mirror/openssl
2
0
Fork 0
mirror of https://github.com/openssl/openssl.git synced 2025-03-19 19:50:42 +08:00
Code Issues Packages Projects Releases Wiki Activity

Avoid undefined behavior with unaligned accesses

Browse Source 
Fixes: #4983

[extended tests]

Reviewed-by: Nicola Tuveri <nic.tuv@gmail.com>
(Merged from https://github.com/openssl/openssl/pull/6074)
This commit is contained in:
Bernd Edlinger 2018-04-24 21:10:13 +02:00
parent c74aaa3920
commit 77286fe3ec
12 changed files with 147 additions and 72 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

4
.travis.yml
View File

@ -103,10 +103,10 @@ jobs:
env: EXTENDED_TEST="yes" CONFIG_OPTS="enable-msan disable-afalgeng -Wno-unused-command-line-argument"
- os: linux
compiler: clang
env: EXTENDED_TEST="yes" CONFIG_OPTS="no-asm enable-ubsan enable-rc5 enable-md2 enable-ssl3 enable-ssl3-method enable-nextprotoneg no-shared enable-buildtest-c++ -fno-sanitize=alignment -DFUZZING_BUILD_MODE_UNSAFE_FOR_PRODUCTION -Wno-unused-command-line-argument" CXX="clang++"
env: EXTENDED_TEST="yes" CONFIG_OPTS="no-asm enable-ubsan enable-rc5 enable-md2 enable-ssl3 enable-ssl3-method enable-nextprotoneg no-shared enable-buildtest-c++ -DFUZZING_BUILD_MODE_UNSAFE_FOR_PRODUCTION -Wno-unused-command-line-argument" CXX="clang++"
- os: linux
compiler: gcc
env: EXTENDED_TEST="yes" CONFIG_OPTS="--debug no-asm enable-ubsan enable-rc5 enable-md2 enable-buildtest-c++ -DPEDANTIC" OPENSSL_TEST_RAND_ORDER=0
env: EXTENDED_TEST="yes" CONFIG_OPTS="--debug no-asm enable-asan enable-ubsan enable-rc5 enable-md2 enable-ec_nistp_64_gcc_128 enable-buildtest-c++" OPENSSL_TEST_RAND_ORDER=0
- os: linux
dist: xenial
addons:

14
crypto/aes/aes_ige.c
View File

@ -18,11 +18,6 @@
#include <openssl/aes.h>
#include "aes_local.h"
#define N_WORDS (AES_BLOCK_SIZE / sizeof(unsigned long))
typedef struct {
unsigned long data[N_WORDS];
} aes_block_t;
/* XXX: probably some better way to do this */
#if defined(__i386__) || defined(__x86_64__)
# define UNALIGNED_MEMOPS_ARE_FAST 1
@ -30,6 +25,15 @@ typedef struct {
# define UNALIGNED_MEMOPS_ARE_FAST 0
#endif
#define N_WORDS (AES_BLOCK_SIZE / sizeof(unsigned long))
typedef struct {
unsigned long data[N_WORDS];
#if defined(__GNUC__) && UNALIGNED_MEMOPS_ARE_FAST
} aes_block_t __attribute((__aligned__(1)));
#else
} aes_block_t;
#endif
#if UNALIGNED_MEMOPS_ARE_FAST
# define load_block(d, s) (d) = *(const aes_block_t *)(s)
# define store_block(d, s) *(aes_block_t *)(d) = (s)

9
crypto/ec/ecp_nistp224.c
View File

@ -75,6 +75,7 @@ typedef uint64_t u64;
*/
typedef uint64_t limb;
typedef uint64_t limb_aX __attribute((__aligned__(1)));
typedef uint128_t widelimb;
typedef limb felem[4];
@ -311,10 +312,10 @@ const EC_METHOD *EC_GFp_nistp224_method(void)
*/
static void bin28_to_felem(felem out, const u8 in[28])
{
out[0] = *((const uint64_t *)(in)) & 0x00ffffffffffffff;
out[1] = (*((const uint64_t *)(in + 7))) & 0x00ffffffffffffff;
out[2] = (*((const uint64_t *)(in + 14))) & 0x00ffffffffffffff;
out[3] = (*((const uint64_t *)(in+20))) >> 8;
out[0] = *((const limb *)(in)) & 0x00ffffffffffffff;
out[1] = (*((const limb_aX *)(in + 7))) & 0x00ffffffffffffff;
out[2] = (*((const limb_aX *)(in + 14))) & 0x00ffffffffffffff;
out[3] = (*((const limb_aX *)(in + 20))) >> 8;
}
static void felem_to_bin28(u8 out[28], const felem in)

