Detecting Malice with ModSecurity: GeoLocation Data

 I would like to introduce a new blog series entitled - Detecting Malice with ModSecurity and will be used as an alternative to the Advanced Topic of the Week blog posts.  The the idea is to take the outstanding web application intrusion/fraud detection concepts introduced by my friend Robert "Rsnake" Hansen in his book Detecting Malice and provide real-life implementation examples using ModSecurity.  Rsnake has given me the green light to provide actual snippets of text/quotes directly from his book!  The format will be that Rsnake will setup the underlying issues and methods to identify bad behavior and then I will present some practical implementations using ModSecurity.

In this installment of Detecting Malice with ModSecurity, we will discuss how to use Geolocation data.

Geolocation section from Detecting Malice

IP addresses aren’t quite like street addresses, but they’re sometimes close. There are a number of projects that correlate, with varying success, IP address information to geographic locations. Some very large retailers help these databases by tying in IP address information collected along with address information given by their users during registration. Others tie in ISP information and geographic information with the IP address – a far less precise method in practice. Yet others still use ping time and hops to get extremely accurate measurement estimates of physical location.

The following information is typically available in GeoIP databases:

 Country

 Region (for some countries)

 City

 Organization (typically from the WHOIS database)

 ISP

 Connection speed

 Domain name

There are lots of reasons you may be interested in the physical location of your users. Advertisers are always on the lookout for ways to make their ads more interesting and meaningful (because that increases their click-through rate), and serving local ads is a good way to achieve that. Relevance is key, and that can be optimized by targeting the results to a user’s location.

Geographic information can also give you clues to the user’s disposition. If you are running a website in the United States and someone from another country is accessing your site, it may tell you quite a bit about the possible intentions of that user. That becomes more interesting if the site is disliked in principle by the population of other countries, or doesn’t do business with that country in particular. For example, many sites are totally blocked by the Chinese’s system of firewalls because the Chinese government believes certain words or thoughts are disruptive to the government’s goals. If someone from China IP space visits your site despite you being blocked by the Chinese government’s firewalls, there is a high likelihood the traffic is state sponsored.

Similarly, if you know that a large percentage of your fraud comes from one location in a particular high crime area of a certain city, it might be useful to know that a user is accessing your site from a cyber café in that same high crime area. Perhaps you can put a hold on that shipment until someone can call the user and verify the recipient. Also if you know that fraud tends to happen in certain types of socio- economic geographies, it could help you heuristically to determine the weighted score of that user. This sort of information may not be available to you in your logs, but could easily be mined through external sources of relevant crime information. Companies like ESRI heavily focus on mapping demographics, for instance.

In terms of history, it doesn’t hurt to look back into time and see if the IP address of the user has ever connected to you before. That information combined with their status with the website can be useful information. Another thing to consider is the type of connection used. If you know the IP space belongs to a DSL provider, it may point to a normal user. If it points to a small hosting provider, and in particular a host that is also a web server, it is highly likely that it has been compromised. Knowing if the IP belongs to a datacenter or a traditional broadband or modem IP range is useful information.

Knowing the physical location can help in other ways as well. If your website does no business in certain states or with certain countries due to legislative issues, it may be useful to know that someone is attempting to access your site from those states. This is often true with online casinos, which cannot do business with people in the United States where the practice is prohibited. Alerting the user of that fact can help reduce the potential legal exposure of doing business.

One concept utilizing physical location is called, “defense condition.” Normally, you allow traffic from everywhere, but if you believe that you're being attacked you can switch to yellow (the meaning of which is specific to the websites in question but, for example, can mean that your website blocks suspicious traffic coming from certain countries) or red (your websites block any traffic from certain countries). While this may seem like a good idea, in some cases this can actually be used against a website to get a company to block other legitimate users, so the use of this concept is cautioned against.

Utilities like MaxMind’s geoipaddress lookup can give you high level information about the location of most IP addresses for free. There are other products that can give you a lot more information about the specific location of the IP address in question, which can be extremely helpful if you are concerned about certain traffic origins.

$ geoiplookup.exe 157.86.173.251 GeoIP Country Edition: BR, Brazil

There has been quite a bit of research into this concept of bad geographic origins. For instance McAfee published a study citing Hong Kong (.hk), China (.cn) and the Philippines (.ph) as the three most likely top level domains to pose a security risk to their visitors. While these kinds of studies are interesting, it’s important not to simply cast blame on an entire country, but rather think about why this is the case – which can give you far more granular and useful information. It should be noted that this study talks about the danger that websites pose to internet surfers, not to websites in general, but there may be a number of ways in which your website can contact or include portions of other sites, so it still may be applicable.

