obfs4/doc/obfs4-spec.txt

   obfs4 (The obfourscator)

0. Introduction

   This is a protocol obfuscation layer for TCP protocols.  Its purpose is to
   keep a third party from telling what protocol is in use based on message
   contents.

   Unlike obfs3, obfs4 attempts to provide authentication and data integrity,
   though it is still designed primarily around providing a layer of
   obfuscation for an existing authenticated protocol like SSH or TLS.

   Like obfs3 and ScrambleSuit, the protocol has 2 phases: in the first phase
   both parties establish keys.  In the second, the parties exchange
   super-enciphered traffic.

1. Motivation

   ScrambleSuit [0] has been developed with the aim of improving the obfs3 [1]
   protocol to provide resilience against active attackers and to disguise
   flow signatures.
 
   ScrambleSuit like the existing obfs3 protocol uses UniformDH for the
   cryptographic handshake, which has severe performance implications due to
   modular exponentiation being a expensive operation.  Additionally, the key
   exchange is not authenticated so it is possible for active attackers to
   mount a man in the middle attack assuming they know the client/bridge
   shared secret (k_B).

   obfs4 attempts to address these shortcomings by using an authenticated key
   exchange mechanism based around the Tor Project's ntor handshake [2].
   Obfuscation of the Curve25519 public keys transmitted over the wire is
   accomplished via the Elligator 2 mapping [3].

2. Threat Model

   The threat model of obfs4 is the threat model of obfs2 [4] with added
   goals/modifications:

   obfs4 offers protection against passive Deep Packet Inspection machines
   that expect the obfs4 protocol.  Such machines should not be able to verify
   the existence of the obfs4 protocol without obtaining the server's Node ID
   and identity public key.

   obfs4 offers protection against active attackers attempting to probe for
   obfs4 servers.  Such machines should not be able to verify the existence
   of an obfs4 server without obtaining the server's Node ID and identity
   public key.

   obfs4 offers protection against active attackers that have obtained the
   server's Node ID and identity public key.  Such machines should not be
   able to impersonate the server without obtaining the server's identity
   private key.

   obfs4 offers protection against some non-content protocol fingerprints,
   specifically the packet size, and optionally packet timing.

   obfs4 provides integrity and confidentiality of the underlying traffic,
   and authentication of the server.

3. Notation and Terminology

   All Curve25519 keys and Elligator 2 representatives are transmitted in the
   Little Endian representation, for ease of integration with current
   Curve25519 and Elligator 2 implementations.  All other numeric fields are
   transmitted as Big Endian (Network byte order) values.

   HMAC-SHA256-128(k, s) is the HMAC-SHA256 digest of s with k as the key,
   truncated to 128 bits.

   x | y is the concatenation of x and y.

   A "byte" is an 8-bit octet.

4. Key Establishment Phase

   As part of the configuration, all obfs4 servers have a 20 byte Node ID
   (NODEID) and Curve25519 keypair (B,b) that is used to establish that the
   client knows about a given server and to authenticate the server.

   The server distributes the public component of the identity key (B) and
   NODEID to the client via an out-of-band mechanism.

   Data sent as part of the handshake are padded to random lengths to attempt to
   obfuscate the initial flow signature.  The constants used are as follows:

     MaximumHandshakeLength = 8192

       Maximum size of a handshake request or response, including padding.

     MarkLength = 16

       Length of M_C/M_S (A HMAC-SHA256-128 digest).

     MACLength = 16

       Length of MAC_C/MAC_S (A HMAC-SHA256-128 digest).

     RepresentativeLength = 32

       Length of a Elligator 2 representative of a Curve25519 public key.

     AuthLength = 32

       Length of the ntor AUTH tag (A HMAC-SHA256 digest).

     InlineSeedFrameLength = 53

       Length of a unpadded TYPE_PRNG_SEED frame.

     ServerHandshakeLength = 96

       The length of the non-padding data in a handshake response.

       RepresentativeLength + AuthLength + MarkLength + MACLength

     ServerMaxPadLength = 8096

       The maximum amount of padding in a handshake response.

