From: Guus Sliepen <guus@tinc-vpn.org>
Date: Sat, 12 Oct 2013 23:02:52 +0000 (+0200)
Subject: Use AES-256-GCM for the SPTPS protocol.
X-Git-Tag: release-1.1pre10~47
X-Git-Url: https://tinc-vpn.org/git/browse?a=commitdiff_plain;h=0da07280882253b792ddf9c6bd8b6690ba585b7a;p=tinc

Use AES-256-GCM for the SPTPS protocol.

It is faster than AES-256-CTR + HMAC-SHA256, especially on Intel chips with AES
and PCLMULQDQ instructions.
---

diff --git a/src/cipher.h b/src/cipher.h
index 17ca6148..bbba5690 100644
--- a/src/cipher.h
+++ b/src/cipher.h
@@ -38,7 +38,12 @@ extern bool cipher_set_counter(cipher_t *, const void *, size_t) __attribute__ (
 extern bool cipher_set_counter_key(cipher_t *, void *) __attribute__ ((__warn_unused_result__));
 extern bool cipher_encrypt(cipher_t *, const void *indata, size_t inlen, void *outdata, size_t *outlen, bool oneshot) __attribute__ ((__warn_unused_result__));
 extern bool cipher_decrypt(cipher_t *, const void *indata, size_t inlen, void *outdata, size_t *outlen, bool oneshot) __attribute__ ((__warn_unused_result__));
-extern bool cipher_counter_xor(cipher_t *, const void *indata, size_t inlen, void *outdata) __attribute__ ((__warn_unused_result__));
+extern bool cipher_gcm_encrypt(cipher_t *, const void *indata, size_t inlen, void *outdata, size_t *outlen) __attribute__ ((__warn_unused_result__));
+extern bool cipher_gcm_encrypt_start(cipher_t *, const void *indata, size_t inlen, void *outdata, size_t *outlen) __attribute__ ((__warn_unused_result__));
+extern bool cipher_gcm_encrypt_finish(cipher_t *, const void *indata, size_t inlen, void *outdata, size_t *outlen) __attribute__ ((__warn_unused_result__));
+extern bool cipher_gcm_decrypt(cipher_t *, const void *indata, size_t inlen, void *outdata, size_t *outlen) __attribute__ ((__warn_unused_result__));
+extern bool cipher_gcm_decrypt_start(cipher_t *, const void *indata, size_t inlen, void *outdata, size_t *outlen) __attribute__ ((__warn_unused_result__));
+extern bool cipher_gcm_decrypt_finish(cipher_t *, const void *indata, size_t inlen, void *outdata, size_t *outlen) __attribute__ ((__warn_unused_result__));
 extern int cipher_get_nid(const cipher_t *);
 extern bool cipher_active(const cipher_t *);
 
diff --git a/src/openssl/cipher.c b/src/openssl/cipher.c
index 5d9bebcd..b01f5267 100644
--- a/src/openssl/cipher.c
+++ b/src/openssl/cipher.c
@@ -84,7 +84,7 @@ size_t cipher_keylength(const cipher_t *cipher) {
 	if(!cipher || !cipher->cipher)
 		return 0;
 
-	return cipher->cipher->key_len + cipher->cipher->block_size;
+	return cipher->cipher->key_len + cipher->cipher->iv_len;
 }
 
 bool cipher_set_key(cipher_t *cipher, void *key, bool encrypt) {
@@ -118,13 +118,12 @@ bool cipher_set_key_from_rsa(cipher_t *cipher, void *key, size_t len, bool encry
 }
 
 bool cipher_set_counter(cipher_t *cipher, const void *counter, size_t len) {
-	if(len > cipher->cipher->block_size - 4) {
+	if(len > cipher->cipher->iv_len - 4) {
 		logger(DEBUG_ALWAYS, LOG_ERR, "Counter too long");
-		abort();
+		return false;
 	}
 
-	memcpy(cipher->counter->counter + cipher->cipher->block_size - len, counter, len);
-	memset(cipher->counter->counter, 0, 4);
+	memcpy(cipher->counter->counter, counter, len);
 	cipher->counter->n = 0;
 
 	return true;
@@ -142,41 +141,96 @@ bool cipher_set_counter_key(cipher_t *cipher, void *key) {
 	else
 		cipher->counter->n = 0;
 
