GPU VulnDB

Database/Firmware, BMC & network fabric

Linux kernel mlxsw: use-after-free updating multicast route stats during route replace

CVSS 7.8CVE-2025-68800Firmware, BMC & network fabriccurated

Impact

The mlxsw driver for Mellanox Spectrum switch ASICs deletes a multicast route list entry during route replace without taking the mutex that protects the list, while a workqueue periodically walks that list to pull stats from the device. KASAN shows a slab-use-after-free in mlxsw_sp_mr_stats_update. This is switch-side code: it runs on Spectrum-based datacenter switches and on hosts running the mlxsw driver, not on a GPU compute node. The practical outcome is a kernel crash on the switch control plane, which takes down a leaf or spine in the fabric that the fleet depends on; the record gives no evidence of controlled exploitation, and triggering it requires control over multicast routing configuration.

Who can reach it

Local on the switch control plane - an operator or process able to change IPv4/IPv6 multicast routes (route replace) on a device running mlxsw. Not reachable from a tenant workload on a GPU node.

What to do

Update the switch OS / host kernel to a build containing the mlxsw spectrum_mr locking fix (commits in the record) and reboot the switch control plane, which means draining traffic to the redundant path in the fabric first. No workaround is given; avoiding multicast route replacement reduces exposure but is not a stated mitigation.

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.