tor-spec.txt

                         Tor Protocol Specification

                              Roger Dingledine
                               Nick Mathewson

Note: This document aims to specify Tor as currently implemented, though it
may take it a little time to become fully up to date. Future versions of Tor
may implement improved protocols, and compatibility is not guaranteed.
We may or may not remove compatibility notes for other obsolete versions of
Tor as they become obsolete.

This specification is not a design document; most design criteria
are not examined.  For more information on why Tor acts as it does,
see tor-design.pdf.

0. Preliminaries

      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
      RFC 2119.

0.1.  Notation and encoding

   PK -- a public key.
   SK -- a private key.
   K  -- a key for a symmetric cipher.

   a|b -- concatenation of 'a' and 'b'.

   [A0 B1 C2] -- a three-byte sequence, containing the bytes with
   hexadecimal values A0, B1, and C2, in that order.

   H(m) -- a cryptographic hash of m.

   We use "byte" and "octet" interchangeably.  Possibly we shouldn't.

0.1.1. Encoding integers

   Unless we explicitly say otherwise below, all numeric values in the
   Tor protocol are encoded in network (big-endian) order.  So a "32-bit
   integer" means a big-endian 32-bit integer; a "2-byte" integer means
   a big-endian 16-bit integer, and so forth.

0.2. Security parameters

   Tor uses a stream cipher, a public-key cipher, the Diffie-Hellman
   protocol, and a hash function.

   KEY_LEN -- the length of the stream cipher's key, in bytes.

   PK_ENC_LEN -- the length of a public-key encrypted message, in bytes.
   PK_PAD_LEN -- the number of bytes added in padding for public-key
     encryption, in bytes. (The largest number of bytes that can be encrypted
     in a single public-key operation is therefore PK_ENC_LEN-PK_PAD_LEN.)

   DH_LEN -- the number of bytes used to represent a member of the
     Diffie-Hellman group.
   DH_SEC_LEN -- the number of bytes used in a Diffie-Hellman private key (x).

   HASH_LEN -- the length of the hash function's output, in bytes.

   PAYLOAD_LEN -- The longest allowable cell payload, in bytes. (509)

   CELL_LEN(v) -- The length of a Tor cell, in bytes, for link protocol
      version v.
       CELL_LEN(v) = 512    if v is less than 4;
                   = 514    otherwise.

0.3. Ciphers

   These are the ciphers we use _unless otherwise specified_.  Several of
   them are deprecated for new use.

   For a stream cipher, unless otherwise specified, we use 128-bit AES in
   counter mode, with an IV of all 0 bytes.  (We also require AES256.)

   For a public-key cipher, unless otherwise specified, we use RSA with
   1024-bit keys and a fixed exponent of 65537.  We use OAEP-MGF1
   padding, with SHA-1 as its digest function.  We leave the optional
   "Label" parameter unset. (For OAEP padding, see
   ftp://ftp.rsasecurity.com/pub/pkcs/pkcs-1/pkcs-1v2-1.pdf)

   We also use the Curve25519 group and the Ed25519 signature format in
   several places.

   For Diffie-Hellman, unless otherwise specified, we use a generator
   (g) of 2.  For the modulus (p), we use the 1024-bit safe prime from
   rfc2409 section 6.2 whose hex representation is:

     "FFFFFFFFFFFFFFFFC90FDAA22168C234C4C6628B80DC1CD129024E08"
     "8A67CC74020BBEA63B139B22514A08798E3404DDEF9519B3CD3A431B"
     "302B0A6DF25F14374FE1356D6D51C245E485B576625E7EC6F44C42E9"
     "A637ED6B0BFF5CB6F406B7EDEE386BFB5A899FA5AE9F24117C4B1FE6"
     "49286651ECE65381FFFFFFFFFFFFFFFF"

   As an optimization, implementations SHOULD choose DH private keys (x) of
   320 bits.  Implementations that do this MUST never use any DH key more
   than once.
   [May other implementations reuse their DH keys?? -RD]
   [Probably not. Conceivably, you could get away with changing DH keys once
   per second, but there are too many oddball attacks for me to be
   comfortable that this is safe. -NM]

   For a hash function, unless otherwise specified, we use SHA-1.

   KEY_LEN=16.
   DH_LEN=128; DH_SEC_LEN=40.
   PK_ENC_LEN=128; PK_PAD_LEN=42.
   HASH_LEN=20.

   We also use SHA256 and SHA3-256 in some places.

   When we refer to "the hash of a public key", unless otherwise
   specified, we mean the SHA-1 hash of the DER encoding of an ASN.1 RSA
   public key (as specified in PKCS.1).

   All "random" values MUST be generated with a cryptographically
   strong pseudorandom number generator seeded from a strong entropy
   source, unless otherwise noted.

0.4. A bad hybrid encryption algorithm, for legacy purposes.

   Some specifications will refer to the "legacy hybrid encryption" of a
   byte sequence M with a public key PK.  It is computed as follows:

      1. If the length of M is no more than PK_ENC_LEN-PK_PAD_LEN,
         pad and encrypt M with PK.
      2. Otherwise, generate a KEY_LEN byte random key K.
         Let M1 = the first PK_ENC_LEN-PK_PAD_LEN-KEY_LEN bytes of M,
         and let M2 = the rest of M.
         Pad and encrypt K|M1 with PK.  Encrypt M2 with our stream cipher,
         using the key K.  Concatenate these encrypted values.

   Note that this "hybrid encryption" approach does not prevent
   an attacker from adding or removing bytes to the end of M. It also
   allows attackers to modify the bytes not covered by the OAEP --
   see Goldberg's PET2006 paper for details.  Do not use it as the basis
   for new protocols! Also note that as used in Tor's protocols, case 1
   never occurs.

1. System overview

   Tor is a distributed overlay network designed to anonymize
   low-latency TCP-based applications such as web browsing, secure shell,
   and instant messaging. Clients choose a path through the network and
   build a ``circuit'', in which each node (or ``onion router'' or ``OR'')
   in the path knows its predecessor and successor, but no other nodes in
   the circuit.  Traffic flowing down the circuit is sent in fixed-size
   ``cells'', which are unwrapped by a symmetric key at each node (like
   the layers of an onion) and relayed downstream.

1.1. Keys and names

   Every Tor relay has multiple public/private keypairs:

   These are 1024-bit RSA keys:

    - A long-term signing-only "Identity key" used to sign documents and
      certificates, and used to establish relay identity.
    - A medium-term TAP "Onion key" used to decrypt onion skins when accepting
      circuit extend attempts.  (See 5.1.)  Old keys MUST be accepted for a
      while after they are no longer advertised.  Because of this,
      relays MUST retain old keys for a while after they're rotated.  (See
      "onion key lifetime parameters" in dir-spec.txt.)
    - A short-term "Connection key" used to negotiate TLS connections.
      Tor implementations MAY rotate this key as often as they like, and
      SHOULD rotate this key at least once a day.

   This is Curve25519 key:

    - A medium-term ntor "Onion key" used to handle onion key handshakes when
      accepting incoming circuit extend requests.  As with TAP onion keys,
      old ntor keys MUST be accepted for at least one week after they are no
      longer advertised.  Because of this, relays MUST retain old keys for a
      while after they're rotated. (See "onion key lifetime parameters" in
      dir-spec.txt.)

   These are Ed25519 keys:

    - A long-term "master identity" key.  This key never
      changes; it is used only to sign the "signing" key below.  It may be
      kept offline.
    - A medium-term "signing" key.  This key is signed by the master identity
      key, and must be kept online.  A new one should be generated
      periodically.  It signs nearly everything else.
    - A short-term "link authentication" key, used to authenticate
      the link handshake: see section 4 below.  This key is signed
      by the "signing" key, and should be regenerated frequently.

   The RSA identity key and Ed25519 master identity key together identify a
   router uniquely.  Once a router has used an Ed25519 master identity key
   together with a given RSA identity key, neither of those keys may ever be
   used with a different key.

2. Connections

   Connections between two Tor relays, or between a client and a relay,
   use TLS/SSLv3 for link authentication and encryption.  All
   implementations MUST support the SSLv3 ciphersuite
   "TLS_DHE_RSA_WITH_AES_128_CBC_SHA" if it is available. They SHOULD
   support better ciphersuites if available.

   There are three ways to perform TLS handshakes with a Tor server.  In
   the first way, "certificates-up-front", both the initiator and
   responder send a two-certificate chain as part of their initial
   handshake.  (This is supported in all Tor versions.)  In the second
   way, "renegotiation", the responder provides a single certificate,
   and the initiator immediately performs a TLS renegotiation.  (This is
   supported in Tor 0.2.0.21 and later.)  And in the third way,
   "in-protocol", the initial TLS negotiation completes, and the
   parties bootstrap themselves to mutual authentication via use of the
   Tor protocol without further TLS handshaking.  (This is supported in
   0.2.3.6-alpha and later.)

   Each of these options provides a way for the parties to learn it is
   available: a client does not need to know the version of the Tor
   server in order to connect to it properly.

   In "certificates up-front" (a.k.a "the v1 handshake"),
   the connection initiator always sends a
   two-certificate chain, consisting of an X.509 certificate using a
   short-term connection public key and a second, self-signed X.509
   certificate containing its identity key.  The other party sends a similar
   certificate chain.  The initiator's ClientHello MUST NOT include any
   ciphersuites other than:

     TLS_DHE_RSA_WITH_AES_256_CBC_SHA
     TLS_DHE_RSA_WITH_AES_128_CBC_SHA
     SSL_DHE_RSA_WITH_3DES_EDE_CBC_SHA

   In "renegotiation" (a.k.a. "the v2 handshake"),
   the connection initiator sends no certificates, and
   the responder sends a single connection certificate.  Once the TLS
   handshake is complete, the initiator renegotiates the handshake, with each
   party sending a two-certificate chain as in "certificates up-front".
   The initiator's ClientHello MUST include at least one ciphersuite not in
   the list above -- that's how the initiator indicates that it can
   handle this handshake.  For other considerations on the initiator's
   ClientHello, see section 2.1 below.

   In "in-protocol" (a.k.a. "the v3 handshake"), the initiator sends no
   certificates, and the
   responder sends a single connection certificate.  The choice of
   ciphersuites must be as in a "renegotiation" handshake.  There are
   additionally a set of constraints on the connection certificate,
   which the initiator can use to learn that the in-protocol handshake
   is in use.  Specifically, at least one of these properties must be
   true of the certificate:

      * The certificate is self-signed
      * Some component other than "commonName" is set in the subject or
        issuer DN of the certificate.
      * The commonName of the subject or issuer of the certificate ends
        with a suffix other than ".net".
      * The certificate's public key modulus is longer than 1024 bits.

   The initiator then sends a VERSIONS cell to the responder, which then
   replies with a VERSIONS cell; they have then negotiated a Tor
   protocol version.  Assuming that the version they negotiate is 3 or higher
   (the only ones specified for use with this handshake right now), the
   responder sends a CERTS cell, an AUTH_CHALLENGE cell, and a NETINFO
   cell to the initiator, which may send either CERTS, AUTHENTICATE,
   NETINFO if it wants to authenticate, or just NETINFO if it does not.

   For backward compatibility between later handshakes and "certificates
   up-front", the ClientHello of an initiator that supports a later
   handshake MUST include at least one ciphersuite other than those listed
   above. The connection responder examines the initiator's ciphersuite list
   to see whether it includes any ciphers other than those included in the
   list above.  If extra ciphers are included, the responder proceeds as in
   "renegotiation" and "in-protocol": it sends a single certificate and
   does not request
   client certificates.  Otherwise (in the case that no extra ciphersuites
   are included in the ClientHello) the responder proceeds as in
   "certificates up-front": it requests client certificates, and sends a
   two-certificate chain.  In either case, once the responder has sent its
   certificate or certificates, the initiator counts them.  If two
   certificates have been sent, it proceeds as in "certificates up-front";
   otherwise, it proceeds as in "renegotiation" or "in-protocol".

   To decide whether to do "renegotiation" or "in-protocol", the
   initiator checks whether the responder's initial certificate matches
   the criteria listed above.

