NFS: Prevent another deadlock in nfs_release_page()
[safe/jmp/linux-2.6] / fs / nfs / nfs4state.c
index 99182b3..6c5ed51 100644 (file)
@@ -60,10 +60,16 @@ const nfs4_stateid zero_stateid;
 
 static LIST_HEAD(nfs4_clientid_list);
 
-static int nfs4_init_client(struct nfs_client *clp, struct rpc_cred *cred)
+int nfs4_init_clientid(struct nfs_client *clp, struct rpc_cred *cred)
 {
-       int status = nfs4_proc_setclientid(clp, NFS4_CALLBACK,
-                       nfs_callback_tcpport, cred);
+       unsigned short port;
+       int status;
+
+       port = nfs_callback_tcpport;
+       if (clp->cl_addr.ss_family == AF_INET6)
+               port = nfs_callback_tcpport6;
+
+       status = nfs4_proc_setclientid(clp, NFS4_CALLBACK, port, cred);
        if (status == 0)
                status = nfs4_proc_setclientid_confirm(clp, cred);
        if (status == 0)
@@ -71,14 +77,12 @@ static int nfs4_init_client(struct nfs_client *clp, struct rpc_cred *cred)
        return status;
 }
 
-static struct rpc_cred *nfs4_get_machine_cred(struct nfs_client *clp)
+struct rpc_cred *nfs4_get_machine_cred_locked(struct nfs_client *clp)
 {
        struct rpc_cred *cred = NULL;
 
-       spin_lock(&clp->cl_lock);
        if (clp->cl_machine_cred != NULL)
                cred = get_rpccred(clp->cl_machine_cred);
-       spin_unlock(&clp->cl_lock);
        return cred;
 }
 
@@ -94,7 +98,7 @@ static void nfs4_clear_machine_cred(struct nfs_client *clp)
                put_rpccred(cred);
 }
 
-struct rpc_cred *nfs4_get_renew_cred(struct nfs_client *clp)
+struct rpc_cred *nfs4_get_renew_cred_locked(struct nfs_client *clp)
 {
        struct nfs4_state_owner *sp;
        struct rb_node *pos;
@@ -110,13 +114,101 @@ struct rpc_cred *nfs4_get_renew_cred(struct nfs_client *clp)
        return cred;
 }
 
-static struct rpc_cred *nfs4_get_setclientid_cred(struct nfs_client *clp)
+#if defined(CONFIG_NFS_V4_1)
+
+static int nfs41_setup_state_renewal(struct nfs_client *clp)
+{
+       int status;
+       struct nfs_fsinfo fsinfo;
+
+       status = nfs4_proc_get_lease_time(clp, &fsinfo);
+       if (status == 0) {
+               /* Update lease time and schedule renewal */
+               spin_lock(&clp->cl_lock);
+               clp->cl_lease_time = fsinfo.lease_time * HZ;
+               clp->cl_last_renewal = jiffies;
+               spin_unlock(&clp->cl_lock);
+
+               nfs4_schedule_state_renewal(clp);
+       }
+
+       return status;
+}
+
+static void nfs4_end_drain_session(struct nfs_client *clp)
+{
+       struct nfs4_session *ses = clp->cl_session;
+       int max_slots;
+
+       if (test_and_clear_bit(NFS4CLNT_SESSION_DRAINING, &clp->cl_state)) {
+               spin_lock(&ses->fc_slot_table.slot_tbl_lock);
+               max_slots = ses->fc_slot_table.max_slots;
+               while (max_slots--) {
+                       struct rpc_task *task;
+
+                       task = rpc_wake_up_next(&ses->fc_slot_table.
+                                               slot_tbl_waitq);
+                       if (!task)
+                               break;
+                       rpc_task_set_priority(task, RPC_PRIORITY_PRIVILEGED);
+               }
+               spin_unlock(&ses->fc_slot_table.slot_tbl_lock);
+       }
+}
+
+static int nfs4_begin_drain_session(struct nfs_client *clp)
+{
+       struct nfs4_session *ses = clp->cl_session;
+       struct nfs4_slot_table *tbl = &ses->fc_slot_table;
+
+       spin_lock(&tbl->slot_tbl_lock);
+       set_bit(NFS4CLNT_SESSION_DRAINING, &clp->cl_state);
+       if (tbl->highest_used_slotid != -1) {
+               INIT_COMPLETION(ses->complete);
+               spin_unlock(&tbl->slot_tbl_lock);
+               return wait_for_completion_interruptible(&ses->complete);
+       }
+       spin_unlock(&tbl->slot_tbl_lock);
+       return 0;
+}
+
+int nfs41_init_clientid(struct nfs_client *clp, struct rpc_cred *cred)
+{
+       int status;
+
+       nfs4_begin_drain_session(clp);
+       status = nfs4_proc_exchange_id(clp, cred);
+       if (status != 0)
+               goto out;
+       status = nfs4_proc_create_session(clp);
+       if (status != 0)
+               goto out;
+       nfs41_setup_state_renewal(clp);
+       nfs_mark_client_ready(clp, NFS_CS_READY);
+out:
+       return status;
+}
+
+struct rpc_cred *nfs4_get_exchange_id_cred(struct nfs_client *clp)
+{
+       struct rpc_cred *cred;
+
+       spin_lock(&clp->cl_lock);
+       cred = nfs4_get_machine_cred_locked(clp);
+       spin_unlock(&clp->cl_lock);
+       return cred;
+}
+
+#endif /* CONFIG_NFS_V4_1 */
+
+struct rpc_cred *nfs4_get_setclientid_cred(struct nfs_client *clp)
 {
        struct nfs4_state_owner *sp;
        struct rb_node *pos;
        struct rpc_cred *cred;
 
