New Spectre attack exploits CPU predictions left by replaced code

Researchers demonstrated root-password-hash leaks on Intel-based Linux systems. They report Linux and Oracle patches, while browser and runtime experiments showed important limits.

Researchers demonstrated root-password-hash leaks on Intel-based Linux systems. They report Linux and Oracle patches, while browser and runtime experiments showed important limits.
Researchers have demonstrated a new Spectre attack that can expose secrets when software replaces generated code but the processor retains predictions associated with the old code. Called Branch Target Reuse (BTR), the technique recovered the root account’s password hash in Intel-based Linux proof-of-concept attacks. The researchers report that Linux and Oracle have released mitigations.
The team, from Vrije Universiteit in the Netherlands and Scuola Superiore Sant’Anna in Italy, examined Linux’s classic BPF filtering system, Oracle GraalVM and Mozilla SpiderMonkey. Their results establish different levels of attack feasibility in each, rather than a working attack against every system studied.
Processors predict where a program will run next and may begin executing instructions before confirming that prediction. This is speculative execution. Spectre attacks use traces left by that work to infer secrets, even when the processor does not ultimately commit the predicted instructions’ results.
Just-in-time (JIT) compilers generate machine code while programs run. They can discard that code and later put different code in the same memory. The researchers found that processors can recognize the changed code without clearing old predictions about where a program will jump.
BTR exploits that mismatch. A stale prediction can send speculative execution to an unintended position in the replacement code, potentially skipping a security check or interpreting instruction bytes differently. The researchers confirmed stale-prediction behavior on all processors they tested, covering Intel, AMD and Arm. Their complete Linux attacks, however, were demonstrated on Intel systems.
On Linux, the team built two complete proof-of-concept exploits using classic BPF (cBPF), a small filtering system used in network filtering. Unlike the more powerful eBPF JIT, which is restricted to privileged users, cBPF remains available to unprivileged programs. That makes the demonstrated attack relevant to local users without administrator privileges.
The Linux demonstration recovered the root account’s password hash—a derived value used to check a password, not the plain-text password—from the “su” process. The team also bypassed cBPF’s optional constant-blinding defense, which is intended to prevent attackers from planting instructions through fixed values in generated code. These were research demonstrations, not reports of attacks against victims.
SpiderMonkey is Firefox’s JavaScript and WebAssembly engine. Here, the potential attacker is a malicious web page controlling the code that runs in the browser. The researchers’ Intel-based proof of concept showed that predictions survived code removal and replacement, but they explicitly say a complete browser exploit requires further work.
Oracle’s GraalVM provides a sandbox that restricts untrusted code, such as a plugin or user-supplied script. The researchers investigated using BTR to jump past the operation that confines memory access to that sandbox. Although code reused memory addresses reliably, GraalVM’s compilation and garbage collection—the cleanup of memory no longer needed—erased stale predictions before the team could use them in its experiments.
According to the researchers, Linux developers added an x86 mitigation that uses IBPB, a mechanism for clearing branch-prediction state, on all processor cores when a cBPF program reuses memory previously occupied by executed cBPF or eBPF code. The change also discourages that reuse. Oracle’s GraalVM mitigation randomizes JIT code-cache locations to make memory-region reuse harder.
The researchers say Linux and Oracle patches have been released, but neither report identifies affected or fixed release numbers. Users should check their operating-system or runtime vendor’s advisory for the applicable patched release; the upstream Linux change alone does not establish delivery in a particular distribution.
Mozilla considered IBPB-based mitigations but is prioritizing completion and deployment of site isolation, according to the researchers. Site isolation aims to keep different sites in separate browser processes. The team says Firefox’s incomplete deployment can leave other tabs sharing an address space with a malicious page, making their data a potential target. Whether BTR can turn that exposure into a complete browser attack remains unresolved.

Email Protection Gateway releases before 9.4.1 are affected. The security updates also fix 11 other critical vulnerabilities in Kiteworks’ Core and gateway components.
2 min

Apps using the old mailbox interface face disruption. Administrators can temporarily keep approved apps running, but access ends on April 1, 2027.
3 min

Tom Watson has renewed claims of benefits from the Federated Data Platform. The figures compare NHS performance before and after its introduction, without accounting for other factors that could explain the changes.
2 min