All of lore.kernel.org
 help / color / mirror / Atom feed
* [PATCH] crypto: serpent: improve __serpent_setkey with UBSAN
@ 2017-07-31 20:43 Arnd Bergmann
  2017-08-09 13:56 ` Herbert Xu
  0 siblings, 1 reply; 2+ messages in thread
From: Arnd Bergmann @ 2017-07-31 20:43 UTC (permalink / raw)
  To: Herbert Xu, David S. Miller; +Cc: Arnd Bergmann, linux-crypto, linux-kernel

When UBSAN is enabled, we get a very large stack frame for
__serpent_setkey, when the register allocator ends up using more registers
than it has, and has to spill temporary values to the stack. The code
was originally optimized for in-order x86-32 CPU implementations using
older compilers, but it now runs into a highly suboptimal case on all
CPU architectures, as seen by this warning:

crypto/serpent_generic.c: In function '__serpent_setkey':
crypto/serpent_generic.c:436:1: error: the frame size of 2720 bytes is larger than 2048 bytes [-Werror=frame-larger-than=]

Disabling -fsanitize=alignment would avoid that warning, presumably the
option turns off a optimization step that is required for getting the
register allocation right, but there is no easy way to do that on gcc-7
(gcc-8 introduces a function attribute for this).

I tried to figure out a way to modify the source code instead, and noticed
that the two stages of the setkey() function (keyiter and sbox) each are
fine by themselves, but not when combined into one function. Splitting
out the entire sbox into a separate function also happens to work fine
with all compilers I tried (arm, arm64 and x86).

The setkey function uses a strange way to handle offsets into the key
array, using both negative and positive index values, as well as adjusting
the array pointer back and forth. I have checked that this actually
makes no difference to modern compilers, but I left that untouched
to make the patch easier to review and to keep the code closer to
the reference implementation.

Link: https://patchwork.kernel.org/patch/9189575/
Signed-off-by: Arnd Bergmann <arnd@arndb.de>
---
 crypto/serpent_generic.c | 77 ++++++++++++++++++++++++++----------------------
 1 file changed, 41 insertions(+), 36 deletions(-)

diff --git a/crypto/serpent_generic.c b/crypto/serpent_generic.c
index 94970a794975..7c3382facc82 100644
--- a/crypto/serpent_generic.c
+++ b/crypto/serpent_generic.c
@@ -229,6 +229,46 @@
 	x4 ^= x2;					\
 	})
 
