January 2016 - tor-dev - lists.torproject.org

Proposal xxx: Filtering malicious rendezvous points at hidden service server side
by s7r 26 Jan '16

26 Jan '16

-----BEGIN PGP SIGNED MESSAGE----- Hash: SHA256 Hello, Inspired by asn, dgoulet and Mike Perry discussing alternate path lengths for proposal 247, I wrote a proposal that could fit in nicely with proposal 247, maybe even allow us to make the 3rd hop wild (random) so we don't have to worry about HSDirs + HSDir fetch/post circuits and introduction points + introduction points circuits linking short term layer 3 guards. Please find the proposal attached as well as inline for comments. Looking forward for reviews. If I've structured it wrong or something I apologize in advance - will do better next time. Filename: xxx-malicious-rendezvous-point-filter.txt Title: Filtering malicious rendezvous points at hidden service server side Authors: s7r [ 0x837FA52C81265B11 ] Created: 23 January 2016 Status: Draft 0. Definitions client = an user running Tor as a client, which connects to a hidden service. hidden service server = the Tor instance hosting the hidden service. rendezvous point = the relay which acts as the rendezvous meeting point in a circuit between a hidden service client and a hidden service server. 1. Motivation A hidden service circuit (rendezvous circuit) is always initiated by the client so the relay which should act as the rendezvous meeting point is selected by the client. We assume that any client can be an attacker and will try to make the hidden service server connect to a malicious rendezvous point under his control. The attacker is also a Sybil (holds an unknown % of the bandwidth in the Tor network). By making the hidden service server build many circuits to his evil rendezvous points, the attacker gets a high probability that the hidden service server will eventually pick his evil relays in a circuit, so the attacker will trivially perform a successful hidden service guard discovery attack or, with more luck, discover the real location of the hidden service server. The hidden service server can only defend itself by building a 3 hop circuit to the rendezvous point, but in practice this is not always enough. 2. Logic for filtering malicious rendezvous points To be able to make fair assumptions about which rendezvous point might be malicious or not we are going to use the probability theory and statistics, precisely the binominal distribution equation [0]. In simple words, we count and keep track of how many rendezvous circuits a hidden service server built and to which rendezvous points. Then, based on the weight (middle probability fraction) of each rendezvous point, we determine if one was insanely overpicked by clients. Think of this like: Alice plays a game with Bob in which Bob has the success probability to draw a certain number n%. In practice however, he shows a success rate of more than (n*2)% (2 times bigger), Alice assumes Bob is not playing nice and will refuse to play with him for the next 24 hours or so. Fast example: A hidden service server built 100 rendezvous circuits in the last 24 hours, out of which in 7 of them the rendezvous point was a relay with the middle probability fraction of 0.5%. We assume that rendezvous point is malicious and refuse to build rendezvous circuits to it for the next 24 hours. 3. Equations and formulas 3.1 Probability of different clients genuinely picking the same rendezvous point n = total number of established rendezvous circuits. k = number of clients that could have genuinely picked the same rendezvous point. p = probability of picking that rendezvous point per client (middle probability fraction of the relay) divided to 100. H = Probability mass function expanded. C = The result of the equation, the value we need to determine multiplied with 100 to have the percent value. Formulas: H = n! / k!^(n - k)! C = H * p^k * (1 - p)^n - k Example: n = 100 k = 9 p = 0.5% / 100 = 0.005 H = 100! / 9!^91! = 20023492720 C = 20023492720 * 0.000000000000000000001953125 * 0.63372, ~ 0.00000000002478376524710625 * 100, ~ 0.000000002478376524710625% 3.2 Expected number of clients to pick the same rendezvous point n = Total number of established rendezvous circuits. p = Probability of picking that rendezvous point per client (middle probability fraction of the relay) divided to 100. We multiply this with 2 intentionally so we will allow 2 times more circuits than we would expect based on this formula - this should reduce the error margin enough. C = The result of the equation, the value we need to determine. Formula: C = n * p Example: n = 100 p = 0.5% / 100 = 0.005 * 2 = 0.01 C = 100 * 0.01 = 1 4. Implementation and variables We keep track in the persistent state of a hidden service server of the total number of built rendezvous circuits as well as a counter for number of rendezvous circuits per rendezvous point. Variables: REND_FILTER_MONITOR_PERIOD = count and track the rendezvous circuits built within this latest period. We set it to 24 hours. REND_FILTER_BAN_PERIOD = for how long we are going to ban a malicious rendezvous point. We set it to 24 hours. REND_FILTER_TOLERANCE = how many circuits we allow per rendezvous point regardless of its middle probability fraction. We set it to 4. REND_FILTER_BAN_ACTIVATION = when we should ban a malicious rendezvous point. To determine this, we are going to get from REND_FILTER_MONITOR_PERIOD the total number of established rendezvous circuits as well as the number of established rendezvous circuits with a given rendezvous point and apply the formula from 3.2 with approximation to the next whole number (example: 5.01 = 6; 5.99 = 6). Example: A hidden service server built 100 circuits in the last 24 hours out of which 5 to a rendezvous point with the middle probability fraction 1% (p = 0.02), so the expected number of circuits we expect is 2, but allow up to maximum 4 (if REND_FILTER_TOLERANCE > REND_FILTER_BAN_ACTIVATION, REND_FILTER_TOLERANCE prevails). In this case we consider this too much of a coincidence and ban this rendezvous point for the next 24 hours. We keep this ban in our persistent state with a timestamp so we'll know when it expires. 