TODO 14 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347
  1. Legend:
  2. SPEC!! - Not specified
  3. SPEC - Spec not finalized
  4. NICK - nick claims
  5. ARMA - arma claims
  6. - Not done
  7. * Top priority
  8. . Partially done
  9. o Done
  10. D Deferred
  11. X Abandoned
  12. For 0.0.9:
  13. o Solve the MSVC nuisance where __FILE__ contains the full path.
  14. People are getting confused about why their errors are coming from
  15. C:\Documents and Settings\Nick Mathewson\My Documents\src\tor .
  16. N&R. bring tor-spec up to date
  17. N&R. make loglevels info,debug less noisy
  18. o OS X package (and bundle?)
  19. o Working RPMs
  20. o Get win32 servers working, or find out why it isn't happening now.
  21. o Why can't win32 find a cpuworker?
  22. For 0.0.9.3:
  23. - All tasks marked for 0093 in flyspray.
  24. o Backport performance improvement (stop calling getttimeofday for
  25. each cell)
  26. o Tor startup script should be installed by default on OSX.
  27. o Setup instructions for OSX.
  28. - Server instructions for OSX and Windows operators.
  29. ************************ For Post 0.0.9 *****************************
  30. - config option to publish what ports you listen on, beyond ORPort/DirPort
  31. - https proxy for OR CONNECT stuff
  32. - choose entry node to be one you're already connected to?
  33. Tier one:
  34. o Move to our new version system.
  35. - Changes for forward compatibility
  36. - If a version is later than the last in its series, but a version
  37. in the next series is recommended, that doesn't mean it's bad.
  38. - Bugfixes
  39. o fix dfc/weasel's intro point bug
  40. - when we haven't explicitly sent a socks reject, sending one in
  41. connection_about_to_close_connection() fails because we never give it
  42. a chance to flush. right answer is to do the socks reply manually in
  43. each appropriate case, and then about-to-close-connection can simply
  44. warn us if we forgot one.
  45. - Documentation
  46. - Convert man pages to pod, or whatever's right. Alternatively, find
  47. a man2html that actually works.
  48. o Macintosh HOWTO page.
  49. - Evangelism
  50. - Get more nodes running on 80 and 443.
  51. - Get epic, aclu, etc running nodes.
  52. - Dirservers and server descs: small, backward-compatible changes
  53. - support hostnames as well as IPs for authdirservers.
  54. - If we have a trusted directory on port 80, stop falling back to
  55. forbidden ports when fascistfirewall blocks all good dirservers.
  56. - GPSLocation optional config string.
  57. - SOCKS enhancements
  58. - niels's "did it fail because conn refused or timeout or what"
  59. relay end feature.
  60. - bug: if the exit node fails to create a socket (e.g. because it
  61. has too many open), we will get a generic stream end response.
  62. - Windows
  63. N - Make millisecond accuracy work on win32
  64. X Switch to WSA*Event code as a better poll replacement. Or maybe just
  65. do libevent?
  66. - Code cleanup
  67. X Make more configuration variables into CSVs.
  68. - Make configure.in handle cross-compilation
  69. - Have NULL_REP_IS_ZERO_BYTES default to 1.
  70. - Make with-ssl-dir disable search for ssl.
  71. - Support
  72. o Bug tracker.
  73. - Exit hostname support
  74. - cache .foo.exit names better, or differently, or not.
  75. - IPv6 support
  76. - teach connection_ap_handshake_socks_reply() about ipv6 and friends
  77. so connection_ap_handshake_socks_resolved() doesn't also need
  78. to know about them.
  79. - Packaging
  80. - Figure out how to make the rpm not strip the binaries it makes.
  81. Tier two:
  82. - Efficiency/speed improvements.
  83. o Handle pools of waiting circuits better.
  84. o Limit number of circuits that we preemptively generate based on past
  85. behavior; use same limits in circuit_expire_old_circuits().
  86. - Write limiting; configurable token buckets.
  87. - Make it harder to circumvent bandwidth caps: look at number of bytes
  88. sent across sockets, not number sent inside TLS stream.
  89. . Switch to libevent
  90. o Evaluate libevent
  91. o Convert socket handling
  92. o Convert signal handling
  93. o Convert timers
  94. o Update configure.in
  95. o Remove fakepoll
  96. - Hold-open-until-flushed now works by accident; it should work by
  97. design.
  98. - The logic for reading from TLS sockets is likely to overrun the
  99. bandwidth buckets under heavy load. (Really, the logic was
  100. never right in the first place.) Also, we should audit all users
  101. of get_pending_bytes().
  102. - Make sure it works on more platforms.
  103. - Find a way to make sure we have libevent 1.0 or later.
  104. - Check return from event_set, event_add, event_del.
