fixes

CMakeLists.txt

@@ -12,7 +12,7 @@ function(set_static_flags)
     set(CMAKE_EXE_LINKER_FLAGS "${CMAKE_EXE_LINKER_FLAGS} -static-libgcc -static-libstdc++")
   endif()
 endfunction(set_static_flags)
-  
+
 include_directories(src contrib/epee/include external "${CMAKE_BINARY_DIR}/version")
 
 if(APPLE)
@@ -21,6 +21,11 @@ endif()
 
 set(STATIC ${MSVC} CACHE BOOL "Link libraries statically")
 
+if (UNIX AND NOT APPLE)
+  # Note that at the time of this writing the -Wstrict-prototypes flag added below will make this fail
+  find_package(Threads REQUIRED)
+endif()
+
 if(MSVC)
   add_definitions("/bigobj /MP /W3 /GS- /D_CRT_SECURE_NO_WARNINGS /wd4996 /wd4345 /D_WIN32_WINNT=0x0600 /DWIN32_LEAN_AND_MEAN /DGTEST_HAS_TR1_TUPLE=0 /FIinline_c.h /D__SSE4_1__")
   # set(CMAKE_C_FLAGS "${CMAKE_C_FLAGS} /Dinline=__inline")
@@ -44,6 +49,11 @@ else()
   else()
     set(WARNINGS "${WARNINGS} -Wlogical-op -Wno-error=maybe-uninitialized")
   endif()
+  # Since gcc 4.9 the LTO format is non-standard (slim), so we need the gcc-specific ar and ranlib binaries
+  if(CMAKE_CXX_COMPILER_ID STREQUAL "GNU" AND NOT (CMAKE_CXX_COMPILER_VERSION VERSION_LESS 4.9.0))
+    set(CMAKE_AR "gcc-ar")
+    set(CMAKE_RANLIB "gcc-ranlib")
+  endif()
   if(MINGW)
     set(WARNINGS "${WARNINGS} -Wno-error=unused-value")
     set(MINGW_FLAG "-DWIN32_LEAN_AND_MEAN")
@@ -70,7 +80,7 @@ else()
     set(DEBUG_FLAGS "-g3 -O0")
   endif()
   set(RELEASE_FLAGS "-Ofast -DNDEBUG -Wno-unused-variable")
-  if(NOT APPLE)
+  if(NOT CMAKE_CXX_COMPILER_ID STREQUAL "Clang")
     # There is a clang bug that does not allow to compile code that uses AES-NI intrinsics if -flto is enabled
     set(RELEASE_FLAGS "${RELEASE_FLAGS} -flto")
   endif()

Doxyfile

@@ -32,7 +32,7 @@ DOXYFILE_ENCODING      = UTF-8
 # title of most generated pages and in a few other places.
 # The default value is: My Project.
 
-PROJECT_NAME           = "Bitmonero"
+PROJECT_NAME           = "Aeon"
 
 # The PROJECT_NUMBER tag can be used to enter a project or revision number. This
 # could be handy for archiving the generated documentation or if some version

src/CMakeLists.txt

@@ -30,13 +30,13 @@ add_library(cryptonote_core ${CRYPTONOTE_CORE})
 add_executable(daemon ${DAEMON} ${P2P} ${CRYPTONOTE_PROTOCOL})
 add_executable(connectivity_tool ${CONN_TOOL})
 add_executable(simpleminer ${MINER})
-target_link_libraries(daemon rpc cryptonote_core crypto common upnpc-static ${Boost_LIBRARIES})
-target_link_libraries(connectivity_tool cryptonote_core crypto common ${Boost_LIBRARIES})
-target_link_libraries(simpleminer cryptonote_core crypto common ${Boost_LIBRARIES})
+target_link_libraries(daemon rpc cryptonote_core crypto common upnpc-static ${CMAKE_THREAD_LIBS_INIT} ${Boost_LIBRARIES})
+target_link_libraries(connectivity_tool cryptonote_core crypto common ${CMAKE_THREAD_LIBS_INIT} ${Boost_LIBRARIES})
+target_link_libraries(simpleminer cryptonote_core crypto common ${CMAKE_THREAD_LIBS_INIT} ${Boost_LIBRARIES})
 add_library(rpc ${RPC})
 add_library(wallet ${WALLET})
 add_executable(simplewallet ${SIMPLEWALLET} )
-target_link_libraries(simplewallet wallet rpc cryptonote_core crypto common upnpc-static ${Boost_LIBRARIES})
+target_link_libraries(simplewallet wallet rpc cryptonote_core crypto common upnpc-static ${CMAKE_THREAD_LIBS_INIT} ${Boost_LIBRARIES})
 add_dependencies(daemon version)
 add_dependencies(rpc version)
 add_dependencies(simplewallet version)

src/common/command_line.cpp

@@ -3,9 +3,21 @@
 // file COPYING or http://www.opensource.org/licenses/mit-license.php.
 
