Changes between Version 31 and Version 32 of thesis

Apr 17, 2011 6:52:13 PM (10 years ago)


  • thesis

    v31 v32  
    66*please add*
     8< FIrst of all thanks a lot for your reply! I'm going to add the comments and hopefully mark them obviously. We will in the next couple of days pass around the current version of the paper to be published.
    1010Specific comments:
    1313   A routerInfo does not contain a "self-signed certificate". It contains a pair of public keys and a "null certificate", which is really just a placeholder for future stuff. It is, however, signed by one of the router's private keys.
     15< Thanks and fixed.[[BR]][[BR]]
    1517Sec. 3.3 Eepsites:
    1618   The "identifier" (what we call a "destination") is actually 387 bytes in binary and 516 characters when encoded in Base 64, not 517 bytes. Leasesets are looked up in the netDB using the 32-byte SHA256 Hash of the destination.
     20< Thanks and fixed.[[BR]]
    1822Sec, 4.1:
    1923   Long paths might be much harder than in ref. 22; our absolute limit is 7 hops max (due to the max length of the tunnel build request message). As of 0.8.4, here are further restrictions enforced: Router A will not build a tunnel A-A. An unmodified router B will not build a tunnel A-B-A (although a hostile B could build this tunnel). A tunnel A-B-C-A cannot be prevented even with non-hostile B and C. So the longest without cooperation would be A-B-C-A-D-E-A. More complex long paths using multiple tunnels or garlic messages may be possible, although there are also maximum message expiration times enforced.
     25< We removed the claim that long tunnels might work.[[BR]][[BR]]
    2127   Actually, I2P doesn't use peers from the same /16 in the same tunnel. It does allow multiples in the fast tier. Since your attack doesn't require two attackers in the same tunnel, the /16 restriction may not be relevant here.
     29< You need /16 restriction for inbound tunnels. Here the "guard" node is the last chosen. Discussion in the paper.[[BR]][[BR]]
    2331Fig. 4.2: outbound tunnel labeled as inbound; The "monitor peers" from Fig. 4.1 with red and black stripes are now labeled "A" in this figure, which is confusing.
     33< Great catch![[BR]][[BR]]
    2535Sec. 5:
    2636   You say that each peer was configured for 64 KBps max but isn't that true only for the 40 attack peers? What was the bandwidth configuration for the 30 monitor peers? Was 64 KBps really high enough to be included in the victim's fast tier? Or were the monitor peers modified such that they would only provide "fast" service for the victim?
     38< Both, monitor and attack peers run with 64 KB/s. Monitor peers just participate in the network and behave as any other node. (No special behavior to the victim)[[BR]][[BR]]
    2840Table 5.2: The network size is estimated to be about 2500 uniques per day, and about 6000 - 7000 uniques per month (source )
     42< Good to know.[[BR]][[BR]]
    3044Figures 5.4 and 5.5: What's the difference between these two figures? Just two different examples?
     46< Yes.[[BR]][[BR]]
    3248Table 5.5:
    3349    What about 3-hop, which is the default for eepsites? 2-hop eepsites are not very secure and 1-hop is trivial. Really need data for 3-hop.
     51< 3-hop requires handling of A-F-A-V-B-C-D case (A=attacker, V=victim, B/C/D=bystanders, F=false-positive), which is easy (looking at traffic volumes not changing) but was not (yet) implemented (hence no data presented); preliminary data shows that the overall signal strength is not significantly disminished. 1-hop and 2-hop data sufficiently strong. Bottom line: If we have time to add the data, it will be added.[[BR]][[BR]]
    3553    "for the duration of the measurement": How long was it? minutes, hours, days? The time-to-deanonymize would be good to include here. It isn't clear if you deanonymize in one tunnel lifetime (10 minutes) or it takes multiple successful placements of the monitor peers over a long time period.
