CVE-2026-64535Critical UAF in Linux Kernel (NVMe/TCP)
In the Linux kernel, the following vulnerability has been resolved: nvmet-tcp: Fix potential UAF when ddgst mismatch Shivam Kumar found via vulnerability testing: When data digest is enabled on an NVMe/TCP connection and a digest mismatch occurs on a non-final H2C_DATA PDU during an R2T-based data transfer, the digest error handler in nvmet_tcp_try_recv_ddgst() calls nvmet_req_uninit() — which performs percpu_ref_put() on the submission queue — but does NOT mark the command as completed. It does not set cqe->status, does not modify rbytes_done, and does not clear any flag. When the subsequent fatal error triggers queue teardown, nvmet_tcp_uninit_data_in_cmds() iterates all commands, checks nvmet_tcp_need_data_in() for each one, and finds that the already-uninited command still appears to need data (because rbytes_done < transfer_len and cqe->status == 0). It therefore calls nvmet_req_uninit() a second time on the same command — a double percpu_ref_put against a single percpu_ref_get.
View on NVDAnalysis
Vulnerabilidad de tipo Use-After-Free (UAF) con impacto crítico (CVSS 9.8) en la implementación de NVMe sobre TCP del kernel de Linux. Un atacante remoto podría provocar un fallo del sistema o ejecución de código mediante el envío de paquetes malformados durante transferencias de datos.
Relevant roles
Severity
CVSS:3.1/AV:N/AC:L/PR:N/UI:N/S:U/C:H/I:H/A:HEPSS
Technical description
In the Linux kernel, the following vulnerability has been resolved: nvmet-tcp: Fix potential UAF when ddgst mismatch Shivam Kumar found via vulnerability testing: When data digest is enabled on an NVMe/TCP connection and a digest mismatch occurs on a non-final H2C_DATA PDU during an R2T-based data transfer, the digest error handler in nvmet_tcp_try_recv_ddgst() calls nvmet_req_uninit() — which performs percpu_ref_put() on the submission queue — but does NOT mark the command as completed. It does not set cqe->status, does not modify rbytes_done, and does not clear any flag. When the subsequent fatal error triggers queue teardown, nvmet_tcp_uninit_data_in_cmds() iterates all commands, checks nvmet_tcp_need_data_in() for each one, and finds that the already-uninited command still appears to need data (because rbytes_done < transfer_len and cqe->status == 0). It therefore calls nvmet_req_uninit() a second time on the same command — a double percpu_ref_put against a single percpu_ref_get.
References
- https://git.kernel.org/stable/c/088ee46c18d99baef453afd74181dd40ade044ad
- https://git.kernel.org/stable/c/6f9442983a3e4227afd1c83a5251ddbca585ea21
- https://git.kernel.org/stable/c/96fe2513df590e74b04253a45089cae75569570e
- https://git.kernel.org/stable/c/dbbd07d0a7020b80f6a7028e561908f7b83b3d5a
- https://git.kernel.org/stable/c/e091ff83d962f9ed00d9bd70443676de9fe98bdc