A Survey Anonymity and Anonymous File-Sharing Tom Chothia (Joint work with Konstantinos Chatzikokolakis)

Outline of Talk The theory of anonymity. Designs for anonymity. Anonymous file-sharing software. Some early results from the analysis of file- sharing software.

Introduction This is a light weight introduction to anonymity: –Definitions –Design –Real Systems –Some Analysis of the Systems Next week you will see more on the technical definitions and modeling with process calculi.

The Theory of Anonymity Anonymity means different things to different users. The right definitions are key to understand any system. “On the Internet nobody knows you’re a dog”

The Theory of Anonymity Anonymity is a difficult notion to define. –Systems have multiple agents –which have different views of the system –and wish to hide different actions –to variable levels. Sometimes you just want some doubt, sometimes you want to act unseen.

The Theory of Anonymity In a system of anonymous communication you can be: –A sender –A receive / responder –A helpful node in the system –An outsider (who may see all or just some of the communications). We might want anonymity for any of these, from any of these.

Example: Anonymous File- Sharing One node sends a request for a file (sender) Other nodes receive this request (the nodes) Maybe one of the nodes replies with a file (receiver/responder). The attacker may be any of these or an outside observer. ?

Example: Anonymous File- Sharing The user may wish to hide –that they are offering files –that they are taking part in data transfer –that they are running the software at all. The user may want to have plausible deniability or go complete unnoticed. ?

The Theory of Anonymity There are many definitions. Some are “too weak”, –Delov-Yao style “Provable Anonymity” Some are “too strong”, –Information flow. There will be more on these definitions next week.

Levels of Anonymity Reiter and Rubin provide the classification: Beyond suspicion: the user appears no more likely to have acted than any other. Probable innocence: the user appears no more likely to have acted than to not to have. Possible innocence: there is a nontrivial probability that it was not the user.

Beyond suspicion All users are Beyond suspicion: Prob Users ABCDE

Beyond suspicion Only B and D are Beyond suspicion: Prob Users ABCDE

Beyond suspicion Now, only B is Beyond suspicion: Prob Users ABCDE

Probable Innocence All users are Probably Innocence Prob Users ABCDE 50%

Probable Innocence All users are Probably Innocence Prob Users ABCDE 50%

Probable Innocence All users are Probably Innocence Prob Users ABCDE 50%

Probable Innocence All users are Probably Innocence Prob Users ABCDE 50%

Probable Innocence All users are Probably Innocence Prob Users ABCDE 50%

Example: The Anonymizer An Internet connection reveals your IP number. The Anonymizer promise “Anonymity” Connection made via The Anonymizer. The Server see only the Anonymizer. S ? The Anonymizer

Example: The Anonymizer The sender is Beyond Suspicion to the server. The server knows The Anonymizer is being used. If there is enough other traffic, you are Probably Innocence to a global observer. The global observer knows you are using the “The Anonymizer” There is no anonymity to the “The Anonymizer”

Example: The Anonymizer From the small print: … we disclose personal information only in the good faith belief that we are required to do so by law, or that doing so is reasonably necessary … … Note to European Customers: The information you provide us will be transferred outside the European Economic Area

Summary: The Theory of Anonymity There are many agents in a system each of which have different views. There are a number of different actions. We need to define the level of anonymity an user has when performing a certain action, given the attacker’s view of the system.

Outline of Talk The theory of Anonymity. Designs for anonymity. Anonymous file-sharing software. Some early results from the analysis of file- sharing software.

Theoretical Designs for Anonymity We have seen an example of anonymity from a Proxy. In Friend-to-Friend networks: –nodes have fixed neighbours, –only direct neighbours know IP addresses, –nodes act as proxies for there neighbours. Anonymity to your neighbour is by trust or by claiming you are just acting as a proxy.

