Merge remote-tracking branch 'remotes/berrange/tags/qcrypto-next-pull-request' into staging

Improve performance of XTS cipher mode impl

The XTS cipher mode performance is approximately doubled and test
coverage is improved.

# gpg: Signature made Wed 24 Oct 2018 19:05:08 BST
# gpg:                using RSA key BE86EBB415104FDF
# gpg: Good signature from "Daniel P. Berrange <dan@berrange.com>"
# gpg:                 aka "Daniel P. Berrange <berrange@redhat.com>"
# Primary key fingerprint: DAF3 A6FD B26B 6291 2D0E  8E3F BE86 EBB4 1510 4FDF

* remotes/berrange/tags/qcrypto-next-pull-request:
  crypto: add testing for unaligned buffers with XTS cipher mode
  crypto: refactor XTS cipher mode test suite
  crypto: annotate xts_tweak_encdec as inlineable
  crypto: convert xts_mult_x to use xts_uint128 type
  crypto: convert xts_tweak_encdec to use xts_uint128 type
  crypto: introduce a xts_uint128 data type
  crypto: remove code duplication in tweak encrypt/decrypt
  crypto: expand algorithm coverage for cipher benchmark

Signed-off-by: Peter Maydell <peter.maydell@linaro.org>

3 files changed, 402 insertions, 173 deletions

diff --git a/crypto/xts.c b/crypto/xts.c
index 9521234..4277ad4 100644
--- a/crypto/xts.c
+++ b/crypto/xts.c
@@ -24,52 +24,75 @@
  */
 
 #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);
 }
 
 
 /**
- * xts_tweak_uncrypt:
+ * xts_tweak_encdec:
  * @param ctxt: the cipher context
  * @param func: the cipher function
- * @src: buffer providing the cipher text of XTS_BLOCK_SIZE bytes
- * @dst: buffer to output the plain text of XTS_BLOCK_SIZE bytes
+ * @src: buffer providing the input text of XTS_BLOCK_SIZE bytes
+ * @dst: buffer to output the output text of XTS_BLOCK_SIZE bytes
  * @iv: the initialization vector tweak of XTS_BLOCK_SIZE bytes
  *
- * Decrypt data with a tweak
+ * Encrypt/decrypt data with a tweak
  */
-static void xts_tweak_decrypt(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);
@@ -85,7 +108,7 @@ void xts_decrypt(const void *datactx,
                  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 */
@@ -102,72 +125,53 @@ void xts_decrypt(const void *datactx,
     }
 
     /* encrypt the iv */
-    encfunc(tweakctx, XTS_BLOCK_SIZE, T, iv);
-
-    for (i = 0; i < lim; i++) {
-        xts_tweak_decrypt(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_decrypt(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_decrypt(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);
-}
-
-
-/**
- * xts_tweak_crypt:
- * @param ctxt: the cipher context
- * @param func: the cipher function
- * @src: buffer providing the plain text of XTS_BLOCK_SIZE bytes
- * @dst: buffer to output the cipher text of XTS_BLOCK_SIZE bytes
- * @iv: the initialization vector tweak of XTS_BLOCK_SIZE bytes
- *
- * Encrypt data with a tweak
- */
-static void xts_tweak_encrypt(const void *ctx,
-                              xts_cipher_func *func,
-                              const uint8_t *src,
-                              uint8_t *dst,
-                              uint8_t *iv)
-{
-    unsigned long x;
-
-    /* tweak encrypt block i */
-    for (x = 0; x < XTS_BLOCK_SIZE; x++) {
-        dst[x] = src[x] ^ iv[x];
-    }
-
-    func(ctx, XTS_BLOCK_SIZE, dst, dst);
-
-    for (x = 0; x < XTS_BLOCK_SIZE; x++) {
-        dst[x] = dst[x] ^ iv[x];
-    }
-
-    /* LFSR the tweak */
-    xts_mult_x(iv);
+    decfunc(tweakctx, XTS_BLOCK_SIZE, iv, T.b);
 }
 