     55< 4 hours test data. (Addressed in the paper) Every point is a deanonymization as we described it. Tunnel participation in inbound and outbound tunnel, for 10 minutes.[[BR]][[BR]]
    3757Sec 6 Discussion:
    3959  The I2P network is still relatively small but is growing quickly. How about a prediction or sensitivity analysis for a network 10X, 100X larger? The analysis starts with "a" monitor peers out of the victim's fast pool of 30 peers. There's no analysis or discussion of how many monitor peers of a given bandwidth you need in the entire network to attain the number "a" in the victim's fast tier.
     61< We don't need a constant number of monitors in the victim's fast tier all the time, and obviously this depends on how many attackers, how many monitors, how fast the rest of the network is AND on the configuration of the victim.  Too many factors for a controlled experiment with a clear answer.  What we show is that an adversary controlling X% of your network can use a DoS on the fast tier-estimate to have MORE than X% of his peers in the victim's fast tier.[[BR]][[BR]]
    4163  In fact, most fast peers are from a Class "O" (greater than 128 KBytes/sec) group of routers and those are about 20% of the network - so there's perhaps 400 peers that could potentially be in the fast group in today's network of 2000 - 3000 routers.
     65< QUESTION: How would an I2P router with 64 KB/s even obtain a speed value measurement > 64k for any other peer?
     66< Also, even Class "O" peers would be DoSed by our attacker and the victim would (eventually) choose our non-class-O monitor nodes.[[BR]][[BR]]
    4368  So isn't this really about an adversary taking over a large proportion of the entire network, or at least of the network's fast routers? Is I2P any more vulnerable at X % hostile peers compared to other networks? Once you have a large number of hostile fast peers in the network, is the traffic analysis of your attack any quicker or more reliable than other attacks, e.g. first and last node in a tunnel (ref: "one ping enough" paper or blog post about Tor)
     70< Really neither.  It is about leaking information about lease sets (vuln. 1), performance-based selection (vuln. 2), long-running services (vuln. 3) and a lack of defense (no guard nodes) and about the possibility of using DDoS attack to change the network-view of a *single*, targeted peer (who's identity is NOT known a-priori, see below).[[BR]][[BR]]
    4572  Also not discussed - effect of leaseset size (number of leases or inbound tunnels) which is user-configurable from 1 to 6. It also is configurably dynamic, with less leases when the server is idle. A high number of leases makes it quicker for an adversary to enumerate the fast peers. I assume you used the default setting of 2 leases for your experimental victim.
     74< Our victim was configured for only 1 lease, more leases would have made our attack easier, so we went for the worst-case here. But yes, we mention it now ;-).[[BR]][[BR]]
    4776  Also not discussed - you started from the end, i.e. the identity of the victim, then your whole experiment was to confirm it. To find the victim from scratch would require another O(n) in time or resources, where n is the size of the network, i.e. you have to run the experiment on each router in the network.
     78< Not correct.  While we did the experiment against an Eepsite that we had set up so that we could be sure it worked and minimize impact on real sites, a real adversary would not have n-times more work; the adversary does NOT start with a particular router in mind (this is NOT a confirmation attack) but rather starts with a hostname, looks up leases and then does HTTP requests via the tunnels specified in the lease and DoS on entry nodes from the lease.  So while we knew the correct answer, the statistical analysis code / detection routines did not at all have any particular router as a target in mind when the code was executed.[[BR]][[BR]]
    4980  Unidirectional tunnels as a "bad design decision":
    5081   A bold and absolute statement not fully supported by the paper. It clearly mitigates other attacks and it's not clear how to trade off the risk of the attack described here (either currently or after implementing one or more of the recommendations below) with attacks on a bidirectional tunnel architecture.
     83< Yes, this is a bit of a bold statement, but only because we focus on one particular attack / system design aspect, and we might be missing out other issues.  However, we're not aware of any research showing advantages of uni-directional tunnels at this point that would mitigate the issues raised by our analysis. Please let us know if we miss something.[[BR]][[BR]]
    5285   Once the attacker's routers are a large portion of the victim's fast tier (e.g. 'one ping enough'), all sorts of analysis and attacks are possible, and many would be the same or easier with bidirectional tunnels. While we appreciate the innovation of your timing analysis attack with our unidirectional tunnels, the victim is eventually owned via any number of attacks when his fast tier is overtaken.
