tinc/src/sptps.c

644 lines
19 KiB
C
Raw Normal View History

Start of "Simple Peer-To-Peer Security" protocol. Encryption and authentication of the meta connection is spread out over meta.c and protocol_auth.c. The new protocol was added there as well, leading to spaghetti code. To improve things, the new protocol will now be implemented in sptps.[ch]. The goal is to have a very simplified version of TLS. There is a record layer, and there are only two record types: application data and handshake messages. The handshake message contains a random nonce, an ephemeral ECDH public key, and an ECDSA signature over the former. After the ECDH public keys are exchanged, a shared secret is calculated, and a TLS style PRF is used to generate the key material for the cipher and HMAC algorithm, and further communication is encrypted and authenticated. A lot of the simplicity comes from the fact that both sides must have each other's public keys in advance, and there are no options to choose. There will be one fixed cipher suite, and both peers always authenticate each other. (Inspiration taken from Ian Grigg's hypotheses[0].) There might be some compromise in the future, to enable or disable encryption, authentication and compression, but there will be no choice of algorithms. This will allow SPTPS to be built with a few embedded crypto algorithms instead of linking with huge crypto libraries. The API is also kept simple. There is a start and a stop function. All data necessary to make the connection work is passed in the start function. Instead having both send- and receive-record functions, there is a send-record function and a receive-data function. The latter will pass protocol data received from the peer to the SPTPS implementation, which will in turn call a receive-record callback function when necessary. This hides all the handshaking from the application, and is completely independent from any event loop or socket characteristics. [0] http://iang.org/ssl/hn_hypotheses_in_secure_protocol_design.html
2011-07-24 13:44:51 +00:00
/*
sptps.c -- Simple Peer-to-Peer Security
Copyright (C) 2011-2015 Guus Sliepen <guus@tinc-vpn.org>,
2010 Brandon L. Black <blblack@gmail.com>
Start of "Simple Peer-To-Peer Security" protocol. Encryption and authentication of the meta connection is spread out over meta.c and protocol_auth.c. The new protocol was added there as well, leading to spaghetti code. To improve things, the new protocol will now be implemented in sptps.[ch]. The goal is to have a very simplified version of TLS. There is a record layer, and there are only two record types: application data and handshake messages. The handshake message contains a random nonce, an ephemeral ECDH public key, and an ECDSA signature over the former. After the ECDH public keys are exchanged, a shared secret is calculated, and a TLS style PRF is used to generate the key material for the cipher and HMAC algorithm, and further communication is encrypted and authenticated. A lot of the simplicity comes from the fact that both sides must have each other's public keys in advance, and there are no options to choose. There will be one fixed cipher suite, and both peers always authenticate each other. (Inspiration taken from Ian Grigg's hypotheses[0].) There might be some compromise in the future, to enable or disable encryption, authentication and compression, but there will be no choice of algorithms. This will allow SPTPS to be built with a few embedded crypto algorithms instead of linking with huge crypto libraries. The API is also kept simple. There is a start and a stop function. All data necessary to make the connection work is passed in the start function. Instead having both send- and receive-record functions, there is a send-record function and a receive-data function. The latter will pass protocol data received from the peer to the SPTPS implementation, which will in turn call a receive-record callback function when necessary. This hides all the handshaking from the application, and is completely independent from any event loop or socket characteristics. [0] http://iang.org/ssl/hn_hypotheses_in_secure_protocol_design.html
2011-07-24 13:44:51 +00:00
This program is free software; you can redistribute it and/or modify
it under the terms of the GNU General Public License as published by
the Free Software Foundation; either version 2 of the License, or
(at your option) any later version.
This program is distributed in the hope that it will be useful,
but WITHOUT ANY WARRANTY; without even the implied warranty of
MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
GNU General Public License for more details.
You should have received a copy of the GNU General Public License along
with this program; if not, write to the Free Software Foundation, Inc.,
51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
*/
#include "system.h"
#include "chacha-poly1305/chacha-poly1305.h"
Start of "Simple Peer-To-Peer Security" protocol. Encryption and authentication of the meta connection is spread out over meta.c and protocol_auth.c. The new protocol was added there as well, leading to spaghetti code. To improve things, the new protocol will now be implemented in sptps.[ch]. The goal is to have a very simplified version of TLS. There is a record layer, and there are only two record types: application data and handshake messages. The handshake message contains a random nonce, an ephemeral ECDH public key, and an ECDSA signature over the former. After the ECDH public keys are exchanged, a shared secret is calculated, and a TLS style PRF is used to generate the key material for the cipher and HMAC algorithm, and further communication is encrypted and authenticated. A lot of the simplicity comes from the fact that both sides must have each other's public keys in advance, and there are no options to choose. There will be one fixed cipher suite, and both peers always authenticate each other. (Inspiration taken from Ian Grigg's hypotheses[0].) There might be some compromise in the future, to enable or disable encryption, authentication and compression, but there will be no choice of algorithms. This will allow SPTPS to be built with a few embedded crypto algorithms instead of linking with huge crypto libraries. The API is also kept simple. There is a start and a stop function. All data necessary to make the connection work is passed in the start function. Instead having both send- and receive-record functions, there is a send-record function and a receive-data function. The latter will pass protocol data received from the peer to the SPTPS implementation, which will in turn call a receive-record callback function when necessary. This hides all the handshaking from the application, and is completely independent from any event loop or socket characteristics. [0] http://iang.org/ssl/hn_hypotheses_in_secure_protocol_design.html
2011-07-24 13:44:51 +00:00
#include "crypto.h"
#include "ecdh.h"
#include "ecdsa.h"
2012-10-14 13:37:24 +00:00
#include "logger.h"
Start of "Simple Peer-To-Peer Security" protocol. Encryption and authentication of the meta connection is spread out over meta.c and protocol_auth.c. The new protocol was added there as well, leading to spaghetti code. To improve things, the new protocol will now be implemented in sptps.[ch]. The goal is to have a very simplified version of TLS. There is a record layer, and there are only two record types: application data and handshake messages. The handshake message contains a random nonce, an ephemeral ECDH public key, and an ECDSA signature over the former. After the ECDH public keys are exchanged, a shared secret is calculated, and a TLS style PRF is used to generate the key material for the cipher and HMAC algorithm, and further communication is encrypted and authenticated. A lot of the simplicity comes from the fact that both sides must have each other's public keys in advance, and there are no options to choose. There will be one fixed cipher suite, and both peers always authenticate each other. (Inspiration taken from Ian Grigg's hypotheses[0].) There might be some compromise in the future, to enable or disable encryption, authentication and compression, but there will be no choice of algorithms. This will allow SPTPS to be built with a few embedded crypto algorithms instead of linking with huge crypto libraries. The API is also kept simple. There is a start and a stop function. All data necessary to make the connection work is passed in the start function. Instead having both send- and receive-record functions, there is a send-record function and a receive-data function. The latter will pass protocol data received from the peer to the SPTPS implementation, which will in turn call a receive-record callback function when necessary. This hides all the handshaking from the application, and is completely independent from any event loop or socket characteristics. [0] http://iang.org/ssl/hn_hypotheses_in_secure_protocol_design.html
2011-07-24 13:44:51 +00:00
#include "prf.h"
#include "sptps.h"
unsigned int sptps_replaywin = 16;
/*
Nonce MUST be exchanged first (done)
Signatures MUST be done over both nonces, to guarantee the signature is fresh
Otherwise: if ECDHE key of one side is compromised, it can be reused!
Add explicit tag to beginning of structure to distinguish the client and server when signing. (done)
Sign all handshake messages up to ECDHE kex with long-term public keys. (done)
2014-05-18 18:47:04 +00:00
HMACed KEX finished message to prevent downgrade attacks and prove you have the right key material (done by virtue of Ed25519 over the whole ECDHE exchange?)
Explicit close message needs to be added.
Maybe do add some alert messages to give helpful error messages? Not more than TLS sends.
2011-10-06 07:34:34 +00:00
Use counter mode instead of OFB. (done)
Make sure ECC operations are fixed time (aka prevent side-channel attacks).
*/
2012-10-14 13:37:24 +00:00
void sptps_log_quiet(sptps_t *s, int s_errno, const char *format, va_list ap) {
}
void sptps_log_stderr(sptps_t *s, int s_errno, const char *format, va_list ap) {
vfprintf(stderr, format, ap);
fputc('\n', stderr);
}
void (*sptps_log)(sptps_t *s, int s_errno, const char *format, va_list ap) = sptps_log_stderr;
// Log an error message.
2012-10-14 13:37:24 +00:00
static bool error(sptps_t *s, int s_errno, const char *format, ...) {
if(format) {
va_list ap;
va_start(ap, format);
sptps_log(s, s_errno, format, ap);
va_end(ap);
}
Start of "Simple Peer-To-Peer Security" protocol. Encryption and authentication of the meta connection is spread out over meta.c and protocol_auth.c. The new protocol was added there as well, leading to spaghetti code. To improve things, the new protocol will now be implemented in sptps.[ch]. The goal is to have a very simplified version of TLS. There is a record layer, and there are only two record types: application data and handshake messages. The handshake message contains a random nonce, an ephemeral ECDH public key, and an ECDSA signature over the former. After the ECDH public keys are exchanged, a shared secret is calculated, and a TLS style PRF is used to generate the key material for the cipher and HMAC algorithm, and further communication is encrypted and authenticated. A lot of the simplicity comes from the fact that both sides must have each other's public keys in advance, and there are no options to choose. There will be one fixed cipher suite, and both peers always authenticate each other. (Inspiration taken from Ian Grigg's hypotheses[0].) There might be some compromise in the future, to enable or disable encryption, authentication and compression, but there will be no choice of algorithms. This will allow SPTPS to be built with a few embedded crypto algorithms instead of linking with huge crypto libraries. The API is also kept simple. There is a start and a stop function. All data necessary to make the connection work is passed in the start function. Instead having both send- and receive-record functions, there is a send-record function and a receive-data function. The latter will pass protocol data received from the peer to the SPTPS implementation, which will in turn call a receive-record callback function when necessary. This hides all the handshaking from the application, and is completely independent from any event loop or socket characteristics. [0] http://iang.org/ssl/hn_hypotheses_in_secure_protocol_design.html
2011-07-24 13:44:51 +00:00
errno = s_errno;
return false;
}
2012-10-14 13:37:24 +00:00
static void warning(sptps_t *s, const char *format, ...) {
va_list ap;
va_start(ap, format);
sptps_log(s, 0, format, ap);
va_end(ap);
}
// 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 void *data, uint16_t len) {
char buffer[len + 21UL];
// Create header with sequence number, length and record type
uint32_t seqno = s->outseqno++;
uint32_t netseqno = ntohl(seqno);
memcpy(buffer, &netseqno, 4);
buffer[4] = type;
memcpy(buffer + 5, data, len);
if(s->outstate) {
// If first handshake has finished, encrypt and HMAC
chacha_poly1305_encrypt(s->outcipher, seqno, buffer + 4, len + 1, buffer + 4, NULL);
return s->send_data(s->handle, type, buffer, len + 21UL);
} else {
// Otherwise send as plaintext
return s->send_data(s->handle, type, buffer, len + 5UL);
}
}
// Send a record (private version, accepts all record types, handles encryption and authentication).
static bool send_record_priv(sptps_t *s, uint8_t type, const void *data, uint16_t len) {
if(s->datagram)
return send_record_priv_datagram(s, type, data, len);
char buffer[len + 19UL];
Start of "Simple Peer-To-Peer Security" protocol. Encryption and authentication of the meta connection is spread out over meta.c and protocol_auth.c. The new protocol was added there as well, leading to spaghetti code. To improve things, the new protocol will now be implemented in sptps.[ch]. The goal is to have a very simplified version of TLS. There is a record layer, and there are only two record types: application data and handshake messages. The handshake message contains a random nonce, an ephemeral ECDH public key, and an ECDSA signature over the former. After the ECDH public keys are exchanged, a shared secret is calculated, and a TLS style PRF is used to generate the key material for the cipher and HMAC algorithm, and further communication is encrypted and authenticated. A lot of the simplicity comes from the fact that both sides must have each other's public keys in advance, and there are no options to choose. There will be one fixed cipher suite, and both peers always authenticate each other. (Inspiration taken from Ian Grigg's hypotheses[0].) There might be some compromise in the future, to enable or disable encryption, authentication and compression, but there will be no choice of algorithms. This will allow SPTPS to be built with a few embedded crypto algorithms instead of linking with huge crypto libraries. The API is also kept simple. There is a start and a stop function. All data necessary to make the connection work is passed in the start function. Instead having both send- and receive-record functions, there is a send-record function and a receive-data function. The latter will pass protocol data received from the peer to the SPTPS implementation, which will in turn call a receive-record callback function when necessary. This hides all the handshaking from the application, and is completely independent from any event loop or socket characteristics. [0] http://iang.org/ssl/hn_hypotheses_in_secure_protocol_design.html
2011-07-24 13:44:51 +00:00
// Create header with sequence number, length and record type
uint32_t seqno = s->outseqno++;
Start of "Simple Peer-To-Peer Security" protocol. Encryption and authentication of the meta connection is spread out over meta.c and protocol_auth.c. The new protocol was added there as well, leading to spaghetti code. To improve things, the new protocol will now be implemented in sptps.[ch]. The goal is to have a very simplified version of TLS. There is a record layer, and there are only two record types: application data and handshake messages. The handshake message contains a random nonce, an ephemeral ECDH public key, and an ECDSA signature over the former. After the ECDH public keys are exchanged, a shared secret is calculated, and a TLS style PRF is used to generate the key material for the cipher and HMAC algorithm, and further communication is encrypted and authenticated. A lot of the simplicity comes from the fact that both sides must have each other's public keys in advance, and there are no options to choose. There will be one fixed cipher suite, and both peers always authenticate each other. (Inspiration taken from Ian Grigg's hypotheses[0].) There might be some compromise in the future, to enable or disable encryption, authentication and compression, but there will be no choice of algorithms. This will allow SPTPS to be built with a few embedded crypto algorithms instead of linking with huge crypto libraries. The API is also kept simple. There is a start and a stop function. All data necessary to make the connection work is passed in the start function. Instead having both send- and receive-record functions, there is a send-record function and a receive-data function. The latter will pass protocol data received from the peer to the SPTPS implementation, which will in turn call a receive-record callback function when necessary. This hides all the handshaking from the application, and is completely independent from any event loop or socket characteristics. [0] http://iang.org/ssl/hn_hypotheses_in_secure_protocol_design.html
2011-07-24 13:44:51 +00:00
uint16_t netlen = htons(len);
memcpy(buffer, &netlen, 2);
buffer[2] = type;
memcpy(buffer + 3, data, len);
Start of "Simple Peer-To-Peer Security" protocol. Encryption and authentication of the meta connection is spread out over meta.c and protocol_auth.c. The new protocol was added there as well, leading to spaghetti code. To improve things, the new protocol will now be implemented in sptps.[ch]. The goal is to have a very simplified version of TLS. There is a record layer, and there are only two record types: application data and handshake messages. The handshake message contains a random nonce, an ephemeral ECDH public key, and an ECDSA signature over the former. After the ECDH public keys are exchanged, a shared secret is calculated, and a TLS style PRF is used to generate the key material for the cipher and HMAC algorithm, and further communication is encrypted and authenticated. A lot of the simplicity comes from the fact that both sides must have each other's public keys in advance, and there are no options to choose. There will be one fixed cipher suite, and both peers always authenticate each other. (Inspiration taken from Ian Grigg's hypotheses[0].) There might be some compromise in the future, to enable or disable encryption, authentication and compression, but there will be no choice of algorithms. This will allow SPTPS to be built with a few embedded crypto algorithms instead of linking with huge crypto libraries. The API is also kept simple. There is a start and a stop function. All data necessary to make the connection work is passed in the start function. Instead having both send- and receive-record functions, there is a send-record function and a receive-data function. The latter will pass protocol data received from the peer to the SPTPS implementation, which will in turn call a receive-record callback function when necessary. This hides all the handshaking from the application, and is completely independent from any event loop or socket characteristics. [0] http://iang.org/ssl/hn_hypotheses_in_secure_protocol_design.html
2011-07-24 13:44:51 +00:00
if(s->outstate) {
Start of "Simple Peer-To-Peer Security" protocol. Encryption and authentication of the meta connection is spread out over meta.c and protocol_auth.c. The new protocol was added there as well, leading to spaghetti code. To improve things, the new protocol will now be implemented in sptps.[ch]. The goal is to have a very simplified version of TLS. There is a record layer, and there are only two record types: application data and handshake messages. The handshake message contains a random nonce, an ephemeral ECDH public key, and an ECDSA signature over the former. After the ECDH public keys are exchanged, a shared secret is calculated, and a TLS style PRF is used to generate the key material for the cipher and HMAC algorithm, and further communication is encrypted and authenticated. A lot of the simplicity comes from the fact that both sides must have each other's public keys in advance, and there are no options to choose. There will be one fixed cipher suite, and both peers always authenticate each other. (Inspiration taken from Ian Grigg's hypotheses[0].) There might be some compromise in the future, to enable or disable encryption, authentication and compression, but there will be no choice of algorithms. This will allow SPTPS to be built with a few embedded crypto algorithms instead of linking with huge crypto libraries. The API is also kept simple. There is a start and a stop function. All data necessary to make the connection work is passed in the start function. Instead having both send- and receive-record functions, there is a send-record function and a receive-data function. The latter will pass protocol data received from the peer to the SPTPS implementation, which will in turn call a receive-record callback function when necessary. This hides all the handshaking from the application, and is completely independent from any event loop or socket characteristics. [0] http://iang.org/ssl/hn_hypotheses_in_secure_protocol_design.html
2011-07-24 13:44:51 +00:00
// If first handshake has finished, encrypt and HMAC
chacha_poly1305_encrypt(s->outcipher, seqno, buffer + 2, len + 1, buffer + 2, NULL);
return s->send_data(s->handle, type, buffer, len + 19UL);
Start of "Simple Peer-To-Peer Security" protocol. Encryption and authentication of the meta connection is spread out over meta.c and protocol_auth.c. The new protocol was added there as well, leading to spaghetti code. To improve things, the new protocol will now be implemented in sptps.[ch]. The goal is to have a very simplified version of TLS. There is a record layer, and there are only two record types: application data and handshake messages. The handshake message contains a random nonce, an ephemeral ECDH public key, and an ECDSA signature over the former. After the ECDH public keys are exchanged, a shared secret is calculated, and a TLS style PRF is used to generate the key material for the cipher and HMAC algorithm, and further communication is encrypted and authenticated. A lot of the simplicity comes from the fact that both sides must have each other's public keys in advance, and there are no options to choose. There will be one fixed cipher suite, and both peers always authenticate each other. (Inspiration taken from Ian Grigg's hypotheses[0].) There might be some compromise in the future, to enable or disable encryption, authentication and compression, but there will be no choice of algorithms. This will allow SPTPS to be built with a few embedded crypto algorithms instead of linking with huge crypto libraries. The API is also kept simple. There is a start and a stop function. All data necessary to make the connection work is passed in the start function. Instead having both send- and receive-record functions, there is a send-record function and a receive-data function. The latter will pass protocol data received from the peer to the SPTPS implementation, which will in turn call a receive-record callback function when necessary. This hides all the handshaking from the application, and is completely independent from any event loop or socket characteristics. [0] http://iang.org/ssl/hn_hypotheses_in_secure_protocol_design.html
2011-07-24 13:44:51 +00:00
} else {
// Otherwise send as plaintext
return s->send_data(s->handle, type, buffer, len + 3UL);
Start of "Simple Peer-To-Peer Security" protocol. Encryption and authentication of the meta connection is spread out over meta.c and protocol_auth.c. The new protocol was added there as well, leading to spaghetti code. To improve things, the new protocol will now be implemented in sptps.[ch]. The goal is to have a very simplified version of TLS. There is a record layer, and there are only two record types: application data and handshake messages. The handshake message contains a random nonce, an ephemeral ECDH public key, and an ECDSA signature over the former. After the ECDH public keys are exchanged, a shared secret is calculated, and a TLS style PRF is used to generate the key material for the cipher and HMAC algorithm, and further communication is encrypted and authenticated. A lot of the simplicity comes from the fact that both sides must have each other's public keys in advance, and there are no options to choose. There will be one fixed cipher suite, and both peers always authenticate each other. (Inspiration taken from Ian Grigg's hypotheses[0].) There might be some compromise in the future, to enable or disable encryption, authentication and compression, but there will be no choice of algorithms. This will allow SPTPS to be built with a few embedded crypto algorithms instead of linking with huge crypto libraries. The API is also kept simple. There is a start and a stop function. All data necessary to make the connection work is passed in the start function. Instead having both send- and receive-record functions, there is a send-record function and a receive-data function. The latter will pass protocol data received from the peer to the SPTPS implementation, which will in turn call a receive-record callback function when necessary. This hides all the handshaking from the application, and is completely independent from any event loop or socket characteristics. [0] http://iang.org/ssl/hn_hypotheses_in_secure_protocol_design.html
2011-07-24 13:44:51 +00:00
}
}
// Send an application record.