5. Ban malicious rendezvous points just for rendezvous circuits, not for anything else If we treat as relay as a malicious rendezvous point for a given time period, this only means we are not building circuits with it as the rendezvous point, but we don't exclude it for other purpose circuits, regardless which are those other purpose circuits. We may use them in our exit circuits, consensus fetch circuits, as HSDirs and HSDir fetch / post circuits, as Introduction Points or Introduction Point circuits. 6. Non zero probability to assign the malicious rendezvous point flag to an innocent relay This is indeed a non-zero value and it can be computed with the formula from 3.1, but it's a value so close to zero that we think it's totally worth it. Even if accidentally (low chances) an innocent relay will be banned, this will be something local to the hidden service server. It won't affect that relay at all, nor how other client or hidden service servers treat that relay. It has nothing to do with the network wide consensus as well. A honest client will always retry with a different rendezvous point, so honest clients should not experience reachability issues. 7. Rendezvous points that aren't included in hidden service server's consensus Clients and hidden service servers might have different consensuses. This means that a hidden service server can be genuinely asked to build a rendezvous circuit to a rendezvous point for which it has no middle probability fraction because it's not included in that consensus. We should allow REND_FILTER_TOLERANCE number of circuits to the rendezvous points for which we have no weights at all and if that rendezvous point appears in a later consensus within the REND_FILTER_MONITOR_PERIOD update its middle probability fraction value to ensure the filter will activate and work properly once REND_FILTER_TOLERANCE threshold was reached. 8. Security implications for maintaining such records in the persistent state of a hidden service server An attacker which compromises the location of a hidden service server will access this additional information: total number of rendezvous circuits built within the last REND_FILTER_MONITOR_PERIOD and to which rendezvous points these circuits were. This tradeoff should be worth it as well, since if the location of a hidden service server is compromised it is game over anyway, and this additional information shouldn't be so important to the attacker, except maybe for better determining the popularity of that hidden service (this can be determined many other ways, like the logs from the application running behind the hidden service, network counter, datacenter/ISP level counters and logs, etc.). To avoid co-location linkability between different hidden services, we should keep rendezvous circuits records and counters individually for each hidden service. The logic is designed to adjust by itself dynamically, based on the popularity and usage of a hidden service. If it's a high profile hidden service, experiencing 10000 connections per 24 hours, the REND_FILTER_BAN_ACTIVATION will increase for every rendezvous point. 9. Default option, with possibility to turn off for testing environments and experiments We add a new torrc option named HiddenServiceRendFilter 0|1 and set it to 1 by default. It can be turned off by setting it to 0, if this will be desired in some testing environments or experiments. 10. Other considerations, limitations and compatibility with other existing proposals This proposal won't disable an attacker entirely, but it will make it much harder and more expensive, given that the attacker will have to run more malicious relays to act as rendezvous points, increasing the capacity of the Tor network overall and directly the costs of the attack. Also, the attacker will have to grow the popularity of a hidden service so that he can establish more rendezvous circuits using the same rendezvous points. This will at least be kind of hard if other genuine clients pick the same rendezvous points and get noisy to the hidden service operator (much more traffic, same number or slightly more users on his service). This proposal is compatible with all the other hidden service related proposals to date. For tools like OnionBalance, each backend / fallback Tor instance included in the master hidden service descriptor will just transparently and individually maintain their own tracking vales and ban list. With proposal 260 - Rendezvous Single Onion Services this should be disabled since it makes no sense. Set HiddenServiceRendFilter 0 in a RSOS type hidden service. This proposal should be quite helpful once we implement enough padding between relays, where an attacker might have to own the entire path in order to distinguish dummy traffic from real traffic. This proposal should fit in nicely with proposal 247 - Vanguards and could make it safe to leave the 3rd hop wild (no layer 3 short term guards) and use only layer 2 medium term guards + layer 1 long term guard since an attacker will be able to establish less circuits to his malicious rendezvous point, thus drastically decreasing the chances of the hidden service server to also pick his malicious relay as 3rd hop to the malicious rendezvous point (layer 2 guards discovery). [0]: https://en.wikipedia.org/wiki/Binomial_distribution -----BEGIN PGP SIGNATURE----- Version: GnuPG v2.0.22 (MingW32) iQEcBAEBCAAGBQJWo/K4AAoJEIN/pSyBJlsRAnMIAKFDDFm0DICrog1JecmrO82y N7dJf6tDURjGkUTr5i+mf91S4gTvxYigkT/tWTPH2E65XQGl2qLXo4d1+vgfaJmf H7QxUFy2kt+oFV03pg7SsF1KlMLaIVjlPMC17hPjAL8wnLMPkTICUn+VqWVYqmq+ w4uFPncm42mhkd3EwRq3HCrwq4O7PasWfSBYXmAuIJl6CWDkZqBI+UUj56dqQG6c zpWUMIZpfaMQzEJ438VhpVpiN19Ep5hIV5dJfz4MJ9eyhyas487nRXYyirEWqBs0 VXUsb/m/wgHICLjLukNYJ/+NZUk+Y5mIhL756HIiYUlL8UK85MaJhbozvjhqLnw= =tMjA -----END PGP SIGNATURE-----

