Monero Cryptonight variants, and add one for v7

This is the first variant of many, with the intent to improve
Monero's resistance to ASICs and encourage mining decentralization.

src/crypto/chacha.h

@@ -69,10 +69,10 @@ namespace crypto {
     chacha20(data, length, key.data(), reinterpret_cast<const uint8_t*>(&iv), cipher);
   }
 
-  inline void generate_chacha_key(const void *data, size_t size, chacha_key& key, bool prehashed=false) {
+  inline void generate_chacha_key(const void *data, size_t size, chacha_key& key, int cn_variant = 0, bool prehashed=false) {
     static_assert(sizeof(chacha_key) <= sizeof(hash), "Size of hash must be at least that of chacha_key");
     tools::scrubbed_arr<char, HASH_SIZE> pwd_hash;
-    crypto::cn_slow_hash_pre(data, size, pwd_hash.data(), prehashed);
+    crypto::cn_slow_hash_pre(data, size, pwd_hash.data(), cn_variant, prehashed);
     memcpy(&key, pwd_hash.data(), sizeof(key));
   }
 

src/crypto/hash-ops.h

@@ -79,8 +79,8 @@ enum {
 };
 
 void cn_fast_hash(const void *data, size_t length, char *hash);
-void cn_slow_hash(const void *data, size_t length, char *hash);
+void cn_slow_hash(const void *data, size_t length, char *hash, int variant);
-void cn_slow_hash_pre(const void *data, size_t length, char *hash, bool pre);
+void cn_slow_hash_pre(const void *data, size_t length, char *hash, int variant, bool pre);
 
 void hash_extra_blake(const void *data, size_t length, char *hash);
 void hash_extra_groestl(const void *data, size_t length, char *hash);

src/crypto/hash.h

@@ -71,8 +71,8 @@ namespace crypto {
     return h;
   }
 
-  inline void cn_slow_hash(const void *data, std::size_t length, hash &hash) {
+  inline void cn_slow_hash(const void *data, std::size_t length, hash &hash, int variant = 0) {
-    cn_slow_hash(data, length, reinterpret_cast<char *>(&hash));
+    cn_slow_hash(data, length, reinterpret_cast<char *>(&hash), variant);
   }
 
   inline void tree_hash(const hash *hashes, std::size_t count, hash &root_hash) {

src/crypto/slow-hash.c

@@ -32,6 +32,8 @@
 #include <stddef.h>
 #include <stdint.h>
 #include <string.h>
+#include <stdio.h>
+#include <unistd.h>
 
 #include "common/int-util.h"
 #include "hash-ops.h"
@@ -47,6 +49,29 @@
 extern int aesb_single_round(const uint8_t *in, uint8_t*out, const uint8_t *expandedKey);
 extern int aesb_pseudo_round(const uint8_t *in, uint8_t *out, const uint8_t *expandedKey);
 
