bpo-33164: update blake2 implementation (GH-6286)

1 files changed, 101 insertions, 114 deletions

diff --git a/Modules/_blake2/impl/blake2s.c b/Modules/_blake2/impl/blake2s.c
index 69385dc..0c3636e 100644
--- a/Modules/_blake2/impl/blake2s.c
+++ b/Modules/_blake2/impl/blake2s.c
@@ -1,16 +1,14 @@
 /*
    BLAKE2 reference source code package - optimized C implementations
-  
-   Copyright 2012, Samuel Neves <sneves@dei.uc.pt>.  You may use this under the
-   terms of the CC0, the OpenSSL Licence, or the Apache Public License 2.0, at
-   your option.  The terms of these licenses can be found at:
-  
-   - CC0 1.0 Universal : http://creativecommons.org/publicdomain/zero/1.0
-   - OpenSSL license   : https://www.openssl.org/source/license.html
-   - Apache 2.0        : http://www.apache.org/licenses/LICENSE-2.0
-  
-   More information about the BLAKE2 hash function can be found at
-   https://blake2.net.
+
+   Written in 2012 by Samuel Neves <sneves@dei.uc.pt>
+
+   To the extent possible under law, the author(s) have dedicated all copyright
+   and related and neighboring rights to this software to the public domain
+   worldwide. This software is distributed without any warranty.
+
+   You should have received a copy of the CC0 Public Domain Dedication along with
+   this software. If not, see <http://creativecommons.org/publicdomain/zero/1.0/>.
 */
 
 #include <stdint.h>
@@ -22,18 +20,32 @@
 
 #include "blake2-config.h"
 
+#if defined(_MSC_VER)
+#include <intrin.h>
+#endif
 
+#if defined(HAVE_SSE2)
 #include <emmintrin.h>
+// MSVC only defines  _mm_set_epi64x for x86_64...
+#if defined(_MSC_VER) && !defined(_M_X64)
+static inline __m128i _mm_set_epi64x( const uint64_t u1, const uint64_t u0 )
+{
+  return _mm_set_epi32( u1 >> 32, u1, u0 >> 32, u0 );
+}
+#endif
+#endif
+
+
 #if defined(HAVE_SSSE3)
 #include <tmmintrin.h>
 #endif
-#if defined(HAVE_SSE41)
+#if defined(HAVE_SSE4_1)
 #include <smmintrin.h>
 #endif
 #if defined(HAVE_AVX)
 #include <immintrin.h>
 #endif
-#if defined(HAVE_XOP)
+#if defined(HAVE_XOP) && !defined(_MSC_VER)
 #include <x86intrin.h>
 #endif
 
@@ -61,40 +73,35 @@ static const uint8_t blake2s_sigma[10][16] =
 
 
 /* Some helper functions, not necessarily useful */
-BLAKE2_LOCAL_INLINE(int) blake2s_set_lastnode( blake2s_state *S )
+static inline int blake2s_set_lastnode( blake2s_state *S )
 {
-  S->f[1] = -1;
+  S->f[1] = ~0U;
   return 0;
 }
 
-BLAKE2_LOCAL_INLINE(int) blake2s_clear_lastnode( blake2s_state *S )
+static inline int blake2s_clear_lastnode( blake2s_state *S )
 {
-  S->f[1] = 0;
+  S->f[1] = 0U;
   return 0;
 }
 
-BLAKE2_LOCAL_INLINE(int) blake2s_is_lastblock( const blake2s_state *S )
-{
-  return S->f[0] != 0;
-}
-
-BLAKE2_LOCAL_INLINE(int) blake2s_set_lastblock( blake2s_state *S )
+static inline int blake2s_set_lastblock( blake2s_state *S )
 {
   if( S->last_node ) blake2s_set_lastnode( S );
 
-  S->f[0] = -1;
+  S->f[0] = ~0U;
   return 0;
 }
 
-BLAKE2_LOCAL_INLINE(int) blake2s_clear_lastblock( blake2s_state *S )
+static inline int blake2s_clear_lastblock( blake2s_state *S )
 {
   if( S->last_node ) blake2s_clear_lastnode( S );
 
