minutes, but writing the "valid-after" line in our vote based
on our configured V3AuthVotingInterval: so unless the intervals
matched up, we immediately rejected our own vote because it didn't
start at the voting interval that caused us to construct a vote.
This caused log entries like:
Oct 23 01:16:16.303 [notice] Choosing expected valid-after time
as 2007-10-23 05:30:00: consensus_set=0, interval=1800
...
Oct 23 01:20:01.203 [notice] Choosing valid-after time in vote as
2007-10-23 06:00:00: consensus_set=0, interval=3600
Oct 23 01:20:01.290 [warn] Rejecting vote with valid-after time of
2007-10-23 06:00:00; we were expecting 2007-10-23 05:30:00
Oct 23 01:20:01.291 [warn] Couldn't store my own vote! (I told
myself, 'Bad valid-after time'.)
Nick, you should look at this, as it's your design. :)
svn:r12129
Make authorities start accepting (and advertising their acceptance of) consensus method 2. If all goes well, we'll have a working Unnamed flag. Otherwise, we'll have a fun backtrace.
svn:r12113
Note that consensus-method is not a means for making backward-incompatible format changes. Because any point that confuses Roger will *definitely* confuse many non-Roger readers of dir-spec.txt ;)
svn:r12112
Remove an unused and unneeded layer of abstraction: we only have one store for routers. (I had thought we might need a second one for annotated routers, but that's silly.
svn:r12101
New code (disabled for now) to use the SSL context's cert store instead of using its "extra chain cert" list to get our identity certificate sent. This is a little close to what OpenSSL expects people to do, and it has the advantage that we should be able to keep the id cert from being sent by setting the NO_CHAIN_CERT bit. I have tried turning new code on, and it seemed to work fine.
svn:r12086
More fixes for bad behavior when downloading extrainfos: do not download an ei if we lack the key to verify it, and do not download it if we already got it and found (weirdly) that it didn't match the corresponding server descriptor.
svn:r12071
Respond to INT and TERM SIGNAL commands before we execute the
signal, in case the signal shuts us down. We had a patch in
0.1.2.1-alpha that tried to do this by queueing the response on
the connection's buffer before shutting down, but that really
isn't the same thing. Bug located by Matt Edman.
This is a bug in 0.1.2.x too, but there's no way we should backport
this fix. Speaking of which, can somebody double-check it? :)
svn:r12070
When we decode to use consensus method 2 or later, compute Unnamed and Named more or less as described in 122. Don't actually use consensus method 2 yet, so we can be sure we didn't screw up v1..
svn:r12055
Remember the valid-until time of the most recent consensus that listed
a router, and (if we are a cache) never delete the routerdesc until
that conensus is expired. This is way easier than retaining multiple
consensuses. (Of course, the info isn't retained across restarts,
but that only affects a few caches at a time.)
svn:r12041
