[tor-bugs] #18319 [Core Tor/Tor]: Exclude relays that don't match pinned RSA/Ed key pairs

[Tor Bug Tracker & Wiki]

#18319: Exclude relays that don't match pinned RSA/Ed key pairs
-------------------------------------------------+-------------------------
 Reporter:  teor                                 |          Owner:  andrea
     Type:  defect                               |         Status:
                                                 |  assigned
 Priority:  High                                 |      Milestone:  Tor:
                                                 |  0.3.0.x-final
Component:  Core Tor/Tor                         |        Version:
 Severity:  Normal                               |     Resolution:
 Keywords:  tor-ed25519-proto, nickm-            |  Actual Points:
  deferred-20160905                              |
Parent ID:                                       |         Points:  1
 Reviewer:                                       |        Sponsor:
                                                 |  SponsorU-can
-------------------------------------------------+-------------------------

Comment (by teor):

 Replying to [comment:19 nickm]:
 > ...
 >
 > When we turn on pinning, the most recent journal entry will rule.  So a
 relay will only be excluded from the consensus if its most recently pinned
 Ed25519 key is not the one it uses.  So if somebody switched Ed keys once
 a few months ago, they won't get penalized here.  This only affects them
 if they are switching frequently, or if they switch keys again.
 >
 > The rule for relays becomes:
 > {{{
 > Always use the same Ed25519 identity with the same RSA identity.
 > }}}
 > So, don't switch one unless you also switch the other.  If you lose one,
 don't try to retain the other.
 >
 > Sebastian says:
 > > ...
 > > How will this all work, by the way? My key pinning journal goes back
 one year and has more entries than what is written above, including more
 than just the dirauth above.
 >
 > Once key pinning is turned on, an authority will believe the latest
 entry for any given RSA key.  They will not accept a descriptor signed
 with that RSA identity key unless it also has the provided Ed25519
 identity.  So it only affects the voting, not the consensus.
 > ...

 I see from the manual that AuthDirPinKeys is set on a per-authority basis,
 so it only affects that authority's votes (and so it's not like a
 consensus method, where every authority uses it at the same time).

 '''Activation Timing'''

 What if I run a relay that changes ed keys during the changeover?

 If authorities A, B, C, D set key pinning at hour 1,
  & authorities E, F, G, H set key pinning at hour 2,
 then I have a different ed key pinned on some authorities compared to
 others.

 I guess I need to regenerate both RSA & ed keys in this instance.

 '''Keeping State'''

 Will authorities need to back up their key pinning file?

 If an authority is restored with an empty pinning file, it will regenerate
 its key pinning file based on the descriptors it sees at that time, and
 those descriptors could be different after the restore. (But the other
 authorities will anchor the pinning, if a majority keep their files.)

 '''Test Network / Testing'''

 I've just set AuthDirPinKeys on some of the authorities in the test
 network, and asked the other operators to do the same. It seems to work
 fine. But we don't have any current mismatching or RSA-only relays, so
 this is not as good a test as it could be.

 (It also works fine in chutney, but I'd like to try to match the public
 dirauth options in chutney going forward, see #20513.)

 '''Requiring Ed25519'''

 Also, what are we going to do about `DISABLE_DISABLING_ED25519`?
 It's currently `#undef`, which means that a relay can drop its ed25519 key
 whenever it wants.
 When are we going to turn it on? When 0.2.5 is no longer recommended?

--
Ticket URL: <https://trac.torproject.org/projects/tor/ticket/18319#comment:23>
Tor Bug Tracker & Wiki <https://trac.torproject.org/>
The Tor Project: anonymity online