TODO 16 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393
  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 0.0.9.6:
  13. - Server instructions for OSX and Windows operators.
  14. - Audit all changes to bandwidth buckets for integer over/underflow.
  15. For 0.1.0.x:
  16. Refactoring and infrastructure:
  17. N . Switch to libevent
  18. - Hold-open-until-flushed now works by accident; it should work by
  19. design.
  20. - The logic for reading from TLS sockets is likely to overrun the
  21. bandwidth buckets under heavy load. (Really, the logic was
  22. never right in the first place.) Also, we should audit all users
  23. of get_pending_bytes().
  24. . Find a way to make sure we have libevent 1.0 or later.
  25. o Implement patch to libevent
  26. o Submit patch to niels making this possible.
  27. - Implement Tor side once patch is accepted.
  28. . Log which poll method we're using.
  29. o Implement patch to libevent
  30. o Submit patch to niels making this possible.
  31. - Implement Tor side once patch is accepted.
  32. . Intercept libevent's "log" messages.
  33. o Ask Niels whether a patch would be accepted.
  34. o Implement patch, if so.
  35. - Implement Tor side once patch is accepted.
  36. o Check return from event_set, event_add, event_del.
  37. - Keep pushing to get a windows patch accepted.
  38. Security:
  39. - Make sure logged info is "safe"ish.
  40. Stability
  41. R o Reset uptime when IP changes.
  42. Functionality
  43. N . Implement pending controller features.
  44. o Stubs for new functions.
  45. o GETINFO
  46. o Version
  47. o Descriptor list
  48. o Individual descriptors
  49. o Need to remember descriptors for all routers.
  50. o Replace everything else that remembers serverdescs with
  51. routerinfo.
  52. o List of address mappings
  53. o POSTDESCRIPTOR
  54. o MAPADDRESS
  55. o Map A->B.
  56. o Map DontCare->B.
  57. o Reuse mappings when asked to map DontCare->B for the same B.
  58. o But only when the DontCare is of the same type. :/
  59. o Way to handle overlong messages
  60. o Specify fragmented format
  61. o Implement fragmented format
  62. o Event for "new descriptors"
  63. o Better stream IDs
  64. o Stream status changed: "new" state.
  65. - EXTENDCIRCUIT
  66. R - revised circ selection stuff.
  67. - Implement controller interface.
  68. - ATTACHSTREAM
  69. - Make streams have an 'unattached and not-automatically-attachable'
  70. state. ("Controller managed.")
  71. - Add support to put new streams into this state rather than try to
  72. attach them automatically. ("Hidden" config option.)
  73. - Time out never-attached streams.
  74. - Implement 'attach stream X to circuit Y' logic.
  75. - Tests for new controller features
  76. R . HTTPS proxy for OR CONNECT stuff. (For outgoing SSL connections to
  77. other ORs.)
  78. o Changes for forward compatibility
  79. o If a version is later than the last in its series, but a version
  80. in the next series is recommended, that doesn't mean it's bad.
  81. o Do end reasons better
  82. o Start using RESOURCELIMIT more.
  83. o Try to use MISC a lot less.
  84. o bug: if the exit node fails to create a socket (e.g. because it
  85. has too many open), we will get a generic stream end response.
  86. o Fix on platforms with set_max_file_descriptors.
  87. o niels's "did it fail because conn refused or timeout or what"
  88. relay end feature.
  89. o Realize that unrecognized end reasons are probably features rather than
  90. bugs. (backport to 009x)
  91. o Push the work of sending the end cell deeper into package_raw_inbuf.
  92. (Turns out, if package_raw_inbuf fails, it *can't* send an end cell.)
  93. o Check for any place where we can close an edge connection without
  94. sending an end; see if we should send an end.
  95. N . Feed end reason back into SOCK5 as reasonable.
  96. R o cache .foo.exit names better, or differently, or not.
  97. N - make !advertised_server_mode() ORs fetch dirs less often.
  98. N - Clean up NT service code even more. Document it. Enable it by default.
  99. Make sure it works.
  100. Documentation
  101. N - Document new version system.
  102. r - Correct and clarify the wiki entry on port forwarding.
