tor-spec.txt 45 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414415416417418419420421422423424425426427428429430431432433434435436437438439440441442443444445446447448449450451452453454455456457458459460461462463464465466467468469470471472473474475476477478479480481482483484485486487488489490491492493494495496497498499500501502503504505506507508509510511512513514515516517518519520521522523524525526527528529530531532533534535536537538539540541542543544545546547548549550551552553554555556557558559560561562563564565566567568569570571572573574575576577578579580581582583584585586587588589590591592593594595596597598599600601602603604605606607608609610611612613614615616617618619620621622623624625626627628629630631632633634635636637638639640641642643644645646647648649650651652653654655656657658659660661662663664665666667668669670671672673674675676677678679680681682683684685686687688689690691692693694695696697698699700701702703704705706707708709710711712713714715716717718719720721722723724725726727728729730731732733734735736737738739740741742743744745746747748749750751752753754755756757758759760761762763764765766767768769770771772773774775776777778779780781782783784785786787788789790791792793794795796797798799800801802803804805806807808809810811812813814815816817818819820821822823824825826827828829830831832833834835836837838839840841842843844845846847848849850851852853854855856857858859860861862863864865866867868869870871872873874875876877878879880881882883884885886887888889890891892893894895896897898899900901902903904905906907908909910911912913914915916917918919920921922923924925926927928929930931932933934935936937938939940941942943944945946947948949950951952953954955956957958959960961962963964965966967968969970971972973974975976977978979980981982983984985986987988989990991992993
  1. $Id$
  2. Tor Protocol Specification
  3. Roger Dingledine
  4. Nick Mathewson
  5. Note: This document aims to specify Tor as implemented in 0.2.1.x. Future
  6. versions of Tor may implement improved protocols, and compatibility is not
  7. guaranteed. Compatibility notes are given for versions 0.1.1.15-rc and
  8. later; earlier versions are not compatible with the Tor network as of this
  9. writing.
  10. This specification is not a design document; most design criteria
  11. are not examined. For more information on why Tor acts as it does,
  12. see tor-design.pdf.
  13. 0. Preliminaries
  14. 0.1. Notation and encoding
  15. PK -- a public key.
  16. SK -- a private key.
  17. K -- a key for a symmetric cypher.
  18. a|b -- concatenation of 'a' and 'b'.
  19. [A0 B1 C2] -- a three-byte sequence, containing the bytes with
  20. hexadecimal values A0, B1, and C2, in that order.
  21. All numeric values are encoded in network (big-endian) order.
  22. H(m) -- a cryptographic hash of m.
  23. 0.2. Security parameters
  24. Tor uses a stream cipher, a public-key cipher, the Diffie-Hellman
  25. protocol, and a hash function.
  26. KEY_LEN -- the length of the stream cipher's key, in bytes.
  27. PK_ENC_LEN -- the length of a public-key encrypted message, in bytes.
  28. PK_PAD_LEN -- the number of bytes added in padding for public-key
  29. encryption, in bytes. (The largest number of bytes that can be encrypted
  30. in a single public-key operation is therefore PK_ENC_LEN-PK_PAD_LEN.)
  31. DH_LEN -- the number of bytes used to represent a member of the
  32. Diffie-Hellman group.
  33. DH_SEC_LEN -- the number of bytes used in a Diffie-Hellman private key (x).
  34. HASH_LEN -- the length of the hash function's output, in bytes.
  35. PAYLOAD_LEN -- The longest allowable cell payload, in bytes. (509)
  36. CELL_LEN -- The length of a Tor cell, in bytes.
  37. 0.3. Ciphers
  38. For a stream cipher, we use 128-bit AES in counter mode, with an IV of all
  39. 0 bytes.
  40. For a public-key cipher, we use RSA with 1024-bit keys and a fixed
  41. exponent of 65537. We use OAEP-MGF1 padding, with SHA-1 as its digest
  42. function. We leave the optional "Label" parameter unset. (For OAEP
  43. padding, see ftp://ftp.rsasecurity.com/pub/pkcs/pkcs-1/pkcs-1v2-1.pdf)
  44. For Diffie-Hellman, we use a generator (g) of 2. For the modulus (p), we
  45. use the 1024-bit safe prime from rfc2409 section 6.2 whose hex
  46. representation is:
  47. "FFFFFFFFFFFFFFFFC90FDAA22168C234C4C6628B80DC1CD129024E08"
  48. "8A67CC74020BBEA63B139B22514A08798E3404DDEF9519B3CD3A431B"
  49. "302B0A6DF25F14374FE1356D6D51C245E485B576625E7EC6F44C42E9"
  50. "A637ED6B0BFF5CB6F406B7EDEE386BFB5A899FA5AE9F24117C4B1FE6"
  51. "49286651ECE65381FFFFFFFFFFFFFFFF"
  52. As an optimization, implementations SHOULD choose DH private keys (x) of
  53. 320 bits. Implementations that do this MUST never use any DH key more
  54. than once.
  55. [May other implementations reuse their DH keys?? -RD]
  56. [Probably not. Conceivably, you could get away with changing DH keys once
  57. per second, but there are too many oddball attacks for me to be
  58. comfortable that this is safe. -NM]
  59. For a hash function, we use SHA-1.
  60. KEY_LEN=16.
  61. DH_LEN=128; DH_SEC_LEN=40.
  62. PK_ENC_LEN=128; PK_PAD_LEN=42.
  63. HASH_LEN=20.
  64. When we refer to "the hash of a public key", we mean the SHA-1 hash of the
  65. DER encoding of an ASN.1 RSA public key (as specified in PKCS.1).
  66. All "random" values should be generated with a cryptographically strong
  67. random number generator, unless otherwise noted.
  68. The "hybrid encryption" of a byte sequence M with a public key PK is
  69. computed as follows:
  70. 1. If M is less than PK_ENC_LEN-PK_PAD_LEN, pad and encrypt M with PK.
  71. 2. Otherwise, generate a KEY_LEN byte random key K.
  72. Let M1 = the first PK_ENC_LEN-PK_PAD_LEN-KEY_LEN bytes of M,
  73. and let M2 = the rest of M.
