struct xdr_stream xdr;
        struct nfs4_delegation *args = cb->cb_op;
        struct nfs4_cb_compound_hdr hdr = {
-               .ident = args->dl_ident,
+               .ident = cb->cb_clp->cl_cb_ident,
                .minorversion = cb->cb_minorversion,
        };
 
                        PTR_ERR(client));
                return PTR_ERR(client);
        }
-       nfsd4_set_callback_client(clp, client);
+       clp->cl_cb_ident = conn->cb_ident;
+       clp->cl_cb_client = client;
        return 0;
 
 }
  */
 void nfsd4_probe_callback(struct nfs4_client *clp, struct nfs4_cb_conn *conn)
 {
-       int status;
-
        BUG_ON(atomic_read(&clp->cl_cb_set));
 
-       status = setup_callback_client(clp, conn);
-       if (status) {
-               warn_no_callback_path(clp, status);
-               return;
-       }
+       spin_lock(&clp->cl_lock);
+       memcpy(&clp->cl_cb_conn, conn, sizeof(struct nfs4_cb_conn));
+       set_bit(NFSD4_CLIENT_CB_UPDATE, &clp->cl_cb_flags);
+       spin_unlock(&clp->cl_lock);
        do_probe_callback(clp);
 }
 
 }
 
 /* must be called under the state lock */
-void nfsd4_set_callback_client(struct nfs4_client *clp, struct rpc_clnt *new)
+void nfsd4_shutdown_callback(struct nfs4_client *clp)
 {
-       struct rpc_clnt *old = clp->cl_cb_client;
-
-       clp->cl_cb_client = new;
+       set_bit(NFSD4_CLIENT_KILL, &clp->cl_cb_flags);
        /*
-        * After this, any work that saw the old value of cl_cb_client will
-        * be gone:
+        * Note this won't actually result in a null callback;
+        * instead, nfsd4_do_callback_rpc() will detect the killed
+        * client, destroy the rpc client, and stop:
         */
+       do_probe_callback(clp);
        flush_workqueue(callback_wq);
-       /* So we can safely shut it down: */
-       if (old)
-               rpc_shutdown_client(old);
 }
 
 void nfsd4_release_cb(struct nfsd4_callback *cb)
                cb->cb_ops->rpc_release(cb);
 }
 
+void nfsd4_process_cb_update(struct nfsd4_callback *cb)
+{
+       struct nfs4_cb_conn conn;
+       struct nfs4_client *clp = cb->cb_clp;
+       int err;
+
+       /*
+        * This is either an update, or the client dying; in either case,
+        * kill the old client:
+        */
+       if (clp->cl_cb_client) {
+               rpc_shutdown_client(clp->cl_cb_client);
+               clp->cl_cb_client = NULL;
+       }
+       if (test_bit(NFSD4_CLIENT_KILL, &clp->cl_cb_flags))
+               return;
+       spin_lock(&clp->cl_lock);
+       /*
+        * Only serialized callback code is allowed to clear these
+        * flags; main nfsd code can only set them:
+        */
+       BUG_ON(!clp->cl_cb_flags);
+       clear_bit(NFSD4_CLIENT_CB_UPDATE, &clp->cl_cb_flags);
+       memcpy(&conn, &cb->cb_clp->cl_cb_conn, sizeof(struct nfs4_cb_conn));
+       spin_unlock(&clp->cl_lock);
+
+       err = setup_callback_client(clp, &conn);
+       if (err)
+               warn_no_callback_path(clp, err);
+}
+
 void nfsd4_do_callback_rpc(struct work_struct *w)
 {
        struct nfsd4_callback *cb = container_of(w, struct nfsd4_callback, cb_work);
        struct nfs4_client *clp = cb->cb_clp;
-       struct rpc_clnt *clnt = clp->cl_cb_client;
+       struct rpc_clnt *clnt;
 
