Foreshadow returns to the foreground: Boffins find secrets-spilling speculative execution flaw lives on
(2020/08/07)
- Reference: 1596758411
- News link: https://www.theregister.co.uk/2020/08/07/foreshadow_strikes_back_boffins_find/
- Source link:
Some of the boffins who in 2018 disclosed the data-leaking speculative-execution flaws known as Spectre and Meltdown today contend that attempts to extinguish the vulnerabilities' Foreshadow variant have missed the mark.
In a paper slated to be distributed through ArXiv any time now, Martin Schwarzl, Thomas Schuster, and Daniel Gruss with Graz University of Technology, and Michael Schwarz, with the Helmholtz Center for Information Security, reveal the computer science world has misunderstood the microarchitectural flaw that enables [1]Foreshadow , which can be exploited by malware or a rogue user on a vulnerable system to [2]extract data from supposedly protected areas of memory – such as Intel SGX enclaves, and operating-system kernel and hypervisor addresses.
The paper, Speculative Dereferencing of Registers: Reviving Foreshadow, details how defenses based on this misunderstanding, such as the Meltdown mitigation known as KAISER, don't really work against Foreshadow.
"We discovered that effects reported in several academic papers over the past four years were not correctly understood, leading to incorrect assumptions on countermeasures," said Daniel Grus, assistant professor in the Secure Systems group at the Graz University of Technology, in an email to The Register . "The consequence is that we are able to mount a Foreshadow attack on older kernels patched against Foreshadow with all mitigations enabled and on a fully patched kernel if only Spectre-v2 mitigations are disabled."
One way those defenses might be disabled would be if, in the case of Linux, the kernel flag nospectre_v2 has been set on a server in an attempt to optimize for speed over security.
Foreshadow – an L1 Terminal Fault ( [3]L1TF ) bug in Intel parlance – involves a speculative execution attack to gather private data from an Intel SGX enclave's L1 data cache via a side channel.
It relies on manipulating how information gets prefetched – collected before it's needed – from the cache. But this anticipatory data gathering doesn't take place in the way computer scientists had thought.
Meltdown The Sequel strikes Intel chips – and full mitigation against data-meddling LVI flaw will slash performance [4]READ MORE
"We discovered that this prefetching effect is actually unrelated to software prefetch instructions or hardware prefetching effects due to memory accesses and instead is caused by speculative dereferencing of user-space registers in the kernel," the paper explained.
This misinterpretation of the root cause of the vulnerability has several implications. First, the paper explained, the assumption that Meltdown leaks from the L3 cache or main memory while Foreshadow only leaks from the L1 cache is incorrect.
For both Meltdown and Foreshadow, the data has to be retrieved from the L1 cache, but the use of prefetch gadgets – code constructs found in memory – allows that restriction to be bypassed. That means, the paper stated, that Foreshadow attacks can be mounted on data outside the L1 cache (e.g. the L3 cache) on kernels with the necessary gadgets.
It also means Foreshadow can affect CPU cores from vendors other than Intel, such as AMD, Arm, and IBM, and that the effects can be triggered by JavaScript, though fairly slowly based on the scenarios described in the paper.
The good news is that there is a defense against the Foreshadow L3 attack described in the paper: implementing the Spectre-BTB (Branch Target Buffer) countermeasures.
"Our results show that, for now, retpoline [a defense against poisoned branch target buffer entries] must remain enabled even on recent CPU generations to fully mitigate high impact microarchitectural attacks such as Foreshadow," the paper concluded.
It's thus not surprising that Intel isn't planning on releasing further fixes. "The paper refers to specific scenarios in the Kernel-based Virtual Machine (KVM) where there are options on implementing mitigation settings," an Intel spokesperson said in an email. "Intel recommends enabling both the V2 and L1TF mitigations for additional protection. Intel is not planning to release additional mitigations."
