*/
#include "qemu/osdep.h"
+#include "qemu/bswap.h"
#include "crypto/xts.h"
-static void xts_mult_x(uint8_t *I)
+typedef union {
+ uint8_t b[XTS_BLOCK_SIZE];
+ uint64_t u[2];
+} xts_uint128;
+
+static inline void xts_uint128_xor(xts_uint128 *D,
+ const xts_uint128 *S1,
+ const xts_uint128 *S2)
{
- int x;
- uint8_t t, tt;
+ D->u[0] = S1->u[0] ^ S2->u[0];
+ D->u[1] = S1->u[1] ^ S2->u[1];
+}
- for (x = t = 0; x < 16; x++) {
- tt = I[x] >> 7;
- I[x] = ((I[x] << 1) | t) & 0xFF;
- t = tt;
- }
- if (tt) {
- I[0] ^= 0x87;
+static inline void xts_uint128_cpu_to_les(xts_uint128 *v)
+{
+ cpu_to_le64s(&v->u[0]);
+ cpu_to_le64s(&v->u[1]);
+}
+
+static inline void xts_uint128_le_to_cpus(xts_uint128 *v)
+{
+ le64_to_cpus(&v->u[0]);
+ le64_to_cpus(&v->u[1]);
+}
+
+static void xts_mult_x(xts_uint128 *I)
+{
+ uint64_t tt;
+
+ xts_uint128_le_to_cpus(I);
+
+ tt = I->u[0] >> 63;
+ I->u[0] <<= 1;
+
+ if (I->u[1] >> 63) {
+ I->u[0] ^= 0x87;
}
+ I->u[1] <<= 1;
+ I->u[1] |= tt;
+
+ xts_uint128_cpu_to_les(I);
}
*
* Encrypt/decrypt data with a tweak
*/
-static void xts_tweak_encdec(const void *ctx,
- xts_cipher_func *func,
- const uint8_t *src,
- uint8_t *dst,
- uint8_t *iv)
+static inline void xts_tweak_encdec(const void *ctx,
+ xts_cipher_func *func,
+ const xts_uint128 *src,
+ xts_uint128 *dst,
+ xts_uint128 *iv)
{
- unsigned long x;
-
/* tweak encrypt block i */
- for (x = 0; x < XTS_BLOCK_SIZE; x++) {
- dst[x] = src[x] ^ iv[x];
- }
+ xts_uint128_xor(dst, src, iv);
- func(ctx, XTS_BLOCK_SIZE, dst, dst);
+ func(ctx, XTS_BLOCK_SIZE, dst->b, dst->b);
- for (x = 0; x < XTS_BLOCK_SIZE; x++) {
- dst[x] = dst[x] ^ iv[x];
- }
+ xts_uint128_xor(dst, dst, iv);
/* LFSR the tweak */
xts_mult_x(iv);
uint8_t *dst,
const uint8_t *src)
{
- uint8_t PP[XTS_BLOCK_SIZE], CC[XTS_BLOCK_SIZE], T[XTS_BLOCK_SIZE];
+ xts_uint128 PP, CC, T;
unsigned long i, m, mo, lim;
/* get number of blocks */
}
/* encrypt the iv */
- encfunc(tweakctx, XTS_BLOCK_SIZE, T, iv);
-
- for (i = 0; i < lim; i++) {
- xts_tweak_encdec(datactx, decfunc, src, dst, T);
-
- src += XTS_BLOCK_SIZE;
- dst += XTS_BLOCK_SIZE;
+ encfunc(tweakctx, XTS_BLOCK_SIZE, T.b, iv);
+
+ if (QEMU_PTR_IS_ALIGNED(src, sizeof(uint64_t)) &&
+ QEMU_PTR_IS_ALIGNED(dst, sizeof(uint64_t))) {
+ xts_uint128 *S = (xts_uint128 *)src;
+ xts_uint128 *D = (xts_uint128 *)dst;
+ for (i = 0; i < lim; i++, S++, D++) {
+ xts_tweak_encdec(datactx, decfunc, S, D, &T);
+ }
+ } else {
+ xts_uint128 D;
+
+ for (i = 0; i < lim; i++) {
+ memcpy(&D, src, XTS_BLOCK_SIZE);
+ xts_tweak_encdec(datactx, decfunc, &D, &D, &T);
+ memcpy(dst, &D, XTS_BLOCK_SIZE);
+ src += XTS_BLOCK_SIZE;
+ dst += XTS_BLOCK_SIZE;
+ }
}
/* if length is not a multiple of XTS_BLOCK_SIZE then */
if (mo > 0) {