  103. N - Document where OSX, windows logs go, where stuff is installed.
  104. Installers
  105. N - Vet all pending installer patches
  106. - Win32 installer plus privoxy, sockscap/freecap, etc.
  107. - Vet win32 systray helper code
  108. N - Make OSX man pages go into man directory.
  109. - Make logs do into platform default locations.
  110. Correctness
  111. - Mark bugs for 010 or post 010 in bugtracker.
  112. - Bugfixes
  113. R - when we haven't explicitly sent a socks reject, sending one in
  114. connection_about_to_close_connection() fails because we never give it
  115. a chance to flush. right answer is to do the socks reply manually in
  116. each appropriate case, and then about-to-close-connection can simply
  117. warn us if we forgot one. [Tag this 010 in flyspray.]
  118. R - should retry exitpolicy end streams even if the end cell didn't
  119. resolve the address for you
  120. o Figure out when to reset addressmaps (on hup, on reconfig, etc)
  121. Improvements to self-measurement.
  122. R X round detected bandwidth up to nearest 10KB?
  123. R o client software not upload descriptor until:
  124. . it decides it is reachable
  125. o dirport
  126. . orport
  127. - rule for now: "If you process a CREATE cell that did not come from
  128. your own IP, you are reachable."
  129. o start counting again if your IP ever changes.
  130. o never regenerate identity keys, for now.
  131. o you can set a bit for not-being-an-OR.
  132. * no need to do this yet. few people define their ORPort.
  133. Arguable
  134. N . Reverse DNS
  135. o specify
  136. - implement
  137. r - make min uptime a function of the available choices (say, choose 60th
  138. percentile, not 1 day.)
  139. r - kill dns workers more slowly
  140. r - build testing circuits? going through non-verified nodes?
  141. - config option to publish what ports you listen on, beyond ORPort/DirPort
  142. N - It would be nice to have a FirewalledIPs thing that works like
  143. FirewallPorts.
  144. - If we have a trusted directory on port 80, stop falling back to
  145. forbidden ports when fascistfirewall blocks all good dirservers.
  146. N - Code cleanup
  147. - Make configure.in handle cross-compilation
  148. - Have NULL_REP_IS_ZERO_BYTES default to 1.
  149. - Make with-ssl-dir disable search for ssl.
  150. - Efficiency/speed improvements.
  151. - Write limiting; configurable token buckets.
  152. - Make it harder to circumvent bandwidth caps: look at number of bytes
  153. sent across sockets, not number sent inside TLS stream.
  154. - Hidden service improvements
  155. - Investigate hidden service performance/reliability
  156. No
  157. - choose entry node to be one you're already connected to?
  158. - Convert man pages to pod, or whatever's right.
  159. - support hostnames as well as IPs for authdirservers.
  160. - GPSLocation optional config string.
  161. - Windows
  162. - Make millisecond accuracy work on win32
  163. - IPv6 support
  164. - teach connection_ap_handshake_socks_reply() about ipv6 and friends
  165. so connection_ap_handshake_socks_resolved() doesn't also need
  166. to know about them.
  167. - Let more config options (e.g. ORPort) change dynamically.
  168. - hidserv offerers shouldn't need to define a SocksPort
  169. * figure out what breaks for this, and do it.
  170. - Destroy and truncated cells should have reasons.
  171. - Packaging
  172. - Figure out how to make the rpm not strip the binaries it makes.
  173. - Integrate an http proxy into Tor (maybe as a third class of worker
  174. process), so we can stop shipping with the beast that is Privoxy.
  175. - Implement If-Modified-Since for directories.
  176. - Big, incompatible re-architecting and decentralization of directory
  177. system.
  178. - Only the top of a directory needs to be signed.
  179. - Windows
  180. - Get a controller to launch tor and keep it on the system tray.
  181. For 0.1.1.x:
  182. Decentralizing:
  183. - self-measurement
  184. - remote measurement
  185. - you've been running for an hour
  186. - it's sufficiently satisfied with its bandwidth
  187. - remove approval crap, add blacklisting by IP
  188. - gather more permanent dirservers and put their keys into the code
  189. - ship with a master key, and implement a way to query dirservers for
  190. a blob which is a timestamped signed newest pile of dirservers. put
  191. that on disk and use it on startup rather than the built-in default.