+#define VARIANT1_1(p) \
+  do if (variant > 0) \
+  { \
+    uint8_t tmp = ((const uint8_t*)p)[11]; \
+    uint8_t tmp1 = (tmp>>4)&1, tmp2 = (tmp>>5)&1, tmp3 = tmp1^tmp2; \
+    uint8_t tmp0 = (((const uint8_t*)p)[11] & 1) ? tmp3 : ((((tmp2<<1)|tmp1) + 1)&3); \
+    ((uint8_t*)p)[11] = (((const uint8_t*)p)[11] & 1) ? ((tmp & 0xef) | (tmp0<<4)):((tmp & 0xcf) | (tmp0<<4)); \
+  } while(0)
+
+#define VARIANT1_2(p) \
+  do if (variant > 0) \
+  { \
+    ((uint32_t*)p)[2] ^= nonce; \
+  } while(0)
+
+#define VARIANT1_INIT() \
+  if (variant > 0 && length < 43) \
+  { \
+    fprintf(stderr, "Cryptonight variants need at least 43 bytes of data"); \
+    _exit(1); \
+  } \
+  const uint32_t nonce = variant > 0 ? *(const uint32_t*)(((const uint8_t*)data)+39) : 0
+
 #if !defined NO_AES && (defined(__x86_64__) || (defined(_MSC_VER) && defined(_WIN64)))
 // Optimised code below, uses x86-specific intrinsics, SSE2, AES-NI
 // Fall back to more portable code is down at the bottom
@@ -125,6 +150,7 @@ extern int aesb_pseudo_round(const uint8_t *in, uint8_t *out, const uint8_t *exp
   _mm_store_si128(R128(c), _c); \
   _b = _mm_xor_si128(_b, _c); \
   _mm_store_si128(R128(&hp_state[j]), _b); \
+  VARIANT1_1(&hp_state[j]); \
   j = state_index(c); \
   p = U64(&hp_state[j]); \
   b[0] = p[0]; b[1] = p[1]; \
@@ -134,6 +160,7 @@ extern int aesb_pseudo_round(const uint8_t *in, uint8_t *out, const uint8_t *exp
   p[0] = a[0];  p[1] = a[1]; \
   a[0] ^= b[0]; a[1] ^= b[1]; \
   _b = _c; \
+  VARIANT1_2(&hp_state[j]); \
 
 #if defined(_MSC_VER)
 #define THREADV __declspec(thread)
@@ -515,11 +542,11 @@ void slow_hash_free_state(void)
  * @param length the length in bytes of the data
  * @param hash a pointer to a buffer in which the final 256 bit hash will be stored
  */
-void cn_slow_hash(const void *data, size_t length, char *hash) {
+void cn_slow_hash(const void *data, size_t length, char *hash, int variant) {
-    cn_slow_hash_pre(data,length,hash,false);
+    cn_slow_hash_pre(data,length,hash,variant,false);
 }
 
-void cn_slow_hash_pre(const void *data, size_t length, char *hash, bool prehashed)
+void cn_slow_hash_pre(const void *data, size_t length, char *hash, int variant, bool prehashed)
 {
     RDATA_ALIGN16 uint8_t expandedKey[240];  /* These buffers are aligned to use later with SSE functions */
 
@@ -541,6 +568,8 @@ void cn_slow_hash_pre(const void *data, size_t length, char *hash, bool prehashe
         hash_extra_blake, hash_extra_groestl, hash_extra_jh, hash_extra_skein
     };
 
+    VARIANT1_INIT();
+
     // this isn't supposed to happen, but guard against it for now.
     if(hp_state == NULL)
         slow_hash_allocate_state();
@@ -712,6 +741,7 @@ union cn_slow_hash_state
   vst1q_u8((uint8_t *)c, _c); \
   _b = veorq_u8(_b, _c); \
   vst1q_u8(&hp_state[j], _b); \
+  VARIANT1_1(&hp_state[j]); \
   j = state_index(c); \
   p = U64(&hp_state[j]); \
   b[0] = p[0]; b[1] = p[1]; \
@@ -720,6 +750,7 @@ union cn_slow_hash_state
   p = U64(&hp_state[j]); \
   p[0] = a[0];  p[1] = a[1]; \
   a[0] ^= b[0]; a[1] ^= b[1]; \
+  VARIANT1_2(p); \
   _b = _c; \
 
 
@@ -851,7 +882,7 @@ STATIC INLINE void aes_pseudo_round_xor(const uint8_t *in, uint8_t *out, const u
 	}
 }
 
-void cn_slow_hash(const void *data, size_t length, char *hash)
+void cn_slow_hash(const void *data, size_t length, char *hash, int variant)
 {
     RDATA_ALIGN16 uint8_t expandedKey[240];
     RDATA_ALIGN16 uint8_t hp_state[MEMORY];
@@ -872,6 +903,8 @@ void cn_slow_hash(const void *data, size_t length, char *hash)
         hash_extra_blake, hash_extra_groestl, hash_extra_jh, hash_extra_skein
     };
 
