tor-spec-v2.txt 41 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414415416417418419420421422423424425426427428429430431432433434435436437438439440441442443444445446447448449450451452453454455456457458459460461462463464465466467468469470471472473474475476477478479480481482483484485486487488489490491492493494495496497498499500501502503504505506507508509510511512513514515516517518519520521522523524525526527528529530531532533534535536537538539540541542543544545546547548549550551552553554555556557558559560561562563564565566567568569570571572573574575576577578579580581582583584585586587588589590591592593594595596597598599600601602603604605606607608609610611612613614615616617618619620621622623624625626627628629630631632633634635636637638639640641642643644645646647648649650651652653654655656657658659660661662663664665666667668669670671672673674675676677678679680681682683684685686687688689690691692693694695696697698699700701702703704705706707708709710711712713714715716717718719720721722723724725726727728729730731732733734735736737738739740741742743744745746747748749750751752753754755756757758759760761762763764765766767768769770771772773774775776777778779780781782783784785786787788789790791792793794795796797798799800801802803804805806807808809810811812813814815816817818819820821822823824825826827828829830831832833834835836837838839840841842843844845846847848849850851852853854855856857858859860861862863864865866867868869870871872873874875876877878879880881882883884885886887888889890891892893894895896897898899900901902903904905906907908909910911912913914915916917918919920921922923924925926927928929930931932933934935936937938939940941942
  1. $Id$
  2. Tor Protocol Specification
  3. Roger Dingledine
  4. Nick Mathewson
  5. Note: This document aims to specify Tor as implemented in 0.1.2.1-alpha-dev
  6. and later. Future versions of Tor will implement improved protocols, and
  7. compatibility is not guaranteed.
  8. THIS DOCUMENT IS UNSTABLE. Right now, we're revising the protocol to remove
  9. a few long-standing limitations. For the most stable current version of the
  10. protocol, see tor-spec-v0.txt; current versions of Tor are backward-compatible.
  11. This specification is not a design document; most design criteria
  12. are not examined. For more information on why Tor acts as it does,
  13. see tor-design.pdf.
  14. TODO for v2 revision:
  15. - Fix onionskin handshake scheme to be more mainstream, less nutty.
  16. Can we just do
  17. E(HMAC(g^x), g^x) rather than just E(g^x) ?
  18. No, that has the same flaws as before. We should send
  19. E(g^x, C) with random C and expect g^y, HMAC_C(K=g^xy).
  20. Better ask Ian; probably Stephen too.
  21. - Versioned CREATE and friends
  22. - Length on CREATE and friends
  23. - Versioning on circuits
  24. - Versioning on create cells
  25. - SHA1 is showing its age
  26. - Not being able to upgrade ciphersuites or increase key lengths is
  27. lame.
  28. TODO:
  29. - REASON_CONNECTFAILED should include an IP.
  30. - Copy prose from tor-design to make everything more readable.
  31. - Spec when we should rotate which keys (tls, link, etc)?
  32. 0. Preliminaries
  33. 0.1. Notation and encoding
  34. PK -- a public key.
  35. SK -- a private key.
  36. K -- a key for a symmetric cypher.
  37. a|b -- concatenation of 'a' and 'b'.
  38. [A0 B1 C2] -- a three-byte sequence, containing the bytes with
  39. hexadecimal values A0, B1, and C2, in that order.
  40. All numeric values are encoded in network (big-endian) order.
  41. H(m) -- a cryptographic hash of m.
  42. 0.2. Security parameters
  43. Tor uses a stream cipher, a public-key cipher, the Diffie-Hellman
  44. protocol, and a hash function.
  45. KEY_LEN -- the length of the stream cipher's key, in bytes.
  46. PK_ENC_LEN -- the length of a public-key encrypted message, in bytes.
  47. PK_PAD_LEN -- the number of bytes added in padding for public-key
  48. encryption, in bytes. (The largest number of bytes that can be encrypted
  49. in a single public-key operation is therefore PK_ENC_LEN-PK_PAD_LEN.)
  50. DH_LEN -- the number of bytes used to represent a member of the
  51. Diffie-Hellman group.
  52. DH_SEC_LEN -- the number of bytes used in a Diffie-Hellman private key (x).
  53. HASH_LEN -- the length of the hash function's output, in bytes.
  54. PAYLOAD_LEN -- The longest allowable cell payload, in bytes. (509)
  55. CELL_LEN -- The length of a Tor cell, in bytes.
  56. 0.3. Ciphers
  57. For a stream cipher, we use 128-bit AES in counter mode, with an IV of all
  58. 0 bytes.
  59. For a public-key cipher, we use RSA with 1024-bit keys and a fixed
  60. exponent of 65537. We use OAEP padding, with SHA-1 as its digest
  61. function. (For OAEP padding, see
  62. ftp://ftp.rsasecurity.com/pub/pkcs/pkcs-1/pkcs-1v2-1.pdf)
  63. For Diffie-Hellman, we use a generator (g) of 2. For the modulus (p), we
  64. use the 1024-bit safe prime from rfc2409 section 6.2 whose hex
  65. representation is:
  66. "FFFFFFFFFFFFFFFFC90FDAA22168C234C4C6628B80DC1CD129024E08"
  67. "8A67CC74020BBEA63B139B22514A08798E3404DDEF9519B3CD3A431B"
  68. "302B0A6DF25F14374FE1356D6D51C245E485B576625E7EC6F44C42E9"
  69. "A637ED6B0BFF5CB6F406B7EDEE386BFB5A899FA5AE9F24117C4B1FE6"
  70. "49286651ECE65381FFFFFFFFFFFFFFFF"
  71. As an optimization, implementations SHOULD choose DH private keys (x) of
  72. 320 bits. Implementations that do this MUST never use any DH key more
  73. than once.