+static void __serpent_setkey_sbox(u32 r0, u32 r1, u32 r2, u32 r3, u32 r4, u32 *k)
+{
+	k += 100;
+	S3(r3, r4, r0, r1, r2); store_and_load_keys(r1, r2, r4, r3, 28, 24);
+	S4(r1, r2, r4, r3, r0); store_and_load_keys(r2, r4, r3, r0, 24, 20);
+	S5(r2, r4, r3, r0, r1); store_and_load_keys(r1, r2, r4, r0, 20, 16);
+	S6(r1, r2, r4, r0, r3); store_and_load_keys(r4, r3, r2, r0, 16, 12);
+	S7(r4, r3, r2, r0, r1); store_and_load_keys(r1, r2, r0, r4, 12, 8);
+	S0(r1, r2, r0, r4, r3); store_and_load_keys(r0, r2, r4, r1, 8, 4);
+	S1(r0, r2, r4, r1, r3); store_and_load_keys(r3, r4, r1, r0, 4, 0);
+	S2(r3, r4, r1, r0, r2); store_and_load_keys(r2, r4, r3, r0, 0, -4);
+	S3(r2, r4, r3, r0, r1); store_and_load_keys(r0, r1, r4, r2, -4, -8);
+	S4(r0, r1, r4, r2, r3); store_and_load_keys(r1, r4, r2, r3, -8, -12);
+	S5(r1, r4, r2, r3, r0); store_and_load_keys(r0, r1, r4, r3, -12, -16);
+	S6(r0, r1, r4, r3, r2); store_and_load_keys(r4, r2, r1, r3, -16, -20);
+	S7(r4, r2, r1, r3, r0); store_and_load_keys(r0, r1, r3, r4, -20, -24);
+	S0(r0, r1, r3, r4, r2); store_and_load_keys(r3, r1, r4, r0, -24, -28);
+	k -= 50;
+	S1(r3, r1, r4, r0, r2); store_and_load_keys(r2, r4, r0, r3, 22, 18);
+	S2(r2, r4, r0, r3, r1); store_and_load_keys(r1, r4, r2, r3, 18, 14);
+	S3(r1, r4, r2, r3, r0); store_and_load_keys(r3, r0, r4, r1, 14, 10);
+	S4(r3, r0, r4, r1, r2); store_and_load_keys(r0, r4, r1, r2, 10, 6);
+	S5(r0, r4, r1, r2, r3); store_and_load_keys(r3, r0, r4, r2, 6, 2);
+	S6(r3, r0, r4, r2, r1); store_and_load_keys(r4, r1, r0, r2, 2, -2);
+	S7(r4, r1, r0, r2, r3); store_and_load_keys(r3, r0, r2, r4, -2, -6);
+	S0(r3, r0, r2, r4, r1); store_and_load_keys(r2, r0, r4, r3, -6, -10);
+	S1(r2, r0, r4, r3, r1); store_and_load_keys(r1, r4, r3, r2, -10, -14);
+	S2(r1, r4, r3, r2, r0); store_and_load_keys(r0, r4, r1, r2, -14, -18);
+	S3(r0, r4, r1, r2, r3); store_and_load_keys(r2, r3, r4, r0, -18, -22);
+	k -= 50;
+	S4(r2, r3, r4, r0, r1); store_and_load_keys(r3, r4, r0, r1, 28, 24);
+	S5(r3, r4, r0, r1, r2); store_and_load_keys(r2, r3, r4, r1, 24, 20);
+	S6(r2, r3, r4, r1, r0); store_and_load_keys(r4, r0, r3, r1, 20, 16);
+	S7(r4, r0, r3, r1, r2); store_and_load_keys(r2, r3, r1, r4, 16, 12);
+	S0(r2, r3, r1, r4, r0); store_and_load_keys(r1, r3, r4, r2, 12, 8);
+	S1(r1, r3, r4, r2, r0); store_and_load_keys(r0, r4, r2, r1, 8, 4);
+	S2(r0, r4, r2, r1, r3); store_and_load_keys(r3, r4, r0, r1, 4, 0);
+	S3(r3, r4, r0, r1, r2); storekeys(r1, r2, r4, r3, 0);
+}
+
 int __serpent_setkey(struct serpent_ctx *ctx, const u8 *key,
 		     unsigned int keylen)
 {
@@ -395,42 +435,7 @@ int __serpent_setkey(struct serpent_ctx *ctx, const u8 *key,
 	keyiter(k[23], r1, r0, r3, 131, 31);
 
 	/* Apply S-boxes */
-
-	S3(r3, r4, r0, r1, r2); store_and_load_keys(r1, r2, r4, r3, 28, 24);
-	S4(r1, r2, r4, r3, r0); store_and_load_keys(r2, r4, r3, r0, 24, 20);
-	S5(r2, r4, r3, r0, r1); store_and_load_keys(r1, r2, r4, r0, 20, 16);
-	S6(r1, r2, r4, r0, r3); store_and_load_keys(r4, r3, r2, r0, 16, 12);
-	S7(r4, r3, r2, r0, r1); store_and_load_keys(r1, r2, r0, r4, 12, 8);
-	S0(r1, r2, r0, r4, r3); store_and_load_keys(r0, r2, r4, r1, 8, 4);
-	S1(r0, r2, r4, r1, r3); store_and_load_keys(r3, r4, r1, r0, 4, 0);
-	S2(r3, r4, r1, r0, r2); store_and_load_keys(r2, r4, r3, r0, 0, -4);
-	S3(r2, r4, r3, r0, r1); store_and_load_keys(r0, r1, r4, r2, -4, -8);
-	S4(r0, r1, r4, r2, r3); store_and_load_keys(r1, r4, r2, r3, -8, -12);
-	S5(r1, r4, r2, r3, r0); store_and_load_keys(r0, r1, r4, r3, -12, -16);
-	S6(r0, r1, r4, r3, r2); store_and_load_keys(r4, r2, r1, r3, -16, -20);
-	S7(r4, r2, r1, r3, r0); store_and_load_keys(r0, r1, r3, r4, -20, -24);
-	S0(r0, r1, r3, r4, r2); store_and_load_keys(r3, r1, r4, r0, -24, -28);
-	k -= 50;
-	S1(r3, r1, r4, r0, r2); store_and_load_keys(r2, r4, r0, r3, 22, 18);
-	S2(r2, r4, r0, r3, r1); store_and_load_keys(r1, r4, r2, r3, 18, 14);
-	S3(r1, r4, r2, r3, r0); store_and_load_keys(r3, r0, r4, r1, 14, 10);
-	S4(r3, r0, r4, r1, r2); store_and_load_keys(r0, r4, r1, r2, 10, 6);
-	S5(r0, r4, r1, r2, r3); store_and_load_keys(r3, r0, r4, r2, 6, 2);
-	S6(r3, r0, r4, r2, r1); store_and_load_keys(r4, r1, r0, r2, 2, -2);
-	S7(r4, r1, r0, r2, r3); store_and_load_keys(r3, r0, r2, r4, -2, -6);
-	S0(r3, r0, r2, r4, r1); store_and_load_keys(r2, r0, r4, r3, -6, -10);
-	S1(r2, r0, r4, r3, r1); store_and_load_keys(r1, r4, r3, r2, -10, -14);
-	S2(r1, r4, r3, r2, r0); store_and_load_keys(r0, r4, r1, r2, -14, -18);
-	S3(r0, r4, r1, r2, r3); store_and_load_keys(r2, r3, r4, r0, -18, -22);
-	k -= 50;
-	S4(r2, r3, r4, r0, r1); store_and_load_keys(r3, r4, r0, r1, 28, 24);
-	S5(r3, r4, r0, r1, r2); store_and_load_keys(r2, r3, r4, r1, 24, 20);
-	S6(r2, r3, r4, r1, r0); store_and_load_keys(r4, r0, r3, r1, 20, 16);
-	S7(r4, r0, r3, r1, r2); store_and_load_keys(r2, r3, r1, r4, 16, 12);
-	S0(r2, r3, r1, r4, r0); store_and_load_keys(r1, r3, r4, r2, 12, 8);
-	S1(r1, r3, r4, r2, r0); store_and_load_keys(r0, r4, r2, r1, 8, 4);
-	S2(r0, r4, r2, r1, r3); store_and_load_keys(r3, r4, r0, r1, 4, 0);
-	S3(r3, r4, r0, r1, r2); storekeys(r1, r2, r4, r3, 0);
+	__serpent_setkey_sbox(r0, r1, r2, r3, r4, ctx->expkey);
 