   All new relay implementations of the Tor protocol MUST support
   backwards-compatible renegotiation; clients SHOULD do this too.  If
   this is not possible, new client implementations MUST support both
   "renegotiation" and "in-protocol" and use the router's
   published link protocols list (see dir-spec.txt on the "protocols" entry)
   to decide which to use.

   In all of the above handshake variants, certificates sent in the clear
   SHOULD NOT include any strings to identify the host as a Tor relay. In
   the "renegotiation" and "backwards-compatible renegotiation" steps, the
   initiator SHOULD choose a list of ciphersuites and TLS extensions
   to mimic one used by a popular web browser.

   Even though the connection protocol is identical, we will think of the
   initiator as either an onion router (OR) if it is willing to relay
   traffic for other Tor users, or an onion proxy (OP) if it only handles
   local requests. Onion proxies SHOULD NOT provide long-term-trackable
   identifiers in their handshakes.

   In all handshake variants, once all certificates are exchanged, all
   parties receiving certificates must confirm that the identity key is as
   expected.  If the key is not as expected, the party must close the
   connection.

   (When initiating a connection, if a reasonably live consensus is
   available, then the expected identity key is taken from that
   consensus. But when initiating a connection otherwise, the expected
   identity key is the one given in the hard-coded authority or
   fallback list.  Finally, when creating a connection because of an
   EXTEND/EXTEND2 cell, the expected identity key is the one given in
   the cell.)

   When connecting to an OR, all parties SHOULD reject the connection if that
   OR has a malformed or missing certificate.  When accepting an incoming
   connection, an OR SHOULD NOT reject incoming connections from parties with
   malformed or missing certificates.  (However, an OR should not believe
   that an incoming connection is from another OR unless the certificates
   are present and well-formed.)

   [Before version 0.1.2.8-rc, ORs rejected incoming connections from ORs and
   OPs alike if their certificates were missing or malformed.]

   Once a TLS connection is established, the two sides send cells
   (specified below) to one another.  Cells are sent serially.  Standard
   cells are CELL_LEN(link_proto) bytes long, but variable-length cells
   also exist; see Section 3.  Cells may be sent embedded in TLS records
   of any size or divided across TLS records, but the framing of TLS
   records MUST NOT leak information about the type or contents of the
   cells.

   TLS connections are not permanent. Either side MAY close a connection
   if there are no circuits running over it and an amount of time
   (KeepalivePeriod, defaults to 5 minutes) has passed since the last time
   any traffic was transmitted over the TLS connection.  Clients SHOULD
   also hold a TLS connection with no circuits open, if it is likely that a
   circuit will be built soon using that connection.

   Client-only Tor instances are encouraged to avoid using handshake
   variants that include certificates, if those certificates provide
   any persistent tags to the relays they contact. If clients do use
   certificates, they SHOULD NOT keep using the same certificates when
   their IP address changes.  Clients MAY send certificates using any
   of the above handshake variants.

2.1. Picking TLS ciphersuites

   Clients SHOULD send a ciphersuite list chosen to emulate some popular
   web browser or other program common on the internet. Clients may send
   the "Fixed Cipheruite List" below.  If they do not, they MUST NOT
   advertise any ciphersuite that they cannot actually support, unless that
   cipher is one not supported by OpenSSL 1.0.1.

   The fixed ciphersuite list is:

     TLS1_ECDHE_ECDSA_WITH_AES_256_CBC_SHA
     TLS1_ECDHE_RSA_WITH_AES_256_CBC_SHA
     TLS1_DHE_RSA_WITH_AES_256_SHA
     TLS1_DHE_DSS_WITH_AES_256_SHA
     TLS1_ECDH_RSA_WITH_AES_256_CBC_SHA
     TLS1_ECDH_ECDSA_WITH_AES_256_CBC_SHA
     TLS1_RSA_WITH_AES_256_SHA
     TLS1_ECDHE_ECDSA_WITH_RC4_128_SHA
     TLS1_ECDHE_ECDSA_WITH_AES_128_CBC_SHA
     TLS1_ECDHE_RSA_WITH_RC4_128_SHA
     TLS1_ECDHE_RSA_WITH_AES_128_CBC_SHA
     TLS1_DHE_RSA_WITH_AES_128_SHA
     TLS1_DHE_DSS_WITH_AES_128_SHA
     TLS1_ECDH_RSA_WITH_RC4_128_SHA
     TLS1_ECDH_RSA_WITH_AES_128_CBC_SHA
     TLS1_ECDH_ECDSA_WITH_RC4_128_SHA
     TLS1_ECDH_ECDSA_WITH_AES_128_CBC_SHA
     SSL3_RSA_RC4_128_MD5
     SSL3_RSA_RC4_128_SHA
     TLS1_RSA_WITH_AES_128_SHA
     TLS1_ECDHE_ECDSA_WITH_DES_192_CBC3_SHA
     TLS1_ECDHE_RSA_WITH_DES_192_CBC3_SHA
     SSL3_EDH_RSA_DES_192_CBC3_SHA
     SSL3_EDH_DSS_DES_192_CBC3_SHA
     TLS1_ECDH_RSA_WITH_DES_192_CBC3_SHA
     TLS1_ECDH_ECDSA_WITH_DES_192_CBC3_SHA
     SSL3_RSA_FIPS_WITH_3DES_EDE_CBC_SHA
     SSL3_RSA_DES_192_CBC3_SHA
     [*] The "extended renegotiation is supported" ciphersuite, 0x00ff, is
         not counted when checking the list of ciphersuites.

   If the client sends the Fixed Ciphersuite List, the responder MUST NOT
   select any ciphersuite besides TLS_DHE_RSA_WITH_AES_256_CBC_SHA,
   TLS_DHE_RSA_WITH_AES_128_CBC_SHA, and SSL_DHE_RSA_WITH_3DES_EDE_CBC_SHA:
   such ciphers might not actually be supported by the client.

   If the client sends a v2+ ClientHello with a list of ciphers other then
   the Fixed Ciphersuite List, the responder can trust that the client
   supports every cipher advertised in that list, so long as that ciphersuite
   is also supported by OpenSSL 1.0.1.

   Responders MUST NOT select any TLS ciphersuite that lacks ephemeral keys,
   or whose symmetric keys are less then KEY_LEN bits, or whose digests are
   less than HASH_LEN bits.  Responders SHOULD NOT select any SSLv3
   ciphersuite other than the DHE+3DES suites listed above.

2.2. TLS security considerations

   Implementations MUST NOT allow TLS session resumption -- it can
   exacerbate some attacks (e.g. the "Triple Handshake" attack from
   Feb 2013), and it plays havoc with forward secrecy guarantees.

   Implementations SHOULD NOT allow TLS compression -- although we don't
   know a way to apply a CRIME-style attack to current Tor directly,
   it's a waste of resources.

3. Cell Packet format

   The basic unit of communication for onion routers and onion
   proxies is a fixed-width "cell".

   On a version 1 connection, each cell contains the following
   fields:

        CircID                                [CIRCID_LEN bytes]
        Command                               [1 byte]
        Payload (padded with padding bytes)   [PAYLOAD_LEN bytes]

   On a version 2 or higher connection, all cells are as in version 1
   connections, except for variable-length cells, whose format is:

        CircID                                [CIRCID_LEN octets]
        Command                               [1 octet]
        Length                                [2 octets; big-endian integer]
        Payload (some commands MAY pad)       [Length bytes]

   Most variable-length cells MAY be padded with padding bytes, except
   for VERSIONS cells, which MUST NOT contain any additional bytes.
   (The payload of VPADDING cells consists of padding bytes.)

   On a version 2 connection, variable-length cells are indicated by a
   command byte equal to 7 ("VERSIONS").  On a version 3 or
   higher connection, variable-length cells are indicated by a command
   byte equal to 7 ("VERSIONS"), or greater than or equal to 128.

   CIRCID_LEN is 2 for link protocol versions 1, 2, and 3.  CIRCID_LEN
   is 4 for link protocol version 4 or higher.  The first VERSIONS cell,
   and any cells sent before the first VERSIONS cell, always have
   CIRCID_LEN == 2 for backward compatibility.

   The CircID field determines which circuit, if any, the cell is
   associated with.

   The 'Command' field of a fixed-length cell holds one of the following
   values:

         0 -- PADDING     (Padding)                 (See Sec 7.2)
         1 -- CREATE      (Create a circuit)        (See Sec 5.1)
         2 -- CREATED     (Acknowledge create)      (See Sec 5.1)
         3 -- RELAY       (End-to-end data)         (See Sec 5.5 and 6)
         4 -- DESTROY     (Stop using a circuit)    (See Sec 5.4)
         5 -- CREATE_FAST (Create a circuit, no PK) (See Sec 5.1)
         6 -- CREATED_FAST (Circuit created, no PK) (See Sec 5.1)
         8 -- NETINFO     (Time and address info)   (See Sec 4.5)
         9 -- RELAY_EARLY (End-to-end data; limited)(See Sec 5.6)
         10 -- CREATE2    (Extended CREATE cell)    (See Sec 5.1)
         11 -- CREATED2   (Extended CREATED cell)    (See Sec 5.1)
         12 -- PADDING_NEGOTIATE   (Padding negotiation)    (See Sec 7.2)

    Variable-length command values are:

         7 -- VERSIONS    (Negotiate proto version) (See Sec 4)
         128 -- VPADDING  (Variable-length padding) (See Sec 7.2)
         129 -- CERTS     (Certificates)            (See Sec 4.2)
         130 -- AUTH_CHALLENGE (Challenge value)    (See Sec 4.3)
         131 -- AUTHENTICATE (Client authentication)(See Sec 4.5)
         132 -- AUTHORIZE (Client authorization)    (Not yet used)

   The interpretation of 'Payload' depends on the type of the cell.

      VPADDING/PADDING:
               Payload contains padding bytes.
      CREATE/CREATE2:  Payload contains the handshake challenge.
      CREATED/CREATED2: Payload contains the handshake response.
      RELAY/RELAY_EARLY: Payload contains the relay header and relay body.
      DESTROY: Payload contains a reason for closing the circuit.
               (see 5.4)

   Upon receiving any other value for the command field, an OR must
   drop the cell.  Since more cell types may be added in the future, ORs
   should generally not warn when encountering unrecognized commands.

   The cell is padded up to the cell length with padding bytes.

   Senders set padding bytes depending on the cell's command:

      VERSIONS:  Payload MUST NOT contain padding bytes.
      AUTHORIZE: Payload is unspecified and reserved for future use.
      Other variable-length cells:
                 Payload MAY contain padding bytes at the end of the cell.
                 Padding bytes SHOULD be set to NUL.
      RELAY/RELAY_EARLY: Payload MUST be padded to PAYLOAD_LEN with padding
                  bytes. Padding bytes SHOULD be set to random values.
      Other fixed-length cells:
                 Payload MUST be padded to PAYLOAD_LEN with padding bytes.
                 Padding bytes SHOULD be set to NUL.

   We recommend random padding in RELAY/RELAY_EARLY cells, so that the cell
   content is unpredictable. See the format of relay cells in section 6.1
   for detail.

   For other cells, TLS authenticates cell content, so randomized padding
   bytes are redundant.

   Receivers MUST ignore padding bytes.

   PADDING cells are currently used to implement connection keepalive.
   If there is no other traffic, ORs and OPs send one another a PADDING
   cell every few minutes.

   CREATE, CREATE2, CREATED, CREATED2, and DESTROY cells are used to
   manage circuits; see section 5 below.

   RELAY cells are used to send commands and data along a circuit; see
   section 6 below.

   VERSIONS and NETINFO cells are used to set up connections in link
   protocols v2 and higher; in link protocol v3 and higher, CERTS,
   AUTH_CHALLENGE, and AUTHENTICATE may also be used.  See section 4
   below.

4. Negotiating and initializing connections

   After Tor instances negotiate handshake with either the "renegotiation" or
   "in-protocol" handshakes, they must exchange a set of cells to set up
   the Tor connection and make it "open" and usable for circuits.

   When the renegotiation handshake is used, both parties immediately
   send a VERSIONS cell (4.1 below), and after negotiating a link
   protocol version (which will be 2), each send a NETINFO cell (4.5
   below) to confirm their addresses and timestamps.  No other intervening
   cell types are allowed.

