[tor-bugs] #1653 [Tor Relay]: exit relays don't consider local cell queue when hearing sendme

[Tor Bug Tracker & Wiki]

#1653: exit relays don't consider local cell queue when hearing sendme
-----------------------+----------------------------------------------------
 Reporter:  arma       |       Owner:                    
     Type:  defect     |      Status:  needs_review      
 Priority:  major      |   Milestone:  Tor: 0.2.2.x-final
Component:  Tor Relay  |     Version:                    
 Keywords:             |      Parent:                    
-----------------------+----------------------------------------------------

Comment(by nickm):

 As noted on the #1298 comments by yetonetime, this fix in 8782dcf6a291b
 isn't right.  To consider why, look at the two warning cases in
 8782dcf6a291b:

 {{{
 +      log_info(LD_BUG, "Got a cell added to a cell queue when streams
 were "
 +               "supposed to be blocked; found that %d streams weren't.",
 n);
 }}}

 This can happen when we get a new connection on a blocked circuit: the
 first CONNECTED relay cell will get added, since connections don't start
 out blocked (nor should they).

 {{{
 +      log_info(LD_BUG, "Got a cell added to a cell queue when streams
 were "
 +               "all blocked. We should figure out why.");
 }}}

 This happens on '''every relay''' when a cell queue is high: we can't
 block incoming circuits (only edge connections), so we let them fill the
 circuit queue as much as they want.  We sure don't want to be logging for
 every cell on a busy circuit, and we don't want every cell to make us do a
 loop over the streams.

 I've tried to fix both of these on the latest commit on branch bug1653
 (f89323afda).

-- 
Ticket URL: <https://trac.torproject.org/projects/tor/ticket/1653#comment:9>
Tor Bug Tracker & Wiki <https://trac.torproject.org/>
The Tor Project: anonymity online