NFSD: Never return NFS4ERR_FILE_OPEN when removing a directory

commit 370345b4bd184a49ac68d6591801e5e3605b355a upstream.

RFC 8881 Section 18.25.4 paragraph 5 tells us that the server
should return NFS4ERR_FILE_OPEN only if the target object is an
opened file. This suggests that returning this status when removing
a directory will confuse NFS clients.

This is a version-specific issue; nfsd_proc_remove/rmdir() and
nfsd3_proc_remove/rmdir() already return nfserr_access as
appropriate.

Unfortunately there is no quick way for nfsd4_remove() to determine
whether the target object is a file or not, so the check is done in
in nfsd_unlink() for now.

Reported-by: Trond Myklebust <trondmy@hammerspace.com>
Fixes: 466e16f092 ("nfsd: check for EBUSY from vfs_rmdir/vfs_unink.")
Reviewed-by: Jeff Layton <jlayton@kernel.org>
Cc: stable@vger.kernel.org
Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>

fs/nfsd/vfs.c

@@ -1935,9 +1935,17 @@ out:
 	return err;
 }
 
-/*
- * Unlink a file or directory
- * N.B. After this call fhp needs an fh_put
+/**
+ * nfsd_unlink - remove a directory entry
+ * @rqstp: RPC transaction context
+ * @fhp: the file handle of the parent directory to be modified
+ * @type: enforced file type of the object to be removed
+ * @fname: the name of directory entry to be removed
+ * @flen: length of @fname in octets
+ *
+ * After this call fhp needs an fh_put.
+ *
+ * Returns a generic NFS status code in network byte-order.
  */
 __be32
 nfsd_unlink(struct svc_rqst *rqstp, struct svc_fh *fhp, int type,
@@ -2011,10 +2019,14 @@ out_drop_write:
 	fh_drop_write(fhp);
 out_nfserr:
 	if (host_err == -EBUSY) {
-		/* name is mounted-on. There is no perfect
-		 * error status.
+		/*
+		 * See RFC 8881 Section 18.25.4 para 4: NFSv4 REMOVE
+		 * wants a status unique to the object type.
 		 */
-		err = nfserr_file_open;
+		if (type != S_IFDIR)
+			err = nfserr_file_open;
+		else
+			err = nfserr_acces;
 	}
 out:
 	return err != nfs_ok ? err : nfserrno(host_err);