  74. [May other implementations reuse their DH keys?? -RD]
  75. [Probably not. Conceivably, you could get away with changing DH keys once
  76. per second, but there are too many oddball attacks for me to be
  77. comfortable that this is safe. -NM]
  78. For a hash function, we use SHA-1.
  79. KEY_LEN=16.
  80. DH_LEN=128; DH_SEC_LEN=40.
  81. PK_ENC_LEN=128; PK_PAD_LEN=42.
  82. HASH_LEN=20.
  83. When we refer to "the hash of a public key", we mean the SHA-1 hash of the
  84. DER encoding of an ASN.1 RSA public key (as specified in PKCS.1).
  85. All "random" values should be generated with a cryptographically strong
  86. random number generator, unless otherwise noted.
  87. The "hybrid encryption" of a byte sequence M with a public key PK is
  88. computed as follows:
  89. 1. If M is less than PK_ENC_LEN-PK_PAD_LEN, pad and encrypt M with PK.
  90. 2. Otherwise, generate a KEY_LEN byte random key K.
  91. Let M1 = the first PK_ENC_LEN-PK_PAD_LEN-KEY_LEN bytes of M,
  92. and let M2 = the rest of M.
  93. Pad and encrypt K|M1 with PK. Encrypt M2 with our stream cipher,
  94. using the key K. Concatenate these encrypted values.
  95. [XXX Note that this "hybrid encryption" approach does not prevent
  96. an attacker from adding or removing bytes to the end of M. It also
  97. allows attackers to modify the bytes not covered by the OAEP --
  98. see Goldberg's PET2006 paper for details. We will add a MAC to this
  99. scheme one day. -RD]
  100. 0.4. Other parameter values
  101. CELL_LEN=512
  102. 1. System overview
  103. Tor is a distributed overlay network designed to anonymize
  104. low-latency TCP-based applications such as web browsing, secure shell,
  105. and instant messaging. Clients choose a path through the network and
  106. build a ``circuit'', in which each node (or ``onion router'' or ``OR'')
  107. in the path knows its predecessor and successor, but no other nodes in
  108. the circuit. Traffic flowing down the circuit is sent in fixed-size
  109. ``cells'', which are unwrapped by a symmetric key at each node (like
  110. the layers of an onion) and relayed downstream.
  111. 1.1. Protocol Versioning
  112. The node-to-node TLS-based "OR connection" protocol and the multi-hop
  113. "circuit" protocol are versioned quasi-independently. (Certain versions
  114. of the circuit protocol may require a minimum version of the connection
  115. protocol to be used.)
  116. Version numbers are incremented for backward-incompatible protocol changes
  117. only. Backward-compatible changes are generally implemented by adding
  118. additional fields to existing structures; implementations MUST ignore
  119. fields they do not expect.
  120. Parties negotiate OR connection versions as described below in sections
  121. 4.1 and 4.2.
  122. 2. Connections
  123. Tor uses TLS for link encryption. All implementations MUST support
  124. the TLS ciphersuite "TLS_EDH_RSA_WITH_DES_192_CBC3_SHA", and SHOULD
  125. support "TLS_DHE_RSA_WITH_AES_128_CBC_SHA" if it is available.
  126. Implementations MAY support other ciphersuites, but MUST NOT
  127. support any suite without ephemeral keys, symmetric keys of at
  128. least KEY_LEN bits, and digests of at least HASH_LEN bits.
  129. Even though the connection protocol is identical, we think of the
  130. initiator as either an onion router (OR) if it is willing to relay
  131. traffic for other Tor users, or an onion proxy (OP) if it only handles
  132. local requests. Onion proxies SHOULD NOT provide long-term-trackable
  133. identifiers in their handshakes.
  134. The connection initiator always sends a two-certificate chain,
  135. consisting of a
  136. certificate using a short-term connection key and a second, self-
  137. signed certificate containing the OR's identity key. The commonName of the
  138. first certificate is the OR's nickname, and the commonName of the second
  139. certificate is the OR's nickname, followed by a space and the string
  140. "<identity>".
  141. Implementations running Protocol 1 and earlier use an
  142. organizationName of "Tor" or "TOR". Future implementations (which
  143. support the version negotiation protocol in section 4.1) MUST NOT
  144. have either of these values for their organizationName.
  145. All parties receiving certificates must confirm that the identity key is
  146. as expected. (When initiating a connection, the expected identity key is
  147. the one given in the directory; when creating a connection because of an
  148. EXTEND cell, the expected identity key is the one given in the cell.) If
  149. the key is not as expected, the party must close the connection.
  150. All parties SHOULD reject connections to or from ORs that have malformed
  151. or missing certificates. ORs MAY accept or reject connections from OPs
  152. with malformed or missing certificates.
  153. Once a TLS connection is established, the two sides send cells
  154. (specified below) to one another. Cells are sent serially. All
  155. cells are CELL_LEN bytes long. Cells may be sent embedded in TLS
  156. records of any size or divided across TLS records, but the framing
  157. of TLS records MUST NOT leak information about the type or contents
  158. of the cells.
  159. TLS connections are not permanent. Either side may close a connection
  160. if there are no circuits running over it and an amount of time
  161. (KeepalivePeriod, defaults to 5 minutes) has passed.
  162. (As an exception, directory servers may try to stay connected to all of
  163. the ORs -- though this will be phased out for the Tor 0.1.2.x release.)
