TODO 15 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333
  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. 0.0.9pre4:
  13. o Don't use FascistFirewall if you're going via Tor, or if
  14. you're going via HttpProxy.
  15. o make RecommendedVersions a CONFIG_TYPE_LINELIST option
  16. o node 'groups' that are known to be in the same zone of control.
  17. o Nodes can list their coadministrated nodes.
  18. o If A lists B, it only counts if B also lists A
  19. o Users can list other coadministrated nodes if they like.
  20. o Never choose two coadministrated nodes in the same circuit.
  21. o let tor servers use proxies for port 80 exits
  22. o Use generic port redirector for IP/bits:Port->IP:Port .
  23. o Make use of them when we're doing exit connections.
  24. X We should set things in options to NULL, not rely on memset(...0)
  25. being equivalent.
  26. o We should check for memset(0) setting things to NULL with
  27. autoconf, and then rely on it in the code.
  28. 0.0.9pre5/6:
  29. - per-month byte allowances.
  30. N - Based on bandwidth and per-month allowance, choose a
  31. window within month to be up. Stay up until allowance is
  32. used. Adjust next month's window based on outcome. Hibernate
  33. when we're not up.
  34. R - Hibernate means "stop accepting connections, and start sleeping"
  35. Implement hibernation. Have a separate
  36. about-to-start-hibernating state implemented in similar way to
  37. will shut-down-in-30-seconds state.
  38. - Rendezvous service bug: can we nail it down?
  39. R . bandwidth buckets for write as well as read.
  40. N - Make watchdogged clients check cached-directory mtime to avoid
  41. fetching directory in a tight loop.
  42. - Implement If-Modified-Since for directories.
  43. N - Handle rendezvousing with unverified nodes.
  44. - Specify: Stick rendezvous point's key in INTRODUCE cell.
  45. Bob should _always_ use key from INTRODUCE cell.
  46. - Implement.
  47. R - figure out enclaves, e.g. so we know what to recommend that people
  48. do, and so running a tor server on your website is helpful.
  49. - Do enclaves for same IP only.
  50. - Resolve first, then if IP is an OR, connect to next guy.
  51. N - Pure C tor_resolve
  52. N - the user interface interface
  53. - Skeleton only.
  54. - Implement parts along with trivial fun gui.
  55. N - add ipv6 support.
  56. - Spec issue: if a resolve returns an IP4 and an IP6 address,
  57. which to use?
  58. N&R - Update Spec
  59. R - learn from ben about his openssl-reinitialization-trick to
  60. rotate tls keys without making new connections.
  61. - (Roger grabs Ben next time he sees him on IRC)
  62. - christian grothoff's attack of infinite-length circuit.
  63. the solution is to have a separate 'extend-data' cell type
  64. which is used for the first N data cells, and only
  65. extend-data cells can be extend requests.
  66. - have a pool of circuits available, cannibalize them
  67. for your purposes (e.g. rendezvous, etc).
  68. D nt services on win32.
  69. - Once we have a trusted directory on port 80, stop falling back to
  70. forbidden ports when fascistfirewall blocks all good dirservers.
  71. 0.0.9 and beyond:
  72. - fix sprintf's to snprintf's?
  73. . Make intro points and rendezvous points accept $KEYID in addition
  74. to nicknames.
  75. o Specify
  76. o Implement parsing
  77. - Generate new formats (Not till 007 is dead)
  78. - make loglevel info less noisy
  79. - Facility to automatically choose long-term helper nodes; perhaps
  80. on by default for hidden services.
  81. - Make command-line strict about checking options; make only certain
  82. option prefixes work.
  83. - Split running-routers into 2 lines: verified-routers and
  84. live-routers.
  85. - Rate-limit OR and directory connections overall and per-IP and
  86. maybe per subnet.
  87. - put expiry date on onion-key, so people don't keep trying
  88. old ones that they could know are expired?
  89. * Leave on todo list, see if pre3 onion fixes helped enough.
  90. - should the running-routers list put unverified routers at the
  91. end?
  92. * Cosmetic, don't do it yet.
  93. - make advertised_server_mode() ORs fetch dirs more often.
  94. * not necessary yet.
  95. - Add a notion of nickname->Pubkey binding that's not 'verification'
  96. * eventually, only when needed
  97. - ORs use uniquer default nicknames
  98. * Don't worry about this for now
  99. - Handle full buffers without totally borking
  100. * do this eventually, no rush.
  101. - do resolve before trying to attach the stream
  102. * don't do this for now.
  103. - if destination IP is running a tor node, extend a circuit there
  104. before sending begin.
  105. * don't do this for now. figure out how enclaves work. but do enclaves soon.
