Threat Research & Response Blog
As Vinny Gullotto, our GM blogged earlier in the week, the 11th edition of the Security Intelligence Report (SIRv11) has been released. One of the new areas of research in this release is a study of the most prevalent kinds of vulnerability exploitation and how much of that exploitation is 0-day (short for zero-day, an attack or exploitation of a vulnerability without an available update). We took two paths to find this answer. The first was an analysis of how the top families found by the Microsoft Malicious Software Removal Tool (MSRT) were known to infect systems. We found that none of the top 27 families were known to use 0-day vulnerabilities in 1H11.
The second way we approached this answer was to measure all of the exploit activity tracked by the MMPC through our real-time protection products (such as Microsoft Security Essentials and Forefront Endpoint Protection) and compare the number of attacks that were 0-day at the time (no update available) versus attacks that occurred after the update was made available. We actually gave a month buffer zone (so any exploits that happened during the month in which the update was made available was still counted as 0-day). We expected the percentage to be low, and it was– 0.12 percent to be exact for 1H11. Here's what it looks like in chart form:
Chart 1 - Chart illustrating percentage of exploits that were 0-day in 1H11
One question that we discussed a lot while working on this report was: How do we measure what we don't know and therefore can't see? (In other words, 0-day by definition means you may not know about it.) Great question! Answer: We can't measure what we can't see. However, what we have seen tells us that "secret 0-days" don't stay a secret for very long. Take, for example, a few we tracked in 2010. These attacks nearly always started out as targeted – sometimes reported as affecting only one entity when they were discovered. The trend they have in common is that they broaden to more generalized use (eventually) and we find out about them sooner or later.
The point here is that although it's true that "you don't know what you don't know," our experience tells us that when it comes to 0-day activity, we find out, and often, we find out quite quickly. Things start to unravel rapidly the moment the 0-day affects either a target that's really paying attention or when the attacks start to affect a broader, less targeted audience.
So, even if our estimates for 0-day activity were off by 5 fold, the estimated activity for 1H11 would remain under 1 percent. That's still pretty small.
Most Frequent Exploits
So, now that the question of 0-day is out of the way, let's talk about the broader volumes of exploit activity that were revealed in SIRv11. Although there are many interesting trends in the chart below, I want to focus on a few of them in this blog: Java (and the age of vulnerabilities in general) and Operating System vulnerabilities. If you want details about the other categories in this chart, see the full Security Intelligence Report.
Chart 2 - Exploit activity over a one year period
As we blogged a year ago, in 3Q10, the exploitation of Java vulnerabilities skyrocketed to new levels that we had never seen before. The analysis in SIRv11 shows that Java exploitation remains high and that the targeted vulnerabilities are quite old. The top four Java exploits are CVE-2010-0840, CVE-2008-5353, CVE-2010-0094, and CVE-2009-3867. These CVEs affect the Oracle Sun Java JRE or JDK, and all of them have updates available to fix them now. The most recent, CVE-2010-0094 and CVE-2010-0840, received updates in April 2010 after following a coordinated disclosure process with an external vendor.
Operating System Exploits
The jump in operating system exploits is primarily due to one technique: CVE-2010-2568 (the vulnerability mentioned earlier that was found with Stuxnet). This exploit was picked up by a number of families that were known to abuse Autorun. And, although CVE-2010-2568 has nothing to do with Autorun itself, the behavior is quite similar: the user connects to a USB device and browses the drive, the malware automatically executes (if the user hasn't applied the update to fix the issue, that is). Malware authors must have found this exploit technique alluring. At least, the data certainly seems to indicate that they did. It's also possible that attackers, after Microsoft released updates to harden the Autorun feature on older systems (which did appear to put a dent in their ability to infect users), were searching for ways to broaden their infection rate.
Another interesting aspect in our exploit data on CVE-2010-2568 is the location of the targets. I recently did a talk at Virus Bulletin on the top exploits of 2011, and in that talk, I looked at geographical differences for regions that face the most exposure to exploitation attempts. Several regions that were at the top, Indonesia, Pakistan, and Vietnam, were there because of exploitation attempts for CVE-2010-2568. If you combine those three locations with two more, India and Mexico, those five together represent 52% of all the computers that have reported CVE-2010-2568 attack attempts in the first three quarters of this year. Although I don't have update statistics for these regions, this data might indicate that there are large numbers of systems there that have not yet applied this very important update (MS10-046).
I've talked about a lot of data in this post, and sometimes it's hard to synthesize it. The key point of the exploit analysis in SIRv11 is that older vulnerabilities are what the vast majority of exploitation attempts target (90 percent are more than a year old). The special 0-day section of the report takes this concept even further – we look at how much of the malware infections are actually attributed to the exploit of vulnerabilities in general. (The answer: Less than 6 percent in 1H11.) To find out what the other 94 percent of infections are attributed to, download the report and keep your eye on this blog for more analysis to come.
- Holly Stewart, MMPC