3 Tor Protocol Specification
8 Note: This document aims to specify Tor as implemented in 0.2.1.x. Future
9 versions of Tor may implement improved protocols, and compatibility is not
10 guaranteed. Compatibility notes are given for versions 0.1.1.15-rc and
11 later; earlier versions are not compatible with the Tor network as of this
14 This specification is not a design document; most design criteria
15 are not examined. For more information on why Tor acts as it does,
20 0.1. Notation and encoding
24 K -- a key for a symmetric cypher.
26 a|b -- concatenation of 'a' and 'b'.
28 [A0 B1 C2] -- a three-byte sequence, containing the bytes with
29 hexadecimal values A0, B1, and C2, in that order.
31 All numeric values are encoded in network (big-endian) order.
33 H(m) -- a cryptographic hash of m.
35 0.2. Security parameters
37 Tor uses a stream cipher, a public-key cipher, the Diffie-Hellman
38 protocol, and a hash function.
40 KEY_LEN -- the length of the stream cipher's key, in bytes.
42 PK_ENC_LEN -- the length of a public-key encrypted message, in bytes.
43 PK_PAD_LEN -- the number of bytes added in padding for public-key
44 encryption, in bytes. (The largest number of bytes that can be encrypted
45 in a single public-key operation is therefore PK_ENC_LEN-PK_PAD_LEN.)
47 DH_LEN -- the number of bytes used to represent a member of the
49 DH_SEC_LEN -- the number of bytes used in a Diffie-Hellman private key (x).
51 HASH_LEN -- the length of the hash function's output, in bytes.
53 PAYLOAD_LEN -- The longest allowable cell payload, in bytes. (509)
55 CELL_LEN -- The length of a Tor cell, in bytes.
59 For a stream cipher, we use 128-bit AES in counter mode, with an IV of all
62 For a public-key cipher, we use RSA with 1024-bit keys and a fixed
63 exponent of 65537. We use OAEP-MGF1 padding, with SHA-1 as its digest
64 function. We leave the optional "Label" parameter unset. (For OAEP
65 padding, see ftp://ftp.rsasecurity.com/pub/pkcs/pkcs-1/pkcs-1v2-1.pdf)
67 For Diffie-Hellman, we use a generator (g) of 2. For the modulus (p), we
68 use the 1024-bit safe prime from rfc2409 section 6.2 whose hex
71 "FFFFFFFFFFFFFFFFC90FDAA22168C234C4C6628B80DC1CD129024E08"
72 "8A67CC74020BBEA63B139B22514A08798E3404DDEF9519B3CD3A431B"
73 "302B0A6DF25F14374FE1356D6D51C245E485B576625E7EC6F44C42E9"
74 "A637ED6B0BFF5CB6F406B7EDEE386BFB5A899FA5AE9F24117C4B1FE6"
75 "49286651ECE65381FFFFFFFFFFFFFFFF"
77 As an optimization, implementations SHOULD choose DH private keys (x) of
78 320 bits. Implementations that do this MUST never use any DH key more
80 [May other implementations reuse their DH keys?? -RD]
81 [Probably not. Conceivably, you could get away with changing DH keys once
82 per second, but there are too many oddball attacks for me to be
83 comfortable that this is safe. -NM]
85 For a hash function, we use SHA-1.
88 DH_LEN=128; DH_SEC_LEN=40.
89 PK_ENC_LEN=128; PK_PAD_LEN=42.
92 When we refer to "the hash of a public key", we mean the SHA-1 hash of the
93 DER encoding of an ASN.1 RSA public key (as specified in PKCS.1).
95 All "random" values should be generated with a cryptographically strong
96 random number generator, unless otherwise noted.
98 The "hybrid encryption" of a byte sequence M with a public key PK is
100 1. If M is less than PK_ENC_LEN-PK_PAD_LEN, pad and encrypt M with PK.
101 2. Otherwise, generate a KEY_LEN byte random key K.
102 Let M1 = the first PK_ENC_LEN-PK_PAD_LEN-KEY_LEN bytes of M,
103 and let M2 = the rest of M.
104 Pad and encrypt K|M1 with PK. Encrypt M2 with our stream cipher,
105 using the key K. Concatenate these encrypted values.
106 [XXX Note that this "hybrid encryption" approach does not prevent
107 an attacker from adding or removing bytes to the end of M. It also
108 allows attackers to modify the bytes not covered by the OAEP --
109 see Goldberg's PET2006 paper for details. We will add a MAC to this
112 0.4. Other parameter values
118 Tor is a distributed overlay network designed to anonymize
119 low-latency TCP-based applications such as web browsing, secure shell,
120 and instant messaging. Clients choose a path through the network and
121 build a ``circuit'', in which each node (or ``onion router'' or ``OR'')
122 in the path knows its predecessor and successor, but no other nodes in
123 the circuit. Traffic flowing down the circuit is sent in fixed-size
124 ``cells'', which are unwrapped by a symmetric key at each node (like
125 the layers of an onion) and relayed downstream.
129 Every Tor server has multiple public/private keypairs:
131 - A long-term signing-only "Identity key" used to sign documents and
132 certificates, and used to establish server identity.
133 - A medium-term "Onion key" used to decrypt onion skins when accepting
134 circuit extend attempts. (See 5.1.) Old keys MUST be accepted for at
135 least one week after they are no longer advertised. Because of this,
136 servers MUST retain old keys for a while after they're rotated.