-       cred = nfs4_get_machine_cred(clp);
+       spin_lock(&clp->cl_lock);
+       cred = nfs4_get_machine_cred_locked(clp);
        if (cred != NULL)
                goto out;
        pos = rb_first(&clp->cl_state_owners);
@@ -125,6 +217,7 @@ static struct rpc_cred *nfs4_get_setclientid_cred(struct nfs_client *clp)
                cred = get_rpccred(sp->so_cred);
        }
 out:
+       spin_unlock(&clp->cl_lock);
        return cred;
 }
 
@@ -295,10 +388,6 @@ nfs4_drop_state_owner(struct nfs4_state_owner *sp)
        }
 }
 
-/*
- * Note: must be called with clp->cl_sem held in order to prevent races
- *       with reboot recovery!
- */
 struct nfs4_state_owner *nfs4_get_state_owner(struct nfs_server *server, struct rpc_cred *cred)
 {
        struct nfs_client *clp = server->nfs_client;
@@ -327,10 +416,6 @@ struct nfs4_state_owner *nfs4_get_state_owner(struct nfs_server *server, struct
        return sp;
 }
 
-/*
- * Must be called with clp->cl_sem held in order to avoid races
- * with state recovery...
- */
 void nfs4_put_state_owner(struct nfs4_state_owner *sp)
 {
        struct nfs_client *clp = sp->so_client;
@@ -361,18 +446,18 @@ nfs4_alloc_open_state(void)
 }
 
 void
-nfs4_state_set_mode_locked(struct nfs4_state *state, mode_t mode)
+nfs4_state_set_mode_locked(struct nfs4_state *state, fmode_t fmode)
 {
-       if (state->state == mode)
+       if (state->state == fmode)
                return;
        /* NB! List reordering - see the reclaim code for why.  */
-       if ((mode & FMODE_WRITE) != (state->state & FMODE_WRITE)) {
-               if (mode & FMODE_WRITE)
+       if ((fmode & FMODE_WRITE) != (state->state & FMODE_WRITE)) {
+               if (fmode & FMODE_WRITE)
                        list_move(&state->open_states, &state->owner->so_states);
                else
                        list_move_tail(&state->open_states, &state->owner->so_states);
        }
-       state->state = mode;
+       state->state = fmode;
 }
 
 static struct nfs4_state *
@@ -432,10 +517,6 @@ out:
        return state;
 }
 
-/*
- * Beware! Caller must be holding exactly one
- * reference to clp->cl_sem!
- */
 void nfs4_put_open_state(struct nfs4_state *state)
 {
        struct inode *inode = state->inode;
@@ -456,16 +537,16 @@ void nfs4_put_open_state(struct nfs4_state *state)
 /*
  * Close the current file.
  */
-static void __nfs4_close(struct path *path, struct nfs4_state *state, mode_t mode, int wait)
+static void __nfs4_close(struct path *path, struct nfs4_state *state, fmode_t fmode, int wait)
 {
        struct nfs4_state_owner *owner = state->owner;
        int call_close = 0;
-       int newstate;
+       fmode_t newstate;
 
        atomic_inc(&owner->so_count);
        /* Protect against nfs4_find_state() */
        spin_lock(&owner->so_lock);
-       switch (mode & (FMODE_READ | FMODE_WRITE)) {
+       switch (fmode & (FMODE_READ | FMODE_WRITE)) {
                case FMODE_READ:
                        state->n_rdonly--;
                        break;
@@ -500,14 +581,14 @@ static void __nfs4_close(struct path *path, struct nfs4_state *state, mode_t mod
                nfs4_do_close(path, state, wait);
 }
 