-	memcpy(cipher->counter->counter, (unsigned char *)key + cipher->cipher->key_len, cipher->cipher->block_size);
+	memcpy(cipher->counter->counter, (unsigned char *)key + cipher->cipher->key_len, cipher->cipher->iv_len);
 
 	return true;
 }
 
-bool cipher_counter_xor(cipher_t *cipher, const void *indata, size_t inlen, void *outdata) {
-	if(!cipher->counter) {
-		logger(DEBUG_ALWAYS, LOG_ERR, "Counter not initialized");
+bool cipher_gcm_encrypt_start(cipher_t *cipher, const void *indata, size_t inlen, void *outdata, size_t *outlen) {
+	int len = 0;
+	if(!EVP_EncryptInit_ex(&cipher->ctx, NULL, NULL, NULL, cipher->counter->counter)
+			|| (inlen && !EVP_EncryptUpdate(&cipher->ctx, (unsigned char *)outdata, &len, (unsigned char *)indata, inlen))) {
+		logger(DEBUG_ALWAYS, LOG_ERR, "Error while encrypting: %s", ERR_error_string(ERR_get_error(), NULL));
 		return false;
 	}
+	if(outlen)
+		*outlen = len;
+	return true;
+}
 
-	const unsigned char *in = indata;
-	unsigned char *out = outdata;
-
-	while(inlen--) {
-		// Encrypt the new counter value if we need it
-		if(!cipher->counter->n) {
-			int len;
-			if(!EVP_EncryptUpdate(&cipher->ctx, cipher->counter->block, &len, cipher->counter->counter, cipher->cipher->block_size)) {
-				logger(DEBUG_ALWAYS, LOG_ERR, "Error while encrypting: %s", ERR_error_string(ERR_get_error(), NULL));
-				return false;
-			}
-
-			// Increase the counter value
-			for(int i = 0; i < cipher->cipher->block_size; i++)
-				if(++cipher->counter->counter[i])
-					break;
-		}
+bool cipher_gcm_encrypt_finish(cipher_t *cipher, const void *indata, size_t inlen, void *outdata, size_t *outlen) {
+	int len = 0, pad = 0;
+	if(!(inlen && EVP_EncryptUpdate(&cipher->ctx, (unsigned char *)outdata, &len, (unsigned char *)indata, inlen))
+			|| !EVP_EncryptFinal(&cipher->ctx, (unsigned char *)outdata + len, &pad)) {
+		logger(DEBUG_ALWAYS, LOG_ERR, "Error while encrypting: %s", ERR_error_string(ERR_get_error(), NULL));
+		return false;
+	}
+	EVP_CIPHER_CTX_ctrl(&cipher->ctx, EVP_CTRL_GCM_GET_TAG, 16, (unsigned char *)outdata + len + pad);
+	if(outlen)
+		*outlen = len + pad + 16;
+	return true;
+}
 
-		*out++ = *in++ ^ cipher->counter->block[cipher->counter->n++];
+bool cipher_gcm_encrypt(cipher_t *cipher, const void *indata, size_t inlen, void *outdata, size_t *outlen) {
+	int len = 0, pad = 0;
+	if(!EVP_EncryptInit_ex(&cipher->ctx, NULL, NULL, NULL, cipher->counter->counter) ||
+			!EVP_EncryptUpdate(&cipher->ctx, (unsigned char *)outdata, &len, (unsigned char *)indata, inlen) ||
+			!EVP_EncryptFinal(&cipher->ctx, (unsigned char *)outdata + len, &pad)) {
+		logger(DEBUG_ALWAYS, LOG_ERR, "Error while encrypting: %s", ERR_error_string(ERR_get_error(), NULL));
+		return false;
+	}
+	EVP_CIPHER_CTX_ctrl(&cipher->ctx, EVP_CTRL_GCM_GET_TAG, 16, (unsigned char *)outdata + len + pad);
+	if(outlen)
+		*outlen = len + pad + 16;
+	return true;
+}
 
-		if(cipher->counter->n >= cipher->cipher->block_size)
-			cipher->counter->n = 0;
+bool cipher_gcm_decrypt(cipher_t *cipher, const void *indata, size_t inlen, void *outdata, size_t *outlen) {
+	if(inlen < 16)
+		return false;
+
+	int len = 0, pad = 0;
+	if(!EVP_DecryptInit_ex(&cipher->ctx, NULL, NULL, NULL, cipher->counter->counter)) {
+		logger(DEBUG_ALWAYS, LOG_ERR, "Error while decrypting: %s", ERR_error_string(ERR_get_error(), NULL));
+		return false;
 	}
 
