public inbox for linux-kernel@vger.kernel.org
 help / color / mirror / Atom feed
* 2.6.13rc3: crypto horribly broken on all 64bit archs
@ 2005-07-17 12:20 Andreas Steinmetz
  2005-07-17 20:45 ` Herbert Xu
  2005-07-20  4:41 ` Andy Isaacson
  0 siblings, 2 replies; 4+ messages in thread
From: Andreas Steinmetz @ 2005-07-17 12:20 UTC (permalink / raw)
  To: Linux Kernel Mailinglist, jmorris, herbert

[-- Attachment #1: Type: text/plain, Size: 669 bytes --]

from include/linux/kernel.h:

#define ALIGN(x,a) (((x)+(a)-1)&~((a)-1))

from crypto/cipher.c:

unsigned int alignmask = ...
u8 *src = (u8 *)ALIGN((unsigned long)buffer, alignmask + 1);
...
unsigned int alignmask = ...
u8 *tmp = (u8 *)ALIGN((unsigned long)buffer, alignmask + 1);
...
unsigned int align;
addr = ALIGN(addr, align);
addr += ALIGN(tfm->__crt_alg->cra_ctxsize, align);

The compiler first does ~((a)-1)) and then expands the unsigned int to
unsigned long for the & operation. So we end up with only the lower 32
bits of the address. Who did smoke what to do this? Patch attached.
-- 
Andreas Steinmetz                       SPAMmers use robotrap@domdv.de


[-- Attachment #2: crypto.patch --]
[-- Type: text/plain, Size: 980 bytes --]

--- linux.orig/crypto/cipher.c	2005-07-17 13:35:15.000000000 +0200
+++ linux/crypto/cipher.c	2005-07-17 14:04:00.000000000 +0200
@@ -41,7 +41,7 @@
 			       struct scatter_walk *in,
 			       struct scatter_walk *out, unsigned int bsize)
 {
-	unsigned int alignmask = crypto_tfm_alg_alignmask(desc->tfm);
+	unsigned long alignmask = crypto_tfm_alg_alignmask(desc->tfm);
 	u8 buffer[bsize * 2 + alignmask];
 	u8 *src = (u8 *)ALIGN((unsigned long)buffer, alignmask + 1);
 	u8 *dst = src + bsize;
@@ -160,7 +160,7 @@
 			      unsigned int nbytes)
 {
 	struct crypto_tfm *tfm = desc->tfm;
-	unsigned int alignmask = crypto_tfm_alg_alignmask(tfm);
+	unsigned long alignmask = crypto_tfm_alg_alignmask(tfm);
 	u8 *iv = desc->info;
 
 	if (unlikely(((unsigned long)iv & alignmask))) {
@@ -424,7 +424,7 @@
 	}
 	
 	if (ops->cit_mode == CRYPTO_TFM_MODE_CBC) {
-		unsigned int align;
+		unsigned long align;
 		unsigned long addr;
 	    	
 	    	switch (crypto_tfm_alg_blocksize(tfm)) {

^ permalink raw reply	[flat|nested] 4+ messages in thread

* Re: 2.6.13rc3: crypto horribly broken on all 64bit archs
  2005-07-17 12:20 2.6.13rc3: crypto horribly broken on all 64bit archs Andreas Steinmetz
@ 2005-07-17 20:45 ` Herbert Xu
  2005-07-18 20:24   ` David S. Miller
  2005-07-20  4:41 ` Andy Isaacson
  1 sibling, 1 reply; 4+ messages in thread
From: Herbert Xu @ 2005-07-17 20:45 UTC (permalink / raw)
  To: Andreas Steinmetz; +Cc: Linux Kernel Mailinglist, jmorris

On Sun, Jul 17, 2005 at 02:20:21PM +0200, Andreas Steinmetz wrote:
> 
> The compiler first does ~((a)-1)) and then expands the unsigned int to
> unsigned long for the & operation. So we end up with only the lower 32
> bits of the address. Who did smoke what to do this? Patch attached.

Thanks for the patch Andreas.  A similar patch will be in the
mainline kernel as soon as everybody is back home from Canada.
-- 
Visit Openswan at http://www.openswan.org/
Email: Herbert Xu ~{PmV>HI~} <herbert@gondor.apana.org.au>
Home Page: http://gondor.apana.org.au/~herbert/
PGP Key: http://gondor.apana.org.au/~herbert/pubkey.txt

^ permalink raw reply	[flat|nested] 4+ messages in thread

* Re: 2.6.13rc3: crypto horribly broken on all 64bit archs
  2005-07-17 20:45 ` Herbert Xu
@ 2005-07-18 20:24   ` David S. Miller
  0 siblings, 0 replies; 4+ messages in thread
From: David S. Miller @ 2005-07-18 20:24 UTC (permalink / raw)
  To: herbert; +Cc: ast, linux-kernel, jmorris

From: Herbert Xu <herbert@gondor.apana.org.au>
Date: Mon, 18 Jul 2005 06:45:54 +1000

> On Sun, Jul 17, 2005 at 02:20:21PM +0200, Andreas Steinmetz wrote:
> > 
> > The compiler first does ~((a)-1)) and then expands the unsigned int to
> > unsigned long for the & operation. So we end up with only the lower 32
> > bits of the address. Who did smoke what to do this? Patch attached.
> 
> Thanks for the patch Andreas.  A similar patch will be in the
> mainline kernel as soon as everybody is back home from Canada.

