"No circuits are opened" messages with onion services
If a Tor instance is only doing onion service activity, and circuits time out, then the Tor client thinks that circuits aren't opened and gives them a full minute to complete. It also complains about this in the logs at notice level, like this:
Aug 06 02:55:30.000 [notice] No circuits are opened. Relaxed timeout for circuit 21570 (a Hidden service: Pre-built vanguard circuit 4-hop circuit in state doing handshakes with channel state open) to 60000ms. However, it appears the circuit has timed out anyway.The fix is simple. circuit_any_opened_circuits() in circuitlist.c is only counting circuits as opened if they use the DEFAULT_ROUTE_LEN. We just need to count >= DEFAULT_ROUTE_LEN to count them.
Activity
changed milestone to %Tor: unspecified
https://github.com/torproject/tor/pull/263
Trac:
Status: assigned to needs_review
Description: If a Tor instance is only doing onion service activity, and circuits time out, then the Tor client thinks that circuits aren't opened and gives them a full minute to complete. It also complains about this in the logs at notice level.The fix is simple. circuit_any_opened_circuits() in circuitlist.c is only counting circuits as opened if they use the DEFAULT_ROUTE_LEN. We just need to count >= DEFAULT_ROUTE_LEN to count them.
to
If a Tor instance is only doing onion service activity, and circuits time out, then the Tor client thinks that circuits aren't opened and gives them a full minute to complete. It also complains about this in the logs at notice level, like this:
Aug 06 02:55:30.000 [notice] No circuits are opened. Relaxed timeout for circuit 21570 (a Hidden service: Pre-built vanguard circuit 4-hop circuit in state doing handshakes with channel state open) to 60000ms. However, it appears the circuit has timed out anyway.The fix is simple. circuit_any_opened_circuits() in circuitlist.c is only counting circuits as opened if they use the DEFAULT_ROUTE_LEN. We just need to count >= DEFAULT_ROUTE_LEN to count them.
Replying to mikeperry:
Please
make check-changes:./changes/bug27409: bug number 27409 does not appearTrac:
Status: needs_review to needs_revision-
Yikes sorry I dropped this for so long. Fixups are on the PR comment.
I'm not sure if 0.2.9 backport is worth it -- I and others refactored bits of circuit_expire_building() in 0.3.3 to handle vanguards and improved HS timeout behavior, among other things. At a glance, I could just patch the pre-refactored code similarly, but this seems not serious enough for that.
Unless you think this message is very common in the wild?
Trac:
Status: needs_revision to needs_review I left a question on the pull request about intro and rend circuits.
I think we should backport to 0.2.9, because this bug can cause clients to stall (see #28639 (moved), same bug, different trigger).
I opened #28714 (moved) for the controller 1- and 2-hop circuit case.
I think that code should be fine if my response to teor is accurate.
Also, this notice log ("No circuits are opened.") is something I have in masse on my obfs4 bridge (in Tor Browser):
Dec 04 08:38:43.000 [notice] No circuits are opened. Relaxed timeout for circuit 21741 (a Testing circuit 3-hop circuit in state doing handshakes with channel state open) to 107586ms. However, it appears the circuit has timed out anyway. [9 similar message(s) suppressed in last 3600 seconds]Always the same... I'm not sure this patch will fix that?
I don't think #28714 (moved) will fix it either.
Maybe we should log something different when the circuit is not a user circuit?
mentioned in issue #27066 (moved)
mentioned in issue #28639 (moved)
mentioned in issue #28663 (moved)
moved to tpo/core/tor#27049