Using GeoLocation Data in ModSecurity

ModSecurity supports using geolocation data through the integration with the free MaxMind GeoLite Country or GeoLite City databases. The first step is to download the database of your choice and put it somewhere on the local filesystem where ModSecurity can use it (for example in the same directory as the Core Rule Set).

Before you can use the GeoIP data, you first have to use the SecGeoLookupDb directive:

SecGeoLookupDb

Description: Defines the path to the geographical database file.

Syntax: SecGeoLookupDb /path/to/db

Example Usage: SecGeoLookupDb /usr/local/geo/data/GeoLiteCity.dat

Processing Phase: N/A

Scope: Any

Version: 2.5.0

Dependencies/Notes: Check out maxmind.com for free database files.

Once the SecGeoLookupDb directive is active, you then need to use a rule similar to the following which will pass the client's IP address (REMOTE_ADDR) to the @geoLookup operator:

SecGeoLookupDb /path/to/apache/conf/base_rules/GeoLiteCity.dat
SecRule REMOTE_ADDR "@geoLookup" "phase:1,t:none,pass,nolog"

When this rule is executed, the GEO variable collection data will be populated with the data extracted from the GeoIP DB.  Here is an example debug log section:

[27/Oct/2010:09:49:51 --0400] [localhost/sid#10080d708][rid#1038060a0][/foo.php][4] Recipe: Invoking rule 1009c47d8; [file "/usr/local/apache/conf/modsec_current/base_rules/modsecurity_crs_15_customrules.conf"] [line "30"].
[27/Oct/2010:09:49:51 --0400] [localhost/sid#10080d708][rid#1038060a0][/foo.php][5] Rule 1009c47d8: SecRule "REMOTE_ADDR" "@geoLookup " "phase:1,t:none,nolog,pass"[27/Oct/2010:09:49:51 --0400] [localhost/sid#10080d708][rid#1038060a0][/foo.php][4] Transformation completed in 1 usec.
[27/Oct/2010:09:49:51 --0400] [localhost/sid#10080d708][rid#1038060a0][/foo.php][4] Executing operator "geoLookup" with param "" against REMOTE_ADDR.
[27/Oct/2010:09:49:51 --0400] [localhost/sid#10080d708][rid#1038060a0][/foo.php][9] Target value: "119.192.231.134"
[27/Oct/2010:09:49:51 --0400] [localhost/sid#10080d708][rid#1038060a0][/foo.php][9] GEO: Looking up "119.192.231.134".
[27/Oct/2010:09:49:51 --0400] [localhost/sid#10080d708][rid#1038060a0][/foo.php][9] GEO: Using address "119.192.231.134" (0x77c0e786).
[27/Oct/2010:09:49:51 --0400] [localhost/sid#10080d708][rid#1038060a0][/foo.php][9] GEO: rec="\x77\x31\x31\x00\x53\x65\x6f\x75\x6c\x00\x00\xb0\x32\x21\x2d\xd8\x2e\x00\x00\x00\x00\xa6\x45\x37\x00\x41\x75\x63\x6b\x6c\x61\x6e\x64\x00\x00\x25\xd7\x15\x13\x22\x36\x00\x00\x00\x00\xa6\x46\x35\x00\x4e"
[27/Oct/2010:09:49:51 --0400] [localhost/sid#10080d708][rid#1038060a0][/foo.php][9] GEO: country="\x77"[27/Oct/2010:09:49:51 --0400] [localhost/sid#10080d708][rid#1038060a0][/foo.php][9] GEO: region="\x31\x31\x00"
[27/Oct/2010:09:49:51 --0400] [localhost/sid#10080d708][rid#1038060a0][/foo.php][9] GEO: city="\x53\x65\x6f\x75\x6c\x00"
[27/Oct/2010:09:49:51 --0400] [localhost/sid#10080d708][rid#1038060a0][/foo.php][9] GEO: postal_code="\x00"
[27/Oct/2010:09:49:51 --0400] [localhost/sid#10080d708][rid#1038060a0][/foo.php][9] GEO: latitude="\xb0\x32\x21"
[27/Oct/2010:09:49:51 --0400] [localhost/sid#10080d708][rid#1038060a0][/foo.php][9] GEO: longitude="\x2d\xd8\x2e"
[27/Oct/2010:09:49:51 --0400] [localhost/sid#10080d708][rid#1038060a0][/foo.php][9] GEO: dma/area="\x00\x00\x00"
[27/Oct/2010:09:49:51 --0400] [localhost/sid#10080d708][rid#1038060a0][/foo.php][9] GEO: 119.192.231.134={country_code=KR, country_code3=KOR, country_name=Korea, Republic of, country_continent=AS, region=11, city=Seoul, postal_code=, latitude=37.566399, longitude=126.999702, dma_code=0, area_code=0} [27/Oct/2010:09:49:51 --0400] [localhost/sid#10080d708][rid#1038060a0][/foo.php][4] Operator completed in 85333 usec.
[27/Oct/2010:09:49:51 --0400] [localhost/sid#10080d708][rid#1038060a0][/foo.php][4] Warning. Geo lookup for "119.192.231.134" succeeded. [file "/usr/local/apache/conf/modsec_current/base_rules/modsecurity_crs_15_customrules.conf"] [line "30"]