  106. - Support egd or other non-OS-integrated strong entropy sources
  107. more features, complex:
  108. - password protection for on-disk identity key
  109. . Switch dirservers entries to config lines:
  110. o read in and parse each TrustedDir config line.
  111. o stop reading dirservers file.
  112. o add some default TrustedDir lines if none defined, or if
  113. no torrc.
  114. o remove notion of ->is_trusted_dir from the routerlist. that's
  115. no longer where you look.
  116. o clean up router parsing flow, since it's simpler now?
  117. o when checking signature on a directory, look it up in
  118. options.TrustedDirs, and make sure there's a descriptor
  119. with that nickname, whose key hashes to the fingerprint,
  120. and who correctly signed the directory.
  121. o when fetching a directory, if you want a trusted one,
  122. choose from the trusteddir list.
  123. o which means keeping track of which ones are "up"
  124. ? if you don't need a trusted one, choose from the routerinfo
  125. list if you have one, else from the trusteddir list.
  126. * roger will do the above
  127. - add a listener for a ui
  128. * nick chats with weasel
  129. - and a basic gui
  130. - Have clients and dirservers preserve reputation info over
  131. reboots.
  132. * continue not doing until we have something we need to preserve
  133. - round detected bandwidth up to nearest 10KB?
  134. - client software not upload descriptor until:
  135. - you've been running for an hour
  136. - it's sufficiently satisfied with its bandwidth
  137. - it decides it is reachable
  138. - start counting again if your IP ever changes.
  139. - never regenerate identity keys, for now.
  140. - you can set a bit for not-being-an-OR.
  141. * no need to do this yet. few people define their ORPort.
  142. - authdirserver lists you as running iff:
  143. - he can connect to you
  144. - he has successfully extended to you
  145. - you have sufficient mean-time-between-failures
  146. * keep doing nothing for now.
  147. blue sky:
  148. - Possible to get autoconf to easily install things into ~/.tor?
  149. ongoing:
  150. . rename/rearrange functions for what file they're in
  151. - generalize our transport: add transport.c in preparation for
  152. http, airhook, etc transport.
  153. NICK - investigate sctp for alternate transport.
  154. For September:
  155. NICK . Windows port
  156. o works as client
  157. - deal with pollhup / reached_eof on all platforms
  158. . robust as a client
  159. . works as server
  160. - can be configured
  161. - robust as a server
  162. . Usable as NT service
  163. - docs for building in win
  164. - installer, including all needed libs.
  165. - Docs
  166. . FAQ
  167. o overview of tor. how does it work, what's it do, pros and
  168. cons of using it, why should I use it, etc.
  169. - a howto tutorial with examples
  170. * put a stub on the wiki
  171. o tutorial: how to set up your own tor network
  172. - (need to not hardcode dirservers file in config.c)
  173. * this will be solved when we put dirservers in config lines
  174. - port forwarding howto for ipchains, etc
  175. * roger add to wiki of requests
  176. . correct, update, polish spec
  177. - document the exposed function api?
  178. o document what we mean by socks.
  179. NICK . packages
  180. . rpm
  181. * nick will look at the spec file
  182. - find a long-term rpm maintainer
  183. * roger will start guilting people
  184. - code
  185. - better warn/info messages
  186. o let tor do resolves.
  187. o extend socks4 to do resolves?
  188. o make script to ask tor for resolves
  189. - write howto for setting up tsocks, socat.
  190. - including on osx and win32
  191. - freecap handling
  192. - tsocks
  193. o gather patches, submit to maintainer
  194. * send him a reminder mail and see what's up.
  195. - intercept gethostbyname and others
  196. * add this to tsocks
  197. o do resolve via tor
  198. - redesign and thorough code revamp, with particular eye toward:
  199. - support half-open tcp connections
  200. - conn key rotation
  201. - other transports -- http, airhook
  202. - modular introduction mechanism
  203. - allow non-clique topology
  204. Other details and small and hard things:
  205. - tor should be able to have a pool of outgoing IP addresses
  206. that it is able to rotate through. (maybe)
  207. - tie into squid
  208. - hidserv offerers shouldn't need to define a SocksPort
  209. * figure out what breaks for this, and do it.
  210. - when the client fails to pick an intro point for a hidserv,
  211. it should refetch the hidserv desc.
  212. . should maybe make clients exit(1) when bad things happen?
  213. e.g. clock skew.
  214. - should retry exitpolicy end streams even if the end cell didn't
  215. resolve the address for you
  216. . Make logs handle it better when writing to them fails.
  217. o Dirserver shouldn't put you in running-routers list if you haven't
  218. uploaded a descriptor recently
  219. . Refactor: add own routerinfo to routerlist. Right now, only
  220. router_get_by_nickname knows about 'this router', as a hack to
  221. get circuit_launch_new to do the right thing.