+	EVP_CIPHER_CTX_ctrl(&cipher->ctx, EVP_CTRL_GCM_SET_TAG, 16, (unsigned char *)indata + inlen - 16);
+
+	if(!EVP_DecryptUpdate(&cipher->ctx, (unsigned char *)outdata, &len, (unsigned char *)indata, inlen - 16) ||
+			!EVP_DecryptFinal(&cipher->ctx, (unsigned char *)outdata + len, &pad)) {
+		logger(DEBUG_ALWAYS, LOG_ERR, "Error while decrypting: %s", ERR_error_string(ERR_get_error(), NULL));
+		return false;
+	}
+	if(outlen)
+		*outlen = len;
+	return true;
+}
+
+bool cipher_gcm_decrypt_start(cipher_t *cipher, const void *indata, size_t inlen, void *outdata, size_t *outlen) {
+	int len = 0;
+	if(!EVP_DecryptInit_ex(&cipher->ctx, NULL, NULL, NULL, cipher->counter->counter)
+			|| (inlen && !EVP_DecryptUpdate(&cipher->ctx, (unsigned char *)outdata, &len, (unsigned char *)indata, inlen))) {
+		logger(DEBUG_ALWAYS, LOG_ERR, "Error while decrypting: %s", ERR_error_string(ERR_get_error(), NULL));
+		return false;
+	}
+	if(outlen)
+		*outlen = len;
+	return true;
+}
+
+bool cipher_gcm_decrypt_finish(cipher_t *cipher, const void *indata, size_t inlen, void *outdata, size_t *outlen) {
+	if(inlen < 16)
+		return false;
+
+	EVP_CIPHER_CTX_ctrl(&cipher->ctx, EVP_CTRL_GCM_SET_TAG, 16, (unsigned char *)indata + inlen - 16);
+
+	int len = 0, pad = 0;
+	if((inlen > 16 && !EVP_DecryptUpdate(&cipher->ctx, (unsigned char *)outdata, &len, (unsigned char *)indata, inlen - 16))
+			|| !EVP_DecryptFinal(&cipher->ctx, (unsigned char *)outdata + len, &pad)) {
+		logger(DEBUG_ALWAYS, LOG_ERR, "Error while decrypting: %s", ERR_error_string(ERR_get_error(), NULL));
+		return false;
+	}
 	return true;
 }
 
diff --git a/src/sptps.c b/src/sptps.c
index 62cfb1f9..0369d283 100644
--- a/src/sptps.c
+++ b/src/sptps.c
@@ -22,7 +22,6 @@
 
 #include "cipher.h"
 #include "crypto.h"
-#include "digest.h"
 #include "ecdh.h"
 #include "ecdsa.h"
 #include "logger.h"
@@ -83,34 +82,30 @@ static void warning(sptps_t *s, const char *format, ...) {
 
 // Send a record (datagram version, accepts all record types, handles encryption and authentication).
 static bool send_record_priv_datagram(sptps_t *s, uint8_t type, const char *data, uint16_t len) {
-	char buffer[len + 23UL];
+	char buffer[len + 21UL];
 
 	// Create header with sequence number, length and record type
 	uint32_t seqno = htonl(s->outseqno++);
-	uint16_t netlen = htons(len);
-
-	memcpy(buffer, &netlen, 2);
-	memcpy(buffer + 2, &seqno, 4);
-	buffer[6] = type;
 
-	// Add plaintext (TODO: avoid unnecessary copy)
-	memcpy(buffer + 7, data, len);
+	memcpy(buffer, &seqno, 4);
+	buffer[4] = type;
 
 	if(s->outstate) {
 		// If first handshake has finished, encrypt and HMAC
 		if(!cipher_set_counter(s->outcipher, &seqno, sizeof seqno))
-			return false;
+			return error(s, EINVAL, "Failed to set counter");
 
-		if(!cipher_counter_xor(s->outcipher, buffer + 6, len + 1UL, buffer + 6))
-			return false;
+		if(!cipher_gcm_encrypt_start(s->outcipher, buffer + 4, 1, buffer + 4, NULL))
+			return error(s, EINVAL, "Error encrypting record");
 
-		if(!digest_create(s->outdigest, buffer, len + 7UL, buffer + 7UL + len))
-			return false;
+		if(!cipher_gcm_encrypt_finish(s->outcipher, data, len, buffer + 5, NULL))
+			return error(s, EINVAL, "Error encrypting record");
 
-		return s->send_data(s->handle, type, buffer + 2, len + 21UL);
+		return s->send_data(s->handle, type, buffer, len + 21UL);
 	} else {
 		// Otherwise send as plaintext
-		return s->send_data(s->handle, type, buffer + 2, len + 5UL);
+		memcpy(buffer + 5, data, len);
+		return s->send_data(s->handle, type, buffer, len + 5UL);
 	}
 }
 // Send a record (private version, accepts all record types, handles encryption and authentication).
@@ -118,31 +113,31 @@ static bool send_record_priv(sptps_t *s, uint8_t type, const char *data, uint16_
 	if(s->datagram)
 		return send_record_priv_datagram(s, type, data, len);
 
-	char buffer[len + 23UL];
+	char buffer[len + 19UL];
 
 	// Create header with sequence number, length and record type
 	uint32_t seqno = htonl(s->outseqno++);
 	uint16_t netlen = htons(len);
 
