From: "J. Bruce Fields" <bfields@fieldses.org>
Subject: Re: [PATCH] crypto: scatterwalk_copychunks() fails to advance through scatterlist
Date: Tue, 20 Mar 2007 10:16:14 -0400
Message-ID: <20070320141614.GA12165@fieldses.org>
References: <20070319235508.GH29272@fieldses.org> <20070320051656.GA17711@gondor.apana.org.au>
Mime-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Cc: linux-crypto@vger.kernel.org, nfsv4@linux-nfs.org
To: Herbert Xu <herbert@gondor.apana.org.au>
Content-Disposition: inline
In-Reply-To: <20070320051656.GA17711@gondor.apana.org.au>
Sender: linux-crypto-owner@vger.kernel.org

On Tue, Mar 20, 2007 at 04:16:56PM +1100, Herbert Xu wrote:
> Thanks for the patch.  However I still have a question as to why
> this is happening.
> 
> As far as I can see scatterwalk_copychunks is only called in two
> places.  In both spots it only processes one block of data.  Since
> we set the maximum block size to PAGE_SIZE/8 I don't see how you
> can get an offset of zero and still roll over to the next page
> in scatterwalk_copychunks.

Are the elements of the scatterlists assumed to always be full pages?  I
need to encrypt things that look like, for example:

	sg[0].page = page1
	sg[0].offset = 0
	sg[0].length = 5
	sg[1].page = page2
	sg[1].offset = 0
	sg[1].length = 37

and worse....  I could do this by hand if I had to, but the crypto code,
if it's not designed to handle this sort of thing, seems very close, so
I'd rather enhance it than duplicate a lot of this complicated
scatterlist-traversal stuff.

--b.