Re: [PATCH] loop: fixing cryptoloop troubles.

[Fruhwirth Clemens]

On Sun, Aug 10, 2003 at 04:27:47PM +0200, Christophe Saout wrote:
> Am So, 2003-08-10 um 16.10 schrieb Pascal Brisset:
> 
> > > In loop_transfer_bio the initial vector has been computed only once. For any
> > > situation where more than one bio_vec is present the initial vector will be
> > > wrong. Here is the trivial but important fix. 
> > 
> > Looks good, but:
> > - I doubt this could explain the alteration pattern (1 byte every 512).
> > - Corruption also occured with cipher_null (which ignores the IV).

I could not find a way to explain that strange pattern either. With CBC it
would have to result in total mess if just one bit is flipped. Probably
read/writing is handled with different sized bio_vec.. no idea.

cipher_null does not ignore the IV. The CBC processing takes place no matter
what mapping function (aka electronic codebook) is used. The fact that
cipher_null is an identity mapping does not stop CBC. 

> I personally think that the only way to get things right is to do
> encryption sector by sector (not bvec by bvec) since every sector can
> have its own iv.

That's done anyway. Per convention the transformation module is allowed to
increase the IV every 512 bytes. The IV parameter is only the initial
initial vector ;). 

> I've implemented a crypto target for device-mapper that does this and it
> doesn't seem to suffer from these corruption problems:
> http://marc.theaimsgroup.com/?l=linux-kernel&m=105967481007242&w=2 and a
> slightly updated patch: http://www.saout.de/misc/dm-crypt.diff

Nice! It's definitely a feature worth merging. loop.c used to be the place
where to put this stuff, but why not replace it by newer in-kernel
techniques?

> Should I repost the patch (inline this time) with an additional [PATCH]
> or am I being annoying? Joe Thornber (the dm maintainer) would like to
> see this patch merged.

If you can't get attention for your patch, try to convince someone "more
important". DM maintainer is a good place to start :)

Regards, Clemens

Attachment: pgp00003.pgp
Description: PGP signature