Ants The Ants protocol is for ah- hoc networking. Each node has a pseudo ID. A node broadcasts a request, labeled with its own ID. Nodes record IDs it receives over each connections. A

Ants If another nodes wishes to reply to the request: It sends packets labeled with its own ID The packets are sent along the most used connection for the to ID. A

MIXes MIXes are proxies that forward messages between them A user contacts a MIX to send a message The MIX waits until it has received a number of messages, then forwards them in different order

MIXes It is difficult to trace the route of each message. Provides beyond suspicion S-R unlinkability even w.r.t. a global attacker. Messages have to be delayed (can be solved with dummy traffic). More complicated when sending series of packets

Onion Routing Messages are routed through a number of nodes called Core Onion Routers (COR) The initiator selects the whole route and encrypts the message with all keys in reverse order Each node unwraps a layer (onion) and forwards the message to the next one {{{m} k3 } k2 } k1 {{m} k3 } k2 123 {m} k3 m

Onion Routing Each node only learns the next one in the path Can be used together with MIXing. End-users can run their own COR –Better anonymity or use an existing one –More efficient –User's identity is revealed to the COR

Crowds A crowd is a group of n nodes The initiator selects randomly a node (called forwarder) and forwards the request to it A forwarder: –With prob. 1-p f selects randomly a new node and forwards the request to him –With prob. p f sends the request to the server server

Crowds Beyond suspicion w.r.t. the server Some of the nodes could be corrupted. The initiator could forward the message to a corrupted node. Probable innocence w.r.t. a node (under conditions on the number of corrupted nodes).

Dining Cryptographers Nodes form a ring Each adjacent pair picks a random number Each node broadcasts the sum (xor) of the adjacent numbers The user who wants to send a message also adds the message The total sum (xor) is: r 1 +r 2 +r 2 +r 3 +r 3 + r 4 +r 4 +r 5 +r 5 +r 1 +m = m r1r1 r4r4 r5r5 r3r3 r2r2 r 1 +r 2 r 5 +r 1 r 4 +r 5 r 3 +r 4 r 2 +r 3 +m

Dinning Cryptographers It's impossible to tell who added m. Beyond suspicion even w.r.t. to a global attacker. Very inefficient: everyone must send the same amount of data as the real sender. More info in Catuscia's talk

Mutli-casting Broadcast the message to the whole network. Provides beyond suspicion for the receiver. No anonymity for the sender. Multicasting is an efficient technique for broadcasting messages. but very inefficient to send just one message.

Spoofed UDP IP packets on the Internet contain the IP address of the sender This address is not used by routers, only by higher-level protocols such as TCP UDP does not use this address A random address can be used instead to provide sender anonymity Method prohibited by many ISPs

Summary of methods

Outline of Talk The theory of anonymity. Designs for anonymity. Anonymous file-sharing software. Some early results from the analysis of file- sharing software.

Mute Mute is an open source project based on the Ants protocol. Mute uses a complicated 3 stage time-to-live counter that allows an attack. In Mute all the probabilistic choices are fixed when a node starts. This protects against statistical attacks.

Ants Ants is also an open source project based on the Ants protocol. There is a probabilistic change of dropping a search request. Avoiding some attacks but giving little control over searches. Ants send most reply packets over the best route but sends some by other routes. This is done for efficiency by it also stops some attacks by inside nodes.

Mantis Mantis is an academic project that uses the Ants protocol. But the sender may make its IP address public and receive the file by address spoofed UDP. Hence only the responder is anonymous, but the system is very efficient.

Anonymous Peer-to-Peer File- Sharing (APFS) APFS is based on Onion Routing Volunteer nodes act as proxies. Centralised servers store an “onion routes” for files. Searching is carried out by asking a server for an onion route for a file. Pro: Secure system, Con: Hard to set up and maintain.

Freenet and Free Haven There are a number of “anonymous publishing system”. For example Freenet and the MIX based Free Haven. These systems make the original author of a file anonymous, not the responder. Nodes will often cache files.Therefore you can “trick” a node into storing and “offering” a file.