Yes, the fix is in my tree and will make it's way upstream
as soon as possible.

^ permalink raw reply	[flat|nested] 4+ messages in thread

* Re: 2.6.13rc3: crypto horribly broken on all 64bit archs
  2005-07-17 12:20 2.6.13rc3: crypto horribly broken on all 64bit archs Andreas Steinmetz
  2005-07-17 20:45 ` Herbert Xu
@ 2005-07-20  4:41 ` Andy Isaacson
  1 sibling, 0 replies; 4+ messages in thread
From: Andy Isaacson @ 2005-07-20  4:41 UTC (permalink / raw)
  To: Andreas Steinmetz; +Cc: Linux Kernel Mailinglist, jmorris, herbert

On Sun, Jul 17, 2005 at 02:20:21PM +0200, Andreas Steinmetz wrote:
> from include/linux/kernel.h:
> 
> #define ALIGN(x,a) (((x)+(a)-1)&~((a)-1))
> 
> from crypto/cipher.c:
> 
> unsigned int alignmask = ...
> u8 *src = (u8 *)ALIGN((unsigned long)buffer, alignmask + 1);

The type foolery you suggested is un-obvious, especially at review time.
How about the following instead?

-andy

# HG changeset patch
# User adi@hexapodia.org
# Node ID ff4d8285dcb581b8340b133c22780c4fcc0dabb3
# Parent  2b0b2208676a22741fc3b1681ca9dd555ca40dfd
Add new PTR_ALIGN macro to encapsulate necessary casting, and use it in
the places where ALIGN was used on pointers.

Signed-Off-By: Andy Isaacson <adi@hexapodia.org>

diff -r 2b0b2208676a -r ff4d8285dcb5 crypto/cipher.c
--- a/crypto/cipher.c	Sun Jul 17 03:06:51 2005
+++ b/crypto/cipher.c	Tue Jul 19 18:51:11 2005
@@ -43,7 +43,7 @@
 {
 	unsigned int alignmask = crypto_tfm_alg_alignmask(desc->tfm);
 	u8 buffer[bsize * 2 + alignmask];
-	u8 *src = (u8 *)ALIGN((unsigned long)buffer, alignmask + 1);
+	u8 *src = PTR_ALIGN(buffer, alignmask + 1);
 	u8 *dst = src + bsize;
 	unsigned int n;
 
@@ -166,7 +166,7 @@
 	if (unlikely(((unsigned long)iv & alignmask))) {
 		unsigned int ivsize = tfm->crt_cipher.cit_ivsize;
 		u8 buffer[ivsize + alignmask];
-		u8 *tmp = (u8 *)ALIGN((unsigned long)buffer, alignmask + 1);
+		u8 *tmp = PTR_ALIGN(buffer, alignmask + 1);
 		int err;
 
 		desc->info = memcpy(tmp, iv, ivsize);
diff -r 2b0b2208676a -r ff4d8285dcb5 drivers/crypto/padlock-aes.c
--- a/drivers/crypto/padlock-aes.c	Sun Jul 17 03:06:51 2005
+++ b/drivers/crypto/padlock-aes.c	Tue Jul 19 18:51:11 2005
@@ -287,7 +287,7 @@
 
 static inline struct aes_ctx *aes_ctx(void *ctx)
 {
-	return (struct aes_ctx *)ALIGN((unsigned long)ctx, PADLOCK_ALIGNMENT);
+	return PTR_ALIGN(ctx, PADLOCK_ALIGNMENT);
 }
 
 static int
diff -r 2b0b2208676a -r ff4d8285dcb5 include/linux/kernel.h
--- a/include/linux/kernel.h	Sun Jul 17 03:06:51 2005
+++ b/include/linux/kernel.h	Tue Jul 19 18:51:11 2005
@@ -29,6 +29,8 @@
 
 #define ARRAY_SIZE(x) (sizeof(x) / sizeof((x)[0]))
 #define ALIGN(x,a) (((x)+(a)-1)&~((a)-1))
+#define PTR_ALIGN(p,a) \
+	((void *)(((unsigned long)(p)+(a)-1)&~(((unsigned long)(a)-1))))
 
 #define	KERN_EMERG	"<0>"	/* system is unusable			*/
 #define	KERN_ALERT	"<1>"	/* action must be taken immediately	*/

^ permalink raw reply	[flat|nested] 4+ messages in thread

end of thread, other threads:[~2005-07-20  4:41 UTC | newest]

Thread overview: 4+ messages (download: mbox.gz follow: Atom feed
-- links below jump to the message on this page --
2005-07-17 12:20 2.6.13rc3: crypto horribly broken on all 64bit archs Andreas Steinmetz
2005-07-17 20:45 ` Herbert Xu
2005-07-18 20:24   ` David S. Miller
2005-07-20  4:41 ` Andy Isaacson

This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox