tor-parallel-relay-conn/doc/TODO

$Id$
Legend:
SPEC!!  - Not specified
SPEC    - Spec not finalized
N       - nick claims
R       - arma claims
P       - phobos claims
        - Not done
        * Top priority
        . Partially done
        o Done
        D Deferred
        X Abandoned

Non-Coding, Soon:
N - Mark up spec; note unclear points about servers
N - Clean up dir spec.
N - Mention controller libs someplace.
  D FAQ entry: why gnutls is bad/not good for tor
P - flesh out the rest of the section 6 of the faq
P - gather pointers to livecd distros that include tor
  - put the logo on the website, in source form, so people can put it on
    stickers directly, etc.
R . more pictures from ren. he wants to describe the tor handshake, i want to
    talk about hidden services.
  * clean up the places where our docs are redundant (or worse, obsolete in
    one file and correct elsewhere). agl has a start on a global
    list-of-tor-docs.
P - update windows docs to clarify which versions of windows, and why a
    DOS window, how it's used, for the less technical users
NR- write a spec appendix for 'being nice with tor'
  - tor-in-the-media page
  - Ask schanzle@cas.homelinux.org about a patch for rpm spec fixes against
    tor-0.1.0.7.rc
  - Remove need for HACKING file.


for 0.1.1.x:
N - if they're trying to be a tor server and they're running
    win 98 or win me, give them a message talking about The Bug.

  . Helper nodes
    . More testing and debugging
R   - If your helper nodes are unavailable, don't abandon them unless
      other nodes *are* reachable.

N - Destroy and truncated cells should have reasons.
    - Specify
    - Implement

N - Only use a routerdesc if you recognize its hash.
    o (Must defer till dirservers are upgraded to latest code, which
      actually generates these hashes.)
    - Of course, authdirservers must not do this.
    o If we have a routerdesc for Bob, and he says, "I'm 0.1.0.x", don't
      fetch a new one if it was published in the last 2 hours.
      - How does this interact with the 'recognized hash' rule?
    - Do not ask for any routers until we have 2 networkstatuses.
    - Client side:
      - Keep a record of which hash is most desirable for each router inside
        local_routerstatus_t.
        - If any hash is listed by two or more networkstatuses, the most
          recent such hash is most desirable.
        - Otherwise, the most recent is desirable.
      - Once we've accepted a router, it's okay.
      - Do not accept a router that no networkstatus lists. (This should maybe
        get stricter.)
      - Download by fingerprint.
      - Reset failure count to zero when hash changes.
    - Mirrors and authorities:
      - Every time we hear a new networkstatus, we want every hash it lists.
      - Make sure that we are always willing to keep at least N routerinfos
        per router, where N = number of authorities.
        - Do whatever else is needed to be sure that we don't request
          hashes that would be immediately discarded, or discard hashes
          that would be immediately re-requested.
      - Only fetch routerinfo from an authority that mentions is.
        - Only ask each authority once.
        - Retry soon after failure.
        - We need one bit per routerstatus for "should we download from
          this guy."

N - Should router info have a pointer to routerstatus?
    - We should at least do something about the duplicated fields.

R - Christian Grothoff's attack of infinite-length circuit.
    the solution is to have a separate 'extend-data' cell type
    which is used for the first N data cells, and only
    extend-data cells can be extend requests.
    - Specify, including thought about
    - Implement

R - When we connect to a Tor server, it sends back a signed cell listing
    the IP it believes it is using. Use this to block dvorak's attack.
    Also, this is a fine time to say what time you think it is.
    - Verify that a new cell type is okay with deployed codebase
    - Specify
    - Implement

R - clients prefer to avoid exit nodes for non-exit path positions.

  - find 10 dirservers.
    - What are criteria to be a dirserver?  Write a policy.



Deferred from 0.1.1.x:

  - Automatically determine what ports are reachable and start using
    those, if circuits aren't working and it's a pattern we recognize
    ("port 443 worked once and port 9001 keeps not working").