  164. 3. Cell Packet format
  165. The basic unit of communication for onion routers and onion
  166. proxies is a fixed-width "cell".
  167. On a version 1 connection, each cell contains the following
  168. fields:
  169. CircID [2 bytes]
  170. Command [1 byte]
  171. Payload (padded with 0 bytes) [PAYLOAD_LEN bytes]
  172. On a version 2 connection, each cell contains the following fields:
  173. CircID [3 bytes]
  174. Command [1 byte]
  175. Payload (padded with 0 bytes) [PAYLOAD_LEN bytes]
  176. The CircID field determines which circuit, if any, the cell is
  177. associated with.
  178. The 'Command' field holds one of the following values:
  179. 0 -- PADDING (Padding) (See Sec 7.2)
  180. 1 -- CREATE (Create a circuit) (See Sec 5.1)
  181. 2 -- CREATED (Acknowledge create) (See Sec 5.1)
  182. 3 -- RELAY (End-to-end data) (See Sec 5.5 and 6)
  183. 4 -- DESTROY (Stop using a circuit) (See Sec 5.4)
  184. 5 -- CREATE_FAST (Create a circuit, no PK) (See Sec 5.1)
  185. 6 -- CREATED_FAST (Circuit created, no PK) (See Sec 5.1)
  186. 7 -- VERSIONS (Negotiate versions) (See Sec 4.1)
  187. 8 -- NETINFO (Time and MITM-prevention) (See Sec 4.2)
  188. The interpretation of 'Payload' depends on the type of the cell.
  189. PADDING: Payload is unused.
  190. CREATE: Payload contains the handshake challenge.
  191. CREATED: Payload contains the handshake response.
  192. RELAY: Payload contains the relay header and relay body.
  193. DESTROY: Payload contains a reason for closing the circuit.
  194. (see 5.4)
  195. Upon receiving any other value for the command field, an OR must
  196. drop the cell. [XXXX Versions prior to 0.1.0.?? logged a warning
  197. when dropping the cell; this is bad behavior. -NM]
  198. The payload is padded with 0 bytes.
  199. PADDING cells are currently used to implement connection keepalive.
  200. If there is no other traffic, ORs and OPs send one another a PADDING
  201. cell every few minutes.
  202. CREATE, CREATED, and DESTROY cells are used to manage circuits;
  203. see section 4 below.
  204. RELAY cells are used to send commands and data along a circuit; see
  205. section 5 below.
  206. VERSIONS cells are used to introduce parameters and characteristics of
  207. Tor clients and servers when connections are established.
  208. 4, Connection management
  209. Upon establishing a TLS connection, both parties immediately begin
  210. negotiating a connection protocol version and other connection parameters.
  211. 4.1. VERSIONS cells
  212. When a Tor connection is established, both parties normally send a
  213. VERSIONS cell before sending any other cells. (But see below.)
  214. NumVersions [1 byte]
  215. Versions [NumVersions bytes]
  216. "Versions" is a sequence of NumVersions link connection protocol versions,
  217. each one byte long. Parties should list all of the versions which they
  218. are able and willing to support. Parties can only communicate if they
  219. have some connection protocol version in common.
  220. Version 0.1.x.y-alpha and earlier don't understand VERSIONS cells,
  221. and therefore don't support version negotiation. Thus, waiting until
  222. the other side has sent a VERSIONS cell won't work for these servers:
  223. if they send no cells back, it is impossible to tell whether they
  224. have sent a VERSIONS cell that has been stalled, or whether they have
  225. dropped our own VERSIONS cell as unrecognized. Thus, immediately after
  226. a TLS connection has been established, the parties check whether the
  227. other side has an obsolete certificate (organizationName equal to "Tor"
  228. or "TOR"). If the other party presented an obsolete certificate,
  229. we assume a v0 connection. Otherwise, both parties send VERSIONS
  230. cells listing all their supported versions. Upon receiving the
  231. other party's VERSIONS cell, the implementation begins using the
  232. highest-valued version common to both cells. If the first cell from
  233. the other party is _not_ a VERSIONS cell, we assume a v0 protocol.
  234. Implementations MUST discard cells that are not the first cells sent on a
  235. connection.
  236. 4.2. MITM-prevention and time checking
  237. If we negotiate a v1 connection or higher, the first cell we send SHOULD
  238. be a NETINFO cell. Implementations SHOULD NOT send NETINFO cells at other
  239. times.
  240. A NETINFO cell contains:
  241. Timestamp [4 bytes]
  242. This OR's address [variable]
  243. Other OR's address [variable]
  244. Timestamp is the OR's current Unix time, in seconds since the epoch. If
  245. an implementation receives time values from many validated ORs that
  246. indicate that its clock is skewed, it SHOULD try to warn the
  247. administrator.
  248. Each address contains Type/Length/Value as used in Section 6.4. The first
  249. address is the address of the interface the party sending the VERSIONS cell
  250. used to connect to or accept connections from the other -- we include it
  251. to block a man-in-the-middle attack on TLS that lets an attacker bounce
  252. traffic through his own computers to enable timing and packet-counting
  253. attacks.
  254. The second address is the one that the party sending the VERSIONS cell
  255. believes the other has -- it can be used to learn what your IP address
  256. is if you have no other hints.
  257. 5. Circuit management
  258. 5.1. CREATE and CREATED cells
  259. Users set up circuits incrementally, one hop at a time. To create a
  260. new circuit, OPs send a CREATE cell to the first node, with the
  261. first half of the DH handshake; that node responds with a CREATED
  262. cell with the second half of the DH handshake plus the first 20 bytes
  263. of derivative key data (see section 5.2). To extend a circuit past
  264. the first hop, the OP sends an EXTEND relay cell (see section 5)
  265. which instructs the last node in the circuit to send a CREATE cell
  266. to extend the circuit.
