3 Tor directory protocol for 0.1.1.x series
5 0. Scope and preliminaries
7 This document should eventually be merged into tor-spec.txt and replace
8 the existing notes on directories.
10 This is not a finalized version; what we actually wind up implementing
11 may be very different from the system described here.
15 There are several problems with the way Tor handles directories right
17 1. Directories are very large and use a lot of bandwidth.
18 2. Every directory server is a single point of failure.
19 3. Requiring every client to know every server won't scale.
20 4. Requiring every directory cache to know every server won't scale.
21 5. Our current "verified server" system is kind of nonsensical.
22 6. Getting more directory servers adds more points of failure and
23 worsens possible partitioning attacks.
25 This design tries to solve every problem except problems 3 and 4, and to
26 be compatible with likely eventual solutions to problems 3 and 4.
30 There is no longer any such thing as a "signed directory". Instead,
31 directory servers sign a very compressed 'network status' object that
32 lists the current descriptors and their status, and router descriptors
33 continue to be self-signed by servers. Clients download network status
34 listings periodically, and download router descriptors as needed. ORs
35 upload descriptors relatively infrequently.
37 There are multiple directory servers. Rather than doing anything
38 complicated to coordinate themselves, clients simply rotate through them
39 in order, and only use servers that most of the last several directory
44 The router descriptor format is unchanged from tor-spec.txt.
46 ORs SHOULD generate a new router descriptor whenever any of the
47 following events have occurred:
49 - A period of time (18 hrs by default) has passed since the last
50 time a descriptor was generated.
52 - A descriptor field other than bandwidth or uptime has changed.
54 - Bandwidth has changed by more than +/- 50% from the last time a
55 descriptor was generated, and at least a given interval of time
56 (20 mins by default) has passed since then.
58 - Uptime has been reset.
60 After generating a descriptor, ORs upload it to every directory
65 Directory servers generate, sign, and compress a network-status document
66 as needed. As an optimization, they may rate-limit the number of such
67 documents generated to once every few seconds. Directory servers should
68 rate-limit at least to the point where these documents are generated no
69 faster than once per second.
71 The network status document contains a preamble, a set of router status
72 entries, and a signature, in that order.
74 We use the same meta-format as used for directories and router descriptors
77 The preamble contains:
79 "network-status-version" -- A document format version. For this
80 specification, the version is "2".
81 "dir-source" -- The hostname, current IP address, and directory
82 port of the directory server, separated by spaces.
83 "fingerprint" -- A base16-encoded hash of the signing key's
84 fingerprint, with no additional spaces added.
85 "contact" -- An arbitrary string describing how to contact the
86 directory server's administrator. Administrators should include at
87 least an email address and a PGP fingerprint.
88 "dir-signing-key" -- The directory server's public signing key.
89 "client-versions" -- A comma-separated list of recommended client versions.
90 "server-versions" -- A comma-separated list of recommended server versions.
91 "published" -- The publication time for this network-status object.
92 "dir-options" -- A set of flags separated by spaces:
93 "Names" if this directory server performs name bindings.
94 "Versions" if this directory server recommends software versions.
96 The dir-options entry is optional. The "-versions" entries are required if
97 the "Versions" flag is present. The other entries are required and must
98 appear exactly once. The "network-status-version" entry must appear first;
99 the others may appear in any order.
101 For each router, the router entry contains: (This format is designed for
104 "r" -- followed by the following elements, separated by spaces:
106 - A hash of its identity key, encoded in base64, with trailing =
108 - A hash of its most recent descriptor, encoded in base64, with
109 trailing = signs removed. (The hash is calculated as for
110 computing the signature of a descriptor.)
111 - The publication time of its most recent descriptor.
114 - A directory port (or "0" for none")
115 "s" -- A series of space-separated status flags:
116 "Exit" if the router is useful for building general-purpose exit
118 "Stable" if the router tends to stay up for a long time.
119 "Fast" if the router has high bandwidth.
120 "Running" if the router is currently usable.
121 "Named" if the router's identity-nickname mapping is canonical.
122 "Valid" if the router has been 'validated'.
123 "Authority" if the router is a directory authority.
125 The "r" entry for each router must appear first and is required. The
126 's" entry is optional. Unrecognized flags, or extra elements on the
127 "r" line must be ignored.
129 The signature section contains:
131 "directory-signature". A signature of the rest of the document using
132 the directory server's signing key.
134 We compress the network status list with zlib before transmitting it.
136 4. Directory server operation
138 By default, directory servers remember all non-expired, non-superseded OR
139 descriptors that they have seen.
141 For each OR, a directory server remembers whether the OR was running and
142 functional the last time they tried to connect to it, and possibly other
143 liveness information.