bool sptps_send_record(sptps_t *s, uint8_t type, const void *data, uint16_t len) {
Start of "Simple Peer-To-Peer Security" protocol. Encryption and authentication of the meta connection is spread out over meta.c and protocol_auth.c. The new protocol was added there as well, leading to spaghetti code. To improve things, the new protocol will now be implemented in sptps.[ch]. The goal is to have a very simplified version of TLS. There is a record layer, and there are only two record types: application data and handshake messages. The handshake message contains a random nonce, an ephemeral ECDH public key, and an ECDSA signature over the former. After the ECDH public keys are exchanged, a shared secret is calculated, and a TLS style PRF is used to generate the key material for the cipher and HMAC algorithm, and further communication is encrypted and authenticated. A lot of the simplicity comes from the fact that both sides must have each other's public keys in advance, and there are no options to choose. There will be one fixed cipher suite, and both peers always authenticate each other. (Inspiration taken from Ian Grigg's hypotheses[0].) There might be some compromise in the future, to enable or disable encryption, authentication and compression, but there will be no choice of algorithms. This will allow SPTPS to be built with a few embedded crypto algorithms instead of linking with huge crypto libraries. The API is also kept simple. There is a start and a stop function. All data necessary to make the connection work is passed in the start function. Instead having both send- and receive-record functions, there is a send-record function and a receive-data function. The latter will pass protocol data received from the peer to the SPTPS implementation, which will in turn call a receive-record callback function when necessary. This hides all the handshaking from the application, and is completely independent from any event loop or socket characteristics. [0] http://iang.org/ssl/hn_hypotheses_in_secure_protocol_design.html
2011-07-24 13:44:51 +00:00
// Sanity checks: application cannot send data before handshake is finished,
// and only record types 0..127 are allowed.
if(!s->outstate)
Start of "Simple Peer-To-Peer Security" protocol. Encryption and authentication of the meta connection is spread out over meta.c and protocol_auth.c. The new protocol was added there as well, leading to spaghetti code. To improve things, the new protocol will now be implemented in sptps.[ch]. The goal is to have a very simplified version of TLS. There is a record layer, and there are only two record types: application data and handshake messages. The handshake message contains a random nonce, an ephemeral ECDH public key, and an ECDSA signature over the former. After the ECDH public keys are exchanged, a shared secret is calculated, and a TLS style PRF is used to generate the key material for the cipher and HMAC algorithm, and further communication is encrypted and authenticated. A lot of the simplicity comes from the fact that both sides must have each other's public keys in advance, and there are no options to choose. There will be one fixed cipher suite, and both peers always authenticate each other. (Inspiration taken from Ian Grigg's hypotheses[0].) There might be some compromise in the future, to enable or disable encryption, authentication and compression, but there will be no choice of algorithms. This will allow SPTPS to be built with a few embedded crypto algorithms instead of linking with huge crypto libraries. The API is also kept simple. There is a start and a stop function. All data necessary to make the connection work is passed in the start function. Instead having both send- and receive-record functions, there is a send-record function and a receive-data function. The latter will pass protocol data received from the peer to the SPTPS implementation, which will in turn call a receive-record callback function when necessary. This hides all the handshaking from the application, and is completely independent from any event loop or socket characteristics. [0] http://iang.org/ssl/hn_hypotheses_in_secure_protocol_design.html
2011-07-24 13:44:51 +00:00
return error(s, EINVAL, "Handshake phase not finished yet");
if(type >= SPTPS_HANDSHAKE)
Start of "Simple Peer-To-Peer Security" protocol. Encryption and authentication of the meta connection is spread out over meta.c and protocol_auth.c. The new protocol was added there as well, leading to spaghetti code. To improve things, the new protocol will now be implemented in sptps.[ch]. The goal is to have a very simplified version of TLS. There is a record layer, and there are only two record types: application data and handshake messages. The handshake message contains a random nonce, an ephemeral ECDH public key, and an ECDSA signature over the former. After the ECDH public keys are exchanged, a shared secret is calculated, and a TLS style PRF is used to generate the key material for the cipher and HMAC algorithm, and further communication is encrypted and authenticated. A lot of the simplicity comes from the fact that both sides must have each other's public keys in advance, and there are no options to choose. There will be one fixed cipher suite, and both peers always authenticate each other. (Inspiration taken from Ian Grigg's hypotheses[0].) There might be some compromise in the future, to enable or disable encryption, authentication and compression, but there will be no choice of algorithms. This will allow SPTPS to be built with a few embedded crypto algorithms instead of linking with huge crypto libraries. The API is also kept simple. There is a start and a stop function. All data necessary to make the connection work is passed in the start function. Instead having both send- and receive-record functions, there is a send-record function and a receive-data function. The latter will pass protocol data received from the peer to the SPTPS implementation, which will in turn call a receive-record callback function when necessary. This hides all the handshaking from the application, and is completely independent from any event loop or socket characteristics. [0] http://iang.org/ssl/hn_hypotheses_in_secure_protocol_design.html
2011-07-24 13:44:51 +00:00
return error(s, EINVAL, "Invalid application record type");
return send_record_priv(s, type, data, len);
}
// Send a Key EXchange record, containing a random nonce and an ECDHE public key.
Start of "Simple Peer-To-Peer Security" protocol. Encryption and authentication of the meta connection is spread out over meta.c and protocol_auth.c. The new protocol was added there as well, leading to spaghetti code. To improve things, the new protocol will now be implemented in sptps.[ch]. The goal is to have a very simplified version of TLS. There is a record layer, and there are only two record types: application data and handshake messages. The handshake message contains a random nonce, an ephemeral ECDH public key, and an ECDSA signature over the former. After the ECDH public keys are exchanged, a shared secret is calculated, and a TLS style PRF is used to generate the key material for the cipher and HMAC algorithm, and further communication is encrypted and authenticated. A lot of the simplicity comes from the fact that both sides must have each other's public keys in advance, and there are no options to choose. There will be one fixed cipher suite, and both peers always authenticate each other. (Inspiration taken from Ian Grigg's hypotheses[0].) There might be some compromise in the future, to enable or disable encryption, authentication and compression, but there will be no choice of algorithms. This will allow SPTPS to be built with a few embedded crypto algorithms instead of linking with huge crypto libraries. The API is also kept simple. There is a start and a stop function. All data necessary to make the connection work is passed in the start function. Instead having both send- and receive-record functions, there is a send-record function and a receive-data function. The latter will pass protocol data received from the peer to the SPTPS implementation, which will in turn call a receive-record callback function when necessary. This hides all the handshaking from the application, and is completely independent from any event loop or socket characteristics. [0] http://iang.org/ssl/hn_hypotheses_in_secure_protocol_design.html
2011-07-24 13:44:51 +00:00
static bool send_kex(sptps_t *s) {
size_t keylen = ECDH_SIZE;
// Make room for our KEX message, which we will keep around since send_sig() needs it.
if(s->mykex)
return false;
s->mykex = realloc(s->mykex, 1 + 32 + keylen);
if(!s->mykex)
Start of "Simple Peer-To-Peer Security" protocol. Encryption and authentication of the meta connection is spread out over meta.c and protocol_auth.c. The new protocol was added there as well, leading to spaghetti code. To improve things, the new protocol will now be implemented in sptps.[ch]. The goal is to have a very simplified version of TLS. There is a record layer, and there are only two record types: application data and handshake messages. The handshake message contains a random nonce, an ephemeral ECDH public key, and an ECDSA signature over the former. After the ECDH public keys are exchanged, a shared secret is calculated, and a TLS style PRF is used to generate the key material for the cipher and HMAC algorithm, and further communication is encrypted and authenticated. A lot of the simplicity comes from the fact that both sides must have each other's public keys in advance, and there are no options to choose. There will be one fixed cipher suite, and both peers always authenticate each other. (Inspiration taken from Ian Grigg's hypotheses[0].) There might be some compromise in the future, to enable or disable encryption, authentication and compression, but there will be no choice of algorithms. This will allow SPTPS to be built with a few embedded crypto algorithms instead of linking with huge crypto libraries. The API is also kept simple. There is a start and a stop function. All data necessary to make the connection work is passed in the start function. Instead having both send- and receive-record functions, there is a send-record function and a receive-data function. The latter will pass protocol data received from the peer to the SPTPS implementation, which will in turn call a receive-record callback function when necessary. This hides all the handshaking from the application, and is completely independent from any event loop or socket characteristics. [0] http://iang.org/ssl/hn_hypotheses_in_secure_protocol_design.html
2011-07-24 13:44:51 +00:00
return error(s, errno, strerror(errno));
// Set version byte to zero.
s->mykex[0] = SPTPS_VERSION;
// Create a random nonce.
randomize(s->mykex + 1, 32);
Start of "Simple Peer-To-Peer Security" protocol. Encryption and authentication of the meta connection is spread out over meta.c and protocol_auth.c. The new protocol was added there as well, leading to spaghetti code. To improve things, the new protocol will now be implemented in sptps.[ch]. The goal is to have a very simplified version of TLS. There is a record layer, and there are only two record types: application data and handshake messages. The handshake message contains a random nonce, an ephemeral ECDH public key, and an ECDSA signature over the former. After the ECDH public keys are exchanged, a shared secret is calculated, and a TLS style PRF is used to generate the key material for the cipher and HMAC algorithm, and further communication is encrypted and authenticated. A lot of the simplicity comes from the fact that both sides must have each other's public keys in advance, and there are no options to choose. There will be one fixed cipher suite, and both peers always authenticate each other. (Inspiration taken from Ian Grigg's hypotheses[0].) There might be some compromise in the future, to enable or disable encryption, authentication and compression, but there will be no choice of algorithms. This will allow SPTPS to be built with a few embedded crypto algorithms instead of linking with huge crypto libraries. The API is also kept simple. There is a start and a stop function. All data necessary to make the connection work is passed in the start function. Instead having both send- and receive-record functions, there is a send-record function and a receive-data function. The latter will pass protocol data received from the peer to the SPTPS implementation, which will in turn call a receive-record callback function when necessary. This hides all the handshaking from the application, and is completely independent from any event loop or socket characteristics. [0] http://iang.org/ssl/hn_hypotheses_in_secure_protocol_design.html
2011-07-24 13:44:51 +00:00
// Create a new ECDH public key.
if(!(s->ecdh = ecdh_generate_public(s->mykex + 1 + 32)))
return error(s, EINVAL, "Failed to generate ECDH public key");
Start of "Simple Peer-To-Peer Security" protocol. Encryption and authentication of the meta connection is spread out over meta.c and protocol_auth.c. The new protocol was added there as well, leading to spaghetti code. To improve things, the new protocol will now be implemented in sptps.[ch]. The goal is to have a very simplified version of TLS. There is a record layer, and there are only two record types: application data and handshake messages. The handshake message contains a random nonce, an ephemeral ECDH public key, and an ECDSA signature over the former. After the ECDH public keys are exchanged, a shared secret is calculated, and a TLS style PRF is used to generate the key material for the cipher and HMAC algorithm, and further communication is encrypted and authenticated. A lot of the simplicity comes from the fact that both sides must have each other's public keys in advance, and there are no options to choose. There will be one fixed cipher suite, and both peers always authenticate each other. (Inspiration taken from Ian Grigg's hypotheses[0].) There might be some compromise in the future, to enable or disable encryption, authentication and compression, but there will be no choice of algorithms. This will allow SPTPS to be built with a few embedded crypto algorithms instead of linking with huge crypto libraries. The API is also kept simple. There is a start and a stop function. All data necessary to make the connection work is passed in the start function. Instead having both send- and receive-record functions, there is a send-record function and a receive-data function. The latter will pass protocol data received from the peer to the SPTPS implementation, which will in turn call a receive-record callback function when necessary. This hides all the handshaking from the application, and is completely independent from any event loop or socket characteristics. [0] http://iang.org/ssl/hn_hypotheses_in_secure_protocol_design.html
2011-07-24 13:44:51 +00:00
return send_record_priv(s, SPTPS_HANDSHAKE, s->mykex, 1 + 32 + keylen);
}
2014-05-18 18:47:04 +00:00
// Send a SIGnature record, containing an Ed25519 signature over both KEX records.
static bool send_sig(sptps_t *s) {
size_t keylen = ECDH_SIZE;
size_t siglen = ecdsa_size(s->mykey);
// Concatenate both KEX messages, plus tag indicating if it is from the connection originator, plus label
char msg[(1 + 32 + keylen) * 2 + 1 + s->labellen];
char sig[siglen];
msg[0] = s->initiator;
memcpy(msg + 1, s->mykex, 1 + 32 + keylen);
memcpy(msg + 1 + 33 + keylen, s->hiskex, 1 + 32 + keylen);
memcpy(msg + 1 + 2 * (33 + keylen), s->label, s->labellen);
// Sign the result.
if(!ecdsa_sign(s->mykey, msg, sizeof msg, sig))
return error(s, EINVAL, "Failed to sign SIG record");
Start of "Simple Peer-To-Peer Security" protocol. Encryption and authentication of the meta connection is spread out over meta.c and protocol_auth.c. The new protocol was added there as well, leading to spaghetti code. To improve things, the new protocol will now be implemented in sptps.[ch]. The goal is to have a very simplified version of TLS. There is a record layer, and there are only two record types: application data and handshake messages. The handshake message contains a random nonce, an ephemeral ECDH public key, and an ECDSA signature over the former. After the ECDH public keys are exchanged, a shared secret is calculated, and a TLS style PRF is used to generate the key material for the cipher and HMAC algorithm, and further communication is encrypted and authenticated. A lot of the simplicity comes from the fact that both sides must have each other's public keys in advance, and there are no options to choose. There will be one fixed cipher suite, and both peers always authenticate each other. (Inspiration taken from Ian Grigg's hypotheses[0].) There might be some compromise in the future, to enable or disable encryption, authentication and compression, but there will be no choice of algorithms. This will allow SPTPS to be built with a few embedded crypto algorithms instead of linking with huge crypto libraries. The API is also kept simple. There is a start and a stop function. All data necessary to make the connection work is passed in the start function. Instead having both send- and receive-record functions, there is a send-record function and a receive-data function. The latter will pass protocol data received from the peer to the SPTPS implementation, which will in turn call a receive-record callback function when necessary. This hides all the handshaking from the application, and is completely independent from any event loop or socket characteristics. [0] http://iang.org/ssl/hn_hypotheses_in_secure_protocol_design.html
2011-07-24 13:44:51 +00:00
// Send the SIG exchange record.
return send_record_priv(s, SPTPS_HANDSHAKE, sig, sizeof sig);
Start of "Simple Peer-To-Peer Security" protocol. Encryption and authentication of the meta connection is spread out over meta.c and protocol_auth.c. The new protocol was added there as well, leading to spaghetti code. To improve things, the new protocol will now be implemented in sptps.[ch]. The goal is to have a very simplified version of TLS. There is a record layer, and there are only two record types: application data and handshake messages. The handshake message contains a random nonce, an ephemeral ECDH public key, and an ECDSA signature over the former. After the ECDH public keys are exchanged, a shared secret is calculated, and a TLS style PRF is used to generate the key material for the cipher and HMAC algorithm, and further communication is encrypted and authenticated. A lot of the simplicity comes from the fact that both sides must have each other's public keys in advance, and there are no options to choose. There will be one fixed cipher suite, and both peers always authenticate each other. (Inspiration taken from Ian Grigg's hypotheses[0].) There might be some compromise in the future, to enable or disable encryption, authentication and compression, but there will be no choice of algorithms. This will allow SPTPS to be built with a few embedded crypto algorithms instead of linking with huge crypto libraries. The API is also kept simple. There is a start and a stop function. All data necessary to make the connection work is passed in the start function. Instead having both send- and receive-record functions, there is a send-record function and a receive-data function. The latter will pass protocol data received from the peer to the SPTPS implementation, which will in turn call a receive-record callback function when necessary. This hides all the handshaking from the application, and is completely independent from any event loop or socket characteristics. [0] http://iang.org/ssl/hn_hypotheses_in_secure_protocol_design.html
2011-07-24 13:44:51 +00:00
}
// Generate key material from the shared secret created from the ECDHE key exchange.
static bool generate_key_material(sptps_t *s, const char *shared, size_t len) {
Start of "Simple Peer-To-Peer Security" protocol. Encryption and authentication of the meta connection is spread out over meta.c and protocol_auth.c. The new protocol was added there as well, leading to spaghetti code. To improve things, the new protocol will now be implemented in sptps.[ch]. The goal is to have a very simplified version of TLS. There is a record layer, and there are only two record types: application data and handshake messages. The handshake message contains a random nonce, an ephemeral ECDH public key, and an ECDSA signature over the former. After the ECDH public keys are exchanged, a shared secret is calculated, and a TLS style PRF is used to generate the key material for the cipher and HMAC algorithm, and further communication is encrypted and authenticated. A lot of the simplicity comes from the fact that both sides must have each other's public keys in advance, and there are no options to choose. There will be one fixed cipher suite, and both peers always authenticate each other. (Inspiration taken from Ian Grigg's hypotheses[0].) There might be some compromise in the future, to enable or disable encryption, authentication and compression, but there will be no choice of algorithms. This will allow SPTPS to be built with a few embedded crypto algorithms instead of linking with huge crypto libraries. The API is also kept simple. There is a start and a stop function. All data necessary to make the connection work is passed in the start function. Instead having both send- and receive-record functions, there is a send-record function and a receive-data function. The latter will pass protocol data received from the peer to the SPTPS implementation, which will in turn call a receive-record callback function when necessary. This hides all the handshaking from the application, and is completely independent from any event loop or socket characteristics. [0] http://iang.org/ssl/hn_hypotheses_in_secure_protocol_design.html
2011-07-24 13:44:51 +00:00
// Initialise cipher and digest structures if necessary
if(!s->outstate) {
s->incipher = chacha_poly1305_init();
s->outcipher = chacha_poly1305_init();
if(!s->incipher || !s->outcipher)
return error(s, EINVAL, "Failed to open cipher");
Start of "Simple Peer-To-Peer Security" protocol. Encryption and authentication of the meta connection is spread out over meta.c and protocol_auth.c. The new protocol was added there as well, leading to spaghetti code. To improve things, the new protocol will now be implemented in sptps.[ch]. The goal is to have a very simplified version of TLS. There is a record layer, and there are only two record types: application data and handshake messages. The handshake message contains a random nonce, an ephemeral ECDH public key, and an ECDSA signature over the former. After the ECDH public keys are exchanged, a shared secret is calculated, and a TLS style PRF is used to generate the key material for the cipher and HMAC algorithm, and further communication is encrypted and authenticated. A lot of the simplicity comes from the fact that both sides must have each other's public keys in advance, and there are no options to choose. There will be one fixed cipher suite, and both peers always authenticate each other. (Inspiration taken from Ian Grigg's hypotheses[0].) There might be some compromise in the future, to enable or disable encryption, authentication and compression, but there will be no choice of algorithms. This will allow SPTPS to be built with a few embedded crypto algorithms instead of linking with huge crypto libraries. The API is also kept simple. There is a start and a stop function. All data necessary to make the connection work is passed in the start function. Instead having both send- and receive-record functions, there is a send-record function and a receive-data function. The latter will pass protocol data received from the peer to the SPTPS implementation, which will in turn call a receive-record callback function when necessary. This hides all the handshaking from the application, and is completely independent from any event loop or socket characteristics. [0] http://iang.org/ssl/hn_hypotheses_in_secure_protocol_design.html
2011-07-24 13:44:51 +00:00
}
// Allocate memory for key material
size_t keylen = 2 * CHACHA_POLY1305_KEYLEN;
Start of "Simple Peer-To-Peer Security" protocol. Encryption and authentication of the meta connection is spread out over meta.c and protocol_auth.c. The new protocol was added there as well, leading to spaghetti code. To improve things, the new protocol will now be implemented in sptps.[ch]. The goal is to have a very simplified version of TLS. There is a record layer, and there are only two record types: application data and handshake messages. The handshake message contains a random nonce, an ephemeral ECDH public key, and an ECDSA signature over the former. After the ECDH public keys are exchanged, a shared secret is calculated, and a TLS style PRF is used to generate the key material for the cipher and HMAC algorithm, and further communication is encrypted and authenticated. A lot of the simplicity comes from the fact that both sides must have each other's public keys in advance, and there are no options to choose. There will be one fixed cipher suite, and both peers always authenticate each other. (Inspiration taken from Ian Grigg's hypotheses[0].) There might be some compromise in the future, to enable or disable encryption, authentication and compression, but there will be no choice of algorithms. This will allow SPTPS to be built with a few embedded crypto algorithms instead of linking with huge crypto libraries. The API is also kept simple. There is a start and a stop function. All data necessary to make the connection work is passed in the start function. Instead having both send- and receive-record functions, there is a send-record function and a receive-data function. The latter will pass protocol data received from the peer to the SPTPS implementation, which will in turn call a receive-record callback function when necessary. This hides all the handshaking from the application, and is completely independent from any event loop or socket characteristics. [0] http://iang.org/ssl/hn_hypotheses_in_secure_protocol_design.html
2011-07-24 13:44:51 +00:00
s->key = realloc(s->key, keylen);
if(!s->key)
return error(s, errno, strerror(errno));
// Create the HMAC seed, which is "key expansion" + session label + server nonce + client nonce
char seed[s->labellen + 64 + 13];
strcpy(seed, "key expansion");
if(s->initiator) {
memcpy(seed + 13, s->mykex + 1, 32);
memcpy(seed + 45, s->hiskex + 1, 32);
Start of "Simple Peer-To-Peer Security" protocol. Encryption and authentication of the meta connection is spread out over meta.c and protocol_auth.c. The new protocol was added there as well, leading to spaghetti code. To improve things, the new protocol will now be implemented in sptps.[ch]. The goal is to have a very simplified version of TLS. There is a record layer, and there are only two record types: application data and handshake messages. The handshake message contains a random nonce, an ephemeral ECDH public key, and an ECDSA signature over the former. After the ECDH public keys are exchanged, a shared secret is calculated, and a TLS style PRF is used to generate the key material for the cipher and HMAC algorithm, and further communication is encrypted and authenticated. A lot of the simplicity comes from the fact that both sides must have each other's public keys in advance, and there are no options to choose. There will be one fixed cipher suite, and both peers always authenticate each other. (Inspiration taken from Ian Grigg's hypotheses[0].) There might be some compromise in the future, to enable or disable encryption, authentication and compression, but there will be no choice of algorithms. This will allow SPTPS to be built with a few embedded crypto algorithms instead of linking with huge crypto libraries. The API is also kept simple. There is a start and a stop function. All data necessary to make the connection work is passed in the start function. Instead having both send- and receive-record functions, there is a send-record function and a receive-data function. The latter will pass protocol data received from the peer to the SPTPS implementation, which will in turn call a receive-record callback function when necessary. This hides all the handshaking from the application, and is completely independent from any event loop or socket characteristics. [0] http://iang.org/ssl/hn_hypotheses_in_secure_protocol_design.html
2011-07-24 13:44:51 +00:00
} else {
memcpy(seed + 13, s->hiskex + 1, 32);
memcpy(seed + 45, s->mykex + 1, 32);
Start of "Simple Peer-To-Peer Security" protocol. Encryption and authentication of the meta connection is spread out over meta.c and protocol_auth.c. The new protocol was added there as well, leading to spaghetti code. To improve things, the new protocol will now be implemented in sptps.[ch]. The goal is to have a very simplified version of TLS. There is a record layer, and there are only two record types: application data and handshake messages. The handshake message contains a random nonce, an ephemeral ECDH public key, and an ECDSA signature over the former. After the ECDH public keys are exchanged, a shared secret is calculated, and a TLS style PRF is used to generate the key material for the cipher and HMAC algorithm, and further communication is encrypted and authenticated. A lot of the simplicity comes from the fact that both sides must have each other's public keys in advance, and there are no options to choose. There will be one fixed cipher suite, and both peers always authenticate each other. (Inspiration taken from Ian Grigg's hypotheses[0].) There might be some compromise in the future, to enable or disable encryption, authentication and compression, but there will be no choice of algorithms. This will allow SPTPS to be built with a few embedded crypto algorithms instead of linking with huge crypto libraries. The API is also kept simple. There is a start and a stop function. All data necessary to make the connection work is passed in the start function. Instead having both send- and receive-record functions, there is a send-record function and a receive-data function. The latter will pass protocol data received from the peer to the SPTPS implementation, which will in turn call a receive-record callback function when necessary. This hides all the handshaking from the application, and is completely independent from any event loop or socket characteristics. [0] http://iang.org/ssl/hn_hypotheses_in_secure_protocol_design.html
2011-07-24 13:44:51 +00:00
}
memcpy(seed + 77, s->label, s->labellen);
Start of "Simple Peer-To-Peer Security" protocol. Encryption and authentication of the meta connection is spread out over meta.c and protocol_auth.c. The new protocol was added there as well, leading to spaghetti code. To improve things, the new protocol will now be implemented in sptps.[ch]. The goal is to have a very simplified version of TLS. There is a record layer, and there are only two record types: application data and handshake messages. The handshake message contains a random nonce, an ephemeral ECDH public key, and an ECDSA signature over the former. After the ECDH public keys are exchanged, a shared secret is calculated, and a TLS style PRF is used to generate the key material for the cipher and HMAC algorithm, and further communication is encrypted and authenticated. A lot of the simplicity comes from the fact that both sides must have each other's public keys in advance, and there are no options to choose. There will be one fixed cipher suite, and both peers always authenticate each other. (Inspiration taken from Ian Grigg's hypotheses[0].) There might be some compromise in the future, to enable or disable encryption, authentication and compression, but there will be no choice of algorithms. This will allow SPTPS to be built with a few embedded crypto algorithms instead of linking with huge crypto libraries. The API is also kept simple. There is a start and a stop function. All data necessary to make the connection work is passed in the start function. Instead having both send- and receive-record functions, there is a send-record function and a receive-data function. The latter will pass protocol data received from the peer to the SPTPS implementation, which will in turn call a receive-record callback function when necessary. This hides all the handshaking from the application, and is completely independent from any event loop or socket characteristics. [0] http://iang.org/ssl/hn_hypotheses_in_secure_protocol_design.html
2011-07-24 13:44:51 +00:00
// Use PRF to generate the key material
if(!prf(shared, len, seed, s->labellen + 64 + 13, s->key, keylen))
return error(s, EINVAL, "Failed to generate key material");
Start of "Simple Peer-To-Peer Security" protocol. Encryption and authentication of the meta connection is spread out over meta.c and protocol_auth.c. The new protocol was added there as well, leading to spaghetti code. To improve things, the new protocol will now be implemented in sptps.[ch]. The goal is to have a very simplified version of TLS. There is a record layer, and there are only two record types: application data and handshake messages. The handshake message contains a random nonce, an ephemeral ECDH public key, and an ECDSA signature over the former. After the ECDH public keys are exchanged, a shared secret is calculated, and a TLS style PRF is used to generate the key material for the cipher and HMAC algorithm, and further communication is encrypted and authenticated. A lot of the simplicity comes from the fact that both sides must have each other's public keys in advance, and there are no options to choose. There will be one fixed cipher suite, and both peers always authenticate each other. (Inspiration taken from Ian Grigg's hypotheses[0].) There might be some compromise in the future, to enable or disable encryption, authentication and compression, but there will be no choice of algorithms. This will allow SPTPS to be built with a few embedded crypto algorithms instead of linking with huge crypto libraries. The API is also kept simple. There is a start and a stop function. All data necessary to make the connection work is passed in the start function. Instead having both send- and receive-record functions, there is a send-record function and a receive-data function. The latter will pass protocol data received from the peer to the SPTPS implementation, which will in turn call a receive-record callback function when necessary. This hides all the handshaking from the application, and is completely independent from any event loop or socket characteristics. [0] http://iang.org/ssl/hn_hypotheses_in_secure_protocol_design.html
2011-07-24 13:44:51 +00:00
return true;
}
// Send an ACKnowledgement record.
Start of "Simple Peer-To-Peer Security" protocol. Encryption and authentication of the meta connection is spread out over meta.c and protocol_auth.c. The new protocol was added there as well, leading to spaghetti code. To improve things, the new protocol will now be implemented in sptps.[ch]. The goal is to have a very simplified version of TLS. There is a record layer, and there are only two record types: application data and handshake messages. The handshake message contains a random nonce, an ephemeral ECDH public key, and an ECDSA signature over the former. After the ECDH public keys are exchanged, a shared secret is calculated, and a TLS style PRF is used to generate the key material for the cipher and HMAC algorithm, and further communication is encrypted and authenticated. A lot of the simplicity comes from the fact that both sides must have each other's public keys in advance, and there are no options to choose. There will be one fixed cipher suite, and both peers always authenticate each other. (Inspiration taken from Ian Grigg's hypotheses[0].) There might be some compromise in the future, to enable or disable encryption, authentication and compression, but there will be no choice of algorithms. This will allow SPTPS to be built with a few embedded crypto algorithms instead of linking with huge crypto libraries. The API is also kept simple. There is a start and a stop function. All data necessary to make the connection work is passed in the start function. Instead having both send- and receive-record functions, there is a send-record function and a receive-data function. The latter will pass protocol data received from the peer to the SPTPS implementation, which will in turn call a receive-record callback function when necessary. This hides all the handshaking from the application, and is completely independent from any event loop or socket characteristics. [0] http://iang.org/ssl/hn_hypotheses_in_secure_protocol_design.html
2011-07-24 13:44:51 +00:00
static bool send_ack(sptps_t *s) {
return send_record_priv(s, SPTPS_HANDSHAKE, "", 0);
Start of "Simple Peer-To-Peer Security" protocol. Encryption and authentication of the meta connection is spread out over meta.c and protocol_auth.c. The new protocol was added there as well, leading to spaghetti code. To improve things, the new protocol will now be implemented in sptps.[ch]. The goal is to have a very simplified version of TLS. There is a record layer, and there are only two record types: application data and handshake messages. The handshake message contains a random nonce, an ephemeral ECDH public key, and an ECDSA signature over the former. After the ECDH public keys are exchanged, a shared secret is calculated, and a TLS style PRF is used to generate the key material for the cipher and HMAC algorithm, and further communication is encrypted and authenticated. A lot of the simplicity comes from the fact that both sides must have each other's public keys in advance, and there are no options to choose. There will be one fixed cipher suite, and both peers always authenticate each other. (Inspiration taken from Ian Grigg's hypotheses[0].) There might be some compromise in the future, to enable or disable encryption, authentication and compression, but there will be no choice of algorithms. This will allow SPTPS to be built with a few embedded crypto algorithms instead of linking with huge crypto libraries. The API is also kept simple. There is a start and a stop function. All data necessary to make the connection work is passed in the start function. Instead having both send- and receive-record functions, there is a send-record function and a receive-data function. The latter will pass protocol data received from the peer to the SPTPS implementation, which will in turn call a receive-record callback function when necessary. This hides all the handshaking from the application, and is completely independent from any event loop or socket characteristics. [0] http://iang.org/ssl/hn_hypotheses_in_secure_protocol_design.html
2011-07-24 13:44:51 +00:00
}
// Receive an ACKnowledgement record.
Start of "Simple Peer-To-Peer Security" protocol. Encryption and authentication of the meta connection is spread out over meta.c and protocol_auth.c. The new protocol was added there as well, leading to spaghetti code. To improve things, the new protocol will now be implemented in sptps.[ch]. The goal is to have a very simplified version of TLS. There is a record layer, and there are only two record types: application data and handshake messages. The handshake message contains a random nonce, an ephemeral ECDH public key, and an ECDSA signature over the former. After the ECDH public keys are exchanged, a shared secret is calculated, and a TLS style PRF is used to generate the key material for the cipher and HMAC algorithm, and further communication is encrypted and authenticated. A lot of the simplicity comes from the fact that both sides must have each other's public keys in advance, and there are no options to choose. There will be one fixed cipher suite, and both peers always authenticate each other. (Inspiration taken from Ian Grigg's hypotheses[0].) There might be some compromise in the future, to enable or disable encryption, authentication and compression, but there will be no choice of algorithms. This will allow SPTPS to be built with a few embedded crypto algorithms instead of linking with huge crypto libraries. The API is also kept simple. There is a start and a stop function. All data necessary to make the connection work is passed in the start function. Instead having both send- and receive-record functions, there is a send-record function and a receive-data function. The latter will pass protocol data received from the peer to the SPTPS implementation, which will in turn call a receive-record callback function when necessary. This hides all the handshaking from the application, and is completely independent from any event loop or socket characteristics. [0] http://iang.org/ssl/hn_hypotheses_in_secure_protocol_design.html
2011-07-24 13:44:51 +00:00
static bool receive_ack(sptps_t *s, const char *data, uint16_t len) {
if(len)
return error(s, EIO, "Invalid ACK record length");
Start of "Simple Peer-To-Peer Security" protocol. Encryption and authentication of the meta connection is spread out over meta.c and protocol_auth.c. The new protocol was added there as well, leading to spaghetti code. To improve things, the new protocol will now be implemented in sptps.[ch]. The goal is to have a very simplified version of TLS. There is a record layer, and there are only two record types: application data and handshake messages. The handshake message contains a random nonce, an ephemeral ECDH public key, and an ECDSA signature over the former. After the ECDH public keys are exchanged, a shared secret is calculated, and a TLS style PRF is used to generate the key material for the cipher and HMAC algorithm, and further communication is encrypted and authenticated. A lot of the simplicity comes from the fact that both sides must have each other's public keys in advance, and there are no options to choose. There will be one fixed cipher suite, and both peers always authenticate each other. (Inspiration taken from Ian Grigg's hypotheses[0].) There might be some compromise in the future, to enable or disable encryption, authentication and compression, but there will be no choice of algorithms. This will allow SPTPS to be built with a few embedded crypto algorithms instead of linking with huge crypto libraries. The API is also kept simple. There is a start and a stop function. All data necessary to make the connection work is passed in the start function. Instead having both send- and receive-record functions, there is a send-record function and a receive-data function. The latter will pass protocol data received from the peer to the SPTPS implementation, which will in turn call a receive-record callback function when necessary. This hides all the handshaking from the application, and is completely independent from any event loop or socket characteristics. [0] http://iang.org/ssl/hn_hypotheses_in_secure_protocol_design.html
2011-07-24 13:44:51 +00:00
if(s->initiator) {
if(!chacha_poly1305_set_key(s->incipher, s->key))
return error(s, EINVAL, "Failed to set counter");
} else {
if(!chacha_poly1305_set_key(s->incipher, s->key + CHACHA_POLY1305_KEYLEN))
return error(s, EINVAL, "Failed to set counter");
}
free(s->key);
s->key = NULL;
s->instate = true;
return true;
Start of "Simple Peer-To-Peer Security" protocol. Encryption and authentication of the meta connection is spread out over meta.c and protocol_auth.c. The new protocol was added there as well, leading to spaghetti code. To improve things, the new protocol will now be implemented in sptps.[ch]. The goal is to have a very simplified version of TLS. There is a record layer, and there are only two record types: application data and handshake messages. The handshake message contains a random nonce, an ephemeral ECDH public key, and an ECDSA signature over the former. After the ECDH public keys are exchanged, a shared secret is calculated, and a TLS style PRF is used to generate the key material for the cipher and HMAC algorithm, and further communication is encrypted and authenticated. A lot of the simplicity comes from the fact that both sides must have each other's public keys in advance, and there are no options to choose. There will be one fixed cipher suite, and both peers always authenticate each other. (Inspiration taken from Ian Grigg's hypotheses[0].) There might be some compromise in the future, to enable or disable encryption, authentication and compression, but there will be no choice of algorithms. This will allow SPTPS to be built with a few embedded crypto algorithms instead of linking with huge crypto libraries. The API is also kept simple. There is a start and a stop function. All data necessary to make the connection work is passed in the start function. Instead having both send- and receive-record functions, there is a send-record function and a receive-data function. The latter will pass protocol data received from the peer to the SPTPS implementation, which will in turn call a receive-record callback function when necessary. This hides all the handshaking from the application, and is completely independent from any event loop or socket characteristics. [0] http://iang.org/ssl/hn_hypotheses_in_secure_protocol_design.html
2011-07-24 13:44:51 +00:00
}
// Receive a Key EXchange record, respond by sending a SIG record.
Start of "Simple Peer-To-Peer Security" protocol. Encryption and authentication of the meta connection is spread out over meta.c and protocol_auth.c. The new protocol was added there as well, leading to spaghetti code. To improve things, the new protocol will now be implemented in sptps.[ch]. The goal is to have a very simplified version of TLS. There is a record layer, and there are only two record types: application data and handshake messages. The handshake message contains a random nonce, an ephemeral ECDH public key, and an ECDSA signature over the former. After the ECDH public keys are exchanged, a shared secret is calculated, and a TLS style PRF is used to generate the key material for the cipher and HMAC algorithm, and further communication is encrypted and authenticated. A lot of the simplicity comes from the fact that both sides must have each other's public keys in advance, and there are no options to choose. There will be one fixed cipher suite, and both peers always authenticate each other. (Inspiration taken from Ian Grigg's hypotheses[0].) There might be some compromise in the future, to enable or disable encryption, authentication and compression, but there will be no choice of algorithms. This will allow SPTPS to be built with a few embedded crypto algorithms instead of linking with huge crypto libraries. The API is also kept simple. There is a start and a stop function. All data necessary to make the connection work is passed in the start function. Instead having both send- and receive-record functions, there is a send-record function and a receive-data function. The latter will pass protocol data received from the peer to the SPTPS implementation, which will in turn call a receive-record callback function when necessary. This hides all the handshaking from the application, and is completely independent from any event loop or socket characteristics. [0] http://iang.org/ssl/hn_hypotheses_in_secure_protocol_design.html
2011-07-24 13:44:51 +00:00
static bool receive_kex(sptps_t *s, const char *data, uint16_t len) {
// Verify length of the HELLO record
if(len != 1 + 32 + ECDH_SIZE)
return error(s, EIO, "Invalid KEX record length");
// Ignore version number for now.
// Make a copy of the KEX message, send_sig() and receive_sig() need it
if(s->hiskex)
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));
memcpy(s->hiskex, data, len);
return send_sig(s);
}
// Receive a SIGnature record, verify it, if it passed, compute the shared secret and calculate the session keys.
static bool receive_sig(sptps_t *s, const char *data, uint16_t len) {
Start of "Simple Peer-To-Peer Security" protocol. Encryption and authentication of the meta connection is spread out over meta.c and protocol_auth.c. The new protocol was added there as well, leading to spaghetti code. To improve things, the new protocol will now be implemented in sptps.[ch]. The goal is to have a very simplified version of TLS. There is a record layer, and there are only two record types: application data and handshake messages. The handshake message contains a random nonce, an ephemeral ECDH public key, and an ECDSA signature over the former. After the ECDH public keys are exchanged, a shared secret is calculated, and a TLS style PRF is used to generate the key material for the cipher and HMAC algorithm, and further communication is encrypted and authenticated. A lot of the simplicity comes from the fact that both sides must have each other's public keys in advance, and there are no options to choose. There will be one fixed cipher suite, and both peers always authenticate each other. (Inspiration taken from Ian Grigg's hypotheses[0].) There might be some compromise in the future, to enable or disable encryption, authentication and compression, but there will be no choice of algorithms. This will allow SPTPS to be built with a few embedded crypto algorithms instead of linking with huge crypto libraries. The API is also kept simple. There is a start and a stop function. All data necessary to make the connection work is passed in the start function. Instead having both send- and receive-record functions, there is a send-record function and a receive-data function. The latter will pass protocol data received from the peer to the SPTPS implementation, which will in turn call a receive-record callback function when necessary. This hides all the handshaking from the application, and is completely independent from any event loop or socket characteristics. [0] http://iang.org/ssl/hn_hypotheses_in_secure_protocol_design.html
2011-07-24 13:44:51 +00:00
size_t keylen = ECDH_SIZE;
size_t siglen = ecdsa_size(s->hiskey);
Start of "Simple Peer-To-Peer Security" protocol. Encryption and authentication of the meta connection is spread out over meta.c and protocol_auth.c. The new protocol was added there as well, leading to spaghetti code. To improve things, the new protocol will now be implemented in sptps.[ch]. The goal is to have a very simplified version of TLS. There is a record layer, and there are only two record types: application data and handshake messages. The handshake message contains a random nonce, an ephemeral ECDH public key, and an ECDSA signature over the former. After the ECDH public keys are exchanged, a shared secret is calculated, and a TLS style PRF is used to generate the key material for the cipher and HMAC algorithm, and further communication is encrypted and authenticated. A lot of the simplicity comes from the fact that both sides must have each other's public keys in advance, and there are no options to choose. There will be one fixed cipher suite, and both peers always authenticate each other. (Inspiration taken from Ian Grigg's hypotheses[0].) There might be some compromise in the future, to enable or disable encryption, authentication and compression, but there will be no choice of algorithms. This will allow SPTPS to be built with a few embedded crypto algorithms instead of linking with huge crypto libraries. The API is also kept simple. There is a start and a stop function. All data necessary to make the connection work is passed in the start function. Instead having both send- and receive-record functions, there is a send-record function and a receive-data function. The latter will pass protocol data received from the peer to the SPTPS implementation, which will in turn call a receive-record callback function when necessary. This hides all the handshaking from the application, and is completely independent from any event loop or socket characteristics. [0] http://iang.org/ssl/hn_hypotheses_in_secure_protocol_design.html
2011-07-24 13:44:51 +00:00
// Verify length of KEX record.
if(len != siglen)
Start of "Simple Peer-To-Peer Security" protocol. Encryption and authentication of the meta connection is spread out over meta.c and protocol_auth.c. The new protocol was added there as well, leading to spaghetti code. To improve things, the new protocol will now be implemented in sptps.[ch]. The goal is to have a very simplified version of TLS. There is a record layer, and there are only two record types: application data and handshake messages. The handshake message contains a random nonce, an ephemeral ECDH public key, and an ECDSA signature over the former. After the ECDH public keys are exchanged, a shared secret is calculated, and a TLS style PRF is used to generate the key material for the cipher and HMAC algorithm, and further communication is encrypted and authenticated. A lot of the simplicity comes from the fact that both sides must have each other's public keys in advance, and there are no options to choose. There will be one fixed cipher suite, and both peers always authenticate each other. (Inspiration taken from Ian Grigg's hypotheses[0].) There might be some compromise in the future, to enable or disable encryption, authentication and compression, but there will be no choice of algorithms. This will allow SPTPS to be built with a few embedded crypto algorithms instead of linking with huge crypto libraries. The API is also kept simple. There is a start and a stop function. All data necessary to make the connection work is passed in the start function. Instead having both send- and receive-record functions, there is a send-record function and a receive-data function. The latter will pass protocol data received from the peer to the SPTPS implementation, which will in turn call a receive-record callback function when necessary. This hides all the handshaking from the application, and is completely independent from any event loop or socket characteristics. [0] http://iang.org/ssl/hn_hypotheses_in_secure_protocol_design.html
2011-07-24 13:44:51 +00:00
return error(s, EIO, "Invalid KEX record length");
// Concatenate both KEX messages, plus tag indicating if it is from the connection originator
char msg[(1 + 32 + keylen) * 2 + 1 + s->labellen];
msg[0] = !s->initiator;
memcpy(msg + 1, s->hiskex, 1 + 32 + keylen);
memcpy(msg + 1 + 33 + keylen, s->mykex, 1 + 32 + keylen);
memcpy(msg + 1 + 2 * (33 + keylen), s->label, s->labellen);
Start of "Simple Peer-To-Peer Security" protocol. Encryption and authentication of the meta connection is spread out over meta.c and protocol_auth.c. The new protocol was added there as well, leading to spaghetti code. To improve things, the new protocol will now be implemented in sptps.[ch]. The goal is to have a very simplified version of TLS. There is a record layer, and there are only two record types: application data and handshake messages. The handshake message contains a random nonce, an ephemeral ECDH public key, and an ECDSA signature over the former. After the ECDH public keys are exchanged, a shared secret is calculated, and a TLS style PRF is used to generate the key material for the cipher and HMAC algorithm, and further communication is encrypted and authenticated. A lot of the simplicity comes from the fact that both sides must have each other's public keys in advance, and there are no options to choose. There will be one fixed cipher suite, and both peers always authenticate each other. (Inspiration taken from Ian Grigg's hypotheses[0].) There might be some compromise in the future, to enable or disable encryption, authentication and compression, but there will be no choice of algorithms. This will allow SPTPS to be built with a few embedded crypto algorithms instead of linking with huge crypto libraries. The API is also kept simple. There is a start and a stop function. All data necessary to make the connection work is passed in the start function. Instead having both send- and receive-record functions, there is a send-record function and a receive-data function. The latter will pass protocol data received from the peer to the SPTPS implementation, which will in turn call a receive-record callback function when necessary. This hides all the handshaking from the application, and is completely independent from any event loop or socket characteristics. [0] http://iang.org/ssl/hn_hypotheses_in_secure_protocol_design.html
2011-07-24 13:44:51 +00:00
// Verify signature.
if(!ecdsa_verify(s->hiskey, msg, sizeof msg, data))
return error(s, EIO, "Failed to verify SIG record");
Start of "Simple Peer-To-Peer Security" protocol. Encryption and authentication of the meta connection is spread out over meta.c and protocol_auth.c. The new protocol was added there as well, leading to spaghetti code. To improve things, the new protocol will now be implemented in sptps.[ch]. The goal is to have a very simplified version of TLS. There is a record layer, and there are only two record types: application data and handshake messages. The handshake message contains a random nonce, an ephemeral ECDH public key, and an ECDSA signature over the former. After the ECDH public keys are exchanged, a shared secret is calculated, and a TLS style PRF is used to generate the key material for the cipher and HMAC algorithm, and further communication is encrypted and authenticated. A lot of the simplicity comes from the fact that both sides must have each other's public keys in advance, and there are no options to choose. There will be one fixed cipher suite, and both peers always authenticate each other. (Inspiration taken from Ian Grigg's hypotheses[0].) There might be some compromise in the future, to enable or disable encryption, authentication and compression, but there will be no choice of algorithms. This will allow SPTPS to be built with a few embedded crypto algorithms instead of linking with huge crypto libraries. The API is also kept simple. There is a start and a stop function. All data necessary to make the connection work is passed in the start function. Instead having both send- and receive-record functions, there is a send-record function and a receive-data function. The latter will pass protocol data received from the peer to the SPTPS implementation, which will in turn call a receive-record callback function when necessary. This hides all the handshaking from the application, and is completely independent from any event loop or socket characteristics. [0] http://iang.org/ssl/hn_hypotheses_in_secure_protocol_design.html
2011-07-24 13:44:51 +00:00
// Compute shared secret.
char shared[ECDH_SHARED_SIZE];
if(!ecdh_compute_shared(s->ecdh, s->hiskex + 1 + 32, shared))
return error(s, EINVAL, "Failed to compute ECDH shared secret");
s->ecdh = NULL;
Start of "Simple Peer-To-Peer Security" protocol. Encryption and authentication of the meta connection is spread out over meta.c and protocol_auth.c. The new protocol was added there as well, leading to spaghetti code. To improve things, the new protocol will now be implemented in sptps.[ch]. The goal is to have a very simplified version of TLS. There is a record layer, and there are only two record types: application data and handshake messages. The handshake message contains a random nonce, an ephemeral ECDH public key, and an ECDSA signature over the former. After the ECDH public keys are exchanged, a shared secret is calculated, and a TLS style PRF is used to generate the key material for the cipher and HMAC algorithm, and further communication is encrypted and authenticated. A lot of the simplicity comes from the fact that both sides must have each other's public keys in advance, and there are no options to choose. There will be one fixed cipher suite, and both peers always authenticate each other. (Inspiration taken from Ian Grigg's hypotheses[0].) There might be some compromise in the future, to enable or disable encryption, authentication and compression, but there will be no choice of algorithms. This will allow SPTPS to be built with a few embedded crypto algorithms instead of linking with huge crypto libraries. The API is also kept simple. There is a start and a stop function. All data necessary to make the connection work is passed in the start function. Instead having both send- and receive-record functions, there is a send-record function and a receive-data function. The latter will pass protocol data received from the peer to the SPTPS implementation, which will in turn call a receive-record callback function when necessary. This hides all the handshaking from the application, and is completely independent from any event loop or socket characteristics. [0] http://iang.org/ssl/hn_hypotheses_in_secure_protocol_design.html
2011-07-24 13:44:51 +00:00
// Generate key material from shared secret.
if(!generate_key_material(s, shared, sizeof shared))
Start of "Simple Peer-To-Peer Security" protocol. Encryption and authentication of the meta connection is spread out over meta.c and protocol_auth.c. The new protocol was added there as well, leading to spaghetti code. To improve things, the new protocol will now be implemented in sptps.[ch]. The goal is to have a very simplified version of TLS. There is a record layer, and there are only two record types: application data and handshake messages. The handshake message contains a random nonce, an ephemeral ECDH public key, and an ECDSA signature over the former. After the ECDH public keys are exchanged, a shared secret is calculated, and a TLS style PRF is used to generate the key material for the cipher and HMAC algorithm, and further communication is encrypted and authenticated. A lot of the simplicity comes from the fact that both sides must have each other's public keys in advance, and there are no options to choose. There will be one fixed cipher suite, and both peers always authenticate each other. (Inspiration taken from Ian Grigg's hypotheses[0].) There might be some compromise in the future, to enable or disable encryption, authentication and compression, but there will be no choice of algorithms. This will allow SPTPS to be built with a few embedded crypto algorithms instead of linking with huge crypto libraries. The API is also kept simple. There is a start and a stop function. All data necessary to make the connection work is passed in the start function. Instead having both send- and receive-record functions, there is a send-record function and a receive-data function. The latter will pass protocol data received from the peer to the SPTPS implementation, which will in turn call a receive-record callback function when necessary. This hides all the handshaking from the application, and is completely independent from any event loop or socket characteristics. [0] http://iang.org/ssl/hn_hypotheses_in_secure_protocol_design.html
2011-07-24 13:44:51 +00:00
return false;
free(s->mykex);
free(s->hiskex);
s->mykex = NULL;
s->hiskex = NULL;
// Send cipher change record
if(s->outstate && !send_ack(s))
return false;
Start of "Simple Peer-To-Peer Security" protocol. Encryption and authentication of the meta connection is spread out over meta.c and protocol_auth.c. The new protocol was added there as well, leading to spaghetti code. To improve things, the new protocol will now be implemented in sptps.[ch]. The goal is to have a very simplified version of TLS. There is a record layer, and there are only two record types: application data and handshake messages. The handshake message contains a random nonce, an ephemeral ECDH public key, and an ECDSA signature over the former. After the ECDH public keys are exchanged, a shared secret is calculated, and a TLS style PRF is used to generate the key material for the cipher and HMAC algorithm, and further communication is encrypted and authenticated. A lot of the simplicity comes from the fact that both sides must have each other's public keys in advance, and there are no options to choose. There will be one fixed cipher suite, and both peers always authenticate each other. (Inspiration taken from Ian Grigg's hypotheses[0].) There might be some compromise in the future, to enable or disable encryption, authentication and compression, but there will be no choice of algorithms. This will allow SPTPS to be built with a few embedded crypto algorithms instead of linking with huge crypto libraries. The API is also kept simple. There is a start and a stop function. All data necessary to make the connection work is passed in the start function. Instead having both send- and receive-record functions, there is a send-record function and a receive-data function. The latter will pass protocol data received from the peer to the SPTPS implementation, which will in turn call a receive-record callback function when necessary. This hides all the handshaking from the application, and is completely independent from any event loop or socket characteristics. [0] http://iang.org/ssl/hn_hypotheses_in_secure_protocol_design.html
2011-07-24 13:44:51 +00:00
// TODO: only set new keys after ACK has been set/received
Start of "Simple Peer-To-Peer Security" protocol. Encryption and authentication of the meta connection is spread out over meta.c and protocol_auth.c. The new protocol was added there as well, leading to spaghetti code. To improve things, the new protocol will now be implemented in sptps.[ch]. The goal is to have a very simplified version of TLS. There is a record layer, and there are only two record types: application data and handshake messages. The handshake message contains a random nonce, an ephemeral ECDH public key, and an ECDSA signature over the former. After the ECDH public keys are exchanged, a shared secret is calculated, and a TLS style PRF is used to generate the key material for the cipher and HMAC algorithm, and further communication is encrypted and authenticated. A lot of the simplicity comes from the fact that both sides must have each other's public keys in advance, and there are no options to choose. There will be one fixed cipher suite, and both peers always authenticate each other. (Inspiration taken from Ian Grigg's hypotheses[0].) There might be some compromise in the future, to enable or disable encryption, authentication and compression, but there will be no choice of algorithms. This will allow SPTPS to be built with a few embedded crypto algorithms instead of linking with huge crypto libraries. The API is also kept simple. There is a start and a stop function. All data necessary to make the connection work is passed in the start function. Instead having both send- and receive-record functions, there is a send-record function and a receive-data function. The latter will pass protocol data received from the peer to the SPTPS implementation, which will in turn call a receive-record callback function when necessary. This hides all the handshaking from the application, and is completely independent from any event loop or socket characteristics. [0] http://iang.org/ssl/hn_hypotheses_in_secure_protocol_design.html
2011-07-24 13:44:51 +00:00
if(s->initiator) {
if(!chacha_poly1305_set_key(s->outcipher, s->key + CHACHA_POLY1305_KEYLEN))
return error(s, EINVAL, "Failed to set key");
Start of "Simple Peer-To-Peer Security" protocol. Encryption and authentication of the meta connection is spread out over meta.c and protocol_auth.c. The new protocol was added there as well, leading to spaghetti code. To improve things, the new protocol will now be implemented in sptps.[ch]. The goal is to have a very simplified version of TLS. There is a record layer, and there are only two record types: application data and handshake messages. The handshake message contains a random nonce, an ephemeral ECDH public key, and an ECDSA signature over the former. After the ECDH public keys are exchanged, a shared secret is calculated, and a TLS style PRF is used to generate the key material for the cipher and HMAC algorithm, and further communication is encrypted and authenticated. A lot of the simplicity comes from the fact that both sides must have each other's public keys in advance, and there are no options to choose. There will be one fixed cipher suite, and both peers always authenticate each other. (Inspiration taken from Ian Grigg's hypotheses[0].) There might be some compromise in the future, to enable or disable encryption, authentication and compression, but there will be no choice of algorithms. This will allow SPTPS to be built with a few embedded crypto algorithms instead of linking with huge crypto libraries. The API is also kept simple. There is a start and a stop function. All data necessary to make the connection work is passed in the start function. Instead having both send- and receive-record functions, there is a send-record function and a receive-data function. The latter will pass protocol data received from the peer to the SPTPS implementation, which will in turn call a receive-record callback function when necessary. This hides all the handshaking from the application, and is completely independent from any event loop or socket characteristics. [0] http://iang.org/ssl/hn_hypotheses_in_secure_protocol_design.html
2011-07-24 13:44:51 +00:00
} else {
if(!chacha_poly1305_set_key(s->outcipher, s->key))
return error(s, EINVAL, "Failed to set key");
Start of "Simple Peer-To-Peer Security" protocol. Encryption and authentication of the meta connection is spread out over meta.c and protocol_auth.c. The new protocol was added there as well, leading to spaghetti code. To improve things, the new protocol will now be implemented in sptps.[ch]. The goal is to have a very simplified version of TLS. There is a record layer, and there are only two record types: application data and handshake messages. The handshake message contains a random nonce, an ephemeral ECDH public key, and an ECDSA signature over the former. After the ECDH public keys are exchanged, a shared secret is calculated, and a TLS style PRF is used to generate the key material for the cipher and HMAC algorithm, and further communication is encrypted and authenticated. A lot of the simplicity comes from the fact that both sides must have each other's public keys in advance, and there are no options to choose. There will be one fixed cipher suite, and both peers always authenticate each other. (Inspiration taken from Ian Grigg's hypotheses[0].) There might be some compromise in the future, to enable or disable encryption, authentication and compression, but there will be no choice of algorithms. This will allow SPTPS to be built with a few embedded crypto algorithms instead of linking with huge crypto libraries. The API is also kept simple. There is a start and a stop function. All data necessary to make the connection work is passed in the start function. Instead having both send- and receive-record functions, there is a send-record function and a receive-data function. The latter will pass protocol data received from the peer to the SPTPS implementation, which will in turn call a receive-record callback function when necessary. This hides all the handshaking from the application, and is completely independent from any event loop or socket characteristics. [0] http://iang.org/ssl/hn_hypotheses_in_secure_protocol_design.html
2011-07-24 13:44:51 +00:00
}
return true;
}
// Force another Key EXchange (for testing purposes).
bool sptps_force_kex(sptps_t *s) {
if(!s->outstate || s->state != SPTPS_SECONDARY_KEX)
return error(s, EINVAL, "Cannot force KEX in current state");
s->state = SPTPS_KEX;
return send_kex(s);
}
// Receive a handshake record.
Start of "Simple Peer-To-Peer Security" protocol. Encryption and authentication of the meta connection is spread out over meta.c and protocol_auth.c. The new protocol was added there as well, leading to spaghetti code. To improve things, the new protocol will now be implemented in sptps.[ch]. The goal is to have a very simplified version of TLS. There is a record layer, and there are only two record types: application data and handshake messages. The handshake message contains a random nonce, an ephemeral ECDH public key, and an ECDSA signature over the former. After the ECDH public keys are exchanged, a shared secret is calculated, and a TLS style PRF is used to generate the key material for the cipher and HMAC algorithm, and further communication is encrypted and authenticated. A lot of the simplicity comes from the fact that both sides must have each other's public keys in advance, and there are no options to choose. There will be one fixed cipher suite, and both peers always authenticate each other. (Inspiration taken from Ian Grigg's hypotheses[0].) There might be some compromise in the future, to enable or disable encryption, authentication and compression, but there will be no choice of algorithms. This will allow SPTPS to be built with a few embedded crypto algorithms instead of linking with huge crypto libraries. The API is also kept simple. There is a start and a stop function. All data necessary to make the connection work is passed in the start function. Instead having both send- and receive-record functions, there is a send-record function and a receive-data function. The latter will pass protocol data received from the peer to the SPTPS implementation, which will in turn call a receive-record callback function when necessary. This hides all the handshaking from the application, and is completely independent from any event loop or socket characteristics. [0] http://iang.org/ssl/hn_hypotheses_in_secure_protocol_design.html
2011-07-24 13:44:51 +00:00
static bool receive_handshake(sptps_t *s, const char *data, uint16_t len) {
// Only a few states to deal with handshaking.
switch(s->state) {
case SPTPS_SECONDARY_KEX:
// We receive a secondary KEX request, first respond by sending our own.
if(!send_kex(s))
return false;
case SPTPS_KEX:
// We have sent our KEX request, we expect our peer to sent one as well.
Start of "Simple Peer-To-Peer Security" protocol. Encryption and authentication of the meta connection is spread out over meta.c and protocol_auth.c. The new protocol was added there as well, leading to spaghetti code. To improve things, the new protocol will now be implemented in sptps.[ch]. The goal is to have a very simplified version of TLS. There is a record layer, and there are only two record types: application data and handshake messages. The handshake message contains a random nonce, an ephemeral ECDH public key, and an ECDSA signature over the former. After the ECDH public keys are exchanged, a shared secret is calculated, and a TLS style PRF is used to generate the key material for the cipher and HMAC algorithm, and further communication is encrypted and authenticated. A lot of the simplicity comes from the fact that both sides must have each other's public keys in advance, and there are no options to choose. There will be one fixed cipher suite, and both peers always authenticate each other. (Inspiration taken from Ian Grigg's hypotheses[0].) There might be some compromise in the future, to enable or disable encryption, authentication and compression, but there will be no choice of algorithms. This will allow SPTPS to be built with a few embedded crypto algorithms instead of linking with huge crypto libraries. The API is also kept simple. There is a start and a stop function. All data necessary to make the connection work is passed in the start function. Instead having both send- and receive-record functions, there is a send-record function and a receive-data function. The latter will pass protocol data received from the peer to the SPTPS implementation, which will in turn call a receive-record callback function when necessary. This hides all the handshaking from the application, and is completely independent from any event loop or socket characteristics. [0] http://iang.org/ssl/hn_hypotheses_in_secure_protocol_design.html
2011-07-24 13:44:51 +00:00
if(!receive_kex(s, data, len))
return false;
s->state = SPTPS_SIG;
Start of "Simple Peer-To-Peer Security" protocol. Encryption and authentication of the meta connection is spread out over meta.c and protocol_auth.c. The new protocol was added there as well, leading to spaghetti code. To improve things, the new protocol will now be implemented in sptps.[ch]. The goal is to have a very simplified version of TLS. There is a record layer, and there are only two record types: application data and handshake messages. The handshake message contains a random nonce, an ephemeral ECDH public key, and an ECDSA signature over the former. After the ECDH public keys are exchanged, a shared secret is calculated, and a TLS style PRF is used to generate the key material for the cipher and HMAC algorithm, and further communication is encrypted and authenticated. A lot of the simplicity comes from the fact that both sides must have each other's public keys in advance, and there are no options to choose. There will be one fixed cipher suite, and both peers always authenticate each other. (Inspiration taken from Ian Grigg's hypotheses[0].) There might be some compromise in the future, to enable or disable encryption, authentication and compression, but there will be no choice of algorithms. This will allow SPTPS to be built with a few embedded crypto algorithms instead of linking with huge crypto libraries. The API is also kept simple. There is a start and a stop function. All data necessary to make the connection work is passed in the start function. Instead having both send- and receive-record functions, there is a send-record function and a receive-data function. The latter will pass protocol data received from the peer to the SPTPS implementation, which will in turn call a receive-record callback function when necessary. This hides all the handshaking from the application, and is completely independent from any event loop or socket characteristics. [0] http://iang.org/ssl/hn_hypotheses_in_secure_protocol_design.html
2011-07-24 13:44:51 +00:00
return true;
case SPTPS_SIG:
Start of "Simple Peer-To-Peer Security" protocol. Encryption and authentication of the meta connection is spread out over meta.c and protocol_auth.c. The new protocol was added there as well, leading to spaghetti code. To improve things, the new protocol will now be implemented in sptps.[ch]. The goal is to have a very simplified version of TLS. There is a record layer, and there are only two record types: application data and handshake messages. The handshake message contains a random nonce, an ephemeral ECDH public key, and an ECDSA signature over the former. After the ECDH public keys are exchanged, a shared secret is calculated, and a TLS style PRF is used to generate the key material for the cipher and HMAC algorithm, and further communication is encrypted and authenticated. A lot of the simplicity comes from the fact that both sides must have each other's public keys in advance, and there are no options to choose. There will be one fixed cipher suite, and both peers always authenticate each other. (Inspiration taken from Ian Grigg's hypotheses[0].) There might be some compromise in the future, to enable or disable encryption, authentication and compression, but there will be no choice of algorithms. This will allow SPTPS to be built with a few embedded crypto algorithms instead of linking with huge crypto libraries. The API is also kept simple. There is a start and a stop function. All data necessary to make the connection work is passed in the start function. Instead having both send- and receive-record functions, there is a send-record function and a receive-data function. The latter will pass protocol data received from the peer to the SPTPS implementation, which will in turn call a receive-record callback function when necessary. This hides all the handshaking from the application, and is completely independent from any event loop or socket characteristics. [0] http://iang.org/ssl/hn_hypotheses_in_secure_protocol_design.html
2011-07-24 13:44:51 +00:00
// If we already sent our secondary public ECDH key, we expect the peer to send his.