33
crypto/ec/ecp_nistp521.c
View File

@ -131,6 +131,7 @@ static const felem_bytearray nistp521_curve_params[5] = {
#define NLIMBS 9
typedef uint64_t limb;
typedef limb limb_aX __attribute((__aligned__(1)));
typedef limb felem[NLIMBS];
typedef uint128_t largefelem[NLIMBS];
@ -144,14 +145,14 @@ static const limb bottom58bits = 0x3ffffffffffffff;
static void bin66_to_felem(felem out, const u8 in[66])
{
out[0] = (*((limb *) & in[0])) & bottom58bits;
out[1] = (*((limb *) & in[7]) >> 2) & bottom58bits;
out[2] = (*((limb *) & in[14]) >> 4) & bottom58bits;
out[3] = (*((limb *) & in[21]) >> 6) & bottom58bits;
out[4] = (*((limb *) & in[29])) & bottom58bits;
out[5] = (*((limb *) & in[36]) >> 2) & bottom58bits;
out[6] = (*((limb *) & in[43]) >> 4) & bottom58bits;
out[7] = (*((limb *) & in[50]) >> 6) & bottom58bits;
out[8] = (*((limb *) & in[58])) & bottom57bits;
out[1] = (*((limb_aX *) & in[7]) >> 2) & bottom58bits;
out[2] = (*((limb_aX *) & in[14]) >> 4) & bottom58bits;
out[3] = (*((limb_aX *) & in[21]) >> 6) & bottom58bits;
out[4] = (*((limb_aX *) & in[29])) & bottom58bits;
out[5] = (*((limb_aX *) & in[36]) >> 2) & bottom58bits;
out[6] = (*((limb_aX *) & in[43]) >> 4) & bottom58bits;
out[7] = (*((limb_aX *) & in[50]) >> 6) & bottom58bits;
out[8] = (*((limb_aX *) & in[58])) & bottom57bits;
}
/*
@ -162,14 +163,14 @@ static void felem_to_bin66(u8 out[66], const felem in)
{
memset(out, 0, 66);
(*((limb *) & out[0])) = in[0];
(*((limb *) & out[7])) |= in[1] << 2;
(*((limb *) & out[14])) |= in[2] << 4;
(*((limb *) & out[21])) |= in[3] << 6;
(*((limb *) & out[29])) = in[4];
(*((limb *) & out[36])) |= in[5] << 2;
(*((limb *) & out[43])) |= in[6] << 4;
(*((limb *) & out[50])) |= in[7] << 6;
(*((limb *) & out[58])) = in[8];
(*((limb_aX *) & out[7])) |= in[1] << 2;
(*((limb_aX *) & out[14])) |= in[2] << 4;
(*((limb_aX *) & out[21])) |= in[3] << 6;
(*((limb_aX *) & out[29])) = in[4];
(*((limb_aX *) & out[36])) |= in[5] << 2;
(*((limb_aX *) & out[43])) |= in[6] << 4;
(*((limb_aX *) & out[50])) |= in[7] << 6;
(*((limb_aX *) & out[58])) = in[8];
}
/* BN_to_felem converts an OpenSSL BIGNUM into an felem */

