Bug 507795

I'm not able to reproduce this issue.

I used same steps as you wrote but after step 7 (service named reload) information that named was reloaded was written to the log.

Could you please try to reproduce this issue without mounting /dev/log to chroot? It shouldn't be needed.

All rotated logfiles have to be removed before next logrotate (possible a logrotate bug?)

Do this before logrotate:

find /var/log -name \*-`date +%Y%m%d` -exec rm -f {} \;

After inspection this is a rsyslogd problem. New rsyslogd closes and reopens /dev/log when it gets -HUP (logrotate sends it). This behavior obviously breaks all applicantions which use chroot and syslog because there is no way how to get access to the new /dev/log file.

[root@evileye x86_64]# stat /dev/log 
  File: `/dev/log'
  Size: 0               Blocks: 0          IO Block: 4096   socket
Device: dh/13d  Inode: 46602       Links: 1
Access: (0666/srw-rw-rw-)  Uid: (    0/    root)   Gid: (    0/    root)
Access: 2009-06-24 16:02:53.775622723 +0200
Modify: 2009-06-24 16:02:53.775622723 +0200
Change: 2009-06-24 16:02:53.775622723 +0200
[root@evileye x86_64]# killall -HUP rsyslogd
[root@evileye x86_64]# stat /dev/log 
  File: `/dev/log'
  Size: 0               Blocks: 0          IO Block: 4096   socket
Device: dh/13d  Inode: 46622       Links: 1
Access: (0666/srw-rw-rw-)  Uid: (    0/    root)   Gid: (    0/    root)
Access: 2009-06-24 16:03:09.011613370 +0200
Modify: 2009-06-24 16:03:09.011613370 +0200
Change: 2009-06-24 16:03:09.011613370 +0200


I think there are two possible solutions for this problem:

1. Use functionality from rsyslog-3.21.11-HUPisRestart.patch and set
"$HUPisRestart off" in the default rsyslog.conf

2. patch source to make sure /dev/log is opened only once

Both approaches solve the problem. Reassigning to rsyslog for further inspection. Note that RHEL5 syslogd has "correct" behavior by default, it doesn't reopen the /dev/log.

Rsyslog indeed reopens the /dev/log socket, but this can't be easily prevented if you want to be able to restart the daemon.

You can configure it to create an additional socket in the daemon's chroot with the $AddUnixListenSocket directive, though. See upstream doc [1] for more info.

Reassigning back to bind.

[1] http://www.rsyslog.com/doc-imuxsock.html

An solution for automatic chroot creation can be to include /etc/sysconfig/named configuration into rsyslog init script, then start initscript with environment set for named and update default rsyslog.conf with:

if $environ contains "ROOTDIR" then $AddUnixListenSocket /var/named/chroot/dev/log

There is no if $direxist()... or something similar in current rsyslog.

But I think it's not best solution. :(

bind-9.6.1-5.P1.fc11 has been submitted as an update for Fedora 11.
http://admin.fedoraproject.org/updates/bind-9.6.1-5.P1.fc11

bind-9.6.1-5.P1.fc11 has been pushed to the Fedora 11 testing repository.  If problems still persist, please make note of it in this bug report.
 If you want to test the update, you can install it with 
 su -c 'yum --enablerepo=updates-testing update bind'.  You can provide feedback for this update here: http://admin.fedoraproject.org/updates/F11/FEDORA-2009-9677

bind-9.6.1-6.P1.fc11 has been pushed to the Fedora 11 stable repository.  If problems still persist, please make note of it in this bug report.

I'm seeing the same thing in EL5.5.  I'm guessing that bug 516998 is the EL5 clone of this one, but I do not have permission to access it.  Can something be done in EL5 to fix this?  I'm not sure what the fix was to the F11 package or if it could be applied to bind-9.3.6-4.P1.el5_5.3.