Waste Waste is a friend-to-friend network. It is designed for small groups (under 50 nodes). The sender and receive are known to network insiders, but anonymous to an outside attacker. Dummy traffic traffic is sent between nodes whenever they are idle.

Tor Tor is an anonymous transport layer. It does not implement a file-sharing but file- sharing software can be run on top of it. Tor implements onion routing without MIXes. Its possible that a program run on top of Tor will reveal its IP address.

Some Other Systems AP3Crowds Mislove et al. Entropy Freenet entrop.stop1984.com GNUnetMIXes gnunet.org I2P Onion routing Nodezilla Freenet Napshare Ants napshare.sourceforge.net SSMP Secret sharing Dingledine et al. & onion routing There are others!

Outline of Talk The theory of anonymity. Designs for anonymity. Anonymous file-sharing software. Some early results for the analysis of file- sharing software.

Goals for Anonymous File- Sharing using Ants The attacker is a node in the network and must discover the pseudo ID of its nieghbours. Sender (requesting files) is Probable Innocence to nodes and responder. Responder (offering files for download) is Probable Innocence to nodes and sender.

The Model The model of the network is a connected weighted graph. The weights are the times it takes for a message to travel along that connection. Travel times are fixed. A single attacker, no timed-based attacks. No time-to-live counter.

The Attackers View Its connections and the real addresses of the nodes each of these connections leads too. The pseudo IDs from the messages it has seen. For each pseudo ID, the ordered over which the attacker receives message The ``to'' and ``from'' pseudo address of all the messages past across it.

The Attackers View The attacker may also send messages. It can form message out of its own random values, its own address or any address is has seen. In particular, it can send messages the “wrong way”.

Time-Based Attacks The quickest reply along any connection will come from the direct neighbour. The attacker may try random request, and note the reply times. The pseudo ID with the fastest reply time over any connection is assume to be the neighbour. If a node shares any files at all, it is not anonymous to its neighbour.

Result Assuming no timed-based attacks, there is still a problem: The attacker might just see one pseudo ID over a connection. Or have a unique pseudo ID “bounced back”. i.e., anonymity depends on how the nodes are connected.

Result One node on its own is not anonymous. Only node one node fastest along a connection is not anonymous. N A

Result Active attacks allow more discrimination. A receives two IDs first over each connection. But N3 and N4 are bounced back Therefore the attack can identify N1 and N2. N1 A N3 N2 N4

Result If we assume that the attackers neighbours might never share files then Ants is anonymous. Otherwise: –The Ants protocol can be broken by a timed attack. –If any connection is not used by at least two different pairs of nodes to communicate then the nodes on this connection are not anonymous to each other.

Protected Addresses Attacker can make a message with another node’s pseudo ID as the from address. This lets it disrupt communication. We can generate a key pair and use the authentication key as the pseudo ID. The sender signs the message ID. Hence the attacker cannot fake messages.

Other Kinds of Attack Global Attacker System Membership Time-to-Live Attacks (Mute, Mantis) Multiple Attackers (Mute) Statistical Attacks (MIXes) Forced Repeat (Crowds) Nodes Joining and Leaving Denial of Service (Mute)

Outline of Talk The theory of Anonymity. Designs for Anonymity Anonymous file-sharing software Some early results for the analysis of file- sharing software.

Further Work Ants Protocol: –Finish formal model and testing, –Time delays, –Deciding when a network is safe, –MIXes for file-sharing. General purpose formal methods for anonymous systems.

Questions?

Example: Anonymous File- Sharing The user may wish to hide –that they are offering files –that they are taking part in data transfer –that they are running the software at all. The user may want to have plausible deniability or go complete unnoticed.

Example: Anonymous File- Sharing The user may wish to hide –that they are offering files –that they are taking part in data transfer –that they are running the software at all. The user may want to have plausible deniability or go complete unnoticed.

Forced Repeat Attack: Crowds

Time-to-live Attack: Mute

Time-to-live Attack: Mantis