Snapshotting our way to CVE
Reporting by Quarkslab BlogRead the original at blog.quarkslab.com
Executive Summary
A new LibAFL-Qemu based fuzzer was developed to investigate complex state machines, specifically targeting libslirp's TCP code to explore the vulnerability CVE-2026-9539. This vulnerability is described as a heartbleed-like Out-Of-Bound (OOB) Read, allowing data leakage across connections due to an integer underflow condition within libslirp's TCP handling logic. A key challenge noted is that the OOB read occurs within a custom allocator, rendering traditional sanitizers like Address Sanitizer (ASAN) blind to the issue.
The authors present two primary methods for detection: using UndefinedBehaviorSanitizer (UBSAN) to catch the integer overflow side effect, or implementing manual memory poisoning/unpoisoning techniques with ASAN to detect the OOB access internally. To manage the stateful nature of TCP fuzzing, the fuzzer employs a custom mechanism inspired by FitM, utilizing LibAFL-Qemu's SnapshotManager. This allows the system to capture internal state changes based on hooks monitoring the TCP state field, enabling the exploration of different state transitions incrementally, much like state machine traversal.
Facts Only
The fuzzer was developed by Jérémy Marchand.
The target of the fuzzer is libslirp's TCP code.
The vulnerability investigated is CVE-2026-9539, an OOB Read in libslirp.
CVE-2026-9539 was reported to libslirp in May 2026.
The vulnerability involves an integer underflow while testing the TCP urgent pointer against an intermediate buffer size.
The OOB read leaks data from other connections, including leftovers from previously freed memory.
The bug remains invisible to Address Sanitizer (ASAN) because the read occurs inside a custom allocator.
Detection can be achieved via UndefinedBehaviorSanitizer (UBSAN) for the integer overflow side effect.
The fuzzer utilizes LibAFL-Qemu's SnapshotModule and SnapshotManager for incremental state exploration.
The fuzzer uses a lain-based custom mutator to generate TCP packets.
A limitation noted is that QASAN does not function with multiple snapshots.
Full Take
The research demonstrates a significant evolution in vulnerability discovery, moving beyond simple memory corruption crashes to target subtle, state-dependent logic errors within complex protocol stacks like TCP. The core conceptual shift is the transition from "crash-finding" to "state-exploration fuzzing." The decision to build on LibAFL-Qemu's snapshotting capabilities mirrors sophisticated techniques seen in complex system emulation, suggesting that future vulnerability hunting in opaque, heavily stateful components will increasingly require temporal state management rather than just input mutation. The author posits that failure to detect CVE-2026-9539 stemmed from tool limitations—specifically, ASAN's inability to monitor internal allocator behavior—rather than a fundamental gap in fuzzing methodology.
The dependency on external tooling (like Notify) to reconstruct and run the input chains post-discovery highlights a critical architectural trade-off: achieving deep state visibility via snapshots sacrifices seamless, automated, in-fuzzer detection pipelines. This pushes the security workflow from an integrated black box to a complex, multi-stage engineering process: generate $\rightarrow$ record $\rightarrow$ export $\rightarrow$ reconstruct $\rightarrow$ test. This places a higher burden on the security engineer to manage the reproducibility lifecycle. The narrative thus implicitly suggests that advanced resilience against subtle bugs requires instrumentation that tracks logical invariants (like buffer validity or state transitions) rather than merely tracking memory permissions.
Patterns detected: none
Autonomy check: No autonomy signals.
From the original · Quarkslab Blog
Author Jérémy Marchand Category Fuzzing Tags fuzzing, tool, vulnerability, CVE-2026-9539, QEMU, TCP/IP, 2026 This blog post presents a LibAFL-Qemu based way to explore state machines and why we failed to find CVE-2026-9539. Introduction In this blogpost we present a LibAFL-Qemu based fuzzer we developed with the goal of targeting (complex) state machines using incremental snapshots.Read the full story at blog.quarkslab.com
Sentinel unavailable
The automated check of the source article's wording did not complete, so no result is shown.
This looks only at the wording of the original source article, not at this page's AI-written sections. A small local AI model made this estimate. It has not been checked against known human and machine texts, so treat it as provisional. It cannot show who wrote an article.