 
@@ -180,7 +184,7 @@ void xts_encrypt(const void *datactx,
                  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 */
@@ -197,34 +201,50 @@ void xts_encrypt(const void *datactx,
     }
 
     /* 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_encrypt(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_encrypt(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_encrypt(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);
 }
diff --git a/tests/benchmark-crypto-cipher.c b/tests/benchmark-crypto-cipher.c
index f5a0d0b..67fdf8c 100644
--- a/tests/benchmark-crypto-cipher.c
+++ b/tests/benchmark-crypto-cipher.c
@@ -15,17 +15,27 @@
 #include "crypto/init.h"
 #include "crypto/cipher.h"
 
-static void test_cipher_speed(const void *opaque)
+static void test_cipher_speed(size_t chunk_size,
+                              QCryptoCipherMode mode,
+                              QCryptoCipherAlgorithm alg)
 {
     QCryptoCipher *cipher;
     Error *err = NULL;
     double total = 0.0;
-    size_t chunk_size = (size_t)opaque;
     uint8_t *key = NULL, *iv = NULL;
     uint8_t *plaintext = NULL, *ciphertext = NULL;
-    size_t nkey = qcrypto_cipher_get_key_len(QCRYPTO_CIPHER_ALG_AES_128);
-    size_t niv = qcrypto_cipher_get_iv_len(QCRYPTO_CIPHER_ALG_AES_128,
-                                           QCRYPTO_CIPHER_MODE_CBC);
+    size_t nkey;
+    size_t niv;
+
+    if (!qcrypto_cipher_supports(alg, mode)) {
+        return;
+    }
+
+    nkey = qcrypto_cipher_get_key_len(alg);
+    niv = qcrypto_cipher_get_iv_len(alg, mode);
+    if (mode == QCRYPTO_CIPHER_MODE_XTS) {
+        nkey *= 2;
+    }
 
     key = g_new0(uint8_t, nkey);
     memset(key, g_test_rand_int(), nkey);
@@ -38,14 +48,14 @@ static void test_cipher_speed(const void *opaque)
     plaintext = g_new0(uint8_t, chunk_size);
     memset(plaintext, g_test_rand_int(), chunk_size);
 
-    cipher = qcrypto_cipher_new(QCRYPTO_CIPHER_ALG_AES_128,
-                                QCRYPTO_CIPHER_MODE_CBC,
+    cipher = qcrypto_cipher_new(alg, mode,
                                 key, nkey, &err);
     g_assert(cipher != NULL);
 
-    g_assert(qcrypto_cipher_setiv(cipher,
-                                  iv, niv,
-                                  &err) == 0);
+    if (mode != QCRYPTO_CIPHER_MODE_ECB)
+        g_assert(qcrypto_cipher_setiv(cipher,
+                                      iv, niv,
+                                      &err) == 0);
 
     g_test_timer_start();
     do {
@@ -55,13 +65,26 @@ static void test_cipher_speed(const void *opaque)
                                         chunk_size,
                                         &err) == 0);
         total += chunk_size;
-    } while (g_test_timer_elapsed() < 5.0);
+    } while (g_test_timer_elapsed() < 1.0);
 
     total /= MiB;
-    g_print("cbc(aes128): ");
-    g_print("Testing chunk_size %zu bytes ", chunk_size);
-    g_print("done: %.2f MB in %.2f secs: ", total, g_test_timer_last());
-    g_print("%.2f MB/sec\n", total / g_test_timer_last());
+    g_print("Enc chunk %zu bytes ", chunk_size);
+    g_print("%.2f MB/sec ", total / g_test_timer_last());
+
+    total = 0.0;
+    g_test_timer_start();
+    do {
+        g_assert(qcrypto_cipher_decrypt(cipher,
+                                        plaintext,
+                                        ciphertext,
+                                        chunk_size,
+                                        &err) == 0);
+        total += chunk_size;
+    } while (g_test_timer_elapsed() < 1.0);
+
+    total /= MiB;
+    g_print("Dec chunk %zu bytes ", chunk_size);
+    g_print("%.2f MB/sec ", total / g_test_timer_last());
 
     qcrypto_cipher_free(cipher);
     g_free(plaintext);
@@ -70,19 +93,99 @@ static void test_cipher_speed(const void *opaque)
     g_free(key);
 }
 
