Older proposals in need of discussion: 149 ("Using data from NETINFO cells")

[Nick Mathewson <nickm@xxxxxxxxxxxxxx>]

Last August I sent out proposal 149, but it never got much discussion
on this list.  (Looking at the archives, I can't find any.)  Here's
the paragraph that hasn't much been implemented:

   We need to think about attackers here.  Just because a router tells
   us that we have a given IP or a given clock skew doesn't mean that
   it's true.  We believe this information only if we've heard it from
   a majority of the routers we've connected to recently, including at
   least 3 routers chosen at random.  Routers only believe this
   information if the majority includes at least one authority.

This isn't so good:

   - It's less secure than we would like for clients.  If I can block
     your network connection selectively,  I can make you only connect
     to routers I control, who can lie to you about your IP and the
     time.

   - It's sometimes useless for clients.  If a client's timestamp is
     in the distant past or future, it may not believe in the quality
     of _any_ router info, and so not actually try to connect to
     anybody and learn the time via a NETINFO cell.  Or it might fail
     because the other side's x509 certs will look invalid.  Or it
     might fail because its own x509 certs will be invalid.

   - It's less functional than we want for servers.  Most servers
     don't go out of their way to talk to the authorities, and so they
     don't often get time/ip info this way.

So I propose that we amend 149 as follows:

   1) Clients and servers alike should believe the time from a netinfo
     if all of the following hold:
      - They have heard similar times from a majority of the servers
        they have connected to.
      - The majority contains at least 3 servers not in the same
        family.
      - A majority of the connections that the node has attempted were
        successful.       

   2) Clients don't care about their own IP.  Servers should consider
      testing an IP given from a netinfo cell if it meets the criteria
      for time differences in 1) above, and they don't already have a
      working IP.

   3) We should stop rejecting connections entirely because of expired
      or no-longer-valid x509 certificates.  Instead, we should allow
      the connection to continue, but not believe the identity of the
      other side.

   4) We should check the code to see what clients do with seemingly
      expired or future consensus documents.  Clients should always
      provisionally accept a consensus if it is newer than any
      consensus they have.  They should then contact nodes until they
      know what time it is.

Thoughts?
-- 
Nick