  105. - Integrate an http proxy into Tor (maybe as a third class of worker
  106. process), so we can stop shipping with the beast that is Privoxy.
  107. - QOI
  108. - Let more config options (e.g. ORPort) change dynamically.
  109. - Dirservers and server descs: small, backward-compatible changes
  110. - make advertised_server_mode() ORs fetch dirs more often.
  111. - Implement If-Modified-Since for directories.
  112. - Big, incompatible re-architecting and decentralization of directory
  113. system.
  114. - Only the top of a directory needs to be signed.
  115. - Windows
  116. N - Clean up NT service code; make it work
  117. - Get a controller to launch tor and keep it on the system tray.
  118. - Win32 installer plus privoxy, sockscap/freecap, etc.
  119. - Controller enhancements.
  120. o Implement SIGNAL feature so windows can hup, shutdown, etc.
  121. - controller should have 'getinfo' command to query about rephist,
  122. about rendezvous status, etc.
  123. N - Handle rendezvousing with unverified nodes.
  124. - Specify: Stick rendezvous point's key in INTRODUCE cell.
  125. Bob should _always_ use key from INTRODUCE cell.
  126. - Implement.
  127. N - IPv6 support (For exit addresses)
  128. - Spec issue: if a resolve returns an IP4 and an IP6 address,
  129. which to use?
  130. - Add to exit policy code
  131. - Make tor_gethostbyname into tor_getaddrinfo
  132. - Make everything that uses uint32_t as an IP address change to use
  133. a generalize address struct.
  134. - Change relay cell types to accept new addresses.
  135. - Add flag to serverdescs to tell whether IPv6 is supported.
  136. - When should servers
  137. - Security fixes
  138. - christian grothoff's attack of infinite-length circuit.
  139. the solution is to have a separate 'extend-data' cell type
  140. which is used for the first N data cells, and only
  141. extend-data cells can be extend requests.
  142. - Make sure logged information is 'safe'.
  143. - Code cleanup
  144. . rename/rearrange functions for what file they're in
  145. - fix router_get_by_* functions so they can get ourselves too,
  146. and audit everything to make sure rend and intro points are
  147. just as likely to be us as not.
  148. - Bugfixes
  149. - hidserv offerers shouldn't need to define a SocksPort
  150. * figure out what breaks for this, and do it.
  151. - should retry exitpolicy end streams even if the end cell didn't
  152. resolve the address for you
  153. - tor should be able to have a pool of outgoing IP addresses
  154. that it is able to rotate through. (maybe)
  155. Packaging, docs, etc:
  156. - Exit node caching: tie into squid or other caching web proxy.
  157. Deferred until needed:
  158. - Do something to prevent spurious EXTEND cells from making middleman
  159. nodes connect all over. Rate-limit failed connections, perhaps?
  160. - Limit to 2 dir, 2 OR, N SOCKS connections per IP.
  161. - Handle full buffers without totally borking
  162. * do this eventually, no rush.
  163. - Rate-limit OR and directory connections overall and per-IP and
  164. maybe per subnet.
  165. - DoS protection: TLS puzzles, public key ops, bandwidth exhaustion.
  166. - Have clients and dirservers preserve reputation info over
  167. reboots.
  168. - round detected bandwidth up to nearest 10KB?
  169. - client software not upload descriptor until:
  170. - you've been running for an hour
  171. - it's sufficiently satisfied with its bandwidth
  172. - it decides it is reachable
  173. - start counting again if your IP ever changes.
  174. - never regenerate identity keys, for now.
  175. - you can set a bit for not-being-an-OR.
  176. * no need to do this yet. few people define their ORPort.
  177. - authdirserver lists you as running iff:
  178. - he can connect to you
  179. - he has successfully extended to you
  180. - you have sufficient mean-time-between-failures
  181. * keep doing nothing for now.
  182. - Include HTTP status messages in logging (see parse_http_response).
  183. Blue sky or deferred indefinitely:
  184. - Support egd or other non-OS-integrated strong entropy sources
  185. - password protection for on-disk identity key
  186. - Possible to get autoconf to easily install things into ~/.tor?
  187. - server descriptor declares min log level, clients avoid servers
  188. that are too loggy.
  189. - put expiry date on onion-key, so people don't keep trying
  190. old ones that they could know are expired?
  191. - Add a notion of nickname->Pubkey binding that's not 'verification'
  192. - Conn key rotation.
  193. - Need a relay teardown cell, separate from one-way ends.
  194. Big tasks that would demonstrate progress:
  195. - Facility to automatically choose long-term helper nodes; perhaps
  196. on by default for hidden services.
  197. - patch privoxy and socks protocol to pass strings to the browser.
  198. - patch tsocks with our current patches + gethostbyname, getpeername, etc.
  199. - make freecap (or whichever) do what we want.
  200. - scrubbing proxies for protocols other than http.