-int main(int argc, char **argv)
+
+static void test_cipher_speed_ecb_aes_128(const void *opaque)
+{
+    size_t chunk_size = (size_t)opaque;
+    test_cipher_speed(chunk_size,
+                      QCRYPTO_CIPHER_MODE_ECB,
+                      QCRYPTO_CIPHER_ALG_AES_128);
+}
+
+static void test_cipher_speed_ecb_aes_256(const void *opaque)
 {
-    size_t i;
-    char name[64];
+    size_t chunk_size = (size_t)opaque;
+    test_cipher_speed(chunk_size,
+                      QCRYPTO_CIPHER_MODE_ECB,
+                      QCRYPTO_CIPHER_ALG_AES_256);
+}
+
+static void test_cipher_speed_cbc_aes_128(const void *opaque)
+{
+    size_t chunk_size = (size_t)opaque;
+    test_cipher_speed(chunk_size,
+                      QCRYPTO_CIPHER_MODE_CBC,
+                      QCRYPTO_CIPHER_ALG_AES_128);
+}
 
+static void test_cipher_speed_cbc_aes_256(const void *opaque)
+{
+    size_t chunk_size = (size_t)opaque;
+    test_cipher_speed(chunk_size,
+                      QCRYPTO_CIPHER_MODE_CBC,
+                      QCRYPTO_CIPHER_ALG_AES_256);
+}
+
+static void test_cipher_speed_ctr_aes_128(const void *opaque)
+{
+    size_t chunk_size = (size_t)opaque;
+    test_cipher_speed(chunk_size,
+                      QCRYPTO_CIPHER_MODE_CTR,
+                      QCRYPTO_CIPHER_ALG_AES_128);
+}
+
+static void test_cipher_speed_ctr_aes_256(const void *opaque)
+{
+    size_t chunk_size = (size_t)opaque;
+    test_cipher_speed(chunk_size,
+                      QCRYPTO_CIPHER_MODE_CTR,
+                      QCRYPTO_CIPHER_ALG_AES_256);
+}
+
+static void test_cipher_speed_xts_aes_128(const void *opaque)
+{
+    size_t chunk_size = (size_t)opaque;
+    test_cipher_speed(chunk_size,
+                      QCRYPTO_CIPHER_MODE_XTS,
+                      QCRYPTO_CIPHER_ALG_AES_128);
+}
+
+static void test_cipher_speed_xts_aes_256(const void *opaque)
+{
+    size_t chunk_size = (size_t)opaque;
+    test_cipher_speed(chunk_size,
+                      QCRYPTO_CIPHER_MODE_XTS,
+                      QCRYPTO_CIPHER_ALG_AES_256);
+}
+
+
+int main(int argc, char **argv)
+{
     g_test_init(&argc, &argv, NULL);
     g_assert(qcrypto_init(NULL) == 0);
 
-    for (i = 512; i <= 64 * KiB; i *= 2) {
-        memset(name, 0 , sizeof(name));
-        snprintf(name, sizeof(name), "/crypto/cipher/speed-%zu", i);
-        g_test_add_data_func(name, (void *)i, test_cipher_speed);
-    }
+#define ADD_TEST(mode, cipher, keysize, chunk)                          \
+    g_test_add_data_func(                                               \
+        "/crypto/cipher/" #mode "-" #cipher "-" #keysize "/chunk-" #chunk, \
+        (void *)chunk,                                                  \
+        test_cipher_speed_ ## mode ## _ ## cipher ## _ ## keysize)
+
+#define ADD_TESTS(chunk)                        \
+    do {                                        \
+        ADD_TEST(ecb, aes, 128, chunk);         \
+        ADD_TEST(ecb, aes, 256, chunk);         \
+        ADD_TEST(cbc, aes, 128, chunk);         \
+        ADD_TEST(cbc, aes, 256, chunk);         \
+        ADD_TEST(ctr, aes, 128, chunk);         \
+        ADD_TEST(ctr, aes, 256, chunk);         \
+        ADD_TEST(xts, aes, 128, chunk);         \
+        ADD_TEST(xts, aes, 256, chunk);         \
+    } while (0)
+
+    ADD_TESTS(512);
+    ADD_TESTS(4096);
+    ADD_TESTS(16384);
+    ADD_TESTS(65536);
 
