Which versions of Cloud Hypervisor are affected by CVE-2026-45782?

Cloud Hypervisor versions 21.0 through 51.1 contain a critical use-after-free vulnerability (CVE-2026-45782, CVSS 8.9) that enables malicious guest VMs to corrupt host memory and potentially escape…. A patch is available.

How to fix or mitigate CVE-2026-45782?

Within 24 hours: Inventory all systems running Cloud Hypervisor versions 21.0-51.1 and assess guest VM trust model. Within 7 days: Implement compensating controls including guest VM isolation, source validation, and enhanced host monitoring; restrict Cloud Hypervisor to trusted guest-only environments if operationally feasible. Within 30 days: Monitor Cloud Hypervisor project announcements for patched releases; establish upgrade timeline to non-vulnerable versions when available.

Cloud Hypervisor CVE-2026-45782

Q: What is the CVSS score of CVE-2026-45782?

CVE-2026-45782 has a CVSS 4.0 base score of 8.9 (High). CVSS vector: CVSS:4.0/AV:L/AC:L/AT:P/PR:N/UI:N/VC:H/VI:H/VA:H/SC:H/SI:H/SA:H/E:X/CR:X/IR:X/AR:X/MAV:X/MAC:X/MAT:X/MPR:X/MUI:X/MVC:X/MVI:X/MVA:X/MSC:X/MSI:X/MSA:X/S:X/AU:X/R:X/V:X/RE:X/U:X. EPSS exploitation probability: 0.0%.

HIGH

Use After Free (CWE-416)

2026-06-09 GitHub_M

Information Disclosure Use After Free Memory Corruption Suse

8.9

CVSS 4.0 · NVD

Severity by source

NVD PRIMARY

8.9 HIGH

CVSS:4.0/AV:L/AC:L/AT:P/PR:N/UI:N/VC:H/VI:H/VA:H/SC:H/SI:H/SA:H/E:X/CR:X/IR:X/AR:X/MAV:X/MAC:X/MAT:X/MPR:X/MUI:X/MVC:X/MVI:X/MVA:X/MSC:X/MSI:X/MSA:X/S:X/AU:X/R:X/V:X/RE:X/U:X

Primary rating from NVD · only source for this CVE.

CVSS VectorNVD

CVSS:4.0/AV:L/AC:L/AT:P/PR:N/UI:N/VC:H/VI:H/VA:H/SC:H/SI:H/SA:H/E:X/CR:X/IR:X/AR:X/MAV:X/MAC:X/MAT:X/MPR:X/MUI:X/MVC:X/MVI:X/MVA:X/MSC:X/MSI:X/MSA:X/S:X/AU:X/R:X/V:X/RE:X/U:X

Attack Vector

Local

Attack Complexity

Low

Privileges Required

None

User Interaction

None

Scope

Lifecycle Timeline

Source Code Evidence Fetched

Jun 10, 2026 - 01:53 vuln.today

Analysis Generated

Jun 10, 2026 - 01:53 vuln.today

CVSS changed

Jun 10, 2026 - 00:22 NVD

8.9 (HIGH)

CVE Published

Jun 09, 2026 - 22:53 nvd

UNKNOWN (no severity yet)

DescriptionNVD

Cloud Hypervisor is a Virtual Machine Monitor for Cloud workloads. From version 21.0 to before version 51.2, a guest can cause a use-after-free in the cloud-hypervisor process by submitting two virtio-block descriptor chains that reuse the same head_index while asynchronous block I/O is enabled (e.g. io_uring, aio). When the kernel completes the duplicate operation before the original, the completion path frees a bounce buffer that the kernel is still actively reading from or writing to, corrupting the freed memory. This issue has been patched in versions 51.2 and 52.0.

AnalysisAI

Use-after-free in Cloud Hypervisor versions 21.0 through 51.1 allows a malicious guest VM to corrupt host memory in the cloud-hypervisor VMM process by racing duplicate virtio-block descriptor chains against the host's asynchronous I/O completion path. The flaw carries a CVSS 4.0 score of 8.9 with high impact on both the affected VMM and subsequent system scope, indicating a credible VM escape primitive. No public exploit identified at time of analysis, and the issue is not listed in CISA KEV.

Technical ContextAI

Cloud Hypervisor is a Rust-based Virtual Machine Monitor (VMM) targeted at cloud workloads, derived from rust-vmm crates and competing with QEMU/Firecracker. The bug is a classic CWE-416 use-after-free in the block device backend's bounce-buffer logic: when a guest descriptor's host virtual address does not meet the disk backend's alignment requirement, the VMM allocates an aligned heap buffer (AlignedOperation) and submits the I/O to an async backend such as io_uring or libaio. According to the upstream PR diff (cloud-hypervisor#8220) and commit 1314ac8, the original implementation stored the bounce buffer inside the Request keyed by descriptor head_index; a guest that submits two descriptor chains reusing the same head_index can cause the completion of the duplicate to drop and deallocate the buffer while the kernel still holds an outstanding read/write reference to it. The fix extracts AlignedOperation into its own module with owned drop semantics and adds an io_uring submission-queue capacity check in raw_async.rs to refuse partially-pushable batches.

RemediationAI

Vendor-released patch: upgrade to Cloud Hypervisor 51.2 (for the 51.x line) or 52.0 (current), per the GHSA-f47p-p25q-83rh advisory and the release tags above; the underlying fix lands in PR #8220 and commit 1314ac883c641f1045bbb06dec0de045a3894baa. If immediate upgrade is not possible, the most directly targeted compensating control is to disable asynchronous block I/O backends in guest disk configuration - switch the block device backend from io_uring/aio to synchronous I/O - which removes the race window described in the advisory at the cost of significant disk throughput and IOPS regression for I/O-heavy guests. Operators running multi-tenant or untrusted-guest workloads should prioritize the patch over the workaround, and where neither is feasible, restricting host placement so that only trusted workloads run on un-patched hypervisors limits blast radius but does not eliminate the bug.