Lightweight Obfuscated Datagram Protocol (LODP) Yawning Angel (XXX: Preliminary Draft. If anyone implements this, I will laugh at them, because there are probably problems, and parts (all?) of it may suck. In particular, open questions/to-be-encorporated feedback is appended to the end of the document.) 0. Introduction This is an encrypted wire transport protocol designed to be a fingerprinting resistant alternative to Datagram TLS. Its purpose is to keep a third party from being able to tell what protocol is in use based on message contents and to provide authentication and data integrity. 1. Motivation There is a large volume of work that allows for "reliable bulk data transfer over UDP", by developing a lightweight secure obfuscated datagram transport protocol, it would be possible to leverage existing research in these areas. Examples of candidates for LODP encapsulation, in no particular order, are SCTP-over-UDP, Reliable Data Protocol (RFC1151), enet, and UDT. (XXX: Add something covering I2P's SSU and Dust) 2. LODP Threat Model The threat model for LODP is identical to the threat model of obfs3 [OBFS3 THREAT MODEL], with added goals: LODP offers protection against active scanning machines that expect the LODP protocol. Such machines should not be able to verify the existence of the LODP protocol without compromising the authenticated encryption used in session initialization. LODP offers protection against active attackers that attempt to mount a man-in-the-middle attack. The cryptographic handshake will fail, and such attacks will be detected. LODP does not protect against adversaries that are capable of measuring protocol entropy, or those that employ a protocol whitelist. Attacks on the out-of-band shared secret distribution mechanism (Eg: Tor's BridgeDB) are likewise outside the scope of the protocol design, however the most that an attacker that obtains a endpoint's shared secret is able to do is to censor the protocol. 3. Notation and Constants The key words "MUST", "MUST NOT", "REQUIRED", "SHALL", "SHALL NOT", "SHOULD", "SHOULD NOT", "RECOMMENDED", "MAY", and "OPTIONAL" in this document are to be interpreted as described in [KEYWORDS]. All packet fields are in network byte order. A byte is a 8 bit octet. Let a | b be the concatenation of a with b. All random numbers used in this protocol MUST come from a cryptographically strong random number generator. 4. Overview of LODP LODP attempts to make all traffic look like random noise. The session handshake uses MAC and bulk encryption keys derived from the server's public Curve25519 key to mask the packet payload, and negotiates a ephemeral session key to provide actual data authentication, integrity and perfect forward secrecy. In the event that the server's long term identity key is compromised, the most that an attacker can do is identify the LODP protocol. The LODP packet format provides for added arbitrary random padding, and undersized packets are silently ignored to allow the implementer flexibility when attempting to mask the packet size. To mitigate the potential of an attacker to mount a denial of service attack given the server's host key, LODP uses a 4 way handshake similar to SCTP or DTLS. (XXX: Does this section need more stuff???) 5. Maximum Transmission Unit and Path MTU Discovery The MTU including IP and UDP headers MUST be a minimum of 576 bytes for IPv4 endpoints, and 1280 bytes for IPv6 endpoints. Implementations MAY implement PMTU discovery and MAY chose to send larger packets if the underlying data link layer is capable of doing so. Note: Implementations that chose to implement PMTU discover MUST ensure that the traffic profile is not trivially fingerprinted. Implementations SHOULD make attempts to avoid IP fragmentation. As LODP is designed as the transport for an encapsulated higher level transport, implementations MAY allow the higher level protocol to handle PMTU discovery. (XXX: The peer could advertise a MTU when handshaking but not sure how useful that is going to be.) (XXX: Technically the MTUs I chose will never be fragmented, but there's probably screwed up routers/hosts on the internet that aren't spec compliant, not sure "thou shalt not fragment" is better as a SHOULD or a MUST) 6. Cryptographic Algorithms and Keys LODP uses several cryptographic algorithms as part of its protocol. All packets are encrypted with XChaCha/20 [XCHACHA], and then the encrypted payload is authenticated with HMAC-SHA2-256 [RFC6234] algorithm (Encrypt-Then-MAC). IVs MUST be randomly generated and MUST NOT be reused for a given key. XChaCha/20 IV = 24 bytes in the packet header Ephemeral session keys are generated during the initial handshake via a modified version of the