Broadcom says they had audited their driver code and released a patch prior to Jon Ellch's October demonstration: The flaw released in a exploitable form on Saturday by the Month of Kernel Bugs (MokB) project allows a physically proximate user to hijack a Windows and some Linux and Unix machines using certain Broadcom Wi-Fi drivers. Via email, Broadcom told me that they had not received information from Ellch prior to his Blue Hat demonstration last month. Rather, the company took seriously this August's demonstration by Ellch and colleague David Maynor of a fuzzing technique for throwing a lot of bad data at a wireless device to see what stuck (and broke). (This post was updated Nov. 16 with new information.)

While Ellch apparently told Brian Krebs of Security Fix, among others, wasn't represented correctly--his chain of events is: found flaw, demonstrated it, notified Broadcom, was told flaw was fixed, released proof with MoKB. Via email Ellch explained that he had notified Broadcom after showing a simple, early version of it at the Microsoft-sponsored Blue Hat conference when he had fleshed out the full exploit a few days later; he had found the flaw days before the event.

With Microsoft's assistance, he notified Broadcom, which, according to Ellch's email chain that he published on his site, led to Broadcom eventually telling him they'd already patched it. But they weren't quite clear in their response that the patch wasn't released yet by all their manufacturing partners, which is why Ellch was comfortable releasing the exploit when he did.

(Note: Ellch and I had a pleasant email exchange after he posted this item to his news page; my first email to him never arrived. Thank you, internets! Some of what he quotes from this post has been modified to reflect his response; other parts have been clarified as it was easy to read the previous final paragraph as accusing Ellch, rather than the MoKB, as making a poor disclosure.)

A Broadcom spokesperson, Heather Roberts, explained earlier this week that Broadcom had audited its own code following that August demo, found a number of flaws revealed in this manner, released patches to manufacturers, and themselves contacted Ellch after hearing of his demonstration. That last statement appears inaccurate based on Ellch's email archive, but it may have referred to the follow-up Ellch received after several exchanges. It's not very material, however.

Roberts said that Broadcom won't release its own list of patched drivers, because they don't want to sabotage the methods by which their manufacturing partners talk to their own customers and distribute their own necessarily customized patches. To date, only Linksys has a driver that others have confirmed fixes this flaw; it's unclear which particular computers, adapters, and software releases are affected except by examining the Broadcom driver release number on an individual computer.

Broadcom's Roberts also said that the company has not seen a demonstration of the root-level access that was claimed for this exploit, stated by Krebs as, "An attacker could use the flaw to take complete control over any vulnerable machine located within a few hundred feet." Roberts said, "Broadcom is evaluating the exploits posted to determine if they do indeed give this level of access. We do not think it's likely because  someone would need to know exactly what driver was in use to make a successful attack. Most likely, the system will just crash."

Ellch said that it's actually quite easy to weaponize this exploit for other drivers, because only the length of the SSID varies before the driver fails. Thus, using software to vary the driver length would quickly allow a successful exploit, at which point the payload could be arranged to load in the same exchange that overwhelms the driver. Ellch explained via email that there are any number of payloads, including something as simple as a "connect-back shell," that could be installed via this attack. Some payloads may require a reboot after they're executed to obtain full control. However, Ellch said the Broadcom attack would disable the Wi-Fi adapter, and I would thus expect a user to reboot in order to get on a network, thus finishing the payload's installation process.

A connect-back shell, by the way, is pretty simple: the payload launches terminal-like access by initiating an outbound connection from the compromised computer to a location of the attacker's choice. The attacker can then execute any command that the legitimate root user could.

I've criticized the MoKB elsewhere, because of their lack of requirement for an advance disclosure policy on bugs they are listing. Typically, security researchers inform a company, leave some reasonable period, then demonstrate a flaw, often leaving the actual details out until a patch is released. The MoKB project allows this but doesn't require it.

A confluence of Broadcom have already patched the driver and some miscommunication on their part makes it clear that Ellch was trying to follow good disclosure practices.