137 - A short-term "Connection key" used to negotiate TLS connections.
138 Tor implementations MAY rotate this key as often as they like, and
139 SHOULD rotate this key at least once a day.
141 Tor servers are also identified by "nicknames"; these are specified in
146 Connections between two Tor servers, or between a client and a server,
147 use TLS/SSLv3 for link authentication and encryption. All
148 implementations MUST support the SSLv3 ciphersuite
149 "SSL_DHE_RSA_WITH_3DES_EDE_CBC_SHA", and SHOULD support the TLS
150 ciphersuite "TLS_DHE_RSA_WITH_AES_128_CBC_SHA" if it is available.
152 There are three acceptable ways to perform a TLS handshake when
153 connecting to a Tor server: "certificates up-front", "renegotiation", and
154 "backwards-compatible renegotiation". ("Backwards-compatible
155 renegotiation" is, as the name implies, compatible with both other
158 Before Tor 0.2.0.21, only "certificates up-front" was supported. In Tor
159 0.2.0.21 or later, "backwards-compatible renegotiation" is used.
161 In "certificates up-front", the connection initiator always sends a
162 two-certificate chain, consisting of an X.509 certificate using a
163 short-term connection public key and a second, self- signed X.509
164 certificate containing its identity key. The other party sends a similar
165 certificate chain. The initiator's ClientHello MUST NOT include any
166 ciphersuites other than:
167 TLS_DHE_RSA_WITH_AES_256_CBC_SHA
168 TLS_DHE_RSA_WITH_AES_128_CBC_SHA
169 SSL_DHE_RSA_WITH_3DES_EDE_CBC_SHA
170 SSL_DHE_DSS_WITH_3DES_EDE_CBC_SHA
172 In "renegotiation", the connection initiator sends no certificates, and
173 the responder sends a single connection certificate. Once the TLS
174 handshake is complete, the initiator renegotiates the handshake, with each
175 parties sending a two-certificate chain as in "certificates up-front".
176 The initiator's ClientHello MUST include at least once ciphersuite not in
177 the list above. The responder SHOULD NOT select any ciphersuite besides
178 those in the list above.
179 [The above "should not" is because some of the ciphers that
180 clients list may be fake.]
182 In "backwards-compatible renegotiation", the connection initiator's
183 ClientHello MUST include at least one ciphersuite other than those listed
184 above. The connection responder examines the initiator's ciphersuite list
185 to see whether it includes any ciphers other than those included in the
186 list above. If extra ciphers are included, the responder proceeds as in
187 "renegotiation": it sends a single certificate and does not request
188 client certificates. Otherwise (in the case that no extra ciphersuites
189 are included in the ClientHello) the responder proceeds as in
190 "certificates up-front": it requests client certificates, and sends a
191 two-certificate chain. In either case, once the responder has sent its
192 certificate or certificates, the initiator counts them. If two
193 certificates have been sent, it proceeds as in "certificates up-front";
194 otherwise, it proceeds as in "renegotiation".
196 All new implementations of the Tor server protocol MUST support
197 "backwards-compatible renegotiation"; clients SHOULD do this too. If
198 this is not possible, new client implementations MUST support both
199 "renegotiation" and "certificates up-front" and use the router's
200 published link protocols list (see dir-spec.txt on the "protocols" entry)
201 to decide which to use.
203 In all of the above handshake variants, certificates sent in the clear
204 SHOULD NOT include any strings to identify the host as a Tor server. In
205 the "renegotation" and "backwards-compatible renegotiation", the
206 initiator SHOULD chose a list of ciphersuites and TLS extensions chosen
207 to mimic one used by a popular web browser.
209 Responders MUST NOT select any TLS ciphersuite that lacks ephemeral keys,
210 or whose symmetric keys are less then KEY_LEN bits, or whose digests are
211 less than HASH_LEN bits. Responders SHOULD NOT select any SSLv3
212 ciphersuite other than those listed above.
214 Even though the connection protocol is identical, we will think of the
215 initiator as either an onion router (OR) if it is willing to relay
216 traffic for other Tor users, or an onion proxy (OP) if it only handles
217 local requests. Onion proxies SHOULD NOT provide long-term-trackable
218 identifiers in their handshakes.
220 In all handshake variants, once all certificates are exchanged, all
221 parties receiving certificates must confirm that the identity key is as
222 expected. (When initiating a connection, the expected identity key is
223 the one given in the directory; when creating a connection because of an
224 EXTEND cell, the expected identity key is the one given in the cell.) If
225 the key is not as expected, the party must close the connection.
227 When connecting to an OR, all parties SHOULD reject the connection if that
228 OR has a malformed or missing certificate. When accepting an incoming
229 connection, an OR SHOULD NOT reject incoming connections from parties with
230 malformed or missing certificates. (However, an OR should not believe
231 that an incoming connection is from another OR unless the certificates
232 are present and well-formed.)