 #include "command_line.h"
+#include "string_tools.h"
 
 namespace command_line
 {
+  std::string input_line(const std::string& prompt)
+  {
+    std::cout << prompt;
+
+    std::string buf;
+    std::getline(std::cin, buf);
+
+    return epee::string_tools::trim(buf);
+
+  }
+
   const arg_descriptor<bool> arg_help = {"help", "Produce help message"};
   const arg_descriptor<bool> arg_version = {"version", "Output version information"};
   const arg_descriptor<std::string> arg_data_dir = {"data-dir", "Specify data directory"};

src/common/command_line.h

@@ -14,6 +14,9 @@
 
 namespace command_line
 {
+
+  std::string input_line(const std::string& prompt);
+
   template<typename T, bool required = false>
   struct arg_descriptor;
 

src/crypto/slow-hash.c

@@ -13,8 +13,9 @@
 
 #include <emmintrin.h>
 
-#if defined(_MSC_VER) || defined(__INTEL_COMPILER)
+#if defined(_MSC_VER)
 #include <intrin.h>
+#include <Windows.h>
 #define STATIC
 #define INLINE __inline
 #if !defined(RDATA_ALIGN16)
@@ -22,6 +23,7 @@
 #endif
 #else
 #include <wmmintrin.h>
+#include <sys/mman.h>
 #define STATIC static
 #define INLINE inline
 #if !defined(RDATA_ALIGN16)
@@ -29,16 +31,66 @@
 #endif
 #endif
 
+#if defined(__INTEL_COMPILER)
+#define ASM __asm__
+#elif !defined(_MSC_VER)
+#define ASM __asm__
+#else
+#define ASM __asm
+#endif
+
 #define MEMORY         (1 << 21) // 2MB scratchpad
 #define ITER           (1 << 20)
 #define AES_BLOCK_SIZE  16
 #define AES_KEY_SIZE    32
 #define INIT_SIZE_BLK   8
 #define INIT_SIZE_BYTE (INIT_SIZE_BLK * AES_BLOCK_SIZE)
+#define TOTALBLOCKS (MEMORY / AES_BLOCK_SIZE)
 
 #define U64(x) ((uint64_t *) (x))
 #define R128(x) ((__m128i *) (x))
 
+#define state_index(x) (((*((uint64_t *)x) >> 4) & (TOTALBLOCKS - 1)) << 4)
+#if defined(_MSC_VER)
+#if !defined(_WIN64)
+#define __mul() lo = mul128(c[0], b[0], &hi);
+#else
+#define __mul() lo = _umul128(c[0], b[0], &hi);
+#endif
+#else
+#if defined(__x86_64__)
+#define __mul() ASM("mulq %3\n\t" : "=d"(hi), "=a"(lo) : "%a" (c[0]), "rm" (b[0]) : "cc");
+#else
+#define __mul() lo = mul128(c[0], b[0], &hi);
+#endif
+#endif
+
+#define pre_aes() \
+    j = state_index(a); \
+	_c = _mm_load_si128(R128(&hp_state[j])); \
+	_a = _mm_load_si128(R128(a)); \
+ 
+// dga's optimized scratchpad twiddling
+#define post_aes() \
+	_mm_store_si128(R128(c), _c); \
+	_b = _mm_xor_si128(_b, _c); \
+	_mm_store_si128(R128(&hp_state[j]), _b); \
+	j = state_index(c); \
+	p = U64(&hp_state[j]); \
+	b[0] = p[0]; b[1] = p[1]; \
+	__mul(); \
+	a[0] += hi; a[1] += lo; \
+	p = U64(&hp_state[j]); \
+	p[0] = a[0];  p[1] = a[1]; \
+	a[0] ^= b[0]; a[1] ^= b[1]; \
+	_b = _c; \
+ 
+#if defined(_MSC_VER)
+#define THREADV __declspec(thread)
+#else
+#define THREADV __thread
+#endif
+
 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);
 
@@ -54,59 +106,26 @@ union cn_slow_hash_state
 };
 #pragma pack(pop)
 
-#if defined(_MSC_VER) || defined(__INTEL_COMPILER)
+THREADV uint8_t *hp_state = NULL;
+THREADV int hp_allocated = 0;
+
+#if defined(_MSC_VER)
 #define cpuid(info,x)    __cpuidex(info,x,0)
 #else
 void cpuid(int CPUInfo[4], int InfoType)
 {
-    __asm__ __volatile__
+    ASM __volatile__
     (
-        "cpuid":
+    "cpuid":
         "=a" (CPUInfo[0]),
         "=b" (CPUInfo[1]),
         "=c" (CPUInfo[2]),
         "=d" (CPUInfo[3]) :
-        "a" (InfoType), "c" (0)
-    );
+            "a" (InfoType), "c" (0)
+        );
 }
 #endif
 
-STATIC INLINE void mul(const uint8_t *a, const uint8_t *b, uint8_t *res)
-{
-    uint64_t a0, b0;
-    uint64_t hi, lo;
-
-    a0 = U64(a)[0];
-    b0 = U64(b)[0];
-    lo = mul128(a0, b0, &hi);
-    U64(res)[0] = hi;
-    U64(res)[1] = lo;
-}
-
-STATIC INLINE void sum_half_blocks(uint8_t *a, const uint8_t *b)
-{
-    uint64_t a0, a1, b0, b1;
-    a0 = U64(a)[0];
-    a1 = U64(a)[1];
-    b0 = U64(b)[0];
-    b1 = U64(b)[1];
-    a0 += b0;
-    a1 += b1;
-    U64(a)[0] = a0;
-    U64(a)[1] = a1;
-}
-
-STATIC INLINE void swap_blocks(uint8_t *a, uint8_t *b)
-{
-    uint64_t t[2];
-    U64(t)[0] = U64(a)[0];
-    U64(t)[1] = U64(a)[1];
-    U64(a)[0] = U64(b)[0];
-    U64(a)[1] = U64(b)[1];
-    U64(b)[0] = U64(t)[0];
-    U64(b)[1] = U64(t)[1];
-}
-
 STATIC INLINE void xor_blocks(uint8_t *a, const uint8_t *b)
 {
     U64(a)[0] ^= U64(b)[0];
@@ -125,74 +144,253 @@ STATIC INLINE int check_aes_hw(void)
     return supported = cpuid_results[2] & (1 << 25);
 }
 