  267. The payload for a CREATE cell is an 'onion skin', which consists
  268. of the first step of the DH handshake data (also known as g^x).
  269. This value is hybrid-encrypted (see 0.3) to Bob's public key, giving
  270. an onion-skin of:
  271. PK-encrypted:
  272. Padding padding [PK_PAD_LEN bytes]
  273. Symmetric key [KEY_LEN bytes]
  274. First part of g^x [PK_ENC_LEN-PK_PAD_LEN-KEY_LEN bytes]
  275. Symmetrically encrypted:
  276. Second part of g^x [DH_LEN-(PK_ENC_LEN-PK_PAD_LEN-KEY_LEN)
  277. bytes]
  278. The relay payload for an EXTEND relay cell consists of:
  279. Address [4 bytes]
  280. Port [2 bytes]
  281. Onion skin [DH_LEN+KEY_LEN+PK_PAD_LEN bytes]
  282. Identity fingerprint [HASH_LEN bytes]
  283. The port and address field denote the IPV4 address and port of the next
  284. onion router in the circuit; the public key hash is the hash of the PKCS#1
  285. ASN1 encoding of the next onion router's identity (signing) key. (See 0.3
  286. above.) (Including this hash allows the extending OR verify that it is
  287. indeed connected to the correct target OR, and prevents certain
  288. man-in-the-middle attacks.)
  289. The payload for a CREATED cell, or the relay payload for an
  290. EXTENDED cell, contains:
  291. DH data (g^y) [DH_LEN bytes]
  292. Derivative key data (KH) [HASH_LEN bytes] <see 5.2 below>
  293. The CircID for a CREATE cell is an arbitrarily chosen 2-byte integer,
  294. selected by the node (OP or OR) that sends the CREATE cell. To prevent
  295. CircID collisions, when one OR sends a CREATE cell to another, it chooses
  296. from only one half of the possible values based on the ORs' public
  297. identity keys: if the sending OR has a lower key, it chooses a CircID with
  298. an MSB of 0; otherwise, it chooses a CircID with an MSB of 1.
  299. Public keys are compared numerically by modulus.
  300. As usual with DH, x and y MUST be generated randomly.
  301. [
  302. To implement backward-compatible version negotiation, parties MUST
  303. drop CREATE cells with all-[00] onion-skins.
  304. ]
  305. 5.1.1. CREATE_FAST/CREATED_FAST cells
  306. When initializing the first hop of a circuit, the OP has already
  307. established the OR's identity and negotiated a secret key using TLS.
  308. Because of this, it is not always necessary for the OP to perform the
  309. public key operations to create a circuit. In this case, the
  310. OP MAY send a CREATE_FAST cell instead of a CREATE cell for the first
  311. hop only. The OR responds with a CREATED_FAST cell, and the circuit is
  312. created.
  313. A CREATE_FAST cell contains:
  314. Key material (X) [HASH_LEN bytes]
  315. A CREATED_FAST cell contains:
  316. Key material (Y) [HASH_LEN bytes]
  317. Derivative key data [HASH_LEN bytes] (See 5.2 below)
  318. The values of X and Y must be generated randomly.
  319. [Versions of Tor before 0.1.0.6-rc did not support these cell types;
  320. clients should not send CREATE_FAST cells to older Tor servers.]
  321. If an OR sees a circuit created with CREATE_FAST, the OR is sure to be the
  322. first hop of a circuit. ORs SHOULD reject attempts to create streams with
  323. RELAY_BEGIN exiting the circuit at the first hop: letting Tor be used as a
  324. single hop proxy makes exit nodes a more attractive target for compromise.
  325. 5.2. Setting circuit keys
  326. Once the handshake between the OP and an OR is completed, both can
  327. now calculate g^xy with ordinary DH. Before computing g^xy, both client
  328. and server MUST verify that the received g^x or g^y value is not degenerate;
  329. that is, it must be strictly greater than 1 and strictly less than p-1
  330. where p is the DH modulus. Implementations MUST NOT complete a handshake
  331. with degenerate keys. Implementations MUST NOT discard other "weak"
  332. g^x values.
  333. (Discarding degenerate keys is critical for security; if bad keys
  334. are not discarded, an attacker can substitute the server's CREATED
  335. cell's g^y with 0 or 1, thus creating a known g^xy and impersonating
  336. the server. Discarding other keys may allow attacks to learn bits of
  337. the private key.)
  338. (The mainline Tor implementation, in the 0.1.1.x-alpha series, discarded
  339. all g^x values less than 2^24, greater than p-2^24, or having more than
  340. 1024-16 identical bits. This served no useful purpose, and we stopped.)
  341. If CREATE or EXTEND is used to extend a circuit, the client and server
  342. base their key material on K0=g^xy, represented as a big-endian unsigned
  343. integer.
  344. If CREATE_FAST is used, the client and server base their key material on
  345. K0=X|Y.
  346. From the base key material K0, they compute KEY_LEN*2+HASH_LEN*3 bytes of
  347. derivative key data as
  348. K = H(K0 | [00]) | H(K0 | [01]) | H(K0 | [02]) | ...
  349. The first HASH_LEN bytes of K form KH; the next HASH_LEN form the forward
  350. digest Df; the next HASH_LEN 41-60 form the backward digest Db; the next
  351. KEY_LEN 61-76 form Kf, and the final KEY_LEN form Kb. Excess bytes from K
  352. are discarded.