From this point on, you can use the geographic information from the GEO collection in your rules.

GeoLocation Usage Strategies

Allowed GEO Locations

If your clients only come from some specific geographic regions, then you could create a rule such as the following which would block clients who are not coming from US-based IP addresses:

SecRule GEO:COUNTRY_CODE3 "!@streq USA" "phase:1,t:none,log,deny,msg:'Client IP not from USA'"

DefCon Level - Static

The vast majority of organizations are global and would not want to implement this type of restriction for day-to-day operations.  This type of rule is useful, however, when the organization wants to initiate a more restrictive "Defense Condition Level" when under a direct attack.  You could make the previous rule conditional and only used when under an increased DefCon Level:  

SecAction "phase:1,t:none,nolog,pass,setvar:tx.defcon_level=1"
SecGeoLookupDb /path/to/apache/conf/base_rules/GeoLiteCity.dat
SecRule REMOTE_ADDR "@geoLookup" "phase:1,t:none,pass,nolog"
SecRule GEO:COUNTRY_CODE3 "!@streq USA" "chain,phase:1,t:none,log,deny,msg:'Client IP not from USA'"      SecRule TX:DEFCON_LEVEL "@streq 1"

DefCon Level - Dynamic

While this example works, one of the big challenges in effectively using the concept of DefCon Levels with ModSecurity rules is the fact that you would have to initiate an Apache restart/reload in order for these DefCon Level variable data to be used.  This becomes a real scaling challenge if you are protecting a large server farm.  Under these types of scenarios, you can actually utilize the concept of dynamic rules which are activated based on web requests from authorized sources.  

Imagine you are running a SOC and you identify that your web sites are under attack and you want to initiate the new DefCon Level restriction rules to only allow requests from specific geographic regions.  Rather then needing to push out new Apache/ModSecurity configs to all web servers and initiate restarts, you could alternatively initiate a web request from an authorized web client system from within the SOC.  This client would have a specific IP address, User-Agent String and make a request to a specific URL with a specific parameter payload.  When this request is received by ModSecurity, the rules would then set a new DefCon level variable in a persistent Global collection.  When this happens, the GeoIP restriction rules would become active.  Once the threat has passed and the DefCon Level has been restored, a 2nd request can be sent out to all web servers to remove the DefCon Level variable and return the rules to their normal processing.  Again, the advantage of this approach is that it is a much faster method to toggle on/off a subset of rules vs having to update Apache/ModSecurity configs and initiate restarts.

Here are some example rules that implement this basic concept:

SecGeoLookupDb /path/to/apache/conf/base_rules/GeoLiteCity.dat
SecAction "phase:1.t:none,nolog,pass,setsid:global" 
SecRule REQUEST_FILENAME "/defcon_control" "chain,phase:1,t:none,log,msg:'DefCon Level 1 - GeoIP Restrictions Enabled.'"
    SecRule ARGS:DEFCON_LEVEL "@streq 1" "chain"
        SecRule REMOTE_ADDR "^192\.168\.1\.101$" "setvar:session.defcon_level=1" 
SecRule REQUEST_FILENAME "/defcon_control" "chain,phase:1,t:none,log,msg:'DefCon Level 1 - GeoIP Restrictions Disabled.'" 
    SecRule ARGS:DEFCON_LEVEL "5" "chain"
         SecRule REMOTE_ADDR "^192\.168\.1\.101$" "setvar:session.defcon_level=5" 
SecRule REMOTE_ADDR "@geoLookup" "phase:1,t:none,pass,nolog" 
SecRule GEO:COUNTRY_CODE3 "!@streq USA" "chain,phase:1,t:none,log,deny,msg:'Client IP not from USA'" 
    SecRule SESSION:DEFCON_LEVEL "@streq 1"

