Re: Doing crypto in small stack buffers (bluetooth vs vmalloc-stack crash, etc)

On Wed, Jun 22, 2016 at 2:48 PM, Andy Lutomirski [off-list ref] wrote:

On Tue, Jun 21, 2016 at 5:52 PM, Andy Lutomirski [off-list ref] wrote:

quoted

On Tue, Jun 21, 2016 at 5:42 PM, Herbert Xu [off-list ref] wrote:

quoted

On Tue, Jun 21, 2016 at 10:43:40AM -0700, Andy Lutomirski wrote:

quoted

Is there a straightforward way that bluetooth and, potentially, other
drivers can just do synchronous crypto in a small buffer specified by
its virtual address?  The actual cryptography part of the crypto code
already works this way, but I can't find an API for it.

Yes, single block users should use crypto_cipher_encrypt_one, an
example would be drivers/md/dm-crypt.c.

Aha!  I expected something like that to exist, but I couldn't find it.
I'll change the two offenders I've found so far to use it.

Before I do this, can you explain what the difference is between
crypto_cipher and crypto_skcipher?  net/bluetooth/smp.c currently uses
crypto_alloc_skcipher, which you added in:

commit 71af2f6bb22a4bf42663e10f1d8913d4967ed07f
Author: Herbert Xu [off-list ref]
Date:   Sun Jan 24 21:18:30 2016 +0800

    Bluetooth: Use skcipher and hash

Am I just supposed to replace "skcipher" with "cipher" everywhere?

It looks like I'm supposed to that and to use "aes" instead of "ebc(aes)".

*However*, the other offender I've found (net/rxrpc/rxkad.c) uses
"pcbc(fcrypt)", which doesn't appear to be usable with this API.  Is
there no way to say "I want synchronous crypto on this VA range" using
the skcipher API?

--Andy


-- 
Andy Lutomirski
AMA Capital Management, LLC