  353. KH is used in the handshake response to demonstrate knowledge of the
  354. computed shared key. Df is used to seed the integrity-checking hash
  355. for the stream of data going from the OP to the OR, and Db seeds the
  356. integrity-checking hash for the data stream from the OR to the OP. Kf
  357. is used to encrypt the stream of data going from the OP to the OR, and
  358. Kb is used to encrypt the stream of data going from the OR to the OP.
  359. 5.3. Creating circuits
  360. When creating a circuit through the network, the circuit creator
  361. (OP) performs the following steps:
  362. 1. Choose an onion router as an exit node (R_N), such that the onion
  363. router's exit policy includes at least one pending stream that
  364. needs a circuit (if there are any).
  365. 2. Choose a chain of (N-1) onion routers
  366. (R_1...R_N-1) to constitute the path, such that no router
  367. appears in the path twice.
  368. 3. If not already connected to the first router in the chain,
  369. open a new connection to that router.
  370. 4. Choose a circID not already in use on the connection with the
  371. first router in the chain; send a CREATE cell along the
  372. connection, to be received by the first onion router.
  373. 5. Wait until a CREATED cell is received; finish the handshake
  374. and extract the forward key Kf_1 and the backward key Kb_1.
  375. 6. For each subsequent onion router R (R_2 through R_N), extend
  376. the circuit to R.
  377. To extend the circuit by a single onion router R_M, the OP performs
  378. these steps:
  379. 1. Create an onion skin, encrypted to R_M's public key.
  380. 2. Send the onion skin in a relay EXTEND cell along
  381. the circuit (see section 5).
  382. 3. When a relay EXTENDED cell is received, verify KH, and
  383. calculate the shared keys. The circuit is now extended.
  384. When an onion router receives an EXTEND relay cell, it sends a CREATE
  385. cell to the next onion router, with the enclosed onion skin as its
  386. payload. The initiating onion router chooses some circID not yet
  387. used on the connection between the two onion routers. (But see
  388. section 5.1. above, concerning choosing circIDs based on
  389. lexicographic order of nicknames.)
  390. When an onion router receives a CREATE cell, if it already has a
  391. circuit on the given connection with the given circID, it drops the
  392. cell. Otherwise, after receiving the CREATE cell, it completes the
  393. DH handshake, and replies with a CREATED cell. Upon receiving a
  394. CREATED cell, an onion router packs it payload into an EXTENDED relay
  395. cell (see section 5), and sends that cell up the circuit. Upon
  396. receiving the EXTENDED relay cell, the OP can retrieve g^y.
  397. (As an optimization, OR implementations may delay processing onions
  398. until a break in traffic allows time to do so without harming
  399. network latency too greatly.)
  400. 5.4. Tearing down circuits
  401. Circuits are torn down when an unrecoverable error occurs along
  402. the circuit, or when all streams on a circuit are closed and the
  403. circuit's intended lifetime is over. Circuits may be torn down
  404. either completely or hop-by-hop.
  405. To tear down a circuit completely, an OR or OP sends a DESTROY
  406. cell to the adjacent nodes on that circuit, using the appropriate
  407. direction's circID.
  408. Upon receiving an outgoing DESTROY cell, an OR frees resources
  409. associated with the corresponding circuit. If it's not the end of
  410. the circuit, it sends a DESTROY cell for that circuit to the next OR
  411. in the circuit. If the node is the end of the circuit, then it tears
  412. down any associated edge connections (see section 6.1).
  413. After a DESTROY cell has been processed, an OR ignores all data or
  414. destroy cells for the corresponding circuit.
  415. To tear down part of a circuit, the OP may send a RELAY_TRUNCATE cell
  416. signaling a given OR (Stream ID zero). That OR sends a DESTROY
  417. cell to the next node in the circuit, and replies to the OP with a
  418. RELAY_TRUNCATED cell.
  419. When an unrecoverable error occurs along one connection in a
  420. circuit, the nodes on either side of the connection should, if they
  421. are able, act as follows: the node closer to the OP should send a
  422. RELAY_TRUNCATED cell towards the OP; the node farther from the OP
  423. should send a DESTROY cell down the circuit.
  424. The payload of a RELAY_TRUNCATED or DESTROY cell contains a single octet,
  425. describing why the circuit is being closed or truncated. When sending a
  426. TRUNCATED or DESTROY cell because of another TRUNCATED or DESTROY cell,
  427. the error code should be propagated. The origin of a circuit always sets
  428. this error code to 0, to avoid leaking its version.
  429. The error codes are:
  430. 0 -- NONE (No reason given.)
  431. 1 -- PROTOCOL (Tor protocol violation.)
  432. 2 -- INTERNAL (Internal error.)
  433. 3 -- REQUESTED (A client sent a TRUNCATE command.)
  434. 4 -- HIBERNATING (Not currently operating; trying to save bandwidth.)
  435. 5 -- RESOURCELIMIT (Out of memory, sockets, or circuit IDs.)
  436. 6 -- CONNECTFAILED (Unable to reach server.)
  437. 7 -- OR_IDENTITY (Connected to server, but its OR identity was not
  438. as expected.)
  439. 8 -- OR_CONN_CLOSED (The OR connection that was carrying this circuit
  440. died.)
  441. 9 -- FINISHED (The circuit has expired for being dirty or old.)
  442. 10 -- TIMEOUT (Circuit construction took too long)
  443. 11 -- DESTROYED (The circuit was destroyed w/o client TRUNCATE)
  444. 12 -- NOSUCHSERVICE (Request for unknown hidden service)
  445. [Versions of Tor prior to 0.1.0.11 didn't send reasons; implementations
  446. MUST accept empty TRUNCATED and DESTROY cells.]
  447. 5.5. Routing relay cells
  448. When an OR receives a RELAY cell, it checks the cell's circID and
  449. determines whether it has a corresponding circuit along that
  450. connection. If not, the OR drops the RELAY cell.