19
crypto/modes/cbc128.c
View File

@ -15,6 +15,12 @@
# define STRICT_ALIGNMENT 0
#endif
#if defined(__GNUC__) && !STRICT_ALIGNMENT
typedef size_t size_t_aX __attribute((__aligned__(1)));
#else
typedef size_t size_t_aX;
#endif
void CRYPTO_cbc128_encrypt(const unsigned char *in, unsigned char *out,
size_t len, const void *key,
unsigned char ivec[16], block128_f block)
@ -40,8 +46,8 @@ void CRYPTO_cbc128_encrypt(const unsigned char *in, unsigned char *out,
} else {
while (len >= 16) {
for (n = 0; n < 16; n += sizeof(size_t))
*(size_t *)(out + n) =
*(size_t *)(in + n) ^ *(size_t *)(iv + n);
*(size_t_aX *)(out + n) =
*(size_t_aX *)(in + n) ^ *(size_t_aX *)(iv + n);
(*block) (out, out, key);
iv = out;
len -= 16;
@ -96,7 +102,8 @@ void CRYPTO_cbc128_decrypt(const unsigned char *in, unsigned char *out,
}
} else if (16 % sizeof(size_t) == 0) { /* always true */
while (len >= 16) {
size_t *out_t = (size_t *)out, *iv_t = (size_t *)iv;
size_t_aX *out_t = (size_t_aX *)out;
size_t_aX *iv_t = (size_t_aX *)iv;
(*block) (in, out, key);
for (n = 0; n < 16 / sizeof(size_t); n++)
@ -125,8 +132,10 @@ void CRYPTO_cbc128_decrypt(const unsigned char *in, unsigned char *out,
}
} else if (16 % sizeof(size_t) == 0) { /* always true */
while (len >= 16) {
size_t c, *out_t = (size_t *)out, *ivec_t = (size_t *)ivec;
const size_t *in_t = (const size_t *)in;
size_t c;
size_t_aX *out_t = (size_t_aX *)out;
size_t_aX *ivec_t = (size_t_aX *)ivec;
const size_t_aX *in_t = (const size_t_aX *)in;
(*block) (in, tmp.c, key);
for (n = 0; n < 16 / sizeof(size_t); n++) {

22
crypto/modes/ccm128.c
View File

@ -11,6 +11,14 @@
#include <openssl/crypto.h>
#include "crypto/modes.h"
#ifndef STRICT_ALIGNMENT
# ifdef __GNUC__
typedef u64 u64_a1 __attribute((__aligned__(1)));
# else
typedef u64 u64_a1;
# endif
#endif
/*
* First you setup M and L parameters and pass the key schedule. This is
* called once per session setup...
@ -170,8 +178,8 @@ int CRYPTO_ccm128_encrypt(CCM128_CONTEXT *ctx,
ctx->cmac.u[0] ^= temp.u[0];
ctx->cmac.u[1] ^= temp.u[1];
#else
ctx->cmac.u[0] ^= ((u64 *)inp)[0];
ctx->cmac.u[1] ^= ((u64 *)inp)[1];
ctx->cmac.u[0] ^= ((u64_a1 *)inp)[0];
ctx->cmac.u[1] ^= ((u64_a1 *)inp)[1];
#endif
(*block) (ctx->cmac.c, ctx->cmac.c, key);
(*block) (ctx->nonce.c, scratch.c, key);
@ -181,8 +189,8 @@ int CRYPTO_ccm128_encrypt(CCM128_CONTEXT *ctx,
temp.u[1] ^= scratch.u[1];
memcpy(out, temp.c, 16);
#else
((u64 *)out)[0] = scratch.u[0] ^ ((u64 *)inp)[0];
((u64 *)out)[1] = scratch.u[1] ^ ((u64 *)inp)[1];
((u64_a1 *)out)[0] = scratch.u[0] ^ ((u64_a1 *)inp)[0];
((u64_a1 *)out)[1] = scratch.u[1] ^ ((u64_a1 *)inp)[1];
#endif
inp += 16;
out += 16;
@ -254,8 +262,10 @@ int CRYPTO_ccm128_decrypt(CCM128_CONTEXT *ctx,
ctx->cmac.u[1] ^= (scratch.u[1] ^= temp.u[1]);
memcpy(out, scratch.c, 16);
#else
ctx->cmac.u[0] ^= (((u64 *)out)[0] = scratch.u[0] ^ ((u64 *)inp)[0]);
ctx->cmac.u[1] ^= (((u64 *)out)[1] = scratch.u[1] ^ ((u64 *)inp)[1]);
ctx->cmac.u[0] ^= (((u64_a1 *)out)[0]
= scratch.u[0] ^ ((u64_a1 *)inp)[0]);
ctx->cmac.u[1] ^= (((u64_a1 *)out)[1]
= scratch.u[1] ^ ((u64_a1 *)inp)[1]);
#endif
(*block) (ctx->cmac.c, ctx->cmac.c, key);