  192. - threshold belief from clients about up-ness
  193. - a way for clients to get fresh enough server descriptors
  194. - a way for clients to partition the set of servers in a safe way:
  195. so they don't have to learn all of them but so they're not easily
  196. partitionable.
  197. Tier two:
  198. N - Handle rendezvousing with unverified nodes.
  199. - Specify: Stick rendezvous point's key in INTRODUCE cell.
  200. Bob should _always_ use key from INTRODUCE cell.
  201. - Implement.
  202. N - IPv6 support (For exit addresses)
  203. - Spec issue: if a resolve returns an IP4 and an IP6 address,
  204. which to use?
  205. - Add to exit policy code
  206. - Make tor_gethostbyname into tor_getaddrinfo
  207. - Make everything that uses uint32_t as an IP address change to use
  208. a generalize address struct.
  209. - Change relay cell types to accept new addresses.
  210. - Add flag to serverdescs to tell whether IPv6 is supported.
  211. - Security fixes
  212. - christian grothoff's attack of infinite-length circuit.
  213. the solution is to have a separate 'extend-data' cell type
  214. which is used for the first N data cells, and only
  215. extend-data cells can be extend requests.
  216. - Code cleanup
  217. o fix router_get_by_* functions so they can get ourselves too ...
  218. - and audit everything to make sure rend and intro points are
  219. just as likely to be us as not.
  220. - tor should be able to have a pool of outgoing IP addresses
  221. that it is able to rotate through. (maybe)
  222. Packaging, docs, etc:
  223. - Exit node caching: tie into squid or other caching web proxy.
  224. Deferred until needed:
  225. - Do something to prevent spurious EXTEND cells from making middleman
  226. nodes connect all over. Rate-limit failed connections, perhaps?
  227. - Limit to 2 dir, 2 OR, N SOCKS connections per IP.
  228. - Handle full buffers without totally borking
  229. * do this eventually, no rush.
  230. - Rate-limit OR and directory connections overall and per-IP and
  231. maybe per subnet.
  232. - DoS protection: TLS puzzles, public key ops, bandwidth exhaustion.
  233. - Have clients and dirservers preserve reputation info over
  234. reboots.
  235. - authdirserver lists you as running iff:
  236. - he can connect to you
  237. - he has successfully extended to you
  238. - you have sufficient mean-time-between-failures
  239. * keep doing nothing for now.
  240. - Include HTTP status messages in logging (see parse_http_response).
  241. Blue sky or deferred indefinitely:
  242. - Support egd or other non-OS-integrated strong entropy sources
  243. - password protection for on-disk identity key
  244. - Possible to get autoconf to easily install things into ~/.tor?
  245. - server descriptor declares min log level, clients avoid servers
  246. that are too loggy.
  247. - put expiry date on onion-key, so people don't keep trying
  248. old ones that they could know are expired?
  249. - Add a notion of nickname->Pubkey binding that's not 'verification'
  250. - Conn key rotation.
  251. - Need a relay teardown cell, separate from one-way ends.
  252. Big tasks that would demonstrate progress:
  253. - Facility to automatically choose long-term helper nodes; perhaps
  254. on by default for hidden services.
  255. - patch privoxy and socks protocol to pass strings to the browser.
  256. - patch tsocks with our current patches + gethostbyname, getpeername, etc.
  257. - make freecap (or whichever) do what we want.
  258. - scrubbing proxies for protocols other than http.
  259. - Find an smtp proxy?
  260. . Get socks4a support into Mozilla
  261. - figure out enclaves, e.g. so we know what to recommend that people
  262. do, and so running a tor server on your website is helpful.
  263. - Do enclaves for same IP only.
  264. - Resolve first, then if IP is an OR, extend to him first.
  265. - implement a trivial fun gui to demonstrate our control interface.
  266. ************************ Roadmap for 2004-2005 **********************
  267. Hard problems that need to be solved:
  268. - Separating node discovery from routing.
  269. - Arranging membership management for independence.
