Also update it with actual decisions and parameters.

George Kadianakis authored 11 years ago and Nick Mathewson committed 11 years ago

Also, add an example of the SOCKS parameters.

Spec confusion pointed out by atagar in #9588.

These are mainly about how to normalize requests and responses, plus a
few security notes.

Also, make it a bit more extensible, in case we do get multi-packet
responses working in the future.

This was originally a DNSSEC proposal, but the round trip issue makes
it look as though this proposal is not sufficient to implement DNSSEC
for practical use.

In our tor-dev@ discussion Karsten mentioned that we could request at most 96
descriptors at a time when polling by their fingerprints...

https://lists.torproject.org/pipermail/tor-dev/2013-June/005005.html

This is a hardcoded limit in tor, so noting it in our spec...

https://gitweb.torproject.org/tor.git/blob/HEAD:/src/or/routerlist.c#l4435

The dir-spec describes a dirport method for retrieving all microdescriptors.
While this is a fine idea it was never implemented. Moving the spec portion
that describes this to the attic for now.

https://trac.torproject.org/9271

Also specify how pluggable transport arguments are passed from the
transport proxy to Tor and then to the bridge authority using
extra-info descriptors.

Tor moved from v2 to v3 directory information in commit 3ad6dc0e. Prior to that
Tor's 'GETINFO ns/*' options provided v2 information, but since then it's been
v3. For more information see...

https://trac.torproject.org/7953

Technically this is a tor bug, but it happens to be a convenient one so simply
updating our spec. In practice controller authors haven't noticed this
discrepancy because v3 router status entries are a superset of the fields in
v2.

Damian Johnson authored 12 years ago and Nick Mathewson committed 11 years ago

Mike has expressed on ticket #6872 and irc that he based bandwidth-weights
after the 'params' entry and would like new bandwidth-weights to be permitted.
Changing the spec as follows...

- Allowing for unspecified new 'Keyword=Int32' pairs, keeping the 'params'
  requrement that they're sorted.

- Making all values optional, including the current ones. This is a change, but
  will give Mike greater flexability and avoid creating a headache of 'these
  weights are mandatory in consensus-method X and these other are mandatory in
  consensus-method Y'.

Implemented in tor in b933360ee848873db6c051eabe5aecd01b3f67a3