ntor [NTOR] handshake. The 32 byte shared secret is then stretched to create the required HMAC-SHA2-256 and XChaCha/20 keys. The modified ntor handshake like the original uses the Curve25519 ECDH [CURVE 25519] and SHA2-256 algorithms. Every end point additionally has a public Curve25519 key (IntroKey) that is used to verify router identity and session establishment. To provide fingerprinting and active scanning resistance, initial semi-public HMAC-SHA2-256 and XChaCha/20 keys are derived via a key stretching algorithm from the public Curve25519 key. The key stretching algorithm used is HKDF-SHA256 [RFC6234] used as the HMAC/hash algorithm. Intro Keys: Salt = "LODP-HKDF-Intro-SHA256" PRK = HKDF-Extract(Salt, PublicCurve25519Key) IntroKey = HKDF-Expand(PRK, Salt, 64) IntroMacKey = IntroKey[0:31] IntroXChaChaKey = IntroKey[32:63] Session Keys: Salt = "LODP-HKDF-Session-SHA256" PRK = SharedSecret SessionKey = HKDF-Expand(PRK, Salt, 128) InitiatorMacKey = SessionKey[0:31] (Client->Server) InitiatorXChaChaKey = SessionKey[32:63] ResponderMacKey = SessionKey[64:95] (Server->Client) ResponderXChaChaKey = SessionKey[96:127] Note: The HKDF-Extract step is carried out as part of the handshake. A rough overview of all the keys, their origins and lifetimes, and their uses are as follows: Host Curve25519 Identity Key: Long term Curve25519 key used for proving host identity and as a component of the modified ntor handshake. Intro HMAC-SHA2-256 Key: Long term MAC key used for providing fingerprinting resistance and to defeat active scanning attacks, derived via HKDF from the Curve25519 key. Intro XChaCha/20 Key: Long term cipher key used for providing fingerprinting resistance and to defeat active scanning attacks, derived via HDFK from the Curve25519 key. Session HMAC-SHA2-256 Key: Session MAC key used for data integrity. Generated during the handshake. Session XChaCha/20 Key: Session cipher key used for data integrity/secrecy. Generated during the handshake. Note: For any given session there are 2 HMAC-SHA2-256 keys and 2 XChaCha/20 keys. One set is used for Initiator->Responder traffic, and the other is used for Responder->Initiator traffic. To connect to a remote peer, the initiator obtains the Host Curve25519 Identity key, and derives the Intro HMAC-SHA2-256 and XChaCha/20 keys. The initiator will then connect to the remote peer and generate the ephemeral session MAC/cipher keys via the handshake process. Once the handshake is completed, all further data sent over the connection is authenticated and encrypted with the ephemeral session keys. The distribution method for the Identity key is left up to the implementers. Fingerprinting resistance is dependent on the security of the Identity key, so the distribution mechanism SHOULD be secure. (XXX: SHA-2 can be replaced with something else) (XXX: Consider HMAC-SHA2-256-128 as the MAC, might need rekeying then.) 7. Packet Format A LODP packet is comprised of a common header, the payload and optional arbitrary padding. The payload contains either control or user data. 0 1 2 3 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | | + HMAC-SHA2-256 MAC + ... + (256 bits) + | | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | | + IV + ... + (192 bits) + | | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | Type | Flags | Length | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ \ \ / Packet Data / \ \ +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ \ \ / Arbitrary amount of uninterpreted data / \ \ +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ All integer fields in a LODP packet MUST be transmitted in network byte order. HMAC-SHA-256 MAC: 256 bits This field contains the HMAC-SHA-256 MAC of the LODP packet. The HMAC algorithm is applied over the beginning of the IV field to the end of the packet (The encrypted optional padding is also validated). IV: 192 bits This field contains the Initialization Vector used for the XChaCha/20 cipher. IVs MUST NOT be reused. Type: 8 bits (unsigned integer) The type of the packet. The values of the Type are defined as follows: Type Value Type ----- ------------- 0 - Data (DATA) 1 - Initiation (INIT) 2 - Initiation Acknowledgement (INIT ACK) 3 - Handshake (HANDSHAKE) 4 - Handshake Acknowledgement (HANDSHAKE ACK) 5 - Heartbeat (HEARTBEAT) 6 - Heartbeat Acknowledgement (HEARTBEAT ACK) Flags: 8 bits (unsigned integer) Type specific flags. Length: 16 bits (unsigned integer) Payload length including the Type, Flag and Length fields ("4" for types with no data). Packet Data: variable length The actual packet data. Arbitrary amount of uninterpreted data: variable length Optional padding. Implementations MAY append random padding to each packet to mask the packet size signature of the upper layer protocol. Implementations MUST NOT interpret this data beyond verifying the MAC and as part of the bulk decryption. The XChaCha/20 encryption begins from the end of the IV to the end of the packet (The optional padding is also encrypted). 