  222. . Scrubbing proxies
  223. - Find an smtp proxy?
  224. . Get socks4a support into Mozilla
  225. - Need a relay teardown cell, separate from one-way ends.
  226. - Make it harder to circumvent bandwidth caps: look at number of bytes
  227. sent across sockets, not number sent inside TLS stream.
  228. - fix router_get_by_* functions so they can get ourselves too,
  229. and audit everything to make sure rend and intro points are
  230. just as likely to be us as not.
  231. ***************************Future tasks:****************************
  232. Rendezvous and hidden services:
  233. make it fast:
  234. - preemptively build and start rendezvous circs.
  235. - preemptively build n-1 hops of intro circs?
  236. - cannibalize general circs?
  237. make it reliable:
  238. - standby/hotswap/redundant services.
  239. - store stuff to disk? dirservers forget service descriptors when
  240. they restart; nodes offering hidden services forget their chosen
  241. intro points when they restart.
  242. make it robust:
  243. - auth mechanisms to let midpoint and bob selectively choose
  244. connection requests.
  245. make it scalable:
  246. - right now the hidserv store/lookup system is run by the dirservers;
  247. this won't scale.
  248. Tor scalability:
  249. Relax clique assumptions.
  250. Redesign how directories are handled.
  251. - Separate running-routers lookup from descriptor list lookup.
  252. - Resolve directory agreement somehow.
  253. - Cache directory on all servers.
  254. Find and remove bottlenecks
  255. - Address linear searches on e.g. circuit and connection lists.
  256. Reputation/memory system, so dirservers can measure people,
  257. and so other people can verify their measurements.
  258. - Need to measure via relay, so it's not distinguishable.
  259. Bandwidth-aware path selection. So people with T3's are picked
  260. more often than people with DSL.
  261. Reliability-aware node selection. So people who are stable are
  262. preferred for long-term circuits such as intro and rend circs,
  263. and general circs for irc, aim, ssh, etc.
  264. Let dissidents get to Tor servers via Tor users. ("Backbone model")
  265. Anonymity improvements:
  266. Is abandoning the circuit the only option when an extend fails, or
  267. can we do something without impacting anonymity too much?
  268. Is exiting from the middle of the circuit always a bad idea?
  269. Helper nodes. Decide how to use them to improve safety.
  270. DNS resolution: need to make tor support resolve requests. Need to write
  271. a script and an interface (including an extension to the socks
  272. protocol) so we can ask it to do resolve requests. Need to patch
  273. tsocks to intercept gethostbyname, else we'll continue leaking it.
  274. Improve path selection algorithms based on routing-zones paper. Be sure
  275. to start and end circuits in different ASs. Ideally, consider AS of
  276. source and destination -- maybe even enter and exit via nearby AS.
  277. Intermediate model, with some delays and mixing.
  278. Add defensive dropping regime?
  279. Make it more correct:
  280. Handle half-open connections: right now we don't support all TCP
  281. streams, at least according to the protocol. But we handle all that
  282. we've seen in the wild.
  283. Support IPv6.
  284. Efficiency/speed/robustness:
  285. Congestion control. Is our current design sufficient once we have heavy
  286. use? Need to measure and tweak, or maybe overhaul.
  287. Allow small cells and large cells on the same network?
  288. Cell buffering and resending. This will allow us to handle broken
  289. circuits as long as the endpoints don't break, plus will allow
  290. connection (tls session key) rotation.
  291. Implement Morphmix, so we can compare its behavior, complexity, etc.
  292. Use cpuworker for more heavy lifting.
  293. - Signing (and verifying) hidserv descriptors
  294. - Signing (and verifying) intro/rend requests
  295. - Signing (and verifying) router descriptors
  296. - Signing (and verifying) directories
  297. - Doing TLS handshake (this is very hard to separate out, though)
  298. Buffer size pool: allocate a maximum size for all buffers, not
  299. a maximum size for each buffer. So we don't have to give up as
  300. quickly (and kill the thickpipe!) when there's congestion.
  301. Exit node caching: tie into squid or other caching web proxy.
  302. Other transport. HTTP, udp, rdp, airhook, etc. May have to do our own
  303. link crypto, unless we can bully openssl into it.
  304. P2P Tor:
  305. Do all the scalability stuff above, first.
  306. Incentives to relay. Not so hard.
  307. Incentives to allow exit. Possibly quite hard.
  308. Sybil defenses without having a human bottleneck.
  309. How to gather random sample of nodes.
  310. How to handle nodelist recommendations.
  311. Consider incremental switches: a p2p tor with only 50 users has
  312. different anonymity properties than one with 10k users, and should
  313. be treated differently.