  74. Pad and encrypt K|M1 with PK. Encrypt M2 with our stream cipher,
  75. using the key K. Concatenate these encrypted values.
  76. [XXX Note that this "hybrid encryption" approach does not prevent
  77. an attacker from adding or removing bytes to the end of M. It also
  78. allows attackers to modify the bytes not covered by the OAEP --
  79. see Goldberg's PET2006 paper for details. We will add a MAC to this
  80. scheme one day. -RD]
  81. 0.4. Other parameter values
  82. CELL_LEN=512
  83. 1. System overview
  84. Tor is a distributed overlay network designed to anonymize
  85. low-latency TCP-based applications such as web browsing, secure shell,
  86. and instant messaging. Clients choose a path through the network and
  87. build a ``circuit'', in which each node (or ``onion router'' or ``OR'')
  88. in the path knows its predecessor and successor, but no other nodes in
  89. the circuit. Traffic flowing down the circuit is sent in fixed-size
  90. ``cells'', which are unwrapped by a symmetric key at each node (like
  91. the layers of an onion) and relayed downstream.
  92. 1.1. Keys and names
  93. Every Tor server has multiple public/private keypairs:
  94. - A long-term signing-only "Identity key" used to sign documents and
  95. certificates, and used to establish server identity.
  96. - A medium-term "Onion key" used to decrypt onion skins when accepting
  97. circuit extend attempts. (See 5.1.) Old keys MUST be accepted for at
  98. least one week after they are no longer advertised. Because of this,
  99. servers MUST retain old keys for a while after they're rotated.
  100. - A short-term "Connection key" used to negotiate TLS connections.
  101. Tor implementations MAY rotate this key as often as they like, and
  102. SHOULD rotate this key at least once a day.
  103. Tor servers are also identified by "nicknames"; these are specified in
  104. dir-spec.txt.
  105. 2. Connections
  106. Connections between two Tor servers, or between a client and a server,
  107. use TLS/SSLv3 for link authentication and encryption. All
  108. implementations MUST support the SSLv3 ciphersuite
  109. "SSL_DHE_RSA_WITH_3DES_EDE_CBC_SHA", and SHOULD support the TLS
  110. ciphersuite "TLS_DHE_RSA_WITH_AES_128_CBC_SHA" if it is available.
  111. There are three acceptable ways to perform a TLS handshake when
  112. connecting to a Tor server: "certificates up-front", "renegotiation", and
  113. "backwards-compatible renegotiation". ("Backwards-compatible
  114. renegotiation" is, as the name implies, compatible with both other
  115. handshake types.)
  116. Before Tor 0.2.0.21, only "certificates up-front" was supported. In Tor
  117. 0.2.0.21 or later, "backwards-compatible renegotiation" is used.
  118. In "certificates up-front", the connection initiator always sends a
  119. two-certificate chain, consisting of an X.509 certificate using a
  120. short-term connection public key and a second, self- signed X.509
  121. certificate containing its identity key. The other party sends a similar
  122. certificate chain. The initiator's ClientHello MUST NOT include any
  123. ciphersuites other than:
  124. TLS_DHE_RSA_WITH_AES_256_CBC_SHA
  125. TLS_DHE_RSA_WITH_AES_128_CBC_SHA
  126. SSL_DHE_RSA_WITH_3DES_EDE_CBC_SHA
  127. SSL_DHE_DSS_WITH_3DES_EDE_CBC_SHA
  128. In "renegotiation", the connection initiator sends no certificates, and
  129. the responder sends a single connection certificate. Once the TLS
  130. handshake is complete, the initiator renegotiates the handshake, with each
  131. parties sending a two-certificate chain as in "certificates up-front".
  132. The initiator's ClientHello MUST include at least once ciphersuite not in
  133. the list above. The responder SHOULD NOT select any ciphersuite besides
  134. those in the list above.
  135. [The above "should not" is because some of the ciphers that
  136. clients list may be fake.]
  137. In "backwards-compatible renegotiation", the connection initiator's
  138. ClientHello MUST include at least one ciphersuite other than those listed
  139. above. The connection responder examines the initiator's ciphersuite list
  140. to see whether it includes any ciphers other than those included in the
  141. list above. If extra ciphers are included, the responder proceeds as in
  142. "renegotiation": it sends a single certificate and does not request
  143. client certificates. Otherwise (in the case that no extra ciphersuites
  144. are included in the ClientHello) the responder proceeds as in
  145. "certificates up-front": it requests client certificates, and sends a
  146. two-certificate chain. In either case, once the responder has sent its
  147. certificate or certificates, the initiator counts them. If two
  148. certificates have been sent, it proceeds as in "certificates up-front";
  149. otherwise, it proceeds as in "renegotiation".
  150. All new implementations of the Tor server protocol MUST support
  151. "backwards-compatible renegotiation"; clients SHOULD do this too. If
  152. this is not possible, new client implementations MUST support both
  153. "renegotiation" and "certificates up-front" and use the router's
  154. published link protocols list (see dir-spec.txt on the "protocols" entry)
  155. to decide which to use.
  156. In all of the above handshake variants, certificates sent in the clear
  157. SHOULD NOT include any strings to identify the host as a Tor server. In
  158. the "renegotation" and "backwards-compatible renegotiation", the
  159. initiator SHOULD chose a list of ciphersuites and TLS extensions chosen
  160. to mimic one used by a popular web browser.
  161. Responders MUST NOT select any TLS ciphersuite that lacks ephemeral keys,
  162. or whose symmetric keys are less then KEY_LEN bits, or whose digests are
  163. less than HASH_LEN bits. Responders SHOULD NOT select any SSLv3
  164. ciphersuite other than those listed above.
  165. Even though the connection protocol is identical, we will think of the
  166. initiator as either an onion router (OR) if it is willing to relay
  167. traffic for other Tor users, or an onion proxy (OP) if it only handles
  168. local requests. Onion proxies SHOULD NOT provide long-term-trackable
  169. identifiers in their handshakes.