     87< This is all a question of degree: how fast can the adversary be how certain of the identity of the user.  You are right that in general, the attack still applies to bidirectional tunnels as well.[[BR]][[BR]]
    5489   Also, given future increases in network size and  implementation of recommendations, the tradeoff of analysis time vs. false-positive rate may change. For example, a 30x increase in analysis time for unidirectional tunnels is not insignificant. What if the fast tier size was increased to 1000? Then the time would be 1000x.
     91< Yes, and accuracy would be x1000000.  So while now the adversary might want to observe the victim in the correct position in 100 tunnels for 99.9999% certainty, the same might be achieved with just a single observation once the fast tier grows to 1000 peers (numbers given are random values, not based on specific facts, other than the x1000000).  Also, given that the adversary has "infinite" time (Eepsites are online a long time), the time increase is likely to be not so dramatic, whereas accuracy is something that might in general be harder to improve.[[BR]][[BR]]
    5693  Paper's recommendations:
    5996      Possibilities: Increase 45 sec evaluation cycle, increase 30-peer fast max and/or 75-peer high-cap max. Downside: larger group makes it more likely for an attacker to eventually be in a tunnel.
     98< We did not suggest to simply change the tier sizes or evaluation cycles.  The suggestion was rather to limit churn, for example by saying that per evaluation cycle, only one peer can be removed from a given tier (adding more might be OK during startup).  You might impose further limits, for example that at most 5 peers can be removed (and not re-added) within 10 cycles, etc.  That would force the adversary to maintain any DoS for longer, still allow you to have the "fast" tier be small and fast AND possibly help your network stability.  45s might also be a bit fast, but changing that value alone would neither be enough nor go to the heart of our suggestion.[[BR]][[BR]]
    61100      Not a possibility: Increasing 10-minute tunnel lifetime (unfortunately it is essentially hard-coded in the network now)
     102< This was not suggested and would likely rather be harmfull (more data points for statisitcal analysis for the adversary).[[BR]][[BR]]
    63104   2) Distributed HTTP services:
    64105      This is supported via "multihoming", whereby multiple routers may host an eepsite. This requires some additional setup, and of course requires the user to operate multiple routers. Truly distributed hosting is under development through a port of the tahoe-lafs distributed file system to I2P.
     107< Okay, we mention is ongoing process as possible solution.[[BR]][[BR]]
    66109   3) Use random peers for leases (guard nodes):
    67110      By this you mean, I think, using random peers outside the fast tier for the inbound tunnel's gateway. We could also keep these peers semi-constant, or more stable, by attempting to recreate the same tunnel at expiration, while still changing them on rejection. This could be done either from the fast tier or by using a random peer. Perhaps each destination could maintain its own "guard tier" that changes slowly.
     112< You seem to confuse guard nodes (those closest to the victim) and 'entry nodes' (those advertised in the leaseSet).  Our suggestion was to pick a completely RANDOM (not-tier, not-fixed) node for the 'entry nodes', so that the adversary cannot learn about your fast tier from the leaseSet. Obviously, there are trade-offs involved.[[BR]][[BR]]
    69114      Benefits / downsides? What happens if an adversary attacks guard nodes (either in I2P or Tor)?
     116< Again, the point here was that the adversary should not be able to figure out the inbound nodes in the first place by not leaking information about likely fast-tier nodes in the leaseSet. We clarified this in the text.[[BR]][[BR]]
    71118  Additional possible changes to I2P not mentioned:
    74121      Changes made in 0.8.4, more coming in 0.8.5, further research needed
     123< Yes.[[BR]][[BR]]
    76125   2) Connection limits (limiting number of requests per client in a minute, hour, or day) are supported in i2p but not enabled by default, these limits, if enabled, would prevent a single client from making requests every 15 seconds forever, although a distributed attack would still be possible. In addition, the server sends a connection reset if the limit is exceeded, it would have to drop the request instead.
