Researchers at the Columbia University School of Engineering and Applied Science have showcased two new approaches to providing computers with memory protection without sacrificing performance – and they're being implemented in silicon by the US Air Force Research Lab.
Modern processors are things of magic, but like any magic they can sometimes work in unexpected ways. Take the Spectre and Meltdown families of vulnerabilities, for example: speculative execution frameworks added to improve performance have turned into a boon for ne'er-do-wells looking to access secrets hidden in supposedly protected memory regions.
In the years since their disclosure, fixes for Spectre, Meltdown, and a whole host of related vulnerabilities have been released. For some workloads, though, the cure can be worse than the disease: a report released this week found specific workloads running 1.6-2x slower than on the same platform without the fixes in place.
"Memory safety has been a problem for nearly 40 years and numerous solutions have been proposed," said Simha Sethumadhavan, associate professor at Columbia University.
"We believe that memory safety continues to be a problem because it does not distribute the burden in a fair manner among software engineers and end-users. With these two papers, we believe we have found the right balance of burdens."
"No-FAT and ZeRØ are two major steps toward putting an end to a longstanding problem," agreed Miguel Arroyo, PhD, co-lead author of the two papers. "Memory safety attacks cost the cyber community millions of dollars. Now we can avoid that and keep everyone's data safe – it's a win-win!"
The first of the papers, ZeRØ (you can watch the presentation here), presents a novel set of memory instructions and a new encoding scheme for metadata, designed to protect both the code and the pointers of a computing system. Requiring, its inventors have claimed, only minor changes to the architecture and being easy to add to modern processors, ZeRØ provides protection with zero measured performance loss – hence the name.
"ZeRØ offers memory security at no cost and it is a perfect complement to systems that mitigate memory attacks," said fourth-year PhD student Mohamed Tarek, co-author on the papers. "The keys to widespread adoption of security techniques are low performance overhead and convenience."
The second paper, No-FAT, aims to reduce the overhead of checking memory safety in the first place. Designed to dramatically boost the speed of fuzzing analyses, an automated approach to finding security vulnerabilities, No-FAT uses memory binning to improve performance.
- Do you want speed or security as expected? Spectre CPU defenses can cripple performance on Linux in tests
- TimeCache aims to block side-channel cache attacks – without hurting performance
- Google emits data-leaking proof-of-concept Spectre exploit for Intel CPUs to really get everyone's attention
- Intel CPU interconnects can be exploited by malware to leak encryption keys and other info, academic study finds
As an added bonus, No-FAT introduces protections of its own against certain types of speculative execution vulnerabilities – including the original Spectre variant – at a very low performance cost, measured at around 8 per cent on the SPEC CPU2017 benchmark. Better still, it can boost the performance of garbage collection in memory-safe languages and may provide a platform for enhancing the predictability of memory prefetch and dynamic RAM (DRAM) controllers.
Premium on ideas
The lofty claims made by the researchers appear to have caught the attention of the US military complex. The work was partially funded with grants from the Office of Naval Research and the US Air Force, along with a Qualcomm Innovation Fellowship and what the university describes as "a gift from Bloomberg," and is being implemented in a processor design at the Air Force Research Lab.
"Ransomware is currently impacting businesses all across the globe and is getting out of hand," ESET UK cybersecurity expert Jake Moore told The Register. "The usual methods of mitigating impact are clearly not up to scratch or keeping up with the standards required. Therefore, any new ideas are hotly contended and happily received.
"With a constant deluge of attacks this new research looks, in theory, to be advantageous in thwarting attempts – but the real proof is in the pudding, and only time will tell if it can withstand the barrage of daily attacks. However, what we have learnt is that threat actors are persistent in their ways and circumnavigation is in their blood."