TLBleed is latest Intel CPU flaw to surface: But don't expect it to be fixed
TLBleed is latest Intel CPU flaw to surface: But don't
expect it to be fixed
Researchers find a new side-channel attack against a
performance-enhancing feature in Intel CPUs.
By Liam Tung | June 26, 2018 -- 12:39 GMT (05:39 PDT)
Intel won't be patching a newly revealed side-channel
vulnerability in its CPUs, even though it could be used to leak encryption keys
for signing a message.
The flaw, which will be presented at the Black Hat USA
2018 conference, is why OpenBSD recently decided to disable hyperthreading on
Intel CPUs.
The OpenBSD project's chief, Theo de Raadt, said he
dropped support for the feature after viewing the paper from researchers at the
Systems and Network Security Group at Vrije Universiteit Amsterdam.
The Register reported on Friday that the paper details an
attack on Intel's Hyper-Threading technology to reliably extract a 256-bit
EdDSA encryption key used for cryptographically signing data.
The researchers argue that their attack, dubbed TLBleed,
is able to leak the keys from another program in no less than 98 percent of
tests, depending on the Intel CPU architecture. The leak happens when the key
is being used to sign data.
As the attack relies on Intel's Hyper-Threading, this
side-channel flaw differs from Spectre and Meltdown, which exploit speculative
execution. Intel's Hyper-Threading technology is available on Intel Core, Core
vPro, Core M, and Xeon processors.
In a publicly available summary, the researchers note
that the side-channel attack leaks information from the Translation Lookaside
Buffer (TLB), a special type of memory cache that stores recent translations
that map virtual to physical memory addresses.
If Hyper-Threading is enabled, a single core can execute
multiple threads simultaneously for performance gains, but that core also
shares the same memory caches and TLB.
The attack makes it possible for one thread to see how
another accesses the CPU through TLB and use this information to work out
secrets from another program stored in shared RAM.
"Our TLBleed exploit successfully leaks a 256-bit
EdDSA key from cryptographic signing code, which would be safe from cache
attacks with cache isolation turned on, but would no longer be safe with
TLBleed. We achieve a 98 percent success rate after just a single observation
of signing operation on a co-resident hyperthread and just 17 seconds of
analysis time."
The researchers say their attack is able to extract this
key while a program is signing a message with the libgcrypt cryptographic
library.
However, to exploit the flaw, an attacker would already
need to have malware running on a target system or be logged in. But the
vulnerability could pose a threat to virtual machines on a public cloud, which
could be exploited from another instance on the same machine.
Intel appears unlikely to patch the bug and did not award
the researchers payment under its side-channel bug bounty. The company has said
its cache attack protections are sufficient to block TLBleed attacks.
However, Ban Gras, one of the researchers behind TLBleed,
said in a tweet that the attack shows that cache side-channel protections, such
as cash isolation, are not enough.
Intel told ZDNet that it had been made aware of the Vrije
Universiteit research and TLBleed, which it stressed is unrelated to Spectre or
Meltdown.
"Research on side-channel analysis methods often
focuses on manipulating and measuring the characteristics (eg, timing) of
shared hardware resources. These measurements can potentially allow researchers
to extract information about the software and related data," Intel said in
a statement.
"Software or software libraries such as Intel
Integrated Performance Primitives Cryptography version U3.1, written to ensure
constant execution time and data independent cache traces, should be immune to
TLBleed."
Comments
Post a Comment