-  S->f[0] = 0;
+  S->f[0] = 0U;
   return 0;
 }
 
-BLAKE2_LOCAL_INLINE(int) blake2s_increment_counter( blake2s_state *S, const uint32_t inc )
+static inline int blake2s_increment_counter( blake2s_state *S, const uint32_t inc )
 {
   uint64_t t = ( ( uint64_t )S->t[1] << 32 ) | S->t[0];
   t += inc;
@@ -104,91 +111,114 @@ BLAKE2_LOCAL_INLINE(int) blake2s_increment_counter( blake2s_state *S, const uint
 }
 
 
-/* Parameter-related functions */
-BLAKE2_LOCAL_INLINE(int) blake2s_param_set_digest_length( blake2s_param *P, const uint8_t digest_length )
+// Parameter-related functions
+static inline int blake2s_param_set_digest_length( blake2s_param *P, const uint8_t digest_length )
 {
   P->digest_length = digest_length;
   return 0;
 }
 
-BLAKE2_LOCAL_INLINE(int) blake2s_param_set_fanout( blake2s_param *P, const uint8_t fanout )
+static inline int blake2s_param_set_fanout( blake2s_param *P, const uint8_t fanout )
 {
   P->fanout = fanout;
   return 0;
 }
 
-BLAKE2_LOCAL_INLINE(int) blake2s_param_set_max_depth( blake2s_param *P, const uint8_t depth )
+static inline int blake2s_param_set_max_depth( blake2s_param *P, const uint8_t depth )
 {
   P->depth = depth;
   return 0;
 }
 
-BLAKE2_LOCAL_INLINE(int) blake2s_param_set_leaf_length( blake2s_param *P, const uint32_t leaf_length )
+static inline int blake2s_param_set_leaf_length( blake2s_param *P, const uint32_t leaf_length )
 {
   P->leaf_length = leaf_length;
   return 0;
 }
 
-BLAKE2_LOCAL_INLINE(int) blake2s_param_set_node_offset( blake2s_param *P, const uint64_t node_offset )
+static inline int blake2s_param_set_node_offset( blake2s_param *P, const uint64_t node_offset )
 {
   store48( P->node_offset, node_offset );
   return 0;
 }
 
-BLAKE2_LOCAL_INLINE(int) blake2s_param_set_node_depth( blake2s_param *P, const uint8_t node_depth )
+static inline int blake2s_param_set_node_depth( blake2s_param *P, const uint8_t node_depth )
 {
   P->node_depth = node_depth;
   return 0;
 }
 
-BLAKE2_LOCAL_INLINE(int) blake2s_param_set_inner_length( blake2s_param *P, const uint8_t inner_length )
+static inline int blake2s_param_set_inner_length( blake2s_param *P, const uint8_t inner_length )
 {
   P->inner_length = inner_length;
   return 0;
 }
 
-BLAKE2_LOCAL_INLINE(int) blake2s_param_set_salt( blake2s_param *P, const uint8_t salt[BLAKE2S_SALTBYTES] )
+static inline int blake2s_param_set_salt( blake2s_param *P, const uint8_t salt[BLAKE2S_SALTBYTES] )
 {
   memcpy( P->salt, salt, BLAKE2S_SALTBYTES );
   return 0;
 }
 
-BLAKE2_LOCAL_INLINE(int) blake2s_param_set_personal( blake2s_param *P, const uint8_t personal[BLAKE2S_PERSONALBYTES] )
+static inline int blake2s_param_set_personal( blake2s_param *P, const uint8_t personal[BLAKE2S_PERSONALBYTES] )
 {
   memcpy( P->personal, personal, BLAKE2S_PERSONALBYTES );
   return 0;
 }
 
-BLAKE2_LOCAL_INLINE(int) blake2s_init0( blake2s_state *S )
+static inline int blake2s_init0( blake2s_state *S )
 {
-  int i;
   memset( S, 0, sizeof( blake2s_state ) );
 