7.1. Data (DATA) The following format MUST be used for Data packets. 0 1 2 3 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | Type = 0 | Reserved | Length | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ \ \ / User Data / \ \ +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ Reserved: 8 bits Should be set to all '0's and ignored by the receiver. User Data: variable length This is the user payload data. The session keys derived during the handshake MUST be used for the MAC and encryption of DATA packets. DATA packets received that are MACed or Encrypted with the IntroKeys MUST be discarded. 7.2. Initiation (INIT) The following format MUST be used for Initiation packets. 0 1 2 3 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | Type = 1 | Reserved | Length = 68 | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | | + Initiator HMAC-SHA2-256 MAC IntroKey + ... + (256 bits) + | | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | | + Initiator XChaCha/20 IntroKey + ... + (256 bits) + | | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | Time Stamp | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ Reserved: 8 bits Should be set to all '0's and ignored by the receiver. Initiator HMAC-SHA2-256 MAC IntroKey: 256 bits The HMAC-SHA2-256 MAC key to use when sending the INIT ACK. Initiator XChaCha/20 IntroKey: 256 bits The XChaCha/20 stream cipher key to use when sending the INIT ACK. Time Stamp: 32 bits (unsigned integer) The number of seconds from the Unix Epoch. Implementations MAY add a small random offset to value transmitted in this field to mask the true system time. The Intro MAC and XChaCha/20 keys derived from the responder's Curve25519 Identity key MUST be used for the MAC and encryption of the INIT packet. INIT packets received that are MACed or Encrypted with other keys MUST be discarded. Implementations MAY chose to use ephemeral MAC/XChaCha keys for the keys transmitted in a INIT ACK packet. 7.3. Initiation Acknowledgement (INIT ACK) The following format MUST be used for Initiation Acknowledgement packets. 0 1 2 3 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | Type = 2 | Reserved | Length | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | Initiator Port | Initiator Address Length | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | Initiator Address (32 or 128 bits) | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ \ \ / Cookie / \ \ +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ Reserved: 8 bits Should be set to all '0's and ignored by the receiver. Initiator Port: 16 bits (unsigned integer) The IP Port that the INIT packet was received from. Initiator Address Length: 16 bits (unsigned integer) The length of the IP address that the INIT packet was received from. "4" for IPv4 addresses, "16" for IPv6 addresses. Initiator Address: 32/128 bits (unsigned integer) The IP address that the INIT packet was received from. Cookie: variable length An opaque cookie. The format is up to the implementation however implementations SHOULD generate cookies that can be validated without maintaining per-connection state, and SHOULD contain timestamping information to prevent cookies from being reused. The IntroKeys received in the INIT packet that triggered the INIT ACK MUST be used for the MAC and encryption of INIT ACK packets. 7.4. Handshake (HANDSHAKE) The following format MUST be used for Handshake packets. 0 1 2 3 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | Type = 3 | Reserved | Length | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | | + Initiator HMAC-SHA2-256 MAC IntroKey + ... + (256 bits) + | | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | | + Initiator XChaCha/20 IntroKey + ... + (256 bits) + | | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | | + Initiator Curve25519 Public Key + ... + (256 bits) + | | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ \ \ / Cookie / \ \ +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ Reserved: 8 bits Should be set to all '0's and ignored by the receiver. Initiator HMAC-SHA2-256 MAC IntroKey: 256 bits The HMAC-SHA2-256 MAC key to use when sending the HANDSHAKE ACK. This key MUST be identical to the key transmitted in the INIT packet. Initiator XChaCha/20 IntroKey: 256 bits The XChaCha/20 stream cipher key to use when sending the HANDSHAKE ACK. This key MUST be identical to the key transmitted in the INIT packet. Initiator Curve25519 Public Key: 256 bits The Curve25519 public key to use when deriving the session keys. Cookie: variable length The opaque cookie received in the INIT ACK, echoed exactly. The Intro MAC and XChaCha/20 keys derived from the responder's Curve25519 Identity key MUST be used for the MAC and encryption of the HANDSHAKE packet. HANDSHAKE packets received that are MACed or Encrypted with other keys MUST be discarded. 