Re: [PATCH v4 0/6] Automatic NFSv4 state revocation on filesystem unmount

[Jeff Layton <jlayton@kernel.org>]

On Wed, 2026-03-18 at 10:15 -0400, Chuck Lever wrote:
> When an NFS server exports a filesystem and clients hold NFSv4
> state (opens, locks, delegations), unmounting the underlying
> filesystem fails with EBUSY. The /proc/fs/nfsd/unlock_fs procfs
> interface revokes all NFSv4 state on a given filesystem's
> superblock, but it operates at whole-filesystem granularity and
> offers no structured way to extend the operation to other scopes.
> 
> This series adds an NFSD_CMD_UNLOCK netlink command that supports
> typed, validated attributes and multiple scope values. The
> initial "ip" scope provides a netlink equivalent of
> write_unlock_ip, releasing NLM locks held by a specific client
> address. A "filesystem" scope parallels write_unlock_fs, accepting
> a path string and revoking NFSv4 state, NLM locks, async COPY
> operations, and cached file handles associated with that path.
> 
> The filesystem scope revokes state at export granularity rather
> than superblock granularity: when multiple exports share a
> filesystem, only state referencing files under the specified
> export root is affected. The subtree check walks all dentry
> aliases of each inode so that hard-linked files outside the
> export are not incorrectly matched. When the export root is the
> filesystem root, the subtree check is elided.
> 
> The exportfs user space command can invoke NFSD_CMD_UNLOCK with
> the filesystem scope after an explicit "exportfs -u", enabling
> automated state cleanup without custom scripting or manual
> procfs writes.
> 

I'll note that the mountd/exportd/exportfs nfs-utils patchset that I
sent recently adds basic netlink support to exportfs. The userland bits
for this should nestle on top of that nicely.

> ---
> Changes since v3:
> - All VFS changes replaced with new netlink "unlock" operation
> 
> Changes since v2:
> - Replace fs_pin with an SRCU umount notifier chain in VFS
> - Merge the pending COPY cancellation patch
> - Replace xa_cmpxchg() with xa_insert()
> - Use cancel_work_sync() instead of flush_workqueue()
> - Remove rcu_barrier()
> - Correct misleading claims in kdoc comments and commit messages
> 
> Changes since v1:
> - Explain why drop_client() is being renamed
> - Finish implementing revocation on umount
> - Rename pin_insert_group
> - Clarified log output and code comments
> - Hold nfsd_mutex while closing nfsd_files
> 
> ---
> Chuck Lever (6):
>       NFSD: Extract revoke_one_stid() utility function
>       NFSD: Add NFSD_CMD_UNLOCK netlink command with ip scope
>       NFSD: Add filesystem scope to NFSD_CMD_UNLOCK
>       NFSD: Refactor find_one_sb_stid() into find_next_sb_stid()
>       NFSD: Add export-scoped state revocation
>       NFSD: Add nfsd_file_close_export() for file cache cleanup
> 
>  Documentation/netlink/specs/nfsd.yaml |  39 ++++++
>  fs/nfsd/filecache.c                   |  77 ++++++++++++
>  fs/nfsd/filecache.h                   |   2 +
>  fs/nfsd/netlink.c                     |  14 +++
>  fs/nfsd/netlink.h                     |   1 +
>  fs/nfsd/nfs4state.c                   | 219 +++++++++++++++++++++-------------
>  fs/nfsd/nfsctl.c                      | 105 +++++++++++++++-
>  fs/nfsd/state.h                       |   7 ++
>  fs/nfsd/trace.h                       |  13 +-
>  include/uapi/linux/nfsd_netlink.h     |  19 +++
>  10 files changed, 406 insertions(+), 90 deletions(-)
> ---
> base-commit: f338e77383789c0cae23ca3d48adcc5e9e137e3c
> change-id: 20260318-umount-kills-nfsv4-state-138218f2f4e0
> 
> Best regards,
> --  
> Chuck Lever

-- 
Jeff Layton <jlayton@kernel.org>