-  for( i = 0; i < 8; ++i ) S->h[i] = blake2s_IV[i];
+  for( int i = 0; i < 8; ++i ) S->h[i] = blake2s_IV[i];
 
   return 0;
 }
 
+#define blake2s_init BLAKE2_IMPL_NAME(blake2s_init)
+#define blake2s_init_param BLAKE2_IMPL_NAME(blake2s_init_param)
+#define blake2s_init_key BLAKE2_IMPL_NAME(blake2s_init_key)
+#define blake2s_update BLAKE2_IMPL_NAME(blake2s_update)
+#define blake2s_final BLAKE2_IMPL_NAME(blake2s_final)
+#define blake2s BLAKE2_IMPL_NAME(blake2s)
+
+#if defined(__cplusplus)
+extern "C" {
+#endif
+  int blake2s_init( blake2s_state *S, size_t outlen );
+  int blake2s_init_param( blake2s_state *S, const blake2s_param *P );
+  int blake2s_init_key( blake2s_state *S, size_t outlen, const void *key, size_t keylen );
+  int blake2s_update( blake2s_state *S, const uint8_t *in, size_t inlen );
+  int blake2s_final( blake2s_state *S, uint8_t *out, size_t outlen );
+  int blake2s( uint8_t *out, const void *in, const void *key, size_t outlen, size_t inlen, size_t keylen );
+#if defined(__cplusplus)
+}
+#endif
+
+
 /* init2 xors IV with input parameter block */
 int blake2s_init_param( blake2s_state *S, const blake2s_param *P )
 {
-  /*blake2s_init0( S ); */
-  const uint8_t * v = ( const uint8_t * )( blake2s_IV );
-  const uint8_t * p = ( const uint8_t * )( P );
-  uint8_t * h = ( uint8_t * )( S->h );
-  int i;
+  uint8_t *p, *h, *v;
+  //blake2s_init0( S );
+  v = ( uint8_t * )( blake2s_IV );
+  h = ( uint8_t * )( S->h );
+  p = ( uint8_t * )( P );
   /* IV XOR ParamBlock */
   memset( S, 0, sizeof( blake2s_state ) );
 
-  for( i = 0; i < BLAKE2S_OUTBYTES; ++i ) h[i] = v[i] ^ p[i];
+  for( int i = 0; i < BLAKE2S_OUTBYTES; ++i ) h[i] = v[i] ^ p[i];
 
+  S->outlen = P->digest_length;
   return 0;
 }
 
 
 /* Some sort of default parameter block initialization, for sequential blake2s */
-int blake2s_init( blake2s_state *S, const uint8_t outlen )
+int blake2s_init( blake2s_state *S, size_t outlen )
 {
+  if ( ( !outlen ) || ( outlen > BLAKE2S_OUTBYTES ) ) return -1;
+
   const blake2s_param P =
   {
     outlen,
@@ -202,14 +232,16 @@ int blake2s_init( blake2s_state *S, const uint8_t outlen )
     {0},
     {0}
   };
-  /* Move interval verification here? */
-  if ( ( !outlen ) || ( outlen > BLAKE2S_OUTBYTES ) ) return -1;
   return blake2s_init_param( S, &P );
 }
 
 
-int blake2s_init_key( blake2s_state *S, const uint8_t outlen, const void *key, const uint8_t keylen )
+int blake2s_init_key( blake2s_state *S, size_t outlen, const void *key, size_t keylen )
 {
+  if ( ( !outlen ) || ( outlen > BLAKE2S_OUTBYTES ) ) return -1;
+
+  if ( ( !key ) || ( !keylen ) || keylen > BLAKE2S_KEYBYTES ) return -1;
+
   const blake2s_param P =
   {
     outlen,
@@ -224,11 +256,6 @@ int blake2s_init_key( blake2s_state *S, const uint8_t outlen, const void *key, c
     {0}
   };
 