+    VARIANT1_INIT();
+
     /* CryptoNight Step 1:  Use Keccak1600 to initialize the 'state' (and 'text') buffers from the data. */
 
     hash_process(&state.hs, data, length);
@@ -1045,7 +1078,7 @@ STATIC INLINE void xor_blocks(uint8_t* a, const uint8_t* b)
   U64(a)[1] ^= U64(b)[1];
 }
 
-void cn_slow_hash(const void *data, size_t length, char *hash)
+void cn_slow_hash(const void *data, size_t length, char *hash, int variant)
 {
     uint8_t text[INIT_SIZE_BYTE];
     uint8_t a[AES_BLOCK_SIZE];
@@ -1064,6 +1097,8 @@ void cn_slow_hash(const void *data, size_t length, char *hash)
         hash_extra_blake, hash_extra_groestl, hash_extra_jh, hash_extra_skein
     };
 
+    VARIANT1_INIT();
+
 #ifndef FORCE_USE_HEAP
     uint8_t long_state[MEMORY];
 #else
@@ -1103,6 +1138,7 @@ void cn_slow_hash(const void *data, size_t length, char *hash)
       xor_blocks(b, p);
       swap_blocks(b, p);
       swap_blocks(a, b);
+      VARIANT1_1(p);
 
       // Iteration 2
       p = &long_state[state_index(a)];
@@ -1112,6 +1148,7 @@ void cn_slow_hash(const void *data, size_t length, char *hash)
       swap_blocks(b, p);
       xor_blocks(b, p);
       swap_blocks(a, b);
+      VARIANT1_2(p);
     }
 
     memcpy(text, state.init, INIT_SIZE_BYTE);
@@ -1216,7 +1253,7 @@ union cn_slow_hash_state {
 };
 #pragma pack(pop)
 
-void cn_slow_hash(const void *data, size_t length, char *hash) {
+void cn_slow_hash(const void *data, size_t length, char *hash, int variant) {
   uint8_t long_state[MEMORY];
   union cn_slow_hash_state state;
   uint8_t text[INIT_SIZE_BYTE];
@@ -1228,6 +1265,8 @@ void cn_slow_hash(const void *data, size_t length, char *hash) {
   uint8_t aes_key[AES_KEY_SIZE];
   oaes_ctx *aes_ctx;
 
+  VARIANT1_INIT();
+
   hash_process(&state.hs, data, length);
   memcpy(text, state.init, INIT_SIZE_BYTE);
   memcpy(aes_key, state.hs.b, AES_KEY_SIZE);
@@ -1260,6 +1299,7 @@ void cn_slow_hash(const void *data, size_t length, char *hash) {
     copy_block(&long_state[j * AES_BLOCK_SIZE], c);
     assert(j == e2i(a, MEMORY / AES_BLOCK_SIZE));
     swap_blocks(a, b);
+    VARIANT1_1(&long_state[j * AES_BLOCK_SIZE]);
     /* Iteration 2 */
     j = e2i(a, MEMORY / AES_BLOCK_SIZE);
     copy_block(c, &long_state[j * AES_BLOCK_SIZE]);
@@ -1270,6 +1310,7 @@ void cn_slow_hash(const void *data, size_t length, char *hash) {
     copy_block(&long_state[j * AES_BLOCK_SIZE], c);
     assert(j == e2i(a, MEMORY / AES_BLOCK_SIZE));
     swap_blocks(a, b);
+    VARIANT1_2(&long_state[j * AES_BLOCK_SIZE]);
   }
 
   memcpy(text, state.init, INIT_SIZE_BYTE);

src/cryptonote_basic/cryptonote_format_utils.cpp

@@ -1022,7 +1022,8 @@ namespace cryptonote
       return true;
     }
     blobdata bd = get_block_hashing_blob(b);
-    crypto::cn_slow_hash(bd.data(), bd.size(), res);
+    const int cn_variant = b.major_version >= 7 ? b.major_version - 6 : 0;
+    crypto::cn_slow_hash(bd.data(), bd.size(), res, cn_variant);
     return true;
   }
   //---------------------------------------------------------------

src/device/device_ledger.cpp

@@ -511,7 +511,7 @@ namespace hw {
 
         char prekey[200];
         memmove(prekey, &this->buffer_recv[0], 200);
-        crypto::generate_chacha_key(&prekey[0], sizeof(prekey), key, true);
+        crypto::generate_chacha_key(&prekey[0], sizeof(prekey), key, 0, true);
 