  451. Otherwise, if the OR is not at the OP edge of the circuit (that is,
  452. either an 'exit node' or a non-edge node), it de/encrypts the payload
  453. with the stream cipher, as follows:
  454. 'Forward' relay cell (same direction as CREATE):
  455. Use Kf as key; decrypt.
  456. 'Back' relay cell (opposite direction from CREATE):
  457. Use Kb as key; encrypt.
  458. Note that in counter mode, decrypt and encrypt are the same operation.
  459. The OR then decides whether it recognizes the relay cell, by
  460. inspecting the payload as described in section 6.1 below. If the OR
  461. recognizes the cell, it processes the contents of the relay cell.
  462. Otherwise, it passes the decrypted relay cell along the circuit if
  463. the circuit continues. If the OR at the end of the circuit
  464. encounters an unrecognized relay cell, an error has occurred: the OR
  465. sends a DESTROY cell to tear down the circuit.
  466. When a relay cell arrives at an OP, the OP decrypts the payload
  467. with the stream cipher as follows:
  468. OP receives data cell:
  469. For I=N...1,
  470. Decrypt with Kb_I. If the payload is recognized (see
  471. section 6..1), then stop and process the payload.
  472. For more information, see section 6 below.
  473. 6. Application connections and stream management
  474. 6.1. Relay cells
  475. Within a circuit, the OP and the exit node use the contents of
  476. RELAY packets to tunnel end-to-end commands and TCP connections
  477. ("Streams") across circuits. End-to-end commands can be initiated
  478. by either edge; streams are initiated by the OP.
  479. The payload of each unencrypted RELAY cell consists of:
  480. Relay command [1 byte]
  481. 'Recognized' [2 bytes]
  482. StreamID [2 bytes]
  483. Digest [4 bytes]
  484. Length [2 bytes]
  485. Data [CELL_LEN-14 bytes]
  486. The relay commands are:
  487. 1 -- RELAY_BEGIN [forward]
  488. 2 -- RELAY_DATA [forward or backward]
  489. 3 -- RELAY_END [forward or backward]
  490. 4 -- RELAY_CONNECTED [backward]
  491. 5 -- RELAY_SENDME [forward or backward] [sometimes control]
  492. 6 -- RELAY_EXTEND [forward] [control]
  493. 7 -- RELAY_EXTENDED [backward] [control]
  494. 8 -- RELAY_TRUNCATE [forward] [control]
  495. 9 -- RELAY_TRUNCATED [backward] [control]
  496. 10 -- RELAY_DROP [forward or backward] [control]
  497. 11 -- RELAY_RESOLVE [forward]
  498. 12 -- RELAY_RESOLVED [backward]
  499. 13 -- RELAY_BEGIN_DIR [forward]
  500. Commands labelled as "forward" must only be sent by the originator
  501. of the circuit. Commands labelled as "backward" must only be sent by
  502. other nodes in the circuit back to the originator. Commands marked
  503. as either can be sent either by the originator or other nodes.
  504. The 'recognized' field in any unencrypted relay payload is always set
  505. to zero; the 'digest' field is computed as the first four bytes of
  506. the running digest of all the bytes that have been destined for
  507. this hop of the circuit or originated from this hop of the circuit,
  508. seeded from Df or Db respectively (obtained in section 5.2 above),
  509. and including this RELAY cell's entire payload (taken with the digest
  510. field set to zero).
  511. When the 'recognized' field of a RELAY cell is zero, and the digest
  512. is correct, the cell is considered "recognized" for the purposes of
  513. decryption (see section 5.5 above).
  514. (The digest does not include any bytes from relay cells that do
  515. not start or end at this hop of the circuit. That is, it does not
  516. include forwarded data. Therefore if 'recognized' is zero but the
  517. digest does not match, the running digest at that node should
  518. not be updated, and the cell should be forwarded on.)
  519. All RELAY cells pertaining to the same tunneled stream have the
  520. same stream ID. StreamIDs are chosen arbitrarily by the OP. RELAY
  521. cells that affect the entire circuit rather than a particular
  522. stream use a StreamID of zero -- they are marked in the table above
  523. as "[control]" style cells. (Sendme cells are marked as "sometimes
  524. control" because they can take include a StreamID or not depending
  525. on their purpose -- see Section 7.)
  526. The 'Length' field of a relay cell contains the number of bytes in
  527. the relay payload which contain real payload data. The remainder of
  528. the payload is padded with NUL bytes.
  529. If the RELAY cell is recognized but the relay command is not
  530. understood, the cell must be dropped and ignored. Its contents
  531. still count with respect to the digests, though. [Before
  532. 0.1.1.10, Tor closed circuits when it received an unknown relay
  533. command. Perhaps this will be more forward-compatible. -RD]
  534. 6.2. Opening streams and transferring data
  535. To open a new anonymized TCP connection, the OP chooses an open
  536. circuit to an exit that may be able to connect to the destination
  537. address, selects an arbitrary StreamID not yet used on that circuit,
  538. and constructs a RELAY_BEGIN cell with a payload encoding the address
  539. and port of the destination host. The payload format is:
  540. ADDRESS | ':' | PORT | [00]
  541. where ADDRESS can be a DNS hostname, or an IPv4 address in
  542. dotted-quad format, or an IPv6 address surrounded by square brackets;
  543. and where PORT is encoded in decimal.