-	memcpy(buffer, &seqno, 4);
-	memcpy(buffer + 4, &netlen, 2);
-	buffer[6] = type;
-
-	// Add plaintext (TODO: avoid unnecessary copy)
-	memcpy(buffer + 7, data, len);
+	memcpy(buffer, &netlen, 2);
+	buffer[2] = type;
 
 	if(s->outstate) {
 		// If first handshake has finished, encrypt and HMAC
-		if(!cipher_counter_xor(s->outcipher, buffer + 4, len + 3UL, buffer + 4))
-			return false;
+		if(!cipher_set_counter(s->outcipher, &seqno, 4))
+			return error(s, EINVAL, "Failed to set counter");
+
+		if(!cipher_gcm_encrypt_start(s->outcipher, buffer, 3, buffer, NULL))
+			return error(s, EINVAL, "Error encrypting record");
 
-		if(!digest_create(s->outdigest, buffer, len + 7UL, buffer + 7UL + len))
-			return false;
+		if(!cipher_gcm_encrypt_finish(s->outcipher, data, len, buffer + 3, NULL))
+			return error(s, EINVAL, "Error encrypting record");
 
-		return s->send_data(s->handle, type, buffer + 4, len + 19UL);
+		return s->send_data(s->handle, type, buffer, len + 19UL);
 	} else {
 		// Otherwise send as plaintext
-		return s->send_data(s->handle, type, buffer + 4, len + 3UL);
+		memcpy(buffer + 3, data, len);
+		return s->send_data(s->handle, type, buffer, len + 3UL);
 	}
 }
 
@@ -165,7 +160,7 @@ static bool send_kex(sptps_t *s) {
 
 	// Make room for our KEX message, which we will keep around since send_sig() needs it.
 	if(s->mykex)
-		abort();
+		return false;
 	s->mykex = realloc(s->mykex, 1 + 32 + keylen);
 	if(!s->mykex)
 		return error(s, errno, strerror(errno));
@@ -178,7 +173,7 @@ static bool send_kex(sptps_t *s) {
 
 	// Create a new ECDH public key.
 	if(!(s->ecdh = ecdh_generate_public(s->mykex + 1 + 32)))
-		return false;
+		return error(s, EINVAL, "Failed to generate ECDH public key");
 
 	return send_record_priv(s, SPTPS_HANDSHAKE, s->mykex, 1 + 32 + keylen);
 }
@@ -199,7 +194,7 @@ static bool send_sig(sptps_t *s) {
 
 	// Sign the result.
 	if(!ecdsa_sign(s->mykey, msg, sizeof msg, sig))
-		return false;
+		return error(s, EINVAL, "Failed to sign SIG record");
 
 	// Send the SIG exchange record.
 	return send_record_priv(s, SPTPS_HANDSHAKE, sig, sizeof sig);
@@ -209,16 +204,14 @@ static bool send_sig(sptps_t *s) {
 static bool generate_key_material(sptps_t *s, const char *shared, size_t len) {
 	// Initialise cipher and digest structures if necessary
 	if(!s->outstate) {
-		s->incipher = cipher_open_by_name("aes-256-ecb");
-		s->outcipher = cipher_open_by_name("aes-256-ecb");
-		s->indigest = digest_open_by_name("sha256", 16);
-		s->outdigest = digest_open_by_name("sha256", 16);
-		if(!s->incipher || !s->outcipher || !s->indigest || !s->outdigest)
-			return false;
+		s->incipher = cipher_open_by_name("aes-256-gcm");
+		s->outcipher = cipher_open_by_name("aes-256-gcm");
+		if(!s->incipher || !s->outcipher)
+			return error(s, EINVAL, "Failed to open cipher");
 	}
 
 	// Allocate memory for key material
-	size_t keylen = digest_keylength(s->indigest) + digest_keylength(s->outdigest) + cipher_keylength(s->incipher) + cipher_keylength(s->outcipher);
+	size_t keylen = cipher_keylength(s->incipher) + cipher_keylength(s->outcipher);
 
 	s->key = realloc(s->key, keylen);
 	if(!s->key)
@@ -238,7 +231,7 @@ static bool generate_key_material(sptps_t *s, const char *shared, size_t len) {
 