   When the in-protocol handshake is used, the initiator sends a
   VERSIONS cell to indicate that it will not be renegotiating.  The
   responder sends a VERSIONS cell, a CERTS cell (4.2 below) to give the
   initiator the certificates it needs to learn the responder's
   identity, an AUTH_CHALLENGE cell (4.3) that the initiator must include
   as part of its answer if it chooses to authenticate, and a NETINFO
   cell (4.5).  As soon as it gets the CERTS cell, the initiator knows
   whether the responder is correctly authenticated.  At this point the
   initiator behaves differently depending on whether it wants to
   authenticate or not. If it does not want to authenticate, it MUST
   send a NETINFO cell.  If it does want to authenticate, it MUST send a
   CERTS cell, an AUTHENTICATE cell (4.4), and a NETINFO.  When this
   handshake is in use, the first cell must be VERSIONS, VPADDING, or
   AUTHORIZE, and no other cell type is allowed to intervene besides
   those specified, except for VPADDING cells.

   The AUTHORIZE cell type is reserved for future use by scanning-resistance
   designs.

   [Tor versions before 0.2.3.11-alpha did not recognize the AUTHORIZE cell,
   and did not permit any command other than VERSIONS as the first cell of
   the in-protocol handshake.]

4.1. Negotiating versions with VERSIONS cells

   There are multiple instances of the Tor link connection protocol.  Any
   connection negotiated using the "certificates up front" handshake (see
   section 2 above) is "version 1".  In any connection where both parties
   have behaved as in the "renegotiation" handshake, the link protocol
   version must be 2.  In any connection where both parties have behaved
   as in the "in-protocol" handshake, the link protocol must be 3 or higher.

   To determine the version, in any connection where the "renegotiation"
   or "in-protocol" handshake was used (that is, where the responder
   sent only one certificate at first and where the initiator did not
   send any certificates in the first negotiation), both parties MUST
   send a VERSIONS cell.  In "renegotiation", they send a VERSIONS cell
   right after the renegotiation is finished, before any other cells are
   sent.  In "in-protocol", the initiator sends a VERSIONS cell
   immediately after the initial TLS handshake, and the responder
   replies immediately with a VERSIONS cell. (As an exception to this rule,
   if both sides support the "in-protocol" handshake, either side may send
   VPADDING cells at any time.)

   The payload in a VERSIONS cell is a series of big-endian two-byte
   integers.  Both parties MUST select as the link protocol version the
   highest number contained both in the VERSIONS cell they sent and in the
   versions cell they received.  If they have no such version in common,
   they cannot communicate and MUST close the connection.  Either party MUST
   close the connection if the versions cell is not well-formed (for example,
   if it contains an odd number of bytes).

   Any VERSIONS cells sent after the first VERSIONS cell MUST be ignored.
   (To be interpreted correctly, later VERSIONS cells MUST have a CIRCID_LEN
   matching the version negotiated with the first VERSIONS cell.)

   Since the version 1 link protocol does not use the "renegotiation"
   handshake, implementations MUST NOT list version 1 in their VERSIONS
   cell.  When the "renegotiation" handshake is used, implementations
   MUST list only the version 2.  When the "in-protocol" handshake is
   used, implementations MUST NOT list any version before 3, and SHOULD
   list at least version 3.

   Link protocols differences are:

     1 -- The "certs up front" handshake.
     2 -- Uses the renegotiation-based handshake. Introduces
          variable-length cells.
     3 -- Uses the in-protocol handshake.
     4 -- Increases circuit ID width to 4 bytes.
     5 -- Adds support for link padding and negotiation (padding-spec.txt).


4.2. CERTS cells

   The CERTS cell describes the keys that a Tor instance is claiming
   to have.  It is a variable-length cell.  Its payload format is:

        N: Number of certs in cell            [1 octet]
        N times:
           CertType                           [1 octet]
           CLEN                               [2 octets]
           Certificate                        [CLEN octets]

   Any extra octets at the end of a CERTS cell MUST be ignored.

     Relevant certType values are:
        1: Link key certificate certified by RSA1024 identity
        2: RSA1024 Identity certificate, self-signed.
        3: RSA1024 AUTHENTICATE cell link certificate, signed with RSA1024 key.
        4: Ed25519 signing key, signed with identity key.
        5: TLS link certificate, signed with ed25519 signing key.
        6: Ed25519 AUTHENTICATE cell key, signed with ed25519 signing key.
        7: Ed25519 identity, signed with RSA identity.

   The certificate format for certificate types 1-3 is DER encoded
   X509.  For others, the format is as documented in cert-spec.txt.
   Note that type 7 uses a different format from types 4-6.

   A CERTS cell may have no more than one certificate of each CertType.


   To authenticate the responder as having a given Ed25519,RSA identity key
   combination, the initiator MUST check the following.

     * The CERTS cell contains exactly one CertType 2 "ID" certificate.
     * The CERTS cell contains exactly one CertType 4 Ed25519
       "Id->Signing" cert.
     * The CERTS cell contains exactly one CertType 5 Ed25519
       "Signing->link" certificate.
     * The CERTS cell contains exactly one CertType 7 "RSA->Ed25519"
       cross-certificate.
     * All X.509 certificates above have validAfter and validUntil dates;
       no X.509 or Ed25519 certificates are expired.
     * All certificates are correctly signed.
     * The certified key in the Signing->Link certificate matches the
       SHA256 digest of the certificate that was used to
       authenticate the TLS connection.
     * The identity key listed in the ID->Signing cert was used to
       sign the ID->Signing Cert.
     * The Signing->Link cert was signed with the Signing key listed
       in the ID->Signing cert.
     * The RSA->Ed25519 cross-certificate certifies the Ed25519
       identity, and is signed with the RSA identity listed in the
       "ID" certificate.
     * The certified key in the ID certificate is a 1024-bit RSA key.
     * The RSA ID certificate is correctly self-signed.

   To authenticate the responder as having a given RSA identity only,
   the initiator MUST check the following:

     * The CERTS cell contains exactly one CertType 1 "Link" certificate.
     * The CERTS cell contains exactly one CertType 2 "ID" certificate.
     * Both certificates have validAfter and validUntil dates that
       are not expired.
     * The certified key in the Link certificate matches the
       link key that was used to negotiate the TLS connection.
     * The certified key in the ID certificate is a 1024-bit RSA key.
     * The certified key in the ID certificate was used to sign both
       certificates.
     * The link certificate is correctly signed with the key in the
       ID certificate
     * The ID certificate is correctly self-signed.

   In both cases above, checking these conditions is sufficient to
   authenticate that the initiator is talking to the Tor node with the
   expected identity, as certified in the ID certificate(s).


   To authenticate the initiator as having a given Ed25519,RSA
   identity key combination, the responder MUST check the following:

     * The CERTS cell contains exactly one CertType 2 "ID" certificate.
     * The CERTS cell contains exactly one CertType 4 Ed25519
       "Id->Signing" certificate.
     * The CERTS cell contains exactly one CertType 6 Ed25519
       "Signing->auth" certificate.
     * The CERTS cell contains exactly one CertType 7 "RSA->Ed25519"
       cross-certificate.
     * All X.509 certificates above have validAfter and validUntil dates;
       no X.509 or Ed25519 certificates are expired.
     * All certificates are correctly signed.
     * The identity key listed in the ID->Signing cert was used to
       sign the ID->Signing Cert.
     * The Signing->AUTH cert was signed with the Signing key listed
       in the ID->Signing cert.
     * The RSA->Ed25519 cross-certificate certifies the Ed25519
       identity, and is signed with the RSA identity listed in the
       "ID" certificate.
     * The certified key in the ID certificate is a 1024-bit RSA key.
     * The RSA ID certificate is correctly self-signed.


   To authenticate the initiator as having an RSA identity key only,
   the responder MUST check the following:

     * The CERTS cell contains exactly one CertType 3 "AUTH" certificate.
     * The CERTS cell contains exactly one CertType 2 "ID" certificate.
     * Both certificates have validAfter and validUntil dates that
       are not expired.
     * The certified key in the AUTH certificate is a 1024-bit RSA key.
     * The certified key in the ID certificate is a 1024-bit RSA key.
     * The certified key in the ID certificate was used to sign both
       certificates.
     * The auth certificate is correctly signed with the key in the
       ID certificate.
     * The ID certificate is correctly self-signed.

   Checking these conditions is NOT sufficient to authenticate that the
   initiator has the ID it claims; to do so, the cells in 4.3 and 4.4
   below must be exchanged.


4.3. AUTH_CHALLENGE cells

   An AUTH_CHALLENGE cell is a variable-length cell with the following
   fields:

       Challenge [32 octets]
       N_Methods [2 octets]
       Methods   [2 * N_Methods octets]

   It is sent from the responder to the initiator. Initiators MUST
   ignore unexpected bytes at the end of the cell.  Responders MUST
   generate every challenge independently using a strong RNG or PRNG.

   The Challenge field is a randomly generated string that the
   initiator must sign (a hash of) as part of authenticating.  The
   methods are the authentication methods that the responder will
   accept.  Only two authentication methods are defined right now:
   see 4.4.1 and 4.4.2 below.

4.4. AUTHENTICATE cells

   If an initiator wants to authenticate, it responds to the
   AUTH_CHALLENGE cell with a CERTS cell and an AUTHENTICATE cell.
   The CERTS cell is as a server would send, except that instead of
   sending a CertType 1 (and possibly CertType 5) certs for arbitrary link
   certificates, the initiator sends a CertType 3 (and possibly
   CertType 6) cert for an RSA/Ed25519 AUTHENTICATE key.

   This difference is because we allow any link key type on a TLS
   link, but the protocol described here will only work for specific key
   types as described in 4.4.1 and 4.4.2 below.

   An AUTHENTICATE cell contains the following:

        AuthType                              [2 octets]
        AuthLen                               [2 octets]
        Authentication                        [AuthLen octets]

   Responders MUST ignore extra bytes at the end of an AUTHENTICATE
   cell.  Recognized AuthTypes are 1 and 3, described in the next
   two sections.

   Initiators MUST NOT send an AUTHENTICATE cell before they have
   verified the certificates presented in the responder's CERTS
   cell, and authenticated the responder.

4.4.1. Link authentication type 1: RSA-SHA256-TLSSecret

   If AuthType is 1 (meaning "RSA-SHA256-TLSSecret"), then the
   Authentication field of the AUTHENTICATE cell contains the following:

       TYPE: The characters "AUTH0001" [8 octets]
       CID: A SHA256 hash of the initiator's RSA1024 identity key [32 octets]
       SID: A SHA256 hash of the responder's RSA1024 identity key [32 octets]
       SLOG: A SHA256 hash of all bytes sent from the responder to the
         initiator as part of the negotiation up to and including the
         AUTH_CHALLENGE cell; that is, the VERSIONS cell, the CERTS cell,
         the AUTH_CHALLENGE cell, and any padding cells.  [32 octets]
       CLOG: A SHA256 hash of all bytes sent from the initiator to the
         responder as part of the negotiation so far; that is, the
         VERSIONS cell and the CERTS cell and any padding cells. [32
         octets]
       SCERT: A SHA256 hash of the responder's TLS link certificate. [32
         octets]
       TLSSECRETS: A SHA256 HMAC, using the TLS master secret as the
         secret key, of the following:
           - client_random, as sent in the TLS Client Hello
           - server_random, as sent in the TLS Server Hello
           - the NUL terminated ASCII string:
             "Tor V3 handshake TLS cross-certification"
          [32 octets]
       RAND: A 24 byte value, randomly chosen by the initiator.  (In an
         imitation of SSL3's gmt_unix_time field, older versions of Tor
         sent an 8-byte timestamp as the first 8 bytes of this field;
         new implementations should not do that.) [24 octets]
       SIG: A signature of a SHA256 hash of all the previous fields
         using the initiator's "Authenticate" key as presented.  (As
         always in Tor, we use OAEP-MGF1 padding; see tor-spec.txt
         section 0.3.)
          [variable length]

   To check the AUTHENTICATE cell, a responder checks that all fields
   from TYPE through TLSSECRETS contain their unique
   correct values as described above, and then verifies the signature.
   The server MUST ignore any extra bytes in the signed data after
   the RAND field.