7.5. Handshake Acknowledgement (HANDSHAKE ACK) The following format MUST be used for Handshake Acknowledgement packets. 0 1 2 3 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | Type = 4 | Reserved | Length = 136 | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | | + Responder Curve25519 Public Key + ... + (256 bits) + | | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | | + Handshake Authentication Digest + ... + (256 bits) + | | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ Reserved: 8 bits Should be set to all '0's and ignored by the receiver. Responder Curve25519 Public Key: 256 bits The Curve25519 public key to use when deriving the session keys. Handshake Authentication Digest: 256 bits The SHA2-256 digest (Auth) to used when verifying the remote identity as part of the handshake. The IntroKeys received in the HANDSHAKE packet that triggered the HANDSHAKE ACK MUST be used for the MAC and encryption of HANDSHAKE ACK packets. 7.6. Heartbeat (HEARTBEAT) The following format MUST be used for Heartbeat packets. 0 1 2 3 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | Type = 5 | Reserved | Length | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ \ \ / Heartbeat Data / \ \ +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ Reserved: 8 bits Should be set to all '0's and ignored by the receiver. Heartbeat Data: variable length The data that the receiver should echo in a Heartbeat Acknowledgement. The format is left unspecified. The session keys derived during the handshake MUST be used for the MAC and encryption of HEARTBEAT packets. HEARTBEAT packets received that are MACed or Encrypted with any other keys MUST be discarded. 7.7. Heartbeat Acknowledgement (HEARTBEAT ACK) The following format MUST be used for Heartbeat Acknowledgement packets. 0 1 2 3 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | Type = 6 | Reserved | Length | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ \ \ / Echoed Heartbeat Data / \ \ +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ Reserved: 8 bits Should be set to all '0's and ignored by the receiver. Echoed Heartbeat Data: variable length The Heartbeat Data received in the Heartbeat packet that triggered the Heartbeat Acknowledgement, echoed exactly. The session keys derived during the handshake MUST be used for the MAC and encryption of HEARTBEAT ACK packets. HEARTBEAT ACK packets received that are MACed or Encrypted with any other keys MUST be discarded. 8. The Modified ntor Handshake Set PROTOID == "ntor-lodp-curve25519-sha256-1". Set EXP(a,b) == curve25519(.,b,a), and g = 9. Both parties MUST check that the operation did not procude the point at infinity, and MUST abort the handshake otherwise. Set X,x = The initiator's Curve25519 key pair (X is the public component transmitted in the HANDSHAKE packet). Set Y,y = The responder's Curve25519 key pair (Y is the public component transmitted in the HANDSHAKE ACK packet). Set B,b = The responder's Curve25519 Identity key pair (B is the public component distributed to the initator out-of-band). The responder side of the handshake computes: SecretInput = EXP(X,y) | EXP(X,b) | B | X | Y | PROTOID SharedSecret = SHA2-256(SecretInput, PROTOID | ":key_extract") Verify = SHA2-256(SecretInput, PROTOID | ":key_verify") AuthInput = Verify | B | Y | X | PROTOID | "Responder" Auth = SHA2-256(AuthInput, PROTOID | ":mac") The responder then transmits Y and Auth to the initiator in the HANDSHAKE ACK packet. The initiator side computes: SecretInput = EXP(Y,x) | EXP(B,x) | B | X | Y | PROTOID SharedSecret = SHA2-256(SecretInput, PROTOID | ":key_extract") Verify = SHA2-256(SecretInput, PROTOID | ":key_verify") AuthInput = Verify | B | Y | X | PROTOID | "Responder" Auth = SHA2-256(AuthInput, PROTOID | ":mac") The initiator verifies that the Auth value it computed is identical to the one received in the HANDSHAKE ACK packet. Ephemeral session keys are then derived from the SharedSecret value per section 6. 9. Packet Processing When a new packet has been received, implementations should process it as follows: 0. Optionally packets that are obviously malformed MAY be silently discarded before further processing is applied to conserve resources. 