234 [Before version 0.1.2.8-rc, ORs rejected incoming connections from ORs and
235 OPs alike if their certificates were missing or malformed.]
237 Once a TLS connection is established, the two sides send cells
238 (specified below) to one another. Cells are sent serially. All
239 cells are CELL_LEN bytes long. Cells may be sent embedded in TLS
240 records of any size or divided across TLS records, but the framing
241 of TLS records MUST NOT leak information about the type or contents
244 TLS connections are not permanent. Either side MAY close a connection
245 if there are no circuits running over it and an amount of time
246 (KeepalivePeriod, defaults to 5 minutes) has passed since the last time
247 any traffic was transmitted over the TLS connection. Clients SHOULD
248 also hold a TLS connection with no circuits open, if it is likely that a
249 circuit will be built soon using that connection.
251 (As an exception, directory servers may try to stay connected to all of
252 the ORs -- though this will be phased out for the Tor 0.1.2.x release.)
254 To avoid being trivially distinguished from servers, client-only Tor
255 instances are encouraged but not required to use a two-certificate chain
256 as well. Clients SHOULD NOT keep using the same certificates when
257 their IP address changes. Clients MAY send no certificates at all.
259 3. Cell Packet format
261 The basic unit of communication for onion routers and onion
262 proxies is a fixed-width "cell".
264 On a version 1 connection, each cell contains the following
269 Payload (padded with 0 bytes) [PAYLOAD_LEN bytes]
271 On a version 2 connection, all cells are as in version 1 connections,
272 except for the initial VERSIONS cell, whose format is:
274 Circuit [2 octets; set to 0]
275 Command [1 octet; set to 7 for VERSIONS]
276 Length [2 octets; big-endian integer]
277 Payload [Length bytes]
279 The CircID field determines which circuit, if any, the cell is
282 The 'Command' field holds one of the following values:
283 0 -- PADDING (Padding) (See Sec 7.2)
284 1 -- CREATE (Create a circuit) (See Sec 5.1)
285 2 -- CREATED (Acknowledge create) (See Sec 5.1)
286 3 -- RELAY (End-to-end data) (See Sec 5.5 and 6)
287 4 -- DESTROY (Stop using a circuit) (See Sec 5.4)
288 5 -- CREATE_FAST (Create a circuit, no PK) (See Sec 5.1)
289 6 -- CREATED_FAST (Circuit created, no PK) (See Sec 5.1)
290 7 -- VERSIONS (Negotiate proto version) (See Sec 4)
291 8 -- NETINFO (Time and address info) (See Sec 4)
292 9 -- RELAY_EARLY (End-to-end data; limited) (See sec 5.6)
294 The interpretation of 'Payload' depends on the type of the cell.
295 PADDING: Payload is unused.
296 CREATE: Payload contains the handshake challenge.
297 CREATED: Payload contains the handshake response.
298 RELAY: Payload contains the relay header and relay body.
299 DESTROY: Payload contains a reason for closing the circuit.
301 Upon receiving any other value for the command field, an OR must
302 drop the cell. Since more cell types may be added in the future, ORs
303 should generally not warn when encountering unrecognized commands.
305 The payload is padded with 0 bytes.
307 PADDING cells are currently used to implement connection keepalive.
308 If there is no other traffic, ORs and OPs send one another a PADDING
309 cell every few minutes.
311 CREATE, CREATED, and DESTROY cells are used to manage circuits;
314 RELAY cells are used to send commands and data along a circuit; see
317 VERSIONS and NETINFO cells are used to set up connections. See section 4
320 4. Negotiating and initializing connections
322 4.1. Negotiating versions with VERSIONS cells
324 There are multiple instances of the Tor link connection protocol. Any
325 connection negotiated using the "certificates up front" handshake (see
326 section 2 above) is "version 1". In any connection where both parties
327 have behaved as in the "renegotiation" handshake, the link protocol
328 version is 2 or higher.
330 To determine the version, in any connection where the "renegotiation"
331 handshake was used (that is, where the server sent only one certificate
332 at first and where the client did not send any certificates until
333 renegotiation), both parties MUST send a VERSIONS cell immediately after
334 the renegotiation is finished, before any other cells are sent. Parties
335 MUST NOT send any other cells on a connection until they have received a
338 The payload in a VERSIONS cell is a series of big-endian two-byte
339 integers. Both parties MUST select as the link protocol version the
340 highest number contained both in the VERSIONS cell they sent and in the
341 versions cell they received. If they have no such version in common,
342 they cannot communicate and MUST close the connection.
344 Since the version 1 link protocol does not use the "renegotiation"
345 handshake, implementations MUST NOT list version 1 in their VERSIONS
350 If version 2 or higher is negotiated, each party sends the other a
351 NETINFO cell. The cell's payload is:
354 Other OR's address [variable]
355 Number of addresses [1 byte]
356 This OR's addresses [variable]
358 The address format is a type/length/value sequence as given in section
359 6.4 below. The timestamp is a big-endian unsigned integer number of
360 seconds since the unix epoch.
362 Implementations MAY use the timestamp value to help decide if their
363 clocks are skewed. Initiators MAY use "other OR's address" to help
364 learn which address their connections are originating from, if they do
365 not know it. Initiators SHOULD use "this OR's address" to make sure
366 that they have connected to another OR at its canonical address.
368 [As of 0.2.0.23-rc, implementations use none of the above values.]
371 5. Circuit management
373 5.1. CREATE and CREATED cells
375 Users set up circuits incrementally, one hop at a time. To create a
376 new circuit, OPs send a CREATE cell to the first node, with the
377 first half of the DH handshake; that node responds with a CREATED
378 cell with the second half of the DH handshake plus the first 20 bytes
379 of derivative key data (see section 5.2). To extend a circuit past
380 the first hop, the OP sends an EXTEND relay cell (see section 5)
381 which instructs the last node in the circuit to send a CREATE cell
382 to extend the circuit.
