[tor-dev] Draft Proposal for BridgeDB IPv6 Support

Aaron aagbsn at extc.org
Tue Dec 6 00:38:49 UTC 2011


Attached is a draft document describing proposed changes to BridgeDB to
accommodate the new or-address spec (186-multiple-orports.txt) and IPv6 bridges.

I am especially interested in comments on sections tagged #XXX.

Thanks!

--Aaron
-------------- next part --------------
Filename: xxx-bridgedb-learns-ipv6.txt
Title: BridgeDB Learns IPv6
Author: Aaron Gibson
Created: 5 Dec 2011
Status: Draft

Overview:

  This document outlines what we'll do to make BridgeDB fully support IPv6
  bridges, and fully support IPv6 with the email, https, and bucket
  distributors.

Motivation:

  IPv6 bridges need a BridgeDB too.

What needs to change:

  There are two main tasks that must be completed for BridgeDB to support IPv6.

    1. BridgeDB must be able to parse IPv6 addresses from router descriptors.
    (Currently, BridgeDB does not recognize the or-address line described in
    186-multiple-orports.txt)
    
    2. BridgeDB must decide how to hand out IPv6 addresses.  (Currently,
    BridgeDB distributors are not IPv6 aware, and provide no support for
    distinguishing bridges by address class)

    
1. BridgeDB learns to parse or-address

    BridgeDB must learn how to parse the new or-address line from server
    descriptors. The new or-address line allows a router to specify a list of
    addresses and ports or port-ranges. 
    
    Here is the or-address specification (see: 186-multiple-orports.txt)
          
      or-address SP ADDRESS ":" PORTLIST NL
      ADDRESS = IP6ADDR | IP4ADDR
      IPV6ADDR = an ipv6 address, surrounded by square brackets.
      IPV4ADDR = an ipv4 address, represented as a dotted quad.
      PORTLIST = PORTSPEC | PORTSPEC "," PORTLIST
      PORTSPEC = PORT | PORT "-" PORT
      PORT = a number between 1 and 65535 inclusive.

    BridgeDB must now comprehend and store multiple listening addresses and
    ports. BridgeDB currently assumes that each bridge has only one listen
    address. BridgeDB must be modified to take one of the following approaches:
    
      a. Treat each ADDRESS:PORT combination as a separate bridge entity
      b. Display a subset of each bridges ADDRESS:PORT entries in a response
      c. Display all of each bridges ADDRESS:PORT entries in a response

    Given any address of the bridge you can learn its fingerprint, and use that
    to look up its descriptor at tonga and learn the rest of the addresses. so
    counting a bridge with 5 addresses as 5 bridges makes it more likely to get
    blocked by a smart adversary. but more useful against a weaker adversary.
    #XXX: thanks arma!
    # any other thoughts here? option c. seems to be the simplest to implement.

    BridgeDB should be able to mark specific IP:port pairs as blocked, and
    indicate where it is blocked (e.g. by country code). This requirement is
    complicated by the fact that or-address may specify a range of listening
    ports.

    How are IPv6 Addresses stored in BridgeDB?

      IPv6 Addresses are stored as strings, the same way as IPv4 addresses.
      #XXX: is this better than using the ipaddr.IPAddress class?

    How are Bridges differentiated by address class?

      Bridges are differentiated by the string representation of their IP
      address.

      When BridgeDB needs to make a distinction between IP address classes, the
      python module ipaddr-py (https://code.google.com/p/ipaddr-py/) will be
      used to determine address class.

2. BridgeDB learns how to selectively distribute IPv6 bridges

    BridgeDB's 3 distributors must be able to selectively provide both
    IPv4 and/or IPv6 bridges to clients. Deployment of these distributors must
    take care to mitigate reachability issues due partly to the ongoing
    transition from IPv4 to IPv6. 

    [One such issue is clients who have IPv6 support on their local network but
    the client's ISP does not; such a client may try to reach the IPv6 address
    specified by a AAAA record and fail to connect.]

    The 3 distributor types that BridgeDB currently features are:

      1. HTTPS Distributor
         
        The HTTPS distributor must be able to selectively offer both IPv4 and
        IPv6 bridges to its' clients, and BridgeDB must support both IPv4 and
        IPv6 connections.
        
        #XXX the twisted framework does not currently support ipv6. However,
        # it is possible to place BridgeDB behind a forwarding proxy such as
        # apache or nginx, which will pass the client address to BridgeDB in the
        # X_FORWARDED_FOR header. BridgeDB HTTPS distributor must be able to
        # parse the X_FORWARDED_FOR header for both IPv4 and IPv6 addresses.
        # Additionally, BridgeDB should eventually support IPv6 natively when
        # the twisted framework provides adequate IPv6 support.

        How does bridgedb determine whether to respond with ipv4 or ipv6
        bridges?

          Users select IPv4 or IPv6 bridges by visiting different URLs. An
          informational message added to the BridgeDB response will include the
          other URL. Two separate BridgeDB instances are run, one for each URL.

          Alternately, ipv6 bridges could be requested by visiting
          bridges.tpo/ipv6 or similar URL path scheme.

        Proposed Additional Hostname For IPv6 Bridges

          BridgeDB shall listen for requests on two different hostnames,
          bridges.torproject.org and bridgesv6.torproject.org.

