300 likes | 468 Views
SPATor : Improving Tor Bridges with Single Packet Authorization. Paper Presentation by Carlos Salazar. Low-latency Anonymity Network Data is sent through a chain of relays known as a circuit.
E N D
SPATor: Improving Tor Bridges with Single Packet Authorization Paper Presentation by Carlos Salazar
Low-latency Anonymity Network • Data is sent through a chain of relays known as a circuit • Each packet sent through the circuit is encrypted with the keys for each successive relay, this results in a layering of encryption (like an onion) • All relays are listed in a public directory hosted by a directory authority
More Tor • Adversaries will often restrict access to Tor by blocking all of the public relays. • Unlisted relays have been created as a way to circumvent this problem • Unlisted Tor relays are called bridges • Unfortunately, Tor’s implementation of bridges has some issues…
Tor Bridges • Can be completely unlisted • Information spread by word-of-mouth • Can be distributed by the Tor Project • Valid bridges are tracked by the bridge authority • Bridge info is stored and dispensed to clients by the BridgeDB
Goals of the bridge authority/BridgeDB: • Provide clients with ready access to bridge information • Prevent an adversary from finding a large amount of bridges in a short amount of time
Problems with Bridge distribution • Does not account for attackers controlling multiple IP addresses • A relay operator can discover bridges by attempting to start a bridge connection with any non-relay node that connects to their relay. • The Bridge Aliveness attack could compromise the anonymity of a bridge operator who uses the same machine for web browsing
Bridge Aliveness Attack • Bridge Discovery • Collect a large list of bridge addresses and descriptors • Winnowing • Narrow the list down to possible bridges which may have contributed material to a website by correlating timestamps with bridge aliveness information (by polling their bridge list) • Confirmation • Use circuit clogging and timing attacks to confirm the suspected source of the website contribution
Mitigating the Bridge Aliveness Attack • Remove the close relationship between a bridge operator serving other clients and using Tor as a client (targets winnowing) • Prevent an entry relay from attempting to test whether or not its clients are bridges (targets discovery)
Enter SPATor • Single Packet Authorization + Tor = SPATor • Mitigates the risks of serving as a bridge while using Tor as a client • Obstructs the hoarding of bridge information • Makes it difficult to query bridge aliveness
SPATor requires an extra per-bridge key be provided to users by the BridgeDB • Each key is only valid for a limited amount of time • Clients must prove knowledge of the key to connect to a bridge • The key is used in the SPA protocol (SilentKnock), so that failed authorization attempts do not reveal bridge aliveness
SilentKnock • Conceals the existence of a particular service until the client sends a special TCP SYN packet to the target IP address and port. • Uses covert channels in TCP/IP headers to hide a Message Authentication Code (MAC) in the TCP SYN packet • The MAC is in the lower 3 bytes of the ISN value and the lower byte of the TCP timestamp
Adversary for the Bridge Aliveness Attack • Remote non-global adversary • Can query a large number of bridge descriptors • Can perform bridge aliveness tests by trying to connect to bridges as a client • Has access to a website timestamps of pseudonymous contributions by a bridge operator and can control some content on that site
The Other Adversary • More Capable than the first adversary • Active • Has full control of the local network in which the client is present • Can monitor, inject, replay, shape, and drop packets within the local network • Cannot view or control anything outside the local network
The SPATor Protocol • Uses a pre-shared key to generate MACs • MACs are used to determine whether or not a bridge connection request is legitimate • SPATor affects three parts of a Tor bridge’s life cycle • Bridge registration • Bridge request • Client Connection
Bridge Registration • New bridges in Tor provide the bridge authority with: • IP address • Port • SPATor Adds: • SeedKey (a random 256-bit value) • Used to generate a short-lived MACKey • Key update frequency
Client Connection • Client generates a MAC and embeds it in a TCP SYN packet • Made with the current time and header Data of the inclosing TCP SYN packet • Keyed with a MACKey recently obtained from a bridge authority • Truncated to 32 bits (from 256) • Client sends the special TCP SYN packet
Unlisted Bridges • SPATor is designed to work with unlisted bridges too • The bridge operator must distribute information to clients somehow • Share the SeedKey & update freq out-of-band • Share only the current MACKey • (maybe) offer a SeedKey or MACKey unique to each client’s address
Attacks on SPATor • Bridge Aliveness • Only works if the attacker has access to a large number of IP addresses in a short amount of time • All IP addresses must be distinct /24 network addresses • Determining aliveness by guessing the correct MAC could take over 715 million guesses • A firewall should blacklist an IP that failed too many times
Bridge Client Detection Attacks • A Passive Adversary can: • Learn that a client’s connection timed out • See that the client established a TLS connection • Distinguish Tor traffic from other types of traffic (this is not changed by SPATor)
Active Adversaries • Can attempt to use information from the client to connect to the bridge • Replay a previously seen TCP SYN packet • Requires the adversary to assume control of the client’s IP address • Must be done before the MAC in the SYN packet becomes stale • Hijack the TCP connection as it is being sychronized and act as a Man-in-the-middle
How to prevent replay and MITM • After a TLS connection is established, the client sends the full 256 bit MAC • The bridge should act like it’s something else until this happens • This isn’t currently possible because the TLS certificates for Tor bridges are distinguishable from other TLS certs.
There is a (very small) delay imposed on connections made with SPATor • Just like there is with SilentKnock • This probably doesn’t matter • Conclusive detection of SPATor use based on timing would require that the adversary: • Is fairly close to the network • Has collected a large amount of data • An adversary could prevent SPA from succeeding by altering the sequence numbers on every packet. • This is impractical for most large active firewalls • This doesn’t necessarily provide a way to prove that SPATor is in use, because this will always break some IP extensions
Future Work • Compatibility with other OSs (not just Linux) • SPATor does not work behind a NAT • Handle SYN retransmits correctly • Add changes to bridge authority and BridgeDB • Prevent other Aliveness checks • Pluggable transports?