  170. In all handshake variants, once all certificates are exchanged, all
  171. parties receiving certificates must confirm that the identity key is as
  172. expected. (When initiating a connection, the expected identity key is
  173. the one given in the directory; when creating a connection because of an
  174. EXTEND cell, the expected identity key is the one given in the cell.) If
  175. the key is not as expected, the party must close the connection.
  176. When connecting to an OR, all parties SHOULD reject the connection if that
  177. OR has a malformed or missing certificate. When accepting an incoming
  178. connection, an OR SHOULD NOT reject incoming connections from parties with
  179. malformed or missing certificates. (However, an OR should not believe
  180. that an incoming connection is from another OR unless the certificates
  181. are present and well-formed.)
  182. [Before version 0.1.2.8-rc, ORs rejected incoming connections from ORs and
  183. OPs alike if their certificates were missing or malformed.]
  184. Once a TLS connection is established, the two sides send cells
  185. (specified below) to one another. Cells are sent serially. All
  186. cells are CELL_LEN bytes long. Cells may be sent embedded in TLS
  187. records of any size or divided across TLS records, but the framing
  188. of TLS records MUST NOT leak information about the type or contents
  189. of the cells.
  190. TLS connections are not permanent. Either side MAY close a connection
  191. if there are no circuits running over it and an amount of time
  192. (KeepalivePeriod, defaults to 5 minutes) has passed since the last time
  193. any traffic was transmitted over the TLS connection. Clients SHOULD
  194. also hold a TLS connection with no circuits open, if it is likely that a
  195. circuit will be built soon using that connection.
  196. (As an exception, directory servers may try to stay connected to all of
  197. the ORs -- though this will be phased out for the Tor 0.1.2.x release.)
  198. To avoid being trivially distinguished from servers, client-only Tor
  199. instances are encouraged but not required to use a two-certificate chain
  200. as well. Clients SHOULD NOT keep using the same certificates when
  201. their IP address changes. Clients MAY send no certificates at all.
  202. 3. Cell Packet format
  203. The basic unit of communication for onion routers and onion
  204. proxies is a fixed-width "cell".
  205. On a version 1 connection, each cell contains the following
  206. fields:
  207. CircID [2 bytes]
  208. Command [1 byte]
  209. Payload (padded with 0 bytes) [PAYLOAD_LEN bytes]
  210. On a version 2 connection, all cells are as in version 1 connections,
  211. except for the initial VERSIONS cell, whose format is:
  212. Circuit [2 octets; set to 0]
  213. Command [1 octet; set to 7 for VERSIONS]
  214. Length [2 octets; big-endian integer]
  215. Payload [Length bytes]
  216. The CircID field determines which circuit, if any, the cell is
  217. associated with.
  218. The 'Command' field holds one of the following values:
  219. 0 -- PADDING (Padding) (See Sec 7.2)
  220. 1 -- CREATE (Create a circuit) (See Sec 5.1)
  221. 2 -- CREATED (Acknowledge create) (See Sec 5.1)
  222. 3 -- RELAY (End-to-end data) (See Sec 5.5 and 6)
  223. 4 -- DESTROY (Stop using a circuit) (See Sec 5.4)
  224. 5 -- CREATE_FAST (Create a circuit, no PK) (See Sec 5.1)
  225. 6 -- CREATED_FAST (Circuit created, no PK) (See Sec 5.1)
  226. 7 -- VERSIONS (Negotiate proto version) (See Sec 4)
  227. 8 -- NETINFO (Time and address info) (See Sec 4)
  228. 9 -- RELAY_EARLY (End-to-end data; limited) (See sec 5.6)
  229. The interpretation of 'Payload' depends on the type of the cell.
  230. PADDING: Payload is unused.
  231. CREATE: Payload contains the handshake challenge.
  232. CREATED: Payload contains the handshake response.
  233. RELAY: Payload contains the relay header and relay body.
  234. DESTROY: Payload contains a reason for closing the circuit.
  235. (see 5.4)
  236. Upon receiving any other value for the command field, an OR must
  237. drop the cell. Since more cell types may be added in the future, ORs
  238. should generally not warn when encountering unrecognized commands.
  239. The payload is padded with 0 bytes.
  240. PADDING cells are currently used to implement connection keepalive.
  241. If there is no other traffic, ORs and OPs send one another a PADDING
  242. cell every few minutes.
  243. CREATE, CREATED, and DESTROY cells are used to manage circuits;
  244. see section 5 below.
  245. RELAY cells are used to send commands and data along a circuit; see
  246. section 6 below.
  247. VERSIONS and NETINFO cells are used to set up connections. See section 4
  248. below.
  249. 4. Negotiating and initializing connections
  250. 4.1. Negotiating versions with VERSIONS cells
  251. There are multiple instances of the Tor link connection protocol. Any
  252. connection negotiated using the "certificates up front" handshake (see
  253. section 2 above) is "version 1". In any connection where both parties
  254. have behaved as in the "renegotiation" handshake, the link protocol
  255. version is 2 or higher.
  256. To determine the version, in any connection where the "renegotiation"
  257. handshake was used (that is, where the server sent only one certificate
  258. at first and where the client did not send any certificates until
  259. renegotiation), both parties MUST send a VERSIONS cell immediately after
  260. the renegotiation is finished, before any other cells are sent. Parties
  261. MUST NOT send any other cells on a connection until they have received a
  262. VERSIONS cell.
  263. The payload in a VERSIONS cell is a series of big-endian two-byte
  264. integers. Both parties MUST select as the link protocol version the
  265. highest number contained both in the VERSIONS cell they sent and in the
  266. versions cell they received. If they have no such version in common,
  267. they cannot communicate and MUST close the connection.
  268. Since the version 1 link protocol does not use the "renegotiation"
  269. handshake, implementations MUST NOT list version 1 in their VERSIONS
  270. cell.