 	return 0;
 }
-- 
2.9.0

^ permalink raw reply related	[flat|nested] 2+ messages in thread

* Re: [PATCH] crypto: serpent: improve __serpent_setkey with UBSAN
  2017-07-31 20:43 [PATCH] crypto: serpent: improve __serpent_setkey with UBSAN Arnd Bergmann
@ 2017-08-09 13:56 ` Herbert Xu
  0 siblings, 0 replies; 2+ messages in thread
From: Herbert Xu @ 2017-08-09 13:56 UTC (permalink / raw)
  To: Arnd Bergmann; +Cc: David S. Miller, linux-crypto, linux-kernel

On Mon, Jul 31, 2017 at 10:43:55PM +0200, Arnd Bergmann wrote:
> When UBSAN is enabled, we get a very large stack frame for
> __serpent_setkey, when the register allocator ends up using more registers
> than it has, and has to spill temporary values to the stack. The code
> was originally optimized for in-order x86-32 CPU implementations using
> older compilers, but it now runs into a highly suboptimal case on all
> CPU architectures, as seen by this warning:
> 
> crypto/serpent_generic.c: In function '__serpent_setkey':
> crypto/serpent_generic.c:436:1: error: the frame size of 2720 bytes is larger than 2048 bytes [-Werror=frame-larger-than=]
> 
> Disabling -fsanitize=alignment would avoid that warning, presumably the
> option turns off a optimization step that is required for getting the
> register allocation right, but there is no easy way to do that on gcc-7
> (gcc-8 introduces a function attribute for this).
> 
> I tried to figure out a way to modify the source code instead, and noticed
> that the two stages of the setkey() function (keyiter and sbox) each are
> fine by themselves, but not when combined into one function. Splitting
> out the entire sbox into a separate function also happens to work fine
> with all compilers I tried (arm, arm64 and x86).
> 
> The setkey function uses a strange way to handle offsets into the key
> array, using both negative and positive index values, as well as adjusting
> the array pointer back and forth. I have checked that this actually
> makes no difference to modern compilers, but I left that untouched
> to make the patch easier to review and to keep the code closer to
> the reference implementation.
> 
> Link: https://patchwork.kernel.org/patch/9189575/
> Signed-off-by: Arnd Bergmann <arnd@arndb.de>

Patch applied.  Thanks.
-- 
Email: Herbert Xu <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] 2+ messages in thread

end of thread, other threads:[~2017-08-09 13:57 UTC | newest]

Thread overview: 2+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2017-07-31 20:43 [PATCH] crypto: serpent: improve __serpent_setkey with UBSAN Arnd Bergmann
2017-08-09 13:56 ` Herbert Xu

This is an external index of several public inboxes,
see mirroring instructions on how to clone and mirror
all data and code used by this external index.