       MaximumHandshakeLength - ServerHandshakeLength

     ServerMinPadLength = InlineSeedFrameLength

       The minimum amount of padding in a handshake response.

     ClientHandshakeLength = 64

       The length of the non-padding data in a handshake request.

       RepresentativeLength + MarkLength + MACLength

     ClientMinPadLength = 85

       The minimum amount of padding in a handshake request.

       (ServerHandshakeLength + ServerMinPadLength) - ClientHandshakeLength

     ClientMaxPadLength = 8128

       The maximum amount of padding in a handshake request.

       MaximumHandshakeLength - ClientHandshakeLength

   The amount of padding is chosen such that the smallest possible request and
   response (requests and responses with the minimum amount of padding) are
   equal in size.  For details on the InlineSeedFrameLength, see section 7.

   The client handshake process is as follows.

    1. The client generates an ephemeral Curve25519 keypair X,x and an
       Elligator 2 representative of the public component X'.

    2. The client sends a handshake request to the server where:

           X' = Elligator 2 representative of X (32 bytes)
           P_C = Random padding [ClientMinPadLength, ClientMaxPadLength] bytes
           M_C = HMAC-SHA256-128(B | NODEID, X')
           E = String representation of the number of hours since the UNIX
               epoch
           MAC_C = HMAC-SHA256-128(B | NODEID, X' | P_C | M_C | E)

           clientRequest = X' | P_C | M_C | MAC_C

    3. The client receives the serverResponse from the server.

    4. The client derives M_S from the serverResponse and uses it to locate
       MAC_S in the serverResponse.  It then calculates MAC_S and compares it
       with the value received from the server.  If M_S cannot be found or the
       MAC_S values do not match, the client MUST drop the connection.

    5. The client derives Y from Y' via the Elligator 2 map in the reverse
       direction.

    6. The client completes the client side of the ntor handshake, deriving
       the 256 bit shared secret (KEY_SEED), and the authentication tag
       (AUTH).  The client then compares the derived value of AUTH with that
       contained in the serverResponse.  If the AUTH values do not match, the
       client MUST drop the connection.

   The server handshake process is as follows.

    1. The server receives the clientRequest from the client.

    2. The server derives M_C from the clientRequest and uses it to locate
       MAC_C in the clientRequest.  It then calculates MAC_C and compares it
       with the value received from the client.  If M_C cannot be found or the
       MAC_C values do not match, the server MUST stop processing data from
       the client.

       Implementations MUST derive and compare multiple values of MAC_C with
       "E = {E - 1, E, E + 1}" to account for clock skew between the client
       and server.

       On the event of a failure at this point implementations SHOULD delay
       dropping the TCP connection from the client by a random interval to
       make active probing more difficult.

    3. The server derives X from X' via the Elligator 2 map in the reverse
       direction.

    4. The server generates an ephemeral Curve25519 keypair Y, y and an
       Elligator 2 representative of the public component Y'.

    5. The server completes the server side of the ntor handshake, deriving
       the 256 bit shared secret (KEY_SEED), and the authentication tag
       (AUTH).

    6. The server sends a handshake response to the client where:

           Y' = Elligator 2 Representative of Y (32 bytes)
           AUTH = The ntor authentication tag (32 bytes)
           P_S = Random padding [ServerMinPadLength, ServerMaxPadLength] bytes
           M_S = HMAC-SHA256-128(B | NODEID, Y')
           E' = E from the client request
           MAC_S = HMAC-SHA256-128(B | NODEID, Y' | AUTH | P_S | M_S | E')

           serverResponse = Y' | AUTH | P_S | M_S | MAC_S

   At the point that each side finishes the handshake, they have a 256 bit
   shared secret KEY_SEED that is then extracted/expanded via the ntor KDF to
   produce the 128 bytes of keying material used to encrypt/authenticate the
   data.

   The keying material is used as follows:

     Bytes 000:031 - Server to Client 256 bit NaCl secretbox key.
     Bytes 032:047 - Server to Client 128 bit NaCl secretbox nonce prefix.
     Bytes 048:063 - Server to Client 128 bit SipHash-2-4 key.