        DNS Configuration Details

          bridges.torproject.org shall not have an AAAA record until the
          addition of the record is determined to be sound.

          bridgesv6.torproject.org shall have both an AAAA and A record.
          
          This is to avoid the confused-client scenario described above.

        How does BridgeDB know which URL was requested?

          This section describes how BridgeDB could be modified to support
          requests to both bridges.torproject.org and bridgesv6.torproject.org
          with a single BridgeDB instance.

          A single BridgeDB instance could handle requests to both
          bridges.torproject.org and bridgesv6.torproject.org by checking the
          Host header sent by the browser. The Host header is optional. In
          order to expose the selector explitely BridgeDB must check the query
          string for the following parameters:

            q=ipv4  - Request IPv4 bridges.
            q=ipv6  - Request IPv6 bridges.  

          Parameters may be repeated to select multiple classes, e.g.
          
            q=ipv4&q=ipv6  - Request both IPv4 and IPv6 bridges.

          When no parameters are set, by default BridgeDB must return addresses
          of the same class as the client. This default may promote IPv6 use
          where possible.

        How does someone end up at bridgesv6.torproject.org?

          BridgeDB should include a message at the end of its' response.  
          e.g.

            "Get IPv4 bridges https://bridges.torproject.org"
            "Get IPv6 bridges from https://bridgesv6.torproject.org"
            "You must have IPv6 for these bridges to work."
          #XXX: will users understand what this means?

        How does IPv6 affect address datamining of https distribution?
          A user may be allocated a /128, or a /64.
          An adversary may control a /32 or perhaps larger
          Proposal: Enable reCAPTCHA support by default.

        How do IPv6 addresses work with the IPBasedDistributor?
        #XXX: I need feedback on this
        # do we use all 128 bits here?
        # upper N bits? lower N bits? random or specific N bits?

        How are IPv6 Bridges actually distinguished?

          A new type of BridgeSplitter (sort of like a BridgeHolder)
          is devised; the function of which is to split bridges into different
          rings determined by a filter function.

          The filtering mechanism here is similar to BridgeDB's ipCategories
          implementation, the difference is that both the filters and ring
          names are specified at instance construction.

          The construction of a BridgeSplitter instance should be done by
          passing lists of tuples (ringName,filterFunction) to the constructor.
          This feature allows for dynamically creating filtered BridgeRings, 
          which would prove useful for constructing more complex filters, for
          example, filtering by both address class and reachability from
          specific countries.

          This implementation could increase the rate at which bridges are
          detected and blocked, although the rate could be controlled by the
          frequency that BridgeDB updates its knowledge of blocked bridges.
          
          #XXX: I have some concern about the performance of
          # filtering bridges dynamically for each response. BridgeDB should
          # learn to cache recently used dynamic filters so that the impact of
          # popular requests will be reduced, and BridgeDB does not have to
          # pre-compute or identify which types of requests will be popular.

          The implementation could look similar to the current 'subring'
          implementation; or the current 'ipCategories' implementation. Both of
          the features are implemented using subrings that hold a subset of
          the parent ring's bridges; the subset being defined by a filtering
          function.

          An accompanying Distributor based on the existing IPBasedDistributor
          shall be designed to use the above BridgeSplitter so that sorted
          Bridges are selectable by address type. Because a bridge
          may now have both IPv6 and IPv4 addresses, BridgeDB needs to take
          one of the following approaches when only a single address class is
          requested:

            a. filter addresses of the other address class from the response
            b. include the other addresses in the response

      2. Email Distributor

        The Email Distributor must accept additional new commands parsed from
        the subject or a single line in the body of an email message.

          ipv4  - request IPv4 bridges.
          ipv6  - request IPv6 bridges.

        The default action may be set in bridgedb.conf with the option
        EMAIL_DEFAULT_ADDRESS_CLASS, which must be one of 'ipv6' or 'ipv4'.  If
        the option is not given in the config, EMAIL_DEFAULT_ADDRESS_CLASS shall
        default to 'ipv4'.

        Similar to the IPBasedDistributor, BridgeDB must determine how the
        response should accommodate bridges with both address classes.

      3. Unassigned Distributor and Buckets

        BridgeDB must provide a selector to choose between exporting
        IPv4, IPv6, or both types of bridges.

        BridgeDB currently provides a way to export bucket files with
        unallocated bridges. The existing syntax provides no mechanism to
        differentiate by address class.

        Proposed new FILE_BUCKET syntax:

          A dictionary of dictionaries with the following acceptable keys and
          values.

          'filename_prefix' shall be a unique string used as the output filename
          prefix. This is string is also the key to a dictionary that contains
          the following key/values:

          'address-class' : one of either 'ipv6' or 'ipv4'
          'number'        : an integer > 0

        Users may wish to provide descriptive names,
        e.g.
        
          FILE_BUCKETS = {
                  'filename_prefix': {'address-class': 'ipv6', 'number': 10},
                  'descriptive_name': {'address-class': 'ipv6', 'number': 10},
          }

        Future BridgeDB enhancements may expand these options to include other
        filters.
        #XXX: e.g. buckets of bridges 'not-blocked-in'


More information about the tor-dev mailing list