AMD, which has said customers shouldn't need to apply Foreshadow defenses, did not respond to requests for comment. ®
Get our [5]Tech Resources
[1] https://www.theregister.com/2018/08/14/intel_l1_terminal_fault_bugs/
[2] https://foreshadowattack.eu/
[3] https://www.intel.com/content/www/us/en/architecture-and-technology/l1tf.html
[4] https://www.theregister.com/2020/03/10/lvi_intel_cpu_attack/
[5] https://whitepapers.theregister.com/
In a paper slated to be distributed through ArXiv any time now, Martin Schwarzl, Thomas Schuster, and Daniel Gruss with Graz University of Technology, and Michael Schwarz, with the Helmholtz Center for Information Security, reveal the computer science world has misunderstood the microarchitectural flaw that enables [1]Foreshadow , which can be exploited by malware or a rogue user on a vulnerable system to [2]extract data from supposedly protected areas of memory – such as Intel SGX enclaves, and operating-system kernel and hypervisor addresses.
The paper, Speculative Dereferencing of Registers: Reviving Foreshadow, details how defenses based on this misunderstanding, such as the Meltdown mitigation known as KAISER, don't really work against Foreshadow.
"We discovered that effects reported in several academic papers over the past four years were not correctly understood, leading to incorrect assumptions on countermeasures," said Daniel Grus, assistant professor in the Secure Systems group at the Graz University of Technology, in an email to The Register . "The consequence is that we are able to mount a Foreshadow attack on older kernels patched against Foreshadow with all mitigations enabled and on a fully patched kernel if only Spectre-v2 mitigations are disabled."
One way those defenses might be disabled would be if, in the case of Linux, the kernel flag nospectre_v2 has been set on a server in an attempt to optimize for speed over security.
Foreshadow – an L1 Terminal Fault ( [3]L1TF ) bug in Intel parlance – involves a speculative execution attack to gather private data from an Intel SGX enclave's L1 data cache via a side channel.
It relies on manipulating how information gets prefetched – collected before it's needed – from the cache. But this anticipatory data gathering doesn't take place in the way computer scientists had thought.
Meltdown The Sequel strikes Intel chips – and full mitigation against data-meddling LVI flaw will slash performance [4]READ MORE
"We discovered that this prefetching effect is actually unrelated to software prefetch instructions or hardware prefetching effects due to memory accesses and instead is caused by speculative dereferencing of user-space registers in the kernel," the paper explained.
This misinterpretation of the root cause of the vulnerability has several implications. First, the paper explained, the assumption that Meltdown leaks from the L3 cache or main memory while Foreshadow only leaks from the L1 cache is incorrect.
For both Meltdown and Foreshadow, the data has to be retrieved from the L1 cache, but the use of prefetch gadgets – code constructs found in memory – allows that restriction to be bypassed. That means, the paper stated, that Foreshadow attacks can be mounted on data outside the L1 cache (e.g. the L3 cache) on kernels with the necessary gadgets.
It also means Foreshadow can affect CPU cores from vendors other than Intel, such as AMD, Arm, and IBM, and that the effects can be triggered by JavaScript, though fairly slowly based on the scenarios described in the paper.
The good news is that there is a defense against the Foreshadow L3 attack described in the paper: implementing the Spectre-BTB (Branch Target Buffer) countermeasures.
"Our results show that, for now, retpoline [a defense against poisoned branch target buffer entries] must remain enabled even on recent CPU generations to fully mitigate high impact microarchitectural attacks such as Foreshadow," the paper concluded.
It's thus not surprising that Intel isn't planning on releasing further fixes. "The paper refers to specific scenarios in the Kernel-based Virtual Machine (KVM) where there are options on implementing mitigation settings," an Intel spokesperson said in an email. "Intel recommends enabling both the V2 and L1TF mitigations for additional protection. Intel is not planning to release additional mitigations."
AMD, which has said customers shouldn't need to apply Foreshadow defenses, did not respond to requests for comment. ®
Get our [5]Tech Resources
[1] https://www.theregister.com/2018/08/14/intel_l1_terminal_fault_bugs/
[2] https://foreshadowattack.eu/
[3] https://www.intel.com/content/www/us/en/architecture-and-technology/l1tf.html
[4] https://www.theregister.com/2020/03/10/lvi_intel_cpu_attack/
[5] https://whitepapers.theregister.com/