384 The payload for a CREATE cell is an 'onion skin', which consists
385 of the first step of the DH handshake data (also known as g^x).
386 This value is hybrid-encrypted (see 0.3) to Bob's onion key, giving
389 Padding [PK_PAD_LEN bytes]
390 Symmetric key [KEY_LEN bytes]
391 First part of g^x [PK_ENC_LEN-PK_PAD_LEN-KEY_LEN bytes]
392 Symmetrically encrypted:
393 Second part of g^x [DH_LEN-(PK_ENC_LEN-PK_PAD_LEN-KEY_LEN)
396 The relay payload for an EXTEND relay cell consists of:
399 Onion skin [DH_LEN+KEY_LEN+PK_PAD_LEN bytes]
400 Identity fingerprint [HASH_LEN bytes]
402 The port and address field denote the IPV4 address and port of the next
403 onion router in the circuit; the public key hash is the hash of the PKCS#1
404 ASN1 encoding of the next onion router's identity (signing) key. (See 0.3
405 above.) Including this hash allows the extending OR verify that it is
406 indeed connected to the correct target OR, and prevents certain
407 man-in-the-middle attacks.
409 The payload for a CREATED cell, or the relay payload for an
410 EXTENDED cell, contains:
411 DH data (g^y) [DH_LEN bytes]
412 Derivative key data (KH) [HASH_LEN bytes] <see 5.2 below>
414 The CircID for a CREATE cell is an arbitrarily chosen 2-byte integer,
415 selected by the node (OP or OR) that sends the CREATE cell. To prevent
416 CircID collisions, when one node sends a CREATE cell to another, it chooses
417 from only one half of the possible values based on the ORs' public
418 identity keys: if the sending node has a lower key, it chooses a CircID with
419 an MSB of 0; otherwise, it chooses a CircID with an MSB of 1.
421 (An OP with no public key MAY choose any CircID it wishes, since an OP
422 never needs to process a CREATE cell.)
424 Public keys are compared numerically by modulus.
426 As usual with DH, x and y MUST be generated randomly.
428 5.1.1. CREATE_FAST/CREATED_FAST cells
430 When initializing the first hop of a circuit, the OP has already
431 established the OR's identity and negotiated a secret key using TLS.
432 Because of this, it is not always necessary for the OP to perform the
433 public key operations to create a circuit. In this case, the
434 OP MAY send a CREATE_FAST cell instead of a CREATE cell for the first
435 hop only. The OR responds with a CREATED_FAST cell, and the circuit is
438 A CREATE_FAST cell contains:
440 Key material (X) [HASH_LEN bytes]
442 A CREATED_FAST cell contains:
444 Key material (Y) [HASH_LEN bytes]
445 Derivative key data [HASH_LEN bytes] (See 5.2 below)
447 The values of X and Y must be generated randomly.
449 If an OR sees a circuit created with CREATE_FAST, the OR is sure to be the
450 first hop of a circuit. ORs SHOULD reject attempts to create streams with
451 RELAY_BEGIN exiting the circuit at the first hop: letting Tor be used as a
452 single hop proxy makes exit nodes a more attractive target for compromise.
454 5.2. Setting circuit keys
456 Once the handshake between the OP and an OR is completed, both can
457 now calculate g^xy with ordinary DH. Before computing g^xy, both client
458 and server MUST verify that the received g^x or g^y value is not degenerate;
459 that is, it must be strictly greater than 1 and strictly less than p-1
460 where p is the DH modulus. Implementations MUST NOT complete a handshake
461 with degenerate keys. Implementations MUST NOT discard other "weak"
464 (Discarding degenerate keys is critical for security; if bad keys
465 are not discarded, an attacker can substitute the server's CREATED
466 cell's g^y with 0 or 1, thus creating a known g^xy and impersonating
467 the server. Discarding other keys may allow attacks to learn bits of
470 If CREATE or EXTEND is used to extend a circuit, the client and server
471 base their key material on K0=g^xy, represented as a big-endian unsigned
474 If CREATE_FAST is used, the client and server base their key material on
477 From the base key material K0, they compute KEY_LEN*2+HASH_LEN*3 bytes of
478 derivative key data as
479 K = H(K0 | [00]) | H(K0 | [01]) | H(K0 | [02]) | ...
481 The first HASH_LEN bytes of K form KH; the next HASH_LEN form the forward
482 digest Df; the next HASH_LEN 41-60 form the backward digest Db; the next
483 KEY_LEN 61-76 form Kf, and the final KEY_LEN form Kb. Excess bytes from K
486 KH is used in the handshake response to demonstrate knowledge of the
487 computed shared key. Df is used to seed the integrity-checking hash
488 for the stream of data going from the OP to the OR, and Db seeds the
489 integrity-checking hash for the data stream from the OR to the OP. Kf
490 is used to encrypt the stream of data going from the OP to the OR, and
491 Kb is used to encrypt the stream of data going from the OR to the OP.
493 5.3. Creating circuits
495 When creating a circuit through the network, the circuit creator
496 (OP) performs the following steps:
498 1. Choose an onion router as an exit node (R_N), such that the onion
499 router's exit policy includes at least one pending stream that
500 needs a circuit (if there are any).
