Assertion in timer in SIP transaction: Timer being rescheduled when already running

[bennylp]

Contributed patch from Matt Williams:

We hit a "entry->_timer_id < 1" assert in schedule_w_grp_lock(). This assert is to check that we're not rescheduling a timer that is already running.

We hit this in two different ways: once via a race condition and once because we sometimes set the timer again without cancelling previous timers.

Proposed solutions:

• Race condition: I think the fix is not to send the packet or reschedule the timer if the timer is already running. I've changed tsx_retransmit to drop out immediately if the resched flag was set and the timer was already scheduled.
• reschedule without cancel first: I've coded a fix to always cancel the timer before restarting it, if we're in a state where the timer could already be running.

[bennylp]

In 4567:

Closed #1690: Assertion in timer in SIP transaction: Timer being rescheduled when already running