5 13

Fwd: Orbot v15.1.0-RC-4
by Nathan Freitas 26 Jan '16

26 Jan '16

Orbot update with new built-in bridges (currently using the TBB default list, randomly shuffled!) and some other small fixes for the user interface, reproducible builds, and VPN feature. ----- Original message ----- From: Nathan of Guardian <nathan(a)guardianproject.info> To: guardian-dev(a)lists.mayfirst.org Subject: Orbot v15.1.0-RC-4 Date: Mon, 25 Jan 2016 22:48:01 -0500 -----BEGIN PGP SIGNED MESSAGE----- Hash: SHA256 Fixes based on feedback from our RC-2 release.... this will be up on Play and F-Droid shortly, and from the link below right now. /** 15.1.0-RC-4 / 25-January-2016 / 80491ea95bd62346bf3f3a579d3302dbd84f4ca9 **/ APK: https://guardianproject.info/releases/Orbot-v15.1.0-RC-4.apk SIG: https://guardianproject.info/releases/Orbot-v15.1.0-RC-4.apk.asc Improve Bridge selection to support built-in default bridges * c235e3e tweak the string about bridges a little bit * 2a72814 choose up to 2 bridges from default list randomly * 1dbe5ea make bridge allocation shuffle randomly to distribute load inspired by this work: https://trac.torproject.org/projects/tor/tic * cf1a644 add support for loading default bridges from asset file VPN mode now uses Google's DNS (tunneled through Tor) as default * 9af00fe change VPN mode DNS to use Google's 8.8.8.8 Other UI fixes * ba83559 this should be kbps in fact * ab8709d add orfox icon to replace orweb * 6eb0a93 improvements to Locale changing in app * 0669add re-add appcompat, update to latest, move to Android 23 to build - this is required for latest appcompat, and to address the bu * c05d8e7 remove appcompat, and just support support-v4 Fix our stpcpy implementation to ensure no buffer overruns * 4913b0c better implementation of stpcpy for pre-Android 21 NDK Make the build process better and more reproduceable * 735b298 make-release-build: remove faketime from `ant release` * 461e35d make-release-build: freeze time when running ndk-build * 5c86b5c make-release-build: make sure ndk-build can be found * 58d53ea make-release-build: use strip-nondeterminism to get reproducible build * 5ce1f5f make-release-build: make sure tag signers exist before verifying tags * 72eab39 build jtorctl directly, using a symlink to point to its source code * a6ac016 use symlinks to provide alternate folders for Hebrew and Indonesian * 5fb4e9b update ndk-builds to NDK 4.8 toolchain - -- Nathan of Guardian nathan(a)guardianproject.info - -- Nathan of Guardian nathan(a)guardianproject.info -----BEGIN PGP SIGNATURE----- Version: Mailvelope v1.3.4 Comment: https://www.mailvelope.com wsFcBAEBCAAQBQJWpuxcCRCoARg+abN6qQAABa8P/2AlGCzBOBWhb4rVTvfE 64af8Pw031oJyyFDvtlUeETS1Ct95Zz7prP41ywhItPi/6eLCTwkbcbDA+Zl xDK6eSy6NhyMgh2qrC3US45HvQGjSCA7496IM5T3lL71wEL71sxAIaPh2abP 1akDEiXiJTnRUqFGui0aQkt0rJKQ0GShqJXmDM6qvuMCoiJSJ/ehc6Rnkyam 3/m4j5+IZ5JNpPKPDFjmW5cNFSDr+11SSdWlzjC9QfE8Ebo0rceMgpFASqH+ alzl6dB7y2rA+QeQcFQfwTlckKELq/cNofruV4UYmEIIlt/3dPZkIoIPlxBT i/ozTkmSPyEUurnxmAJQMVhergN/0WzGOWST5ceLABjjuVsoVrCZ7ai4eKiX M2zHxPQdjeLnc1HG5IXYLQcABJ9j9ol84oIH4VfcMPBCSjb205S42ChBx8Tm 9LGSSkn1i7JmVR2BaXaxmhbgoU5X+abumy3fH2uwDS5WCMDAExVw43sb8qM9 pGbDxK/wYd4aBaYZQK9VO7nA48eiv7yIw7uoKOhA7JFOnLg8he0DcngikCIx VNEqaO231Kgj1kN8Uc8+ombelRxb++997gnWfqE2ZHsEt3BpRpSAephJMJyq pxOeW/0tQL860oObACLjIp9xZ5FecZPWB4QqDEBRdOtT4MsgjDRP/Wty8Rzt UF2G =9Fk9 -----END PGP SIGNATURE-----

