[tor-bugs] #31653 [Core Tor/Tor]: padding machine sending padding from relay to client closes circuit

[Tor Bug Tracker & Wiki]

#31653: padding machine sending padding from relay to client closes circuit
-----------------------------+------------------------------------
 Reporter:  pulls            |          Owner:  (none)
     Type:  defect           |         Status:  new
 Priority:  Medium           |      Milestone:  Tor: 0.4.2.x-final
Component:  Core Tor/Tor     |        Version:  Tor: 0.4.1.5
 Severity:  Normal           |     Resolution:
 Keywords:  wtf-pad circpad  |  Actual Points:
Parent ID:                   |         Points:
 Reviewer:                   |        Sponsor:
-----------------------------+------------------------------------

Comment (by mikeperry):

 The codepath in circpad_machine_schedule_padding() for 0 delay uses a
 direct call instead of a scheduled callback.

 This might be causing an out-of-order AES ctr issue where the padding cell
 is being sent before the cell that triggered it, but somehow the AES
 counters are not updated correctly for this ordering. This should not
 happen normally... Are you also using the branch from #29494 by any
 chance? That might mess with your cell ordering in this case...

 As a workaround: Can you try replacing the direct call to
 circpad_send_padding_cell_for_callback() from
 circpad_machine_schedule_padding() for the case where in_usec <0
 (https://github.com/pylls/tor/blob/40c6f9bd887bdec7ed3bb03c690dd3d560321d48/src/core/or/circuitpadding.c#L1485)
 with an assignment of either in_usec = 0 or in_usec = 1?

 This will cause us to unwind back to libevent to call timer_set_cb() on
 the next event loop.

 Hope this helps!

--
Ticket URL: <https://trac.torproject.org/projects/tor/ticket/31653#comment:3>
Tor Bug Tracker & Wiki <https://trac.torproject.org/>
The Tor Project: anonymity online