145 Directory server administrators may label some servers or IPs as
146 blacklisted, and elect not to include them in their network-status lists.
148 Thus, the network-status list includes all non-blacklisted,
149 non-expired, non-superseded descriptors for ORs that the directory has
150 observed at least once to be running.
152 Directory server administrators may decide to support name binding. If
153 they do, then they must maintain a file of nickname-to-identity-key
154 mappings, and try to keep this file consistent with other directory
155 servers. If they don't, they act as clients, and report bindings made by
156 other directory servers (name X is bound to identity Y if at least one
157 binding directory lists it, and no directory binds X to some other Y'.)
159 The authoritative network-status published by a host should be available at:
160 http://<hostname>/tor/status/authority.z
162 An authoritative network-status published by another host with fingerprint
163 <F> should be available at:
164 http://<hostname>/tor/status/fp/<F>.z
166 An authoritative network-status published by other hosts with fingerprints
167 <F1>,<F2>,<F3> should be available at:
168 http://<hostname>/tor/status/fp/<F1>+<F2>+<F3>.z
170 The most recent network-status documents from all known authoritative
171 directories, concatenated, should be available at:
172 http://<hostname>/tor/status/all.z
174 The most recent descriptor for a server whose identity key has a
175 fingerprint of <F> should be available at:
176 http://<hostname>/tor/server/fp/<F>.z
178 The most recent descriptors for servers with fingerprints <F1>,<F2>,<F3>
179 should be available at:
180 http://<hostname>/tor/server/fp/<F1>+<F2>+<F3>.z
182 The descriptor for a server whose digest (in hex) is <D> should be
184 http://<hostname>/tor/server/d/<D>.z
186 The most recent descriptors with digests <D1>,<D2>,<D3> should be
188 http://<hostname>/tor/server/d/<D1>+<D2>+<D3>.z
190 The most recent descriptor for this server should be at:
191 http://<hostname>/tor/server/authority.z
193 A concatenated set of the most recent descriptors for all known servers
194 should be available at:
195 http://<hostname>/tor/server/all.z
197 For debugging, directories MAY expose non-compressed objects at URLs like
198 the above, but without the final ".z".
200 Clients MUST handle compressed concatenated information in two forms:
201 - A concatenated list of zlib-compressed objects.
202 - A zlib-compressed concatenated list of objects.
203 Directory servers MAY generate either format: the former requires less
204 CPU, but the latter requires less bandwidth.
208 Directory caches (most ORs) regularly download network status documents,
209 and republish them at a URL based on the directory server's identity key:
210 http://<hostname>/tor/status/<identity fingerprint>.z
212 A concatenated list of all network-status documents should be available at:
213 http://<hostname>/tor/status/all.z
220 Every OP or OR, including directory servers, acts as a client to the
223 Each client maintains a list of trusted directory servers. Periodically
224 (currently every 20 minutes), the client downloads a new network status. It
225 chooses the directory server from which its current information is most
226 out-of-date, and retries on failure until it finds a running server.
228 When choosing ORs to build circuits, clients proceed as follows:
229 - A server is "listed" if it is listed by more than half of the "live"
230 network status documents the clients have downloaded. (A network
231 status is "live" if it is the most recently downloaded network status
232 document for a given directory server, and the server is a directory
233 server trusted by the client, and the network-status document is no
234 more than D (say, 10) days old.)
235 - A server is "valid" is it is listed as valid by more than half of the
236 "live" downloaded" network-status document.
237 - A server is "running" if it is listed as running by more than
238 half of the "recent" downloaded network-status documents.
239 (A network status is "recent" if it was published in the last
240 60 minutes. If there are fewer than 3 such documents, the most
241 recently published 3 are "recent." If there are fewer than 3 in all,
245 Clients store network status documents so long as they are live.
247 5.1. Scheduling network status downloads
249 This download scheduling algorithm implements the approach described above
250 in a relatively low-state fashion. It reflects the current Tor
253 Clients maintain a list of authorities; each client tries to keep the same
254 list, in the same order.
256 Periodically, on startup, and on HUP, clients check whether they need to
257 download fresh network status documents. The approach is as follows:
258 - If we have under X network status documents newer than OLD, we choose a
259 member of the list at random and try download XX documents starting
261 - Otherwise, if we have no network status documents newer than NEW, we
262 check to see which authority's document we retrieved most recently,
263 and try to retrieve the next authority's document. If we can't, we
264 try the next authority in sequence, and so on.