   Responders MUST NOT accept this AuthType if the initiator has
   claimed to have an Ed25519 identity.

   (There is no AuthType 2: It was reserved but never implemented.)

4.4.2. Link authentication type 3: Ed25519-SHA256-RFC5705.

   If AuthType is 3, meaning "Ed25519-SHA256-RFC5705", the
   Authentication field of the AuthType cell is as below:

   Modified values and new fields below are marked with asterisks.

       TYPE: The characters "AUTH0003" [8 octets]
       CID: A SHA256 hash of the initiator's RSA1024 identity key [32 octets]
       SID: A SHA256 hash of the responder's RSA1024 identity key [32 octets]
       CID_ED: The initiator's Ed25519 identity key [32 octets]
       SID_ED: The responder's Ed25519 identity key, or all-zero. [32 octets]
       SLOG: A SHA256 hash of all bytes sent from the responder to the
         initiator as part of the negotiation up to and including the
         AUTH_CHALLENGE cell; that is, the VERSIONS cell, the CERTS cell,
         the AUTH_CHALLENGE cell, and any padding cells.  [32 octets]
       CLOG: A SHA256 hash of all bytes sent from the initiator to the
         responder as part of the negotiation so far; that is, the
         VERSIONS cell and the CERTS cell and any padding cells. [32
         octets]
       SCERT: A SHA256 hash of the responder's TLS link certificate. [32
         octets]
       TLSSECRETS: The output of an RFC5705 Exporter function on the
         TLS session, using as its inputs:
          - The label string "EXPORTER FOR TOR TLS CLIENT BINDING AUTH0003"
          - The context value equal to the initiator's Ed25519 identity key.
          - The length 32.
            [32 octets]
       RAND: A 24 byte value, randomly chosen by the initiator. [24 octets]
       SIG: A signature of all previous fields using the initiator's
          Ed25519 authentication key (as in the cert with CertType 6).
          [variable length]

   To check the AUTHENTICATE cell, a responder checks that all fields
   from TYPE through TLSSECRETS contain their unique
   correct values as described above, and then verifies the signature.
   The server MUST ignore any extra bytes in the signed data after
   the RAND field.

4.5. NETINFO cells

   If version 2 or higher is negotiated, each party sends the other a
   NETINFO cell.  The cell's payload is:

      TIME       (Timestamp)                     [4 bytes]
      OTHERADDR  (Other OR's address)            [variable]
         ATYPE   (Address type)                  [1 byte]
         ALEN    (Adress length)                 [1 byte]
         AVAL    (Address value in NBO)          [ALEN bytes]
      NMYADDR    (Number of this OR's addresses) [1 byte]
        NMYADDR times:
          ATYPE   (Address type)                 [1 byte]
          ALEN    (Adress length)                [1 byte]
          AVAL    (Address value in NBO))        [ALEN bytes]

   Recognized address types (ATYPE) are:

     [04] IPv4.
     [06] IPv6.

   ALEN MUST be 4 when ATYPE is 0x04 (IPv4) and 16 when ATYPE is 0x06
   (IPv6).

   The timestamp is a big-endian unsigned integer number of seconds
   since the Unix epoch. Implementations MUST ignore unexpected bytes
   at the end of the cell.

   Implementations MAY use the timestamp value to help decide if their
   clocks are skewed.  Initiators MAY use "other OR's address" to help
   learn which address their connections may be originating from, if they do
   not know it; and to learn whether the peer will treat the current
   connection as canonical.  Implementations SHOULD NOT trust these
   values unconditionally, especially when they come from non-authorities,
   since the other party can lie about the time or IP addresses it sees.

   Initiators SHOULD use "this OR's address" to make sure
   that they have connected to another OR at its canonical address.
   (See 5.3.1 below.)

5. Circuit management

5.1. CREATE and CREATED cells

   Users set up circuits incrementally, one hop at a time. To create a
   new circuit, OPs send a CREATE/CREATE2 cell to the first node, with
   the first half of an authenticated handshake; that node responds with
   a CREATED/CREATED2 cell with the second half of the handshake. To
   extend a circuit past the first hop, the OP sends an EXTEND/EXTEND2
   relay cell (see section 5.1.2) which instructs the last node in the
   circuit to send a CREATE/CREATE2 cell to extend the circuit.

   There are two kinds of CREATE and CREATED cells: The older
   "CREATE/CREATED" format, and the newer "CREATE2/CREATED2" format.  The
   newer format is extensible by design; the older one is not.

   A CREATE2 cell contains:

       HTYPE     (Client Handshake Type)     [2 bytes]
       HLEN      (Client Handshake Data Len) [2 bytes]
       HDATA     (Client Handshake Data)     [HLEN bytes]

   A CREATED2 cell contains:

       HLEN      (Server Handshake Data Len) [2 bytes]
       HDATA     (Server Handshake Data)     [HLEN bytes]

   Recognized handshake types are:

       0x0000  TAP  -- the original Tor handshake; see 5.1.3
       0x0001  reserved
       0x0002  ntor -- the ntor+curve25519+sha256 handshake; see 5.1.4


   The format of a CREATE cell is one of the following:

       HDATA     (Client Handshake Data)     [TAP_C_HANDSHAKE_LEN bytes]

   or

       HTAG      (Client Handshake Type Tag) [16 bytes]
       HDATA     (Client Handshake Data)     [TAP_C_HANDSHAKE_LEN-16 bytes]

   The first format is equivalent to a CREATE2 cell with HTYPE of 'tap'
   and length of TAP_C_HANDSHAKE_LEN.  The second format is a way to
   encapsulate new handshake types into the old CREATE cell format for
   migration. See 5.1.2 below. Recognized HTAG values are:

       ntor -- 'ntorNTORntorNTOR'

   The format of a CREATED cell is:

       HDATA     (Server Handshake Data)     [TAP_S_HANDSHAKE_LEN bytes]

   (It's equivalent to a CREATED2 cell with length of TAP_S_HANDSHAKE_LEN.)

   As usual with DH, x and y MUST be generated randomly.

   In general, clients SHOULD use CREATE whenever they are using the TAP
   handshake, and CREATE2 otherwise.  Clients SHOULD NOT send the
   second format of CREATE cells (the one with the handshake type tag)
   to a server directly.

   Servers always reply to a successful CREATE with a CREATED, and to a
   successful CREATE2 with a CREATED2.  On failure, a server sends a
   DESTROY cell to tear down the circuit.

   [CREATE2 is handled by Tor 0.2.4.7-alpha and later.]

5.1.1. Choosing circuit IDs in create cells

   The CircID for a CREATE/CREATE2 cell is an arbitrarily chosen
   nonzero integer, selected by the node (OP or OR) that sends the
   CREATE/CREATE2 cell.  In link protocol 3 or lower, CircIDs are 2
   bytes long; in protocol 4 or higher, CircIDs are 4 bytes long.

   To prevent CircID collisions, when one node sends a CREATE/CREATE2
   cell to another, it chooses from only one half of the possible
   values based on the ORs' public identity keys.

   In link protocol version 3 or lower, if the sending node has a lower
   key, it chooses a CircID with an MSB of 0; otherwise, it chooses a
   CircID with an MSB of 1. (Public keys are compared numerically by
   modulus.)  With protocol version 3 or lower, a client with no public key
   MAY choose any CircID it wishes, since clients never need to process a
   CREATE/CREATE2 cell.

   In link protocol version 4 or higher, whichever node initiated the
   connection sets its MSB to 1, and whichever node didn't initiate the
   connection sets its MSB to 0.

   The CircID value 0 is specifically reserved for cells that do not
   belong to any circuit: CircID 0 must not be used for circuits.  No
   other CircID value, including 0x8000 or 0x80000000, is reserved.

5.1.2. EXTEND and EXTENDED cells

   To extend an existing circuit, the client sends an EXTEND or EXTEND2
   relay cell to the last node in the circuit.

   An EXTEND2 cell's relay payload contains:

       NSPEC      (Number of link specifiers)     [1 byte]
         NSPEC times:
           LSTYPE (Link specifier type)           [1 byte]
           LSLEN  (Link specifier length)         [1 byte]
           LSPEC  (Link specifier)                [LSLEN bytes]
       HTYPE      (Client Handshake Type)         [2 bytes]
       HLEN       (Client Handshake Data Len)     [2 bytes]
       HDATA      (Client Handshake Data)         [HLEN bytes]

   Link specifiers describe the next node in the circuit and how to
   connect to it. Recognized specifiers are:

      [00] TLS-over-TCP, IPv4 address
           A four-byte IPv4 address plus two-byte ORPort
      [01] TLS-over-TCP, IPv6 address
           A sixteen-byte IPv6 address plus two-byte ORPort
      [02] Legacy identity
           A 20-byte SHA1 identity fingerprint. At most one may be listed.
      [03] Ed25519 identity
           A 32-byte Ed25519 identity fingerprint. At most one may
           be listed.

   Nodes MUST ignore unrecognized specifiers, and MUST accept multiple
   instances of specifiers other than 'legacy identity'.

   For purposes of indistinguishability, implementations SHOULD send
   these link specifiers, if using them, in this order: [00], [02], [03],
   [01].

   The relay payload for an EXTEND relay cell consists of:

         Address                       [4 bytes]
         Port                          [2 bytes]
         Onion skin                    [TAP_C_HANDSHAKE_LEN bytes]
         Identity fingerprint          [HASH_LEN bytes]

   The "legacy identity" and "identity fingerprint" fields are the
   SHA1 hash of the PKCS#1 ASN1 encoding of the next onion router's
   identity (signing) key.  (See 0.3 above.)  The "Ed25519 identity"
   field is the Ed25519 identity key of the target node.  Including
   this key information allows the extending OR verify that it is
   indeed connected to the correct target OR, and prevents certain
   man-in-the-middle attacks.

   Extending ORs MUST check _all_ provided identity keys (if they
   recognize the format), and and MUST NOT extend the circuit if the
   target OR did not prove its ownership of any such identity key.
   If only one identity key is provided, but the extending OR knows
   the other (from directory information), then the OR SHOULD also
   enforce that key.

   If an extending OR has a channel with a given Ed25519 ID and RSA
   identity, and receives a request for that Ed25519 ID and a
   different RSA identity, it SHOULD NOT attempt to make another
   connection: it should just fail and DESTROY the circuit.

   After checking relay identities, extending ORs generate a
   CREATE/CREATE2 cell from the contents of the EXTEND/EXTEND2 cell.
   See section 5.3 for details.

   The payload of an EXTENDED cell is the same as the payload of a
   CREATED cell.

   The payload of an EXTENDED2 cell is the same as the payload of a
   CREATED2 cell.

   [Support for EXTEND2/EXTENDED2 was added in Tor 0.2.4.8-alpha.]

   Clients SHOULD use the EXTEND format whenever sending a TAP
   handshake, and MUST use it whenever the EXTEND cell will be handled
   by a node running a version of Tor too old to support EXTEND2.  In
   other cases, clients SHOULD use EXTEND2.

   When generating an EXTEND2 cell, clients SHOULD include the target's
   Ed25519 identity whenever the target has one, and whenever the
   target supports LinkAuth subprotocol version "3". (See section 9.2.)

   When encoding a non-TAP handshake in an EXTEND cell, clients SHOULD
   use the format with 'client handshake type tag'.

5.1.3. The "TAP" handshake

   This handshake uses Diffie-Hellman in Z_p and RSA to compute a set of
   shared keys which the client knows are shared only with a particular
   server, and the server knows are shared with whomever sent the
   original handshake (or with nobody at all).  It's not very fast and
   not very good.  (See Goldberg's "On the Security of the Tor
   Authentication Protocol".)

   Define TAP_C_HANDSHAKE_LEN as DH_LEN+KEY_LEN+PK_PAD_LEN.
   Define TAP_S_HANDSHAKE_LEN as DH_LEN+HASH_LEN.