18
crypto/modes/cfb128.c
View File

@ -11,6 +11,12 @@
#include <openssl/crypto.h>
#include "crypto/modes.h"
#if defined(__GNUC__) && !defined(STRICT_ALIGNMENT)
typedef size_t size_t_aX __attribute((__aligned__(1)));
#else
typedef size_t size_t_aX;
#endif
/*
* The input and output encrypted as though 128bit cfb mode is being used.
* The extra state information to record how much of the 128bit block we have
@ -43,8 +49,9 @@ void CRYPTO_cfb128_encrypt(const unsigned char *in, unsigned char *out,
while (len >= 16) {
(*block) (ivec, ivec, key);
for (; n < 16; n += sizeof(size_t)) {
*(size_t *)(out + n) =
*(size_t *)(ivec + n) ^= *(size_t *)(in + n);
*(size_t_aX *)(out + n) =
*(size_t_aX *)(ivec + n)
^= *(size_t_aX *)(in + n);
}
len -= 16;
out += 16;
@ -92,9 +99,10 @@ void CRYPTO_cfb128_encrypt(const unsigned char *in, unsigned char *out,
while (len >= 16) {
(*block) (ivec, ivec, key);
for (; n < 16; n += sizeof(size_t)) {
size_t t = *(size_t *)(in + n);
*(size_t *)(out + n) = *(size_t *)(ivec + n) ^ t;
*(size_t *)(ivec + n) = t;
size_t t = *(size_t_aX *)(in + n);
*(size_t_aX *)(out + n)
= *(size_t_aX *)(ivec + n) ^ t;
*(size_t_aX *)(ivec + n) = t;
}
len -= 16;
out += 16;

11
crypto/modes/ctr128.c
View File

@ -11,6 +11,12 @@
#include <openssl/crypto.h>
#include "crypto/modes.h"
#if defined(__GNUC__) && !defined(STRICT_ALIGNMENT)
typedef size_t size_t_aX __attribute((__aligned__(1)));
#else
typedef size_t size_t_aX;
#endif
/*
* NOTE: the IV/counter CTR mode is big-endian. The code itself is
* endian-neutral.
@ -97,8 +103,9 @@ void CRYPTO_ctr128_encrypt(const unsigned char *in, unsigned char *out,
(*block) (ivec, ecount_buf, key);
ctr128_inc_aligned(ivec);
for (n = 0; n < 16; n += sizeof(size_t))
*(size_t *)(out + n) =
*(size_t *)(in + n) ^ *(size_t *)(ecount_buf + n);
*(size_t_aX *)(out + n) =
*(size_t_aX *)(in + n)
^ *(size_t_aX *)(ecount_buf + n);
len -= 16;
out += 16;
in += 16;

22
crypto/modes/gcm128.c
View File

@ -12,6 +12,12 @@
#include "internal/cryptlib.h"
#include "crypto/modes.h"
#if defined(__GNUC__) && !defined(STRICT_ALIGNMENT)
typedef size_t size_t_aX __attribute((__aligned__(1)));
#else
typedef size_t size_t_aX;
#endif
#if defined(BSWAP4) && defined(STRICT_ALIGNMENT)
/* redefine, because alignment is ensured */
# undef GETU32
@ -1080,8 +1086,8 @@ int CRYPTO_gcm128_encrypt(GCM128_CONTEXT *ctx,
size_t j = GHASH_CHUNK;
while (j) {
size_t *out_t = (size_t *)out;
const size_t *in_t = (const size_t *)in;
size_t_aX *out_t = (size_t_aX *)out;
const size_t_aX *in_t = (const size_t_aX *)in;
(*block) (ctx->Yi.c, ctx->EKi.c, key);
++ctr;
@ -1107,8 +1113,8 @@ int CRYPTO_gcm128_encrypt(GCM128_CONTEXT *ctx,
size_t j = i;
while (len >= 16) {
size_t *out_t = (size_t *)out;
const size_t *in_t = (const size_t *)in;
size_t_aX *out_t = (size_t_aX *)out;
const size_t_aX *in_t = (const size_t_aX *)in;
(*block) (ctx->Yi.c, ctx->EKi.c, key);
++ctr;
@ -1318,8 +1324,8 @@ int CRYPTO_gcm128_decrypt(GCM128_CONTEXT *ctx,
GHASH(ctx, in, GHASH_CHUNK);
while (j) {
size_t *out_t = (size_t *)out;
const size_t *in_t = (const size_t *)in;
size_t_aX *out_t = (size_t_aX *)out;
const size_t_aX *in_t = (const size_t_aX *)in;
(*block) (ctx->Yi.c, ctx->EKi.c, key);
++ctr;
@ -1343,8 +1349,8 @@ int CRYPTO_gcm128_decrypt(GCM128_CONTEXT *ctx,
if ((i = (len & (size_t)-16))) {
GHASH(ctx, in, i);
while (len >= 16) {
size_t *out_t = (size_t *)out;
const size_t *in_t = (const size_t *)in;
size_t_aX *out_t = (size_t_aX *)out;
const size_t_aX *in_t = (const size_t_aX *)in;
(*block) (ctx->Yi.c, ctx->EKi.c, key);
++ctr;

