Re: postfix/master broken in -current? and 5.0_RC4

[christos%astron.com@localhost (Christos Zoulas)]

In article <grdlgn$3a2$1%ger.gmane.org@localhost>,
Christos Zoulas <christos%astron.com@localhost> wrote:
>In article <ECF45CEE-E3CA-417B-9CD6-B84061DA1D35%wrede.ca@localhost>,
>Andreas Wrede  <andreas%wrede.ca@localhost> wrote:
>>
>>On 06-Apr-09, at 14:34 , Joerg Sonnenberger wrote:
>>
>>> On Mon, Apr 06, 2009 at 08:02:47PM +0200, Markus W Kilbinger wrote:
>>>> After upgrading my qube 2 to actual -current kernel and userland
>>>> (cross compiled from scratch on my amd64 machine) I see postfix/ 
>>>> master
>>>> failing repeatedly mostly after processing/delivering some mails
>>>> successfully (ktrace output):
>>>
>>> I have the same problem. I can trigger it at will with:
>>> mailq
>>> mail someone@somedomain
>>> mailq
>>>
>>> as normal user. The second mailq will kill postfix.
>>
>>
>>Same problem on /i386 5.0_RC4, both with a newly built postfix and one  
>>built under 4.0 some time ago:
>>
>>postfix/master[11000]: panic: master_status_event: read EOF status
>
>Maybe it has to do with the latest fileops changes?

The following seems to restore the previous behavior. Perhaps we should only
drain file descriptor instances that belong to the same process, not to
different ones?

christos

Index: sys_socket.c
===================================================================
RCS file: /cvsroot/src/sys/kern/sys_socket.c,v
retrieving revision 1.59
diff -u -u -r1.59 sys_socket.c
--- sys_socket.c        4 Apr 2009 10:12:51 -0000       1.59
+++ sys_socket.c        6 Apr 2009 21:24:04 -0000
@@ -89,7 +89,7 @@
        .fo_stat = soo_stat,
        .fo_close = soo_close,
        .fo_kqfilter = soo_kqfilter,
-       .fo_drain = soo_drain,
+       .fo_drain = fnullop_drain, /* soo_drain, */
 };
 
 /* ARGSUSED */