1 0

Re: [tor-dev] Release: obfs4proxy-0.0.6
by Nathan Freitas 25 Jan '16

25 Jan '16

On Mon, Jan 25, 2016, at 09:38 AM, Yawning Angel wrote: > benefit is that it is a lot easier to package than meek-client + the > external helper, and this can probably save binary size on things > like Android. Yes, I've been waiting for this. It is especially useful since we will be including both ARM and x86 builds shortly, so the size of PT storage in the app is already going to double. Thanks for the continued hard work on this. +n

1 0

Release: obfs4proxy-0.0.6
by Yawning Angel 25 Jan '16

25 Jan '16

Hello all, I just tagged obfs4proxy-0.0.6. There aren't many significant changes, and the internal changes primarily affect the client side initialization, so those of you that are perfectly content with obfs4proxy-0.0.5 can continue to use the existing version without issue. Tarball/Signature: https://people.torproject.org/~yawning/releases/obfs4proxy/obfs4proxy-0.0.6… https://people.torproject.org/~yawning/releases/obfs4proxy/obfs4proxy-0.0.6… Changes in version 0.0.6 - 2016-01-25: - Delay transport factory initialization till after logging has been initialized. - Add a meek client implementation (WARNING: Does not support using a helper to normalize TLS signatures). The brave people that want to use it can do so as the "meek_lite" transport, with identical bridge lines to the real meek-client. This means "It's blatantly obvious that you're using a HTTPS client written in Go because the ClientHello is rather distinctive". The benefit is that it is a lot easier to package than meek-client + the external helper, and this can probably save binary size on things like Android. An example bridge line would be: Bridge meek_lite 0.0.2.0:1 url=https://meek-reflect.appspot.com/ front=www.google.com It has received moderate testing when I was on a business trip, and appears to function as intended, given the design constraints. There are minor internal API changes (which caused issues delaying the release) motivated by "meek's behavior is unlike most other transports" but I expect such things to be straight forward to those that actually hack on the code. Thanks to the person on Github for filing a pull request which saved me time debugging. Regards, -- Yawning Angel

1 0

Moving prop#250 discussion meeting to Tuesday, 1600 UTC
by Nick Mathewson 25 Jan '16

25 Jan '16

As mentioned on the other thread, George and David (who wrote proposal 250) can't make it at the originally planned time. So instead, we'll be discussing proposal 250 at 11 am eastern (1600 UTC) on Tuesday! (I'll still be around at the planned time to talk about patches and code and whatever.) cheers, -- Nick