11
crypto/modes/ofb128.c
View File

@ -11,6 +11,12 @@
#include <openssl/crypto.h>
#include "crypto/modes.h"
#if defined(__GNUC__) && !defined(STRICT_ALIGNMENT)
typedef size_t size_t_aX __attribute((__aligned__(1)));
#else
typedef size_t size_t_aX;
#endif
/*
* The input and output encrypted as though 128bit ofb mode is being used.
* The extra state information to record how much of the 128bit block we have
@ -41,8 +47,9 @@ void CRYPTO_ofb128_encrypt(const unsigned char *in, unsigned char *out,
while (len >= 16) {
(*block) (ivec, ivec, key);
for (; n < 16; n += sizeof(size_t))
*(size_t *)(out + n) =
*(size_t *)(in + n) ^ *(size_t *)(ivec + n);
*(size_t_aX *)(out + n) =
*(size_t_aX *)(in + n)
^ *(size_t_aX *)(ivec + n);
len -= 16;
out += 16;
in += 16;

24
crypto/modes/xts128.c
View File

@ -11,6 +11,14 @@
#include <openssl/crypto.h>
#include "crypto/modes.h"
#ifndef STRICT_ALIGNMENT
# ifdef __GNUC__
typedef u64 u64_a1 __attribute((__aligned__(1)));
# else
typedef u64 u64_a1;
# endif
#endif
int CRYPTO_xts128_encrypt(const XTS128_CONTEXT *ctx,
const unsigned char iv[16],
const unsigned char *inp, unsigned char *out,
@ -45,8 +53,8 @@ int CRYPTO_xts128_encrypt(const XTS128_CONTEXT *ctx,
scratch.u[0] ^= tweak.u[0];
scratch.u[1] ^= tweak.u[1];
#else
scratch.u[0] = ((u64 *)inp)[0] ^ tweak.u[0];
scratch.u[1] = ((u64 *)inp)[1] ^ tweak.u[1];
scratch.u[0] = ((u64_a1 *)inp)[0] ^ tweak.u[0];
scratch.u[1] = ((u64_a1 *)inp)[1] ^ tweak.u[1];
#endif
(*ctx->block1) (scratch.c, scratch.c, ctx->key1);
#if defined(STRICT_ALIGNMENT)
@ -54,8 +62,8 @@ int CRYPTO_xts128_encrypt(const XTS128_CONTEXT *ctx,
scratch.u[1] ^= tweak.u[1];
memcpy(out, scratch.c, 16);
#else
((u64 *)out)[0] = scratch.u[0] ^= tweak.u[0];
((u64 *)out)[1] = scratch.u[1] ^= tweak.u[1];
((u64_a1 *)out)[0] = scratch.u[0] ^= tweak.u[0];
((u64_a1 *)out)[1] = scratch.u[1] ^= tweak.u[1];
#endif
inp += 16;
out += 16;
@ -128,8 +136,8 @@ int CRYPTO_xts128_encrypt(const XTS128_CONTEXT *ctx,
scratch.u[0] ^= tweak1.u[0];
scratch.u[1] ^= tweak1.u[1];
#else
scratch.u[0] = ((u64 *)inp)[0] ^ tweak1.u[0];
scratch.u[1] = ((u64 *)inp)[1] ^ tweak1.u[1];
scratch.u[0] = ((u64_a1 *)inp)[0] ^ tweak1.u[0];
scratch.u[1] = ((u64_a1 *)inp)[1] ^ tweak1.u[1];
#endif
(*ctx->block1) (scratch.c, scratch.c, ctx->key1);
scratch.u[0] ^= tweak1.u[0];
@ -148,8 +156,8 @@ int CRYPTO_xts128_encrypt(const XTS128_CONTEXT *ctx,
scratch.u[1] ^= tweak.u[1];
memcpy(out, scratch.c, 16);
#else
((u64 *)out)[0] = scratch.u[0] ^ tweak.u[0];
((u64 *)out)[1] = scratch.u[1] ^ tweak.u[1];
((u64_a1 *)out)[0] = scratch.u[0] ^ tweak.u[0];
((u64_a1 *)out)[1] = scratch.u[1] ^ tweak.u[1];
#endif
}

