Keep a detailed audit trail of what's being done on your systems.
Process accounting allows you to keep detailed logs of every command a user runs, including CPU time and memory used. From a security standpoint, this means the system administrator can gather information about what user ran which command and at what time. This is not only very useful in assessing a break-in or local root compromise, but can also be used to spot attempted malicious behavior by normal users of the system. (Remember that intrusions don't always come from the outside.)
To enable process accounting, run these commands:
# mkdir /var/account # touch /var/account/pacct && chmod 660 /var/account/pacct # /sbin/accton /var/account/pacct
Alternatively, if you are running Red Hat or SuSE Linux and have the process accounting package installed, you can run a startup script to enable process accounting. On Red Hat, try this:
# chkconfig psacct on # /sbin/service psacct start
On SuSE, use these commands:
# chkconfig acct on # /sbin/service acct start
The process accounting package provides several programs to make use of the data that is being logged. The ac program analyzes total connect time for users on the system.
Running it without any arguments prints out the number of hours logged by the current user:
[andrew@colossus andrew]$ ac total 106.23
If you want to display connect time for all users who have logged onto the system, use the -p switch:
# ac -p root 0.07 andrew 106.05 total 106.12
The lastcomm command lets you search the accounting logs by username, command name, or terminal:
# lastcomm andrew ls andrew ?? 0.01 secs Mon Dec 15 05:58 rpmq andrew ?? 0.08 secs Mon Dec 15 05:58 sh andrew ?? 0.03 secs Mon Dec 15 05:44 gunzip andrew ?? 0.00 secs Mon Dec 15 05:44 # lastcomm bash bash F andrew ?? 0.00 secs Mon Dec 15 06:44 bash F root stdout 0.01 secs Mon Dec 15 05:20 bash F root stdout 0.00 secs Mon Dec 15 05:20 bash F andrew ?? 0.00 secs Mon Dec 15 05:19
To summarize the accounting information, you can use the sa command. By default it will list all the commands found in the accounting logs and print the number of times that each one has been executed:
# sa 14 0.04re 0.03cp 0avio 1297k troff 7 0.03re 0.03cp 0avio 422k lastcomm 2 63.90re 0.01cp 0avio 983k info 14 34.02re 0.01cp 0avio 959k less 14 0.03re 0.01cp 0avio 1132k grotty 44 0.02re 0.01cp 0avio 432k gunzip
You can also use the -u flag to output per-user statistics:
# sa -u root 0.01 cpu 344k mem 0 io which root 0.00 cpu 1094k mem 0 io bash root 0.07 cpu 1434k mem 0 io rpmq andrew 0.02 cpu 342k mem 0 io id andrew 0.00 cpu 526k mem 0 io bash andrew 0.01 cpu 526k mem 0 io bash andrew 0.03 cpu 378k mem 0 io grep andrew 0.01 cpu 354k mem 0 io id andrew 0.01 cpu 526k mem 0 io bash andrew 0.00 cpu 340k mem 0 io hostname
You can peruse the output of these commands every so often to look for suspicious activity, such as increases in CPU usage or commands that are known to be used for mischief.