 	// Use PRF to generate the key material
 	if(!prf(shared, len, seed, s->labellen + 64 + 13, s->key, keylen))
-		return false;
+		return error(s, EINVAL, "Failed to generate key material");
 
 	return true;
 }
@@ -254,17 +247,11 @@ static bool receive_ack(sptps_t *s, const char *data, uint16_t len) {
 		return error(s, EIO, "Invalid ACK record length");
 
 	if(s->initiator) {
-		bool result
-			= cipher_set_counter_key(s->incipher, s->key)
-			&& digest_set_key(s->indigest, s->key + cipher_keylength(s->incipher), digest_keylength(s->indigest));
-		if(!result)
-			return false;
+		if(!cipher_set_counter_key(s->incipher, s->key))
+			return error(s, EINVAL, "Failed to set counter");
 	} else {
-		bool result
-			= cipher_set_counter_key(s->incipher, s->key + cipher_keylength(s->outcipher) + digest_keylength(s->outdigest))
-			&& digest_set_key(s->indigest, s->key + cipher_keylength(s->outcipher) + digest_keylength(s->outdigest) + cipher_keylength(s->incipher), digest_keylength(s->indigest));
-		if(!result)
-			return false;
+		if(!cipher_set_counter_key(s->incipher, s->key + cipher_keylength(s->outcipher)))
+			return error(s, EINVAL, "Failed to set counter");
 	}
 
 	free(s->key);
@@ -284,7 +271,7 @@ static bool receive_kex(sptps_t *s, const char *data, uint16_t len) {
 
 	// Make a copy of the KEX message, send_sig() and receive_sig() need it
 	if(s->hiskex)
-		abort();
+		return error(s, EINVAL, "Received a second KEX message before first has been processed");
 	s->hiskex = realloc(s->hiskex, len);
 	if(!s->hiskex)
 		return error(s, errno, strerror(errno));
@@ -313,12 +300,12 @@ static bool receive_sig(sptps_t *s, const char *data, uint16_t len) {
 
 	// Verify signature.
 	if(!ecdsa_verify(s->hiskey, msg, sizeof msg, data))
-		return false;
+		return error(s, EIO, "Failed to verify SIG record");
 
 	// Compute shared secret.
 	char shared[ECDH_SHARED_SIZE];
 	if(!ecdh_compute_shared(s->ecdh, s->hiskex + 1 + 32, shared))
-		return false;
+		return error(s, EINVAL, "Failed to compute ECDH shared secret");
 	s->ecdh = NULL;
 
 	// Generate key material from shared secret.
@@ -337,17 +324,11 @@ static bool receive_sig(sptps_t *s, const char *data, uint16_t len) {
 
 	// TODO: only set new keys after ACK has been set/received
 	if(s->initiator) {
-		bool result
-			= cipher_set_counter_key(s->outcipher, s->key + cipher_keylength(s->incipher) + digest_keylength(s->indigest))
-			&& digest_set_key(s->outdigest, s->key + cipher_keylength(s->incipher) + digest_keylength(s->indigest) + cipher_keylength(s->outcipher), digest_keylength(s->outdigest));
-		if(!result)
-			return false;
+		if(!cipher_set_counter_key(s->outcipher, s->key + cipher_keylength(s->incipher)))
+			return error(s, EINVAL, "Failed to set counter");
 	} else {
-		bool result
-			=  cipher_set_counter_key(s->outcipher, s->key)
-			&& digest_set_key(s->outdigest, s->key + cipher_keylength(s->outcipher), digest_keylength(s->outdigest));
-		if(!result)
-			return false;
+		if(!cipher_set_counter_key(s->outcipher, s->key))
+			return error(s, EINVAL, "Failed to set counter");
 	}
 
 	return true;
@@ -407,15 +388,11 @@ static bool receive_handshake(sptps_t *s, const char *data, uint16_t len) {
 // Check datagram for valid HMAC
 bool sptps_verify_datagram(sptps_t *s, const char *data, size_t len) {
 	if(!s->instate || len < 21)
-		return false;
-
-	char buffer[len + 23];
-	uint16_t netlen = htons(len - 21);
+		return error(s, EIO, "Received short packet");
 
-	memcpy(buffer, &netlen, 2);
-	memcpy(buffer + 2, data, len);
+	// TODO: just decrypt without updating the replay window
 
-	return digest_verify(s->indigest, buffer, len - 14, buffer + len - 14);
+	return true;
 }
 
 // Receive incoming data, datagram version.
@@ -441,16 +418,16 @@ static bool sptps_receive_data_datagram(sptps_t *s, const char *data, size_t len
 		return receive_handshake(s, data + 5, len - 5);
 	}
 
-	// Check HMAC.
-	uint16_t netlen = htons(len - 21);
+	// Decrypt
 
-	char buffer[len + 23];
+	char buffer[len];
 
-	memcpy(buffer, &netlen, 2);
-	memcpy(buffer + 2, data, len);
+	if(!cipher_set_counter(s->incipher, data, sizeof seqno))
+		return error(s, EINVAL, "Failed to set counter");
+	size_t outlen;
 
-	if(!digest_verify(s->indigest, buffer, len - 14, buffer + len - 14))
-		return error(s, EIO, "Invalid HMAC");
+	if(!cipher_gcm_decrypt(s->incipher, data + 4, len - 4, buffer, &outlen))
+		return error(s, EIO, "Failed to decrypt and verify packet");
 
