Configuration Tracking

Robert Ess

What is this machine's IP address? How much disk does it have? How much swap space? Has the configuration changed lately? These are typical questions you and others probably often ask about your installed base of UNIX workstations. The shell script, sys.conf, shown in Listing 1 is a very simplistic approach toward maintaining a flat file database of all your installed workstations. If configured into the bootup sequence, the script can notify you when a machine comes online for the first time and will let you know when the hardware or software configuration for that machine changes. Figure 1 shows a sample output from sys.conf from an HP9000, Model 725 running HP-UX.

How does it work?

The sys.conf script is a series of simple shell functions, and making the script functional on several platforms required some ugly case statements. This script could probably have been written in 30 lines of Perl or less, but since I don't know Perl, I wrote it as a Bourne shell script with shell functions. This also allowed for greater portability.

The script first checks to see that the user running the script is root; it then divines the operating system that the script is running on. Each OS has something different about it, such as the Ethernet interface name, the swap command, or arguments to the mail command, so the first case construct in the Define_Vars shell function is used to build those commands with the appropriate syntax.

To determine the machine's hardware serial number from a script, I created a file in /usr/tmp, host_name.sn, in which host_name is the actual hostname of that machine. In the Machine_Data shell function, you will need to insert your own server's host names to preclude any user being reported as the logged in id. For example,

Machine_Data(){
MACHINE='uname -m'
OSREV='uname -r'

case 'hostname' in
serv01|serv02|serv03)            LASTLOGIN="Server";;
*) LASTLOGIN='$PSCMD | egrep -v "lp|daemon|root|UID" | \
awk '{print $1}' | head -1';;
esac

I use three different methods to try to obtain the primary user for a machine. The first method involves checking the process table for the first nonsystem user. If no one is logged in at the time the script is run, it uses the last utility to check /etc/wtmp for the last non-root console login. If this is unsuccessful, the script checks the existing baseline file from previous runs to see whether a user name has been captured. If none of these is successful, LASTLOGIN is set to Unknown.

Next, the Network_Data function is called. The operation of this function is straightforward. It reports the number of Ethernet interfaces, their IP addresses, the netmask and broadcast values, and the default router. These are all trivial to obtain, but are nicely contained in one shell function.

In Disk_Data, vendor-specific commands are used to report on the locally attached disks. HP-UX, Solaris, and SunOS all use different commands to report on device and filesystem swap.

There are also shell functions for each of the operating systems that allow you to tailor the output to your liking. HP has the handy ioscan command for scanning the hardware bus. Also, the installed fileset information is easily perused and grepped. In HP-UX 10.x, the swlist utility provides a very informative list of installed products for a machine. Solaris uses the showrev -p command to list installed patches, and I use the output of pkginfo for product information.

Implementation

Ideally, sys.conf should reside on an NFS-mounted filesystem accessible by all of your workstations. I put ours on an Auspex file server, and have a single file that is called by all of the workstations through cron. Thus, I can easily make changes to all of the workstations at one time.

If you want to put this script into your machine's startup sequence, I suggest the localrc function in HP-UX 9.x and /etc/rc2.d/S99sysconf as the last startup script called in Solaris. When the script is run for the very first time, it checks for /usr/tmp/'hostname'.bl. If it is not found, the script assumes this to be a new install, and mail is sent to $ADMIN.

Each night via cron, the script is run, and the resulting output file is diffed against the baseline file, /usr/tmp/'hostname'.bl. If any differences are found, they are mailed to the $ADMIN, informing him/her of any changes taken place on the machine since the previous night's run. The previous configuration is then moved to a timestamped file, which provides a history of the machine's configuration.

I use a trusted host to rcp all of the baseline files to a common repository on our Auspex server. Then, I write a simple script that will display the desired system's configuration.

So, when the boss walks in and says, "I need a count of all the SPARC 20s with 500 Mb or more of disk, 64 Mb or more of RAM, how many HP 725s we have with less than 400 Mb, and who owns that HP-UX 10.01 box out in Bloomfield. Oh, by the way, I need it in 15 minutes," I don't panic. I run my script, give him a printout, and go fight the next fire.

About the Author

Bob Ess has been in the computing business long enough to know better. He is currently a senior UNIX systems administrator at DSC Communication Corporation in Plano, Texas. He can be reached at ress@spd.dsccc.com