if(!receive_sig(s, data, len))
Start of "Simple Peer-To-Peer Security" protocol. Encryption and authentication of the meta connection is spread out over meta.c and protocol_auth.c. The new protocol was added there as well, leading to spaghetti code. To improve things, the new protocol will now be implemented in sptps.[ch]. The goal is to have a very simplified version of TLS. There is a record layer, and there are only two record types: application data and handshake messages. The handshake message contains a random nonce, an ephemeral ECDH public key, and an ECDSA signature over the former. After the ECDH public keys are exchanged, a shared secret is calculated, and a TLS style PRF is used to generate the key material for the cipher and HMAC algorithm, and further communication is encrypted and authenticated. A lot of the simplicity comes from the fact that both sides must have each other's public keys in advance, and there are no options to choose. There will be one fixed cipher suite, and both peers always authenticate each other. (Inspiration taken from Ian Grigg's hypotheses[0].) There might be some compromise in the future, to enable or disable encryption, authentication and compression, but there will be no choice of algorithms. This will allow SPTPS to be built with a few embedded crypto algorithms instead of linking with huge crypto libraries. The API is also kept simple. There is a start and a stop function. All data necessary to make the connection work is passed in the start function. Instead having both send- and receive-record functions, there is a send-record function and a receive-data function. The latter will pass protocol data received from the peer to the SPTPS implementation, which will in turn call a receive-record callback function when necessary. This hides all the handshaking from the application, and is completely independent from any event loop or socket characteristics. [0] http://iang.org/ssl/hn_hypotheses_in_secure_protocol_design.html
2011-07-24 13:44:51 +00:00
return false;
if(s->outstate)
s->state = SPTPS_ACK;
else {
s->outstate = true;
if(!receive_ack(s, NULL, 0))
return false;
s->receive_record(s->handle, SPTPS_HANDSHAKE, NULL, 0);
s->state = SPTPS_SECONDARY_KEX;
}
Start of "Simple Peer-To-Peer Security" protocol. Encryption and authentication of the meta connection is spread out over meta.c and protocol_auth.c. The new protocol was added there as well, leading to spaghetti code. To improve things, the new protocol will now be implemented in sptps.[ch]. The goal is to have a very simplified version of TLS. There is a record layer, and there are only two record types: application data and handshake messages. The handshake message contains a random nonce, an ephemeral ECDH public key, and an ECDSA signature over the former. After the ECDH public keys are exchanged, a shared secret is calculated, and a TLS style PRF is used to generate the key material for the cipher and HMAC algorithm, and further communication is encrypted and authenticated. A lot of the simplicity comes from the fact that both sides must have each other's public keys in advance, and there are no options to choose. There will be one fixed cipher suite, and both peers always authenticate each other. (Inspiration taken from Ian Grigg's hypotheses[0].) There might be some compromise in the future, to enable or disable encryption, authentication and compression, but there will be no choice of algorithms. This will allow SPTPS to be built with a few embedded crypto algorithms instead of linking with huge crypto libraries. The API is also kept simple. There is a start and a stop function. All data necessary to make the connection work is passed in the start function. Instead having both send- and receive-record functions, there is a send-record function and a receive-data function. The latter will pass protocol data received from the peer to the SPTPS implementation, which will in turn call a receive-record callback function when necessary. This hides all the handshaking from the application, and is completely independent from any event loop or socket characteristics. [0] http://iang.org/ssl/hn_hypotheses_in_secure_protocol_design.html
2011-07-24 13:44:51 +00:00
return true;
case SPTPS_ACK:
// We expect a handshake message to indicate transition to the new keys.
Start of "Simple Peer-To-Peer Security" protocol. Encryption and authentication of the meta connection is spread out over meta.c and protocol_auth.c. The new protocol was added there as well, leading to spaghetti code. To improve things, the new protocol will now be implemented in sptps.[ch]. The goal is to have a very simplified version of TLS. There is a record layer, and there are only two record types: application data and handshake messages. The handshake message contains a random nonce, an ephemeral ECDH public key, and an ECDSA signature over the former. After the ECDH public keys are exchanged, a shared secret is calculated, and a TLS style PRF is used to generate the key material for the cipher and HMAC algorithm, and further communication is encrypted and authenticated. A lot of the simplicity comes from the fact that both sides must have each other's public keys in advance, and there are no options to choose. There will be one fixed cipher suite, and both peers always authenticate each other. (Inspiration taken from Ian Grigg's hypotheses[0].) There might be some compromise in the future, to enable or disable encryption, authentication and compression, but there will be no choice of algorithms. This will allow SPTPS to be built with a few embedded crypto algorithms instead of linking with huge crypto libraries. The API is also kept simple. There is a start and a stop function. All data necessary to make the connection work is passed in the start function. Instead having both send- and receive-record functions, there is a send-record function and a receive-data function. The latter will pass protocol data received from the peer to the SPTPS implementation, which will in turn call a receive-record callback function when necessary. This hides all the handshaking from the application, and is completely independent from any event loop or socket characteristics. [0] http://iang.org/ssl/hn_hypotheses_in_secure_protocol_design.html
2011-07-24 13:44:51 +00:00
if(!receive_ack(s, data, len))
return false;
s->receive_record(s->handle, SPTPS_HANDSHAKE, NULL, 0);
s->state = SPTPS_SECONDARY_KEX;
Start of "Simple Peer-To-Peer Security" protocol. Encryption and authentication of the meta connection is spread out over meta.c and protocol_auth.c. The new protocol was added there as well, leading to spaghetti code. To improve things, the new protocol will now be implemented in sptps.[ch]. The goal is to have a very simplified version of TLS. There is a record layer, and there are only two record types: application data and handshake messages. The handshake message contains a random nonce, an ephemeral ECDH public key, and an ECDSA signature over the former. After the ECDH public keys are exchanged, a shared secret is calculated, and a TLS style PRF is used to generate the key material for the cipher and HMAC algorithm, and further communication is encrypted and authenticated. A lot of the simplicity comes from the fact that both sides must have each other's public keys in advance, and there are no options to choose. There will be one fixed cipher suite, and both peers always authenticate each other. (Inspiration taken from Ian Grigg's hypotheses[0].) There might be some compromise in the future, to enable or disable encryption, authentication and compression, but there will be no choice of algorithms. This will allow SPTPS to be built with a few embedded crypto algorithms instead of linking with huge crypto libraries. The API is also kept simple. There is a start and a stop function. All data necessary to make the connection work is passed in the start function. Instead having both send- and receive-record functions, there is a send-record function and a receive-data function. The latter will pass protocol data received from the peer to the SPTPS implementation, which will in turn call a receive-record callback function when necessary. This hides all the handshaking from the application, and is completely independent from any event loop or socket characteristics. [0] http://iang.org/ssl/hn_hypotheses_in_secure_protocol_design.html
2011-07-24 13:44:51 +00:00
return true;
// TODO: split ACK into a VERify and ACK?
Start of "Simple Peer-To-Peer Security" protocol. Encryption and authentication of the meta connection is spread out over meta.c and protocol_auth.c. The new protocol was added there as well, leading to spaghetti code. To improve things, the new protocol will now be implemented in sptps.[ch]. The goal is to have a very simplified version of TLS. There is a record layer, and there are only two record types: application data and handshake messages. The handshake message contains a random nonce, an ephemeral ECDH public key, and an ECDSA signature over the former. After the ECDH public keys are exchanged, a shared secret is calculated, and a TLS style PRF is used to generate the key material for the cipher and HMAC algorithm, and further communication is encrypted and authenticated. A lot of the simplicity comes from the fact that both sides must have each other's public keys in advance, and there are no options to choose. There will be one fixed cipher suite, and both peers always authenticate each other. (Inspiration taken from Ian Grigg's hypotheses[0].) There might be some compromise in the future, to enable or disable encryption, authentication and compression, but there will be no choice of algorithms. This will allow SPTPS to be built with a few embedded crypto algorithms instead of linking with huge crypto libraries. The API is also kept simple. There is a start and a stop function. All data necessary to make the connection work is passed in the start function. Instead having both send- and receive-record functions, there is a send-record function and a receive-data function. The latter will pass protocol data received from the peer to the SPTPS implementation, which will in turn call a receive-record callback function when necessary. This hides all the handshaking from the application, and is completely independent from any event loop or socket characteristics. [0] http://iang.org/ssl/hn_hypotheses_in_secure_protocol_design.html
2011-07-24 13:44:51 +00:00
default:
return error(s, EIO, "Invalid session state %d", s->state);
Start of "Simple Peer-To-Peer Security" protocol. Encryption and authentication of the meta connection is spread out over meta.c and protocol_auth.c. The new protocol was added there as well, leading to spaghetti code. To improve things, the new protocol will now be implemented in sptps.[ch]. The goal is to have a very simplified version of TLS. There is a record layer, and there are only two record types: application data and handshake messages. The handshake message contains a random nonce, an ephemeral ECDH public key, and an ECDSA signature over the former. After the ECDH public keys are exchanged, a shared secret is calculated, and a TLS style PRF is used to generate the key material for the cipher and HMAC algorithm, and further communication is encrypted and authenticated. A lot of the simplicity comes from the fact that both sides must have each other's public keys in advance, and there are no options to choose. There will be one fixed cipher suite, and both peers always authenticate each other. (Inspiration taken from Ian Grigg's hypotheses[0].) There might be some compromise in the future, to enable or disable encryption, authentication and compression, but there will be no choice of algorithms. This will allow SPTPS to be built with a few embedded crypto algorithms instead of linking with huge crypto libraries. The API is also kept simple. There is a start and a stop function. All data necessary to make the connection work is passed in the start function. Instead having both send- and receive-record functions, there is a send-record function and a receive-data function. The latter will pass protocol data received from the peer to the SPTPS implementation, which will in turn call a receive-record callback function when necessary. This hides all the handshaking from the application, and is completely independent from any event loop or socket characteristics. [0] http://iang.org/ssl/hn_hypotheses_in_secure_protocol_design.html
2011-07-24 13:44:51 +00:00
}
}
static bool sptps_check_seqno(sptps_t *s, uint32_t seqno, bool update_state) {
// Replay protection using a sliding window of configurable size.
// s->inseqno is expected sequence number
// seqno is received sequence number
// s->late[] is a circular buffer, a 1 bit means a packet has not been received yet
// The circular buffer contains bits for sequence numbers from s->inseqno - s->replaywin * 8 to (but excluding) s->inseqno.
if(s->replaywin) {
if(seqno != s->inseqno) {
if(seqno >= s->inseqno + s->replaywin * 8) {
// Prevent packets that jump far ahead of the queue from causing many others to be dropped.
bool farfuture = s->farfuture < s->replaywin >> 2;
if (update_state)
s->farfuture++;
if(farfuture)
return update_state ? error(s, EIO, "Packet is %d seqs in the future, dropped (%u)\n", seqno - s->inseqno, s->farfuture) : false;
// Unless we have seen lots of them, in which case we consider the others lost.
if(update_state)
warning(s, "Lost %d packets\n", seqno - s->inseqno);
if (update_state) {
// Mark all packets in the replay window as being late.
memset(s->late, 255, s->replaywin);
}
} else if (seqno < s->inseqno) {
// If the sequence number is farther in the past than the bitmap goes, or if the packet was already received, drop it.
if((s->inseqno >= s->replaywin * 8 && seqno < s->inseqno - s->replaywin * 8) || !(s->late[(seqno / 8) % s->replaywin] & (1 << seqno % 8)))
return update_state ? error(s, EIO, "Received late or replayed packet, seqno %d, last received %d\n", seqno, s->inseqno) : false;
} else if (update_state) {
// We missed some packets. Mark them in the bitmap as being late.
for(int i = s->inseqno; i < seqno; i++)
s->late[(i / 8) % s->replaywin] |= 1 << i % 8;
}
}
if (update_state) {
// Mark the current packet as not being late.
s->late[(seqno / 8) % s->replaywin] &= ~(1 << seqno % 8);
s->farfuture = 0;
}
}
if (update_state) {
if(seqno >= s->inseqno)
s->inseqno = seqno + 1;
if(!s->inseqno)
s->received = 0;
else
s->received++;
}
return true;
}
2012-07-31 18:36:35 +00:00
// Check datagram for valid HMAC
bool sptps_verify_datagram(sptps_t *s, const void *data, size_t len) {
2012-07-31 18:36:35 +00:00
if(!s->instate || len < 21)
return error(s, EIO, "Received short packet");
2012-07-31 18:36:35 +00:00
uint32_t seqno;
memcpy(&seqno, data, 4);
seqno = ntohl(seqno);
if (!sptps_check_seqno(s, seqno, false))
return false;
2012-07-31 18:36:35 +00:00
char buffer[len];
size_t outlen;
return chacha_poly1305_decrypt(s->incipher, seqno, data + 4, len - 4, buffer, &outlen);
2012-07-31 18:36:35 +00:00
}
// Receive incoming data, datagram version.
static bool sptps_receive_data_datagram(sptps_t *s, const char *data, size_t len) {
if(len < (s->instate ? 21 : 5))
return error(s, EIO, "Received short packet");
uint32_t seqno;
memcpy(&seqno, data, 4);
seqno = ntohl(seqno);
data += 4; len -= 4;
if(!s->instate) {
2012-10-14 13:37:24 +00:00
if(seqno != s->inseqno)
return error(s, EIO, "Invalid packet seqno: %d != %d", seqno, s->inseqno);
s->inseqno = seqno + 1;
uint8_t type = *(data++); len--;
if(type != SPTPS_HANDSHAKE)
return error(s, EIO, "Application record received before handshake finished");
return receive_handshake(s, data, len);
}
// Decrypt
2013-01-14 12:08:35 +00:00
char buffer[len];
size_t outlen;
if(!chacha_poly1305_decrypt(s->incipher, seqno, data, len, buffer, &outlen))
return error(s, EIO, "Failed to decrypt and verify packet");
2013-01-14 12:08:35 +00:00
if(!sptps_check_seqno(s, seqno, true))
return false;
// Append a NULL byte for safety.
buffer[outlen] = 0;
data = buffer;
len = outlen;
uint8_t type = *(data++); len--;
if(type < SPTPS_HANDSHAKE) {
if(!s->instate)
return error(s, EIO, "Application record received before handshake finished");
if(!s->receive_record(s->handle, type, data, len))
return false;
} else if(type == SPTPS_HANDSHAKE) {
if(!receive_handshake(s, data, len))
return false;
} else {
return error(s, EIO, "Invalid record type %d", type);
}
return true;
}
// Receive incoming data. Check if it contains a complete record, if so, handle it.
size_t sptps_receive_data(sptps_t *s, const void *data, size_t len) {
size_t total_read = 0;
if(!s->state)
return error(s, EIO, "Invalid session state zero");
if(s->datagram)
return sptps_receive_data_datagram(s, data, len) ? len : false;
// First read the 2 length bytes.
if(s->buflen < 2) {
size_t toread = 2 - s->buflen;
if(toread > len)
toread = len;
Start of "Simple Peer-To-Peer Security" protocol. Encryption and authentication of the meta connection is spread out over meta.c and protocol_auth.c. The new protocol was added there as well, leading to spaghetti code. To improve things, the new protocol will now be implemented in sptps.[ch]. The goal is to have a very simplified version of TLS. There is a record layer, and there are only two record types: application data and handshake messages. The handshake message contains a random nonce, an ephemeral ECDH public key, and an ECDSA signature over the former. After the ECDH public keys are exchanged, a shared secret is calculated, and a TLS style PRF is used to generate the key material for the cipher and HMAC algorithm, and further communication is encrypted and authenticated. A lot of the simplicity comes from the fact that both sides must have each other's public keys in advance, and there are no options to choose. There will be one fixed cipher suite, and both peers always authenticate each other. (Inspiration taken from Ian Grigg's hypotheses[0].) There might be some compromise in the future, to enable or disable encryption, authentication and compression, but there will be no choice of algorithms. This will allow SPTPS to be built with a few embedded crypto algorithms instead of linking with huge crypto libraries. The API is also kept simple. There is a start and a stop function. All data necessary to make the connection work is passed in the start function. Instead having both send- and receive-record functions, there is a send-record function and a receive-data function. The latter will pass protocol data received from the peer to the SPTPS implementation, which will in turn call a receive-record callback function when necessary. This hides all the handshaking from the application, and is completely independent from any event loop or socket characteristics. [0] http://iang.org/ssl/hn_hypotheses_in_secure_protocol_design.html
2011-07-24 13:44:51 +00:00
memcpy(s->inbuf + s->buflen, data, toread);
Start of "Simple Peer-To-Peer Security" protocol. Encryption and authentication of the meta connection is spread out over meta.c and protocol_auth.c. The new protocol was added there as well, leading to spaghetti code. To improve things, the new protocol will now be implemented in sptps.[ch]. The goal is to have a very simplified version of TLS. There is a record layer, and there are only two record types: application data and handshake messages. The handshake message contains a random nonce, an ephemeral ECDH public key, and an ECDSA signature over the former. After the ECDH public keys are exchanged, a shared secret is calculated, and a TLS style PRF is used to generate the key material for the cipher and HMAC algorithm, and further communication is encrypted and authenticated. A lot of the simplicity comes from the fact that both sides must have each other's public keys in advance, and there are no options to choose. There will be one fixed cipher suite, and both peers always authenticate each other. (Inspiration taken from Ian Grigg's hypotheses[0].) There might be some compromise in the future, to enable or disable encryption, authentication and compression, but there will be no choice of algorithms. This will allow SPTPS to be built with a few embedded crypto algorithms instead of linking with huge crypto libraries. The API is also kept simple. There is a start and a stop function. All data necessary to make the connection work is passed in the start function. Instead having both send- and receive-record functions, there is a send-record function and a receive-data function. The latter will pass protocol data received from the peer to the SPTPS implementation, which will in turn call a receive-record callback function when necessary. This hides all the handshaking from the application, and is completely independent from any event loop or socket characteristics. [0] http://iang.org/ssl/hn_hypotheses_in_secure_protocol_design.html
2011-07-24 13:44:51 +00:00
total_read += toread;
s->buflen += toread;
len -= toread;
data += toread;
2012-10-10 15:17:49 +00:00
// Exit early if we don't have the full length.
