GPU VulnDB

Database/Firmware, BMC & network fabric

Linux kernel mlxsw: failed LAG index allocation leaks a LAG reference on Spectrum switches

CVE-2026-72308Firmware, BMC & network fabriccurated

Impact

mlxsw is the driver that runs NVIDIA Spectrum switch ASICs under Linux-based network OSes, so this lives on the switch, not on the compute node. When mlxsw_sp_port_lag_index_get() fails, mlxsw_sp_port_lag_join() returns the error without dropping the reference taken by the earlier mlxsw_sp_lag_get(), while every other error path in the function goes through the cleanup label that releases it. Repeated failures therefore leak LAG objects that are never freed, and a switch that has exhausted its LAG capacity cannot bring up new port channels. On a leaf carrying storage or east-west GPU traffic that is expensive to recover from, because clearing the leak needs a switch reload and that takes the rack's links with it.

Who can reach it

Local and privileged on the switch control plane: an operator or automation with CAP_NET_ADMIN repeatedly enslaving ports to a bond when no LAG index is available. Not reachable from data-plane traffic and not reachable by a tenant on a GPU node.

What to do

Move to a stable kernel or NOS image that carries the fix and reload the switch. The leaked references cannot be reclaimed at runtime, so a switch already in that state needs a reboot regardless - schedule it per-switch behind the fabric's normal redundancy. The record contains kernel commits only; no NOS version mapping is published in it.

References

Related entries

All Firmware, BMC & network fabric 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.