-  /* Move interval verification here? */
-  if ( ( !outlen ) || ( outlen > BLAKE2S_OUTBYTES ) ) return -1;
-
-  if ( ( !key ) || ( !keylen ) || keylen > BLAKE2S_KEYBYTES ) return -1;
-
   if( blake2s_init_param( S, &P ) < 0 )
     return -1;
 
@@ -243,11 +270,11 @@ int blake2s_init_key( blake2s_state *S, const uint8_t outlen, const void *key, c
 }
 
 
-BLAKE2_LOCAL_INLINE(int) blake2s_compress( blake2s_state *S, const uint8_t block[BLAKE2S_BLOCKBYTES] )
+static inline int blake2s_compress( blake2s_state *S, const uint8_t block[BLAKE2S_BLOCKBYTES] )
 {
   __m128i row1, row2, row3, row4;
   __m128i buf1, buf2, buf3, buf4;
-#if defined(HAVE_SSE41)
+#if defined(HAVE_SSE4_1)
   __m128i t0, t1;
 #if !defined(HAVE_XOP)
   __m128i t2;
@@ -258,7 +285,7 @@ BLAKE2_LOCAL_INLINE(int) blake2s_compress( blake2s_state *S, const uint8_t block
   const __m128i r8 = _mm_set_epi8( 12, 15, 14, 13, 8, 11, 10, 9, 4, 7, 6, 5, 0, 3, 2, 1 );
   const __m128i r16 = _mm_set_epi8( 13, 12, 15, 14, 9, 8, 11, 10, 5, 4, 7, 6, 1, 0, 3, 2 );
 #endif
-#if defined(HAVE_SSE41)
+#if defined(HAVE_SSE4_1)
   const __m128i m0 = LOADU( block +  00 );
   const __m128i m1 = LOADU( block +  16 );
   const __m128i m2 = LOADU( block +  32 );
@@ -300,8 +327,8 @@ BLAKE2_LOCAL_INLINE(int) blake2s_compress( blake2s_state *S, const uint8_t block
   return 0;
 }
 
-/* inlen now in bytes */
-int blake2s_update( blake2s_state *S, const uint8_t *in, uint64_t inlen )
+
+int blake2s_update( blake2s_state *S, const uint8_t *in, size_t inlen )
 {
   while( inlen > 0 )
   {
@@ -310,11 +337,11 @@ int blake2s_update( blake2s_state *S, const uint8_t *in, uint64_t inlen )
 
     if( inlen > fill )
     {
-      memcpy( S->buf + left, in, fill ); /* Fill buffer */
+      memcpy( S->buf + left, in, fill ); // Fill buffer
       S->buflen += fill;
       blake2s_increment_counter( S, BLAKE2S_BLOCKBYTES );
-      blake2s_compress( S, S->buf ); /* Compress */
-      memcpy( S->buf, S->buf + BLAKE2S_BLOCKBYTES, BLAKE2S_BLOCKBYTES ); /* Shift buffer left */
+      blake2s_compress( S, S->buf ); // Compress
+      memcpy( S->buf, S->buf + BLAKE2S_BLOCKBYTES, BLAKE2S_BLOCKBYTES ); // Shift buffer left
       S->buflen -= BLAKE2S_BLOCKBYTES;
       in += fill;
       inlen -= fill;
@@ -322,7 +349,7 @@ int blake2s_update( blake2s_state *S, const uint8_t *in, uint64_t inlen )
     else /* inlen <= fill */
     {
       memcpy( S->buf + left, in, inlen );
-      S->buflen += inlen; /* Be lazy, do not compress */
+      S->buflen += inlen; // Be lazy, do not compress
       in += inlen;
       inlen -= inlen;
     }
@@ -331,24 +358,19 @@ int blake2s_update( blake2s_state *S, const uint8_t *in, uint64_t inlen )
   return 0;
 }
 
-/* Is this correct? */
-int blake2s_final( blake2s_state *S, uint8_t *out, uint8_t outlen )
-{
-  uint8_t buffer[BLAKE2S_OUTBYTES] = {0};
-  int i;
 