502 2. Choose a chain of (N-1) onion routers
503 (R_1...R_N-1) to constitute the path, such that no router
504 appears in the path twice.
506 3. If not already connected to the first router in the chain,
507 open a new connection to that router.
509 4. Choose a circID not already in use on the connection with the
510 first router in the chain; send a CREATE cell along the
511 connection, to be received by the first onion router.
513 5. Wait until a CREATED cell is received; finish the handshake
514 and extract the forward key Kf_1 and the backward key Kb_1.
516 6. For each subsequent onion router R (R_2 through R_N), extend
519 To extend the circuit by a single onion router R_M, the OP performs
522 1. Create an onion skin, encrypted to R_M's public onion key.
524 2. Send the onion skin in a relay EXTEND cell along
525 the circuit (see section 5).
527 3. When a relay EXTENDED cell is received, verify KH, and
528 calculate the shared keys. The circuit is now extended.
530 When an onion router receives an EXTEND relay cell, it sends a CREATE
531 cell to the next onion router, with the enclosed onion skin as its
532 payload. As special cases, if the extend cell includes a digest of
533 all zeroes, or asks to extend back to the relay that sent the extend
534 cell, the circuit will fail and be torn down. The initiating onion
535 router chooses some circID not yet used on the connection between the
536 two onion routers. (But see section 5.1. above, concerning choosing
537 circIDs based on lexicographic order of nicknames.)
539 When an onion router receives a CREATE cell, if it already has a
540 circuit on the given connection with the given circID, it drops the
541 cell. Otherwise, after receiving the CREATE cell, it completes the
542 DH handshake, and replies with a CREATED cell. Upon receiving a
543 CREATED cell, an onion router packs it payload into an EXTENDED relay
544 cell (see section 5), and sends that cell up the circuit. Upon
545 receiving the EXTENDED relay cell, the OP can retrieve g^y.
547 (As an optimization, OR implementations may delay processing onions
548 until a break in traffic allows time to do so without harming
549 network latency too greatly.)
551 5.3.1. Canonical connections
553 It is possible for an attacker to launch a man-in-the-middle attack
554 against a connection by telling OR Alice to extend to OR Bob at some
555 address X controlled by the attacker. The attacker cannot read the
556 encrypted traffic, but the attacker is now in a position to count all
557 bytes sent between Alice and Bob (assuming Alice was not already
560 To prevent this, when an OR we gets an extend request, it SHOULD use an
561 existing OR connection if the ID matches, and ANY of the following
563 - The IP matches the requested IP.
564 - The OR knows that the IP of the connection it's using is canonical
565 because it was listed in the NETINFO cell.
566 - The OR knows that the IP of the connection it's using is canonical
567 because it was listed in the server descriptor.
569 [This is not implemented in Tor 0.2.0.23-rc.]
571 5.4. Tearing down circuits
573 Circuits are torn down when an unrecoverable error occurs along
574 the circuit, or when all streams on a circuit are closed and the
575 circuit's intended lifetime is over. Circuits may be torn down
576 either completely or hop-by-hop.
578 To tear down a circuit completely, an OR or OP sends a DESTROY
579 cell to the adjacent nodes on that circuit, using the appropriate
582 Upon receiving an outgoing DESTROY cell, an OR frees resources
583 associated with the corresponding circuit. If it's not the end of
584 the circuit, it sends a DESTROY cell for that circuit to the next OR
585 in the circuit. If the node is the end of the circuit, then it tears
586 down any associated edge connections (see section 6.1).
588 After a DESTROY cell has been processed, an OR ignores all data or
589 destroy cells for the corresponding circuit.
591 To tear down part of a circuit, the OP may send a RELAY_TRUNCATE cell
592 signaling a given OR (Stream ID zero). That OR sends a DESTROY
593 cell to the next node in the circuit, and replies to the OP with a
594 RELAY_TRUNCATED cell.
596 When an unrecoverable error occurs along one connection in a
597 circuit, the nodes on either side of the connection should, if they
598 are able, act as follows: the node closer to the OP should send a
599 RELAY_TRUNCATED cell towards the OP; the node farther from the OP
600 should send a DESTROY cell down the circuit.
602 The payload of a RELAY_TRUNCATED or DESTROY cell contains a single octet,
603 describing why the circuit is being closed or truncated. When sending a
604 TRUNCATED or DESTROY cell because of another TRUNCATED or DESTROY cell,
605 the error code should be propagated. The origin of a circuit always sets
606 this error code to 0, to avoid leaking its version.
609 0 -- NONE (No reason given.)
610 1 -- PROTOCOL (Tor protocol violation.)
611 2 -- INTERNAL (Internal error.)
612 3 -- REQUESTED (A client sent a TRUNCATE command.)
613 4 -- HIBERNATING (Not currently operating; trying to save bandwidth.)
614 5 -- RESOURCELIMIT (Out of memory, sockets, or circuit IDs.)
615 6 -- CONNECTFAILED (Unable to reach server.)
616 7 -- OR_IDENTITY (Connected to server, but its OR identity was not
618 8 -- OR_CONN_CLOSED (The OR connection that was carrying this circuit
620 9 -- FINISHED (The circuit has expired for being dirty or old.)