if(s->buflen < 2)
return total_read;
Start of "Simple Peer-To-Peer Security" protocol. Encryption and authentication of the meta connection is spread out over meta.c and protocol_auth.c. The new protocol was added there as well, leading to spaghetti code. To improve things, the new protocol will now be implemented in sptps.[ch]. The goal is to have a very simplified version of TLS. There is a record layer, and there are only two record types: application data and handshake messages. The handshake message contains a random nonce, an ephemeral ECDH public key, and an ECDSA signature over the former. After the ECDH public keys are exchanged, a shared secret is calculated, and a TLS style PRF is used to generate the key material for the cipher and HMAC algorithm, and further communication is encrypted and authenticated. A lot of the simplicity comes from the fact that both sides must have each other's public keys in advance, and there are no options to choose. There will be one fixed cipher suite, and both peers always authenticate each other. (Inspiration taken from Ian Grigg's hypotheses[0].) There might be some compromise in the future, to enable or disable encryption, authentication and compression, but there will be no choice of algorithms. This will allow SPTPS to be built with a few embedded crypto algorithms instead of linking with huge crypto libraries. The API is also kept simple. There is a start and a stop function. All data necessary to make the connection work is passed in the start function. Instead having both send- and receive-record functions, there is a send-record function and a receive-data function. The latter will pass protocol data received from the peer to the SPTPS implementation, which will in turn call a receive-record callback function when necessary. This hides all the handshaking from the application, and is completely independent from any event loop or socket characteristics. [0] http://iang.org/ssl/hn_hypotheses_in_secure_protocol_design.html
2011-07-24 13:44:51 +00:00
// Get the length bytes
2012-02-25 14:18:15 +00:00
memcpy(&s->reclen, s->inbuf, 2);
s->reclen = ntohs(s->reclen);
2012-02-25 14:18:15 +00:00
// If we have the length bytes, ensure our buffer can hold the whole request.
s->inbuf = realloc(s->inbuf, s->reclen + 19UL);
if(!s->inbuf)
return error(s, errno, strerror(errno));
Start of "Simple Peer-To-Peer Security" protocol. Encryption and authentication of the meta connection is spread out over meta.c and protocol_auth.c. The new protocol was added there as well, leading to spaghetti code. To improve things, the new protocol will now be implemented in sptps.[ch]. The goal is to have a very simplified version of TLS. There is a record layer, and there are only two record types: application data and handshake messages. The handshake message contains a random nonce, an ephemeral ECDH public key, and an ECDSA signature over the former. After the ECDH public keys are exchanged, a shared secret is calculated, and a TLS style PRF is used to generate the key material for the cipher and HMAC algorithm, and further communication is encrypted and authenticated. A lot of the simplicity comes from the fact that both sides must have each other's public keys in advance, and there are no options to choose. There will be one fixed cipher suite, and both peers always authenticate each other. (Inspiration taken from Ian Grigg's hypotheses[0].) There might be some compromise in the future, to enable or disable encryption, authentication and compression, but there will be no choice of algorithms. This will allow SPTPS to be built with a few embedded crypto algorithms instead of linking with huge crypto libraries. The API is also kept simple. There is a start and a stop function. All data necessary to make the connection work is passed in the start function. Instead having both send- and receive-record functions, there is a send-record function and a receive-data function. The latter will pass protocol data received from the peer to the SPTPS implementation, which will in turn call a receive-record callback function when necessary. This hides all the handshaking from the application, and is completely independent from any event loop or socket characteristics. [0] http://iang.org/ssl/hn_hypotheses_in_secure_protocol_design.html
2011-07-24 13:44:51 +00:00
// Exit early if we have no more data to process.
if(!len)
return total_read;
}
Start of "Simple Peer-To-Peer Security" protocol. Encryption and authentication of the meta connection is spread out over meta.c and protocol_auth.c. The new protocol was added there as well, leading to spaghetti code. To improve things, the new protocol will now be implemented in sptps.[ch]. The goal is to have a very simplified version of TLS. There is a record layer, and there are only two record types: application data and handshake messages. The handshake message contains a random nonce, an ephemeral ECDH public key, and an ECDSA signature over the former. After the ECDH public keys are exchanged, a shared secret is calculated, and a TLS style PRF is used to generate the key material for the cipher and HMAC algorithm, and further communication is encrypted and authenticated. A lot of the simplicity comes from the fact that both sides must have each other's public keys in advance, and there are no options to choose. There will be one fixed cipher suite, and both peers always authenticate each other. (Inspiration taken from Ian Grigg's hypotheses[0].) There might be some compromise in the future, to enable or disable encryption, authentication and compression, but there will be no choice of algorithms. This will allow SPTPS to be built with a few embedded crypto algorithms instead of linking with huge crypto libraries. The API is also kept simple. There is a start and a stop function. All data necessary to make the connection work is passed in the start function. Instead having both send- and receive-record functions, there is a send-record function and a receive-data function. The latter will pass protocol data received from the peer to the SPTPS implementation, which will in turn call a receive-record callback function when necessary. This hides all the handshaking from the application, and is completely independent from any event loop or socket characteristics. [0] http://iang.org/ssl/hn_hypotheses_in_secure_protocol_design.html
2011-07-24 13:44:51 +00:00
// Read up to the end of the record.
size_t toread = s->reclen + (s->instate ? 19UL : 3UL) - s->buflen;
if(toread > len)
toread = len;
Start of "Simple Peer-To-Peer Security" protocol. Encryption and authentication of the meta connection is spread out over meta.c and protocol_auth.c. The new protocol was added there as well, leading to spaghetti code. To improve things, the new protocol will now be implemented in sptps.[ch]. The goal is to have a very simplified version of TLS. There is a record layer, and there are only two record types: application data and handshake messages. The handshake message contains a random nonce, an ephemeral ECDH public key, and an ECDSA signature over the former. After the ECDH public keys are exchanged, a shared secret is calculated, and a TLS style PRF is used to generate the key material for the cipher and HMAC algorithm, and further communication is encrypted and authenticated. A lot of the simplicity comes from the fact that both sides must have each other's public keys in advance, and there are no options to choose. There will be one fixed cipher suite, and both peers always authenticate each other. (Inspiration taken from Ian Grigg's hypotheses[0].) There might be some compromise in the future, to enable or disable encryption, authentication and compression, but there will be no choice of algorithms. This will allow SPTPS to be built with a few embedded crypto algorithms instead of linking with huge crypto libraries. The API is also kept simple. There is a start and a stop function. All data necessary to make the connection work is passed in the start function. Instead having both send- and receive-record functions, there is a send-record function and a receive-data function. The latter will pass protocol data received from the peer to the SPTPS implementation, which will in turn call a receive-record callback function when necessary. This hides all the handshaking from the application, and is completely independent from any event loop or socket characteristics. [0] http://iang.org/ssl/hn_hypotheses_in_secure_protocol_design.html
2011-07-24 13:44:51 +00:00
memcpy(s->inbuf + s->buflen, data, toread);
total_read += toread;
s->buflen += toread;
len -= toread;
data += toread;
Start of "Simple Peer-To-Peer Security" protocol. Encryption and authentication of the meta connection is spread out over meta.c and protocol_auth.c. The new protocol was added there as well, leading to spaghetti code. To improve things, the new protocol will now be implemented in sptps.[ch]. The goal is to have a very simplified version of TLS. There is a record layer, and there are only two record types: application data and handshake messages. The handshake message contains a random nonce, an ephemeral ECDH public key, and an ECDSA signature over the former. After the ECDH public keys are exchanged, a shared secret is calculated, and a TLS style PRF is used to generate the key material for the cipher and HMAC algorithm, and further communication is encrypted and authenticated. A lot of the simplicity comes from the fact that both sides must have each other's public keys in advance, and there are no options to choose. There will be one fixed cipher suite, and both peers always authenticate each other. (Inspiration taken from Ian Grigg's hypotheses[0].) There might be some compromise in the future, to enable or disable encryption, authentication and compression, but there will be no choice of algorithms. This will allow SPTPS to be built with a few embedded crypto algorithms instead of linking with huge crypto libraries. The API is also kept simple. There is a start and a stop function. All data necessary to make the connection work is passed in the start function. Instead having both send- and receive-record functions, there is a send-record function and a receive-data function. The latter will pass protocol data received from the peer to the SPTPS implementation, which will in turn call a receive-record callback function when necessary. This hides all the handshaking from the application, and is completely independent from any event loop or socket characteristics. [0] http://iang.org/ssl/hn_hypotheses_in_secure_protocol_design.html
2011-07-24 13:44:51 +00:00
// If we don't have a whole record, exit.
if(s->buflen < s->reclen + (s->instate ? 19UL : 3UL))
return total_read;
Start of "Simple Peer-To-Peer Security" protocol. Encryption and authentication of the meta connection is spread out over meta.c and protocol_auth.c. The new protocol was added there as well, leading to spaghetti code. To improve things, the new protocol will now be implemented in sptps.[ch]. The goal is to have a very simplified version of TLS. There is a record layer, and there are only two record types: application data and handshake messages. The handshake message contains a random nonce, an ephemeral ECDH public key, and an ECDSA signature over the former. After the ECDH public keys are exchanged, a shared secret is calculated, and a TLS style PRF is used to generate the key material for the cipher and HMAC algorithm, and further communication is encrypted and authenticated. A lot of the simplicity comes from the fact that both sides must have each other's public keys in advance, and there are no options to choose. There will be one fixed cipher suite, and both peers always authenticate each other. (Inspiration taken from Ian Grigg's hypotheses[0].) There might be some compromise in the future, to enable or disable encryption, authentication and compression, but there will be no choice of algorithms. This will allow SPTPS to be built with a few embedded crypto algorithms instead of linking with huge crypto libraries. The API is also kept simple. There is a start and a stop function. All data necessary to make the connection work is passed in the start function. Instead having both send- and receive-record functions, there is a send-record function and a receive-data function. The latter will pass protocol data received from the peer to the SPTPS implementation, which will in turn call a receive-record callback function when necessary. This hides all the handshaking from the application, and is completely independent from any event loop or socket characteristics. [0] http://iang.org/ssl/hn_hypotheses_in_secure_protocol_design.html
2011-07-24 13:44:51 +00:00
// Update sequence number.
uint32_t seqno = s->inseqno++;
// Check HMAC and decrypt.
if(s->instate) {
if(!chacha_poly1305_decrypt(s->incipher, seqno, s->inbuf + 2UL, s->reclen + 17UL, s->inbuf + 2UL, NULL))
return error(s, EINVAL, "Failed to decrypt and verify record");
}
Start of "Simple Peer-To-Peer Security" protocol. Encryption and authentication of the meta connection is spread out over meta.c and protocol_auth.c. The new protocol was added there as well, leading to spaghetti code. To improve things, the new protocol will now be implemented in sptps.[ch]. The goal is to have a very simplified version of TLS. There is a record layer, and there are only two record types: application data and handshake messages. The handshake message contains a random nonce, an ephemeral ECDH public key, and an ECDSA signature over the former. After the ECDH public keys are exchanged, a shared secret is calculated, and a TLS style PRF is used to generate the key material for the cipher and HMAC algorithm, and further communication is encrypted and authenticated. A lot of the simplicity comes from the fact that both sides must have each other's public keys in advance, and there are no options to choose. There will be one fixed cipher suite, and both peers always authenticate each other. (Inspiration taken from Ian Grigg's hypotheses[0].) There might be some compromise in the future, to enable or disable encryption, authentication and compression, but there will be no choice of algorithms. This will allow SPTPS to be built with a few embedded crypto algorithms instead of linking with huge crypto libraries. The API is also kept simple. There is a start and a stop function. All data necessary to make the connection work is passed in the start function. Instead having both send- and receive-record functions, there is a send-record function and a receive-data function. The latter will pass protocol data received from the peer to the SPTPS implementation, which will in turn call a receive-record callback function when necessary. This hides all the handshaking from the application, and is completely independent from any event loop or socket characteristics. [0] http://iang.org/ssl/hn_hypotheses_in_secure_protocol_design.html
2011-07-24 13:44:51 +00:00
// Append a NULL byte for safety.
s->inbuf[s->reclen + 3UL] = 0;
Start of "Simple Peer-To-Peer Security" protocol. Encryption and authentication of the meta connection is spread out over meta.c and protocol_auth.c. The new protocol was added there as well, leading to spaghetti code. To improve things, the new protocol will now be implemented in sptps.[ch]. The goal is to have a very simplified version of TLS. There is a record layer, and there are only two record types: application data and handshake messages. The handshake message contains a random nonce, an ephemeral ECDH public key, and an ECDSA signature over the former. After the ECDH public keys are exchanged, a shared secret is calculated, and a TLS style PRF is used to generate the key material for the cipher and HMAC algorithm, and further communication is encrypted and authenticated. A lot of the simplicity comes from the fact that both sides must have each other's public keys in advance, and there are no options to choose. There will be one fixed cipher suite, and both peers always authenticate each other. (Inspiration taken from Ian Grigg's hypotheses[0].) There might be some compromise in the future, to enable or disable encryption, authentication and compression, but there will be no choice of algorithms. This will allow SPTPS to be built with a few embedded crypto algorithms instead of linking with huge crypto libraries. The API is also kept simple. There is a start and a stop function. All data necessary to make the connection work is passed in the start function. Instead having both send- and receive-record functions, there is a send-record function and a receive-data function. The latter will pass protocol data received from the peer to the SPTPS implementation, which will in turn call a receive-record callback function when necessary. This hides all the handshaking from the application, and is completely independent from any event loop or socket characteristics. [0] http://iang.org/ssl/hn_hypotheses_in_secure_protocol_design.html
2011-07-24 13:44:51 +00:00
uint8_t type = s->inbuf[2];
Start of "Simple Peer-To-Peer Security" protocol. Encryption and authentication of the meta connection is spread out over meta.c and protocol_auth.c. The new protocol was added there as well, leading to spaghetti code. To improve things, the new protocol will now be implemented in sptps.[ch]. The goal is to have a very simplified version of TLS. There is a record layer, and there are only two record types: application data and handshake messages. The handshake message contains a random nonce, an ephemeral ECDH public key, and an ECDSA signature over the former. After the ECDH public keys are exchanged, a shared secret is calculated, and a TLS style PRF is used to generate the key material for the cipher and HMAC algorithm, and further communication is encrypted and authenticated. A lot of the simplicity comes from the fact that both sides must have each other's public keys in advance, and there are no options to choose. There will be one fixed cipher suite, and both peers always authenticate each other. (Inspiration taken from Ian Grigg's hypotheses[0].) There might be some compromise in the future, to enable or disable encryption, authentication and compression, but there will be no choice of algorithms. This will allow SPTPS to be built with a few embedded crypto algorithms instead of linking with huge crypto libraries. The API is also kept simple. There is a start and a stop function. All data necessary to make the connection work is passed in the start function. Instead having both send- and receive-record functions, there is a send-record function and a receive-data function. The latter will pass protocol data received from the peer to the SPTPS implementation, which will in turn call a receive-record callback function when necessary. This hides all the handshaking from the application, and is completely independent from any event loop or socket characteristics. [0] http://iang.org/ssl/hn_hypotheses_in_secure_protocol_design.html
2011-07-24 13:44:51 +00:00
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 + 3, s->reclen))
return false;
} else if(type == SPTPS_HANDSHAKE) {
if(!receive_handshake(s, s->inbuf + 3, s->reclen))
return false;
} else {
return error(s, EIO, "Invalid record type %d", type);
Start of "Simple Peer-To-Peer Security" protocol. Encryption and authentication of the meta connection is spread out over meta.c and protocol_auth.c. The new protocol was added there as well, leading to spaghetti code. To improve things, the new protocol will now be implemented in sptps.[ch]. The goal is to have a very simplified version of TLS. There is a record layer, and there are only two record types: application data and handshake messages. The handshake message contains a random nonce, an ephemeral ECDH public key, and an ECDSA signature over the former. After the ECDH public keys are exchanged, a shared secret is calculated, and a TLS style PRF is used to generate the key material for the cipher and HMAC algorithm, and further communication is encrypted and authenticated. A lot of the simplicity comes from the fact that both sides must have each other's public keys in advance, and there are no options to choose. There will be one fixed cipher suite, and both peers always authenticate each other. (Inspiration taken from Ian Grigg's hypotheses[0].) There might be some compromise in the future, to enable or disable encryption, authentication and compression, but there will be no choice of algorithms. This will allow SPTPS to be built with a few embedded crypto algorithms instead of linking with huge crypto libraries. The API is also kept simple. There is a start and a stop function. All data necessary to make the connection work is passed in the start function. Instead having both send- and receive-record functions, there is a send-record function and a receive-data function. The latter will pass protocol data received from the peer to the SPTPS implementation, which will in turn call a receive-record callback function when necessary. This hides all the handshaking from the application, and is completely independent from any event loop or socket characteristics. [0] http://iang.org/ssl/hn_hypotheses_in_secure_protocol_design.html
2011-07-24 13:44:51 +00:00
}
s->buflen = 0;
return total_read;
Start of "Simple Peer-To-Peer Security" protocol. Encryption and authentication of the meta connection is spread out over meta.c and protocol_auth.c. The new protocol was added there as well, leading to spaghetti code. To improve things, the new protocol will now be implemented in sptps.[ch]. The goal is to have a very simplified version of TLS. There is a record layer, and there are only two record types: application data and handshake messages. The handshake message contains a random nonce, an ephemeral ECDH public key, and an ECDSA signature over the former. After the ECDH public keys are exchanged, a shared secret is calculated, and a TLS style PRF is used to generate the key material for the cipher and HMAC algorithm, and further communication is encrypted and authenticated. A lot of the simplicity comes from the fact that both sides must have each other's public keys in advance, and there are no options to choose. There will be one fixed cipher suite, and both peers always authenticate each other. (Inspiration taken from Ian Grigg's hypotheses[0].) There might be some compromise in the future, to enable or disable encryption, authentication and compression, but there will be no choice of algorithms. This will allow SPTPS to be built with a few embedded crypto algorithms instead of linking with huge crypto libraries. The API is also kept simple. There is a start and a stop function. All data necessary to make the connection work is passed in the start function. Instead having both send- and receive-record functions, there is a send-record function and a receive-data function. The latter will pass protocol data received from the peer to the SPTPS implementation, which will in turn call a receive-record callback function when necessary. This hides all the handshaking from the application, and is completely independent from any event loop or socket characteristics. [0] http://iang.org/ssl/hn_hypotheses_in_secure_protocol_design.html
2011-07-24 13:44:51 +00:00
}
// Start a SPTPS session.