 	// Replay protection using a sliding window of configurable size.
 	// s->inseqno is expected sequence number
@@ -492,26 +469,19 @@ static bool sptps_receive_data_datagram(sptps_t *s, const char *data, size_t len
 	else
 		s->received++;
 
-	// Decrypt.
-	memcpy(&seqno, buffer + 2, 4);
-	if(!cipher_set_counter(s->incipher, &seqno, sizeof seqno))
-		return false;
-	if(!cipher_counter_xor(s->incipher, buffer + 6, len - 4, buffer + 6))
-		return false;
-
 	// Append a NULL byte for safety.
-	buffer[len - 14] = 0;
+	buffer[len - 20] = 0;
 
-	uint8_t type = buffer[6];
+	uint8_t type = buffer[0];
 
 	if(type < SPTPS_HANDSHAKE) {
 		if(!s->instate)
 			return error(s, EIO, "Application record received before handshake finished");
-		if(!s->receive_record(s->handle, type, buffer + 7, len - 21))
-			return false;
+		if(!s->receive_record(s->handle, type, buffer + 1, len - 21))
+			abort();
 	} else if(type == SPTPS_HANDSHAKE) {
-		if(!receive_handshake(s, buffer + 7, len - 21))
-			return false;
+		if(!receive_handshake(s, buffer + 1, len - 21))
+			abort();
 	} else {
 		return error(s, EIO, "Invalid record type %d", type);
 	}
@@ -529,8 +499,8 @@ bool sptps_receive_data(sptps_t *s, const char *data, size_t len) {
 
 	while(len) {
 		// First read the 2 length bytes.
-		if(s->buflen < 6) {
-			size_t toread = 6 - s->buflen;
+		if(s->buflen < 2) {
+			size_t toread = 2 - s->buflen;
 			if(toread > len)
 				toread = len;
 
@@ -541,36 +511,39 @@ bool sptps_receive_data(sptps_t *s, const char *data, size_t len) {
 			data += toread;
 
 			// Exit early if we don't have the full length.
-			if(s->buflen < 6)
+			if(s->buflen < 2)
 				return true;
 
+			// Update sequence number.
+
+			uint32_t seqno = htonl(s->inseqno++);
+
 			// Decrypt the length bytes
 
 			if(s->instate) {
-				if(!cipher_counter_xor(s->incipher, s->inbuf + 4, 2, &s->reclen))
-					return false;
+				if(!cipher_set_counter(s->incipher, &seqno, 4))
+					return error(s, EINVAL, "Failed to set counter");
+
+				if(!cipher_gcm_decrypt_start(s->incipher, s->inbuf, 2, &s->reclen, NULL))
+					return error(s, EINVAL, "Failed to decrypt record");
 			} else {
-				memcpy(&s->reclen, s->inbuf + 4, 2);
+				memcpy(&s->reclen, s->inbuf, 2);
 			}
 
 			s->reclen = ntohs(s->reclen);
 
 			// If we have the length bytes, ensure our buffer can hold the whole request.
-			s->inbuf = realloc(s->inbuf, s->reclen + 23UL);
+			s->inbuf = realloc(s->inbuf, s->reclen + 19UL);
 			if(!s->inbuf)
 				return error(s, errno, strerror(errno));
 
-			// Add sequence number.
-			uint32_t seqno = htonl(s->inseqno++);
-			memcpy(s->inbuf, &seqno, 4);
-
 			// Exit early if we have no more data to process.
 			if(!len)
 				return true;
 		}
 
 		// Read up to the end of the record.
-		size_t toread = s->reclen + (s->instate ? 23UL : 7UL) - s->buflen;
+		size_t toread = s->reclen + (s->instate ? 19UL : 3UL) - s->buflen;
 		if(toread > len)
 			toread = len;
 
@@ -580,36 +553,33 @@ bool sptps_receive_data(sptps_t *s, const char *data, size_t len) {
 		data += toread;
 
 		// If we don't have a whole record, exit.
-		if(s->buflen < s->reclen + (s->instate ? 23UL : 7UL))
+		if(s->buflen < s->reclen + (s->instate ? 19UL : 3UL))
 			return true;
 
 		// Check HMAC and decrypt.
 		if(s->instate) {
-			if(!digest_verify(s->indigest, s->inbuf, s->reclen + 7UL, s->inbuf + s->reclen + 7UL))
-				return error(s, EIO, "Invalid HMAC");
-
-			if(!cipher_counter_xor(s->incipher, s->inbuf + 6UL, s->reclen + 1UL, s->inbuf + 6UL))
-				return false;
+			if(!cipher_gcm_decrypt_finish(s->incipher, s->inbuf + 2UL, s->reclen + 17UL, s->inbuf + 2UL, NULL))
+				return error(s, EINVAL, "Failed to decrypt and verify record");
 		}
 