     return g_test_run();
 }
diff --git a/tests/test-crypto-xts.c b/tests/test-crypto-xts.c
index 1f1412c..6fb61cf 100644
--- a/tests/test-crypto-xts.c
+++ b/tests/test-crypto-xts.c
@@ -1,7 +1,7 @@
 /*
  * QEMU Crypto XTS cipher mode
  *
- * Copyright (c) 2015-2016 Red Hat, Inc.
+ * Copyright (c) 2015-2018 Red Hat, Inc.
  *
  * This library is free software; you can redistribute it and/or
  * modify it under the terms of the GNU Lesser General Public
@@ -340,70 +340,161 @@ static void test_xts_aes_decrypt(const void *ctx,
 static void test_xts(const void *opaque)
 {
     const QCryptoXTSTestData *data = opaque;
-    unsigned char out[512], Torg[16], T[16];
+    uint8_t out[512], Torg[16], T[16];
     uint64_t seq;
-    int j;
-    unsigned long len;
     struct TestAES aesdata;
     struct TestAES aestweak;
 
-    for (j = 0; j < 2; j++) {
-        /* skip the cases where
-         * the length is smaller than 2*blocklen
-         * or the length is not a multiple of 32
-         */
-        if ((j == 1) && ((data->PTLEN < 32) || (data->PTLEN % 32))) {
-            continue;
-        }
-        len = data->PTLEN / 2;
-
-        AES_set_encrypt_key(data->key1, data->keylen / 2 * 8, &aesdata.enc);
-        AES_set_decrypt_key(data->key1, data->keylen / 2 * 8, &aesdata.dec);
-        AES_set_encrypt_key(data->key2, data->keylen / 2 * 8, &aestweak.enc);
-        AES_set_decrypt_key(data->key2, data->keylen / 2 * 8, &aestweak.dec);
-
-        seq = data->seqnum;
-        STORE64L(seq, Torg);
-        memset(Torg + 8, 0, 8);
-
-        memcpy(T, Torg, sizeof(T));
-        if (j == 0) {
-            xts_encrypt(&aesdata, &aestweak,
-                        test_xts_aes_encrypt,
-                        test_xts_aes_decrypt,
-                        T, data->PTLEN, out, data->PTX);
-        } else {
-            xts_encrypt(&aesdata, &aestweak,
-                        test_xts_aes_encrypt,
-                        test_xts_aes_decrypt,
-                        T, len, out, data->PTX);
-            xts_encrypt(&aesdata, &aestweak,
-                        test_xts_aes_encrypt,
-                        test_xts_aes_decrypt,
-                        T, len, &out[len], &data->PTX[len]);
-        }
+    AES_set_encrypt_key(data->key1, data->keylen / 2 * 8, &aesdata.enc);
+    AES_set_decrypt_key(data->key1, data->keylen / 2 * 8, &aesdata.dec);
+    AES_set_encrypt_key(data->key2, data->keylen / 2 * 8, &aestweak.enc);
+    AES_set_decrypt_key(data->key2, data->keylen / 2 * 8, &aestweak.dec);
 