         #ifdef DEBUG_HWDEVICE
         hw::ledger::check32("generate_chacha_key", "key", (char*)key_x.data(), (char*)key.data());

tests/hash/CMakeLists.txt

@@ -43,7 +43,7 @@ set_property(TARGET hash-tests
   PROPERTY
     FOLDER "tests")
 
-foreach (hash IN ITEMS fast slow tree extra-blake extra-groestl extra-jh extra-skein)
+foreach (hash IN ITEMS fast slow slow-1 tree extra-blake extra-groestl extra-jh extra-skein)
   add_test(
     NAME    "hash-${hash}"
     COMMAND hash-tests "${hash}" "${CMAKE_CURRENT_SOURCE_DIR}/tests-${hash}.txt")

tests/hash/main.cpp

@@ -51,6 +51,12 @@ extern "C" {
     }
     tree_hash((const char (*)[32]) data, length >> 5, hash);
   }
+  static void cn_slow_hash_0(const void *data, size_t length, char *hash) {
+    return cn_slow_hash(data, length, hash, 0);
+  }
+  static void cn_slow_hash_1(const void *data, size_t length, char *hash) {
+    return cn_slow_hash(data, length, hash, 1);
+  }
 }
 POP_WARNINGS
 
@@ -58,9 +64,10 @@ extern "C" typedef void hash_f(const void *, size_t, char *);
 struct hash_func {
   const string name;
   hash_f &f;
-} hashes[] = {{"fast", cn_fast_hash}, {"slow", cn_slow_hash}, {"tree", hash_tree},
+} hashes[] = {{"fast", cn_fast_hash}, {"slow", cn_slow_hash_0}, {"tree", hash_tree},
   {"extra-blake", hash_extra_blake}, {"extra-groestl", hash_extra_groestl},
-  {"extra-jh", hash_extra_jh}, {"extra-skein", hash_extra_skein}};
+  {"extra-jh", hash_extra_jh}, {"extra-skein", hash_extra_skein},
+  {"slow-1", cn_slow_hash_1}};
 
 int main(int argc, char *argv[]) {
   hash_f *f;

tests/hash/tests-slow-1.txt

@@ -0,0 +1,5 @@
+b46aab7facce3eb4b4679d1d526f2d500d43736988b7881853c4c0c4af04ac0e 00000000000000000000000000000000000000000000000000000000000000000000000000000000000000
+f0f0ab8a50a809ccd82c76996e494b20a1e01e90ed4814a27db2558511559091 00000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000
+13310922971e1112b382357aeaa8b0bf0cd941dc6f5e398979f3fa77d13749bd 8519e039172b0d70e5ca7b3383d6b3167315a422747b73f019cf9528f0fde341fd0f2a63030ba6450525cf6de31837669af6f1df8131faf50aaab8d3a7405589
+0d9f641d14a782748ed5548f72a20c83c4a2cfe606015ad9eda5b36a00947d72 37a636d7dafdf259b7287eddca2f58099e98619d2f99bdb8969d7b14498102cc065201c8be90bd777323f449848b215d2977c92c4c1c2da36ab46b2e389689ed97c18fec08cd3b03235c5e4c62a37ad88c7b67932495a71090e85dd4020a9300
+d74a0c9b603aa9f9af0bfec0c36d3d383c14930a0ee2f08dab44536fccda5bee 38274c97c45a172cfc97679870422e3a1ab0784960c60514d816271415c306ee3a3ed1a77e31f6a885c3cb