GPU VulnDB

Database/Kernel, userspace & hypervisor

Linux kernel SCSI target: use-after-free reading the iSCSI ISID after the PR OUT parameter list is unmapped

CVSS 9.8CVE-2026-72083Kernel, userspace & hypervisorcurated

Impact

core_scsi3_emulate_pro_register_and_move() parses an iSCSI TransportID and keeps iport_ptr as a raw pointer into the mapped PERSISTENT RESERVE OUT parameter list, then unmaps that buffer before dereferencing the pointer in strcmp(), __core_scsi3_locate_pr_reg() and core_scsi3_alloc_registration(). When the parameter list exceeds one page the mapping was made with vmap(), so the unmap tears down the kernel virtual address and every later read hits unmapped memory. Persistent reservations are exactly what shared-LUN setups use, so this is reachable on the storage nodes that back clustered filesystems and boot LUNs for a GPU fleet. The practical outcome is an oops on the target node, taking its LUNs away from every initiator at once.

Who can reach it

An authenticated SCSI initiator with access to a LUN on the target that can issue PERSISTENT RESERVE OUT / REGISTER AND MOVE with a parameter list longer than 4096 bytes carrying an iSCSI-format TransportID.

What to do

Apply the stable-kernel fix, which keeps the parameter list mapped until iport_ptr is no longer needed and unmaps once on the success path. Kernel update plus reboot on every node running the LIO target; schedule around the initiators that depend on its LUNs. No vendor advisory beyond the kernel stable commits is present in the record.

References

Related entries

All Kernel, userspace & hypervisor entries

This entry is curated: imported from vendor advisories with machine assistance, not yet individually verified. Confirm against your vendor's advisory before acting, and report anything wrong.