-        g_assert(memcmp(out, data->CTX, data->PTLEN) == 0);
-
-        memcpy(T, Torg, sizeof(T));
-        if (j == 0) {
-            xts_decrypt(&aesdata, &aestweak,
-                        test_xts_aes_encrypt,
-                        test_xts_aes_decrypt,
-                        T, data->PTLEN, out, data->CTX);
-        } else {
-            xts_decrypt(&aesdata, &aestweak,
-                        test_xts_aes_encrypt,
-                        test_xts_aes_decrypt,
-                        T, len, out, data->CTX);
-            xts_decrypt(&aesdata, &aestweak,
-                        test_xts_aes_encrypt,
-                        test_xts_aes_decrypt,
-                        T, len, &out[len], &data->CTX[len]);
-        }
+    seq = data->seqnum;
+    STORE64L(seq, Torg);
+    memset(Torg + 8, 0, 8);
 
-        g_assert(memcmp(out, data->PTX, data->PTLEN) == 0);
-    }
+    memcpy(T, Torg, sizeof(T));
+    xts_encrypt(&aesdata, &aestweak,
+                test_xts_aes_encrypt,
+                test_xts_aes_decrypt,
+                T, data->PTLEN, out, data->PTX);
+
+    g_assert(memcmp(out, data->CTX, data->PTLEN) == 0);
+
+    memcpy(T, Torg, sizeof(T));
+    xts_decrypt(&aesdata, &aestweak,
+                test_xts_aes_encrypt,
+                test_xts_aes_decrypt,
+                T, data->PTLEN, out, data->CTX);
+
+    g_assert(memcmp(out, data->PTX, data->PTLEN) == 0);
+}
+
+
+static void test_xts_split(const void *opaque)
+{
+    const QCryptoXTSTestData *data = opaque;
+    uint8_t out[512], Torg[16], T[16];
+    uint64_t seq;
+    unsigned long len = data->PTLEN / 2;
+    struct TestAES aesdata;
+    struct TestAES aestweak;
+
+    AES_set_encrypt_key(data->key1, data->keylen / 2 * 8, &aesdata.enc);
+    AES_set_decrypt_key(data->key1, data->keylen / 2 * 8, &aesdata.dec);
+    AES_set_encrypt_key(data->key2, data->keylen / 2 * 8, &aestweak.enc);
+    AES_set_decrypt_key(data->key2, data->keylen / 2 * 8, &aestweak.dec);
+
+    seq = data->seqnum;
+    STORE64L(seq, Torg);
+    memset(Torg + 8, 0, 8);
+
+    memcpy(T, Torg, sizeof(T));
+    xts_encrypt(&aesdata, &aestweak,
+                test_xts_aes_encrypt,
+                test_xts_aes_decrypt,
+                T, len, out, data->PTX);
+    xts_encrypt(&aesdata, &aestweak,
+                test_xts_aes_encrypt,
+                test_xts_aes_decrypt,
+                T, len, &out[len], &data->PTX[len]);
+
+    g_assert(memcmp(out, data->CTX, data->PTLEN) == 0);
+
+    memcpy(T, Torg, sizeof(T));
+    xts_decrypt(&aesdata, &aestweak,
+                test_xts_aes_encrypt,
+                test_xts_aes_decrypt,
+                T, len, out, data->CTX);
+    xts_decrypt(&aesdata, &aestweak,
+                test_xts_aes_encrypt,
+                test_xts_aes_decrypt,
+                T, len, &out[len], &data->CTX[len]);
+
+    g_assert(memcmp(out, data->PTX, data->PTLEN) == 0);
+}
+
+
+static void test_xts_unaligned(const void *opaque)
+{
+#define BAD_ALIGN 3
+    const QCryptoXTSTestData *data = opaque;
+    uint8_t in[512 + BAD_ALIGN], out[512 + BAD_ALIGN];
+    uint8_t Torg[16], T[16 + BAD_ALIGN];
+    uint64_t seq;
+    struct TestAES aesdata;
+    struct TestAES aestweak;
+
+    AES_set_encrypt_key(data->key1, data->keylen / 2 * 8, &aesdata.enc);
