I just started using owasp rules and got tons of false positives. Example someone in the description field has written:
"we are going to select some users tomorrow for our job platform."
This is detected as sql injection attack (id 950007). Well it is not. It is valid comment. I have tons of this kind false positives.
First I have set up SecRuleEngine DetectionOnly to gather information. Then I started using "SecRuleUpdateTargetById 950007 !ARGS:desc" or "SecRuleRemoveById 950007" and I already spend a day for this. modsec_audit.log is alreay > 100MB of size.
I am interested from your experience, how long do you fine tune it (roughly). After you turn it on, do you still get false positives and how do you manage to add white lists on time (do you analyze the logs daily) ?
I need this info to tell by boss the estimation for this task. It seems that will be long lasting.
Totally depends on your site, your technology and your test infrastructure. The OWASP CRS is very noisy by default and does require a LOT of tweaking. Incidentally there is some work going on this and next version might have a normal and a paranoid mode, to hopefully reduce false positives.
To give an example I look after a reasonably sized site with a mixture of static pages and a number of apps written in wide variety of technologies (legacy code - urgh!) and a fair amount of visitors.
Luckily I had a nightly regression run in our preproduction environment with good coverage, so that was my first port of call. I released ModSecurity there after some initial testing, in DetectionOnly mode and tweaked it over a month maybe until I'd addressed all of the issues and was comfortable moving to prod. This wasn't a full month of continuous work of course but 30-60 mins on most days to check the previous nights run, tweak the rules appropriately and set it up for next night's run (damn cookies with their random strings!).
Next up I did the same in production, and pretty much immediately ran into issues with free text feedback fields like you have (of course I didn't see most of these in regression runs). That took a lot of tweaking (had to turn off a lot of SQL Injection rules for those fields). I also got a lot of insight how many bots and scripts run against our site! Most were harmless or Wordpress exploit attempts (luckily I don't run Wordpress), so no real risk to my site, but still an eye opener. I monitored the logs hourly initially (paranoid!), then daily, and then weekly.
I would say from memory that it took another 3 months or so until I was fully comfortable turning it on fully and checked it a lot over the next few days. Luckily all hard work paid off and very few false positives.
Since then it's been fairly stable and very few false alerts - mostly dues to bad data (e.g. email@@example.com entered as an email address for a field which didn't validate email addresses properly) and I often left those place and fixed the field validation instead.
Some of the common issues and rules I had to tweak are given here: Modsecurity: Excessive false positives (note you may not need or want to turn off all these rules in your site).
We have Splunk installed on our web servers (basically a tool which sucks up log files and can then be searched or automatically alert or report on issues). So set up a few alerts for when the more troublesome, free text fields fields caused a ModSecurity block (have corrected one or two more false positives there), and also on volume (so we get an alert when a threshold passed and could see we were under a sustained attack - happens few times a year) and weekly/monthly reporting.
So a good 4-5 months to implement from scratch end to end with maybe 30-40 man days work over that time. But it was a very complicated site and I had no prior ModSecurity/WAF experience. On plus side learned a lot about web technologies, ModSecurity and got regexpr-blindness from staring at some of the rules! :-)