-STATIC INLINE void aesni_pseudo_round(const uint8_t *in, uint8_t *out,
-                                      const uint8_t *expandedKey)
+STATIC INLINE void aes_256_assist1(__m128i* t1, __m128i * t2)
+{
+    __m128i t4;
+    *t2 = _mm_shuffle_epi32(*t2, 0xff);
+    t4 = _mm_slli_si128(*t1, 0x04);
+    *t1 = _mm_xor_si128(*t1, t4);
+    t4 = _mm_slli_si128(t4, 0x04);
+    *t1 = _mm_xor_si128(*t1, t4);
+    t4 = _mm_slli_si128(t4, 0x04);
+    *t1 = _mm_xor_si128(*t1, t4);
+    *t1 = _mm_xor_si128(*t1, *t2);
+}
+
+STATIC INLINE void aes_256_assist2(__m128i* t1, __m128i * t3)
+{
+    __m128i t2, t4;
+    t4 = _mm_aeskeygenassist_si128(*t1, 0x00);
+    t2 = _mm_shuffle_epi32(t4, 0xaa);
+    t4 = _mm_slli_si128(*t3, 0x04);
+    *t3 = _mm_xor_si128(*t3, t4);
+    t4 = _mm_slli_si128(t4, 0x04);
+    *t3 = _mm_xor_si128(*t3, t4);
+    t4 = _mm_slli_si128(t4, 0x04);
+    *t3 = _mm_xor_si128(*t3, t4);
+    *t3 = _mm_xor_si128(*t3, t2);
+}
+
+STATIC INLINE void aes_expand_key(const uint8_t *key, uint8_t *expandedKey)
+{
+    __m128i *ek = R128(expandedKey);
+    __m128i t1, t2, t3;
+
+    t1 = _mm_loadu_si128(R128(key));
+    t3 = _mm_loadu_si128(R128(key + 16));
+
+    ek[0] = t1;
+    ek[1] = t3;
+
+    t2 = _mm_aeskeygenassist_si128(t3, 0x01);
+    aes_256_assist1(&t1, &t2);
+    ek[2] = t1;
+    aes_256_assist2(&t1, &t3);
+    ek[3] = t3;
+
+    t2 = _mm_aeskeygenassist_si128(t3, 0x02);
+    aes_256_assist1(&t1, &t2);
+    ek[4] = t1;
+    aes_256_assist2(&t1, &t3);
+    ek[5] = t3;
+
+    t2 = _mm_aeskeygenassist_si128(t3, 0x04);
+    aes_256_assist1(&t1, &t2);
+    ek[6] = t1;
+    aes_256_assist2(&t1, &t3);
+    ek[7] = t3;
+
+    t2 = _mm_aeskeygenassist_si128(t3, 0x08);
+    aes_256_assist1(&t1, &t2);
+    ek[8] = t1;
+    aes_256_assist2(&t1, &t3);
+    ek[9] = t3;
+
+    t2 = _mm_aeskeygenassist_si128(t3, 0x10);
+    aes_256_assist1(&t1, &t2);
+    ek[10] = t1;
+}
+
+STATIC INLINE void aes_pseudo_round(const uint8_t *in, uint8_t *out,
+                                    const uint8_t *expandedKey, int nblocks)
 {
     __m128i *k = R128(expandedKey);
     __m128i d;
+    int i;
 