  201. - Find an smtp proxy?
  202. . Get socks4a support into Mozilla
  203. N - Reverse DNS: specify and implement.
  204. - figure out enclaves, e.g. so we know what to recommend that people
  205. do, and so running a tor server on your website is helpful.
  206. - Do enclaves for same IP only.
  207. - Resolve first, then if IP is an OR, extend to him first.
  208. - implement a trivial fun gui to demonstrate our control interface.
  209. ************************ Roadmap for 2004-2005 **********************
  210. Hard problems that need to be solved:
  211. - Separating node discovery from routing.
  212. - Arranging membership management for independence.
  213. Sybil defenses without having a human bottleneck.
  214. How to gather random sample of nodes.
  215. How to handle nodelist recommendations.
  216. Consider incremental switches: a p2p tor with only 50 users has
  217. different anonymity properties than one with 10k users, and should
  218. be treated differently.
  219. - Measuring performance of other nodes. Measuring whether they're up.
  220. - Choosing exit node by meta-data, e.g. country.
  221. - Incentives to relay; incentives to exit.
  222. - Allowing dissidents to relay through Tor clients.
  223. - How to intercept, or not need to intercept, dns queries locally.
  224. - Improved anonymity:
  225. - Experiment with mid-latency systems. How do they impact usability,
  226. how do they impact safety?
  227. - Understand how powerful fingerprinting attacks are, and experiment
  228. with ways to foil them (long-range padding?).
  229. - Come up with practical approximations to picking entry and exit in
  230. different routing zones.
  231. - Find ideal churn rate for helper nodes; how safe is it?
  232. - What info squeaks by Privoxy? Are other scrubbers better?
  233. - Attacking freenet-gnunet/timing-delay-randomness-arguments.
  234. - Is abandoning the circuit the only option when an extend fails, or
  235. can we do something without impacting anonymity too much?
  236. - Is exiting from the middle of the circuit always a bad idea?
  237. Sample Publicity Landmarks:
  238. - we have N servers / N users
  239. - we have servers at epic and aclu and foo
  240. - hidden services are robust and fast
  241. - a more decentralized design
  242. - tor win32 installer works
  243. - win32 tray icon for end-users
  244. - tor server works on win32
  245. - win32 service for servers
  246. - mac installer works
  247. ***************************Future tasks:****************************
  248. Rendezvous and hidden services:
  249. make it fast:
  250. o preemptively build and start rendezvous circs.
  251. o preemptively build n-1 hops of intro circs?
  252. o cannibalize general circs?
  253. make it reliable:
  254. - standby/hotswap/redundant services.
  255. - store stuff to disk? dirservers forget service descriptors when
  256. they restart; nodes offering hidden services forget their chosen
  257. intro points when they restart.
  258. make it robust:
  259. - auth mechanisms to let midpoint and bob selectively choose
  260. connection requests.
  261. make it scalable:
  262. - robust decentralized storage for hidden service descriptors.
  263. make it accessible:
  264. - web proxy gateways to let normal people browse hidden services.
  265. Tor scalability:
  266. Relax clique assumptions.
  267. Redesign how directories are handled.
  268. - Resolve directory agreement somehow.
  269. Find and remove bottlenecks
  270. - Address linear searches on e.g. circuit and connection lists.
  271. Reputation/memory system, so dirservers can measure people,
  272. and so other people can verify their measurements.
  273. - Need to measure via relay, so it's not distinguishable.
  274. Let dissidents get to Tor servers via Tor users. ("Backbone model")
  275. Make it more correct:
  276. Handle half-open connections: right now we don't support all TCP
  277. streams, at least according to the protocol. But we handle all that
  278. we've seen in the wild.
  279. Support IPv6.
  280. Efficiency/speed/robustness:
  281. Congestion control. Is our current design sufficient once we have heavy
  282. use? Need to measure and tweak, or maybe overhaul.
  283. Allow small cells and large cells on the same network?
  284. Cell buffering and resending. This will allow us to handle broken
  285. circuits as long as the endpoints don't break, plus will allow
  286. connection (tls session key) rotation.
  287. Implement Morphmix, so we can compare its behavior, complexity, etc.
  288. Use cpuworker for more heavy lifting.
  289. - Signing (and verifying) hidserv descriptors
  290. - Signing (and verifying) intro/rend requests
  291. - Signing (and verifying) router descriptors
  292. - Signing (and verifying) directories
  293. - Doing TLS handshake (this is very hard to separate out, though)
  294. Buffer size pool: allocate a maximum size for all buffers, not
  295. a maximum size for each buffer. So we don't have to give up as
  296. quickly (and kill the thickpipe!) when there's congestion.
  297. Other transport. HTTP, udp, rdp, airhook, etc. May have to do our own
  298. link crypto, unless we can bully openssl into it.