N . Additional controller features
      o Find a way to make event info more extensible
      - change circuit status events to give more details, like purpose,
        whether they're internal, when they become dirty, when they become
        too dirty for further circuits, etc.
R       - What do we want here, exactly?
N       - Specify and implement it.
      - Change stream status events analogously.
R       - What do we want here, exactly?
N       - Specify and implement it.
      - Make other events "better".
      - Change stream status events analogously.
R       - What do we want here, exactly?
N       - Specify and implement it.
      - Make other events "better" analogously
R       - What do we want here, exactly?
N       - Specify and implement it.
      . Expose more information via getinfo:
        - import and export rendezvous descriptors
        - Review all static fields for additional candidates
      - Allow EXTENDCIRCUIT to unknown server.
      - We need some way to adjust server status, and to tell tor not to
        download directories/network-status, and a way to force a download.
      - It would be nice to request address lookups from the controller
        without using SOCKS.
      - Make everything work with hidden services

  X switch accountingmax to count total in+out, not either in or
    out. it's easy to move in this direction (not risky), but hard to
    back out if we decide we prefer it the way it already is. hm.

  - cpu fixes:
    - see if we should make use of truncate to retry
    o hardware accelerator support (configure engines.)
    o hardware accelerator support (use instead of aes.c when reasonable)
      - Benchmark this somehow to see whether using EVP_foo is slower in the
        non-engine case than AES_foo.  If so, check for AES engine and fall
        back to AES_foo when it's not found.
R   - kill dns workers more slowly

  . Directory changes
    o recommended-versions for client / server ?
    . Some back-out mechanism for auto-approval
      o dirservers have blacklist of IPs and keys they hate
      - a way of rolling back approvals to before a timestamp
        - Consider minion-like fingerprint file/log combination.

    - Decentralization
      o Dirservers publish compressed network-status objects.
        o Support retrieving several-at-once
      o Everyone downloads network-status objects
        o Clients: from all directories, round-robin
          o Basic implementation: disable until 0.1.1.x is out.
          o On failure, mark trusted_dir_server as having failed
          o Retry, up to a point.
          X Launch retry immediately on failure.
        o Parse them
        o Cache them, reload on restart
        o Serve cached directories
      o Directories expose individual descriptors
        X By 'if-newer-than' (Does the spec require this??)
        o Support compression.
      o Alice acts on network-status objects
        o Alice downloads descriptors as needed.
          o Figure out what's needed
          o Store it
            o Implement store
            o Implement reload-from-store
            o Store downloaded descriptors
          o Download it
            o As-needed if we have 2 network-status objs.
            o Download "all" if we have less than 2 network-status objs.
              (This has vulnerabilities if we're not careful)
            o Call directory_has_arrived as needed; rename it.
            o Set has_fetched_directory properly.
          o Retry descriptors on failure
          o Give up after a while.
          - But try again after a long while (???)
        o Check software versions according to some sane plan.
          - Warn again after 24 hours.
        o Alice sets descriptor status from network-status
          o Implement
          o Use
      o Routerdesc download changes
        o Refactor combined-status to be its own type.
        o Change rule from "do not launch new connections when one exists" to
          "do not request any fingerprint that we're currently requesting."
        o Launch connections every minute, or whenever a download fails
        o Retry failed routerdescs after 0, 1, 5, 10 minutes.
          o Mirrors retry harder and more often. (0, 0, 1, 1, 2, 5, and 15)
        o Reset failure count every 60 minutes
        o Drop fallback to download-all.  Also, always split download.
        o Use has_fetched_directory sanely, whatever that means.
      o Downgrade new directory events from notice to info
      o Call dirport_is_reachable from somewhere else.
      o Networkstatus should list who's an authority.
      o Add nickname element to dirserver line.  Log this along with IP:Port.
      o Warn when using non-default directory servers.
      o When giving up on a non-finished dir request, log how many bytes
        dropped, to see whether it's worthwhile to use partial info.

    - config option to publish what ports you listen on, beyond
      ORPort/DirPort.  It should support ranges and bit prefixes (?) too.
      - Parse this.
      - Relay this in networkstatus.

    X Make authorities rate-limit logging their complaints about given
      servers?
    o All versions of Tor should get cosmetic changes rate-limited.
    o Pick directories from networkstatus objects, not from routerlist.
      o But! We can't do this easily, since we want to know about platform,
        and networkstatus doesn't tell us Tor version.  Can we solve this?
        Should we do it by adding flags to networkstatus or what?