  271. 4.2. NETINFO cells
  272. If version 2 or higher is negotiated, each party sends the other a
  273. NETINFO cell. The cell's payload is:
  274. Timestamp [4 bytes]
  275. Other OR's address [variable]
  276. Number of addresses [1 byte]
  277. This OR's addresses [variable]
  278. The address format is a type/length/value sequence as given in section
  279. 6.4 below. The timestamp is a big-endian unsigned integer number of
  280. seconds since the unix epoch.
  281. Implementations MAY use the timestamp value to help decide if their
  282. clocks are skewed. Initiators MAY use "other OR's address" to help
  283. learn which address their connections are originating from, if they do
  284. not know it. Initiators SHOULD use "this OR's address" to make sure
  285. that they have connected to another OR at its canonical address.
  286. [As of 0.2.0.23-rc, implementations use none of the above values.]
  287. 5. Circuit management
  288. 5.1. CREATE and CREATED cells
  289. Users set up circuits incrementally, one hop at a time. To create a
  290. new circuit, OPs send a CREATE cell to the first node, with the
  291. first half of the DH handshake; that node responds with a CREATED
  292. cell with the second half of the DH handshake plus the first 20 bytes
  293. of derivative key data (see section 5.2). To extend a circuit past
  294. the first hop, the OP sends an EXTEND relay cell (see section 5)
  295. which instructs the last node in the circuit to send a CREATE cell
  296. to extend the circuit.
  297. The payload for a CREATE cell is an 'onion skin', which consists
  298. of the first step of the DH handshake data (also known as g^x).
  299. This value is hybrid-encrypted (see 0.3) to Bob's onion key, giving
  300. an onion-skin of:
  301. PK-encrypted:
  302. Padding [PK_PAD_LEN bytes]
  303. Symmetric key [KEY_LEN bytes]
  304. First part of g^x [PK_ENC_LEN-PK_PAD_LEN-KEY_LEN bytes]
  305. Symmetrically encrypted:
  306. Second part of g^x [DH_LEN-(PK_ENC_LEN-PK_PAD_LEN-KEY_LEN)
  307. bytes]
  308. The relay payload for an EXTEND relay cell consists of:
  309. Address [4 bytes]
  310. Port [2 bytes]
  311. Onion skin [DH_LEN+KEY_LEN+PK_PAD_LEN bytes]
  312. Identity fingerprint [HASH_LEN bytes]
  313. The port and address field denote the IPV4 address and port of the next
  314. onion router in the circuit; the public key hash is the hash of the PKCS#1
  315. ASN1 encoding of the next onion router's identity (signing) key. (See 0.3
  316. above.) Including this hash allows the extending OR verify that it is
  317. indeed connected to the correct target OR, and prevents certain
  318. man-in-the-middle attacks.
  319. The payload for a CREATED cell, or the relay payload for an
  320. EXTENDED cell, contains:
  321. DH data (g^y) [DH_LEN bytes]
  322. Derivative key data (KH) [HASH_LEN bytes] <see 5.2 below>
  323. The CircID for a CREATE cell is an arbitrarily chosen 2-byte integer,
  324. selected by the node (OP or OR) that sends the CREATE cell. To prevent
  325. CircID collisions, when one node sends a CREATE cell to another, it chooses
  326. from only one half of the possible values based on the ORs' public
  327. identity keys: if the sending node has a lower key, it chooses a CircID with
  328. an MSB of 0; otherwise, it chooses a CircID with an MSB of 1.
  329. (An OP with no public key MAY choose any CircID it wishes, since an OP
  330. never needs to process a CREATE cell.)
  331. Public keys are compared numerically by modulus.
  332. As usual with DH, x and y MUST be generated randomly.
  333. 5.1.1. CREATE_FAST/CREATED_FAST cells
  334. When initializing the first hop of a circuit, the OP has already
  335. established the OR's identity and negotiated a secret key using TLS.
  336. Because of this, it is not always necessary for the OP to perform the
  337. public key operations to create a circuit. In this case, the
  338. OP MAY send a CREATE_FAST cell instead of a CREATE cell for the first
  339. hop only. The OR responds with a CREATED_FAST cell, and the circuit is
  340. created.
  341. A CREATE_FAST cell contains:
  342. Key material (X) [HASH_LEN bytes]
  343. A CREATED_FAST cell contains:
  344. Key material (Y) [HASH_LEN bytes]
  345. Derivative key data [HASH_LEN bytes] (See 5.2 below)
  346. The values of X and Y must be generated randomly.
  347. If an OR sees a circuit created with CREATE_FAST, the OR is sure to be the
  348. first hop of a circuit. ORs SHOULD reject attempts to create streams with
  349. RELAY_BEGIN exiting the circuit at the first hop: letting Tor be used as a
  350. single hop proxy makes exit nodes a more attractive target for compromise.
  351. 5.2. Setting circuit keys
  352. Once the handshake between the OP and an OR is completed, both can
  353. now calculate g^xy with ordinary DH. Before computing g^xy, both client
  354. and server MUST verify that the received g^x or g^y value is not degenerate;
  355. that is, it must be strictly greater than 1 and strictly less than p-1
  356. where p is the DH modulus. Implementations MUST NOT complete a handshake
  357. with degenerate keys. Implementations MUST NOT discard other "weak"
  358. g^x values.
  359. (Discarding degenerate keys is critical for security; if bad keys
  360. are not discarded, an attacker can substitute the server's CREATED
  361. cell's g^y with 0 or 1, thus creating a known g^xy and impersonating
  362. the server. Discarding other keys may allow attacks to learn bits of
  363. the private key.)
  364. If CREATE or EXTEND is used to extend a circuit, the client and server
  365. base their key material on K0=g^xy, represented as a big-endian unsigned
  366. integer.
  367. If CREATE_FAST is used, the client and server base their key material on
  368. K0=X|Y.
  369. From the base key material K0, they compute KEY_LEN*2+HASH_LEN*3 bytes of
  370. derivative key data as
  371. K = H(K0 | [00]) | H(K0 | [01]) | H(K0 | [02]) | ...
  372. The first HASH_LEN bytes of K form KH; the next HASH_LEN form the forward
  373. digest Df; the next HASH_LEN 41-60 form the backward digest Db; the next
  374. KEY_LEN 61-76 form Kf, and the final KEY_LEN form Kb. Excess bytes from K
  375. are discarded.