1. If there is (a) existing session(s) associated with the packet's Address/Port combination. a) The MAC is verified with the SessionKey in the existing TCB. If the verification fails the packet is processed as part of a new handshake (Step 2). Note: There may be more than one Session originating from a given Address/Port. It is left up to implementations if they allow this or not, however implementations MUST account for retransmitted HANDSHAKE packets in the event that HANDSHAKE ACKs are lost. b) The packet is decrypted with the cipher SessionKey in the existing TCB. c) If the packet type is DATA, HEARTBEAT or HEARTBEAT ACK, the packet is processed. Otherwise the packet MUST be silently discarded. Note: Implementations MAY ignore HEARTBEAT packets if under load or to prevent excess traffic. 2. If there is no existing session associated with the packet's Address/Port combination. a) The MAC is verified with the MAC IntroKey. If the verification fails, the packet MUST be silenty discarded. b) The packet is decrypted with the cipher IntroKey. c) If the packet type is INIT, INIT ACK, HANDSHAKE or HANDSHAKE ACK, the packet is processed (See "Session Establishment"). Otherwise the packet MUST be silently discarded. 10. Session Establishment The session establishment process and the keys used for each packet is as follows: Initiator Responder --------- --------- [Obtain the responder Identity key] [Derive the responder IntroKeys] [Create TCB] INIT (IntroKeys) --------> [INIT packet timestamp is verified to be sufficiently fresh] [Create cryptographically strong cookie] <-------- INIT ACK (Initiator IntroKeys contained in INIT) HANDSHAKE (IntroKeys) --------> [Validate cookie] [Create TCB] [Complete ntor handshake] [Connection state -> ESTABLISHED] <-------- HANDSHAKE ACK (Initiator IntroKeys contained in HANDSHAKE) [Complete ntor handshake] [Validate ntor digest] [Connection state -> ESTABLISHED] DATA (Initiator SessionKeys) <-------> DATA (Responder SessionKeys) The responder MUST NOT allocate any session state till it receives a valid HANDSHAKE packet from the initiator. Implementations MUST allow for a reasonable clock skew between initiator and responder when verifying the INIT packet timestamp to allow for the possibility that the clocks are out of sync (Due to systems not using a time synchronization service or implementations chosing to send a value with a random offset to avoid client identification). The responder MUST NOT automatically retransmit HANDSHAKE ACKs. The initiator SHOULD handle the possibility of a HANDSHAKE ACK getting lost by retransmitting HANDSHAKE packets. The responder MAY cache the payload of the HANDSHAKE ACK packet to conserve CPU, however the retransmitted HANDSHAKE ACK MUST have unique IV. 11. Session Teardown As LODP is designed for encapsulated transport, session teardown is generally handled by the upper level protocol. Implementations MUST allow the user to specify that a given session has terminated, and reclaim resources when this occurs. Implementations MAY additionally implement detection of idle connections by using the HEARTBEAT and HEARTBEAT ACK packets. 12. Security Considerations In the event that an attacker can obtain the responder's Identity key, it is able to decrypt the handshake's bulk encryption and thus identify LODP traffic. However this event implies that the distribution mechanism or one of the peers participating in protocol has been attacked, and thus censoring LODP is possible due to the attacker knowing the IP address/Port associated with the responder. Additionally as ephemeral session keys are used, this does not compromise the secrecy or integrity of the transported data. Most stateful protocols are vulnerable to SYN flood like DDOS attacks. In LODP this attack is mitigated by the use of a 4 way handshake preventing INIT packets with spoofed source addresses from consuming server resources apart from the processing power required to create INIT ACK responses. As randomized IVs are used, the odds of collisions due to the birthday problem cannot be ignored. However since the IV is 192 bits, the odds of this occurring are astronomically small given a cryptographically strong random number generator. The protocol is vulnerable to replay attacks unless proper defenses are present. The susceptibility to this is dependent on what state the connection is in. The handshake is protected against replay attacks by the inclusion of a timestamp field in the INIT packet and by the opaque cookie included in the INIT ACK and HANDSHAKE packets. Implementations SHOULD incorporate additional defenses such as rejecting duplicate IVs. Data packets have no replay protection under the assumption that the encapsulated higher level protocol handles this on it's own. As most protocols that attempt to provide reliable data services include sequence numbering, common use cases should be adequately defended against this attack. 