October 18, 2009, 10:12 PM — I know that chances are no one rushed to remove all SQL injection vulnerabilities from their Web applications after I warned in my column last month how serious they can be . I know that most people, even if they recognized the importance of the vulnerability, would consider how much work was involved and conclude that there must be an easier way. Perhaps they'd think that their network intrusion-detection system (NIDS) could handle the job.
Well, I'm back this month to puncture that balloon.
NIDSs are largely ineffective against a determined attacker using techniques like SQL injection attacks.
As I said last month, attackers love targeting SQL services because that's where some of the most interesting stuff resides in many applications. Now imagine a high-value target coupled with security tools that largely fail to detect attacks. How is this possible? Think back to my description of the dreaded "or 1=1" attack.
It's trivially easy to tell our NIDSs to look out for the "or 1=1" string. But don't be fooled, since "2=2" and "3=3" work equally well. For that matter, so do "foo=foo" and "xyzzy=xyzzy." See where this is going?
That's right, using static signature definitions -- like the ones used by many popular NIDS products today -- will fail every time, since you'd need to write an infinite number of signature strings to recognize every attack. And those "n=n" attacks are the most trivial of SQL injection attacks at that.
The core of the problem is that all network-based intrusion-detection tools lack the context to be able to effectively determine whether or not an HTTP request contains malicious inputs.
And let's not stop there -- there's another dark truth of the IDS world lurking about. You know how we security folks have been telling the software folks to use SSL encryption to secure sensitive data as it traverses the network? Well, encrypting the network traffic -- which is a good practice -- effectively puts a lens cap on our surveillance cameras. NIDS products cannot see inside an SSL-encrypted packet any more than our adversaries can.
By SSL-encrypting our sensitive data, we're robbing our NIDS sensors of the ability to look into the data for signs of malicious payloads. Even if they were capable of detecting SQL injection (and other attacks) all the time, SSL would prevent that from working.
So does that mean we should toss in the towel and stop using NIDS and SSL? Of course not. It does mean, though, that we need to be keenly aware of the strengths and the limitations of the technologies we're using.
There remain many things that NIDS tools are extremely effective at detecting. Outbreaks of (known) viruses, worms, malware, etc., can stand out like a sore thumb on a network that is being monitored using NIDS tools.
Even many novel attacks can be detected using NIDS tools. For example, botnet and other malware often installs network services on infected computers. When attackers connect to these services, a desktop computer behaves -- at a network level -- much like a "server." Most modern NIDS tools will notice that sort of network activity even when it's caused by previously unknown malware.
But don't be fooled for a moment that NIDS products are effective at detecting all Web application-level attacks. The only place to put real application security measures is inside the applications themselves. That kind of security cannot be bought and bolted on post facto.
Similarly, we mustn't give up on SSL either. It remains the most ubiquitous and accepted network encryption technology available today. We really must use it to protect sensitive data in transit.
Does that mean our NIDS products are doomed to never be able to pierce the SSL veil? In many or most cases, yes it does. Now, in enterprise-class architectures, we can offload the SSL processing to front-end processors and place our NIDS sensors behind those processors, but that may well be beyond the reach of many smaller companies.
We can also consider host-based intrusion detection systems (HIDS) on our application processors, but be warned that many HIDS products are woefully inadequate at detecting application-layer attacks. Again, there is no substitute for placing application-layer defenses -- including intrusion detection -- inside an application.
And let's not also neglect another hugely beneficial aspect of using SSL -- authentication. In most applications, only the server authenticates to the client, but even that is a value-added service. And in cases where client certificates are used in addition to server certificates, SSL provides us with seriously strong mutual authentication.