  376. KH is used in the handshake response to demonstrate knowledge of the
  377. computed shared key. Df is used to seed the integrity-checking hash
  378. for the stream of data going from the OP to the OR, and Db seeds the
  379. integrity-checking hash for the data stream from the OR to the OP. Kf
  380. is used to encrypt the stream of data going from the OP to the OR, and
  381. Kb is used to encrypt the stream of data going from the OR to the OP.
  382. 5.3. Creating circuits
  383. When creating a circuit through the network, the circuit creator
  384. (OP) performs the following steps:
  385. 1. Choose an onion router as an exit node (R_N), such that the onion
  386. router's exit policy includes at least one pending stream that
  387. needs a circuit (if there are any).
  388. 2. Choose a chain of (N-1) onion routers
  389. (R_1...R_N-1) to constitute the path, such that no router
  390. appears in the path twice.
  391. 3. If not already connected to the first router in the chain,
  392. open a new connection to that router.
  393. 4. Choose a circID not already in use on the connection with the
  394. first router in the chain; send a CREATE cell along the
  395. connection, to be received by the first onion router.
  396. 5. Wait until a CREATED cell is received; finish the handshake
  397. and extract the forward key Kf_1 and the backward key Kb_1.
  398. 6. For each subsequent onion router R (R_2 through R_N), extend
  399. the circuit to R.
  400. To extend the circuit by a single onion router R_M, the OP performs
  401. these steps:
  402. 1. Create an onion skin, encrypted to R_M's public onion key.
  403. 2. Send the onion skin in a relay EXTEND cell along
  404. the circuit (see section 5).
  405. 3. When a relay EXTENDED cell is received, verify KH, and
  406. calculate the shared keys. The circuit is now extended.
  407. When an onion router receives an EXTEND relay cell, it sends a CREATE
  408. cell to the next onion router, with the enclosed onion skin as its
  409. payload. As special cases, if the extend cell includes a digest of
  410. all zeroes, or asks to extend back to the relay that sent the extend
  411. cell, the circuit will fail and be torn down. The initiating onion
  412. router chooses some circID not yet used on the connection between the
  413. two onion routers. (But see section 5.1. above, concerning choosing
  414. circIDs based on lexicographic order of nicknames.)
  415. When an onion router receives a CREATE cell, if it already has a
  416. circuit on the given connection with the given circID, it drops the
  417. cell. Otherwise, after receiving the CREATE cell, it completes the
  418. DH handshake, and replies with a CREATED cell. Upon receiving a
  419. CREATED cell, an onion router packs it payload into an EXTENDED relay
  420. cell (see section 5), and sends that cell up the circuit. Upon
  421. receiving the EXTENDED relay cell, the OP can retrieve g^y.
  422. (As an optimization, OR implementations may delay processing onions
  423. until a break in traffic allows time to do so without harming
  424. network latency too greatly.)
  425. 5.3.1. Canonical connections
  426. It is possible for an attacker to launch a man-in-the-middle attack
  427. against a connection by telling OR Alice to extend to OR Bob at some
  428. address X controlled by the attacker. The attacker cannot read the
  429. encrypted traffic, but the attacker is now in a position to count all
  430. bytes sent between Alice and Bob (assuming Alice was not already
  431. connected to Bob.)
  432. To prevent this, when an OR we gets an extend request, it SHOULD use an
  433. existing OR connection if the ID matches, and ANY of the following
  434. conditions hold:
  435. - The IP matches the requested IP.
  436. - The OR knows that the IP of the connection it's using is canonical
  437. because it was listed in the NETINFO cell.
  438. - The OR knows that the IP of the connection it's using is canonical
  439. because it was listed in the server descriptor.
  440. [This is not implemented in Tor 0.2.0.23-rc.]
  441. 5.4. Tearing down circuits
  442. Circuits are torn down when an unrecoverable error occurs along
  443. the circuit, or when all streams on a circuit are closed and the
  444. circuit's intended lifetime is over. Circuits may be torn down
  445. either completely or hop-by-hop.
  446. To tear down a circuit completely, an OR or OP sends a DESTROY
  447. cell to the adjacent nodes on that circuit, using the appropriate
  448. direction's circID.
  449. Upon receiving an outgoing DESTROY cell, an OR frees resources
  450. associated with the corresponding circuit. If it's not the end of
  451. the circuit, it sends a DESTROY cell for that circuit to the next OR
  452. in the circuit. If the node is the end of the circuit, then it tears
  453. down any associated edge connections (see section 6.1).
  454. After a DESTROY cell has been processed, an OR ignores all data or
  455. destroy cells for the corresponding circuit.
  456. To tear down part of a circuit, the OP may send a RELAY_TRUNCATE cell
  457. signaling a given OR (Stream ID zero). That OR sends a DESTROY
  458. cell to the next node in the circuit, and replies to the OP with a
  459. RELAY_TRUNCATED cell.
  460. When an unrecoverable error occurs along one connection in a
  461. circuit, the nodes on either side of the connection should, if they
  462. are able, act as follows: the node closer to the OP should send a
  463. RELAY_TRUNCATED cell towards the OP; the node farther from the OP
  464. should send a DESTROY cell down the circuit.
  465. The payload of a RELAY_TRUNCATED or DESTROY cell contains a single octet,
  466. describing why the circuit is being closed or truncated. When sending a
  467. TRUNCATED or DESTROY cell because of another TRUNCATED or DESTROY cell,
  468. the error code should be propagated. The origin of a circuit always sets
  469. this error code to 0, to avoid leaking its version.
  470. The error codes are:
  471. 0 -- NONE (No reason given.)
  472. 1 -- PROTOCOL (Tor protocol violation.)
  473. 2 -- INTERNAL (Internal error.)
  474. 3 -- REQUESTED (A client sent a TRUNCATE command.)
  475. 4 -- HIBERNATING (Not currently operating; trying to save bandwidth.)
  476. 5 -- RESOURCELIMIT (Out of memory, sockets, or circuit IDs.)