-  if( outlen > BLAKE2S_OUTBYTES )
-    return -1;
+int blake2s_final( blake2s_state *S, uint8_t *out, size_t outlen )
+{
+  uint8_t buffer[BLAKE2S_OUTBYTES];
 
-  if( blake2s_is_lastblock( S ) )
-    return -1;
+  if(outlen != S->outlen ) return -1;
 
   if( S->buflen > BLAKE2S_BLOCKBYTES )
   {
     blake2s_increment_counter( S, BLAKE2S_BLOCKBYTES );
     blake2s_compress( S, S->buf );
     S->buflen -= BLAKE2S_BLOCKBYTES;
-    memmove( S->buf, S->buf + BLAKE2S_BLOCKBYTES, S->buflen );
+    memcpy( S->buf, S->buf + BLAKE2S_BLOCKBYTES, S->buflen );
   }
 
   blake2s_increment_counter( S, ( uint32_t )S->buflen );
@@ -356,15 +378,14 @@ int blake2s_final( blake2s_state *S, uint8_t *out, uint8_t outlen )
   memset( S->buf + S->buflen, 0, 2 * BLAKE2S_BLOCKBYTES - S->buflen ); /* Padding */
   blake2s_compress( S, S->buf );
 
-  for( i = 0; i < 8; ++i ) /* Output full hash to temp buffer */
+  for( int i = 0; i < 8; ++i ) /* Output full hash to temp buffer */
     store32( buffer + sizeof( S->h[i] ) * i, S->h[i] );
 
   memcpy( out, buffer, outlen );
   return 0;
 }
 
-/* inlen, at least, should be uint64_t. Others can be size_t. */
-int blake2s( uint8_t *out, const void *in, const void *key, const uint8_t outlen, const uint64_t inlen, uint8_t keylen )
+int blake2s( uint8_t *out, const void *in, const void *key, size_t outlen, size_t inlen, size_t keylen )
 {
   blake2s_state S[1];
 
@@ -388,48 +409,14 @@ int blake2s( uint8_t *out, const void *in, const void *key, const uint8_t outlen
     if( blake2s_init( S, outlen ) < 0 ) return -1;
   }
 
-  blake2s_update( S, ( const uint8_t * )in, inlen );
-  blake2s_final( S, out, outlen );
-  return 0;
+  if( blake2s_update( S, ( uint8_t * )in, inlen ) < 0) return -1;
+  return blake2s_final( S, out, outlen );
 }
 
 #if defined(SUPERCOP)
 int crypto_hash( unsigned char *out, unsigned char *in, unsigned long long inlen )
 {
-  return blake2s( out, in, NULL, BLAKE2S_OUTBYTES, inlen, 0 );
-}
-#endif
-
-#if defined(BLAKE2S_SELFTEST)
-#include <string.h>
-#include "blake2-kat.h"
-int main( int argc, char **argv )
-{
-  uint8_t key[BLAKE2S_KEYBYTES];
-  uint8_t buf[KAT_LENGTH];
-  size_t i;
-
-  for( i = 0; i < BLAKE2S_KEYBYTES; ++i )
-    key[i] = ( uint8_t )i;
-
-  for( i = 0; i < KAT_LENGTH; ++i )
-    buf[i] = ( uint8_t )i;
-
-  for( i = 0; i < KAT_LENGTH; ++i )
-  {
-    uint8_t hash[BLAKE2S_OUTBYTES];
-
-    if( blake2s( hash, buf, key, BLAKE2S_OUTBYTES, i, BLAKE2S_KEYBYTES ) < 0 ||
-        0 != memcmp( hash, blake2s_keyed_kat[i], BLAKE2S_OUTBYTES ) )
-    {
-      puts( "error" );
-      return -1;
-    }
-  }
-
-  puts( "ok" );
-  return 0;
+  return blake2s( out, in, NULL, BLAKE2S_OUTBYTES, (size_t)inlen, 0 );
 }
 #endif
 
-