  544. [What is the [00] for? -NM]
  545. [It's so the payload is easy to parse out with string funcs -RD]
  546. Upon receiving this cell, the exit node resolves the address as
  547. necessary, and opens a new TCP connection to the target port. If the
  548. address cannot be resolved, or a connection can't be established, the
  549. exit node replies with a RELAY_END cell. (See 6.4 below.)
  550. Otherwise, the exit node replies with a RELAY_CONNECTED cell, whose
  551. payload is in one of the following formats:
  552. The IPv4 address to which the connection was made [4 octets]
  553. A number of seconds (TTL) for which the address may be cached [4 octets]
  554. or
  555. Four zero-valued octets [4 octets]
  556. An address type (6) [1 octet]
  557. The IPv6 address to which the connection was made [16 octets]
  558. A number of seconds (TTL) for which the address may be cached [4 octets]
  559. [XXXX Versions of Tor before 0.1.1.6 ignore and do not generate the TTL
  560. field. No version of Tor currently generates the IPv6 format.
  561. Tor servers before 0.1.2.0 set the TTL field to a fixed value. Later
  562. versions set the TTL to the last value seen from a DNS server, and expire
  563. their own cached entries after a fixed interval. This prevents certain
  564. attacks.]
  565. The OP waits for a RELAY_CONNECTED cell before sending any data.
  566. Once a connection has been established, the OP and exit node
  567. package stream data in RELAY_DATA cells, and upon receiving such
  568. cells, echo their contents to the corresponding TCP stream.
  569. RELAY_DATA cells sent to unrecognized streams are dropped.
  570. Relay RELAY_DROP cells are long-range dummies; upon receiving such
  571. a cell, the OR or OP must drop it.
  572. 6.2.1. Opening a directory stream
  573. If a Tor server is a directory server, it should respond to a
  574. RELAY_BEGIN_DIR cell as if it had received a BEGIN cell requesting a
  575. connection to its directory port. RELAY_BEGIN_DIR cells ignore exit
  576. policy, since the stream is local to the Tor process.
  577. If the Tor server is not running a directory service, it should respond
  578. with a REASON_NOTDIRECTORY RELAY_END cell.
  579. Clients MUST generate an all-zero payload for RELAY_BEGIN_DIR cells,
  580. and servers MUST ignore the payload.
  581. [RELAY_BEGIN_DIR was not supported before Tor 0.1.2.2-alpha; clients
  582. SHOULD NOT send it to routers running earlier versions of Tor.]
  583. 6.3. Closing streams
  584. When an anonymized TCP connection is closed, or an edge node
  585. encounters error on any stream, it sends a 'RELAY_END' cell along the
  586. circuit (if possible) and closes the TCP connection immediately. If
  587. an edge node receives a 'RELAY_END' cell for any stream, it closes
  588. the TCP connection completely, and sends nothing more along the
  589. circuit for that stream.
  590. The payload of a RELAY_END cell begins with a single 'reason' byte to
  591. describe why the stream is closing, plus optional data (depending on
  592. the reason.) The values are:
  593. 1 -- REASON_MISC (catch-all for unlisted reasons)
  594. 2 -- REASON_RESOLVEFAILED (couldn't look up hostname)
  595. 3 -- REASON_CONNECTREFUSED (remote host refused connection) [*]
  596. 4 -- REASON_EXITPOLICY (OR refuses to connect to host or port)
  597. 5 -- REASON_DESTROY (Circuit is being destroyed)
  598. 6 -- REASON_DONE (Anonymized TCP connection was closed)
  599. 7 -- REASON_TIMEOUT (Connection timed out, or OR timed out
  600. while connecting)
  601. 8 -- (unallocated) [**]
  602. 9 -- REASON_HIBERNATING (OR is temporarily hibernating)
  603. 10 -- REASON_INTERNAL (Internal error at the OR)
  604. 11 -- REASON_RESOURCELIMIT (OR has no resources to fulfill request)
  605. 12 -- REASON_CONNRESET (Connection was unexpectedly reset)
  606. 13 -- REASON_TORPROTOCOL (Sent when closing connection because of
  607. Tor protocol violations.)
  608. 14 -- REASON_NOTDIRECTORY (Client sent RELAY_BEGIN_DIR to a
  609. non-directory server.)
  610. (With REASON_EXITPOLICY, the 4-byte IPv4 address or 16-byte IPv6 address
  611. forms the optional data; no other reason currently has extra data.
  612. As of 0.1.1.6, the body also contains a 4-byte TTL.)
  613. OPs and ORs MUST accept reasons not on the above list, since future
  614. versions of Tor may provide more fine-grained reasons.
  615. [*] Older versions of Tor also send this reason when connections are
  616. reset.
  617. [**] Due to a bug in versions of Tor through 0095, error reason 8 must
  618. remain allocated until that version is obsolete.
  619. --- [The rest of this section describes unimplemented functionality.]
  620. Because TCP connections can be half-open, we follow an equivalent
  621. to TCP's FIN/FIN-ACK/ACK protocol to close streams.
  622. An exit connection can have a TCP stream in one of three states:
  623. 'OPEN', 'DONE_PACKAGING', and 'DONE_DELIVERING'. For the purposes
  624. of modeling transitions, we treat 'CLOSED' as a fourth state,
  625. although connections in this state are not, in fact, tracked by the
  626. onion router.
  627. A stream begins in the 'OPEN' state. Upon receiving a 'FIN' from
  628. the corresponding TCP connection, the edge node sends a 'RELAY_FIN'
  629. cell along the circuit and changes its state to 'DONE_PACKAGING'.