621 10 -- TIMEOUT (Circuit construction took too long)
622 11 -- DESTROYED (The circuit was destroyed w/o client TRUNCATE)
623 12 -- NOSUCHSERVICE (Request for unknown hidden service)
625 5.5. Routing relay cells
627 When an OR receives a RELAY or RELAY_EARLY cell, it checks the cell's
628 circID and determines whether it has a corresponding circuit along that
629 connection. If not, the OR drops the cell.
631 Otherwise, if the OR is not at the OP edge of the circuit (that is,
632 either an 'exit node' or a non-edge node), it de/encrypts the payload
633 with the stream cipher, as follows:
634 'Forward' relay cell (same direction as CREATE):
635 Use Kf as key; decrypt.
636 'Back' relay cell (opposite direction from CREATE):
637 Use Kb as key; encrypt.
638 Note that in counter mode, decrypt and encrypt are the same operation.
640 The OR then decides whether it recognizes the relay cell, by
641 inspecting the payload as described in section 6.1 below. If the OR
642 recognizes the cell, it processes the contents of the relay cell.
643 Otherwise, it passes the decrypted relay cell along the circuit if
644 the circuit continues. If the OR at the end of the circuit
645 encounters an unrecognized relay cell, an error has occurred: the OR
646 sends a DESTROY cell to tear down the circuit.
648 When a relay cell arrives at an OP, the OP decrypts the payload
649 with the stream cipher as follows:
650 OP receives data cell:
652 Decrypt with Kb_I. If the payload is recognized (see
653 section 6..1), then stop and process the payload.
655 For more information, see section 6 below.
657 5.6. Handling relay_early cells
659 A RELAY_EARLY cell is designed to limit the length any circuit can reach.
660 When an OR receives a RELAY_EARLY cell, and the next node in the circuit
661 is speaking v2 of the link protocol or later, the OR relays the cell as a
662 RELAY_EARLY cell. Otherwise, it relays it as a RELAY cell.
664 If a node ever receives more than 8 RELAY_EARLY cells on a given circuit,
665 it SHOULD close the circuit.
667 When speaking v2 of the link protocol or later, clients MUST only send
668 EXTEND cells inside RELAY_EARLY cells. Clients SHOULD send the first ~8
669 RELAY cells that are not targeted at the first hop of any circuit as
670 RELAY_EARLY cells too, in order to partially conceal the circuit length.
672 [In a future version of Tor, servers will reject any EXTEND cell not
673 received in a RELAY_EARLY cell. See proposal 110.]
675 6. Application connections and stream management
679 Within a circuit, the OP and the exit node use the contents of
680 RELAY packets to tunnel end-to-end commands and TCP connections
681 ("Streams") across circuits. End-to-end commands can be initiated
682 by either edge; streams are initiated by the OP.
684 The payload of each unencrypted RELAY cell consists of:
685 Relay command [1 byte]
686 'Recognized' [2 bytes]
690 Data [CELL_LEN-14 bytes]
692 The relay commands are:
693 1 -- RELAY_BEGIN [forward]
694 2 -- RELAY_DATA [forward or backward]
695 3 -- RELAY_END [forward or backward]
696 4 -- RELAY_CONNECTED [backward]
697 5 -- RELAY_SENDME [forward or backward] [sometimes control]
698 6 -- RELAY_EXTEND [forward] [control]
699 7 -- RELAY_EXTENDED [backward] [control]
700 8 -- RELAY_TRUNCATE [forward] [control]
701 9 -- RELAY_TRUNCATED [backward] [control]
702 10 -- RELAY_DROP [forward or backward] [control]
703 11 -- RELAY_RESOLVE [forward]
704 12 -- RELAY_RESOLVED [backward]
705 13 -- RELAY_BEGIN_DIR [forward]
707 32..40 -- Used for hidden services; see rend-spec.txt.
709 Commands labelled as "forward" must only be sent by the originator
710 of the circuit. Commands labelled as "backward" must only be sent by
711 other nodes in the circuit back to the originator. Commands marked
712 as either can be sent either by the originator or other nodes.
714 The 'recognized' field in any unencrypted relay payload is always set
715 to zero; the 'digest' field is computed as the first four bytes of
716 the running digest of all the bytes that have been destined for
717 this hop of the circuit or originated from this hop of the circuit,
718 seeded from Df or Db respectively (obtained in section 5.2 above),
719 and including this RELAY cell's entire payload (taken with the digest
722 When the 'recognized' field of a RELAY cell is zero, and the digest
723 is correct, the cell is considered "recognized" for the purposes of
724 decryption (see section 5.5 above).
726 (The digest does not include any bytes from relay cells that do
727 not start or end at this hop of the circuit. That is, it does not
728 include forwarded data. Therefore if 'recognized' is zero but the
729 digest does not match, the running digest at that node should
730 not be updated, and the cell should be forwarded on.)
732 All RELAY cells pertaining to the same tunneled stream have the
733 same stream ID. StreamIDs are chosen arbitrarily by the OP. RELAY
734 cells that affect the entire circuit rather than a particular
735 stream use a StreamID of zero -- they are marked in the table above
736 as "[control]" style cells. (Sendme cells are marked as "sometimes
737 control" because they can take include a StreamID or not depending
738 on their purpose -- see Section 7.)
740 The 'Length' field of a relay cell contains the number of bytes in
741 the relay payload which contain real payload data. The remainder of
742 the payload is padded with NUL bytes.