13. References [OBFS3 THREAT MODEL] Kadianakis, G., "Threat model for the obfs3 obfuscation protocol", (https://gitweb.torproject.org/pluggable-transports/ obfsproxy.git/blob/HEAD:/doc/obfs3/ obfs3-threat-model.txt) [KEYWORDS] Bradner, S., "Key words for use in RFCs to Indicate Requirement Levels", BCP 14, RFC 2119, March 1997. [XCHACHA] Bernstein, D. J., "Extending the Salsa20 nonce", (http://cr.yp.to/snuffle/xsalsa-20110204.pdf) [RFC6234] Eastlake, D. and T. Hansen, "US Secure Hash Algorithms (SHA and SHA-based HMAC and HKDF)", RFC 6234, May 2011. [CURVE 25519] Bernstein, D. J., "A state-of-the-art Diffie-Hellman function", (https://cr.yp.to/ecdh.html) [NTOR] Matthewson, N., "Improved circuit-creation key exchange", (https://gitweb.torproject.org/torspec.git/blob_plain/HEAD: /proposals/216-ntor-handshake.txt) 14. Acknowledgements The author would like to thank Roger Dingledine, George Kadianakis, Mike Perry, Nick Mathewson, Brandon Wiley, Zachary Weinberg and the various denizens of #tor-dev for their useful comments. Much of LODP's design has been inspired by the SSU protocol developed by the I2P project and the Stream Control Transmission Protocol. (XXX: Did I miss anyone?) XXX: Stuff that I want input on (my comments below) * HMAC-SHA2-256 is kind of heavy, in terms of packet header overhead (truncate?) and processing speed (Different algorithm?). The inital draft used Poly1305-AES, but people didn't seem to like that because it pulled in AES (side channel attacks are sad). Some people have suggested using OCB (patent issues), but that requires finding a non-suck block cypher (I could use ChaCha in a LIONESS construct but then I might as well just use AES CCM or something) * The fact that I have 64 bytes worth of headers is kind of worrying. If it's an issue I could just use ChaCha/20 (64 bit IV) and truncate the HMAC output down to 128 bits and half the header size (32 bytes) but I will need to add support for rekeying. Just using truncated HMAC cuts the header down to 48 bytes. Note that the large header does not weaken the fingerprinting resistance as undersized packets are silently ignored so implementations are free to send undersized packets as cover traffic). * Is adding support for rekeying a good idea? I don't foresee IV collisions to be a realistic problem unless I switch the bulk crypto algorithm to ChaCha. * Per zwol: elligator instead of preshared secret? (No implementation, I'm not smart enough to do it, not sure if applying elligator to all incoming unknown packets will kill the server's CPU or not. Probaby a bigger hit than HMAC-SHA2-256.... Additionally that still requires some form of shared secret or I'll be vulnerable to active probing attacks, and using a shared secret in itself isn't a issue for Tor PTs because an attacker obtaining the shared secret also gives up the IP/Port to a censor, and we lose anyway.) * Is it ok to derive the Intro HMAC/encryption keys from the public identity key? It's relatively unlikely than an IV would get reused for the 2 packets that it's relevant for and there are more cost effective attacks to defeat the fingerprinting resistance ("just attack bridgedb")... I mean, I *could* do something like "append the date to the HMACed init/handshake data" and rotate the HMAC/encrpytion intro keys once aday without too much extra overhead. With a 192 bit IV, unless certain platforms have a utterly horrific PRNG, this is unlikely. * Is this actually secure? I'm not a cryptographer. Are the primitives I'm using safe? XXX: Stuff that's probably ok, but want a second opinion on * The ntor variant I use doesn't have a notion of NODEID, but otherwise is near identical (Some of the personalization strings I use are different). Did I inadvertently break the security? (Don't think so) * Is it ok to say "the encapsulated transport should handle graceful teardown, implementations can implement an idle timer to reap stale connections"? * Should I generate errors on handshake failure or is saying "Well, it fails when your timer expires sufficient"? * Per zwol: look at what crypto_box in NaCl does and adjust my bulk crypto. (It looks like the same thing that I do, at least for crypto_securebox, just with Poly1305-AES instead of HMAC-SHA2-256) * The INIT packet includes a second granular timestamp. This is just to prevent replay attacks from trivially detecting the protocol, so the time doesn't need to be all that accurate (+- 5-10 mins should be a reasonable window). So the spec says that the client can add random jitter to the value to prevent leaking it's system clock.