  - packaging and ui stuff:
    . multiple sample torrc files
    - uninstallers
      . for os x
    . figure out how to make nt service stuff work?
      . Document it.
    . Add version number to directory.
N   - Vet all pending installer patches
      - Win32 installer plus privoxy, sockscap/freecap, etc.
      - Vet win32 systray helper code

  - document:
    - torcp needs more attention in the tor-doc-win32.
    - recommend gaim.
    - unrecommend IE because of ftp:// bug.
    - torrc.complete.in needs attention?

  o Dump "ports" from routerparse?

  o Let more config options (e.g. ORPort) change dynamically.
  o Add TTLs to DNS-related replies, and use them (when present) to adjust
    addressmap values.
  - Bind to random port when making outgoing connections to Tor servers,
    to reduce remote sniping attacks.
  - Have new people be in limbo and need to demonstrate usefulness
    before we approve them.
  - Clients should estimate their skew as median of skew from servers
    over last N seconds.
  - Security
    - Alices avoid duplicate class C nodes.
    - Analyze how bad the partitioning is or isn't.

  . Update the hidden service stuff for the new dir approach.
    - switch to an ascii format, maybe sexpr?
    - authdirservers publish blobs of them.
    - other authdirservers fetch these blobs.
    - hidserv people have the option of not uploading their blobs.
    - you can insert a blob via the controller.
    - and there's some amount of backwards compatibility.
    - teach clients, intro points, and hidservs about auth mechanisms.
    - come up with a few more auth mechanisms.

  . Come up with a coherent strategy for bandwidth buckets and TLS. (The
    logic for reading from TLS sockets is likely to overrun the bandwidth
    buckets under heavy load.  (Really, the logic was never right in the
    first place.)  Also, we should audit all users of get_pending_bytes().)
      - Make it harder to circumvent bandwidth caps: look at number of bytes
        sent across sockets, not number sent inside TLS stream.

  o Research memory use on Linux: what's happening?
    X Is it threading?  (Maybe, maybe not)
    X Is it the buf_shrink bug? (Quite possibly)
    o Instrument the 0.1.1 code to figure out where our memory is going;
      apply the results. (all platforms?)

  - Make router_is_general_exit() a bit smarter once we're sure what it's for.

  - Directory "helper".

  - rewrite how libevent does select() on win32 so it's not so very slow.
  o enclaves (at least preliminary)
  - Write limiting; separate token bucket for write
  - Audit everything to make sure rend and intro points are just as likely to
    be us as not.
  - Do something to prevent spurious EXTEND cells from making middleman
    nodes connect all over.  Rate-limit failed connections, perhaps?

Future version:
  - Limit to 2 dir, 2 OR, N SOCKS connections per IP.
  - Handle full buffers without totally borking
  - Rate-limit OR and directory connections overall and per-IP and
    maybe per subnet.
  - Hold-open-until-flushed now works by accident; it should work by
    design.
  - DoS protection: TLS puzzles, public key ops, bandwidth exhaustion.
    - Specify?
  - tor-resolve script should use socks5 to get better error messages.
  - make min uptime a function of the available choices (say, choose 60th
    percentile, not 1 day.)
  - Track uptime as %-of-time-up, as well as time-since-last-down.
  - hidserv offerers shouldn't need to define a SocksPort
    * figure out what breaks for this, and do it.
  - auth mechanisms to let hidden service midpoint and responder filter
    connection requests.
  - Relax clique assumptions.
  - start handling server descriptors without a socksport?
  - tor should be able to have a pool of outgoing IP addresses
    that it is able to rotate through. (maybe)

Blue-sky:
  - Patch privoxy and socks protocol to pass strings to the browser.
  - Standby/hotswap/redundant hidden services.
  - Robust decentralized storage for hidden service descriptors.
  - The "China problem"
  - Allow small cells and large cells on the same network?
  - Cell buffering and resending. This will allow us to handle broken
    circuits as long as the endpoints don't break, plus will allow
    connection (tls session key) rotation.
  - Implement Morphmix, so we can compare its behavior, complexity, etc.
  - Other transport. HTTP, udp, rdp, airhook, etc. May have to do our own
    link crypto, unless we can bully openssl into it.
  . Conn key rotation (we switch to a new one after a week, but
    old circuits don't get any benefit from this).
  - Need a relay teardown cell, separate from one-way ends.
    (Pending a user who needs this)
  - Handle half-open connections: right now we don't support all TCP
    streams, at least according to the protocol. But we handle all that
    we've seen in the wild.
    (Pending a user who needs this)