{"id":"CVE-2026-89798","published":"2026-09-16T11:16:44.660","lastModified":"2026-09-21T14:17:26.890","description":"In the Linux kernel, the following vulnerability has been resolved:\n\nrpcrdma: arm rn_done before publishing the notification\n\nrpcrdma_rn_register() inserts @rn into rd_xa with xa_alloc() before\nstoring the caller's callback in rn->rn_done. The xarray makes @rn\nreachable to rpcrdma_remove_one(), which walks rd_xa and invokes\nrn->rn_done(rn) for every registered notification. A device removal\nthat races a fresh registration can therefore observe @rn with\nrn_done still NULL, because the notification objects are zero\nallocated by their owners, and call through a NULL function pointer.\n\nStore rn->rn_done before xa_alloc() publishes @rn. The xarray's\nstore-side and load-side ordering then guarantees that any CPU which\nfinds @rn in rd_xa also observes the armed callback.\n\nrpcrdma_rn_unregister() treats a non-NULL rn_done as the sentinel\nfor a completed registration, so the early store must not survive a\nfailed registration. Clear rn_done again when xa_alloc() fails.\nWere it left set, the failed-accept cleanup path would call\nrpcrdma_rn_unregister() on an @rn that was never inserted, erasing\nan unrelated rd_xa slot and underflowing rd_kref.","cvssScore":null,"cvssSeverity":null,"cvssVector":null,"cwes":[],"vendors":[],"products":[],"references":[{"url":"https://git.kernel.org/stable/c/3c97b8e76ca2bba9e8770571413aed694068e78e","tags":[]},{"url":"https://git.kernel.org/stable/c/5b06f706374c37375bdff9d21cc10e61df925a92","tags":[]},{"url":"https://git.kernel.org/stable/c/ea0408273ccf5df1fb52d9a1b9db4d31600dcb36","tags":[]},{"url":"https://git.kernel.org/stable/c/f9fe3cad43c42d874a388583552b39b411d886eb","tags":[]}],"exploitRefs":[],"hasPoc":false,"ai":null}