   The payload for a CREATE cell is an 'onion skin', which consists of
   the first step of the DH handshake data (also known as g^x).  This
   value is encrypted using the "legacy hybrid encryption" algorithm
   (see 0.4 above) to the server's onion key, giving a client handshake:

       PK-encrypted:
         Padding                       [PK_PAD_LEN bytes]
         Symmetric key                 [KEY_LEN bytes]
         First part of g^x             [PK_ENC_LEN-PK_PAD_LEN-KEY_LEN bytes]
       Symmetrically encrypted:
         Second part of g^x            [DH_LEN-(PK_ENC_LEN-PK_PAD_LEN-KEY_LEN)
                                           bytes]

   The payload for a CREATED cell, or the relay payload for an
   EXTENDED cell, contains:

         DH data (g^y)                 [DH_LEN bytes]
         Derivative key data (KH)      [HASH_LEN bytes]   <see 5.2 below>

   Once the handshake between the OP and an OR is completed, both can
   now calculate g^xy with ordinary DH.  Before computing g^xy, both parties
   MUST verify that the received g^x or g^y value is not degenerate;
   that is, it must be strictly greater than 1 and strictly less than p-1
   where p is the DH modulus.  Implementations MUST NOT complete a handshake
   with degenerate keys.  Implementations MUST NOT discard other "weak"
   g^x values.

   (Discarding degenerate keys is critical for security; if bad keys
   are not discarded, an attacker can substitute the OR's CREATED
   cell's g^y with 0 or 1, thus creating a known g^xy and impersonating
   the OR. Discarding other keys may allow attacks to learn bits of
   the private key.)

   Once both parties have g^xy, they derive their shared circuit keys
   and 'derivative key data' value via the KDF-TOR function in 5.2.1.

5.1.4. The "ntor" handshake

   This handshake uses a set of DH handshakes to compute a set of
   shared keys which the client knows are shared only with a particular
   server, and the server knows are shared with whomever sent the
   original handshake (or with nobody at all).  Here we use the
   "curve25519" group and representation as specified in "Curve25519:
   new Diffie-Hellman speed records" by D. J. Bernstein.

   [The ntor handshake was added in Tor 0.2.4.8-alpha.]

   In this section, define:

      H(x,t) as HMAC_SHA256 with message x and key t.
      H_LENGTH  = 32.
      ID_LENGTH = 20.
      G_LENGTH  = 32
      PROTOID   = "ntor-curve25519-sha256-1"
      t_mac     = PROTOID | ":mac"
      t_key     = PROTOID | ":key_extract"
      t_verify  = PROTOID | ":verify"
      MULT(a,b) = the multiplication of the curve25519 point 'a' by the
                  scalar 'b'.
      G         = The preferred base point for curve25519 ([9])
      KEYGEN()  = The curve25519 key generation algorithm, returning
                  a private/public keypair.
      m_expand  = PROTOID | ":key_expand"
      KEYID(A)  = A

   To perform the handshake, the client needs to know an identity key
   digest for the server, and an ntor onion key (a curve25519 public
   key) for that server. Call the ntor onion key "B".  The client
   generates a temporary keypair:

       x,X = KEYGEN()

   and generates a client-side handshake with contents:

       NODEID      Server identity digest  [ID_LENGTH bytes]
       KEYID       KEYID(B)                [H_LENGTH bytes]
       CLIENT_PK   X                       [G_LENGTH bytes]

   The server generates a keypair of y,Y = KEYGEN(), and uses its ntor
   private key 'b' to compute:

     secret_input = EXP(X,y) | EXP(X,b) | ID | B | X | Y | PROTOID
     KEY_SEED = H(secret_input, t_key)
     verify = H(secret_input, t_verify)
     auth_input = verify | ID | B | Y | X | PROTOID | "Server"

   The server's handshake reply is:

       SERVER_PK   Y                       [G_LENGTH bytes]
       AUTH        H(auth_input, t_mac)    [H_LENGTH bytes]

   The client then checks Y is in G^* [see NOTE below], and computes

     secret_input = EXP(Y,x) | EXP(B,x) | ID | B | X | Y | PROTOID
     KEY_SEED = H(secret_input, t_key)
     verify = H(secret_input, t_verify)
     auth_input = verify | ID | B | Y | X | PROTOID | "Server"

   The client verifies that AUTH == H(auth_input, t_mac).

   Both parties check that none of the EXP() operations produced the
   point at infinity. [NOTE: This is an adequate replacement for
   checking Y for group membership, if the group is curve25519.]

   Both parties now have a shared value for KEY_SEED.  They expand this
   into the keys needed for the Tor relay protocol, using the KDF
   described in 5.2.2 and the tag m_expand.

5.1.5. CREATE_FAST/CREATED_FAST cells

   When initializing the first hop of a circuit, the OP has already
   established the OR's identity and negotiated a secret key using TLS.
   Because of this, it is not always necessary for the OP to perform the
   public key operations to create a circuit.  In this case, the
   OP MAY send a CREATE_FAST cell instead of a CREATE cell for the first
   hop only.  The OR responds with a CREATED_FAST cell, and the circuit is
   created.

   A CREATE_FAST cell contains:

       Key material (X)    [HASH_LEN bytes]

   A CREATED_FAST cell contains:

       Key material (Y)    [HASH_LEN bytes]
       Derivative key data [HASH_LEN bytes] (See 5.2.1 below)

   The values of X and Y must be generated randomly.

   Once both parties have X and Y, they derive their shared circuit keys
   and 'derivative key data' value via the KDF-TOR function in 5.2.1.

   The CREATE_FAST handshake is currently deprecated whenever it is not
   necessary; the migration is controlled by the "usecreatefast"
   networkstatus parameter as described in dir-spec.txt.

   [Tor 0.3.1.1-alpha and later disable CREATE_FAST by default.]

5.2. Setting circuit keys

5.2.1. KDF-TOR

   This key derivation function is used by the TAP and CREATE_FAST
   handshakes, and in the current hidden service protocol. It shouldn't
   be used for new functionality.

   If the TAP handshake is used to extend a circuit, both parties
   base their key material on K0=g^xy, represented as a big-endian unsigned
   integer.

   If CREATE_FAST is used, both parties base their key material on
   K0=X|Y.

   From the base key material K0, they compute KEY_LEN*2+HASH_LEN*3 bytes of
   derivative key data as

       K = H(K0 | [00]) | H(K0 | [01]) | H(K0 | [02]) | ...

   The first HASH_LEN bytes of K form KH; the next HASH_LEN form the forward
   digest Df; the next HASH_LEN 41-60 form the backward digest Db; the next
   KEY_LEN 61-76 form Kf, and the final KEY_LEN form Kb.  Excess bytes from K
   are discarded.

   KH is used in the handshake response to demonstrate knowledge of the
   computed shared key. Df is used to seed the integrity-checking hash
   for the stream of data going from the OP to the OR, and Db seeds the
   integrity-checking hash for the data stream from the OR to the OP. Kf
   is used to encrypt the stream of data going from the OP to the OR, and
   Kb is used to encrypt the stream of data going from the OR to the OP.

5.2.2. KDF-RFC5869

   For newer KDF needs, Tor uses the key derivation function HKDF from
   RFC5869, instantiated with SHA256.  (This is due to a construction
   from Krawczyk.)  The generated key material is:

       K = K_1 | K_2 | K_3 | ...

       Where H(x,t) is HMAC_SHA256 with value x and key t
         and K_1     = H(m_expand | INT8(1) , KEY_SEED )
         and K_(i+1) = H(K_i | m_expand | INT8(i+1) , KEY_SEED )
         and m_expand is an arbitrarily chosen value,
         and INT8(i) is a octet with the value "i".

   In RFC5869's vocabulary, this is HKDF-SHA256 with info == m_expand,
   salt == t_key, and IKM == secret_input.

   When used in the ntor handshake, the first HASH_LEN bytes form the
   forward digest Df; the next HASH_LEN form the backward digest Db; the
   next KEY_LEN form Kf, the next KEY_LEN form Kb, and the final
   DIGEST_LEN bytes are taken as a nonce to use in the place of KH in the
   hidden service protocol.  Excess bytes from K are discarded.

5.3. Creating circuits

   When creating a circuit through the network, the circuit creator
   (OP) performs the following steps:

      1. Choose an onion router as an end node (R_N):
         * N MAY be 1 for non-anonymous directory mirror, introduction point,
           or service rendezvous connections.
         * N SHOULD be 3 or more for anonymous connections.
         Some end nodes accept streams (see 6.1), others are introduction
         or rendezvous points (see rend-spec-{v2,v3}.txt).

      2. Choose a chain of (N-1) onion routers (R_1...R_N-1) to constitute
         the path, such that no router appears in the path twice.

      3. If not already connected to the first router in the chain,
         open a new connection to that router.

      4. Choose a circID not already in use on the connection with the
         first router in the chain; send a CREATE/CREATE2 cell along
         the connection, to be received by the first onion router.

      5. Wait until a CREATED/CREATED2 cell is received; finish the
         handshake and extract the forward key Kf_1 and the backward
         key Kb_1.

      6. For each subsequent onion router R (R_2 through R_N), extend
         the circuit to R.

   To extend the circuit by a single onion router R_M, the OP performs
   these steps:

      1. Create an onion skin, encrypted to R_M's public onion key.

      2. Send the onion skin in a relay EXTEND/EXTEND2 cell along
         the circuit (see sections 5.1.2 and 5.5).

      3. When a relay EXTENDED/EXTENDED2 cell is received, verify KH,
         and calculate the shared keys.  The circuit is now extended.

   When an onion router receives an EXTEND relay cell, it sends a CREATE
   cell to the next onion router, with the enclosed onion skin as its
   payload.

   When an onion router receives an EXTEND2 relay cell, it sends a CREATE2
   cell to the next onion router, with the enclosed HLEN, HTYPE, and HDATA
   as its payload. The initiating onion router chooses some circID not yet
   used on the connection between the two onion routers. (But see section
   5.1.1 above, concerning choosing circIDs.)

   As special cases, if the EXTEND/EXTEND2 cell includes a legacy identity, or
   identity fingerprint of all zeroes, or asks to extend back to the relay
   that sent the extend cell, the circuit will fail and be torn down.

   Ed25519 identity keys are not required in EXTEND2 cells, so all zero
   keys SHOULD be accepted. If the extending relay knows the ed25519 key from
   the consensus, it SHOULD also check that key. (See section 5.1.2.)

   If an EXTEND2 cell contains the ed25519 key of the relay that sent the
   extend cell, the circuit will fail and be torn down.

   When an onion router receives a CREATE/CREATE2 cell, if it already has a
   circuit on the given connection with the given circID, it drops the
   cell. Otherwise, after receiving the CREATE/CREATE2 cell, it completes
   the specified handshake, and replies with a CREATED/CREATED2 cell.

   Upon receiving a CREATED/CREATED2 cell, an onion router packs it payload
   into an EXTENDED/EXTENDED2 relay cell (see section 5.1.2), and sends
   that cell up the circuit. Upon receiving the EXTENDED/EXTENDED2 relay
   cell, the OP can retrieve the handshake material.

   (As an optimization, OR implementations may delay processing onions
   until a break in traffic allows time to do so without harming
   network latency too greatly.)

5.3.1. Canonical connections

   It is possible for an attacker to launch a man-in-the-middle attack
   against a connection by telling OR Alice to extend to OR Bob at some
   address X controlled by the attacker.  The attacker cannot read the
   encrypted traffic, but the attacker is now in a position to count all
   bytes sent between Alice and Bob (assuming Alice was not already
   connected to Bob.)

   To prevent this, when an OR gets an extend request, it SHOULD use an
   existing OR connection if the ID matches, and ANY of the following
   conditions hold:

       - The IP matches the requested IP.
       - The OR knows that the IP of the connection it's using is canonical
         because it was listed in the NETINFO cell.

    ORs SHOULD NOT check the IPs that are listed in the server descriptor.
    Trusting server IPs makes it easier to covertly impersonate a relay, after
    stealing its keys.

5.4. Tearing down circuits

   Circuits are torn down when an unrecoverable error occurs along
   the circuit, or when all streams on a circuit are closed and the
   circuit's intended lifetime is over.

   ORs SHOULD also tear down circuits which attempt to create:

   * streams with RELAY_BEGIN, or
   * rendezvous points with ESTABLISH_RENDEZVOUS,
   ending at the first hop. Letting Tor be used as a single hop proxy makes
   exit and rendezvous nodes a more attractive target for compromise.

