## translation metadata # Revision: $Revision: 16901 $ #include "head.wmi" TITLE="Tor: Volunteer"

A few things everyone can do now:

  1. Please consider running a relay to help the Tor network grow.
  2. Tell your friends! Get them to run relays. Get them to run hidden services. Get them to tell their friends.
  3. If you like Tor's goals, please take a moment to donate to support further Tor development. We're also looking for more sponsors — if you know any companies, NGOs, agencies, or other organizations that want anonymity / privacy / communications security, let them know about us.
  4. We're looking for more good examples of Tor users and Tor use cases. If you use Tor for a scenario or purpose not yet described on that page, and you're comfortable sharing it with us, we'd love to hear from you.

Supporting Applications

  1. We need more good ways to intercept DNS requests so they don't "leak" their request to a local observer while we're trying to be anonymous. (This happens because the application does the DNS resolve before going to the SOCKS proxy.)
  2. Tsocks/dsocks items:
    • We need to apply all our tsocks patches and maintain a new fork. We'll host it if you want.
    • We should patch Dug Song's "dsocks" program to use Tor's mapaddress commands from the controller interface, so we don't waste a whole round-trip inside Tor doing the resolve before connecting.
    • We need to make our torify script detect which of tsocks or dsocks is installed, and call them appropriately. This probably means unifying their interfaces, and might involve sharing code between them or discarding one entirely.
  3. People running relays tell us they want to have one BandwidthRate during some part of the day, and a different BandwidthRate at other parts of the day. Rather than coding this inside Tor, we should have a little script that speaks via the Tor Controller Interface, and does a setconf to change the bandwidth rate. There is one for Unix and Mac already (it uses bash and cron), but Windows users still need a solution.
  4. Tor can exit the Tor network from a particular exit node, but we should be able to specify just a country and have something automatically pick. The best bet is to fetch Blossom's directory also, and run a local Blossom client that fetches this directory securely (via Tor and checking its signature), intercepts .country.blossom hostnames, and does the right thing.
  5. Speaking of geolocation data, somebody should draw a map of the Earth with a pin-point for each Tor relay. Bonus points if it updates as the network grows and changes. Unfortunately, the easy ways to do this involve sending all the data to Google and having them draw the map for you. How much does this impact privacy, and do we have any other good options?

Documentation

  1. Please help Matt Edman with the documentation and how-tos for his Tor controller, Vidalia.
  2. Evaluate and document our list of programs that can be configured to use Tor.
  3. We need better documentation for dynamically intercepting connections and sending them through Tor. tsocks (Linux), dsocks (BSD), and freecap (Windows) seem to be good candidates, as would better use of our new TransPort feature.
  4. We have a huge list of potentially useful programs that interface to Tor. Which ones are useful in which situations? Please help us test them out and document your results.
  5. Help translate the web page and documentation into other languages. See the translation guidelines if you want to help out. We especially need Arabic or Farsi translations, for the many Tor users in censored areas.

Good Coding Projects

Here is a list of ideas that were proposed for the Google Summer of Code 2008 but have not been put into practice. Some of the current proposals might also be short on developers. If you think you can help out, let us know!

  1. Tor Node Scanner improvements
    Similar to the SoaT exit scanner (or perhaps even during exit scanning), statistics can be gathered about the reliability of nodes. Nodes that fail too high a percentage of their circuits should not be given Guard status. Perhaps they should have their reported bandwidth penalized by some ratio as well, or just get marked as Invalid. In addition, nodes that exhibit a very low average stream capacity but advertise a very high node bandwidth can also be marked as Invalid. Much of this statistics gathering is already done, it just needs to be transformed into something that can be reported to the Directory Authorities to blacklist/penalize nodes in such a way that clients will listen.
    In addition, these same statistics can be gathered about the traffic through a node. Events can be added to the Tor Control Protocol to report if a circuit extend attempt through the node succeeds or fails, and passive statistics can be gathered on both bandwidth and reliability of other nodes via a node-based monitor using these events. Such a scanner would also report information on oddly-behaving nodes to the Directory Authorities, but a communication channel for this currently does not exist and would need to be developed as well.
  2. Help track the overall Tor Network status
    It would be great to set up an automated system for tracking network health over time, graphing it, etc. Part of this project would involve inventing better metrics for assessing network health and growth. Is the average uptime of the network increasing? How many relays are qualifying for Guard status this month compared to last month? What's the turnover in terms of new relays showing up and relays shutting off? Periodically people collect brief snapshots, but where it gets really interesting is when we start tracking data points over time.
    Data could be collected from the "Tor Node Scanner" item above, from the server descriptors that each relay publishes, and from other sources. Results over time could be integrated into one of the Tor Status web pages, or be kept separate. Speaking of the Tor Status pages, take a look at Roger's Tor Status wish list.
  3. Better Debian/Ubuntu Packaging for Tor+Vidalia
    Vidalia currently doesn't play nicely on Debian and Ubuntu with the default Tor packages. The current Tor packages automatically start Tor as a daemon running as the debian-tor user and (sensibly) do not have a ControlPort defined in the default torrc. Consequently, Vidalia will try to start its own Tor process since it could not connect to the existing Tor, and Vidalia's Tor process will then exit with an error message the user likely doesn't understand since Tor cannot bind its listening ports — they're already in use by the original Tor daemon.
    The current solution involves either telling the user to stop the existing Tor daemon and let Vidalia start its own Tor process, or explaining to the user how to set a control port and password in their torrc. A better solution on Debian would be to use Tor's ControlSocket, which allows Vidalia to talk to Tor via a Unix domain socket, and could possibly be enabled by default in Tor's Debian packages. Vidalia can then authenticate to Tor using filesystem-based (cookie) authentication if the user running Vidalia is also in the debian-tor group.
    This project will first involve adding support for Tor's ControlSocket to Vidalia. The student will then develop and test Debian and Ubuntu packages for Vidalia that conform to Debian's packaging standards and make sure they work well with the existing Tor packages. We can also set up an apt repository to host the new Vidalia packages.
    The next challenge would be to find an intuitive usable way for Vidalia to be able to change Tor's configuration (torrc) even though it is located in /etc/tor/torrc and thus immutable. The best idea we've come up with so far is to feed Tor a new configuration via the ControlSocket when Vidalia starts, but that's bad because Tor starts each boot with a different configuration than the user wants. The second best idea we've come up with is for Vidalia to write out a temporary torrc file and ask the user to manually move it to /etc/tor/torrc, but that's bad because users shouldn't have to mess with files directly.
    A person undertaking this project should have prior knowledge of Debian package management and some C++ development experience. Previous experience with Qt is helpful, but not required.
  4. Improving Tor's ability to resist censorship
    The Tor 0.2.0.x series makes significant improvements in resisting national and organizational censorship. But Tor still needs better mechanisms for some parts of its anti-censorship design. For example, current Tors can only listen on a single address/port combination at a time. There's a proposal to address this limitation and allow clients to connect to any given Tor on multiple addresses and ports, but it needs more work. Another anti-censorship project (far more difficult) is to try to make Tor more scanning-resistant. Right now, an adversary can identify Tor bridges just by trying to connect to them, following the Tor protocol, and seeing if they respond. To solve this, bridges could act like webservers (HTTP or HTTPS) when contacted by port-scanning tools, and not act like bridges until the user provides a bridge-specific key.
    This project involves a lot of research and design. One of the big challenges will be identifying and crafting approaches that can still resist an adversary even after the adversary knows the design, and then trading off censorship resistance with usability and robustness.