32
crypto/whrlpool/wp_block.c
View File

@ -70,6 +70,20 @@ typedef unsigned long long u64;
# undef STRICT_ALIGNMENT
#endif
#ifndef STRICT_ALIGNMENT
# ifdef __GNUC__
typedef u64 u64_a1 __attribute((__aligned__(1)));
# else
typedef u64 u64_a1;
# endif
#endif
#if defined(__GNUC__) && !defined(STRICT_ALIGNMENT)
typedef u64 u64_aX __attribute((__aligned__(1)));
#else
typedef u64 u64_aX;
#endif
#undef SMALL_REGISTER_BANK
#if defined(__i386) || defined(__i386__) || defined(_M_IX86)
# define SMALL_REGISTER_BANK
@ -197,13 +211,13 @@ typedef unsigned long long u64;
# define LL(c0,c1,c2,c3,c4,c5,c6,c7) c0,c1,c2,c3,c4,c5,c6,c7, \
c0,c1,c2,c3,c4,c5,c6,c7
# define C0(K,i) (((u64*)(Cx.c+0))[2*K.c[(i)*8+0]])
# define C1(K,i) (((u64*)(Cx.c+7))[2*K.c[(i)*8+1]])
# define C2(K,i) (((u64*)(Cx.c+6))[2*K.c[(i)*8+2]])
# define C3(K,i) (((u64*)(Cx.c+5))[2*K.c[(i)*8+3]])
# define C4(K,i) (((u64*)(Cx.c+4))[2*K.c[(i)*8+4]])
# define C5(K,i) (((u64*)(Cx.c+3))[2*K.c[(i)*8+5]])
# define C6(K,i) (((u64*)(Cx.c+2))[2*K.c[(i)*8+6]])
# define C7(K,i) (((u64*)(Cx.c+1))[2*K.c[(i)*8+7]])
# define C1(K,i) (((u64_a1*)(Cx.c+7))[2*K.c[(i)*8+1]])
# define C2(K,i) (((u64_a1*)(Cx.c+6))[2*K.c[(i)*8+2]])
# define C3(K,i) (((u64_a1*)(Cx.c+5))[2*K.c[(i)*8+3]])
# define C4(K,i) (((u64_a1*)(Cx.c+4))[2*K.c[(i)*8+4]])
# define C5(K,i) (((u64_a1*)(Cx.c+3))[2*K.c[(i)*8+5]])
# define C6(K,i) (((u64_a1*)(Cx.c+2))[2*K.c[(i)*8+6]])
# define C7(K,i) (((u64_a1*)(Cx.c+1))[2*K.c[(i)*8+7]])
#endif
static const
@ -537,7 +551,7 @@ void whirlpool_block(WHIRLPOOL_CTX *ctx, const void *inp, size_t n)
} else
# endif
{
const u64 *pa = (const u64 *)p;
const u64_aX *pa = (const u64_aX *)p;
S.q[0] = (K.q[0] = H->q[0]) ^ pa[0];
S.q[1] = (K.q[1] = H->q[1]) ^ pa[1];
S.q[2] = (K.q[2] = H->q[2]) ^ pa[2];
@ -775,7 +789,7 @@ void whirlpool_block(WHIRLPOOL_CTX *ctx, const void *inp, size_t n)
} else
# endif
{
const u64 *pa = (const u64 *)p;
const u64_aX *pa = (const u64_aX *)p;
H->q[0] ^= S.q[0] ^ pa[0];
H->q[1] ^= S.q[1] ^ pa[1];
H->q[2] ^= S.q[2] ^ pa[2];