   ORs MAY use multiple methods to check if they are the first hop:

   * If an OR sees a circuit created with CREATE_FAST, the OR is sure to be
     the first hop of a circuit.
   * If an OR is the responder, and the initiator:
     * did not authenticate the link, or
     * authenticated with a key that is not in the consensus,
     then the OR is probably the first hop of a circuit (or the second hop of
     a circuit via a bridge relay).

   Circuits may be torn down either completely or hop-by-hop.

   To tear down a circuit completely, an OR or OP sends a DESTROY
   cell to the adjacent nodes on that circuit, using the appropriate
   direction's circID.

   Upon receiving an outgoing DESTROY cell, an OR frees resources
   associated with the corresponding circuit. If it's not the end of
   the circuit, it sends a DESTROY cell for that circuit to the next OR
   in the circuit. If the node is the end of the circuit, then it tears
   down any associated edge connections (see section 6.1).

   After a DESTROY cell has been processed, an OR ignores all data or
   destroy cells for the corresponding circuit.

   To tear down part of a circuit, the OP may send a RELAY_TRUNCATE cell
   signaling a given OR (Stream ID zero).  That OR sends a DESTROY
   cell to the next node in the circuit, and replies to the OP with a
   RELAY_TRUNCATED cell.

   [Note: If an OR receives a TRUNCATE cell and it has any RELAY cells
   still queued on the circuit for the next node it will drop them
   without sending them.  This is not considered conformant behavior,
   but it probably won't get fixed until a later version of Tor.  Thus,
   clients SHOULD NOT send a TRUNCATE cell to a node running any current
   version of Tor if a) they have sent relay cells through that node,
   and b) they aren't sure whether those cells have been sent on yet.]

   When an unrecoverable error occurs along one connection in a
   circuit, the nodes on either side of the connection should, if they
   are able, act as follows:  the node closer to the OP should send a
   RELAY_TRUNCATED cell towards the OP; the node farther from the OP
   should send a DESTROY cell down the circuit.

   The payload of a DESTROY cell contains a single octet, describing the
   reason that the circuit was closed. Similarly, the data of a
   RELAY_TRUNCATED cell also contains this single octet "reason" field. When
   sending a TRUNCATED or DESTROY cell because of another TRUNCATED or
   DESTROY cell, the error code should be propagated. The origin of a circuit
   always sets this error code to 0, to avoid leaking its version.

   The error codes are:

     0 -- NONE            (No reason given.)
     1 -- PROTOCOL        (Tor protocol violation.)
     2 -- INTERNAL        (Internal error.)
     3 -- REQUESTED       (A client sent a TRUNCATE command.)
     4 -- HIBERNATING     (Not currently operating; trying to save bandwidth.)
     5 -- RESOURCELIMIT   (Out of memory, sockets, or circuit IDs.)
     6 -- CONNECTFAILED   (Unable to reach relay.)
     7 -- OR_IDENTITY     (Connected to relay, but its OR identity was not
                           as expected.)
     8 -- CHANNEL_CLOSED  (The OR connection that was carrying this circuit
                           died.)
     9 -- FINISHED        (The circuit has expired for being dirty or old.)
    10 -- TIMEOUT         (Circuit construction took too long)
    11 -- DESTROYED       (The circuit was destroyed w/o client TRUNCATE)
    12 -- NOSUCHSERVICE   (Request for unknown hidden service)

5.5. Routing relay cells

5.5.1. Circuit ID Checks

   When a node wants to send a RELAY or RELAY_EARLY cell, it checks the cell's
   circID and determines whether the corresponding circuit along that
   connection is still open. If not, the node drops the cell.

   When a node receives a RELAY or RELAY_EARLY cell, it checks the cell's
   circID and determines whether it has a corresponding circuit along
   that connection. If not, the node drops the cell.

5.5.2. Forward Direction

   The forward direction is the direction that CREATE/CREATE2 cells
   are sent.

5.5.2.1. Routing from the Origin

   When a relay cell is sent from an OP, the OP encrypts the payload
   with the stream cipher as follows:

      OP sends relay cell:
         For I=N...1, where N is the destination node:
            Encrypt with Kf_I.
         Transmit the encrypted cell to node 1.

5.5.2.2. Relaying Forward at Onion Routers

   When a forward relay cell is received by an OR, it decrypts the payload
   with the stream cipher, as follows:

      'Forward' relay cell:
         Use Kf as key; decrypt.

   The OR then decides whether it recognizes the relay cell, by
   inspecting the payload as described in section 6.1 below.  If the OR
   recognizes the cell, it processes the contents of the relay cell.
   Otherwise, it passes the decrypted relay cell along the circuit if
   the circuit continues.  If the OR at the end of the circuit
   encounters an unrecognized relay cell, an error has occurred: the OR
   sends a DESTROY cell to tear down the circuit.

   For more information, see section 6 below.

5.5.3. Backward Direction

   The backward direction is the opposite direction from
   CREATE/CREATE2 cells.

5.5.3.1. Relaying Backward at Onion Routers

   When a backward relay cell is received by an OR, it encrypts the payload
   with the stream cipher, as follows:

      'Backward' relay cell:
         Use Kb as key; encrypt.

5.5.3. Routing to the Origin

   When a relay cell arrives at an OP, the OP decrypts the payload
   with the stream cipher as follows:

         OP receives relay cell from node 1:
            For I=1...N, where N is the final node on the circuit:
                Decrypt with Kb_I.
                If the payload is recognized (see section 6.1), then:
                    The sending node is I.
                    Stop and process the payload.

5.6. Handling relay_early cells

   A RELAY_EARLY cell is designed to limit the length any circuit can reach.
   When an OR receives a RELAY_EARLY cell, and the next node in the circuit
   is speaking v2 of the link protocol or later, the OR relays the cell as a
   RELAY_EARLY cell.  Otherwise, older Tors will relay it as a RELAY cell.

   If a node ever receives more than 8 RELAY_EARLY cells on a given
   outbound circuit, it SHOULD close the circuit. If it receives any
   inbound RELAY_EARLY cells, it MUST close the circuit immediately.

   When speaking v2 of the link protocol or later, clients MUST only send
   EXTEND/EXTEND2 cells inside RELAY_EARLY cells.  Clients SHOULD send the first ~8
   RELAY cells that are not targeted at the first hop of any circuit as
   RELAY_EARLY cells too, in order to partially conceal the circuit length.

   [Starting with Tor 0.2.3.11-alpha, relays should reject any
   EXTEND/EXTEND2 cell not received in a RELAY_EARLY cell.]

6. Application connections and stream management

6.1. Relay cells

   Within a circuit, the OP and the end node use the contents of
   RELAY packets to tunnel end-to-end commands and TCP connections
   ("Streams") across circuits.  End-to-end commands can be initiated
   by either edge; streams are initiated by the OP.

   End nodes that accept streams may be:
   * exit relays (RELAY_BEGIN, anonymous),
   * directory servers (RELAY_BEGIN_DIR, anonymous or non-anonymous),
   * onion services (RELAY_BEGIN, anonymous via a rendezvous point).

   The payload of each unencrypted RELAY cell consists of:

         Relay command           [1 byte]
         'Recognized'            [2 bytes]
         StreamID                [2 bytes]
         Digest                  [4 bytes]
         Length                  [2 bytes]
         Data                    [Length bytes]
         Padding                 [PAYLOAD_LEN - 11 - Length bytes]

   The relay commands are:

         1 -- RELAY_BEGIN     [forward]
         2 -- RELAY_DATA      [forward or backward]
         3 -- RELAY_END       [forward or backward]
         4 -- RELAY_CONNECTED [backward]
         5 -- RELAY_SENDME    [forward or backward] [sometimes control]
         6 -- RELAY_EXTEND    [forward]             [control]
         7 -- RELAY_EXTENDED  [backward]            [control]
         8 -- RELAY_TRUNCATE  [forward]             [control]
         9 -- RELAY_TRUNCATED [backward]            [control]
        10 -- RELAY_DROP      [forward or backward] [control]
        11 -- RELAY_RESOLVE   [forward]
        12 -- RELAY_RESOLVED  [backward]
        13 -- RELAY_BEGIN_DIR [forward]
        14 -- RELAY_EXTEND2   [forward]             [control]
        15 -- RELAY_EXTENDED2 [backward]            [control]

        32..40 -- Used for hidden services; see rend-spec-{v2,v3}.txt.

        41..42 -- Used for circuit padding; see Section 3 of padding-spec.txt.

   Commands labelled as "forward" must only be sent by the originator
   of the circuit. Commands labelled as "backward" must only be sent by
   other nodes in the circuit back to the originator. Commands marked
   as either can be sent either by the originator or other nodes.

   The 'recognized' field is used as a simple indication that the cell
   is still encrypted. It is an optimization to avoid calculating
   expensive digests for every cell. When sending cells, the unencrypted
   'recognized' MUST be set to zero.

   When receiving and decrypting cells the 'recognized' will always be
   zero if we're the endpoint that the cell is destined for.  For cells
   that we should relay, the 'recognized' field will usually be nonzero,
   but will accidentally be zero with P=2^-16.

   When handling a relay cell, if the 'recognized' in field in a
   decrypted relay payload is zero, the 'digest' field is computed as
   the first four bytes of the running digest of all the bytes that have
   been destined for this hop of the circuit or originated from this hop
   of the circuit, seeded from Df or Db respectively (obtained in
   section 5.2 above), and including this RELAY cell's entire payload
   (taken with the digest field set to zero).  Note that these digests
   _do_ include the padding bytes at the end of the cell, not only those up
   to "Len".  If the digest is correct, the cell is considered "recognized"
   for the purposes of decryption (see section 5.5 above).

   (The digest does not include any bytes from relay cells that do
   not start or end at this hop of the circuit. That is, it does not
   include forwarded data. Therefore if 'recognized' is zero but the
   digest does not match, the running digest at that node should
   not be updated, and the cell should be forwarded on.)

   All RELAY cells pertaining to the same tunneled stream have the same
   stream ID.  StreamIDs are chosen arbitrarily by the OP.  No stream
   may have a StreamID of zero. Rather, RELAY cells that affect the
   entire circuit rather than a particular stream use a StreamID of zero
   -- they are marked in the table above as "[control]" style
   cells. (Sendme cells are marked as "sometimes control" because they
   can include a StreamID or not depending on their purpose -- see
   Section 7.)

   The 'Length' field of a relay cell contains the number of bytes in
   the relay payload which contain real payload data. The remainder of
   the unencrypted payload is padded with padding bytes. Implementations
   handle padding bytes of unencrypted relay cells as they do padding
   bytes for other cell types; see Section 3.

   The 'Padding' field is used to make relay cell contents unpredictable, to
   avoid certain attacks (see proposal 289 for rationale). Implementations
   SHOULD fill this field with four zero-valued bytes, followed by as many
   random bytes as will fit.  (If there are fewer than 4 bytes for padding,
   then they should all be filled with zero.

   Implementations MUST NOT rely on the contents of the 'Padding' field.

   If the RELAY cell is recognized but the relay command is not
   understood, the cell must be dropped and ignored. Its contents
   still count with respect to the digests and flow control windows, though.

6.2. Opening streams and transferring data

   To open a new anonymized TCP connection, the OP chooses an open
   circuit to an exit that may be able to connect to the destination
   address, selects an arbitrary StreamID not yet used on that circuit,
   and constructs a RELAY_BEGIN cell with a payload encoding the address
   and port of the destination host.  The payload format is:

         ADDRPORT [nul-terminated string]
         FLAGS    [4 bytes]

   ADDRPORT is made of ADDRESS | ':' | PORT | [00]

   where  ADDRESS can be a DNS hostname, or an IPv4 address in
   dotted-quad format, or an IPv6 address surrounded by square brackets;
   and where PORT is a decimal integer between 1 and 65535, inclusive.

