Two security researchers fleshed out details today at the Black Hat conference in Washington D.C. of a method they disclosed earlier this year for circumventing Intel Corp.'s new Trusted Execution Technology (TXT) security software.
The two-stage attack against TXT (PDF document), which is designed to protect data on PCs, was disclosed in January by Joanna Rutkowska and Rafal Wojtczuk of the security research firm Invisible Things Lab in Poland.
When first disclosed, they said they had discovered a design flaw in TXT and certain implementation errors in some associated Intel system software that allowed them to bypass any of TXT's security protections. They also released proof-of-concept code showing how an attacker could use their method to compromise Intel's implementation of the trusted boot process for Xen and Linux operating systems.
In their presentation at the Black Hat hacker conference today, the researchers offered more details on their attack method. The also noted that patching BIOS software would address the system software vulnerabilities, but added that no easy measure was currently available for tackling the TXT problem.
Intel confirmed the issue, which affects mobile, desktop, and server motherboards, “without providing any more details about which exact models are vulnerable,” the researches wrote in their presentation. “We suspect it might affect all recent Intel motherboards and BIOSes.”
Intel's TXT, previously code-named LaGrande, is a relatively new technology designed to provide a trusted way for loading and launching system software such as an operating system kernel or a Virtualization Machine Monitor on a system.
The technology is expected to greatly reduce the risk of software being compromised by system-level malware threats such as root-kits. Intel's vPro processor platform currently implements TXT, and PCs based on the technology have been shipping for about a year now.
Rutkowska and Wojtczuk said the problems with TXT stem primarily from Intel's implementation of an especially critical software component known as System Management Mode (SMM) Memory. The errors allow an attacker to infect SMM memory and inject shellcode of their choice into it, they said. For the proof-of-concept attack, they used shellcode that added a backdoor to Xen hypervisor software.
Because TXT does not validate the SMM memory while software is being loaded, any malware that is hidden in SMM survives TXT's trusted launch process and can compromise the software that has just been launched.
Once compromised in that fashion, malware could completely bypass “everything the trusted boot is supposed to provide,” Rutkowska said in an e-mail. “We can attack TXT only by combining those two [implementation and design flaws.]”
“We used our SMM attacks to compromise the TXT boot process,” Rutkowska said. “We could also use them to create SMM rootkits.”
Rutkowska and Wojtczuk said that finding a way to compromise SMM is not easy, particularly on modern systems. But they said they have found several implementation errors in SMM that could allow an attacker to potentially introduce malicious code.
Specific details of some of the new SMM attacks will be presented later this year at the Black Hat conference in Las Vegas, after Intel has had a chance to patch the underlying vulnerabilities.
Rutkowska said Intel was in the process of patching its BIOS software to deal with vulnerabilities in the SMM software. But she said dealing with the TXT problem could be slightly trickier because it involves the development of something called the SMM Transfer Monitor (STM), “a special hypervisor that should be written by OEM/BIOS vendors,” in addition to Intel, also a BIOS vendor.
Spokesman George Alfs confirmed Intel was working with the researchers on the issue. “We are not aware of any active exploits in the wild as described in their research, but we take all reports seriously,” Alfs said.