  5. Tor/Polipo/Vidalia Auto-Update Framework
    We're in need of a good authenticated-update framework. Vidalia already has the ability to notice when the user is running an outdated or unrecommended version of Tor, using signed statements inside the Tor directory information. Currently, Vidalia simply pops up a little message box that lets the user know they should manually upgrade. The goal of this project would be to extend Vidalia with the ability to also fetch and install the updated Tor software for the user. We should do the fetches via Tor when possible, but also fall back to direct fetches in a smart way. Time permitting, we would also like to be able to update other applications included in the bundled installers, such as Polipo and Vidalia itself.
    To complete this project, the student will first need to first investigate the existing auto-update frameworks (e.g., Sparkle on OS X) to evaluate their strengths, weaknesses, security properties, and ability to be integrated into Vidalia. If none are found to be suitable, the student will design their own auto-update framework, document the design, and then discuss the design with other developers to assess any security issues. The student will then implement their framework (or integrate an existing one) and test it.
    A person undertaking this project should have good C++ development experience. Previous experience with Qt is helpful, but not required. One should also have a good understanding of common security practices, such as package signature verification. Good writing ability is also important for this project, since a vital step of the project will be producing a design document to review and discuss with others prior to implementation.
  6. An Improved and More Usable Network Map in Vidalia
    One of Vidalia's existing features is a network map that shows the user the approximate geographic location of relays in the Tor network and plots the paths the user's traffic takes as it is tunneled through the Tor network. The map is currently not very interactive and has rather poor graphics. Instead, we would like to leverage KDE's Marble widget that gives us a better quality map and enables improved interactivity, such as allowing the user to click on individual relays or circuits to display additional information. We might also consider adding the ability for users to click on a particular relay or a country containing one or more Tor exit relays and say, "I want my connections to foo.com to exit from here."
    This project will first involve getting familiar with Vidalia and the Marble widget's API. One will then integrate the widget into Vidalia and customize Marble to be better suited for our application, such as making circuits clickable, storing cached map data in Vidalia's own data directory, and customizing some of the widget's dialogs.
    A person undertaking this project should have good C++ development experience. Previous experience with Qt and CMake is helpful, but not required.
  7. Tor Controller Status Event Interface
    There are a number of status changes inside Tor of which the user may need to be informed. For example, if the user is trying to set up his Tor as a relay and Tor decides that its ports are not reachable from outside the user's network, we should alert the user. Currently, all the user gets is a couple log messages in Vidalia's 'message log' window, which they likely never see since they don't receive a notification that something has gone wrong. Even if the user does actually look at the message log, most of the messages make little sense to the novice user.
    Tor has the ability to inform Vidalia of many such status changes, and we recently implemented support for a couple of these events. Still, there are many more status events the user should be informed of and we need a better UI for actually displaying them to the user.
    The goal of this project then is to design and implement a UI for displaying Tor status events to the user. For example, we might put a little badge on Vidalia's tray icon that alerts the user to new status events they should look at. Double-clicking the icon could bring up a dialog that summarizes recent status events in simple terms and maybe suggests a remedy for any negative events if they can be corrected by the user. Of course, this is just an example and one is free to suggest another approach.
    A person undertaking this project should have good UI design and layout and some C++ development experience. Previous experience with Qt and Qt's Designer will be very helpful, but are not required. Some English writing ability will also be useful, since this project will likely involve writing small amounts of help documentation that should be understandable by non-technical users. Bonus points for some graphic design/Photoshop fu, since we might want/need some shiny new icons too.
  8. Improvements on our active browser configuration tester - https://check.torproject.org/
    We currently have a functional web page to detect if Tor is working. It has a few places where it falls short. It requires improvements with regard to default languages and functionality. It currently only responds in English. In addition, it is a hack of a perl script that should have never seen the light of day. It should probably be rewritten in python with multi-lingual support in mind. It currently uses the Tor DNS exit list and should continue to do so in the future. It currently result in certain false positives and these should be discovered, documented, and fixed where possible. Anyone working on this project should be interested in DNS, basic perl or preferably python programming skills, and will have to interact minimally with Tor to test their code.