-    d = _mm_loadu_si128(R128(in));
-    d = _mm_aesenc_si128(d, *R128(&k[0]));
-    d = _mm_aesenc_si128(d, *R128(&k[1]));
-    d = _mm_aesenc_si128(d, *R128(&k[2]));
-    d = _mm_aesenc_si128(d, *R128(&k[3]));
-    d = _mm_aesenc_si128(d, *R128(&k[4]));
-    d = _mm_aesenc_si128(d, *R128(&k[5]));
-    d = _mm_aesenc_si128(d, *R128(&k[6]));
-    d = _mm_aesenc_si128(d, *R128(&k[7]));
-    d = _mm_aesenc_si128(d, *R128(&k[8]));
-    d = _mm_aesenc_si128(d, *R128(&k[9]));
-    _mm_storeu_si128((R128(out)), d);
+    for(i = 0; i < nblocks; i++)
+    {
+        d = _mm_loadu_si128(R128(in + i * AES_BLOCK_SIZE));
+        d = _mm_aesenc_si128(d, *R128(&k[0]));
+        d = _mm_aesenc_si128(d, *R128(&k[1]));
+        d = _mm_aesenc_si128(d, *R128(&k[2]));
+        d = _mm_aesenc_si128(d, *R128(&k[3]));
+        d = _mm_aesenc_si128(d, *R128(&k[4]));
+        d = _mm_aesenc_si128(d, *R128(&k[5]));
+        d = _mm_aesenc_si128(d, *R128(&k[6]));
+        d = _mm_aesenc_si128(d, *R128(&k[7]));
+        d = _mm_aesenc_si128(d, *R128(&k[8]));
+        d = _mm_aesenc_si128(d, *R128(&k[9]));
+        _mm_storeu_si128((R128(out + i * AES_BLOCK_SIZE)), d);
+    }
+}
+
+STATIC INLINE void aes_pseudo_round_xor(const uint8_t *in, uint8_t *out,
+                                        const uint8_t *expandedKey, const uint8_t *xor, int nblocks)
+{
+    __m128i *k = R128(expandedKey);
+    __m128i *x = R128(xor);
+    __m128i d;
+    int i;
+
+    for(i = 0; i < nblocks; i++)
+    {
+        d = _mm_loadu_si128(R128(in + i * AES_BLOCK_SIZE));
+        d = _mm_xor_si128(d, *R128(x++));
+        d = _mm_aesenc_si128(d, *R128(&k[0]));
+        d = _mm_aesenc_si128(d, *R128(&k[1]));
+        d = _mm_aesenc_si128(d, *R128(&k[2]));
+        d = _mm_aesenc_si128(d, *R128(&k[3]));
+        d = _mm_aesenc_si128(d, *R128(&k[4]));
+        d = _mm_aesenc_si128(d, *R128(&k[5]));
+        d = _mm_aesenc_si128(d, *R128(&k[6]));
+        d = _mm_aesenc_si128(d, *R128(&k[7]));
+        d = _mm_aesenc_si128(d, *R128(&k[8]));
+        d = _mm_aesenc_si128(d, *R128(&k[9]));
+        _mm_storeu_si128((R128(out + i * AES_BLOCK_SIZE)), d);
+    }
+}
+
+#if defined(_MSC_VER)
+BOOL SetLockPagesPrivilege(HANDLE hProcess, BOOL bEnable)
+{
+    struct
+    {
+        DWORD count;
+        LUID_AND_ATTRIBUTES privilege[1];
+    } info;
+
+    HANDLE token;
+    if(!OpenProcessToken(hProcess, TOKEN_ADJUST_PRIVILEGES, &token))
+        return FALSE;
+
+    info.count = 1;
+    info.privilege[0].Attributes = bEnable ? SE_PRIVILEGE_ENABLED : 0;
+
+    if(!LookupPrivilegeValue(NULL, SE_LOCK_MEMORY_NAME, &(info.privilege[0].Luid)))
+        return FALSE;
+
+    if(!AdjustTokenPrivileges(token, FALSE, (PTOKEN_PRIVILEGES) &info, 0, NULL, NULL))
+        return FALSE;
+
+    if (GetLastError() != ERROR_SUCCESS)
+        return FALSE;
+
+    CloseHandle(token);
+
+    return TRUE;
+
+}
+#endif
+
+void slow_hash_allocate_state(void)
+{
+    int state = 0;
+    if(hp_state != NULL)
+        return;
+
+#if defined(_MSC_VER)
+    SetLockPagesPrivilege(GetCurrentProcess(), TRUE);
+    hp_state = (uint8_t *) VirtualAlloc(hp_state, MEMORY, MEM_LARGE_PAGES |
+                                        MEM_COMMIT | MEM_RESERVE, PAGE_READWRITE);
+#else
+#if defined(__APPLE__)
+    hp_state = mmap(0, MEMORY, PROT_READ | PROT_WRITE,
+                    MAP_PRIVATE | MAP_ANON, 0, 0);    
+#else
+    hp_state = mmap(0, MEMORY, PROT_READ | PROT_WRITE,
+                    MAP_PRIVATE | MAP_ANONYMOUS | MAP_HUGETLB, 0, 0);
+#endif
+    if(hp_state == MAP_FAILED)
+        hp_state = NULL;
+#endif
+    hp_allocated = 1;
+    if(hp_state == NULL)
+    {
+        hp_allocated = 0;
+        hp_state = (uint8_t *) malloc(MEMORY);
+    }
+}
+
+void slow_hash_free_state(void)
+{
+    if(hp_state == NULL)
+        return;
+
+    if(!hp_allocated)
+        free(hp_state);
+    else
+    {
+#if defined(_MSC_VER)
+        VirtualFree(hp_state, MEMORY, MEM_RELEASE);
+#else
+        munmap(hp_state, MEMORY);
+#endif
+    }
+
+    hp_state = NULL;
+    hp_allocated = 0;
 }
 
 void cn_slow_hash(const void *data, size_t length, char *hash)
 {
-    uint8_t long_state[MEMORY];
-    uint8_t text[INIT_SIZE_BYTE];
-    uint8_t a[AES_BLOCK_SIZE];
-    uint8_t b[AES_BLOCK_SIZE];
-    uint8_t d[AES_BLOCK_SIZE];
-    uint8_t aes_key[AES_KEY_SIZE];
-    RDATA_ALIGN16 uint8_t expandedKey[256];
+    RDATA_ALIGN16 uint8_t expandedKey[240];
 