  630. Upon receiving a 'RELAY_FIN' cell, an edge node sends a 'FIN' to
  631. the corresponding TCP connection (e.g., by calling
  632. shutdown(SHUT_WR)) and changing its state to 'DONE_DELIVERING'.
  633. When a stream in already in 'DONE_DELIVERING' receives a 'FIN', it
  634. also sends a 'RELAY_FIN' along the circuit, and changes its state
  635. to 'CLOSED'. When a stream already in 'DONE_PACKAGING' receives a
  636. 'RELAY_FIN' cell, it sends a 'FIN' and changes its state to
  637. 'CLOSED'.
  638. If an edge node encounters an error on any stream, it sends a
  639. 'RELAY_END' cell (if possible) and closes the stream immediately.
  640. 6.4. Remote hostname lookup
  641. To find the address associated with a hostname, the OP sends a
  642. RELAY_RESOLVE cell containing the hostname to be resolved. (For a reverse
  643. lookup, the OP sends a RELAY_RESOLVE cell containing an in-addr.arpa
  644. address.) The OR replies with a RELAY_RESOLVED cell containing a status
  645. byte, and any number of answers. Each answer is of the form:
  646. Type (1 octet)
  647. Length (1 octet)
  648. Value (variable-width)
  649. TTL (4 octets)
  650. "Length" is the length of the Value field.
  651. "Type" is one of:
  652. 0x00 -- Hostname
  653. 0x04 -- IPv4 address
  654. 0x06 -- IPv6 address
  655. 0xF0 -- Error, transient
  656. 0xF1 -- Error, nontransient
  657. If any answer has a type of 'Error', then no other answer may be given.
  658. The RELAY_RESOLVE cell must use a nonzero, distinct streamID; the
  659. corresponding RELAY_RESOLVED cell must use the same streamID. No stream
  660. is actually created by the OR when resolving the name.
  661. 7. Flow control
  662. 7.1. Link throttling
  663. Each node should do appropriate bandwidth throttling to keep its
  664. user happy.
  665. Communicants rely on TCP's default flow control to push back when they
  666. stop reading.
  667. 7.2. Link padding
  668. Link padding can be created by sending PADDING cells along the
  669. connection; relay cells of type "DROP" can be used for long-range
  670. padding.
  671. Currently nodes are not required to do any sort of link padding or
  672. dummy traffic. Because strong attacks exist even with link padding,
  673. and because link padding greatly increases the bandwidth requirements
  674. for running a node, we plan to leave out link padding until this
  675. tradeoff is better understood.
  676. 7.3. Circuit-level flow control
  677. To control a circuit's bandwidth usage, each OR keeps track of
  678. two 'windows', consisting of how many RELAY_DATA cells it is
  679. allowed to package for transmission, and how many RELAY_DATA cells
  680. it is willing to deliver to streams outside the network.
  681. Each 'window' value is initially set to 1000 data cells
  682. in each direction (cells that are not data cells do not affect
  683. the window). When an OR is willing to deliver more cells, it sends a
  684. RELAY_SENDME cell towards the OP, with Stream ID zero. When an OR
  685. receives a RELAY_SENDME cell with stream ID zero, it increments its
  686. packaging window.
  687. Each of these cells increments the corresponding window by 100.
  688. The OP behaves identically, except that it must track a packaging
  689. window and a delivery window for every OR in the circuit.
  690. An OR or OP sends cells to increment its delivery window when the
  691. corresponding window value falls under some threshold (900).
  692. If a packaging window reaches 0, the OR or OP stops reading from
  693. TCP connections for all streams on the corresponding circuit, and
  694. sends no more RELAY_DATA cells until receiving a RELAY_SENDME cell.
  695. [this stuff is badly worded; copy in the tor-design section -RD]
  696. 7.4. Stream-level flow control
  697. Edge nodes use RELAY_SENDME cells to implement end-to-end flow
  698. control for individual connections across circuits. Similarly to
  699. circuit-level flow control, edge nodes begin with a window of cells
  700. (500) per stream, and increment the window by a fixed value (50)
  701. upon receiving a RELAY_SENDME cell. Edge nodes initiate RELAY_SENDME
  702. cells when both a) the window is <= 450, and b) there are less than
  703. ten cell payloads remaining to be flushed at that edge.
  704. A.1. Differences between spec and implementation
  705. - The current specification requires all ORs to have IPv4 addresses, but
  706. allows servers to exit and resolve to IPv6 addresses, and to declare IPv6
  707. addresses in their exit policies. The current codebase has no IPv6
  708. support at all.
  709. B. Things that should change in a later version of the Tor protocol
  710. B.1. ... but which will require backward-incompatible change
  711. - Circuit IDs should be longer.
  712. - IPv6 everywhere.
  713. - Maybe, keys should be longer.
  714. - Maybe, key-length should be adjustable. How to do this without
  715. making anonymity suck?
  716. - Drop backward compatibility.
  717. - We should use a 128-bit subgroup of our DH prime.
  718. - Handshake should use HMAC.
  719. - Multiple cell lengths.
  720. - Ability to split circuits across paths (If this is useful.)
  721. - SENDME windows should be dynamic.
  722. - Directory
  723. - Stop ever mentioning socks ports
  724. B.1. ... and that will require no changes
  725. - Mention multiple addr/port combos
  726. - Advertised outbound IP?
  727. - Migrate streams across circuits.
  728. B.2. ... and that we have no idea how to do.
  729. - UDP (as transport)
  730. - UDP (as content)
  731. - Use a better AES mode that has built-in integrity checking,
  732. doesn't grow with the number of hops, is not patented, and
  733. is implemented and maintained by smart people.