    If you want to make the project more exciting and involve more design and coding, take a look at proposal 131-verify-tor-usage.txt.
  9. Improvements on our DNS Exit List service - http://exitlist.torproject.org/
    The exitlist software is written by our fabulous anonymous contributer Tup. It's a DNS server written in Haskell that supports part of our exitlist design document. Currently, it is functional and it is used by check.torproject.org and other users. The issues that are outstanding are mostly aesthetic. This wonderful service could use a much better website using the common Tor theme. It would be best served with better documentation for common services that use an RBL. It could use more publicity. A person working on this project should be interested in DNS, basic RBL configuration for popular services, and writing documentation. The person would require minimal Tor interaction — testing their own documentation at the very least. Furthermore, it would be useful if they were interested in Haskell and wanted to implement more of the torel-design.txt suggestions.
  10. Testing integration of Tor with web browsers for our end users
    The Tor project currently lacks a solid test suite to ensure that a user has a properly and safely configured web browser. It should test for as many known issues as possible. It should attempt to decloak the user in any way possible. Two current webpages that track these kinds of issues are run by Greg Fleischer and HD Moore. Greg keeps a nice list of issues along with their proof of concept code, bug issues, etc. HD Moore runs the metasploit decloak website. A person interested in defending Tor could start by collecting as many workable and known methods for decloaking a Tor user. (This page may be helpful as a start.) One should be familiar with the common pitfalls but possibly have new methods in mind for implementing decloaking issues. The website should ensure that it tells a user what their problem is. It should help them to fix the problem or direct them to the proper support channels. The person should also be closely familiar with using Tor and how to prevent Tor information leakage.
  11. Libevent and Tor integration improvements
    Tor should make better use of the more recent features of Niels Provos's Libevent library. Tor already uses Libevent for its low-level asynchronous IO calls, and could also use Libevent's increasingly good implementations of network buffers and of HTTP. This wouldn't be simply a matter of replacing Tor's internal calls with calls to Libevent: instead, we'll need to refactor Tor to use Libevent calls that do not follow the same models as Tor's existing backends. Also, we'll need to add missing functionality to Libevent as needed — most difficult likely will be adding OpenSSL support on top of Libevent's buffer abstraction. Also tricky will be adding rate-limiting to Libevent.
  12. Tuneup Tor!
    Right now, Tor relays measure and report their own bandwidth, and Tor clients choose which relays to use in part based on that bandwidth. This approach is vulnerable to attacks where relays lie about their bandwidth; to address this, Tor currently caps the maximum bandwidth it's willing to believe any relay provides. This is a limited fix, and a waste of bandwidth capacity to boot. Instead, Tor should possibly measure bandwidth in a more distributed way, perhaps as described in the "A Tune-up for Tor" paper by Snader and Borisov. One could use current testing code to double-check this paper's findings and verify the extent to which they dovetail with Tor as deployed in the wild, and determine good ways to incorporate them into their suggestions Tor network without adding too much communications overhead between relays and directory authorities.
  13. Improve our unit testing process
    Tor needs to be far more tested. This is a multi-part effort. To start with, our unit test coverage should rise substantially, especially in the areas outside the utility functions. This will require significant refactoring of some parts of Tor, in order to dissociate as much logic as possible from globals.
    Additionally, we need to automate our performance testing. We've got buildbot to automate our regular integration and compile testing already (though we need somebody to set it up on Windows), but we need to get our network simulation tests (as built in TorFlow: see the "Tor Node Scanner improvements" item) updated for more recent versions of Tor, and designed to launch a test network either on a single machine, or across several, so we can test changes in performance on machines in different roles automatically.
  14. Help revive an independent Tor client implementation
    Reanimate one of the approaches to implement a Tor client in Java, e.g. the OnionCoffee project, and make it run on Android. The first step would be to port the existing code and execute it in an Android environment. Next, the code should be updated to support the newer Tor protocol versions like the v3 directory protocol. Further, support for requesting or even providing Tor hidden services would be neat, but not required.
    A perspective developer should be able to understand and write new Java code, including a Java cryptography API. Being able to read C code would be helpful, too. One should be willing to read the existing documentation, implement code based on it, and refine the documentation when things are underdocumented. This project is mostly about coding and to a small degree about design.
