Why crawling doesn't matter
This post isn't intended to be a retort to Jeremiah Grossman's post last month on Why crawling matters, but more of a follow-up post to my latest blog entry on Why pen-testing doesn't matter. Hint: both pen-testing and crawling are still important/matter, but my CPSL process described in my last post leans towards a new, idealistic approach on the vulnerability problem.
I've covered web crawling before, in Scraping the web for fun and profit as well as when I mentioned papers written by Shreeraj Shah in my longest-yet post on 2007 Security Testing tools in review. Ajax crawling was only partially covered, so we'll continue to expand on that topic (as well as other topics) in this blog entry. Another motivator for posting information about crawling is that I'm seeing some repeated information over at blogs such as securitytechscience.com/blog -- especially in their latest on Screen Scraping.
Securitytechscience is interesting, because they are trying to build a web application security scanner. In their "Screen Scraping" article, they mention a TBD release of sts-scanner, and the author hints at a future solution for crawling Web 2.0. It will be interesting to hear more about what their solution is, although judging by their language (Ruby), library (Mechanize with Hpricot and/or RubyfulSoup), and available resources (i.e. downloadable source code) -- we can make some assumptions.
Also in the "Screen Scraping" article, the author makes mention that Wapiti is using RubyfulSoup? Maybe I took this out of context, but he says, "Rubyfulsoup is also a very powerful Ruby screen scraping package, offering similar parsing structures like HTree but without the complex XPath query language. I noticed that Wapiti recently incorporated this into its package". I think maybe he means BeautifulSoup, since Wapiti is based on Python. I do fail to understand how XPath is complex, and it's also very rarely slow (notable exception: Selenium Core's XPath expressions in IE are incredibly slow because they are non-native, more detail on this below).
Casey, a tssci-security local, brought up a comment in my post on "Scraping the web for fun and profit" that pointed me to an article on issues with XPath in the Firequark section entitled XPath vs. CSS Selector. While Casey may be onto something using scrapi (even though it is limited to using CSS expressions), this Firequark article appears to be somewhat wrong when it comes to comparing XPath and CSS expressions. Dealing with broken markup is simply something that needs to be dealt with if you're using an application driver. However, it appears that Firequark is browser driven, so while not wrong -- it still fails forward thinking behavior. Firequark (having "fire" in the name, and being based on Firebug) is likely a Firefox add-on, so it will only work in Firefox. Of course, CSS selectors are much slower than the native DOM expressions built into Firefox version 3 (by a lot!) as we'll come to see later in this post.
People who have no formal background in development or quality testing usually refer to protocol drivers as Web 1.0 crawlers and application drivers as Web 2.0 crawlers. Often, Web 1.0 crawlers are "good enough". Assuming there is no Ajax or other active client-side logic, there really isn't a need to crawl with an application driver. Protocol drivers are much less overhead, and therefore can be much faster. Building a speedy Web 1.0 crawler can be best accomplished by using Perl's FEAR::API, which can be even further enhanced by using larbin to provide C++ parallel crawling. This would be nice functionality to include for any web application security scanner.
As mentioned earlier, Shreeraj Shah has some articles on crawling Ajax and Web 2.0. In one article, he uses RBNarcissus (an application driver) for a basic example, and then moves into an extended example using Watir. Watir is a browser driver, but only for Internet Explorer, and only using the Ruby language. Confusingly, I discovered at a Java Users Group meeting that Watij and WatiN both exist, to drive IE from Java and .NET respectively. Firefox functional testers felt left out, so some clever person wrote FireWatir (good name, too) to drive Firefox browsers, but only using Ruby. The thought crossed my mind that it would be better if somebody would write a browser driver that worked with multiple-browsers using multiple-languages, because learning four completely different tools that only drove one specific browser seemed like quite a bit of a learning curve, and wasted time. This doesn't appeal to lazy programmers, does it?
After playing with RBNarcissus, Watir, and Selenium Core -- it's hard to pick up traction about where to go next (although Shreeraj and OpenQA both led me towards Sahi, I skipped it over as it appeared too similar in its approach to Selenium). HtmlUnit isn't exactly a functional testing tool or approachable as an Ajax crawling solution (at least not just yet). I really liked the Selenium IDE about a year ago, which is a great little Firefox extension. After checking out the NIST SAMATE Web Application Vulnerability Scanners project page, I noticed a link that I hadn't seen before under Free/OpenSource Tools: TestMaker by Pushtotest. After further exploration, I discovered that another tool, TestGen4Web (another Firefox extension, be sure to check out the picture on that page), was capable of translating between HttpUnit, SimpleTest (i.e. TestMaker), and Selenium test cases. Yet I was still looking for something better.
After reading the Jason Huggins' blog (the author of Selenium) one day, he pointed attention towards a new project, WebDriver -- which is basically like using HtmlUnit as a standalone application. The biggest draw to WebDriver is that it not only can act as a very powerful application driver, but also has drivers specifically for Firefox and IE. This provides all the benefits of both application drivers and browser drivers: full browser JSE and plugin support through the browser drivers, while the speed and extensibility of an application driver. After exploring WebDriver, I later learned that Jason Huggins moved from ThoughtWorks to Google, where he built a grid testing system using Selenium RC. Apparently Selenium RC Grid is both very buggy and very cool!
