Enable use of NaCl for faster X64 Salsa20 implementations. Also include binary for OSX for easy build. Blazingly fast.

6 files changed, 99 insertions, 41 deletions

diff --git a/node/Identity.cpp b/node/Identity.cpp
index 89fdb836..d1b21e9c 100644
--- a/node/Identity.cpp
+++ b/node/Identity.cpp
@@ -45,7 +45,7 @@ static inline void _computeMemoryHardHash(const void *publicKey,unsigned int pub
 	// ordinary Salsa20 is randomly seekable. This is good for a cipher
 	// but is not what we want for sequential memory-harndess.
 	memset(genmem,0,ZT_IDENTITY_GEN_MEMORY);
-	Salsa20 s20(digest,256,(char *)digest + 32);
+	Salsa20 s20(digest,(char *)digest + 32);
 	s20.crypt20((char *)genmem,(char *)genmem,64);
 	for(unsigned long i=64;i<ZT_IDENTITY_GEN_MEMORY;i+=64) {
 		unsigned long k = i - 64;
diff --git a/node/Node.cpp b/node/Node.cpp
index 1bc96cca..55fb4e72 100644
--- a/node/Node.cpp
+++ b/node/Node.cpp
@@ -66,9 +66,9 @@ Node::Node(void *uptr,void *tptr,const struct ZT_Node_Callbacks *callbacks,uint6
 	memset(_lastIdentityVerification,0,sizeof(_lastIdentityVerification));
 
 	// Use Salsa20 alone as a high-quality non-crypto PRNG
-	char foo[32];
-	Utils::getSecureRandom(foo,32);
-	_prng.init(foo,256,foo);
+	char foo[64];
+	Utils::getSecureRandom(foo,64);
+	_prng.init(foo,foo + 32);
 	memset(_prngStream,0,sizeof(_prngStream));
 	_prng.crypt12(_prngStream,_prngStream,sizeof(_prngStream));
 
diff --git a/node/Packet.cpp b/node/Packet.cpp
index 756f3140..31c46e82 100644
--- a/node/Packet.cpp
+++ b/node/Packet.cpp
@@ -1074,7 +1074,7 @@ void Packet::armor(const void *key,bool encryptPayload,unsigned int counter)
 	setCipher(encryptPayload ? ZT_PROTO_CIPHER_SUITE__C25519_POLY1305_SALSA2012 : ZT_PROTO_CIPHER_SUITE__C25519_POLY1305_NONE);
 
 	_salsa20MangleKey((const unsigned char *)key,mangledKey);
-	Salsa20 s20(mangledKey,256,data + ZT_PACKET_IDX_IV);
+	Salsa20 s20(mangledKey,data + ZT_PACKET_IDX_IV);
 
 	// MAC key is always the first 32 bytes of the Salsa20 key stream
 	// This is the same construction DJB's NaCl library uses
@@ -1098,7 +1098,7 @@ bool Packet::dearmor(const void *key)
 
 	if ((cs == ZT_PROTO_CIPHER_SUITE__C25519_POLY1305_NONE)||(cs == ZT_PROTO_CIPHER_SUITE__C25519_POLY1305_SALSA2012)) {
 		_salsa20MangleKey((const unsigned char *)key,mangledKey);
-		Salsa20 s20(mangledKey,256,data + ZT_PACKET_IDX_IV);
+		Salsa20 s20(mangledKey,data + ZT_PACKET_IDX_IV);
 
 		s20.crypt12(ZERO_KEY,macKey,sizeof(macKey));
 		Poly1305::compute(mac,payload,payloadLen,macKey);
@@ -1120,7 +1120,7 @@ void Packet::cryptField(const void *key,unsigned int start,unsigned int len)
 	uint8_t iv[8];
 	for(int i=0;i<8;++i) iv[i] = data[i];
 	iv[7] &= 0xf8; // mask off least significant 3 bits of packet ID / IV since this is unset when this function gets called
-	Salsa20 s20(key,256,iv);
+	Salsa20 s20(key,iv);
 	s20.crypt12(data + start,data + start,len);
 }
 
diff --git a/node/Salsa20.cpp b/node/Salsa20.cpp
index 1a4641f7..1e2b4b0f 100644
--- a/node/Salsa20.cpp
+++ b/node/Salsa20.cpp
@@ -10,6 +10,8 @@
 #include "Constants.hpp"
 #include "Salsa20.hpp"
 