  15. Bring moniTor to life
    Implement a top-like management tool for Tor relays. The purpose of such a tool would be to monitor a local Tor relay via its control port and include useful system information of the underlying machine. When running this tool, it would dynamically update its content like top does for Linux processes. This or-dev post might be a good first read.
    A person interested in this should be familiar with or willing to learn about administering a Tor relay and configuring it via its control port. As an initial prototype is written in Python, some knowledge about writing Python code would be helpful, too. This project is one part about identifying requirements to such a tool and designing its interface, and one part lots of coding.
  16. Torbutton improvements
    Torbutton has a number of improvements that can be made in the post-1.2 timeframe. Most of these are documented as feature requests in the Torbutton flyspray section. Good examples include: stripping off node.exit on http headers, more fine-grained control over formfill blocking, improved referrer spoofing based on the domain of the site (a-la refcontrol extension), tighter integration with Vidalia for reporting Tor status, a New Identity button with Tor integration and multiple identity management, and anything else you might think of.
    This work would be independent coding in Javascript and the fun world of XUL, with not too much involvement in the Tor internals.
  17. Porting Polipo to Windows
    Help port Polipo to Windows. Example topics to tackle include: 1) handle spaces in path names and understand the filesystem namespace — that is, where application data, personal data, and program data typically reside in various versions of Windows. 2) the ability to handle ipv6 communications. 3) the ability to asynchronously query name servers, find the system nameservers, and manage netbios and dns queries. 4) use native regex capabilities of Windows, rather than using 3rd party GNU regex libraries. 5) manage events and buffers natively (i.e. in Unix-like OSes, Polipo defaults to 25% of ram, in Windows it's whatever the config specifies). 6) some sort of GUI config and reporting tool, bonus if it has a systray icon with right clickable menu options. Double bonus if it's cross-platform compatible.
  18. Make our diagrams beautiful and automated
    We need a way to generate the website diagrams (for example, the "How Tor Works" pictures on the overview page from source, so we can translate them as UTF-8 text rather than edit them by hand with Gimp. We might want to integrate this as an wml file so translations are easy and images are generated in multiple languages whenever we build the website.
  19. Improve the LiveCD offerings for the Tor community
    How can we make the Incognito LiveCD easier to maintain, improve, and document?
  20. Rework and extend Blossom
    Rework and extend Blossom (a tool for monitoring and selecting appropriate Tor circuits based upon exit node requirements specified by the user) to gather data in a self-contained way, with parameters easily configurable by the user. Blossom is presently implemented as a single Python script that interfaces with Tor using the Controller interface and depends upon metadata about Tor nodes obtained via external processes, such as a webpage indicating status of the nodes plus publically available data from DNS, whois, etc. This project has two parts: (1) Determine which additional metadata may be useful and rework Blossom so that it cleanly obtains the metadata on its own rather than depend upon external scripts (this may, for example, involve additional threads or inter-process communication), and (2) develop a means by which the user can easily configure Blossom, starting with a configuration file and possibly working up to a web configuration engine. Knowledge of Tor and Python are important; knowledge of TCP, interprocess communication, and Perl will also be helpful. An interest in network neutrality is important as well, since the principles of evaluating and understanding internet inconsistency are at the core of the Blossom effort.
  21. Improve Blossom: Allow users to qualitatively describe exit nodes they desire
    Develop and implement a means of affording Blossom users the ability to qualitatively describe the exit node that they want. The Internet is an inconsistent place: some Tor exit nodes see the world differently than others. As presently implemented, Blossom (a tool for monitoring and selecting appropriate Tor circuits based upon exit node requirements specified by the user) lacks a sufficiently rich language to describe how the different vantage points are different. For example, some exit nodes may have an upstream network that filters certain kinds of traffic or certain websites. Other exit nodes may provide access to special content as a result of their location, perhaps as a result of discrimination on the part of the content providers themselves. This project has two parts: (1) develop a language for describing characteristics of networks in which exit nodes reside, and (2) incorporate this language into Blossom so that users can select Tor paths based upon the description. Knowledge of Tor and Python are important; knowledge of TCP, interprocess communication, and Perl will also be helpful. An interest in network neutrality is important as well, since the principles of evaluating and understanding internet inconsistency are at the core of the Blossom effort.
  22. Bring up new ideas!
    Don't like any of these? Look at the Tor development roadmap for more ideas.