+    uint8_t text[INIT_SIZE_BYTE];
+    RDATA_ALIGN16 uint64_t a[2];
+    RDATA_ALIGN16 uint64_t b[2];
+    RDATA_ALIGN16 uint64_t c[2];
+    RDATA_ALIGN16 uint8_t aes_key[AES_KEY_SIZE];
     union cn_slow_hash_state state;
+    __m128i _a, _b, _c;
+    uint64_t hi, lo;
 
     size_t i, j;
-    uint8_t *p = NULL;
+    uint64_t *p = NULL;
     oaes_ctx *aes_ctx;
-
     int useAes = check_aes_hw();
+
     static void (*const extra_hashes[4])(const void *, size_t, char *) =
     {
         hash_extra_blake, hash_extra_groestl, hash_extra_jh, hash_extra_skein
     };
 
+    // this isn't supposed to happen, but guard against it for now.
+    if(hp_state == NULL)
+        slow_hash_allocate_state();
+
     hash_process(&state.hs, data, length);
     memcpy(text, state.init, INIT_SIZE_BYTE);
 
-    aes_ctx = (oaes_ctx *) oaes_alloc();
-    oaes_key_import_data(aes_ctx, state.hs.b, AES_KEY_SIZE);
-
-    // use aligned data
-    memcpy(expandedKey, aes_ctx->key->exp_data, aes_ctx->key->exp_data_len);
-
     if(useAes)
     {
+        aes_expand_key(state.hs.b, expandedKey);
         for(i = 0; i < MEMORY / INIT_SIZE_BYTE; i++)
         {
-            for(j = 0; j < INIT_SIZE_BLK; j++)
-                aesni_pseudo_round(&text[AES_BLOCK_SIZE * j], &text[AES_BLOCK_SIZE * j], expandedKey);
-            memcpy(&long_state[i * INIT_SIZE_BYTE], text, INIT_SIZE_BYTE);
+            aes_pseudo_round(text, text, expandedKey, INIT_SIZE_BLK);
+            memcpy(&hp_state[i * INIT_SIZE_BYTE], text, INIT_SIZE_BYTE);
         }
     }
     else
     {
+        aes_ctx = (oaes_ctx *) oaes_alloc();
+        oaes_key_import_data(aes_ctx, state.hs.b, AES_KEY_SIZE);
         for(i = 0; i < MEMORY / INIT_SIZE_BYTE; i++)
         {
             for(j = 0; j < INIT_SIZE_BLK; j++)
-                aesb_pseudo_round(&text[AES_BLOCK_SIZE * j], &text[AES_BLOCK_SIZE * j], expandedKey);
+                aesb_pseudo_round(&text[AES_BLOCK_SIZE * j], &text[AES_BLOCK_SIZE * j], aes_ctx->key->exp_data);
 
-            memcpy(&long_state[i * INIT_SIZE_BYTE], text, INIT_SIZE_BYTE);
+            memcpy(&hp_state[i * INIT_SIZE_BYTE], text, INIT_SIZE_BYTE);
         }
     }
 
@@ -201,60 +399,52 @@ void cn_slow_hash(const void *data, size_t length, char *hash)
     U64(b)[0] = U64(&state.k[16])[0] ^ U64(&state.k[48])[0];
     U64(b)[1] = U64(&state.k[16])[1] ^ U64(&state.k[48])[1];
 
-    for(i = 0; i < ITER / 2; i++)
-    {
-				#define TOTALBLOCKS (MEMORY / AES_BLOCK_SIZE)
-				#define state_index(x) (((*((uint64_t *)x) >> 4) & (TOTALBLOCKS - 1)) << 4)
-
-        // Iteration 1
-        p = &long_state[state_index(a)];
-
-        if(useAes)
-            _mm_storeu_si128(R128(p), _mm_aesenc_si128(_mm_loadu_si128(R128(p)), _mm_loadu_si128(R128(a))));
-        else
-            aesb_single_round(p, p, a);
-
-        xor_blocks(b, p);
-        swap_blocks(b, p);
-        swap_blocks(a, b);
-
-        // Iteration 2
-        p = &long_state[state_index(a)];
-
-        mul(a, p, d);
-        sum_half_blocks(b, d);
-        swap_blocks(b, p);
-        xor_blocks(b, p);
-        swap_blocks(a, b);
-    }
-
-    memcpy(text, state.init, INIT_SIZE_BYTE);
-    oaes_key_import_data(aes_ctx, &state.hs.b[32], AES_KEY_SIZE);
-    memcpy(expandedKey, aes_ctx->key->exp_data, aes_ctx->key->exp_data_len);
+    _b = _mm_load_si128(R128(b));
+    // this is ugly but the branching affects the loop somewhat so put it outside.
     if(useAes)
     {
-        for(i = 0; i < MEMORY / INIT_SIZE_BYTE; i++)
+        for(i = 0; i < ITER / 2; i++)
         {
-            for(j = 0; j < INIT_SIZE_BLK; j++)
-            {
-                xor_blocks(&text[j * AES_BLOCK_SIZE], &long_state[i * INIT_SIZE_BYTE + j * AES_BLOCK_SIZE]);
-                aesni_pseudo_round(&text[j * AES_BLOCK_SIZE], &text[j * AES_BLOCK_SIZE], expandedKey);
-            }
+            pre_aes();
+            _c = _mm_aesenc_si128(_c, _a);
+            // post_aes(), optimized scratchpad twiddling (credits to dga)
+            post_aes();
         }
     }
     else
     {
+        for(i = 0; i < ITER / 2; i++)
+        {
+            pre_aes();
+            aesb_single_round((uint8_t *) &_c, (uint8_t *) &_c, (uint8_t *) &_a);
+            post_aes();
+        }
+    }
+
+    memcpy(text, state.init, INIT_SIZE_BYTE);
+    if(useAes)
+    {
+        aes_expand_key(&state.hs.b[32], expandedKey);
+        for(i = 0; i < MEMORY / INIT_SIZE_BYTE; i++)
+        {
+            // add the xor to the pseudo round
+            aes_pseudo_round_xor(text, text, expandedKey, &hp_state[i * INIT_SIZE_BYTE], INIT_SIZE_BLK);
+        }
+    }
+    else
+    {
+        oaes_key_import_data(aes_ctx, &state.hs.b[32], AES_KEY_SIZE);
         for(i = 0; i < MEMORY / INIT_SIZE_BYTE; i++)
         {
             for(j = 0; j < INIT_SIZE_BLK; j++)
             {
-                xor_blocks(&text[j * AES_BLOCK_SIZE], &long_state[i * INIT_SIZE_BYTE + j * AES_BLOCK_SIZE]);
-                aesb_pseudo_round(&text[AES_BLOCK_SIZE * j], &text[AES_BLOCK_SIZE * j], expandedKey);
+                xor_blocks(&text[j * AES_BLOCK_SIZE], &hp_state[i * INIT_SIZE_BYTE + j * AES_BLOCK_SIZE]);
+                aesb_pseudo_round(&text[AES_BLOCK_SIZE * j], &text[AES_BLOCK_SIZE * j], aes_ctx->key->exp_data);
             }
         }
+        oaes_free((OAES_CTX **) &aes_ctx);
     }
 
