Rui: On Sat, May 26, 2012 at 3:36 AM, Rui Miguel Silva Seabra <rms at 1407.org>wrote: > On Fri, 25 May 2012 22:52:13 +0530 > Arun Khan <knura9 at gmail.com> wrote: > > > I have a client project to implement PCI/DSS compliance. > > Some advice from my practical professional knowledge... > > Excellent post... > > The PCI/DSS auditor has stipulated that the web server, application > > middleware (tomcat), the db server have to be on different systems. > > In addition the auditor has also stipulated that there be a NTP > > server, a "patch" server, > > There is always the scope to be understood. > > If a server has card numbers somewhere, that server in on scope. > So is any other server on the same network segment. > So is any firewall delimiting these network segments. > > Now... if you have a sufficiently large number of systems in scope, > it's more practical to suppose PCI:DSS is in scope on all servers. > > This is what we ended up doing. It was far easier to build everything to be compliant than to selectively push PCI compliant configurations to a handful of servers. > This eases your maintenance as you won't have exceptions to deal with, > or justify, but if you have very few systems in scope rather than most > of the others which aren't, it'll be your decision considering the work > overload. I personally still advise to follow most rules on the non > scoped servers as they are in fact wise rules. > > > The Host OS on all of the above nodes will be CentOS 6.2. > > Not a good practice to say "6.2". Merely applying patches as time goes > on means in some time you'll be running 6.3. Say 6. :) > > > Below is a list of things that would be necessary. > > > > 1. Digital Certificates for each host on the PCI/DSS segment > > 2. SELinux on each Linux host in the PCI/DSS network segment > > Beware that many instructions tell you to disable selinux. I found that > with a little bit of work and the help of audit2why and a few more > selinux commands, you can usually work around bad apps by assuming the > risk of allowing what they need. > > A master will write his own selinux rules according to apps, though. > > We have selinux in our base configuration. The only caveat here is that vendors often will refuse to support an application if selinux is enabled. Though I know very well that selinux itself is not the problem (i.e., it's the policy that needs to be tweaked), the app owners claim that there is no way to figure out what is wrong when selinux is enabled. > > 3. Tripwire/AIDE on each Linux host in the PCI/DSS segment > > I advise OSSEC, rather than those, as it's a much better Host IDS. > > > 4. OS hardening scripts (e.g. Bastille Linux) > > I'm very wary of these generic ones, I usually bet on strongly reducing > the packages installed and defining the security settings straight from > my kickstart install scripts. > > > 5. Firewall > > 6. IDS (Snort) > > 6. Central “syslog” server > > Be careful to send logs under TLS. I found that as a syslog server, > rsyslog on RHEL/CentOS 5 *sucks* and gets you in trouble with ram > exhaustion and crashes. I had to backport from 6 as the idiotic siem > software running on that server demanded series 5 (even though it's > just java *sigh*) and we ran into this issue with rsyslog, which is > quite old under RHEL/CentOS. > > This siem server does not support TLS syslog, only plain UDP/TCP > unecrypted syslog, so one has to use a syslog server to receive under > TLS then forward to the localhost. > > > However, beyond this I would appreciate any comments/feedback / > > suggestion if you or your organization has undergone a PCI/DSS audit > > and what are the gotchas that you encountered, especially with respect > > to CentOS/ open source stack. > > Use sudo extensively. If you have many servers without central password > validation and too little people, it's better to have passwordless sudo > restricted to admins group as identified by access via OpenSSH RSA keys > than having to change your password every month on hundreds of servers. > > We use sudo among other things. Lately we have enabled ACLs to allow specific individuals access to specific configuration files. The main caveat with sudo is finding those applications that allow shell access, etc.. > Restrict your access to root shell, and keep it's password (written by > two persons, each knowing their own half) in a safe where none of you > can access without paper trail. > > Yes, as an admin you can override that, but if you have externalized > logs audited by a separate set of people, your trails may get you in > trouble, so that risk is mitigated. > > > I came across this which kind of brings out issues between the > > implementer and the PCI/DSS auditor. > > < > http://webmasters.stackexchange.com/questions/15098/pci-dss-compliance-for-a-vps-using-centos > > > > I see there some things that are not true, namely WRT CentOS versions. > > It has a lot to do with *how* you do your things, what evidences you > register, whether the auditor is excessively strict and/or knows the > technology and/or does a risk based assessment, how segmented is your > network, and so on. > > And this is so true. It is so much dependent on the auditor and how he/she interprets the spec. Again, thanks for an excellent post.