268 In order to provide human-memorable names for individual server
269 identities, some directory servers bind names to IDs. Clients handle
272 If a client is encountering a name it has not mapped before:
274 If all the "binding" networks-status documents the client has so far
275 received same claim that the name binds to some identity X, and the
276 client has received at least three network-status documents, the client
279 If a client is encountering a name it has mapped before:
281 It uses the last-mapped identity value, unless all of the "binding"
282 network status documents bind the name to some other identity.
284 5.3. Notes on what we do now.
286 THIS SECTION SHOULD BE FOLDED INTO THE EARLIER SECTIONS; THEY ARE WRONG;
289 All downloaded networkstatuses are discarded once they are 10 days old (by
292 Authdirs download each others' networkstatus every
293 AUTHORITY_NS_CACHE_INTERVAL minutes (currently 10).
295 Directory caches download authorities' networkstatus every
296 NONAUTHORITY_NS_CACHE_INTERVAL minutes (currently 10).
298 Clients always try to replace any networkstatus received over
299 NETWORKSTATUS_MAX_VALIDITY ago (currently 2 days). Also, when the most
300 recently received networkstatus is more than
301 NETWORKSTATUS_CLIENT_DL_INTERVAL (30 minutes) old, and we do not have any
302 open directory connections fetching a networkstatus, clients try to
303 download the networkstatus on their list after the most recently received
304 networkstatus, skipping failed networkstatuses. A networkstatus is
305 "failed" if NETWORKSTATUS_N_ALLOWABLE_FAILURES (3) attempts in a row have
308 We do not update router statuses if we have less than half of the
311 A networkstatus is "live" if it is the most recent we have received signed
312 by a given trusted authority.
314 A networkstatus is "recent" if it is "live" and:
315 - it was received in the last DEFAULT_RUNNING_INTERVAL (currently 60
317 OR - it was one of the MIN_TO_INFLUENCE_RUNNING (3) most recently received
320 Authorities always believe their own opinion as to a router's status. For
322 - a router is valid if more than half of the live networkstatuses think
324 - a router is named if more than half of the live networkstatuses from
325 naming authorities think it's named, and they all think it has the
327 - a router is running if more than half of the recent networkstatuses
330 Everyone downloads router descriptors as follows:
332 - If any networkstatus lists a more recently published routerdesc with a
333 different descriptor digest, and no more than
334 MAX_ROUTERDESC_DOWNLOAD_FAILURES attempts to retrieve that routerdesc
335 have failed, then that routerdesc is "downloadable".
337 - Every DirFetchInterval, or whenever a request for routerdescs returns
338 no routerdescs, we launch a set of requests for all downloadable
339 routerdescs. We divide the downloadable routerdescs into groups of no
340 more than DL_PER_REQUEST, and send a request for each group to
341 directory servers chosen independently.
343 - We also launch a request as above when a request for routerdescs
344 fails and we have no directory connections fetching routerdescs.
347 - When to 0-out failure count for networkstatus?
349 - Drop fallback to download-all. Also, always split download.
351 - For versions: if you're listed by more than half of live versioning
352 networkstatuses, good. if less than half of networkstatuses are live,
353 don't do anything. If half are live, and half of less of the
354 versioning ones list you, warn. Only warn once every 24 hours.
356 - For names: warn if an unnamed router is specified by nickname.
357 Rate-limit these warnings.
358 - Also, don't believe N->K if another naming authdir says N->K'.
359 - Revise naming rule: N->K is true if any naming directory says N->K,
360 and no other naming directory says N->K' or N'->K.
362 - Minimum info to build circuits.
364 - Revise: always split requests when we have too little info to build
367 - Describe when router is "out of date". (Any dirserver says so.)
369 - Change rule from "do not launch new connections when one exists" to
370 "do not request any fingerprint that we're currently requesting."
372 - Launch new connections every minute, plus whenever a download fails.
373 - Reset routerdesc failure count after 60 minutes, or when
374 when network comes back on after absence.
375 - Make "I didn't get the one I thought was most recent" a failure.
376 - Retry these every 5 minutes if you're a client.
377 - Mirrors should retry these harder and more often.
378 - If we have a routerdesc for Bob, and he says, "I'm 0.1.0.x", don't
379 fetch a new one if it was published in the last 2 hours. (??)
381 - Describe what we do with old server versions.
383 - If we have less than 16 to download, do not download unless 10 minutes
384 have passed since last download.
386 - Which descriptors do directory servers remember?
390 Client-knowledge partitioning is worrisome. Most versions of this don't
391 seem to be worse than the Danezis-Murdoch tracing attack, since an
392 attacker can't do more than deduce probable exits from entries (or vice
393 versa). But what about when the client connects to A and B but in a
394 different order? How bad can it be partitioned based on its knowledge?