     127< Yes.[[BR]][[BR]]
    78129   3) Would server-imposed response delays help?
     131< We tried to avoid suggesting measures that obviously trade performance for security.  I2P is a low-latency design, so adding artifical delays would seem a rather significant change.  If done right, yes, they would likely make the attack harder --- at the expense of making I2P slower; this risks reducing the size of the userbase and hence might do more harm then good in the end.  Tough call.[[BR]][[BR]]
    80133   4) Disallow multiples from the same /16 in the fast tier
     135< Given that our adversary controls nodes in many /16s, this suggestion would not apply in the context as we presented it, but of course might be a good idea in practice.[[BR]][[BR]]
    82137   5) Increase fast and high-capacity tier maximum sizes
     139< Not clear (time/accuracy trade-offs), and naturally at some point you'd have everyone in all tiers, so again trade-off between (to us not easily quantified) performance benefits from tiers and security issues (from non-random/biased/manipulateable) performance-based peer selection.[[BR]][[BR]]
    84141Sec 7 Conclusion:
    87144   Released March 2, installed in 25% of network by ~March 4, 50% by ~March 6, 75% by ~March 14 (source )
     146< Okay.[[BR]][[BR]]
    89148  2) Relevant changes in 0.8.4 release:
    90149    a) Prevent tunnel-building DOS by a single source. This was done in reaction to the attack.
     151< NOTE: Our DDoS involved 40 peers hitting a single peer.[[BR]][[BR]]
    92153    b) Penalize peers more due to tunnel rejections. This did not change the time constants of the capacity formulas, just changed (a + r) to (a + 2r) in the denominator of the formula in section A.1. However it may have had the effect of reacting faster to a DOS attack. This change was not made in reaction to the attack, but was previously planned and is part of a strategy to spread the traffic across more peers in the network and adjust the forumla in response to network conditions that have changed markedly in the past two years.
     155< Oh, OK, we thought it was a reaction to our activity. Removed.[[BR]][[BR]]
    94157  3) More changes to detect and prevent DOS are upcoming in 0.8.5 (scheduled for release the week of April 18) but these are not a complete solution. A fully distributed tunnel-building DDsS is difficult to prevent completely.
     159< Right, which is why we said that simply trying to prevent DoS was not the best answer.[[BR]][[BR]]
    96161 Reacting to performance changes as a "bad idea":
    97162   A bold and absolute statement not fully supported by the paper. Clearly there are steps we can take to mitigate the attack, especially by detecting and reacting to the DDoS and other items discussed at the end of Section 6. Also, the network is still very small. We aren't making absolute claims about our anonymity especially when the whole network is 2000 - 3000 routers. Yes we still have work to do to ensure anonymity while doing performance-based peer selection and there are tradeoffs involved. But don't write it off as a "bad idea". A more nuanced statement would be that anonymity vs. performance is a fundamental tradeoff, and that the speed of reaction to a peer's perceived performance change affects that tradeoff, and that perhaps i2p's reaction time is too fast for the anonymity that its users expect, and should be adjusted.
     164< That was not our statement.  We said that reacting to performance changes *quickly* was a bad idea.  This refers to the (somewhat theoretical) possibility of the entire I2P fast-tier being replaced by other peers quickly IF the adversary targets the right victims with DoS (in practice, as we have shown, the adversary may not be able to determine the correct set of victims that precisely, but your implementation has no 'speed limit' for the evolution of the tiers (ok, 45s...), which is what this is about).  If you limited churn instead (see above), the reaction would take longer, forcing the adversary to maintain the DoS for a longer time.[[BR]][[BR]]
    99166Sec. A.2 Integration value:
    100167  This isn't used in I2P for anything except a display and isn't relevant to the paper. You may also wish to remove the information about the well-integrated tier from sections 3.2.4, 3.2.5, and B.1.
     169< We'll consider this.  Thanks for confirming that well-integrated is really used for nothing.[[BR]][[BR]]