 		// Append a NULL byte for safety.
-		s->inbuf[s->reclen + 7UL] = 0;
+		s->inbuf[s->reclen + 3UL] = 0;
 
-		uint8_t type = s->inbuf[6];
+		uint8_t type = s->inbuf[2];
 
 		if(type < SPTPS_HANDSHAKE) {
 			if(!s->instate)
 				return error(s, EIO, "Application record received before handshake finished");
-			if(!s->receive_record(s->handle, type, s->inbuf + 7, s->reclen))
+			if(!s->receive_record(s->handle, type, s->inbuf + 3, s->reclen))
 				return false;
 		} else if(type == SPTPS_HANDSHAKE) {
-			if(!receive_handshake(s, s->inbuf + 7, s->reclen))
+			if(!receive_handshake(s, s->inbuf + 3, s->reclen))
 				return false;
 		} else {
 			return error(s, EIO, "Invalid record type %d", type);
 		}
 
-		s->buflen = 4;
+		s->buflen = 0;
 	}
 
 	return true;
@@ -641,8 +611,7 @@ bool sptps_start(sptps_t *s, void *handle, bool initiator, bool datagram, ecdsa_
 		s->inbuf = malloc(7);
 		if(!s->inbuf)
 			return error(s, errno, strerror(errno));
-		s->buflen = 4;
-		memset(s->inbuf, 0, 4);
+		s->buflen = 0;
 	}
 
 	memcpy(s->label, label, labellen);
diff --git a/src/sptps_speed.c b/src/sptps_speed.c
index 69f75997..268e6c9b 100644
--- a/src/sptps_speed.c
+++ b/src/sptps_speed.c
@@ -48,7 +48,8 @@ static void receive_data(sptps_t *sptps) {
 	char buf[4096];
 	int fd = *(int *)sptps->handle;
 	size_t len = recv(fd, buf, sizeof buf, 0);
-	sptps_receive_data(sptps, buf, len);
+	if(!sptps_receive_data(sptps, buf, len))
+		abort();
 }
 
 struct timespec start;
@@ -62,12 +63,12 @@ static void clock_start() {
 	clock_gettime(CLOCK_PROCESS_CPUTIME_ID, &start);
 }
 
-static bool clock_countto(int seconds) {
+static bool clock_countto(double seconds) {
 	clock_gettime(CLOCK_PROCESS_CPUTIME_ID, &end);
 	elapsed = end.tv_sec + end.tv_nsec * 1e-9 - start.tv_sec - start.tv_nsec * 1e-9;
 	if(elapsed < seconds)
 		return ++count;
-		
+
 	rate = count / elapsed;
 	return false;
 }
@@ -77,38 +78,39 @@ int main(int argc, char *argv[]) {
 	ecdh_t *ecdh1, *ecdh2;
 	sptps_t sptps1, sptps2;
 	char buf1[4096], buf2[4096], buf3[4096];
+	double duration = argc > 1 ? atof(argv[1]) : 10;
 
 	crypto_init();
 
 	// Key generation
 
-	fprintf(stderr, "Generating keys for 10 seconds: ");
-	for(clock_start(); clock_countto(10);)
+	fprintf(stderr, "Generating keys for %lg seconds: ", duration);
+	for(clock_start(); clock_countto(duration);)
 		ecdsa_free(ecdsa_generate());
-	fprintf(stderr, "%13.2lf op/s\n", rate);
+	fprintf(stderr, "%17.2lf op/s\n", rate);
 
 	key1 = ecdsa_generate();
 	key2 = ecdsa_generate();
 
 	// ECDSA signatures
 
-	fprintf(stderr, "ECDSA sign for 10 seconds: ");
-	for(clock_start(); clock_countto(10);)
+	fprintf(stderr, "ECDSA sign for %lg seconds: ", duration);
+	for(clock_start(); clock_countto(duration);)
 		ecdsa_sign(key1, buf1, 256, buf2);
-	fprintf(stderr, "%18.2lf op/s\n", rate);
-	
-	fprintf(stderr, "ECDSA verify for 10 seconds: ");
-	for(clock_start(); clock_countto(10);)
+	fprintf(stderr, "%22.2lf op/s\n", rate);
+
+	fprintf(stderr, "ECDSA verify for %lg seconds: ", duration);
+	for(clock_start(); clock_countto(duration);)
 		ecdsa_verify(key1, buf1, 256, buf2);
-	fprintf(stderr, "%16.2lf op/s\n", rate);
+	fprintf(stderr, "%20.2lf op/s\n", rate);
 
 	ecdh1 = ecdh_generate_public(buf1);
-	fprintf(stderr, "ECDH for 10 seconds: ");
-	for(clock_start(); clock_countto(10);) {
+	fprintf(stderr, "ECDH for %lg seconds: ", duration);
+	for(clock_start(); clock_countto(duration);) {
 		ecdh2 = ecdh_generate_public(buf2);
 		ecdh_compute_shared(ecdh2, buf1, buf3);
 	}
-	fprintf(stderr, "%24.2lf op/s\n", rate);
+	fprintf(stderr, "%28.2lf op/s\n", rate);
 	ecdh_free(ecdh1);
 
