{"id":"CVE-2026-93221","published":"2026-09-24T16:17:17.310","lastModified":"2026-09-25T05:17:00.150","description":"In the Linux kernel, the following vulnerability has been resolved:\n\nnfsd: convert nfsd_net boolean flags to unsigned long flags word\n\nnfsd_net contains several boolean fields that are accessed from\nconcurrent contexts without serialization.  In particular,\nnfsd4_end_grace() guards its drain path with a plain bool:\n\n    if (nn->grace_ended)\n            return;\n    nn->grace_ended = true;\n\nThe read and the write are independent, and nothing in struct\nnfsd_net serializes them.  At least two contexts can reach this\ncode with no lock held:\n\n    laundromat path\n      laundry_wq kworker\n        nfs4_laundromat()\n          nfsd4_end_grace()\n\n    RECLAIM_COMPLETE path\n      nfsd compound kthread\n        nfsd4_reclaim_complete()\n          inc_reclaim_complete()\n            nfsd4_end_grace()\n\nBoth callers can observe grace_ended == false on different CPUs,\nboth store true, and both proceed into nfsd4_record_grace_done(),\nwhich invokes the active client_tracking_ops->grace_done callback.\nFor tracking ops that drain reclaim_str_hashtbl (legacy_tracking_ops\nvia nfsd4_recdir_purge_old, and the cld v1+ ops via\nnfsd4_cld_grace_done), grace_done calls nfs4_release_reclaim(),\nwhich walks every bucket of reclaim_str_hashtbl with no lock and\ncalls nfs4_remove_reclaim_record() (list_del + kfree) on each\nentry.  Two concurrent walkers corrupt the list and double-free\nevery nfs4_client_reclaim.  A concurrent nfsd4_find_reclaim_client()\niterating the same bucket reads through freed memory.\n\nA third call site exists in nfs4_state_start_net() on the\nskip_grace startup path, but it runs under nfsd_mutex before any\nclient has connected and before the laundromat's first delayed\nwork fires, so it cannot race with the two callers above.\n\nReplace the scattered boolean fields in nfsd_net with a single\nunsigned long flags word and an enum nfsd_net_flag for the bit\npositions.  The grace_ended race is fixed by using\ntest_and_set_bit(), which is atomic on all architectures.  The\nremaining flags (grace_end_forced, in_grace, somebody_reclaimed,\ntrack_reclaim_completes, nfsd_net_up, lockd_up) are converted to\nuse test_bit/set_bit/clear_bit for consistency.  This avoids\nsub-word cmpxchg issues on architectures like Hexagon that only\nsupport word-sized atomic operations.","cvssScore":8.1,"cvssSeverity":"HIGH","cvssVector":"CVSS:3.1/AV:N/AC:H/PR:N/UI:N/S:U/C:H/I:H/A:H","cwes":[],"vendors":[],"products":[],"references":[{"url":"https://git.kernel.org/stable/c/11a5fe42e1811f793e04ef885b639ea7668f439d","tags":[]},{"url":"https://git.kernel.org/stable/c/df5922fe09a8131c793ffa86adf204999b0470f8","tags":[]}],"exploitRefs":[],"hasPoc":false,"ai":null}