With these rules in place, the following request from source IP 192.168.1.101 would enable the DefCon Level GeoIP restriction rules:

GET /defcon_control?defcon_level=1 HTTP/1.1

Assigning Fraud/Risk Scores 

As Rsnake discussed, there are many different fraud detection resources that have assigned general risk scores to certain geographic regions.  For example, in the Clear Commerce whitepaper entitled Fraud Prevention Guide, they list the top 12 High Risk Countries:

• Ukraine
• Indonesia
• Yugoslavia
• Lithuania
• Egypt
• Romania
• Bulgaria
• Turkey
• Russia
• Pakistan
• Malaysia
• Israel

If you wanted to use this general information and assign an increased anomaly score to requests originating from these countries, you could use a rule similar to the following:

SecRule GEO:COUNTRY_CODE "@pm UA ID YU LT EG RO BG TR RU PK MY IL" "phase:1,t:none,log,pass,msg:'High Risk Fraud Location',setvar:tx.fraud_score=+10"

This rule does not block the request but it does increase the "TX:FRAUD_SCORE" variable data.  The TX data can be increased by other rules as well, such as those doing IP Reputation checks with @rbl, and then it can be evaluated at the end of the request phase when a decisions is made after correlating all detection data. 

Per-User GeoIP Data

Another Fraud related concept that can use GeoIP data is to track the Country/City data used by each user.  You can save this data in a per-user persistent collection and simply raise the fraud_score if the data changes.  This concept can be implemented either across multiple user sessions or for possible session hijacking detection during the course of one session.

Here are some example rules that will save the GEO Country Code data in a persistent User collection.  It will count the number of times that user has connected from that same country and once it is gone over the defined threshold (10), it will then issue alerts when the use logs in from a different country.

SecGeoLookupDb /path/to/apache/conf/base_rules/GeoLiteCity.dat 
SecAction "phase:1.t:none,nolog,pass,setuid:%{args.username}" SecRule REMOTE_ADDR "@geoLookup" "phase:1,t:none,nolog,pass" 
SecRule &USER:GEO_COUNTRY_CODE "@eq 0" "phase:1,t:none,nolog,pass,setvar:user.geo_country_code=%{geo.country_code},setvar:user.geo_country_code_counter=+1" 
SecRule GEO:COUNTRY_CODE "@streq %{user.geo_country_code}" "phase:1,t:none,nolog,pass,setvar:user.geo_country_code_counter=+1" 
SecRule USER:GEO_COUNTRY_CODE_COUNTER "@gt 10" "chain,phase:1,t:none,log,pass,msg:'Geo Country Code Change for User.',logdata:'Username: %{userid}, Expected Country Code: %{user.geo_country_code}, Current Country Code: %{geo.country_code}'"
    SecRule GEO:COUNTRY_CODE "!@streq %{user.geo_country_code}" "setvar:tx.fraud_score=+10"

The final example shows how you can identify if the GEO Country Code data changes during the course of a Session (using JSESSIONID as the example).  The key is to save the GEO:COUNTRY_CODE data when the application issues a Set-Cookie SessionID and then validate it on subsequent requests.

SecGeoLookupDb /path/to/apache/conf/base_rules/GeoLiteCity.dat
SecRule REMOTE_ADDR "@geoLookup" "phase:1,t:none,nolog,pass"
SecRule REQUEST_COOKIES:JSESSIONID ".*" "chain,phase:1,t:none,pass,nolog,auditlog,msg:'Geo Country Code Change During Session.',setsid:%{request_cookies.jsessionid}"
    SecRule GEO:COUNTRY_CODE "!@streq %{session.geo_country_code}"
SecRule RESPONSE_HEADERS:/Set-Cookie2?/ "(?:jsessionid=([^\s]+)\;\s?)" "phase:3,t:none,pass,nolog,capture,setsid:%{TX.2},setvar:session.geo_country_code=%{geo.country_code}"

Hopefully the data presented in this blog post will help users to be able to better utilize GeoIP data.