Re: [PATCH 3/3] sha1: use char type for temporary work buffer

From: Yann Droneaud <ydroneaud@opteya.com>
Date: Sep 13, 2012, 06:11 UTC
Message-ID: <1347516662.1961.23.camel@test.quest-ce.net>
In-Reply-To: <20120912210455.GA30679@sigill.intra.peff.net>

Le mercredi 12 septembre 2012 à 17:04 -0400, Jeff King a écrit :

Show 7 quoted lines

> > > Wouldn't this break all of the code that is planning to index "W" by
> > > 32-bit words (see the definitions of setW in block-sha1/sha1.c)?
> > > 
> > That's not the same "W" ... This part of the code is indeed unclear.
> 
> Sorry, you're right, that's a different work array (though it has the
> identical issue, no?).

No, this one is really accessed as int. But would probably benefit being declared as uint32_t.

> But the point still stands.  Did you audit the
> block-sha1 code to make sure nobody is ever indexing the W array?

Yes. It was the first thing to do before changing its definition (for alignment purpose especially).

> If you didn't, then your change is not safe. If you did, then you should really
> mention that in the commit message.
>

Sorry about this. I thought having the test suite OK was enough to prove this.

Show 8 quoted lines

> > > If that is indeed the problem, wouldn't the simplest fix be using
> > > uint32_t instead of "unsigned int"?
> > 
> > It's another way to fix this oddity, but not simpler.
> 
> It is simpler in the sense that it does not have any side effects (like
> changing how every user of the data structure needs to index it).
>

There's no other user than blk_SHA1_Update()

Show 12 quoted lines

> > > Moreover, would that be sufficient to run on such a platform? At the
> > > very least, "H" above would want the same treatment. And I would not be
> > > surprised if some of the actual code in block-sha1/sha1.c needed
> > > updating, as well.
> > 
> > ctx->H is actually used as an array of integer, so it would benefits of
> > being declared uint32_t for an ILP64 system. This fix would also be
> > required for blk_SHA1_Block() function.
> 
> So...if we are not ready to run on an ILP system after this change, then
> what is the purpose?
>

Readility: in blk_SHA1_Block(), the ctx->W array is used a 64 bytes len
array, so, AFAIK, there's no point of having it defined as a 16 int len.
It's disturbing while reading the code.

This could allows us to change the memcpy() call further:

@@ -246,7 +246,7 @@ void blk_SHA1_Update(blk_SHA_CTX *ctx, const void
*data, unsigned long len)
                unsigned int left = 64 - lenW;
                if (len < left)
                        left = len;
-               memcpy((char *)ctx->W + lenW, data, left);
+               memcpy(ctx->W + lenW, data, left);
                lenW = (lenW + left) & 63;
                if (lenW)

Regards.

-- 
Yann Droneaud
OPTEYA