Changes between Version 52 and Version 53 of doc/FAQUnanswered


Ignore:
Timestamp:
Apr 23, 2010, 4:48:48 AM (9 years ago)
Author:
trac
Comment:

--

Legend:

Unmodified
Added
Removed
Modified
  • doc/FAQUnanswered

    v52 v53  
    88= Unanswered FAQ Questions =
    99
    10 1) '''Why do I keep getting messages telling me that my clock has just jumped ahead and that my circuits will be assumed broken? ''(eg.
    11 Oct 02 10:14:53.619 [notice] Your clock just jumped 1056 seconds forward; assuming established circuits no longer work.'') I've got a cron job to sync the time every eight hours and it's never out by more than a second.'''
    12 
    13 -- On my system, this happens when Vidalia gets into trouble. It seems that when tor and V communicate, tor can wind up waiting for V to respond, or for the V process to be killed.
    14 
    15 
    16 2)'''If an attacker has access to past logs of ISP and any given visted site, does the prng of the tor client allow the attacker to guess which circuit it used next?'''
    17 
    18 Tor uses cryptographically strong random numbers provided by OpenSSL when choosing nodes to use in a circuit. How OpenSSL implements this is operating system specific. If there's a weakness in Tor's method of choosing nodes, it probably isn't in the random number generator.
    19 
    20 *** Hmm, if the prng is deterministic, and you can narrow one result of calling it by knowing what host was chosen, can you, knowing the algorithm, however good it is, thereby narrow the result of the next call to it?  This question might be a bit ignorant; for example the prng might use other data on the client computer instead of following an algorithm to return the next item.
    21 
    22 *** Thanks for your comments, BTW, as a lot of us are wondering the answers to these FAQU.
    23 
    24 *** It has nothing todo with determinism. It is a matter of predictability. Without using real entropy, everything done on conventional computers should be deterministic (try teling that to people who use Windows ;). However, the part that we are interested in is whether someone can predict what Tor is going to next choose.
    25 
    26   Cryptographic number generators have the property that they (shouldn't) give up their internal state by their external outputs quickly. So, unless the attacker gets to see a lot of outputs - node choices - without reseeding from real entropy they are pretty much screwed provided the cryptograpphic prng isn't broken.
    27 
    28 
    29 '''Is there a signal i can send the tor client to tell it to switch circuits immediately?'''
    30 
    31 You can connect to Tor's control port and send "authenticateCRLFsignal newnymCRLF" where CRLF is a carriage return line feed pair.
    32 
    33 '''Why does Firefox/Privoxy/Tor return Privoxy 404 pages so frequently -- almost every time -- when properly configured, even on sites like google.com?  How to mitigate?  The tor process is running fine.'''
    34 (Votes: 2)
    35 
    36 The first problem is that Privoxy doesn't retry in case of
    37 DNS errors. It shows the 404 no such domain message right away.
    38 The second problem is that some browser cache Privoxy's error
    39 messages and Firefox is one of them.
    40 
    41 The Privoxy patch described at
    42 http://www.fabiankeil.de/sourcecode/privoxy/
    43 lets Privoxy retry in case of connection problems
    44 and makes sure, the browser doesn't reuse a cached
    45 error message.
    46 
    47 '''For that matter, why is DNS the main failure mode?  Who is timing out and why?  Can Tor (1) change the timeout, (2) deprecate bad servers, or (3) cache DNS locally so it doesn't have to make a long, slow, failure-prone DNS lookup every time?'''
    48 
    49 You should be sending hostnames to Tor over SOCKS4a or SOCKS5. In that case, the Tor exit node will resolve the hostname before making a connection for you. Unless the exit node is misconfigured, there shouldn't be a problem with DNS resolves timing out.
    50 
    51 *** Actually, for me this happens most of the time.  I am sending through Privoxy, and tried both sockses.  So why would I be hitting so many exit nodes that FREQUENTLY time out on DNS?  Firefox 1.5, most recent stable Tor.
     10'''Why does Firefox/Privoxy/Tor return Privoxy 404 pages so frequently -- almost every time -- when properly configured, even on sites like google.com?  How to mitigate?''' (Votes: 2)
     11
     12Maybe it's just my Mac but I find that Tor has a habit of dying and staying dead which leaves privoxy and firefox non-functional. The work around I've found is that you can open Terminal directly and just start Tor by typing "Tor". I also find that if I start getting a lot of 403's or 504's then I can just kill the Tor instance running in Terminal and re-start it. More often than not, that fixes the problem.
    5213
    5314'''Is the reason that gmail rarely works: gmail, tor, privoxy, firefox, your own bandwidth/latency, tor's bw/latency, or some combination?  Is it fixable?'''
    5415
    55 *** For Gmail for me, it's even worse; I have to try 5 or 6 times before I get a page.
    56 
    57 -- I know that improvements to DNS handling are due in 1.2.x-final (see bug #364).
    58 