Other Coding and Design related ideas

  1. Tor relays don't work well on Windows XP. On Windows, Tor uses the standard select() system call, which uses space in the non-page pool. This means that a medium sized Tor relay will empty the non-page pool, causing havoc and system crashes. We should probably be using overlapped IO instead. One solution would be to teach libevent how to use overlapped IO rather than select() on Windows, and then adapt Tor to the new libevent interface. Christian King made a good start on this in the summer of 2007.
  2. We need to actually start building our blocking-resistance design. This involves fleshing out the design, modifying many different pieces of Tor, adapting Vidalia so it supports the new features, and planning for deployment.
  3. We need a flexible simulator framework for studying end-to-end traffic confirmation attacks. Many researchers have whipped up ad hoc simulators to support their intuition either that the attacks work really well or that some defense works great. Can we build a simulator that's clearly documented and open enough that everybody knows it's giving a reasonable answer? This will spur a lot of new research. See the entry below on confirmation attacks for details on the research side of this task — who knows, when it's done maybe you can help write a paper or three also.
  4. Tor 0.1.1.x and later include support for hardware crypto accelerators via OpenSSL. Nobody has ever tested it, though. Does somebody want to get a card and let us know how it goes?
  5. Perform a security analysis of Tor with "fuzz". Determine if there are good fuzzing libraries out there for what we want. Win fame by getting credit when we put out a new release because of you!
  6. Tor uses TCP for transport and TLS for link encryption. This is nice and simple, but it means all cells on a link are delayed when a single packet gets dropped, and it means we can only reasonably support TCP streams. We have a list of reasons why we haven't shifted to UDP transport, but it would be great to see that list get shorter. We also have a proposed specification for Tor and UDP — please let us know what's wrong with it.
  7. We're not that far from having IPv6 support for destination addresses (at exit nodes). If you care strongly about IPv6, that's probably the first place to start.