bool sptps_start(sptps_t *s, void *handle, bool initiator, bool datagram, ecdsa_t *mykey, ecdsa_t *hiskey, const void *label, size_t labellen, send_data_t send_data, receive_record_t receive_record) {
Start of "Simple Peer-To-Peer Security" protocol. Encryption and authentication of the meta connection is spread out over meta.c and protocol_auth.c. The new protocol was added there as well, leading to spaghetti code. To improve things, the new protocol will now be implemented in sptps.[ch]. The goal is to have a very simplified version of TLS. There is a record layer, and there are only two record types: application data and handshake messages. The handshake message contains a random nonce, an ephemeral ECDH public key, and an ECDSA signature over the former. After the ECDH public keys are exchanged, a shared secret is calculated, and a TLS style PRF is used to generate the key material for the cipher and HMAC algorithm, and further communication is encrypted and authenticated. A lot of the simplicity comes from the fact that both sides must have each other's public keys in advance, and there are no options to choose. There will be one fixed cipher suite, and both peers always authenticate each other. (Inspiration taken from Ian Grigg's hypotheses[0].) There might be some compromise in the future, to enable or disable encryption, authentication and compression, but there will be no choice of algorithms. This will allow SPTPS to be built with a few embedded crypto algorithms instead of linking with huge crypto libraries. The API is also kept simple. There is a start and a stop function. All data necessary to make the connection work is passed in the start function. Instead having both send- and receive-record functions, there is a send-record function and a receive-data function. The latter will pass protocol data received from the peer to the SPTPS implementation, which will in turn call a receive-record callback function when necessary. This hides all the handshaking from the application, and is completely independent from any event loop or socket characteristics. [0] http://iang.org/ssl/hn_hypotheses_in_secure_protocol_design.html
2011-07-24 13:44:51 +00:00
// Initialise struct sptps
memset(s, 0, sizeof *s);
s->handle = handle;
s->initiator = initiator;
s->datagram = datagram;
Start of "Simple Peer-To-Peer Security" protocol. Encryption and authentication of the meta connection is spread out over meta.c and protocol_auth.c. The new protocol was added there as well, leading to spaghetti code. To improve things, the new protocol will now be implemented in sptps.[ch]. The goal is to have a very simplified version of TLS. There is a record layer, and there are only two record types: application data and handshake messages. The handshake message contains a random nonce, an ephemeral ECDH public key, and an ECDSA signature over the former. After the ECDH public keys are exchanged, a shared secret is calculated, and a TLS style PRF is used to generate the key material for the cipher and HMAC algorithm, and further communication is encrypted and authenticated. A lot of the simplicity comes from the fact that both sides must have each other's public keys in advance, and there are no options to choose. There will be one fixed cipher suite, and both peers always authenticate each other. (Inspiration taken from Ian Grigg's hypotheses[0].) There might be some compromise in the future, to enable or disable encryption, authentication and compression, but there will be no choice of algorithms. This will allow SPTPS to be built with a few embedded crypto algorithms instead of linking with huge crypto libraries. The API is also kept simple. There is a start and a stop function. All data necessary to make the connection work is passed in the start function. Instead having both send- and receive-record functions, there is a send-record function and a receive-data function. The latter will pass protocol data received from the peer to the SPTPS implementation, which will in turn call a receive-record callback function when necessary. This hides all the handshaking from the application, and is completely independent from any event loop or socket characteristics. [0] http://iang.org/ssl/hn_hypotheses_in_secure_protocol_design.html
2011-07-24 13:44:51 +00:00
s->mykey = mykey;
s->hiskey = hiskey;
s->replaywin = sptps_replaywin;
if(s->replaywin) {
s->late = malloc(s->replaywin);
if(!s->late)
return error(s, errno, strerror(errno));
2013-08-30 12:22:05 +00:00
memset(s->late, 0, s->replaywin);
}
Start of "Simple Peer-To-Peer Security" protocol. Encryption and authentication of the meta connection is spread out over meta.c and protocol_auth.c. The new protocol was added there as well, leading to spaghetti code. To improve things, the new protocol will now be implemented in sptps.[ch]. The goal is to have a very simplified version of TLS. There is a record layer, and there are only two record types: application data and handshake messages. The handshake message contains a random nonce, an ephemeral ECDH public key, and an ECDSA signature over the former. After the ECDH public keys are exchanged, a shared secret is calculated, and a TLS style PRF is used to generate the key material for the cipher and HMAC algorithm, and further communication is encrypted and authenticated. A lot of the simplicity comes from the fact that both sides must have each other's public keys in advance, and there are no options to choose. There will be one fixed cipher suite, and both peers always authenticate each other. (Inspiration taken from Ian Grigg's hypotheses[0].) There might be some compromise in the future, to enable or disable encryption, authentication and compression, but there will be no choice of algorithms. This will allow SPTPS to be built with a few embedded crypto algorithms instead of linking with huge crypto libraries. The API is also kept simple. There is a start and a stop function. All data necessary to make the connection work is passed in the start function. Instead having both send- and receive-record functions, there is a send-record function and a receive-data function. The latter will pass protocol data received from the peer to the SPTPS implementation, which will in turn call a receive-record callback function when necessary. This hides all the handshaking from the application, and is completely independent from any event loop or socket characteristics. [0] http://iang.org/ssl/hn_hypotheses_in_secure_protocol_design.html
2011-07-24 13:44:51 +00:00
s->label = malloc(labellen);
if(!s->label)
return error(s, errno, strerror(errno));
if(!datagram) {
s->inbuf = malloc(7);
if(!s->inbuf)
return error(s, errno, strerror(errno));
s->buflen = 0;
}
Start of "Simple Peer-To-Peer Security" protocol. Encryption and authentication of the meta connection is spread out over meta.c and protocol_auth.c. The new protocol was added there as well, leading to spaghetti code. To improve things, the new protocol will now be implemented in sptps.[ch]. The goal is to have a very simplified version of TLS. There is a record layer, and there are only two record types: application data and handshake messages. The handshake message contains a random nonce, an ephemeral ECDH public key, and an ECDSA signature over the former. After the ECDH public keys are exchanged, a shared secret is calculated, and a TLS style PRF is used to generate the key material for the cipher and HMAC algorithm, and further communication is encrypted and authenticated. A lot of the simplicity comes from the fact that both sides must have each other's public keys in advance, and there are no options to choose. There will be one fixed cipher suite, and both peers always authenticate each other. (Inspiration taken from Ian Grigg's hypotheses[0].) There might be some compromise in the future, to enable or disable encryption, authentication and compression, but there will be no choice of algorithms. This will allow SPTPS to be built with a few embedded crypto algorithms instead of linking with huge crypto libraries. The API is also kept simple. There is a start and a stop function. All data necessary to make the connection work is passed in the start function. Instead having both send- and receive-record functions, there is a send-record function and a receive-data function. The latter will pass protocol data received from the peer to the SPTPS implementation, which will in turn call a receive-record callback function when necessary. This hides all the handshaking from the application, and is completely independent from any event loop or socket characteristics. [0] http://iang.org/ssl/hn_hypotheses_in_secure_protocol_design.html
2011-07-24 13:44:51 +00:00
memcpy(s->label, label, labellen);
s->labellen = labellen;
s->send_data = send_data;
s->receive_record = receive_record;
// Do first KEX immediately
s->state = SPTPS_KEX;
Start of "Simple Peer-To-Peer Security" protocol. Encryption and authentication of the meta connection is spread out over meta.c and protocol_auth.c. The new protocol was added there as well, leading to spaghetti code. To improve things, the new protocol will now be implemented in sptps.[ch]. The goal is to have a very simplified version of TLS. There is a record layer, and there are only two record types: application data and handshake messages. The handshake message contains a random nonce, an ephemeral ECDH public key, and an ECDSA signature over the former. After the ECDH public keys are exchanged, a shared secret is calculated, and a TLS style PRF is used to generate the key material for the cipher and HMAC algorithm, and further communication is encrypted and authenticated. A lot of the simplicity comes from the fact that both sides must have each other's public keys in advance, and there are no options to choose. There will be one fixed cipher suite, and both peers always authenticate each other. (Inspiration taken from Ian Grigg's hypotheses[0].) There might be some compromise in the future, to enable or disable encryption, authentication and compression, but there will be no choice of algorithms. This will allow SPTPS to be built with a few embedded crypto algorithms instead of linking with huge crypto libraries. The API is also kept simple. There is a start and a stop function. All data necessary to make the connection work is passed in the start function. Instead having both send- and receive-record functions, there is a send-record function and a receive-data function. The latter will pass protocol data received from the peer to the SPTPS implementation, which will in turn call a receive-record callback function when necessary. This hides all the handshaking from the application, and is completely independent from any event loop or socket characteristics. [0] http://iang.org/ssl/hn_hypotheses_in_secure_protocol_design.html
2011-07-24 13:44:51 +00:00
return send_kex(s);
}
// Stop a SPTPS session.
bool sptps_stop(sptps_t *s) {
Start of "Simple Peer-To-Peer Security" protocol. Encryption and authentication of the meta connection is spread out over meta.c and protocol_auth.c. The new protocol was added there as well, leading to spaghetti code. To improve things, the new protocol will now be implemented in sptps.[ch]. The goal is to have a very simplified version of TLS. There is a record layer, and there are only two record types: application data and handshake messages. The handshake message contains a random nonce, an ephemeral ECDH public key, and an ECDSA signature over the former. After the ECDH public keys are exchanged, a shared secret is calculated, and a TLS style PRF is used to generate the key material for the cipher and HMAC algorithm, and further communication is encrypted and authenticated. A lot of the simplicity comes from the fact that both sides must have each other's public keys in advance, and there are no options to choose. There will be one fixed cipher suite, and both peers always authenticate each other. (Inspiration taken from Ian Grigg's hypotheses[0].) There might be some compromise in the future, to enable or disable encryption, authentication and compression, but there will be no choice of algorithms. This will allow SPTPS to be built with a few embedded crypto algorithms instead of linking with huge crypto libraries. The API is also kept simple. There is a start and a stop function. All data necessary to make the connection work is passed in the start function. Instead having both send- and receive-record functions, there is a send-record function and a receive-data function. The latter will pass protocol data received from the peer to the SPTPS implementation, which will in turn call a receive-record callback function when necessary. This hides all the handshaking from the application, and is completely independent from any event loop or socket characteristics. [0] http://iang.org/ssl/hn_hypotheses_in_secure_protocol_design.html
2011-07-24 13:44:51 +00:00
// Clean up any resources.
chacha_poly1305_exit(s->incipher);
chacha_poly1305_exit(s->outcipher);
ecdh_free(s->ecdh);
Start of "Simple Peer-To-Peer Security" protocol. Encryption and authentication of the meta connection is spread out over meta.c and protocol_auth.c. The new protocol was added there as well, leading to spaghetti code. To improve things, the new protocol will now be implemented in sptps.[ch]. The goal is to have a very simplified version of TLS. There is a record layer, and there are only two record types: application data and handshake messages. The handshake message contains a random nonce, an ephemeral ECDH public key, and an ECDSA signature over the former. After the ECDH public keys are exchanged, a shared secret is calculated, and a TLS style PRF is used to generate the key material for the cipher and HMAC algorithm, and further communication is encrypted and authenticated. A lot of the simplicity comes from the fact that both sides must have each other's public keys in advance, and there are no options to choose. There will be one fixed cipher suite, and both peers always authenticate each other. (Inspiration taken from Ian Grigg's hypotheses[0].) There might be some compromise in the future, to enable or disable encryption, authentication and compression, but there will be no choice of algorithms. This will allow SPTPS to be built with a few embedded crypto algorithms instead of linking with huge crypto libraries. The API is also kept simple. There is a start and a stop function. All data necessary to make the connection work is passed in the start function. Instead having both send- and receive-record functions, there is a send-record function and a receive-data function. The latter will pass protocol data received from the peer to the SPTPS implementation, which will in turn call a receive-record callback function when necessary. This hides all the handshaking from the application, and is completely independent from any event loop or socket characteristics. [0] http://iang.org/ssl/hn_hypotheses_in_secure_protocol_design.html
2011-07-24 13:44:51 +00:00
free(s->inbuf);
free(s->mykex);
free(s->hiskex);
Start of "Simple Peer-To-Peer Security" protocol. Encryption and authentication of the meta connection is spread out over meta.c and protocol_auth.c. The new protocol was added there as well, leading to spaghetti code. To improve things, the new protocol will now be implemented in sptps.[ch]. The goal is to have a very simplified version of TLS. There is a record layer, and there are only two record types: application data and handshake messages. The handshake message contains a random nonce, an ephemeral ECDH public key, and an ECDSA signature over the former. After the ECDH public keys are exchanged, a shared secret is calculated, and a TLS style PRF is used to generate the key material for the cipher and HMAC algorithm, and further communication is encrypted and authenticated. A lot of the simplicity comes from the fact that both sides must have each other's public keys in advance, and there are no options to choose. There will be one fixed cipher suite, and both peers always authenticate each other. (Inspiration taken from Ian Grigg's hypotheses[0].) There might be some compromise in the future, to enable or disable encryption, authentication and compression, but there will be no choice of algorithms. This will allow SPTPS to be built with a few embedded crypto algorithms instead of linking with huge crypto libraries. The API is also kept simple. There is a start and a stop function. All data necessary to make the connection work is passed in the start function. Instead having both send- and receive-record functions, there is a send-record function and a receive-data function. The latter will pass protocol data received from the peer to the SPTPS implementation, which will in turn call a receive-record callback function when necessary. This hides all the handshaking from the application, and is completely independent from any event loop or socket characteristics. [0] http://iang.org/ssl/hn_hypotheses_in_secure_protocol_design.html
2011-07-24 13:44:51 +00:00
free(s->key);
free(s->label);
free(s->late);
2012-10-09 14:27:28 +00:00
memset(s, 0, sizeof *s);
Start of "Simple Peer-To-Peer Security" protocol. Encryption and authentication of the meta connection is spread out over meta.c and protocol_auth.c. The new protocol was added there as well, leading to spaghetti code. To improve things, the new protocol will now be implemented in sptps.[ch]. The goal is to have a very simplified version of TLS. There is a record layer, and there are only two record types: application data and handshake messages. The handshake message contains a random nonce, an ephemeral ECDH public key, and an ECDSA signature over the former. After the ECDH public keys are exchanged, a shared secret is calculated, and a TLS style PRF is used to generate the key material for the cipher and HMAC algorithm, and further communication is encrypted and authenticated. A lot of the simplicity comes from the fact that both sides must have each other's public keys in advance, and there are no options to choose. There will be one fixed cipher suite, and both peers always authenticate each other. (Inspiration taken from Ian Grigg's hypotheses[0].) There might be some compromise in the future, to enable or disable encryption, authentication and compression, but there will be no choice of algorithms. This will allow SPTPS to be built with a few embedded crypto algorithms instead of linking with huge crypto libraries. The API is also kept simple. There is a start and a stop function. All data necessary to make the connection work is passed in the start function. Instead having both send- and receive-record functions, there is a send-record function and a receive-data function. The latter will pass protocol data received from the peer to the SPTPS implementation, which will in turn call a receive-record callback function when necessary. This hides all the handshaking from the application, and is completely independent from any event loop or socket characteristics. [0] http://iang.org/ssl/hn_hypotheses_in_secure_protocol_design.html
2011-07-24 13:44:51 +00:00
return true;
}