GPU VulnDB

Database/Kernel, userspace & hypervisor

Linux kernel ntb_transport: oversized transmit buffers leak skbs until the host runs out of memory

CVSS 7.5CVE-2026-80987Kernel, userspace & hypervisorcurated

Impact

ntb_process_tx() treats an oversized buffer as a success and calls tx_handler() with a NULL data pointer, so ntb_netdev neither frees the skb nor takes its enqueue error path and the buffer leaks. Sustained oversized traffic over a non-transparent PCIe bridge link drains host memory, which on a GPU node means the OOM killer arriving in the middle of jobs that cannot be cheaply rescheduled. Only hosts that actually use NTB interconnect with ntb_netdev are affected; most GPU fleets use InfiniBand or Ethernet here and never load this driver.

Who can reach it

A peer that can send traffic across the NTB link to the affected host. No authentication exists at this layer; exposure depends entirely on whether ntb_netdev is in use.

What to do

Apply the stable fix that rejects oversized buffers in ntb_transport_tx_enqueue() with -EMSGSIZE (five stable commits linked - take the one for your branch, plus the preceding netdev patch that frees the skb on that error). Kernel update and node reboot. If ntb_transport/ntb_netdev are not loaded, there is nothing to schedule. No fixed version strings are given 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.