With so much hype about the 'darknet' and 'tor', i still don't see how it can be completely anonymous if it is implemented in the application layer?

If a webserver, using Nginx or something similar, logs access to that server, then irrespective of how many times a request is bounced around it should still be traceable providing you have the access log of the destination server.

And my understanding is that retaining access logs is a legal requirement of at least some industries and countries currently, and could easily be enforced everywhere.

So unless the tor network used servers that refused to log access, then the tor network is not anonymous?

3 Answers 3


What you're talking about is a "Global Passive Adversary" (GPA): An adversary who can see everything all of the time (in your case, read all of the logs).

Tor cannot protect against a GPA, there aren't many (if any?) low-latency anonymity networks that could defeat a GPA. Although there are high latency anonymity networks that absolutely can.

First, you have to cosnider what information the logs contain, second where they are being logged and third who has access to the logs.

Onion routing derives from a form of high latency network called a Mix Network (for a well presented rundown of the history and evolution of mix networks, see this lecture by Tor developer Isis Lovecruft on anonymity networks). The idea is based on indistinguishability, each node has many messages coming in. It mixes them, and sends them off to other relays. An outside observer cannot determine how the mixing went, and thus which input message corresponds to which output message. Infact, in the case of high latency mix networks if only one relay in the chain isn't malicious then it will still be impossible to link sender to destination.

Tor is different, because it is a low latency network which means it's vulnerable to traffic analysis in ways that high latency networks aren't. If an adversary has sufficiently complete traffic logs for both your entry and exit from the network then they could determine who is talking to who.

However, if the adversary doesn't have access to sufficiently detailed logs or network traffic from both entry and exit, they will not be able to connect the input to output.

So, under many circumstances and many adversary models Tor can indeed provide anonymity as an application layer protocol.

Note also that while the Tor network is formed as an overlay network in the application layer, what it carries is transport layer traffic.

  • I'm sorry for a dumb, but right-placed question: where exactly, but except the cryptography basics and museums - where have you seen a passive adversary nowdays?
    – Alexey Vesnin
    Commented Oct 6, 2017 at 18:22
  • 1
    Government and law enforcement adversaries. Have you actually read any indictments from the last 10 years or so? The reason adversaries are using active attacks is because passive doesn't work very effectively, which is exactly what I just explained. If you want to talk about CNE then you're talking about end user software, not the suitability of the network protocol.
    – cacahuatl
    Commented Oct 6, 2017 at 19:16
  • as far as I know, LEO's and governments are using and installing active technologies for last 5+ years, including DPI - and that aims the protocols specifically. End-user software - I agree, it's a very different question and not related. For example, here in Russia government illegally introducing very active DPI and traffic inspection techniques...
    – Alexey Vesnin
    Commented Oct 7, 2017 at 14:43
  • Or do you mean GPA = "Eavesdropper"?
    – Alexey Vesnin
    Commented Oct 7, 2017 at 14:44
  • DPI isn't active, it's passive.
    – cacahuatl
    Commented Oct 9, 2017 at 3:31

there's NO absolute. Anywhere. In anonymity included. There's no "unhackable systems"/algo's/networks - it's just a question of complexity

  • Thanks. that is what I expect. could you elaborate on complexity?
    – Zach Smith
    Commented Oct 4, 2017 at 15:12
  • no problem! please describe your vectors of concern - I'll be more specific about them. Just name it! ;)
    – Alexey Vesnin
    Commented Oct 4, 2017 at 17:13

When a request is "bounced around", Tor doesn't retain the originating identification. So the exit node (the one that contacts a public web server) has no idea where the request came from. And all the web server knows is that a request came from a specific Tor exit node.

Now, if you're doing this with Facebook or your bank or something and you actually log in to the web site over Tor, then sure, they're going to know it's you.

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .