On Sun, Dec 01, 2019 at 11:41:51PM +0100, Klemens Nanni wrote:
> It crashed again:
>
> Dec 1 23:36:50 eru unwind[88042]: startup
> Dec 1 23:38:18 eru unwind[44830]: frontend exiting
> Dec 1 23:38:18 eru unwind[88042]: resolver terminated; signal 11
> Dec 1 23:38:18 eru unwind[88042]: terminating
Sorry, this is the log from the crash that happened right after I
restarted unwind, this time while being in the Air Canada lounge.
Here's the log that corresponds to the backtrace from the previous mail:
Dec 1 19:22:50 eru unwind[44200]: startup
Dec 1 21:43:42 eru unwind[38187]: frontend exiting
Dec 1 21:43:42 eru unwind[57687]: resolver terminated; signal 11
Dec 1 21:43:42 eru unwind[57687]: terminating
As you can see, it's the same crash; looking at the newest crash
(from 23:38:18) one can see that it is the exact same code path:
Program terminated with signal SIGSEGV, Segmentation fault.
#0 try_next_resolver (rq=0x15eba8268200) at
/s/sbin/unwind/resolver.c:788
788 rq->running++;
(gdb) bt
#0 try_next_resolver (rq=0x15eba8268200) at
/s/sbin/unwind/resolver.c:788
#1 0x000015e9782eb0ba in setup_query (query_imsg=<optimized out>) at
/s/sbin/unwind/resolver.c:715
#2 0x000015e9782ea974 in resolver_dispatch_frontend (fd=<optimized
out>, event=<optimized out>, bula=0x15ec61a0d000) at
/s/sbin/unwind/resolver.c:483
#3 0x000015e97838ac5f in event_process_active (base=<optimized out>)
at /usr/src/lib/libevent/event.c:334
#4 event_base_loop (base=0x15ec4e373000, flags=0) at
/usr/src/lib/libevent/event.c:483
#5 0x000015e9782e9c56 in resolver (debug=<optimized out>,
verbose=<optimized out>) at /s/sbin/unwind/resolver.c:383
#6 0x000015e9782f2ce2 in main (argc=0, argv=0x7f7fffff1588) at
/s/sbin/unwind/unwind.c:173
(gdb) l
783 evtimer_add(&rq->timer_ev, &tv);
784
785 if (resolve(res, query_imsg->qname, query_imsg->t,
786 query_imsg->c, query_imsg, resolve_done) != 0)
787 goto err;
788 rq->running++;
789
790 return 0;
791
792 err:
(gdb) p rq
$1 = (struct running_query *) 0x15eba8268200
(gdb) p *rq
Cannot access memory at address 0x15eba8268200
So why does evtimer_add() work on derefencing `rq' but the increment
fails, even though resolve() in between apparently does not touch it?
While sitting at the lounge, I'm running unwind with DEBUG='-g3 -O0' to
get a better backtrace.