  270. Sybil defenses without having a human bottleneck.
  271. How to gather random sample of nodes.
  272. How to handle nodelist recommendations.
  273. Consider incremental switches: a p2p tor with only 50 users has
  274. different anonymity properties than one with 10k users, and should
  275. be treated differently.
  276. - Measuring performance of other nodes. Measuring whether they're up.
  277. - Choosing exit node by meta-data, e.g. country.
  278. - Incentives to relay; incentives to exit.
  279. - Allowing dissidents to relay through Tor clients.
  280. - How to intercept, or not need to intercept, dns queries locally.
  281. - Improved anonymity:
  282. - Experiment with mid-latency systems. How do they impact usability,
  283. how do they impact safety?
  284. - Understand how powerful fingerprinting attacks are, and experiment
  285. with ways to foil them (long-range padding?).
  286. - Come up with practical approximations to picking entry and exit in
  287. different routing zones.
  288. - Find ideal churn rate for helper nodes; how safe is it?
  289. - What info squeaks by Privoxy? Are other scrubbers better?
  290. - Attacking freenet-gnunet/timing-delay-randomness-arguments.
  291. - Is abandoning the circuit the only option when an extend fails, or
  292. can we do something without impacting anonymity too much?
  293. - Is exiting from the middle of the circuit always a bad idea?
  294. Sample Publicity Landmarks:
  295. - we have N servers / N users
  296. - we have servers at epic and aclu and foo
  297. - hidden services are robust and fast
  298. - a more decentralized design
  299. - tor win32 installer works
  300. - win32 tray icon for end-users
  301. - tor server works on win32
  302. - win32 service for servers
  303. - mac installer works
  304. ***************************Future tasks:****************************
  305. Rendezvous and hidden services:
  306. make it fast:
  307. o preemptively build and start rendezvous circs.
  308. o preemptively build n-1 hops of intro circs?
  309. o cannibalize general circs?
  310. make it reliable:
  311. - standby/hotswap/redundant services.
  312. - store stuff to disk? dirservers forget service descriptors when
  313. they restart; nodes offering hidden services forget their chosen
  314. intro points when they restart.
  315. make it robust:
  316. - auth mechanisms to let midpoint and bob selectively choose
  317. connection requests.
  318. make it scalable:
  319. - robust decentralized storage for hidden service descriptors.
  320. make it accessible:
  321. - web proxy gateways to let normal people browse hidden services.
  322. Tor scalability:
  323. Relax clique assumptions.
  324. Redesign how directories are handled.
  325. - Resolve directory agreement somehow.
  326. Find and remove bottlenecks
  327. - Address linear searches on e.g. circuit and connection lists.
  328. Reputation/memory system, so dirservers can measure people,
  329. and so other people can verify their measurements.
  330. - Need to measure via relay, so it's not distinguishable.
  331. Let dissidents get to Tor servers via Tor users. ("Backbone model")
  332. Make it more correct:
  333. Handle half-open connections: right now we don't support all TCP
  334. streams, at least according to the protocol. But we handle all that
  335. we've seen in the wild.
  336. Support IPv6.
  337. Efficiency/speed/robustness:
  338. Congestion control. Is our current design sufficient once we have heavy
  339. use? Need to measure and tweak, or maybe overhaul.
  340. Allow small cells and large cells on the same network?
  341. Cell buffering and resending. This will allow us to handle broken
  342. circuits as long as the endpoints don't break, plus will allow
  343. connection (tls session key) rotation.
  344. Implement Morphmix, so we can compare its behavior, complexity, etc.
  345. Use cpuworker for more heavy lifting.
  346. - Signing (and verifying) hidserv descriptors
  347. - Signing (and verifying) intro/rend requests
  348. - Signing (and verifying) router descriptors
  349. - Signing (and verifying) directories
  350. - Doing TLS handshake (this is very hard to separate out, though)
  351. Buffer size pool: allocate a maximum size for all buffers, not
  352. a maximum size for each buffer. So we don't have to give up as
  353. quickly (and kill the thickpipe!) when there's congestion.
  354. Other transport. HTTP, udp, rdp, airhook, etc. May have to do our own
  355. link crypto, unless we can bully openssl into it.