kqueue crash on FreeBSD with 2.2.25

[Edgar Pettijohn]

On 16-07-03 03:30:36, Timo Sirainen wrote:
> On 02 Jul 2016, at 03:30, Adam Weinberger <adamw at adamw.org> wrote:
> > 
> >>> Jul  1 10:07:27 imap dovecot: master: Panic: kevent(EV_ADD, READ, 54) failed: Bad file descriptor
> >>> 
> >>> It's not dumping core, and I get the message even with "protocols ="
> >>> 
> >>> Downgrading back to 2.2.24 fixes it. What info would be helpful for me to provide here?
> >> 
> >> gdb backtrace: http://dovecot.org/bugreport.html#coredumps
> >> 
> > 
> > Fuller logs attached. Thanks for your help, Timo.
> > 
> > #6  0x2815b23a in i_panic (format=0x281ccf7a "kevent(EV_ADD, READ, %d) failed: %m") at failures.c:275
> > #7  0x28185e10 in io_loop_handle_add (io=0x288843a0) at ioloop-kqueue.c:67
> > #8  0x281815a8 in io_add_file (fd=56, condition=IO_READ, source_linenum=244, callback=0x2818a7d0 <signal_read>, context=0x0) at ioloop.c:59
> > #9  0x281813a6 in io_add (fd=56, condition=IO_READ, source_linenum=244, callback=0x2818a7d0 <signal_read>, context=0x0) at ioloop.c:81
> > #10 0x2818a666 in lib_signals_set_handler (signo=1, flags=3, handler=0x804e6c0 <sig_settings_reload>, context=0x0) at lib-signals.c:243
> > #11 0x0804e129 in main_init (set=0x2881d098) at main.c:518
> > #12 0x0804d562 in main (argc=3, argv=0xffffae08) at main.c:890
> 
> Weird. I haven't touched anything even close to that. Also kqueue works on my OSX. The only thing that comes to my mind is https://github.com/dovecot/core/commit/fde7b8a03bf91cfa5bb7ca3e84545386243fa0d2
> 
> Does it happen to work if you use the attached patch?
> 
The patch doesn't apply cleanly for me on OpenBSD -current.

Hmm...  Looks like a unified diff to me...
The text leading up to this was:
--------------------------
|commit 3fea4d5988de365503df44a9b067e3b181cac65c
|Author: Timo Sirainen <timo.sirainen at dovecot.fi>
|Date:   Thu Jun 30 14:08:24 2016 +0300
|
|    master: Stopping didn't close dead-pipes early enough.
|    
|    This caused stopping to be too slow. Broken by 0153cf542.
|
|diff --git a/src/master/service-monitor.c b/src/master/service-monitor.c
|index 19655c2..44cdb26 100644
|--- a/src/master/service-monitor.c
|+++ b/src/master/service-monitor.c
--------------------------
Patching file service-monitor.c using Plan A...
Hunk #1 succeeded at 513 (offset 8 lines).
Hunk #2 failed at 538.
Hunk #3 succeeded at 595 with fuzz 2 (offset 8 lines).
1 out of 3 hunks failed--saving rejects to service-monitor.c.rej
Hmm...  The next patch looks like a unified diff to me...
The text leading up to this was:
--------------------------
|commit fde7b8a03bf91cfa5bb7ca3e84545386243fa0d2
|Author: Timo Sirainen <timo.sirainen at dovecot.fi>
|Date:   Tue Jun 28 23:38:11 2016 +0300
|
|    master: Added support for stopping specific services.
|    
|    We need to have a per-service fd for detecting a dead master.
|
|diff --git a/src/master/service-monitor.c b/src/master/service-monitor.c
|index bf4e931..fb4f64f 100644
|--- a/src/master/service-monitor.c
|+++ b/src/master/service-monitor.c
--------------------------
Patching file service-monitor.c using Plan A...
Reversed (or previously applied) patch detected!  Assume -R? [y]

If I select [y] and continue to build I get lots of build errors.


-- 
Edgar Pettijohn