1 0

New Shadow versions now available
by Rob Jansen 24 Jan '16

24 Jan '16

Hello, Shadow versions 1.11.0 and 1.11.1 are now released! https://github.com/shadow/shadow/releases/tag/v1.11.0 https://github.com/shadow/shadow/releases/tag/v1.11.1 These versions represent a huge engineering effort to provide transparent support for plug-ins that use pthreads and that make blocking IO or system calls. We are using a fork of GNU portable threads (http://www.gnu.org/software/pth/) to enable this thread and blocking support. Shadow now literally just calls `tor_main()` and handles the rest internally. More details can be found in the release notes here: http://mailman.cs.umn.edu/archives/shadow-support/2016-January/000475.html All the best, Rob

1 0

Proposal 251 (netflow padding) meeting logs, minutes, and action items
by Mike Perry 22 Jan '16

22 Jan '16

We had our meeting about the netflow padding defense proposal and implementation today: https://gitweb.torproject.org/torspec.git/tree/proposals/251-netflow-paddin… Here's the IRC logs: http://meetbot.debian.net/tor-dev/2016/tor-dev.2016-01-22-16.00.log.html The minutes+action item summary is as follows: - Write a torspec branch to describe the defense and changes since the proposal. The protocol changes will go in tor-spec.txt, and the behavior/usage will go in padding-spec.txt. Right now, I do not plan to also update Prop#251 itself, as the spec will be canonical. - In padding-spec.txt, we need to describe the options for disabling padding, reducing padding, and connection lifespan controls as these have changed since Prop#251 was written. - We are not going to use channel padding to obscure circuit setup (https://trac.torproject.org/projects/tor/ticket/17591) as fingerprinting is more of a risk from the guard node rather than someone watching TLS. It is better handled through multi-hop padding. That ticket needs to be closed or replaced with one to write a separate proposal for multi-hop padding. - We discussed overhead bounds and if we believe there are adversaries that exist that this will protect against (who won't/aren't doing full take for example). We believe data retention regulation and ISP practices strongly indicates that there are. - Bridges will be padded like clients are. - We discussed using behavioral checks to differentiate clients from relays instead of the consensus checks. I do not like this option, as it introduces protocol side effects that seem sloppy and prone to cause problems in the future (like CREATE_FAST did). I am not sure if we came to an agreement here. - We need to be sure to describe the protocol negotiation and forward compatibility properties of the padding negotiation in detail in tor-spec.txt. After the meeting, Nick and I discussed some implementation details. We need to improve the callback handling and maybe also upgrade libevent's timer algorithm. Nick is working on a new data structure to look up channels in O(1), so we can fix a callback channel dropping bug he spotted in code review, and eliminate the need to reach into channeltls_t from channelpadding.c. Nick also has code review comments in https://trac.torproject.org/projects/tor/ticket/16861#comment:39 that I have mostly fixed, minus the callback issue. I will be posting a reply and a new branch later today. I think that's it. Please reply with questions, comments, or additions, or if I missed anything. -- Mike Perry

1 0

Entry/Exit node selection
by Evan d'Entremont 22 Jan '16

22 Jan '16

What threat is Tor trying to defeat? Region locking or nation states? If the former, then great, select an exit country, or just use a VPN. If the latter, perhaps that actual threat profile should be taken into account. Is there any reason why Tor doesn't select exit nodes which are as close as possible to the intended host? If I connect to Tor and request a resource from a server on ISP A, would in not make sense to enforce an exit node also on ISP A, or if not, as close as possible? As well, entry guards should be as close as possible to the user, limiting the ability of others to log the connection. In short, it's safer that only my ISP see a connection rather than my ISP, a backbone provider, the entry guard's ISP, etc. Systems like XKeyscore wouldn't even see the traffic in this case. It seems that selecting an exit country may actually be detrimental to anonymity by forcing traffic over the (monitored) internet backbone. Evan d'Entremont

4 6

SyslogIdentityTag considered to be backported to 0.2.7?
by nusenu 21 Jan '16

21 Jan '16

Hi, I like weasel's [1] SyslogIdentityTag feature that is available in debian 0.2.7 packages only. The ticket [1] suggests that it will be included in tor 0.2.8.x (milestone). Would you consider to backport it to 0.2.7 as well so other platforms can make use of it before we see tor 0.2.8 (without having to ask every package maintainer to maybe ship packages with the patch applied)? thanks! [1] https://trac.torproject.org/projects/tor/ticket/17194 https://gitweb.torproject.org/debian/tor.git/tree/debian/patches/20-upstrea…

2 1

Is it possible to specify voluntary delays in my Tor client?
by Virgil Griffith 20 Jan '16

20 Jan '16

I.e., if I want the extra resistance to traffic analysis that higher latency connections provide, is there a way to specify that in my Tor config? -V

5 6