  477. 6 -- CONNECTFAILED (Unable to reach server.)
  478. 7 -- OR_IDENTITY (Connected to server, but its OR identity was not
  479. as expected.)
  480. 8 -- OR_CONN_CLOSED (The OR connection that was carrying this circuit
  481. died.)
  482. 9 -- FINISHED (The circuit has expired for being dirty or old.)
  483. 10 -- TIMEOUT (Circuit construction took too long)
  484. 11 -- DESTROYED (The circuit was destroyed w/o client TRUNCATE)
  485. 12 -- NOSUCHSERVICE (Request for unknown hidden service)
  486. 5.5. Routing relay cells
  487. When an OR receives a RELAY or RELAY_EARLY cell, it checks the cell's
  488. circID and determines whether it has a corresponding circuit along that
  489. connection. If not, the OR drops the cell.
  490. Otherwise, if the OR is not at the OP edge of the circuit (that is,
  491. either an 'exit node' or a non-edge node), it de/encrypts the payload
  492. with the stream cipher, as follows:
  493. 'Forward' relay cell (same direction as CREATE):
  494. Use Kf as key; decrypt.
  495. 'Back' relay cell (opposite direction from CREATE):
  496. Use Kb as key; encrypt.
  497. Note that in counter mode, decrypt and encrypt are the same operation.
  498. The OR then decides whether it recognizes the relay cell, by
  499. inspecting the payload as described in section 6.1 below. If the OR
  500. recognizes the cell, it processes the contents of the relay cell.
  501. Otherwise, it passes the decrypted relay cell along the circuit if
  502. the circuit continues. If the OR at the end of the circuit
  503. encounters an unrecognized relay cell, an error has occurred: the OR
  504. sends a DESTROY cell to tear down the circuit.
  505. When a relay cell arrives at an OP, the OP decrypts the payload
  506. with the stream cipher as follows:
  507. OP receives data cell:
  508. For I=N...1,
  509. Decrypt with Kb_I. If the payload is recognized (see
  510. section 6..1), then stop and process the payload.
  511. For more information, see section 6 below.
  512. 5.6. Handling relay_early cells
  513. A RELAY_EARLY cell is designed to limit the length any circuit can reach.
  514. When an OR receives a RELAY_EARLY cell, and the next node in the circuit
  515. is speaking v2 of the link protocol or later, the OR relays the cell as a
  516. RELAY_EARLY cell. Otherwise, it relays it as a RELAY cell.
  517. If a node ever receives more than 8 RELAY_EARLY cells on a given
  518. outbound circuit, it SHOULD close the circuit. (For historical reasons,
  519. we don't limit the number of inbound RELAY_EARLY cells; they should
  520. be harmless anyway because clients won't accept extend requests. See
  521. bug 1038.)
  522. When speaking v2 of the link protocol or later, clients MUST only send
  523. EXTEND cells inside RELAY_EARLY cells. Clients SHOULD send the first ~8
  524. RELAY cells that are not targeted at the first hop of any circuit as
  525. RELAY_EARLY cells too, in order to partially conceal the circuit length.
  526. [In a future version of Tor, servers will reject any EXTEND cell not
  527. received in a RELAY_EARLY cell. See proposal 110.]
  528. 6. Application connections and stream management
  529. 6.1. Relay cells
  530. Within a circuit, the OP and the exit node use the contents of
  531. RELAY packets to tunnel end-to-end commands and TCP connections
  532. ("Streams") across circuits. End-to-end commands can be initiated
  533. by either edge; streams are initiated by the OP.
  534. The payload of each unencrypted RELAY cell consists of:
  535. Relay command [1 byte]
  536. 'Recognized' [2 bytes]
  537. StreamID [2 bytes]
  538. Digest [4 bytes]
  539. Length [2 bytes]
  540. Data [CELL_LEN-14 bytes]
  541. The relay commands are:
  542. 1 -- RELAY_BEGIN [forward]
  543. 2 -- RELAY_DATA [forward or backward]
  544. 3 -- RELAY_END [forward or backward]
  545. 4 -- RELAY_CONNECTED [backward]
  546. 5 -- RELAY_SENDME [forward or backward] [sometimes control]
  547. 6 -- RELAY_EXTEND [forward] [control]
  548. 7 -- RELAY_EXTENDED [backward] [control]
  549. 8 -- RELAY_TRUNCATE [forward] [control]
  550. 9 -- RELAY_TRUNCATED [backward] [control]
  551. 10 -- RELAY_DROP [forward or backward] [control]
  552. 11 -- RELAY_RESOLVE [forward]
  553. 12 -- RELAY_RESOLVED [backward]
  554. 13 -- RELAY_BEGIN_DIR [forward]
  555. 32..40 -- Used for hidden services; see rend-spec.txt.
  556. Commands labelled as "forward" must only be sent by the originator
  557. of the circuit. Commands labelled as "backward" must only be sent by
  558. other nodes in the circuit back to the originator. Commands marked
  559. as either can be sent either by the originator or other nodes.
  560. The 'recognized' field in any unencrypted relay payload is always set
  561. to zero; the 'digest' field is computed as the first four bytes of
  562. the running digest of all the bytes that have been destined for
  563. this hop of the circuit or originated from this hop of the circuit,
  564. seeded from Df or Db respectively (obtained in section 5.2 above),
  565. and including this RELAY cell's entire payload (taken with the digest
  566. field set to zero).
  567. When the 'recognized' field of a RELAY cell is zero, and the digest
  568. is correct, the cell is considered "recognized" for the purposes of
  569. decryption (see section 5.5 above).
  570. (The digest does not include any bytes from relay cells that do
  571. not start or end at this hop of the circuit. That is, it does not
  572. include forwarded data. Therefore if 'recognized' is zero but the
  573. digest does not match, the running digest at that node should
  574. not be updated, and the cell should be forwarded on.)
