CVE-2026-99790: NFSD: stop registering with rpcbind after a failure in listener_set
In the Linux kernel, the following vulnerability has been resolved:
NFSD: stop registering with rpcbind after a failure in listenerset
nfsdnllistenersetdoit() calls svcxprtcreatefromsa() once for each requested listener and passes flags of 0, so every listener registers with rpcbind on its own. A rpcbind that accepts the connection and never replies therefore costs one timeout for each entry. With the cap of 1024 entries the request can hold nfsdmutex for about 34 minutes, which is roughly 17 times the hung-task threshold.
One failure is enough to know that the next call will not fare better. Read svcrpcbfailurecount() before the create loop, and pass SVCSOCKANONYMOUS for the rest of the request once the count moves.
Affected Software
Remediation
Recommended actions to resolve this vulnerability, in priority order.
- Upgrade
Upgrade
linux/Kernelto a version that resolves this vulnerability.Fixed in 6.12.118 - Upgrade
Upgrade
linux/Kernelto a version that resolves this vulnerability.Fixed in 6.18.64 - Upgrade
Upgrade
linux/Kernelto a version that resolves this vulnerability.Fixed in 7.3.5 - Upgrade
Upgrade
debian/linuxto a version that resolves this vulnerability.Fixed in 6.1.176-1Fixed in 6.1.192-2Fixed in 7.3.5-1 - Compensating control
In nfsd_nl_listener_set_doit(), read svc_rpcb_failure_count() before the listener creation loop; once the count changes, pass SVC_SOCK_ANONYMOUS for the remaining listeners so they do not register with rpcbind.
Event History
Frequently Asked Questions
What conditions are needed to trigger the prolonged lock hold?
An NFSD listener-set request must create multiple requested listeners while rpcbind accepts a connection but does not reply. Before the fix, each listener could incur its own rpcbind timeout while nfsd_mutex remains held.
How severe can the delay become?
With the stated cap of 1024 listener entries, the request can hold nfsd_mutex for about 34 minutes. The description notes that this is roughly 17 times the hung-task threshold.
What does the fix change when rpcbind fails?
After detecting that the rpcbind failure count has increased, the remaining listeners in that request are created with SVC_SOCK_ANONYMOUS. This stops the remaining listener creations from registering individually with rpcbind.