+#ifndef ZT_USE_LIBSODIUM
+
 #define ROTATE(v,c) (((v) << (c)) | ((v) >> (32 - (c))))
 #define XOR(v,w) ((v) ^ (w))
 #define PLUS(v,w) ((uint32_t)((v) + (w)))
@@ -66,8 +68,7 @@ static const _s20sseconsts _S20SSECONSTANTS;
 
 namespace ZeroTier {
 
-void Salsa20::init(const void *key,unsigned int kbits,const void *iv)
-	throw()
+void Salsa20::init(const void *key,const void *iv)
 {
 #ifdef ZT_SALSA20_SSE
 	const uint32_t *k = (const uint32_t *)key;
@@ -78,14 +79,9 @@ void Salsa20::init(const void *key,unsigned int kbits,const void *iv)
 	_state.i[10] = k[1];
 	_state.i[7] = k[2];
 	_state.i[4] = k[3];
-	if (kbits == 256) {
-		k += 4;
-		_state.i[1] = 0x3320646e;
-		_state.i[2] = 0x79622d32;
-	} else {
-		_state.i[1] = 0x3120646e;
-		_state.i[2] = 0x79622d36;
-	}
+	k += 4;
+	_state.i[1] = 0x3320646e;
+	_state.i[2] = 0x79622d32;
 	_state.i[15] = k[0];
 	_state.i[12] = k[1];
 	_state.i[9] = k[2];
@@ -95,19 +91,14 @@ void Salsa20::init(const void *key,unsigned int kbits,const void *iv)
 	_state.i[5] = 0;
 	_state.i[8] = 0;
 #else
-	const char *constants;
+	const char *const constants = "expand 32-byte k";
 	const uint8_t *k = (const uint8_t *)key;
 
 	_state.i[1] = U8TO32_LITTLE(k + 0);
 	_state.i[2] = U8TO32_LITTLE(k + 4);
 	_state.i[3] = U8TO32_LITTLE(k + 8);
 	_state.i[4] = U8TO32_LITTLE(k + 12);
-	if (kbits == 256) { /* recommended */
-		k += 16;
-		constants = "expand 32-byte k";
-	} else { /* kbits == 128 */
-		constants = "expand 16-byte k";
-	}
+	k += 16;
 	_state.i[5] = U8TO32_LITTLE(constants + 4);
 	_state.i[6] = U8TO32_LITTLE(((const uint8_t *)iv) + 0);
 	_state.i[7] = U8TO32_LITTLE(((const uint8_t *)iv) + 4);
@@ -124,7 +115,6 @@ void Salsa20::init(const void *key,unsigned int kbits,const void *iv)
 }
 
 void Salsa20::crypt12(const void *in,void *out,unsigned int bytes)
-	throw()
 {
 	uint8_t tmp[64];
 	const uint8_t *m = (const uint8_t *)in;
@@ -624,7 +614,6 @@ void Salsa20::crypt12(const void *in,void *out,unsigned int bytes)
 }
 
 void Salsa20::crypt20(const void *in,void *out,unsigned int bytes)
-	throw()
 {
 	uint8_t tmp[64];
 	const uint8_t *m = (const uint8_t *)in;
@@ -1356,3 +1345,5 @@ void Salsa20::crypt20(const void *in,void *out,unsigned int bytes)
 }
 
 } // namespace ZeroTier
+
+#endif // !ZT_USE_LIBSODIUM
diff --git a/node/Salsa20.hpp b/node/Salsa20.hpp
index 6405d450..5e4c68be 100644
--- a/node/Salsa20.hpp
+++ b/node/Salsa20.hpp
@@ -10,10 +10,82 @@
 #include <stdio.h>
 #include <stdint.h>
 #include <stdlib.h>
+#include <string.h>
 
 #include "Constants.hpp"
 #include "Utils.hpp"
 
