Bug#536247: dovecot-common: dovecot-auth segfaults randomly

2009-07-11 Thread Andrew Schulman
On Wed, 08 Jul 2009 13:12:39 -0400, Timo Sirainen t...@iki.fi wrote: On Wed, 2009-07-08 at 10:56 -0400, Andrew Schulman wrote: messages.4.gz:Jul 2 12:26:24 helium kernel: dovecot-auth[26364]: segfault at 10008 ip 7faf466441eb sp 7fff50a3f260 error 4 in libc-2.9.so[7faf465d2000+149000]

Bug#536247: dovecot-common: dovecot-auth segfaults randomly

2009-07-11 Thread Timo Sirainen
On Jul 11, 2009, at 2:37 AM, Andrew Schulman wrote: Good news: I got a core file from a segfault. Bad news: I'm using a standard Debian package with no debugging symbols. Backtrace is below, but it looks as though it's not going to be very useful. I suppose that in order to get a useful

Bug#536247: dovecot-common: dovecot-auth segfaults randomly

2009-07-11 Thread Andrew Schulman
#2 0x7f06ac1f854f in calloc () from /lib/libc.so.6 The crash is in memory allocation, which means something had corrupted memory already. The only way to really solve this would be to run dovecot-auth via valgrind: auth_executable = /usr/bin/valgrind /usr/lib/dovecot/imap/dovecot-auth

Bug#536247: dovecot-common: dovecot-auth segfaults randomly

2009-07-11 Thread Andrew Schulman
OK, I restarted dovecot with auth_executable = /usr/bin/valgrind --log-file=/var/run/dovecot/valgrind.%p /usr/lib/dovecot/dovecot-auth as you suggested. My dovecot-auth process just quit: Jul 11 17:50:33 helium dovecot: auth(default): worker-server(andrex,127.0.0.1): Aborted: Worker process

Bug#536247: dovecot-common: dovecot-auth segfaults randomly

2009-07-11 Thread Timo Sirainen
On Jul 11, 2009, at 6:06 PM, Andrew Schulman wrote: OK, I restarted dovecot with auth_executable = /usr/bin/valgrind --log-file=/var/run/dovecot/ valgrind.%p /usr/lib/dovecot/dovecot-auth as you suggested. My dovecot-auth process just quit: Jul 11 17:50:33 helium dovecot: auth(default):

Bug#536247: dovecot-common: dovecot-auth segfaults randomly

2009-07-08 Thread Andrew Schulman
Package: dovecot-common Version: 1:1.1.16-1 Severity: important Several times a day, dovecot-auth crashes, disconnecting the connected users of course. syslog output from one day is below. Thanks, Andrew. messages.4.gz:Jul 2 12:26:24 helium kernel: dovecot-auth[26364]: segfault at 10008

Bug#536247: dovecot-common: dovecot-auth segfaults randomly

2009-07-08 Thread Timo Sirainen
On Wed, 2009-07-08 at 10:56 -0400, Andrew Schulman wrote: messages.4.gz:Jul 2 12:26:24 helium kernel: dovecot-auth[26364]: segfault at 10008 ip 7faf466441eb sp 7fff50a3f260 error 4 in libc-2.9.so[7faf465d2000+149000] Any chance of getting a gdb backtrace?

Bug#536247: dovecot-common: dovecot-auth segfaults randomly

2009-07-08 Thread Andrew Schulman
Any chance of getting a gdb backtrace? http://dovecot.org/bugreport.html Sure, but it looks a little involved. I'll work on this and report back when I have it. Also show your dovecot -n output. # dovecot -n # 1.1.16: /etc/dovecot/dovecot.conf # OS: Linux 2.6.26 x86_64 Debian squeeze/sid

Bug#536247: dovecot-common: dovecot-auth segfaults randomly

2009-07-08 Thread Timo Sirainen
On Wed, 2009-07-08 at 10:56 -0400, Andrew Schulman wrote: Several times a day, dovecot-auth crashes, disconnecting the connected users of course. syslog output from one day is below. Thanks, Andrew. messages.4.gz:Jul 2 12:26:24 helium kernel: dovecot-auth[26364]: segfault at 10008 ip

Bug#536247: dovecot-common: dovecot-auth segfaults randomly

2009-07-08 Thread Andrew Schulman
On Wed, 08 Jul 2009 13:42:33 -0400, Timo Sirainen t...@iki.fi wrote: On Wed, 2009-07-08 at 10:56 -0400, Andrew Schulman wrote: Several times a day, dovecot-auth crashes, disconnecting the connected users of course. syslog output from one day is below. Thanks, Andrew. messages.4.gz:Jul 2

Bug#536247: dovecot-common: dovecot-auth segfaults randomly

2009-07-08 Thread Timo Sirainen
On Wed, 2009-07-08 at 14:03 -0400, Andrew Schulman wrote: Jul 2 12:26:24 helium dovecot: child 26364 (auth-worker) killed with signal 11 (core dumps disabled) It's possible that the crash comes from PAM plugins. You could try if setting auth_worker_max_request_count to somewhat low value (or