- memcpy(CC, T, XTS_BLOCK_SIZE);
- xts_mult_x(CC);
+ xts_uint128 S, D;
+ memcpy(&CC, &T, XTS_BLOCK_SIZE);
+ xts_mult_x(&CC);
/* PP = tweak decrypt block m-1 */
- xts_tweak_encdec(datactx, decfunc, src, PP, CC);
+ memcpy(&S, src, XTS_BLOCK_SIZE);
+ xts_tweak_encdec(datactx, decfunc, &S, &PP, &CC);
/* Pm = first length % XTS_BLOCK_SIZE bytes of PP */
for (i = 0; i < mo; i++) {
- CC[i] = src[XTS_BLOCK_SIZE + i];
- dst[XTS_BLOCK_SIZE + i] = PP[i];
+ CC.b[i] = src[XTS_BLOCK_SIZE + i];
+ dst[XTS_BLOCK_SIZE + i] = PP.b[i];
}
for (; i < XTS_BLOCK_SIZE; i++) {
- CC[i] = PP[i];
+ CC.b[i] = PP.b[i];
}
/* Pm-1 = Tweak uncrypt CC */
- xts_tweak_encdec(datactx, decfunc, CC, dst, T);
+ xts_tweak_encdec(datactx, decfunc, &CC, &D, &T);
+ memcpy(dst, &D, XTS_BLOCK_SIZE);
}
/* Decrypt the iv back */
- decfunc(tweakctx, XTS_BLOCK_SIZE, iv, T);
+ decfunc(tweakctx, XTS_BLOCK_SIZE, iv, T.b);
}
uint8_t *dst,
const uint8_t *src)
{
- uint8_t PP[XTS_BLOCK_SIZE], CC[XTS_BLOCK_SIZE], T[XTS_BLOCK_SIZE];
+ xts_uint128 PP, CC, T;
unsigned long i, m, mo, lim;
/* get number of blocks */
}
/* encrypt the iv */
- encfunc(tweakctx, XTS_BLOCK_SIZE, T, iv);
+ encfunc(tweakctx, XTS_BLOCK_SIZE, T.b, iv);
+
+ if (QEMU_PTR_IS_ALIGNED(src, sizeof(uint64_t)) &&
+ QEMU_PTR_IS_ALIGNED(dst, sizeof(uint64_t))) {
+ xts_uint128 *S = (xts_uint128 *)src;
+ xts_uint128 *D = (xts_uint128 *)dst;
+ for (i = 0; i < lim; i++, S++, D++) {
+ xts_tweak_encdec(datactx, encfunc, S, D, &T);
+ }
+ } else {
+ xts_uint128 D;
- for (i = 0; i < lim; i++) {
- xts_tweak_encdec(datactx, encfunc, src, dst, T);
+ for (i = 0; i < lim; i++) {
+ memcpy(&D, src, XTS_BLOCK_SIZE);
+ xts_tweak_encdec(datactx, encfunc, &D, &D, &T);
+ memcpy(dst, &D, XTS_BLOCK_SIZE);
- dst += XTS_BLOCK_SIZE;
- src += XTS_BLOCK_SIZE;
+ dst += XTS_BLOCK_SIZE;
+ src += XTS_BLOCK_SIZE;
+ }
}
/* if length is not a multiple of XTS_BLOCK_SIZE then */
if (mo > 0) {
+ xts_uint128 S, D;
/* CC = tweak encrypt block m-1 */
- xts_tweak_encdec(datactx, encfunc, src, CC, T);
+ memcpy(&S, src, XTS_BLOCK_SIZE);
+ xts_tweak_encdec(datactx, encfunc, &S, &CC, &T);
/* Cm = first length % XTS_BLOCK_SIZE bytes of CC */
for (i = 0; i < mo; i++) {
- PP[i] = src[XTS_BLOCK_SIZE + i];
- dst[XTS_BLOCK_SIZE + i] = CC[i];
+ PP.b[i] = src[XTS_BLOCK_SIZE + i];
+ dst[XTS_BLOCK_SIZE + i] = CC.b[i];
}
for (; i < XTS_BLOCK_SIZE; i++) {
- PP[i] = CC[i];
+ PP.b[i] = CC.b[i];
}
/* Cm-1 = Tweak encrypt PP */
- xts_tweak_encdec(datactx, encfunc, PP, dst, T);
+ xts_tweak_encdec(datactx, encfunc, &PP, &D, &T);
+ memcpy(dst, &D, XTS_BLOCK_SIZE);
}
/* Decrypt the iv back */
- decfunc(tweakctx, XTS_BLOCK_SIZE, iv, T);
+ decfunc(tweakctx, XTS_BLOCK_SIZE, iv, T.b);
}
projects / qemu.git / blobdiff