+#ifdef ZT_USE_LIBSODIUM
+
+#include <sodium/crypto_stream_salsa20.h>
+#include <sodium/crypto_stream_salsa2012.h>
+
+namespace ZeroTier {
+
+/**
+ * Salsa20 stream cipher
+ */
+class Salsa20
+{
+public:
+	Salsa20() {}
+	~Salsa20() { Utils::burn(_k,sizeof(_k)); }
+
+	/**
+	 * @param key 256-bit (32 byte) key
+	 * @param iv 64-bit initialization vector
+	 */
+	Salsa20(const void *key,const void *iv)
+	{
+		memcpy(_k,key,32);
+		memcpy(&_iv,iv,8);
+	}
+
+	/**
+	 * Initialize cipher
+	 *
+	 * @param key Key bits
+	 * @param iv 64-bit initialization vector
+	 */
+	inline void init(const void *key,const void *iv)
+	{
+		memcpy(_k,key,32);
+		memcpy(&_iv,iv,8);
+	}
+
+	/**
+	 * Encrypt/decrypt data using Salsa20/12
+	 *
+	 * @param in Input data
+	 * @param out Output buffer
+	 * @param bytes Length of data
+	 */
+	inline void crypt12(const void *in,void *out,unsigned int bytes)
+	{
+		crypto_stream_salsa2012_xor(reinterpret_cast<unsigned char *>(out),reinterpret_cast<const unsigned char *>(in),bytes,reinterpret_cast<const unsigned char *>(&_iv),reinterpret_cast<const unsigned char *>(_k));
+	}
+
+	/**
+	 * Encrypt/decrypt data using Salsa20/20
+	 *
+	 * @param in Input data
+	 * @param out Output buffer
+	 * @param bytes Length of data
+	 */
+	inline void crypt20(const void *in,void *out,unsigned int bytes)
+	{
+		crypto_stream_salsa20_xor(reinterpret_cast<unsigned char *>(out),reinterpret_cast<const unsigned char *>(in),bytes,reinterpret_cast<const unsigned char *>(&_iv),reinterpret_cast<const unsigned char *>(_k));
+	}
+
+private:
+	uint64_t _k[4];
+	uint64_t _iv;
+};
+
+} // namespace ZeroTier
+
+#else // !ZT_USE_LIBSODIUM
+
 #if (!defined(ZT_SALSA20_SSE)) && (defined(__SSE2__) || defined(__WINDOWS__))
 #define ZT_SALSA20_SSE 1
 #endif
@@ -30,30 +102,25 @@ namespace ZeroTier {
 class Salsa20
 {
 public:
-	Salsa20() throw() {}
-
+	Salsa20() {}
 	~Salsa20() { Utils::burn(&_state,sizeof(_state)); }
 
 	/**
-	 * @param key Key bits
-	 * @param kbits Number of key bits: 128 or 256 (recommended)
+	 * @param key 256-bit (32 byte) key
 	 * @param iv 64-bit initialization vector
 	 */
-	Salsa20(const void *key,unsigned int kbits,const void *iv)
-		throw()
+	Salsa20(const void *key,const void *iv)
 	{
-		init(key,kbits,iv);
+		init(key,iv);
 	}
 
 	/**
 	 * Initialize cipher
 	 *
 	 * @param key Key bits
-	 * @param kbits Number of key bits: 128 or 256 (recommended)
 	 * @param iv 64-bit initialization vector
 	 */
-	void init(const void *key,unsigned int kbits,const void *iv)
-		throw();
+	void init(const void *key,const void *iv);
 
 	/**
 	 * Encrypt/decrypt data using Salsa20/12
@@ -62,8 +129,7 @@ public:
 	 * @param out Output buffer
 	 * @param bytes Length of data
 	 */
-	void crypt12(const void *in,void *out,unsigned int bytes)
-		throw();
+	void crypt12(const void *in,void *out,unsigned int bytes);
 
 	/**
 	 * Encrypt/decrypt data using Salsa20/20
@@ -72,8 +138,7 @@ public:
 	 * @param out Output buffer
 	 * @param bytes Length of data
 	 */
-	void crypt20(const void *in,void *out,unsigned int bytes)
-		throw();
+	void crypt20(const void *in,void *out,unsigned int bytes);
 
 private:
 	union {
@@ -86,4 +151,6 @@ private:
 
 } // namespace ZeroTier
 
+#endif // ZT_USE_LIBSODIUM
+
 #endif
diff --git a/node/Utils.cpp b/node/Utils.cpp
index fb448dd6..92d14d19 100644
--- a/node/Utils.cpp
+++ b/node/Utils.cpp
@@ -156,7 +156,7 @@ void Utils::getSecureRandom(void *buf,unsigned int bytes)
 		s20Key[1] = (uint64_t)buf; // address of buf
 		s20Key[2] = (uint64_t)s20Key; // address of s20Key[]
 		s20Key[3] = (uint64_t)&s20; // address of s20
-		s20.init(s20Key,256,s20Key);
+		s20.init(s20Key,s20Key);
 	}
 
 #ifdef __WINDOWS__