  575. All RELAY cells pertaining to the same tunneled stream have the
  576. same stream ID. StreamIDs are chosen arbitrarily by the OP. RELAY
  577. cells that affect the entire circuit rather than a particular
  578. stream use a StreamID of zero -- they are marked in the table above
  579. as "[control]" style cells. (Sendme cells are marked as "sometimes
  580. control" because they can take include a StreamID or not depending
  581. on their purpose -- see Section 7.)
  582. The 'Length' field of a relay cell contains the number of bytes in
  583. the relay payload which contain real payload data. The remainder of
  584. the payload is padded with NUL bytes.
  585. If the RELAY cell is recognized but the relay command is not
  586. understood, the cell must be dropped and ignored. Its contents
  587. still count with respect to the digests, though.
  588. 6.2. Opening streams and transferring data
  589. To open a new anonymized TCP connection, the OP chooses an open
  590. circuit to an exit that may be able to connect to the destination
  591. address, selects an arbitrary StreamID not yet used on that circuit,
  592. and constructs a RELAY_BEGIN cell with a payload encoding the address
  593. and port of the destination host. The payload format is:
  594. ADDRESS | ':' | PORT | [00]
  595. where ADDRESS can be a DNS hostname, or an IPv4 address in
  596. dotted-quad format, or an IPv6 address surrounded by square brackets;
  597. and where PORT is a decimal integer between 1 and 65535, inclusive.
  598. [What is the [00] for? -NM]
  599. [It's so the payload is easy to parse out with string funcs -RD]
  600. Upon receiving this cell, the exit node resolves the address as
  601. necessary, and opens a new TCP connection to the target port. If the
  602. address cannot be resolved, or a connection can't be established, the
  603. exit node replies with a RELAY_END cell. (See 6.4 below.)
  604. Otherwise, the exit node replies with a RELAY_CONNECTED cell, whose
  605. payload is in one of the following formats:
  606. The IPv4 address to which the connection was made [4 octets]
  607. A number of seconds (TTL) for which the address may be cached [4 octets]
  608. or
  609. Four zero-valued octets [4 octets]
  610. An address type (6) [1 octet]
  611. The IPv6 address to which the connection was made [16 octets]
  612. A number of seconds (TTL) for which the address may be cached [4 octets]
  613. [XXXX No version of Tor currently generates the IPv6 format.]
  614. [Tor servers before 0.1.2.0 set the TTL field to a fixed value. Later
  615. versions set the TTL to the last value seen from a DNS server, and expire
  616. their own cached entries after a fixed interval. This prevents certain
  617. attacks.]
  618. The OP waits for a RELAY_CONNECTED cell before sending any data.
  619. Once a connection has been established, the OP and exit node
  620. package stream data in RELAY_DATA cells, and upon receiving such
  621. cells, echo their contents to the corresponding TCP stream.
  622. RELAY_DATA cells sent to unrecognized streams are dropped.
  623. Relay RELAY_DROP cells are long-range dummies; upon receiving such
  624. a cell, the OR or OP must drop it.
  625. 6.2.1. Opening a directory stream
  626. If a Tor server is a directory server, it should respond to a
  627. RELAY_BEGIN_DIR cell as if it had received a BEGIN cell requesting a
  628. connection to its directory port. RELAY_BEGIN_DIR cells ignore exit
  629. policy, since the stream is local to the Tor process.
  630. If the Tor server is not running a directory service, it should respond
  631. with a REASON_NOTDIRECTORY RELAY_END cell.
  632. Clients MUST generate an all-zero payload for RELAY_BEGIN_DIR cells,
  633. and servers MUST ignore the payload.
  634. [RELAY_BEGIN_DIR was not supported before Tor 0.1.2.2-alpha; clients
  635. SHOULD NOT send it to routers running earlier versions of Tor.]
  636. 6.3. Closing streams
  637. When an anonymized TCP connection is closed, or an edge node
  638. encounters error on any stream, it sends a 'RELAY_END' cell along the
  639. circuit (if possible) and closes the TCP connection immediately. If
  640. an edge node receives a 'RELAY_END' cell for any stream, it closes
  641. the TCP connection completely, and sends nothing more along the
  642. circuit for that stream.
  643. The payload of a RELAY_END cell begins with a single 'reason' byte to
  644. describe why the stream is closing, plus optional data (depending on
  645. the reason.) The values are:
  646. 1 -- REASON_MISC (catch-all for unlisted reasons)
  647. 2 -- REASON_RESOLVEFAILED (couldn't look up hostname)
  648. 3 -- REASON_CONNECTREFUSED (remote host refused connection) [*]
  649. 4 -- REASON_EXITPOLICY (OR refuses to connect to host or port)
  650. 5 -- REASON_DESTROY (Circuit is being destroyed)
  651. 6 -- REASON_DONE (Anonymized TCP connection was closed)
  652. 7 -- REASON_TIMEOUT (Connection timed out, or OR timed out
  653. while connecting)
  654. 8 -- (unallocated) [**]
  655. 9 -- REASON_HIBERNATING (OR is temporarily hibernating)
  656. 10 -- REASON_INTERNAL (Internal error at the OR)
  657. 11 -- REASON_RESOURCELIMIT (OR has no resources to fulfill request)
  658. 12 -- REASON_CONNRESET (Connection was unexpectedly reset)
  659. 13 -- REASON_TORPROTOCOL (Sent when closing connection because of
  660. Tor protocol violations.)
  661. 14 -- REASON_NOTDIRECTORY (Client sent RELAY_BEGIN_DIR to a
  662. non-directory server.)
  663. (With REASON_EXITPOLICY, the 4-byte IPv4 address or 16-byte IPv6 address
  664. forms the optional data, along with a 4-byte TTL; no other reason
  665. currently has extra data.)
  666. OPs and ORs MUST accept reasons not on the above list, since future
  667. versions of Tor may provide more fine-grained reasons.
  668. Tors SHOULD NOT send any reason except REASON_MISC for a stream that they
  669. have originated.
  670. [*] Older versions of Tor also send this reason when connections are
  671. reset.
  672. [**] Due to a bug in versions of Tor through 0095, error reason 8 must
  673. remain allocated until that version is obsolete.
  674. --- [The rest of this section describes unimplemented functionality.]