-    oaes_free((OAES_CTX **) &aes_ctx);
     memcpy(state.init, text, INIT_SIZE_BYTE);
     hash_permutation(&state.hs);
     extra_hashes[state.hs.b[0] & 3](&state, 200, hash);

src/cryptonote_config.h

@@ -51,6 +51,8 @@
 #define BLOCKS_SYNCHRONIZING_DEFAULT_COUNT              200    //by default, blocks count in blocks downloading
 #define CRYPTONOTE_PROTOCOL_HOP_RELAX_COUNT             3      //value of hop, after which we use only announce of new block
 
+#define CRYPTONOTE_MEMPOOL_TX_LIVETIME                    86400 //seconds, one day
+#define CRYPTONOTE_MEMPOOL_TX_FROM_ALT_BLOCK_LIVETIME     604800 //seconds, one week
 
 #define P2P_DEFAULT_PORT                                11180
 #define RPC_DEFAULT_PORT                                11181

src/cryptonote_core/checkpoints_create.h

@@ -11,7 +11,9 @@
 
 namespace cryptonote {
   inline bool create_checkpoints(cryptonote::checkpoints& checkpoints)
-  {
+  {     
+    ADD_CHECKPOINT(10000, "1ac1ebd25baf0d6ec593daa3389f1aa7e860ff2cc29f3cf1be586d245b379da4");
+    ADD_CHECKPOINT(15000, "15567af42afc1ed00538f53b5e3822d421e3ed6372ca79f4ea4e3e3bab709a87");
     return true;
   }
 }

src/cryptonote_core/cryptonote_core.cpp

@@ -432,6 +432,18 @@ namespace cryptonote
     return true;
   }
   //-----------------------------------------------------------------------------------------------
+  // Used by the RPC server to check the size of an incoming
+  // block_blob
+  bool core::check_incoming_block_size(const blobdata& block_blob)
+  {
+    if(block_blob.size() > get_max_block_size())
+    {
+      LOG_PRINT_L0("WRONG BLOCK BLOB, too big size " << block_blob.size() << ", rejected");
+      return false;
+    }
+    return true;
+  }
+  //-----------------------------------------------------------------------------------------------
   crypto::hash core::get_tail_id()
   {
     return m_blockchain_storage.get_tail_id();
@@ -514,6 +526,7 @@ namespace cryptonote
 
     m_store_blockchain_interval.do_call(boost::bind(&blockchain_storage::store_blockchain, &m_blockchain_storage));
     m_miner.on_idle();
+    m_mempool.on_idle();
     return true;
   }
   //-----------------------------------------------------------------------------------------------

src/cryptonote_core/cryptonote_core.h

@@ -34,6 +34,7 @@ namespace cryptonote
      bool on_idle();
      bool handle_incoming_tx(const blobdata& tx_blob, tx_verification_context& tvc, bool keeped_by_block);
      bool handle_incoming_block(const blobdata& block_blob, block_verification_context& bvc, bool update_miner_blocktemplate = true);
+     bool check_incoming_block_size(const blobdata& block_blob);
      i_cryptonote_protocol* get_protocol(){return m_pprotocol;}
 
      //-------------------- i_miner_handler -----------------------

src/cryptonote_core/cryptonote_format_utils.cpp

@@ -617,6 +617,7 @@ namespace cryptonote
     //genesis block
     bl = boost::value_initialized<block>();
 