 	// SPTPS authentication phase
@@ -121,8 +123,8 @@ int main(int argc, char *argv[]) {
 
 	struct pollfd pfd[2] = {{fd[0], POLLIN}, {fd[1], POLLIN}};
 
-	fprintf(stderr, "SPTPS authenticate for 10 seconds: ");
-	for(clock_start(); clock_countto(10);) {
+	fprintf(stderr, "SPTPS/TCP authenticate for %lg seconds: ", duration);
+	for(clock_start(); clock_countto(duration);) {
 		sptps_start(&sptps1, fd + 0, true, false, key1, key2, "sptps_speed", 11, send_data, receive_record);
 		sptps_start(&sptps2, fd + 1, false, false, key2, key1, "sptps_speed", 11, send_data, receive_record);
 		while(poll(pfd, 2, 0)) {
@@ -133,8 +135,8 @@ int main(int argc, char *argv[]) {
 		}
 		sptps_stop(&sptps1);
 		sptps_stop(&sptps2);
-	} 
-	fprintf(stderr, "%10.2lf op/s\n", rate * 2);	
+	}
+	fprintf(stderr, "%10.2lf op/s\n", rate * 2);
 
 	// SPTPS data
 
@@ -146,9 +148,61 @@ int main(int argc, char *argv[]) {
 		if(pfd[1].revents)
 			receive_data(&sptps2);
 	}
-	fprintf(stderr, "SPTPS transmit for 10 seconds: ");
-	for(clock_start(); clock_countto(10);) {
-		sptps_send_record(&sptps1, 0, buf1, 1451);
+	fprintf(stderr, "SPTPS/TCP transmit for %lg seconds: ", duration);
+	for(clock_start(); clock_countto(duration);) {
+		if(!sptps_send_record(&sptps1, 0, buf1, 1451))
+			abort();
+		receive_data(&sptps2);
+	}
+	rate *= 2 * 1451 * 8;
+	if(rate > 1e9)
+		fprintf(stderr, "%14.2lf Gbit/s\n", rate / 1e9);
+	else if(rate > 1e6)
+		fprintf(stderr, "%14.2lf Mbit/s\n", rate / 1e6);
+	else if(rate > 1e3)
+		fprintf(stderr, "%14.2lf kbit/s\n", rate / 1e3);
+	sptps_stop(&sptps1);
+	sptps_stop(&sptps2);
+
+	// SPTPS datagram authentication phase
+
+	close(fd[0]);
+	close(fd[1]);
+
+	if(socketpair(AF_UNIX, SOCK_DGRAM, 0, fd)) {
+		fprintf(stderr, "Could not create a UNIX socket pair: %s\n", strerror(errno));
+		return 1;
+	}
+
+	fprintf(stderr, "SPTPS/UDP authenticate for %lg seconds: ", duration);
+	for(clock_start(); clock_countto(duration);) {
+		sptps_start(&sptps1, fd + 0, true, true, key1, key2, "sptps_speed", 11, send_data, receive_record);
+		sptps_start(&sptps2, fd + 1, false, true, key2, key1, "sptps_speed", 11, send_data, receive_record);
+		while(poll(pfd, 2, 0)) {
+			if(pfd[0].revents)
+				receive_data(&sptps1);
+			if(pfd[1].revents)
+				receive_data(&sptps2);
+		}
+		sptps_stop(&sptps1);
+		sptps_stop(&sptps2);
+	}
+	fprintf(stderr, "%10.2lf op/s\n", rate * 2);
+
+	// SPTPS datagram data
+
+	sptps_start(&sptps1, fd + 0, true, true, key1, key2, "sptps_speed", 11, send_data, receive_record);
+	sptps_start(&sptps2, fd + 1, false, true, key2, key1, "sptps_speed", 11, send_data, receive_record);
+	while(poll(pfd, 2, 0)) {
+		if(pfd[0].revents)
+			receive_data(&sptps1);
+		if(pfd[1].revents)
+			receive_data(&sptps2);
+	}
+	fprintf(stderr, "SPTPS/UDP transmit for %lg seconds: ", duration);
+	for(clock_start(); clock_countto(duration);) {
+		if(!sptps_send_record(&sptps1, 0, buf1, 1451))
+			abort();
 		receive_data(&sptps2);
 	}
 	rate *= 2 * 1451 * 8;