  675. Because TCP connections can be half-open, we follow an equivalent
  676. to TCP's FIN/FIN-ACK/ACK protocol to close streams.
  677. An exit connection can have a TCP stream in one of three states:
  678. 'OPEN', 'DONE_PACKAGING', and 'DONE_DELIVERING'. For the purposes
  679. of modeling transitions, we treat 'CLOSED' as a fourth state,
  680. although connections in this state are not, in fact, tracked by the
  681. onion router.
  682. A stream begins in the 'OPEN' state. Upon receiving a 'FIN' from
  683. the corresponding TCP connection, the edge node sends a 'RELAY_FIN'
  684. cell along the circuit and changes its state to 'DONE_PACKAGING'.
  685. Upon receiving a 'RELAY_FIN' cell, an edge node sends a 'FIN' to
  686. the corresponding TCP connection (e.g., by calling
  687. shutdown(SHUT_WR)) and changing its state to 'DONE_DELIVERING'.
  688. When a stream in already in 'DONE_DELIVERING' receives a 'FIN', it
  689. also sends a 'RELAY_FIN' along the circuit, and changes its state
  690. to 'CLOSED'. When a stream already in 'DONE_PACKAGING' receives a
  691. 'RELAY_FIN' cell, it sends a 'FIN' and changes its state to
  692. 'CLOSED'.
  693. If an edge node encounters an error on any stream, it sends a
  694. 'RELAY_END' cell (if possible) and closes the stream immediately.
  695. 6.4. Remote hostname lookup
  696. To find the address associated with a hostname, the OP sends a
  697. RELAY_RESOLVE cell containing the hostname to be resolved with a nul
  698. terminating byte. (For a reverse lookup, the OP sends a RELAY_RESOLVE
  699. cell containing an in-addr.arpa address.) The OR replies with a
  700. RELAY_RESOLVED cell containing a status byte, and any number of
  701. answers. Each answer is of the form:
  702. Type (1 octet)
  703. Length (1 octet)
  704. Value (variable-width)
  705. TTL (4 octets)
  706. "Length" is the length of the Value field.
  707. "Type" is one of:
  708. 0x00 -- Hostname
  709. 0x04 -- IPv4 address
  710. 0x06 -- IPv6 address
  711. 0xF0 -- Error, transient
  712. 0xF1 -- Error, nontransient
  713. If any answer has a type of 'Error', then no other answer may be given.
  714. The RELAY_RESOLVE cell must use a nonzero, distinct streamID; the
  715. corresponding RELAY_RESOLVED cell must use the same streamID. No stream
  716. is actually created by the OR when resolving the name.
  717. 7. Flow control
  718. 7.1. Link throttling
  719. Each client or relay should do appropriate bandwidth throttling to
  720. keep its user happy.
  721. Communicants rely on TCP's default flow control to push back when they
  722. stop reading.
  723. The mainline Tor implementation uses token buckets (one for reads,
  724. one for writes) for the rate limiting.
  725. Since 0.2.0.x, Tor has let the user specify an additional pair of
  726. token buckets for "relayed" traffic, so people can deploy a Tor relay
  727. with strict rate limiting, but also use the same Tor as a client. To
  728. avoid partitioning concerns we combine both classes of traffic over a
  729. given OR connection, and keep track of the last time we read or wrote
  730. a high-priority (non-relayed) cell. If it's been less than N seconds
  731. (currently N=30), we give the whole connection high priority, else we
  732. give the whole connection low priority. We also give low priority
  733. to reads and writes for connections that are serving directory
  734. information. See proposal 111 for details.
  735. 7.2. Link padding
  736. Link padding can be created by sending PADDING cells along the
  737. connection; relay cells of type "DROP" can be used for long-range
  738. padding.
  739. Currently nodes are not required to do any sort of link padding or
  740. dummy traffic. Because strong attacks exist even with link padding,
  741. and because link padding greatly increases the bandwidth requirements
  742. for running a node, we plan to leave out link padding until this
  743. tradeoff is better understood.
  744. 7.3. Circuit-level flow control
  745. To control a circuit's bandwidth usage, each OR keeps track of two
  746. 'windows', consisting of how many RELAY_DATA cells it is allowed to
  747. originate (package for transmission), and how many RELAY_DATA cells
  748. it is willing to consume (receive for local streams). These limits
  749. do not apply to cells that the OR receives from one host and relays
  750. to another.
  751. Each 'window' value is initially set to 1000 data cells
  752. in each direction (cells that are not data cells do not affect
  753. the window). When an OR is willing to deliver more cells, it sends a
  754. RELAY_SENDME cell towards the OP, with Stream ID zero. When an OR
  755. receives a RELAY_SENDME cell with stream ID zero, it increments its
  756. packaging window.
  757. Each of these cells increments the corresponding window by 100.
  758. The OP behaves identically, except that it must track a packaging
  759. window and a delivery window for every OR in the circuit.
  760. An OR or OP sends cells to increment its delivery window when the
  761. corresponding window value falls under some threshold (900).
  762. If a packaging window reaches 0, the OR or OP stops reading from
  763. TCP connections for all streams on the corresponding circuit, and
  764. sends no more RELAY_DATA cells until receiving a RELAY_SENDME cell.
  765. [this stuff is badly worded; copy in the tor-design section -RD]
  766. 7.4. Stream-level flow control
  767. Edge nodes use RELAY_SENDME cells to implement end-to-end flow
  768. control for individual connections across circuits. Similarly to
  769. circuit-level flow control, edge nodes begin with a window of cells
  770. (500) per stream, and increment the window by a fixed value (50)
  771. upon receiving a RELAY_SENDME cell. Edge nodes initiate RELAY_SENDME
  772. cells when both a) the window is <= 450, and b) there are less than
  773. ten cell payloads remaining to be flushed at that edge.
  774. A.1. Differences between spec and implementation
  775. - The current specification requires all ORs to have IPv4 addresses, but
  776. allows servers to exit and resolve to IPv6 addresses, and to declare IPv6
  777. addresses in their exit policies. The current codebase has no IPv6
  778. support at all.