+
     account_public_address ac = boost::value_initialized<account_public_address>();
     std::vector<size_t> sz;
     construct_miner_tx(0, 0, 0, 0, 0, ac, bl.miner_tx); // zero fee in genesis
@@ -630,8 +631,8 @@ namespace cryptonote
     string_tools::parse_hexstr_to_binbuff(genesis_coinbase_tx_hex, tx_bl);
     bool r = parse_and_validate_tx_from_blob(tx_bl, bl.miner_tx);
     CHECK_AND_ASSERT_MES(r, false, "failed to parse coinbase tx from hard coded blob");
-    bl.major_version = 1;
-    bl.minor_version = 0;
+    bl.major_version = CURRENT_BLOCK_MAJOR_VERSION;
+    bl.minor_version = CURRENT_BLOCK_MINOR_VERSION;
     bl.timestamp = 0;
     bl.nonce = 10000;
     miner::find_nonce_for_given_block(bl, 1, 0);

src/cryptonote_core/miner.cpp

@@ -23,7 +23,8 @@ using namespace epee;
 #include "miner.h"
 
 
-
+extern "C" void slow_hash_allocate_state();
+extern "C" void slow_hash_free_state();
 namespace cryptonote
 {
 
@@ -320,6 +321,7 @@ namespace cryptonote
     difficulty_type local_diff = 0;
     uint32_t local_template_ver = 0;
     block b;
+	  slow_hash_allocate_state();
     while(!m_stop)
     {
       if(m_pausers_count)//anti split workaround
@@ -368,6 +370,7 @@ namespace cryptonote
       nonce+=m_threads_total;
       ++m_hashes;
     }
+	  slow_hash_free_state();
     LOG_PRINT_L0("Miner thread stopped ["<< th_local_index << "]");
     return true;
   }

src/cryptonote_core/tx_pool.cpp

@@ -83,6 +83,7 @@ namespace cryptonote
         txd_p.first->second.max_used_block_id = null_hash;
         txd_p.first->second.max_used_block_height = 0;
         txd_p.first->second.kept_by_block = kept_by_block;
+        txd_p.first->second.receive_time = time(nullptr);
         tvc.m_verifivation_impossible = true;
         tvc.m_added_to_pool = true;
       }else
@@ -104,6 +105,7 @@ namespace cryptonote
       txd_p.first->second.max_used_block_height = max_used_block_height;
       txd_p.first->second.last_failed_height = 0;
       txd_p.first->second.last_failed_id = null_hash;
+      txd_p.first->second.receive_time = time(nullptr);
       tvc.m_added_to_pool = true;
 
       if(txd_p.first->second.fee > 0)
@@ -178,6 +180,30 @@ namespace cryptonote
     return true;
   }
   //---------------------------------------------------------------------------------
+  void tx_memory_pool::on_idle()
+  {
+    m_remove_stuck_tx_interval.do_call([this](){return remove_stuck_transactions();});
+  }
+  //---------------------------------------------------------------------------------
+  //proper tx_pool handling courtesy of CryptoZoidberg and Boolberry
+  bool tx_memory_pool::remove_stuck_transactions()
+  {
+    CRITICAL_REGION_LOCAL(m_transactions_lock);
+    for(auto it = m_transactions.begin(); it!= m_transactions.end();)
+    {
+      uint64_t tx_age = time(nullptr) - it->second.receive_time;
+
+      if((tx_age > CRYPTONOTE_MEMPOOL_TX_LIVETIME && !it->second.kept_by_block) || 
+         (tx_age > CRYPTONOTE_MEMPOOL_TX_FROM_ALT_BLOCK_LIVETIME && it->second.kept_by_block) )
+      {
+        LOG_PRINT_L0("Tx " << it->first << " removed from tx pool due to outdated, age: " << tx_age );
+        m_transactions.erase(it++);
+      }else
+        ++it;
+    }
+    return true;
+  }
+  //---------------------------------------------------------------------------------
   size_t tx_memory_pool::get_transactions_count()
   {
     CRITICAL_REGION_LOCAL(m_transactions_lock);

src/cryptonote_core/tx_pool.h

@@ -13,6 +13,7 @@
 
 #include "string_tools.h"
 #include "syncobj.h"
+#include "math_helper.h"
 #include "cryptonote_basic_impl.h"
 #include "verification_context.h"
 #include "crypto/hash.h"
@@ -40,6 +41,7 @@ namespace cryptonote
 
     bool on_blockchain_inc(uint64_t new_block_height, const crypto::hash& top_block_id);
     bool on_blockchain_dec(uint64_t new_block_height, const crypto::hash& top_block_id);
+    void on_idle();
 
     void lock();
     void unlock();
@@ -59,7 +61,7 @@ namespace cryptonote
     /*bool flush_pool(const std::strig& folder);
     bool inflate_pool(const std::strig& folder);*/
 