   The FLAGS value has one or more of the following bits set, where
   "bit 1" is the LSB of the 32-bit value, and "bit 32" is the MSB.
   (Remember that all values in Tor are big-endian (see 0.1.1 above), so
   the MSB of a 4-byte value is the MSB of the first byte, and the LSB
   of a 4-byte value is the LSB of its last byte.)

     bit   meaning
      1 -- IPv6 okay.  We support learning about IPv6 addresses and
           connecting to IPv6 addresses.
      2 -- IPv4 not okay.  We don't want to learn about IPv4 addresses
           or connect to them.
      3 -- IPv6 preferred.  If there are both IPv4 and IPv6 addresses,
           we want to connect to the IPv6 one.  (By default, we connect
           to the IPv4 address.)
      4..32 -- Reserved. Current clients MUST NOT set these. Servers
           MUST ignore them.

   Upon receiving this cell, the exit node resolves the address as
   necessary, and opens a new TCP connection to the target port.  If the
   address cannot be resolved, or a connection can't be established, the
   exit node replies with a RELAY_END cell.  (See 6.3 below.)
   Otherwise, the exit node replies with a RELAY_CONNECTED cell, whose
   payload is in one of the following formats:

       The IPv4 address to which the connection was made [4 octets]
       A number of seconds (TTL) for which the address may be cached [4 octets]

    or

       Four zero-valued octets [4 octets]
       An address type (6)     [1 octet]
       The IPv6 address to which the connection was made [16 octets]
       A number of seconds (TTL) for which the address may be cached [4 octets]

   [Tor exit nodes before 0.1.2.0 set the TTL field to a fixed value.  Later
   versions set the TTL to the last value seen from a DNS server, and expire
   their own cached entries after a fixed interval.  This prevents certain
   attacks.]

   Once a connection has been established, the OP and exit node
   package stream data in RELAY_DATA cells, and upon receiving such
   cells, echo their contents to the corresponding TCP stream.

   If the exit node does not support optimistic data (i.e. its
   version number is before 0.2.3.1-alpha), then the OP MUST wait
   for a RELAY_CONNECTED cell before sending any data.  If the exit
   node supports optimistic data (i.e. its version number is
   0.2.3.1-alpha or later), then the OP MAY send RELAY_DATA cells
   immediately after sending the RELAY_BEGIN cell (and before
   receiving either a RELAY_CONNECTED or RELAY_END cell).

   RELAY_DATA cells sent to unrecognized streams are dropped.  If
   the exit node supports optimistic data, then RELAY_DATA cells it
   receives on streams which have seen RELAY_BEGIN but have not yet
   been replied to with a RELAY_CONNECTED or RELAY_END are queued.
   If the stream creation succeeds with a RELAY_CONNECTED, the queue
   is processed immediately afterwards; if the stream creation fails
   with a RELAY_END, the contents of the queue are deleted.

   Relay RELAY_DROP cells are long-range dummies; upon receiving such
   a cell, the OR or OP must drop it.

6.2.1. Opening a directory stream

   If a Tor relay is a directory server, it should respond to a
   RELAY_BEGIN_DIR cell as if it had received a BEGIN cell requesting a
   connection to its directory port.  RELAY_BEGIN_DIR cells ignore exit
   policy, since the stream is local to the Tor process.

   Directory servers may be:
   * authoritative directories (RELAY_BEGIN_DIR, usually non-anonymous),
   * bridge authoritative directories (RELAY_BEGIN_DIR, anonymous),
   * directory mirrors (RELAY_BEGIN_DIR, usually non-anonymous),
   * onion service directories (RELAY_BEGIN_DIR, anonymous).

   If the Tor relay is not running a directory service, it should respond
   with a REASON_NOTDIRECTORY RELAY_END cell.

   Clients MUST generate an all-zero payload for RELAY_BEGIN_DIR cells,
   and relays MUST ignore the payload.

   In response to a RELAY_BEGIN_DIR cell, relays respond either with a
   RELAY_CONNECTED cell on succcess, or a RELAY_END cell on failure. They
   MUST send a RELAY_CONNECTED cell all-zero payload, and clients MUST ignore
   the payload.

   [RELAY_BEGIN_DIR was not supported before Tor 0.1.2.2-alpha; clients
   SHOULD NOT send it to routers running earlier versions of Tor.]

6.3. Closing streams

   When an anonymized TCP connection is closed, or an edge node
   encounters error on any stream, it sends a 'RELAY_END' cell along the
   circuit (if possible) and closes the TCP connection immediately.  If
   an edge node receives a 'RELAY_END' cell for any stream, it closes
   the TCP connection completely, and sends nothing more along the
   circuit for that stream.

   The payload of a RELAY_END cell begins with a single 'reason' byte to
   describe why the stream is closing.  For some reasons, it contains
   additional data (depending on the reason.)  The values are:

       1 -- REASON_MISC           (catch-all for unlisted reasons)
       2 -- REASON_RESOLVEFAILED  (couldn't look up hostname)
       3 -- REASON_CONNECTREFUSED (remote host refused connection) [*]
       4 -- REASON_EXITPOLICY     (OR refuses to connect to host or port)
       5 -- REASON_DESTROY        (Circuit is being destroyed)
       6 -- REASON_DONE           (Anonymized TCP connection was closed)
       7 -- REASON_TIMEOUT        (Connection timed out, or OR timed out
                                   while connecting)
       8 -- REASON_NOROUTE        (Routing error while attempting to
                                   contact destination)
       9 -- REASON_HIBERNATING    (OR is temporarily hibernating)
      10 -- REASON_INTERNAL       (Internal error at the OR)
      11 -- REASON_RESOURCELIMIT  (OR has no resources to fulfill request)
      12 -- REASON_CONNRESET      (Connection was unexpectedly reset)
      13 -- REASON_TORPROTOCOL    (Sent when closing connection because of
                                   Tor protocol violations.)
      14 -- REASON_NOTDIRECTORY   (Client sent RELAY_BEGIN_DIR to a
                                   non-directory relay.)


   For most reasons, the format is:

      Reason                      [1 byte]

   For REASON_EXITPOLICY, the format is:

      Reason                      [1 byte]
      IPv4 or IPv6 address        [4 bytes or 16 bytes]
      TTL                         [4 bytes]

   OPs and ORs MUST accept reasons not on the above list, since future
   versions of Tor may provide more fine-grained reasons.

   Tors SHOULD NOT send any reason except REASON_MISC for a stream that they
   have originated.

   [*] Older versions of Tor also send this reason when connections are
       reset.

   --- [The rest of this section describes unimplemented functionality.]

   Because TCP connections can be half-open, we follow an equivalent
   to TCP's FIN/FIN-ACK/ACK protocol to close streams.

   An exit (or onion service) connection can have a TCP stream in one of
   three states: 'OPEN', 'DONE_PACKAGING', and 'DONE_DELIVERING'.  For the
   purposes of modeling transitions, we treat 'CLOSED' as a fourth state,
   although connections in this state are not, in fact, tracked by the
   onion router.

   A stream begins in the 'OPEN' state.  Upon receiving a 'FIN' from
   the corresponding TCP connection, the edge node sends a 'RELAY_FIN'
   cell along the circuit and changes its state to 'DONE_PACKAGING'.
   Upon receiving a 'RELAY_FIN' cell, an edge node sends a 'FIN' to
   the corresponding TCP connection (e.g., by calling
   shutdown(SHUT_WR)) and changing its state to 'DONE_DELIVERING'.

   When a stream in already in 'DONE_DELIVERING' receives a 'FIN', it
   also sends a 'RELAY_FIN' along the circuit, and changes its state
   to 'CLOSED'.  When a stream already in 'DONE_PACKAGING' receives a
   'RELAY_FIN' cell, it sends a 'FIN' and changes its state to
   'CLOSED'.

   If an edge node encounters an error on any stream, it sends a
   'RELAY_END' cell (if possible) and closes the stream immediately.

6.4. Remote hostname lookup

   To find the address associated with a hostname, the OP sends a
   RELAY_RESOLVE cell containing the hostname to be resolved with a NUL
   terminating byte. (For a reverse lookup, the OP sends a RELAY_RESOLVE
   cell containing an in-addr.arpa address.) The OR replies with a
   RELAY_RESOLVED cell containing any number of answers. Each answer is
   of the form:

       Type   (1 octet)
       Length (1 octet)
       Value  (variable-width)
       TTL    (4 octets)
   "Length" is the length of the Value field.
   "Type" is one of:

      0x00 -- Hostname
      0x04 -- IPv4 address
      0x06 -- IPv6 address
      0xF0 -- Error, transient
      0xF1 -- Error, nontransient

    If any answer has a type of 'Error', then no other answer may be
    given.

    The 'Value' field encodes the answer:
        IP addresses are given in network order.
        Hostnames are given in standard DNS order ("www.example.com")
          and not NUL-terminated.
        The content of Errors is currently ignored. Relays currently
          set it to the string "Error resolving hostname" with no
          terminating NUL. Implementations MUST ignore this value.

    For backward compatibility, if there are any IPv4 answers, one of those
    must be given as the first answer.

    The RELAY_RESOLVE cell must use a nonzero, distinct streamID; the
    corresponding RELAY_RESOLVED cell must use the same streamID.  No stream
    is actually created by the OR when resolving the name.

7. Flow control

7.1. Link throttling

   Each client or relay should do appropriate bandwidth throttling to
   keep its user happy.

   Communicants rely on TCP's default flow control to push back when they
   stop reading.

   The mainline Tor implementation uses token buckets (one for reads,
   one for writes) for the rate limiting.

   Since 0.2.0.x, Tor has let the user specify an additional pair of
   token buckets for "relayed" traffic, so people can deploy a Tor relay
   with strict rate limiting, but also use the same Tor as a client. To
   avoid partitioning concerns we combine both classes of traffic over a
   given OR connection, and keep track of the last time we read or wrote
   a high-priority (non-relayed) cell. If it's been less than N seconds
   (currently N=30), we give the whole connection high priority, else we
   give the whole connection low priority. We also give low priority
   to reads and writes for connections that are serving directory
   information. See proposal 111 for details.

7.2. Link padding

   Link padding can be created by sending PADDING or VPADDING cells
   along the connection; relay cells of type "DROP" can be used for
   long-range padding.  The payloads of PADDING, VPADDING, or DROP
   cells are filled with padding bytes. See Section 3.

   If the link protocol is version 5 or higher, link level padding is
   enabled as per padding-spec.txt. On these connections, clients may
   negotiate the use of padding with a CELL_PADDING_NEGOTIATE command
   whose format is as follows:

         Version           [1 byte]
         Command           [1 byte]
         ito_low_ms        [2 bytes]
         ito_high_ms       [2 bytes]

   Currently, only version 0 of this cell is defined. In it, the command
   field is either 1 (stop padding) or 2 (start padding). For the start
   padding command, a pair of timeout values specifying a low and a high
   range bounds for randomized padding timeouts may be specified as unsigned
   integer values in milliseconds. The ito_low_ms field must not be lower than
   the current consensus parameter value for nf_ito_low (default: 1500).
   ito_high_ms must be greater than ito_low_ms.

   For more details on padding behavior, see padding-spec.txt.

7.3. Circuit-level flow control

   To control a circuit's bandwidth usage, each OR keeps track of two
   'windows', consisting of how many RELAY_DATA cells it is allowed to
   originate or willing to consume.

   These two windows are respectively named: the package window (packaged for
   transmission) and the deliver window (delivered for local streams).

   Because of our leaky-pipe topology, every relay on the circuit has a pair
   of windows, and the OP has a pair of windows for every relay on the
   circuit. These windows do not apply to relayed cells, however, and a relay
   that is never used for streams will never decrement its window or cause the
   client to decrement a window.

   Each 'window' value is initially set based on the consensus parameter
   'circwindow' in the directory (see dir-spec.txt), or to 1000 data cells if
   no 'circwindow' value is given. In each direction, cells that are not
   RELAY_DATA cells do not affect the window.

   An OR or OP (depending on the stream direction) sends a RELAY_SENDME cell
   to indicate that it is willing to receive more cells when its deliver
   window goes down below a full increment (100). For example, if the window
   started at 1000, it should send a RELAY_SENDME when it reaches 900.

   When an OR or OP receives a RELAY_SENDME, it increments its package window
   by a value of 100 (circuit window increment) and proceeds to sending the
   remaining RELAY_DATA cells.

   If a package window reaches 0, the OR or OP stops reading from TCP
   connections for all streams on the corresponding circuit, and sends no more
   RELAY_DATA cells until receiving a RELAY_SENDME cell.