-       if (clnt == NULL) {
+       if (clp->cl_cb_flags)
+               nfsd4_process_cb_update(cb);
+
+       clnt = clp->cl_cb_client;
+       if (!clnt) {
+               /* Callback channel broken, or client killed; give up: */
                nfsd4_release_cb(cb);
-               return; /* Client is shutting down; give up. */
+               return;
        }
        rpc_call_async(clnt, &cb->cb_msg, RPC_TASK_SOFT | RPC_TASK_SOFTCONN,
                        cb->cb_ops, cb);
 
 {
        struct nfs4_delegation *dp;
        struct nfs4_file *fp = stp->st_file;
-       struct nfs4_cb_conn *conn = &stp->st_stateowner->so_client->cl_cb_conn;
 
        dprintk("NFSD alloc_init_deleg\n");
        /*
        nfs4_file_get_access(fp, O_RDONLY);
        dp->dl_flock = NULL;
        dp->dl_type = type;
-       dp->dl_ident = conn->cb_ident;
        dp->dl_stateid.si_boot = boot_time;
        dp->dl_stateid.si_stateownerid = current_delegid++;
        dp->dl_stateid.si_fileid = 0;
                sop = list_entry(clp->cl_openowners.next, struct nfs4_stateowner, so_perclient);
                release_openowner(sop);
        }
-       nfsd4_set_callback_client(clp, NULL);
+       nfsd4_shutdown_callback(clp);
        if (clp->cl_cb_conn.cb_xprt)
                svc_xprt_put(clp->cl_cb_conn.cb_xprt);
        list_del(&clp->cl_idhash);
        INIT_LIST_HEAD(&clp->cl_delegations);
        INIT_LIST_HEAD(&clp->cl_sessions);
        INIT_LIST_HEAD(&clp->cl_lru);
+       spin_lock_init(&clp->cl_lock);
        INIT_WORK(&clp->cl_cb_null.cb_work, nfsd4_do_callback_rpc);
        clp->cl_time = get_seconds();
        clear_bit(0, &clp->cl_cb_slot_busy);
 
        nfs4_lock_state();
        /* wait for callbacks */
-       nfsd4_set_callback_client(ses->se_client, NULL);
+       nfsd4_shutdown_callback(ses->se_client);
        nfs4_unlock_state();
        nfsd4_put_session(ses);
        status = nfs_ok;
 
        u32                     dl_type;
        time_t                  dl_time;
 /* For recall: */
-       u32                     dl_ident;
        stateid_t               dl_stateid;
        struct knfsd_fh         dl_fh;
        int                     dl_retries;
 
        /* for v4.0 and v4.1 callbacks: */
        struct nfs4_cb_conn     cl_cb_conn;
+#define NFSD4_CLIENT_CB_UPDATE 1
+#define NFSD4_CLIENT_KILL      2
+       unsigned long           cl_cb_flags;
        struct rpc_clnt         *cl_cb_client;
+       u32                     cl_cb_ident;
        atomic_t                cl_cb_set;
        struct nfsd4_callback   cl_cb_null;
 
+       /* for all client information that callback code might need: */
+       spinlock_t              cl_lock;
+
        /* for nfs41 */
        struct list_head        cl_sessions;
        struct nfsd4_clid_slot  cl_cs_slot;     /* create_session slot */
 extern void nfsd4_cb_recall(struct nfs4_delegation *dp);
 extern int nfsd4_create_callback_queue(void);
 extern void nfsd4_destroy_callback_queue(void);
-extern void nfsd4_set_callback_client(struct nfs4_client *, struct rpc_clnt *);
+extern void nfsd4_shutdown_callback(struct nfs4_client *);
 extern void nfs4_put_delegation(struct nfs4_delegation *dp);
 extern __be32 nfs4_make_rec_clidname(char *clidname, struct xdr_netobj *clname);
 extern void nfsd4_init_recdir(char *recdir_name);