Virtual Services HOWTO

6. Syslogd

6.1 Problem

Syslogd is the system logging utility commonly used on UNIX systems. Syslogd is a daemon that opens a special file called a FIFO. A FIFO is a special file that acts like a pipe. Anything that is written to the write side will come out the read side. Syslogd waits for data from the read side. There are C functions that write to the write side. If your program uses these C functions your output will go to syslogd.

Remember that we have used a chroot environment and the FIFO that syslogd is reading from (/dev/log) is not present. That means all the virtual environments will not log to syslogd.

6.2 Solution

Setup Links

Syslogd can look to a different FIFO if you tell it on the command line so run syslogd with the argument:

syslogd -p /virtual/log

Then symlink /dev/log to /virtual/log by:

ln -sf /virtual/log /dev/log

Then hard link all the /dev/log copies to this file by running:

ln -f /virtual/log /virtual/domain1.com/dev/log 

The virtfs script above already does this. Since /virtual is one contiguous disk and the /dev/log's are hard linked they have the same inode number and point to the same data. The chroot cannot stop this so all your virtual /dev/log's will now function. Note that all the messages from all the environments will be logged in one place. However, you can write separate programs to filter out the data.

Syslogd.init

This version of the syslogd.init file hard links the /dev/log's each time you start it because syslogd deletes and creates the /dev/log FIFO each time it runs. Here is a modified syslogd.init file:

#!/bin/sh

. /etc/rc.d/init.d/functions

case "$1" in
  start)
        echo -n "Starting dev log: "
        ln -sf /virtual/log /dev/log
        echo done
        echo -n "Starting system loggers: "
        daemon syslogd -p /virtual/log
        daemon klogd
        echo
        echo -n "Starting virtual dev log: "
        for i in /virtual/*
        do
                if [ ! -d "$i" ]
                then
                        continue
                fi
                if [ "$i" = "/virtual/lost+found" ]
                then
                        continue
                fi
                ln -f /virtual/log $i/dev/log
                echo -n "."
        done
        echo " done"
        touch /var/lock/subsys/syslogd
        ;;
  stop)
        echo -n "Shutting down system loggers: "
        killproc syslogd
        killproc klogd
        echo
        rm -f /var/lock/subsys/syslogd
        ;;
  *)
        echo "Usage: syslogd {start|stop}"
        exit 1
esac

exit 0

6.3 Multiple Syslogd's

One Per Disk

If you run out of space on one filesystem and you have to break up your virtual domains onto different disks remember that hard links will not cross disks. That means you will have to run a separate syslogd for each group of domains on a disk. For example, if you had thirteen domains on /virtual1 and fifteen domains on /virtual2, you would hard link thirteen domains to /virtual1/log and run one syslogd with syslogd -p /virtual1/log and hard link fifteen other domains to /virtual2/log with a syslogd running with syslogd -p /virtual2/log .

One Per Domain

If you do not want to centralize the logs to one place you could also run one syslogd per domain. This wastes process ID's so I do not recommend it but it is easier to implement. You would have to alter your syslogd.init file to run syslogd as chroot /virtual/domain1.com syslogd for each domain. This will run each syslogd within the chroot and the logs will be in /virtual/domain1.com/var/log rather than all combined in /var/log. Do not forget to run a syslogd normally syslogd for the main system and a kernel logger klogd .