744 If the RELAY cell is recognized but the relay command is not
745 understood, the cell must be dropped and ignored. Its contents
746 still count with respect to the digests, though.
748 6.2. Opening streams and transferring data
750 To open a new anonymized TCP connection, the OP chooses an open
751 circuit to an exit that may be able to connect to the destination
752 address, selects an arbitrary StreamID not yet used on that circuit,
753 and constructs a RELAY_BEGIN cell with a payload encoding the address
754 and port of the destination host. The payload format is:
756 ADDRESS | ':' | PORT | [00]
758 where ADDRESS can be a DNS hostname, or an IPv4 address in
759 dotted-quad format, or an IPv6 address surrounded by square brackets;
760 and where PORT is a decimal integer between 1 and 65535, inclusive.
762 [What is the [00] for? -NM]
763 [It's so the payload is easy to parse out with string funcs -RD]
765 Upon receiving this cell, the exit node resolves the address as
766 necessary, and opens a new TCP connection to the target port. If the
767 address cannot be resolved, or a connection can't be established, the
768 exit node replies with a RELAY_END cell. (See 6.4 below.)
769 Otherwise, the exit node replies with a RELAY_CONNECTED cell, whose
770 payload is in one of the following formats:
771 The IPv4 address to which the connection was made [4 octets]
772 A number of seconds (TTL) for which the address may be cached [4 octets]
774 Four zero-valued octets [4 octets]
775 An address type (6) [1 octet]
776 The IPv6 address to which the connection was made [16 octets]
777 A number of seconds (TTL) for which the address may be cached [4 octets]
778 [XXXX No version of Tor currently generates the IPv6 format.]
780 [Tor servers before 0.1.2.0 set the TTL field to a fixed value. Later
781 versions set the TTL to the last value seen from a DNS server, and expire
782 their own cached entries after a fixed interval. This prevents certain
785 The OP waits for a RELAY_CONNECTED cell before sending any data.
786 Once a connection has been established, the OP and exit node
787 package stream data in RELAY_DATA cells, and upon receiving such
788 cells, echo their contents to the corresponding TCP stream.
789 RELAY_DATA cells sent to unrecognized streams are dropped.
791 Relay RELAY_DROP cells are long-range dummies; upon receiving such
792 a cell, the OR or OP must drop it.
794 6.2.1. Opening a directory stream
796 If a Tor server is a directory server, it should respond to a
797 RELAY_BEGIN_DIR cell as if it had received a BEGIN cell requesting a
798 connection to its directory port. RELAY_BEGIN_DIR cells ignore exit
799 policy, since the stream is local to the Tor process.
801 If the Tor server is not running a directory service, it should respond
802 with a REASON_NOTDIRECTORY RELAY_END cell.
804 Clients MUST generate an all-zero payload for RELAY_BEGIN_DIR cells,
805 and servers MUST ignore the payload.
807 [RELAY_BEGIN_DIR was not supported before Tor 0.1.2.2-alpha; clients
808 SHOULD NOT send it to routers running earlier versions of Tor.]
812 When an anonymized TCP connection is closed, or an edge node
813 encounters error on any stream, it sends a 'RELAY_END' cell along the
814 circuit (if possible) and closes the TCP connection immediately. If
815 an edge node receives a 'RELAY_END' cell for any stream, it closes
816 the TCP connection completely, and sends nothing more along the
817 circuit for that stream.
819 The payload of a RELAY_END cell begins with a single 'reason' byte to
820 describe why the stream is closing, plus optional data (depending on
821 the reason.) The values are:
823 1 -- REASON_MISC (catch-all for unlisted reasons)
824 2 -- REASON_RESOLVEFAILED (couldn't look up hostname)
825 3 -- REASON_CONNECTREFUSED (remote host refused connection) [*]
826 4 -- REASON_EXITPOLICY (OR refuses to connect to host or port)
827 5 -- REASON_DESTROY (Circuit is being destroyed)
828 6 -- REASON_DONE (Anonymized TCP connection was closed)
829 7 -- REASON_TIMEOUT (Connection timed out, or OR timed out
831 8 -- (unallocated) [**]
832 9 -- REASON_HIBERNATING (OR is temporarily hibernating)
833 10 -- REASON_INTERNAL (Internal error at the OR)
834 11 -- REASON_RESOURCELIMIT (OR has no resources to fulfill request)
835 12 -- REASON_CONNRESET (Connection was unexpectedly reset)
836 13 -- REASON_TORPROTOCOL (Sent when closing connection because of
837 Tor protocol violations.)
838 14 -- REASON_NOTDIRECTORY (Client sent RELAY_BEGIN_DIR to a
839 non-directory server.)
841 (With REASON_EXITPOLICY, the 4-byte IPv4 address or 16-byte IPv6 address
842 forms the optional data, along with a 4-byte TTL; no other reason
843 currently has extra data.)
845 OPs and ORs MUST accept reasons not on the above list, since future
846 versions of Tor may provide more fine-grained reasons.
848 Tors SHOULD NOT send any reason except REASON_MISC for a stream that they
851 [*] Older versions of Tor also send this reason when connections are
853 [**] Due to a bug in versions of Tor through 0095, error reason 8 must
854 remain allocated until that version is obsolete.
856 --- [The rest of this section describes unimplemented functionality.]