     Bytes 064:095 - Client to Server 256 bit NaCl secretbox key.
     Bytes 096:111 - Client to Server NaCl secretbox nonce prefix.
     Bytes 112:127 - Client to Server 128 bit SipHash-2-4 key.

5. Data Transfer Phase

   Once both sides have completed the handshake, they transfer application
   data broken up into "packets", that are then encrypted and authenticated in
   NaCl crypto_secretbox_xsalsa20poly1305 [5] "frames".

   +------------+----------+--------+--------------+------------+------------+
   |  2 bytes   | 16 bytes | 1 byte |   2 bytes    | (optional) | (optional) |
   | Frame len. |   Tag    |  Type  | Payload len. |  Payload   |  Padding   |
   +------------+----------+--------+--------------+------------+------------+
    \_ Obfs.  _/ \___________ NaCl secretbox (Poly1305/XSalsa20) ___________/

   The frame length refers to the length of the succeeding secretbox.  To
   avoid transmitting identifiable length fields in stream, the frame length
   is obfuscated by XORing the value with the SipHash-2-4[4] digest of the
   secretbox nonce, truncated to 2 bytes.  As the nonce is deterministic,
   decoding the length is done by deriving the nonce that will be used to
   unseal the next secret box, calculating the truncated SipHash-24 digest,
   and XORing the digest with the obfuscated frame length to obtain the
   length of the secretbox.

   The payload length refers to the length of the payload portion of the frame
   and does not include the padding.  It is possible for the payload length to
   be 0 in which case all the remaining data is authenticated and decrypted,
   but ignored.

   The maximum allowed frame length is 1448 bytes, which allows up to 1427
   bytes of useful payload to be transmitted per "frame".

   If unsealing a secretbox ever fails (due to a Tag mismatch), implementations
   MUST drop the connection.

   The type field is used to denote the type of payload (if any) contained in
   each packet.

     TYPE_PAYLOAD (0x00):

         The entire payload is to be treated as application data.

     TYPE_PRNG_SEED (0x01):

         The entire payload is to be treated as seeding material for the
         protocol polymorphism PRNG.  The format is 32 bytes of seeding
         material.

   Implementations SHOULD ignore unknown packet types for the purposes of
   forward compatibility, though each frame MUST still be authenticated and
   decrypted.

6. Protocol Polymorphism

   Implementations MUST implement protocol polymorphism to obfuscate the obfs4
   flow signature.  The implementation should follow that of ScrambleSuit (See
   "ScrambleSuit Protocol Specification", section 4).  Like with ScrambleSuit,
   implementations MAY omit inter-arrival time obfuscation as a performance
   trade-off.

   As an optimization, implementations MAY treat the TYPE_PRNG_SEED frame as
   part of the serverResponse if it always sends the frame immediately
   following the serverResponse body.  If implementations chose to do this,
   the TYPE_PRNG_SEED frame MUST have 0 bytes of padding, and P_S MUST
   be generated with a ServerMinPadLength of 0 (P_S consists of [0,8096]
   bytes of random data).  The calculation of ClientMinPadLength however is
   unchanged (P_C still consists of [85,8128] bytes of random data).
 
7. References

   [0]: https://gitweb.torproject.org/user/phw/scramblesuit.git/blob/HEAD:/doc/scramblesuit-spec.txt

   [1]: https://gitweb.torproject.org/pluggable-transports/obfsproxy.git/blob/HEAD:/doc/obfs3/obfs3-protocol-spec.txt

   [2]: https://gitweb.torproject.org/torspec.git/blob/HEAD:/proposals/216-ntor-handshake.txt

   [3]: http://elligator.cr.yp.to/elligator-20130828.pdf

   [4]: https://gitweb.torproject.org/pluggable-transports/obfsproxy.git/blob/HEAD:/doc/obfs2/obfs2-threat-model.txt

   [5]: http://nacl.cr.yp.to/secretbox.html

   [6]: https://131002.net/siphash/

8. Acknowledgments

   Much of the protocol and this specification document is derived from the
   ScrambleSuit protocol and specification by Philipp Winter.