject: Re: limits within mscash2

It appears to be only 19 bytes (that is the salt_length in the .h file).

If you know where the changes are best placed, I will leave it with you.  if they can ber put into the 'prelim' steps, then by all means that sounds like the best location.

NOTE for all pbkdf2-HMAC code, we really should do that first iteration outside of the main 'fast' loop.  That way, we can do a slower hash of the salt. Once that hash is done, ALL remaining hashing can be done with fast 1 limb code (skipping the first half of the HMAC).  The only time we have potential for a longer crypt is that very first iteration, where the salt creates the end result of the first step.

---- Sayantan Datta <std2048@...il.com> wrote: 
> Hi,
> 
> On Sat, Jun 8, 2013 at 10:40 PM, <jfoug@....net> wrote:
> 
> > Changing the opencl_mscash2 format to allow for salts (user names) longer
> > than 19 bytes will take more work.  Ideally, this would be done 'prior' to
> > jumping into GPU code.  The salt is ONLY used on iteration #0.  This
> > iteration really should be pulled back into the crypt_all function, likely
> > done right after the DCC function, but before calling the pbkdf2. Then
> > instead of providing the salt to the pbkdf2 function, the 'results' of
> > iteration 0 would be provided.  Thus, in the gpu code, iterations 1 to
> > 10240 would be done (but not the first). This will have no impact on
> > overall speed,  AND will keep the GPU code faster, only doing 1 block
> > sha1's, BUT allow multiple block SHA1 code for that very first iteration
> > where the salt value is used.
> 
> 
> Yes it allows for 22 char salts :). Thank you very much. Probably I can
> also make it support 128 byte salts. Mostly some changes need to be made in
> pbkdf2_preprocess kernel. The pbkdf2_iter kernel does 1 to 10240
> iterations.
> 
> Regards,
> Sayantan

Powered by blists - more mailing lists

Confused about mailing lists and their use? Read about mailing lists on Wikipedia and check out these guidelines on proper formatting of your messages.