858 Because TCP connections can be half-open, we follow an equivalent
859 to TCP's FIN/FIN-ACK/ACK protocol to close streams.
861 An exit connection can have a TCP stream in one of three states:
862 'OPEN', 'DONE_PACKAGING', and 'DONE_DELIVERING'. For the purposes
863 of modeling transitions, we treat 'CLOSED' as a fourth state,
864 although connections in this state are not, in fact, tracked by the
867 A stream begins in the 'OPEN' state. Upon receiving a 'FIN' from
868 the corresponding TCP connection, the edge node sends a 'RELAY_FIN'
869 cell along the circuit and changes its state to 'DONE_PACKAGING'.
870 Upon receiving a 'RELAY_FIN' cell, an edge node sends a 'FIN' to
871 the corresponding TCP connection (e.g., by calling
872 shutdown(SHUT_WR)) and changing its state to 'DONE_DELIVERING'.
874 When a stream in already in 'DONE_DELIVERING' receives a 'FIN', it
875 also sends a 'RELAY_FIN' along the circuit, and changes its state
876 to 'CLOSED'. When a stream already in 'DONE_PACKAGING' receives a
877 'RELAY_FIN' cell, it sends a 'FIN' and changes its state to
880 If an edge node encounters an error on any stream, it sends a
881 'RELAY_END' cell (if possible) and closes the stream immediately.
883 6.4. Remote hostname lookup
885 To find the address associated with a hostname, the OP sends a
886 RELAY_RESOLVE cell containing the hostname to be resolved with a nul
887 terminating byte. (For a reverse lookup, the OP sends a RELAY_RESOLVE
888 cell containing an in-addr.arpa address.) The OR replies with a
889 RELAY_RESOLVED cell containing a status byte, and any number of
890 answers. Each answer is of the form:
893 Value (variable-width)
895 "Length" is the length of the Value field.
900 0xF0 -- Error, transient
901 0xF1 -- Error, nontransient
903 If any answer has a type of 'Error', then no other answer may be given.
905 The RELAY_RESOLVE cell must use a nonzero, distinct streamID; the
906 corresponding RELAY_RESOLVED cell must use the same streamID. No stream
907 is actually created by the OR when resolving the name.
913 Each client or relay should do appropriate bandwidth throttling to
916 Communicants rely on TCP's default flow control to push back when they
919 The mainline Tor implementation uses token buckets (one for reads,
920 one for writes) for the rate limiting.
922 Since 0.2.0.x, Tor has let the user specify an additional pair of
923 token buckets for "relayed" traffic, so people can deploy a Tor relay
924 with strict rate limiting, but also use the same Tor as a client. To
925 avoid partitioning concerns we combine both classes of traffic over a
926 given OR connection, and keep track of the last time we read or wrote
927 a high-priority (non-relayed) cell. If it's been less than N seconds
928 (currently N=30), we give the whole connection high priority, else we
929 give the whole connection low priority. We also give low priority
930 to reads and writes for connections that are serving directory
931 information. See proposal 111 for details.
935 Link padding can be created by sending PADDING cells along the
936 connection; relay cells of type "DROP" can be used for long-range
939 Currently nodes are not required to do any sort of link padding or
940 dummy traffic. Because strong attacks exist even with link padding,
941 and because link padding greatly increases the bandwidth requirements
942 for running a node, we plan to leave out link padding until this
943 tradeoff is better understood.
945 7.3. Circuit-level flow control
947 To control a circuit's bandwidth usage, each OR keeps track of two
948 'windows', consisting of how many RELAY_DATA cells it is allowed to
949 originate (package for transmission), and how many RELAY_DATA cells
950 it is willing to consume (receive for local streams). These limits
951 do not apply to cells that the OR receives from one host and relays
954 Each 'window' value is initially set to 1000 data cells
955 in each direction (cells that are not data cells do not affect
956 the window). When an OR is willing to deliver more cells, it sends a
957 RELAY_SENDME cell towards the OP, with Stream ID zero. When an OR
958 receives a RELAY_SENDME cell with stream ID zero, it increments its
961 Each of these cells increments the corresponding window by 100.
963 The OP behaves identically, except that it must track a packaging
964 window and a delivery window for every OR in the circuit.
966 An OR or OP sends cells to increment its delivery window when the
967 corresponding window value falls under some threshold (900).
969 If a packaging window reaches 0, the OR or OP stops reading from
970 TCP connections for all streams on the corresponding circuit, and
971 sends no more RELAY_DATA cells until receiving a RELAY_SENDME cell.
972 [this stuff is badly worded; copy in the tor-design section -RD]
974 7.4. Stream-level flow control
976 Edge nodes use RELAY_SENDME cells to implement end-to-end flow
977 control for individual connections across circuits. Similarly to
978 circuit-level flow control, edge nodes begin with a window of cells
979 (500) per stream, and increment the window by a fixed value (50)
980 upon receiving a RELAY_SENDME cell. Edge nodes initiate RELAY_SENDME
981 cells when both a) the window is <= 450, and b) there are less than
982 ten cell payloads remaining to be flushed at that edge.
984 A.1. Differences between spec and implementation
986 - The current specification requires all ORs to have IPv4 addresses, but
987 allows servers to exit and resolve to IPv6 addresses, and to declare IPv6
988 addresses in their exit policies. The current codebase has no IPv6