As we proceed through this blog post, I hope that you are following along with what I've learned about crawling the web. The answer is that there is no universal answer. Every web application and security test is different: different size, different code requirements, different tests require different and multiple crawls, etc. So using one method of crawling, or one tool is by definition a bad approach. Crawling is not a "one-size fits all" program that you build. Hence, web application security scanners are inherently flawed because they fail to take different scanning engines into account for different reasons. Note that at least one commercial web application security scanner will allow customization to change it from Depth-first search (DFS) mode to Breadth-first search (BFS) mode, but this is only scratching the surface.
Consider for a second that you are attempting to crawl the "Dark Web". You've been tasked as a security professional to scope out what the bad guys are doing. First of all, you're going to want to protect yourself, maybe get yourself your own hijacked IP and ASN from APNIC and announce them out randomly changing BGP-speaking routers. Unlike phishers, you probably won't need to bother with DNS, so you won't need fast-flux networks except for maybe use in your proxy infrastructure. Your enemy is likely to start figuring out that you are crawling his important terrorist websites and will put up protections. So what works best, DFS or BFS in this situation? Neither, the best answer is to crawl by making yourself look like a legitimate dark guest. A few years ago I attended the LayerOne conference in Pasadena, CA -- where I had the pleasure to meet Billy Hoffman and attend his talk on Covert Crawling.
Covert crawling may not be enough, and if an after-thought may be "too little; too late". Consider wily attackers such as Ronald, who would consider strike-back attacks such as Spidering the Spiders. One of my favorite web application security blogs, Palisade, did a two part article on Dodging the spiders and More on dodging spiders. They also include a lot of great links and resources; I especially like the PDF on Stopping automated attack tools. The legality of crawling (especially crawling with attacks) comes into play, especially during strike-back, but what if you are able to purposefully or accidentally delete an entire site's content? Can a spider enter into a binding contract? This case was debated in a Colorado federal district court in `Internet Archive v. Shell <http://blog.ericgoldman.org/archives/2007/03/can_a_spider_en.htm>`_. Be especially careful when crawling financial institutions, especially if you're verifying a vulnerability in a two-factor authentication system in use for 13 million banking customers i.e. the BofA PassMark/SiteKey service.
Many people I know cite the primary reason for crawling (outside of search engines, which is everyone's obvious first answer) is to simply get an idea of the size and layout of a website. David Kierznowski discusses the benefits and drawbacks of using Microsoft Visio to spider a website on his blog, while Daniel Cuthbert always points people towards Freemind, a mind-mapping tool. My personal answer to this is to just ask the developers to do a walkthrough of the application with you, as they usually have understanding and comprehension of their application in ways that diagrams and mind-maps may never elicit. I often use other tools, services, and Firefox add-ons, such as The Wayback Machine, Netcraft, AboutURL, The Scrutinizer, myIPNeighbors, SpiderZilla, Slogger, Goosweep, Paterva, Googlegath, and Fierce. Sometimes I go really deep and start digging through whois-servers.net, ARIN, RWhois, RADB, and trash cans. Of course this list would not be complete without Johnny Long's `Google Hacking for Penetration Testers <http://isbn.nu/9781597491761>`_, which is about to release a second volume/edition slated by the end of the year. I'm curious to find out if they cover any material on Using Google Analytics to Subvert Privacy or Google Analytics: Now with Stolen Search Queries -- and I also expect to follow up these posts with new material, regardless of how it's covered in the new book.
pdp and John Resig have some advanced crawling techniques that are worth additional mention. pdp posted the Top 5 most popular Web2.0 services hackers cannot live without and followed up with another post on 5 generic Yahoo Pipes hackers cannot live without. John also posted information about Selector Speeds and followed up with a post detailing the getElementsByClassName Speed Comparison, leading to additional comparisons on the isobaric blog.
This about covers the latest and greatest in web crawling technology. Before you pick a language, scraping module, and begin writing security checks -- let me point you towards one more resource. Courtenay of the Caboose blog released a Ruby-based spider fuzz plugin in February, along with three updates (April, August, and October). This work seems to go beyond that of what Romain Gaucher was working on with his coverage tool, so be sure to take a look and integrate these concepts into your own future crawlers.
I also have an interest in crawling/walking Flash and RIA apps, so if you have any pointers, please do let us know! I know how to extract links ("grep" with swfdump -D or similar), but any further information/ideas would be very useful. Something like the opposite of hxASM (for disassembling from memory) or at least something better than Flare (for decompiling) would be very interesting. The flashsec.org software list is great, but I still haven't found anything that works well for this (AMF logs from Charles, maybe?). Stefano Di Paola has also yet to release the JAMFProxy and SWFRTAnalyzer tools, but according to his OWASP AppSec presentation, the SWF Runtime Analyzer will be made available on the Minded Security website. Looking forward to this and any other innovations in web application crawling!blog comments powered by Disqus