PCI-DSS, nginx access logs, and credit card masking

PCI-DSS states that the card number must not be displayed in full—no more than the first six and the last four digits may be visible on a screen, a receipt, or on any other media used by the organization.

Sometimes your web server might leave access logs that contain this sort of information though. Here's an nginx access log snippet:

1.2.3.4 - - [02/Dec/2014:14:58:54 +0200] "GET /ccsite/?card_number=5100+1500+0000+0001&amount=10000&cvv=123&holder_name=N.Ame&expiration_month=1&expiration_year=2015&description=Something&_method=POST HTTP/1.1" 200 395 "https://partnersite.com/apps/ccapp?dothis.js" "Mozilla/5.0 (Windows NT 6.1; WOW64; rv:33.0) Gecko/20100101 Firefox/33.0"
1.2.3.4 - - [02/Dec/2014:16:07:46 +0200] "password=12345678&username=my_username" 200 416 "-" "Dalvik/1.6.0 (Linux; U; Android 4.3; C1905 Build/15.4.A.1.9)" 

Ouch, we shouldn't be writing stuff like this to our logs!

So what so we do?

Well, in this example I'm going to install the ngx_http_perl_module and use regular expressions to mask out the credit card and CVV.

First of all, let's see if we have Perl installed. Also, in order for Perl to recompile the modified modules during reconfiguration, it should be built with the -Dusemultiplicity=yes or -Dusethreads=yes parameters. Also, to make Perl leak less memory at run time, it should be built with the -Dusemymalloc=no parameter.

Let's check if we're good to go:

[root@webserver ~]# perl -V:usemultiplicity -V:usemymalloc
usemultiplicity='define';
usemymalloc='n';

Let's install some prerequisites to embed our Perl:

[root@webserver ~]# yum install perl-CPAN perl-ExtUtils-CBuilder perl-ExtUtils-MakeMaker perl-ExtUtils-Embed perl-devel 

Great. Now, let's back up our nginx configuration files and recompile it:

[root@webserver ~]# cp -rf /opt/nginx/conf/ /root/nginx_conf_bak

OK, now we should configure nginx as always, but with the addition of --with-http_perl_module. So in my case it is:

[root@webserver nginx-1.6.2]# ./configure --add-module=../naxsi/naxsi-master/naxsi_src/ --prefix=/opt/nginx --error-log-path=/var/log/nginx/nginx_error.log --http-log-path=/var/log/nginx/nginx_access.log --user=www-data --group=www-data --with-http_addition_module --with-http_geoip_module --with-http_gzip_static_module --with-http_stub_status_module --with-http_realip_module --without-mail_pop3_module --without-mail_smtp_module --without-mail_imap_module --without-http_memcached_module --without-http_ssi_module --without-http_uwsgi_module --without-http_scgi_module --with-http_perl_module

If you, like me, like to change the nginx headers and version to spoof their make to something else, say, IIS, then you should also make sure that the #define NGINX_VERSION in /src/core/nginx.h variable consists of numbers and dots only, for example not 3.4.1 (Unix), but 3.4.1, otherwise you'll encounter compilation problems.

Let's proceed to the installation of our new nginx binary that will now support the nginx perl module:

[root@webserver ~]# make
[root@webserver ~]# make install

Let's restore our old config:

[root@webserver ~]# cd /opt/nginx/conf/
[root@webserver ~]# rm -rf *
[root@webserver conf]# cp -rf /root/nginx_conf_bak/* .
[root@webserver conf]# chown -R www-data:www-data *

Time to do some substitutions using regular expressions on our logs. This needs to be in our http block:

[root@webserver ~]# vi nginx.conf
....
http {
perl_set $anonymize_data '
                sub {
                        my $r = shift;
                        my $req =  $r->request_body;
                        $req =~ s/\?card_number\=(?:4[0-9]{12}(?:[0-9]{3})?|5[1-5][0-9]{14}|3[47][0-9]{13}|3(?:0[0-5]|[68][0-9])[0-9]{11}|6(?:011|5[0-9]{2})[0-9]{12}|(?:2131|1800|35\d{3})\d{11})/\?card_number\=XXXX-XXXX-XXXX-XXXX/g;
                        $req =~ s/\&card_number\=(?:4[0-9]{12}(?:[0-9]{3})?|5[1-5][0-9]{14}|3[47][0-9]{13}|3(?:0[0-5]|[68][0-9])[0-9]{11}|6(?:011|5[0-9]{2})[0-9]{12}|(?:2131|1800|35\d{3})\d{11})/\&card_number\=XXXX-XXXX-XXXX-XXXX/g;
                        $req =~ s/\&cvv\=\d\d\d/\?cvv\=XXX/g;
                        $req =~ s/\&cvv\=\d\d\d/\&cvv\=XXX/g;
                        $req =~ s/password\=\w+/password\=XXXXXXXX/g;
                        return $req;
                } ';

Notice that I cover both scenarios; if the card_number arrives first (?card_number); or as a secondary/tertiary variable (&card_number). I do the same for the CVV. Replace "card_number" and "cvv" with whatever variable names your requests come as.

Finally, you will need to replace your "$request" with "$anonymize_data" in your log_format, like so:

log_format main $remote_addr - $remote_user [$time_local] "$anonymize_data" $status $body_bytes_sent "$http_referer" "$http_user_agent";

And now the card number, the CVV and the user's password get masked.

1.2.3.4 - - [02/Dec/2014:14:58:54 +0200] "GET /ccsite/?card_number=XXXX-XXXX-XXXX-XXXX&amount=10000&cvv=XXX&holder_name=N.Ame&expiration_month=1&expiration_year=2015&description=Something&_method=POST HTTP/1.1" 200 395 "https://partnersite.com/apps/ccapp?dothis.js" "Mozilla/5.0 (Windows NT 6.1; WOW64; rv:33.0) Gecko/20100101 Firefox/33.0"
1.2.3.4 - - [02/Dec/2014:16:07:46 +0200] "password=XXXXXXXX&username=my_username" 200 416 "-" "Dalvik/1.6.0 (Linux; U; Android 4.3; C1905 Build/15.4.A.1.9)"