-#define CURRENT_MEMPOOL_ARCHIVE_VER    7
+#define CURRENT_MEMPOOL_ARCHIVE_VER    8
 
     template<class archive_t>
     void serialize(archive_t & a, const unsigned int version)
@@ -82,9 +84,11 @@ namespace cryptonote
       //
       uint64_t last_failed_height;
       crypto::hash last_failed_id;
+      time_t receive_time;
     };
 
   private:
+  	bool remove_stuck_transactions();
     bool is_transaction_ready_to_go(tx_details& txd);
     typedef std::unordered_map<crypto::hash, tx_details > transactions_container;
     typedef std::unordered_map<crypto::key_image, std::unordered_set<crypto::hash> > key_images_container;
@@ -92,6 +96,7 @@ namespace cryptonote
     epee::critical_section m_transactions_lock;
     transactions_container m_transactions;
     key_images_container m_spent_key_images;
+    epee::math_helper::once_a_time_seconds<30> m_remove_stuck_tx_interval;
 
     //transactions_container m_alternative_transactions;
 
@@ -159,6 +164,7 @@ namespace boost
       ar & td.max_used_block_id;
       ar & td.last_failed_height;
       ar & td.last_failed_id;
+      ar & td.receive_time;
 
     }
   }

src/rpc/core_rpc_server.cpp

@@ -438,9 +438,27 @@ namespace cryptonote
       error_resp.message = "Wrong block blob";
       return false;
     }
-    cryptonote::block_verification_context bvc = AUTO_VAL_INIT(bvc);
-    m_core.handle_incoming_block(blockblob, bvc);
-    if(!bvc.m_added_to_main_chain)
+    
+    // Fixing of high orphan issue for most pools
+    // Thanks Boolberry!
+    block b = AUTO_VAL_INIT(b);
+    if(!parse_and_validate_block_from_blob(blockblob, b))
+    {
+      error_resp.code = CORE_RPC_ERROR_CODE_WRONG_BLOCKBLOB;
+      error_resp.message = "Wrong block blob";
+      return false;
+    }
+
+    // Fix from Boolberry neglects to check block
+    // size, do that with the function below
+    if(!m_core.check_incoming_block_size(blockblob))
+    {
+      error_resp.code = CORE_RPC_ERROR_CODE_WRONG_BLOCKBLOB_SIZE;
+      error_resp.message = "Block bloc size is too big, rejecting block";
+      return false;
+    }
+
+    if(!m_core.handle_block_found(b))
     {
       error_resp.code = CORE_RPC_ERROR_CODE_BLOCK_NOT_ACCEPTED;
       error_resp.message = "Block not accepted";

src/rpc/core_rpc_server_error_codes.h

@@ -13,7 +13,6 @@
 #define CORE_RPC_ERROR_CODE_WRONG_BLOCKBLOB       -6
 #define CORE_RPC_ERROR_CODE_BLOCK_NOT_ACCEPTED    -7
 #define CORE_RPC_ERROR_CODE_CORE_BUSY             -9
-
-
+#define CORE_RPC_ERROR_CODE_WRONG_BLOCKBLOB_SIZE  -10
 
 

tests/CMakeLists.txt

@@ -31,16 +31,16 @@ add_executable(unit_tests ${UNIT_TESTS})
 add_executable(net_load_tests_clt net_load_tests/clt.cpp)
 add_executable(net_load_tests_srv net_load_tests/srv.cpp)
 
-target_link_libraries(core_proxy cryptonote_core common crypto upnpc-static ${Boost_LIBRARIES})
-target_link_libraries(coretests cryptonote_core common crypto ${Boost_LIBRARIES})
+target_link_libraries(core_proxy cryptonote_core common crypto upnpc-static ${CMAKE_THREAD_LIBS_INIT} ${Boost_LIBRARIES})
+target_link_libraries(coretests cryptonote_core common crypto ${CMAKE_THREAD_LIBS_INIT} ${Boost_LIBRARIES})
 target_link_libraries(difficulty-tests cryptonote_core)
-target_link_libraries(functional_tests cryptonote_core wallet common crypto ${Boost_LIBRARIES})
+target_link_libraries(functional_tests cryptonote_core wallet common crypto ${CMAKE_THREAD_LIBS_INIT} ${Boost_LIBRARIES})
 target_link_libraries(hash-tests crypto)
 target_link_libraries(hash-target-tests crypto cryptonote_core)
-target_link_libraries(performance_tests cryptonote_core common crypto ${Boost_LIBRARIES})
-target_link_libraries(unit_tests cryptonote_core common crypto gtest_main ${Boost_LIBRARIES})
-target_link_libraries(net_load_tests_clt cryptonote_core common crypto gtest_main ${Boost_LIBRARIES})
-target_link_libraries(net_load_tests_srv cryptonote_core common crypto gtest_main ${Boost_LIBRARIES})
+target_link_libraries(performance_tests cryptonote_core common crypto ${CMAKE_THREAD_LIBS_INIT} ${Boost_LIBRARIES})
+target_link_libraries(unit_tests cryptonote_core common crypto gtest_main ${CMAKE_THREAD_LIBS_INIT} ${Boost_LIBRARIES})
+target_link_libraries(net_load_tests_clt cryptonote_core common crypto gtest_main ${CMAKE_THREAD_LIBS_INIT} ${Boost_LIBRARIES})
+target_link_libraries(net_load_tests_srv cryptonote_core common crypto gtest_main ${CMAKE_THREAD_LIBS_INIT} ${Boost_LIBRARIES})
 
 if(NOT MSVC)
   set_property(TARGET gtest gtest_main unit_tests net_load_tests_clt net_load_tests_srv APPEND_STRING PROPERTY COMPILE_FLAGS " -Wno-undef -Wno-sign-compare")