Emulated fault injection bypasses CHERI-RISC-V memory-safety checks
Original title: CHERI under Fault Injection: An Emulation-Based Security Evaluation of Capability Hardware Enhanced RISC Instructions
One or two targeted faults enabled overflows, control-flow hijacks, data-only attacks, and use-after-free in QEMU.
A thesis tests whether fault injection can break CHERI's spatial and temporal memory safety in a pure-capability CHERI-RISC-V environment, a threat the architecture's model excludes. Using QEMU, the authors flip capability-metadata bits and skip checks on capability store instructions. Four scenarios succeed with one or two faults, and check skipping alone can hijack control flow or enable data-only attacks with one fault. CHERI still requires fault injection and a valid tagged capability or a second fault because of sentries. Proposed defenses include ECC-protected capability memory, field authentication codes, redundant checks, and revocation-bitmap verification.
Unconfirmed: Reported fault counts are lower bounds from emulation, not measurements on physical hardware.
- Bit flips corrupt capability metadata; a second class skips capability-store checks.
- Check skipping needs a minimum of one fault for two attack classes.
- A single fault can bypass spatial safety, temporal safety, or control-flow integrity.
- Sentries force a valid tagged capability or a second fault to restore the tag.
Impact: One fault can bypass spatial, temporal, or control-flow safety
| Product | Affected |
|---|---|
| CHERI-RISC-V | pure-capability environment |
- 1.Consider ECC on capability memory and authentication codes for capability fields.
- 2.Redundantly execute security-critical checks and verify revocation-bitmap integrity.
From the original article
1 min read at arxiv.orgMemory-safety vulnerabilities continue to dominate real-world exploit chains despite decades of research into software mitigations. CHERI is a hardware architecture that enforces spatial and temporal memory safety by augmenting pointers with hardware-checked metadata, thus preventing software-based exploitation, but its threat model explicitly excludes fault injection.
By Eva Sophia Schreyer · arxiv.org
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.