   If a deliver window goes below 0, the circuit should be torn down.

   Starting with tor-0.4.1.1-alpha, authenticated SENDMEs are supported
   (version 1, see below). This means that both the OR and OP need to remember
   the rolling digest of the cell that precedes (triggers) a RELAY_SENDME.
   This can be known if the package window gets to a multiple of the circuit
   window increment (100).

   When the RELAY_SENDME version 1 arrives, it will contain a digest that MUST
   match the one remembered. This represents a proof that the end point of the
   circuit saw the sent cells. On failure to match, the circuit should be torn
   down.

   To ensure unpredictability, random bytes should be added to at least one
   RELAY_DATA cell within one increment window. In other word, every 100 cells
   (increment), random bytes should be introduced in at least one cell.

7.3.1. SENDME Cell Format

   A circuit-level RELAY_SENDME cell always has its StreamID=0.

   An OR or OP must obey these two consensus parameters in order to know which
   version to emit and accept.

      'sendme_emit_min_version': Minimum version to emit.
      'sendme_accept_min_version': Minimum version to accept.

   If a RELAY_SENDME version is received that is below the minimum accepted
   version, the circuit should be closed.

   The RELAY_SENDME payload contains the following:

      VERSION     [1 byte]
      DATA_LEN    [2 bytes]
      DATA        [DATA_LEN bytes]

   The VERSION tells us what is expected in the DATA section of length
   DATA_LEN and how to handle it. The recognized values are:

      0x00: The rest of the payload should be ignored.

      0x01: Authenticated SENDME. The DATA section MUST contain:

         DIGEST   [20 bytes]

         If the DATA_LEN value is less than 20 bytes, the cell should be
         dropped and the circuit closed. If the value is more than 20 bytes,
         then the first 20 bytes should be read to get the DIGEST value.

         The DIGEST is the rolling digest value from the RELAY_DATA cell that
         immediately preceded (triggered) this RELAY_SENDME. This value is
         matched on the other side from the previous cell sent that the OR/OP
         must remember.

   If the VERSION is unrecognized or below the minimum accepted version (taken
   from the consensus), the circuit should be torn down.

7.4. Stream-level flow control

   Edge nodes use RELAY_SENDME cells to implement end-to-end flow
   control for individual connections across circuits. Similarly to
   circuit-level flow control, edge nodes begin with a window of cells
   (500) per stream, and increment the window by a fixed value (50)
   upon receiving a RELAY_SENDME cell. Edge nodes initiate RELAY_SENDME
   cells when both a) the window is <= 450, and b) there are less than
   ten cell payloads remaining to be flushed at that edge.

   Stream-level RELAY_SENDME cells are distinguished by having nonzero
   StreamID. They are still empty; the body still SHOULD be ignored.


8. Handling resource exhaustion


8.1. Memory exhaustion.

   If RAM becomes low, an OR should begin destroying circuits until
   more memory is free again.  We recommend the following algorithm:

     - Set a threshold amount of RAM to recover at 10% of the total RAM.

     - Sort the circuits by their 'staleness', defined as the age of the
       oldest data queued on the circuit.  This data can be:

          * Bytes that are waiting to flush to or from a stream on that
            circuit.

          * Bytes that are waiting to flush from a connection created with
            BEGIN_DIR.

          * Cells that are waiting to flush or be processed.

     - While we have not yet recovered enough RAM:

          * Free all memory held by the most stale circuit, and send DESTROY
            cells in both directions on that circuit.  Count the amount of
            memory we recovered towards the total.

9. Subprotocol versioning

   This section specifies the Tor subprotocol versioning. They are broken down
   into different types with their current version numbers. Any new version
   number should be added to this section.

   The dir-spec.txt details how those versions are encoded. See the
   "proto"/"pr" line in a descriptor and the "recommended-relay-protocols",
   "required-relay-protocols", "recommended-client-protocols" and
   "required-client-protocols" lines in the vote/consensus format.

   Here are the rules a relay and client should follow when encountering a
   protocol list in the consensus:

      - When a relay lacks a protocol listed in recommended-relay-protocols,
        it should warn its operator that the relay is obsolete.

      - When a relay lacks a protocol listed in required-relay-protocols, it
        should warn its operator as above. If the consensus is newer than the
        date when the software was released or scheduled for release, it must
        not attempt to join the network.

      - When a client lacks a protocol listed in recommended-client-protocols,
        it should warn the user that the client is obsolete.

      - When a client lacks a protocol listed in required-client-protocols,
        it should warn the user as above.  If the consensus is newer than the
        date when the software was released, it must not connect to the
        network.  This implements a "safe forward shutdown" mechanism for
        zombie clients.

      - If a client or relay has a cached consensus telling it that a given
        protocol is required, and it does not implement that protocol, it
        SHOULD NOT try to fetch a newer consensus.

   Software release dates SHOULD be automatically updated as part of the
   release process, to prevent forgetting to move them forward.  Software
   release dates MAY be manually adjusted by maintainers if necessary.

   Starting in version 0.2.9.4-alpha, the initial required protocols for
   clients that we will Recommend and Require are:

      Cons=1-2 Desc=1-2 DirCache=1 HSDir=1 HSIntro=3 HSRend=1 Link=4
      LinkAuth=1 Microdesc=1-2 Relay=2

   For relays we will Require:

      Cons=1 Desc=1 DirCache=1 HSDir=1 HSIntro=3 HSRend=1 Link=3-4
      LinkAuth=1 Microdesc=1 Relay=1-2

   For relays, we will additionally Recommend all protocols which we
   recommend for clients.

9.1. "Link"

   The "link" protocols are those used by clients and relays to initiate and
   receive OR connections and to handle cells on OR connections. The "link"
   protocol versions correspond 1:1 to those versions.

   Two Tor instances can make a connection to each other only if they have at
   least one link protocol in common.

   The current "link" versions are: "1" through "5". See section 4.1 for more
   information. All current Tor versions support "1-3"; versions from
   0.2.4.11-alpha and on support "1-4"; versions from 0.3.1.1-alpha and on
   support "1-5". Eventually we will drop "1" and "2".

9.2. "LinkAuth"

   LinkAuth protocols correspond to varieties of Authenticate cells used for
   the v3+ link protocols.

   Current versions are:

   "1" is the RSA link authentication described in section 4.4.1 above.

   "2" is unused, and reserved by proposal 244.

   "3" is the ed25519 link authentication described in 4.4.2 above.


9.3. "Relay"

   The "relay" protocols are those used to handle CREATE/CREATE2
   cells, and those that handle the various RELAY cell types received
   after a CREATE/CREATE2 cell.  (Except, relay cells used to manage
   introduction and rendezvous points are managed with the "HSIntro"
   and "HSRend" protocols respectively.)

   Current versions are:

   "1" -- supports the TAP key exchange, with all features in Tor 0.2.3.
          Support for CREATE and CREATED and CREATE_FAST and CREATED_FAST
          and EXTEND and EXTENDED.

   "2" -- supports the ntor key exchange, and all features in Tor
          0.2.4.19.  Includes support for CREATE2 and CREATED2 and
          EXTEND2 and EXTENDED2.

          Relay=2 has limited IPv6 support:
            * Clients might not include IPv6 ORPorts in EXTEND2 cells.
            * Relays (and bridges) might not initiate IPv6 connections in
              response to EXTEND2 cells containing IPv6 ORPorts, even if they
              are configured with an IPv6 ORPort.

          However, relays support accepting inbound connections to their IPv6
          ORPorts. And they might extend circuits via authenticated IPv6
          connections to other relays.

   "3" -- relays support extending over IPv6 connections in response to an
          EXTEND2 cell containing an IPv6 ORPort.

          Bridges might not extend over IPv6, because they try to imitate
          client behaviour.

          A successful IPv6 extend requires:
            * Relay subprotocol version 3 (or later) on the extending relay,
            * an IPv6 ORPort on the extending relay,
            * an IPv6 ORPort for the accepting relay in the EXTEND2 cell, and
            * an IPv6 ORPort on the accepting relay.
          (Because different tor instances can have different views of the
          network, these checks should be done when the path is selected.
          Extending relays should only check local IPv6 information, before
          attempting the extend.)

          When relays receive an EXTEND2 cell containing both an IPv4 and an
          IPv6 ORPort, and there is no existing authenticated connection with
          the target relay, the extending relay may choose between IPv4 and
          IPv6 at random. The extending relay might not try the other address,
          if the first connection fails.

          As is the case with other subprotocol versions, tor advertises,
          recommends, or requires support for this protocol version, regardless
          of its current configuration.

          In particular:
            * relays without an IPv6 ORPort, and
            * tor instances that are not relays,
          have the following behaviour, regardless of their configuration:
            * advertise support for "Relay=3" in their descriptor
              (if they are a relay, bridge, or directory authority), and
            * react to consensuses recommending or requiring support for
              "Relay=3".

          This subprotocol version is described in proposal 311, and
          implemented in Tor 0.4.4.1-alpha.

9.4. "HSIntro"

   The "HSIntro" protocol handles introduction points.

   "3" -- supports authentication as of proposal 121 in Tor
          0.2.1.6-alpha.

   "4" -- support ed25519 authentication keys which is defined by the HS v3
          protocol as part of proposal 224 in Tor 0.3.0.4-alpha.

   "5" -- support ESTABLISH_INTRO cell DoS parameters extension for onion
          service version 3 only in Tor 0.4.2.1-alpha.

9.5. "HSRend"

   The "HSRend" protocol handles rendezvous points.

   "1" -- supports all features in Tor 0.0.6.

   "2" -- supports RENDEZVOUS2 cells of arbitrary length as long as they
          have 20 bytes of cookie in Tor 0.2.9.1-alpha.

9.6. "HSDir"

   The "HSDir" protocols are the set of hidden service document types that can
   be uploaded to, understood by, and downloaded from a tor relay, and the set
   of URLs available to fetch them.

   "1" -- supports all features in Tor 0.2.0.10-alpha.

   "2" -- support ed25519 blinded keys request which is defined by the HS v3
          protocol as part of proposal 224 in Tor 0.3.0.4-alpha.

9.7. "DirCache"

   The "DirCache" protocols are the set of documents available for download
   from a directory cache via BEGIN_DIR, and the set of URLs available to
   fetch them.  (This excludes URLs for hidden service objects.)

   "1" -- supports all features in Tor 0.2.4.19.

   "2" -- adds support for consensus diffs.

9.8. "Desc"

   Describes features present or absent in descriptors.

   Most features in descriptors don't require a "Desc" update -- only those
   that need to someday be required.  For example, someday clients will need
   to understand ed25519 identities.

   "1" -- supports all features in Tor 0.2.4.19.

   "2" -- cross-signing with onion-keys, signing with ed25519
   identities.

9.9. "Microdesc"

   Describes features present or absent in microdescriptors.

   Most features in descriptors don't require a "MicroDesc" update -- only
   those that need to someday be required.  These correspond more or less with
   consensus methods.

   "1" -- consensus methods 9 through 20.

   "2" -- consensus method 21 (adds ed25519 keys to microdescs).

9.10. "Cons"

   Describes features present or absent in consensus documents.

   Most features in consensus documents don't require a "Cons" update -- only
   those that need to someday be required.

   These correspond more or less with consensus methods.

   "1" -- consensus methods 9 through 20.

   "2" -- consensus method 21 (adds ed25519 keys to microdescs).

9.11. "Padding"

   Describes the padding capabilities of the relay.

   "1" -- [DEFUNCT] Relay supports circuit-level padding. This version MUST NOT
          be used as it was also enabled in relays that don't actually support
          circuit-level padding. Advertised by Tor versions from
          tor-0.4.0.1-alpha and only up to and including tor-0.4.1.4-rc.

   "2" -- Relay supports the HS circuit setup padding machines (proposal 302).
          Advertised by Tor versions from tor-0.4.1.5 and onwards.

9.12. "FlowCtrl"

   Describes the flow control protocol at the circuit and stream level. If
   there is no FlowCtrl advertised, tor supports the unauthenticated flow
   control features (version 0).

   "1" -- supports authenticated circuit level SENDMEs as of proposal 289 in
          Tor 0.4.1.1-alpha.