Research

  1. The "website fingerprinting attack": make a list of a few hundred popular websites, download their pages, and make a set of "signatures" for each site. Then observe a Tor client's traffic. As you watch him receive data, you quickly approach a guess about which (if any) of those sites he is visiting. First, how effective is this attack on the deployed Tor codebase? Then start exploring defenses: for example, we could change Tor's cell size from 512 bytes to 1024 bytes, we could employ padding techniques like defensive dropping, or we could add traffic delays. How much of an impact do these have, and how much usability impact (using some suitable metric) is there from a successful defense in each case?
  2. The "end-to-end traffic confirmation attack": by watching traffic at Alice and at Bob, we can compare traffic signatures and become convinced that we're watching the same stream. So far Tor accepts this as a fact of life and assumes this attack is trivial in all cases. First of all, is that actually true? How much traffic of what sort of distribution is needed before the adversary is confident he has won? Are there scenarios (e.g. not transmitting much) that slow down the attack? Do some traffic padding or traffic shaping schemes work better than others?
  3. A related question is: Does running a relay/bridge provide additional protection against these timing attacks? Can an external adversary that can't see inside TLS links still recognize individual streams reliably? Does the amount of traffic carried degrade this ability any? What if the client-relay deliberately delayed upstream relayed traffic to create a queue that could be used to mimic timings of client downstream traffic to make it look like it was also relayed? This same queue could also be used for masking timings in client upstream traffic with the techniques from adaptive padding, but without the need for additional traffic. Would such an interleaving of client upstream traffic obscure timings for external adversaries? Would the strategies need to be adjusted for asymmetric links? For example, on asymmetric links, is it actually possible to differentiate client traffic from natural bursts due to their asymmetric capacity? Or is it easier than symmetric links for some other reason?
  4. Repeat Murdoch and Danezis's attack from Oakland 05 on the current Tor network. See if you can learn why it works well on some nodes and not well on others. (My theory is that the fast nodes with spare capacity resist the attack better.) If that's true, then experiment with the RelayBandwidthRate and RelayBandwidthBurst options to run a relay that is used as a client while relaying the attacker's traffic: as we crank down the RelayBandwidthRate, does the attack get harder? What's the right ratio of RelayBandwidthRate to actually capacity? Or is it a ratio at all? While we're at it, does a much larger set of candidate relays increase the false positive rate or other complexity for the attack? (The Tor network is now almost two orders of magnitude larger than it was when they wrote their paper.) Be sure to read Don't Clog the Queue too.
  5. The "routing zones attack": most of the literature thinks of the network path between Alice and her entry node (and between the exit node and Bob) as a single link on some graph. In practice, though, the path traverses many autonomous systems (ASes), and it's not uncommon that the same AS appears on both the entry path and the exit path. Unfortunately, to accurately predict whether a given Alice, entry, exit, Bob quad will be dangerous, we need to download an entire Internet routing zone and perform expensive operations on it. Are there practical approximations, such as avoiding IP addresses in the same /8 network?
  6. Other research questions regarding geographic diversity consider the tradeoff between choosing an efficient circuit and choosing a random circuit. Look at Stephen Rollyson's position paper on how to discard particularly slow choices without hurting anonymity "too much". This line of reasoning needs more work and more thinking, but it looks very promising.
  7. Tor doesn't work very well when relays have asymmetric bandwidth (e.g. cable or DSL). Because Tor has separate TCP connections between each hop, if the incoming bytes are arriving just fine and the outgoing bytes are all getting dropped on the floor, the TCP push-back mechanisms don't really transmit this information back to the incoming streams. Perhaps Tor should detect when it's dropping a lot of outgoing packets, and rate-limit incoming streams to regulate this itself? I can imagine a build-up and drop-off scheme where we pick a conservative rate-limit, slowly increase it until we get lost packets, back off, repeat. We need somebody who's good with networks to simulate this and help design solutions; and/or we need to understand the extent of the performance degradation, and use this as motivation to reconsider UDP transport.
  8. A related topic is congestion control. Is our current design sufficient once we have heavy use? Maybe we should experiment with variable-sized windows rather than fixed-size windows? That seemed to go well in an ssh throughput experiment. We'll need to measure and tweak, and maybe overhaul if the results are good.
  9. Our censorship-resistance goals include preventing an attacker who's looking at Tor traffic on the wire from distinguishing it from normal SSL traffic. Obviously we can't achieve perfect steganography and still remain usable, but for a first step we'd like to block any attacks that can win by observing only a few packets. One of the remaining attacks we haven't examined much is that Tor cells are 512 bytes, so the traffic on the wire may well be a multiple of 512 bytes. How much does the batching and overhead in TLS records blur this on the wire? Do different buffer flushing strategies in Tor affect this? Could a bit of padding help a lot, or is this an attack we must accept?
  10. Tor circuits are built one hop at a time, so in theory we have the ability to make some streams exit from the second hop, some from the third, and so on. This seems nice because it breaks up the set of exiting streams that a given relay can see. But if we want each stream to be safe, the "shortest" path should be at least 3 hops long by our current logic, so the rest will be even longer. We need to examine this performance / security tradeoff.
  11. It's not that hard to DoS Tor relays or directory authorities. Are client puzzles the right answer? What other practical approaches are there? Bonus if they're backward-compatible with the current Tor protocol.
  12. Programs like Torbutton aim to hide your browser's UserAgent string by replacing it with a uniform answer for every Tor user. That way the attacker can't splinter Tor's anonymity set by looking at that header. It tries to pick a string that is commonly used by non-Tor users too, so it doesn't stand out. Question one: how badly do we hurt ourselves by periodically updating the version of Firefox that Torbutton claims to be? If we update it too often, we splinter the anonymity sets ourselves. If we don't update it often enough, then all the Tor users stand out because they claim to be running a quite old version of Firefox. The answer here probably depends on the Firefox versions seen in the wild. Question two: periodically people ask us to cycle through N UserAgent strings rather than stick with one. Does this approach help, hurt, or not matter? Consider: cookies and recognizing Torbutton users by their rotating UserAgents; malicious websites who only attack certain browsers; and whether the answers to question one impact this answer.
  13. Right now Tor clients are willing to reuse a given circuit for ten minutes after it's first used. The goal is to avoid loading down the network with too many circuit extend operations, yet to also avoid having clients use the same circuit for so long that the exit node can build a useful pseudonymous profile of them. Alas, ten minutes is probably way too long, especially if connections from multiple protocols (e.g. IM and web browsing) are put on the same circuit. If we keep fixed the overall number of circuit extends that the network needs to do, are there more efficient and/or safer ways for clients to allocate streams to circuits, or for clients to build preemptive circuits? Perhaps this research item needs to start with gathering some traces of what connections typical clients try to launch, so you have something realistic to try to optimize.
  14. How many bridge relays do you need to know to maintain reachability? We should measure the churn in our bridges. If there is lots of churn, are there ways to keep bridge users more likely to stay connected?

Let us know if you've made progress on any of these!

#include