DPDK vhost-crypto CVE-2026-86848

Requires a host running DPDK with the vhost-crypto library in use and processing indirect descriptors, plus attacker control of a guest VM (low guest-side privilege, PR:L) connected to that vhost-crypto backend. Exploitation is local to the host (AV:L) via the guest and needs no user interaction. Limiting factors: the vhost-crypto feature must actually be enabled/configured (it is not a default for most DPDK deployments), the attacker must already control a guest on a vulnerable host, and the TOCTOU double-fetch requires winning a timing race between the host's two reads of the descriptor next index, which constrains reliability. Multi-tenant or untrusted-guest environments are the realistic exposure surface.

This is a guest-to-host escape vector in DPDK's vhost-crypto library (CWE-367, TOCTOU double-fetch of a guest-controlled indirect-descriptor next index) leading to out-of-bounds access in the host backend. The supplied CVSS 3.1 vector (AV:L/AC:L/PR:L/UI:N/S:C/C:H/I:H/A:H, 8.8) reflects the key danger: a low-privileged but malicious guest crosses a security boundary (S:C) to impact the host, with full C/I/A impact - a classic virtualization-escape risk profile. However, several signals are MISSING: there is no EPSS score, no CISA KEV listing, and no public POC in the provided data, so there is no public exploit identified at time of analysis. The AC:L in the vendor vector is worth scrutiny - double-fetch/TOCTOU races typically require winning a timing window, which often argues for higher attack complexity; the realized impact (OOB access) does not automatically equate to reliable arbitrary code execution, so the C:H/I:H/A:H rating likely represents a worst-case ceiling rather than a demonstrated outcome. Net: this is a genuine priority for operators running DPDK vhost-crypto with untrusted guests (NFV/multi-tenant), but the absence of exploitation evidence (no KEV, no EPSS, no POC) suggests it is currently a patch-on-cycle item rather than an emergency, and the host-escape potential should be weighed against the fact that AV:L requires the attacker to already control a guest.

A tenant with control of a guest VM on a host running DPDK vhost-crypto submits crafted indirect virtqueue descriptors and mutates the descriptor 'next' index between the host's two reads, forcing the host backend to follow an attacker-chosen index and read or write out of bounds. Because the scope changes from guest to host (S:C), successful exploitation can corrupt or disclose host memory and potentially break tenant isolation. No public exploit code is identified in the available data, and the TOCTOU timing window makes reliable exploitation non-trivial.