+    AES_set_decrypt_key(data->key1, data->keylen / 2 * 8, &aesdata.dec);
+    AES_set_encrypt_key(data->key2, data->keylen / 2 * 8, &aestweak.enc);
+    AES_set_decrypt_key(data->key2, data->keylen / 2 * 8, &aestweak.dec);
+
+    seq = data->seqnum;
+    STORE64L(seq, Torg);
+    memset(Torg + 8, 0, 8);
+
+    /* IV not aligned */
+    memcpy(T + BAD_ALIGN, Torg, 16);
+    memcpy(in, data->PTX, data->PTLEN);
+    xts_encrypt(&aesdata, &aestweak,
+                test_xts_aes_encrypt,
+                test_xts_aes_decrypt,
+                T + BAD_ALIGN, data->PTLEN, out, in);
+
+    g_assert(memcmp(out, data->CTX, data->PTLEN) == 0);
+
+    /* plain text not aligned */
+    memcpy(T, Torg, 16);
+    memcpy(in + BAD_ALIGN, data->PTX, data->PTLEN);
+    xts_encrypt(&aesdata, &aestweak,
+                test_xts_aes_encrypt,
+                test_xts_aes_decrypt,
+                T, data->PTLEN, out, in + BAD_ALIGN);
+
+    g_assert(memcmp(out, data->CTX, data->PTLEN) == 0);
+
+    /* cipher text not aligned */
+    memcpy(T, Torg, 16);
+    memcpy(in, data->PTX, data->PTLEN);
+    xts_encrypt(&aesdata, &aestweak,
+                test_xts_aes_encrypt,
+                test_xts_aes_decrypt,
+                T, data->PTLEN, out + BAD_ALIGN, in);
+
+    g_assert(memcmp(out + BAD_ALIGN, data->CTX, data->PTLEN) == 0);
+
+
+    /* IV not aligned */
+    memcpy(T + BAD_ALIGN, Torg, 16);
+    memcpy(in, data->CTX, data->PTLEN);
+    xts_decrypt(&aesdata, &aestweak,
+                test_xts_aes_encrypt,
+                test_xts_aes_decrypt,
+                T + BAD_ALIGN, data->PTLEN, out, in);
+
+    g_assert(memcmp(out, data->PTX, data->PTLEN) == 0);
+
+    /* cipher text not aligned */
+    memcpy(T, Torg, 16);
+    memcpy(in + BAD_ALIGN, data->CTX, data->PTLEN);
+    xts_decrypt(&aesdata, &aestweak,
+                test_xts_aes_encrypt,
+                test_xts_aes_decrypt,
+                T, data->PTLEN, out, in + BAD_ALIGN);
+
+    g_assert(memcmp(out, data->PTX, data->PTLEN) == 0);
+
+    /* plain text not aligned */
+    memcpy(T, Torg, 16);
+    memcpy(in, data->CTX, data->PTLEN);
+    xts_decrypt(&aesdata, &aestweak,
+                test_xts_aes_encrypt,
+                test_xts_aes_decrypt,
+                T, data->PTLEN, out + BAD_ALIGN, in);
+
+    g_assert(memcmp(out + BAD_ALIGN, data->PTX, data->PTLEN) == 0);
 }
 
 
@@ -416,7 +507,22 @@ int main(int argc, char **argv)
     g_assert(qcrypto_init(NULL) == 0);
 
     for (i = 0; i < G_N_ELEMENTS(test_data); i++) {
-        g_test_add_data_func(test_data[i].path, &test_data[i], test_xts);
+        gchar *path = g_strdup_printf("%s/basic", test_data[i].path);
+        g_test_add_data_func(path, &test_data[i], test_xts);
+        g_free(path);
+
+        /* skip the cases where the length is smaller than 2*blocklen
+         * or the length is not a multiple of 32
+         */
+        if ((test_data[i].PTLEN >= 32) && !(test_data[i].PTLEN % 32)) {
+            path = g_strdup_printf("%s/split", test_data[i].path);
+            g_test_add_data_func(path, &test_data[i], test_xts_split);
+            g_free(path);
+        }
+
+        path = g_strdup_printf("%s/unaligned", test_data[i].path);
+        g_test_add_data_func(path, &test_data[i], test_xts_unaligned);
+        g_free(path);
     }
 
     return g_test_run();