A newly discovered unpatchable hardware vulnerability in Xilinx programmable logic products could allow an attacker to break bitstream encryption, and clone intellectual property, change the functionality, and even implant hardware Trojans.
"We exploit a design flaw which piecewise leaks the decrypted bitstream," the researchers said. "In the attack, the FPGA is used as a decryption oracle, while only access to a configuration interface is needed. The attack does not require any sophisticated tools and, depending on the target system, can potentially be launched remotely."
The findings will be presented at the USENIX Security Symposium later this year. The researchers said they privately disclosed the flaws to Xilinx on 24 September 2019. The semiconductor giant, in response, has posted a design advisory acknowledging the vulnerability.
"The complexity of this attack is similar to well known, and proven, DPA attacks against these devices and therefore do not weaken their security posture," the company noted in its alert.
Exploiting CBC Mode to Encrypt and Decrypt Arbitrary Blocks
FPGAs are programmable integrated circuits that can be reconfigured in-field to match a desired application or functionality depending on where it's deployed. Due to their high levels of flexibility, FPGAs have been widely used in the development of 5G mobile networks, consumer electronics, data centers, aerospace, and automotive technology,
It's worth noting that Xilinx and Intel (via its Altera acquisition) dominate the FPGA market, with Xilinx alone accounting for nearly 50% of the market share.
As FPGA designs are encoded in bitstreams, a hardware vulnerability of this scale can have severe consequences, the researchers said.
In contrast to other known side-channel and probing attacks against Xilinx and Altera FPGAs, the novel "low-cost" attack aims to recover and manipulate the bitstream by leveraging the configuration interface (such as SelectMAP or JTAG) to read back data from the FPGA device.
As a feature, the "readback" is meant to help verify that the design was correctly downloaded to the device. But in an attempt to protect the design, the bitstream is encrypted (AES-256 in CBC mode) to prevent readbacks on all external ports.
The attack devised by the researchers sets out to manipulate the encrypted bitstream to redirect its decrypted configuration data to a MultiBoot start address register (WBSTAR or Warm Boot Start Address), which allows switching between images on-the-fly for remote updates and load a fallback bitstream with a known good design into the FPGA device.
But given the use of flash memory to store these components, a reset doesn't clear the register's contents. As a consequence, the confidentiality of the bitstream can be broken as follows:
Create a malicious bitstream and a readout bitstream. The malicious bitstream exploits the malleability of the CBC encryption mode to alter the command in the bitstream, which writes data to the WBSTAR configuration register.
Load the malicious bitstream into the FPGA device
An automatic reset of the FPGA happens due to changes made to the bitstream in step (1) but doesn't reset the WBSTAR contents as it's used for the MultiBoot and fallback feature.
Readback the contents of the WBSTAR register using the readout bitstream.
Manually reset the FPGA device to repeat the above steps and recover the entire encrypted bitstream as 32-bit words.
"In summary, the FPGA, if loaded with the encryption key, decrypts the encrypted bitstream and writes it for the attacker to the readable configuration register," the researchers stated.
"Hence, the FPGA is used as a decryption oracle. The fact that only single 32-bit words can be uncovered in each iteration determines the duration of decrypting a whole bitstream: In our experiments, we are able to uncover a complete Kintex-7 XC7K160T bitstream in 3 hours and 42 minutes, for instance
In the second type of attack, the FPGA can be used to encrypt arbitrary bitstreams — once again taking advantage of the underlying CBC mode — and create a valid message authentication tag (HMAC), thereby breaking the authenticity of the bitstream as well.
According to the researchers, the attacks stem from a pitfall that the data of the encrypted bitstream header is interpreted before it's been verified, thus allowing a malicious bitstream to run on the FPGA's logic fabric.
The Flaw Cannot Be Patched
Considering that the attacks are based on the flaws in the protocol, the researchers noted that "any kind of non-trivial change to the security protocol is not possible without a redesign of the FPGA hardware and is currently not available for 7-Series and Virtex-6 devices."
In addition to recommending hardware developers to subject input data to cryptographic validation and make use of a patchable bitstream encryption engine — both of which are already in place for Xilinx's Zynq-7000, UltraScale, and UltraScale+ devices, a number of countermeasures have been proposed, such as implementing obfuscation schemes or patching the PCB to use FPGA's Revision Select pins to prevent readout from the WBSTAR register.
"We consider this as a severe attack, since (ironically) there is no opportunity to patch the underlying silicon of the cryptographic protocol," the researchers concluded. "We note that the 7-Series have a substantial share of the FPGA market, which makes it even more difficult or impossible to replace these devices."