    5916'''Why is the argument against more than 3 hops that both-ends attacks are the enemy?  Wouldn't it be better to have more than 3 if the enemy cannot mount a both-ends attack?'''
    6017
    61 This is two questions, really.
    62 
    63 To understand it better, think from the perspective of an attacker. They have a choice: they can make a both-ends attack, an attack comprising the hops and both.
    64 
    65 Now, at this point you need to think what attacker you are trying to defend against. The NSA? Sorry, your screwed - find something else. A big company? Some rogue ISP's? Ok, now that's something that Tor may be able to defend against...
    66 
    67 Now, what information are you - the attacker - trying to find? Who is talking to who? Content of some transmissions? Both?
    68 
    69 Well, for finding content of transmissions your best way is to just listen in on some exit nodes.. or create some malicious ones. That's not the attack threat we're trying to defend against, then.
    70 For finding who is talking to who? Ok, let's say that's what we're trying to find out.
    71 
    72 From this perspective, what is needed to mount a both-ends attack? Listening on both ends. If there is some mitigation technique used - like random timing - compromising those both ends could come in handy (at this point, more hops would really be useful). But there isn't - AFAIK - and so we shall suppose that listening on both ends is enough.
    73 
    74 Now, what is the alternative option? Listening to all the hops AS WELL? Oh dear, that's a little harder.. and what does it bring us? Well, we can get slightly more accurate results - maybe - as we can more closely correlate. This, of course, gets yet better if we compromise some of those hops.
    75 
    76 The key word here is slightly. Provided the number of hops is a constant - so you can get a reasonably accurate latency estimate - then, over a longer enough period of time, you will be able to correlate and get your results.
    77 
    78 The key word here is slightly. Provided the number of hops is a constant - so you can get a reasonably accurate latency estimate - then, over a longer enough period of time, you will be able to correlate and get your results.
    79 
    80 So, yes, the hops will make it slightly more secure. But only slightly. And yes, in some cases, it may be really useful.. but in those cases, you probably should be using something with much stronger anonymity than tor (if someone knowing what your saying is unacceptable, don't use tor.. not, at any rate, on it's own).
    81 
    82 Now, then, what are the disadvantages to more hops? More bandwidth is used. Latency is increased.
    83 
    84 Despite this, it's possible that having number of hops as an easily configurable option is not a bad idea... I would guess that there is an excess of middle-man nodes with the recent draconian laws in some parts of the Western oh so free world that make people interested in privacy yet at the same time too scared to actually stick their neck out. In this case, Tor as a network would likely not lose much by doing that.
    85 
    86 *** I ommitted elaboration of the cases where more hops would be really useful (I think I forgot about it..). These cases are generally when traffic originating from an exit node is wanted to be tracked down and a response can be made quickly, but does not have global observer capabilities. In that case, in order to find the entry node, going through each hop is the only sensible solution in a network with > 50 servers. How much you would gain from extra hops is difficult to answer.. it would depend, I think, on how often circuits rotate and the probability of a hop being out of the grasp of the attackers. Hops are, I think, probably most useful when you - or your data - specifically are being targeted..
    87 
    8818'''How can I be sure that sending DNS through tor doesn't get spoofed sites?'''
    8919
    90 Use a method for authenticating the site you want to connect to, such as ensuring an SSL certificate is associated with the domain name and signed by a trusted authority, or verifying an SSH fingerprint out-of-band. Hidden services on Tor are not subject to this problem because their .onion names are self-authenticating as hashes of their public keys.
    91 
    9220'''How does tor relate to ipv6 and how should typical applications handle ipv6 if they use tor (or tor via Privoxy)?'''
    9321
    94 
    95 ---- /!\ '''Edit conflict - other version:''' ----
    96 Like a dog talking to a quasar... I never was good with similes.
    97 
    98 Tor carries TCP data (does it? Maybe it just carries some data that is then transported over TCP... I know, for instance, that it also contains IP and port... TCP doesn't) over another layer - currently IPv4. There is work to make Tor work with IPV6, but I am not aware of it being completed.
    99 
    100 As for how an application would interface with it -- depends. It could use SOCKS; in that case, I think it would have to use SOCKS5 in the event of addressing IPv6 (rather than domains). This could be hacked around by adding a .ipv6 domain - nasty.
    101 
    102 Privoxy? Same as currently, if it handles it..
    10322
    10423'''What version of libevent should I be using?'''
     
