log_analysis

Rationale

I like to go through my system logs. This often lets me know if there's a system problem before it becomes a major issue, and it shows me security issues.

Reviewing system logs has problems:

• Logs contain lots of extraneous stuff that I want to be logged, but that I don't want to sift through when I review logs (ie. routine, error-free daemon operation.)
• Logs contain a lot of repetition, which drowns out the interesting entries.
• Noting repetition can be tricky because each entry usually has extra features to make it unique, such as a date, maybe a PID (ie. for syslog), and maybe application-specific information (ie. sendmail queue IDs.)
• One needs to remember to review them. :)
• One needs to be root to looks at logs for some OSs.
• On most systems, looking at the logs for just one day can be a pain.
• If I attack each box I deal with and write a separate script to do all this, I'll waste a lot of time duplicating effort.
• Writing patterns is a pain even if you know regular expressions.

Security

The program needs to run with permissions to read your log files in order to be useful, which usually means root. It does not default to SUID root, and I recommend not making it SUID, so just run it as root (ie. manually or out of cron). I've tried to avoid temp files everywhere that I can, and in the one case where I do use a temp file, I make sure to use the POSIX tmpnam function instead of trying to make up my own temp file algorithm. The default umask is 077. If you use action commands, there is nothing to stop you from using parts of the log message in insecure ways, so for goodness' sake, be careful.

Local extensions

log_analysis already has lots of rules, but chances are that you have log entries that aren't already covered. So, log_analysis can easily be extended via a local config file, as documented in the log_analysis manpage. There's even an easy way to do modular plug-ins.

Mailing list

Change Log

Incompatible changes

0.44

• no more gui_mode_configuration_disabled or gui_mode_ignore_disabled
• using the "gui_mode_config_savelocal" option will not recognize local modifications made from earlier versions of log_analysis. Be careful!

0.42

• dests may no longer contain backslash

0.38

• date_format defaults to %Y_%m_%d
• -o no longer also outputs to standard out. Add -O for the old behavior.
• config_version is now mandatory

0.37

• end has been replaced with @@end
• the new default sort is "funky". It handles numbers and IPs more cleanly, although it takes longer for large data sets.

0.36

• -F has been changed to -I internal_config
• -D has been changed to -I evals

0.35

• raw_rules no longer allows an optional code hook as its 4th field. If you were using this, please revert to the old version and let me know.

0.34

• Took required_log_file out of the config; supporting it was too messy, and everyone I asked wasn't using it, anyway. If you were using it, please revert to the old version and let me know.

0.29

• Some directive names used "-" instead of "_". These have been changed. The new names are include_dir, include_dir_if_exists, include_if_exists, and block_comment.

Future plans/TODO

• Statistics support. That is, some way to keep track of information beyond simple message counts -- say, so we can know not only how many mail message lines we've seen, but also how many bytes those messages contained. This probably includes the ability to allow each message to count towards multiple categories.
• Persistence. This goes with statistics -- once I can know, say, many bytes I get per day, I would now like a way for log_analysis to keep track of that, and tell me how many we saw this month or this year.
• Support for "state" across lines. That is, some kinds of log messages (most notably sendmail) don't provide all the relevant info in each log message, so you need to keep track of previous log messages. There is no mechanism to handle this.
• A config syntax sophisticated enough to keep up with all this.
• This page sucks. :) Rewrite to be more useful.