    13958'''Can Tor be used in a network that has NO DEFAULT ROUTE?  The only access method from this network is to use a traditional proxy.  Is there a way to chain proxies so that TOR requests are sent outbound via the standard proxy? (votes: 1)'''
    14059
    141 Maybe. If you can get some routes for the Tor servers, then that of course is great.... assuming it must go through the proxy, however, it will need to support sending Tor requests.
    142 
    143 Assuming it does.. AFAIK, Tor doesn't have built-in support directly for this. So, you'll have to make it look to Tor like it really can connect .. err.. normally.
    144 
    145 I can see several ways of doing this.
    146 
    147 You could overload the network commands and have them actually connect to the proxy.. that's easy but a hack.
    148 
    149 You could try doing something with OpenVPN; personally, I don't have any experience with it... but I'm guessing you could do something neat with it (I remember someone setting up an OpenVPN with Tor being used to route things.. so it supports socks, I think, in some way. In that case, assuming the proxy is socks... all done :) Oterwise, you could write an interface to make it socks).
    15060
    15161
     
    17888'''Can anonymity be broken if all Tor servers in the chain are compromised/malicious and so are keeping logs to trace the chain?'''
    17989
    180 Yes. Indeed, if all the servers in a circuit are compromised then they need not even be communicated with.. the entry node can decipher for all the (possibly even non-existent) nodes. In order to maintain a superficial view of anonymity, it would probably be good to forward it to the exit server however.
    18190
    18291'''What system resources does a TOR server use?  The FAQ already dicusses memory a bit.  What about CPU?  Encryption is CPU-intensive.  Specific question I'd like answered: I'll be setting up a TOR node bandwidth-limited to about 256kbps (half my upstream bandwidth).  Will an old 300MHz G3 Mac easily handle this, or will a faster processor be needed? How 'bout a P90?  Presumably, TOR's disk usage and I/O is minimal.''' (Votes: 1)
     
    280189
    281190
    282 Well, you don't appear to allow exits on HTTP ports so I don't know why they blocked you. Fascists? [Ed. Fascists put their belief into the state... I doubt Slashdot does. Really, it's authoritarian]
    283 
    284 Anyway, you should probably block their IP rather than ports that happen to be theirs ;) (which don't exist - surprised tor let you do that..)
    285 
    286 
    287191'''What is the significance of the changes in the Bittorrent Torify HOWTO?'''
    288192
     
    292196
    293197Also, the same page now mentions a technique of using Tor to connect to the tracker only, as opposed to the peers, by including the line --tracker-proxy 127.0.0.1:8118: on the command line. However, I see no documentation of this option in the btdownloadcurses client and I find it a bit suspicious that the format of this option uses a hyphen rather than an underscore as all the other command line options that are listed as being compatible with btdownloadcurses use underscores to separate options with two words such as --check_hashes <arg> or --report_hash_failures <arg>. Is that a typo or an undocumented option that just happens to deviate from the naming convetion of all the other options?
    294 
    295 '''How do you start and stop Tor and Privoxy in OS X (Panther) if you did not install the startup script? (needs to be added to installation instructions)'''
    296 
    297 '''How do you configure the proxy if you are using Tor and Privoxy in OS X (Panther) with a router's firewall and the built-in OS X firewall, e.g. when using Wi-fi to connect to wireless router?  (needs to be added to installation instructions)'''
    298 
    299 '''What to do (troubleshooting) if browsing slows to a crawl with Tor and Privoxy running in OS X?'''
    300 
    301 
    302 '''I am running a Tor server on one computer on a network. Can I stop the other PCs on the same network from being k-lined on QuakeNet?'''
    303 
    304 '''Would it make sense to support binding to multiple ports in Tor server (e.g. to bind to ports 443, 22, 5190 etc.) for clients behind _really_ restrictive firewalls? If this was implemented one day, maybe you could also support binding to multiple specific IP addresses on multihomed servers?'''
    305 
    306198
    307199----