-void nfs4_close_state(struct path *path, struct nfs4_state *state, mode_t mode)
+void nfs4_close_state(struct path *path, struct nfs4_state *state, fmode_t fmode)
 {
-       __nfs4_close(path, state, mode, 0);
+       __nfs4_close(path, state, fmode, 0);
 }
 
-void nfs4_close_sync(struct path *path, struct nfs4_state *state, mode_t mode)
+void nfs4_close_sync(struct path *path, struct nfs4_state *state, fmode_t fmode)
 {
-       __nfs4_close(path, state, mode, 1);
+       __nfs4_close(path, state, fmode, 1);
 }
 
 /*
@@ -545,6 +626,7 @@ static struct nfs4_lock_state *nfs4_alloc_lock_state(struct nfs4_state *state, f
        INIT_LIST_HEAD(&lsp->ls_sequence.list);
        lsp->ls_seqid.sequence = &lsp->ls_sequence;
        atomic_set(&lsp->ls_count, 1);
+       lsp->ls_state = state;
        lsp->ls_owner = fl_owner;
        spin_lock(&clp->cl_lock);
        nfs_alloc_unique_id(&clp->cl_lockowner_id, &lsp->ls_id, 1, 64);
@@ -568,7 +650,6 @@ static void nfs4_free_lock_state(struct nfs4_lock_state *lsp)
  * Return a compatible lock_state. If no initialized lock_state structure
  * exists, return an uninitialized one.
  *
- * The caller must be holding clp->cl_sem
  */
 static struct nfs4_lock_state *nfs4_get_lock_state(struct nfs4_state *state, fl_owner_t owner)
 {
@@ -580,7 +661,6 @@ static struct nfs4_lock_state *nfs4_get_lock_state(struct nfs4_state *state, fl_
                if (lsp != NULL)
                        break;
                if (new != NULL) {
-                       new->ls_state = state;
                        list_add(&new->ls_locks, &state->lock_states);
                        set_bit(LK_STATE_IN_USE, &state->flags);
                        lsp = new;
@@ -631,7 +711,7 @@ static void nfs4_fl_release_lock(struct file_lock *fl)
        nfs4_put_lock_state(fl->fl_u.nfs4_fl.owner);
 }
 
-static struct file_lock_operations nfs4_fl_lock_ops = {
+static const struct file_lock_operations nfs4_fl_lock_ops = {
        .fl_copy_lock = nfs4_fl_copy_lock,
        .fl_release_private = nfs4_fl_release_lock,
 };
@@ -686,16 +766,21 @@ struct nfs_seqid *nfs_alloc_seqid(struct nfs_seqid_counter *counter)
        return new;
 }
 
-void nfs_free_seqid(struct nfs_seqid *seqid)
+void nfs_release_seqid(struct nfs_seqid *seqid)
 {
        if (!list_empty(&seqid->list)) {
                struct rpc_sequence *sequence = seqid->sequence->sequence;
 
                spin_lock(&sequence->lock);
-               list_del(&seqid->list);
+               list_del_init(&seqid->list);
                spin_unlock(&sequence->lock);
                rpc_wake_up(&sequence->wait);
        }
+}
+
+void nfs_free_seqid(struct nfs_seqid *seqid)
+{
+       nfs_release_seqid(seqid);
        kfree(seqid);
 }
 
@@ -735,12 +820,14 @@ static void nfs_increment_seqid(int status, struct nfs_seqid *seqid)
 
 void nfs_increment_open_seqid(int status, struct nfs_seqid *seqid)
 {
-       if (status == -NFS4ERR_BAD_SEQID) {
-               struct nfs4_state_owner *sp = container_of(seqid->sequence,
-                               struct nfs4_state_owner, so_seqid);
+       struct nfs4_state_owner *sp = container_of(seqid->sequence,
+                                       struct nfs4_state_owner, so_seqid);
+       struct nfs_server *server = sp->so_server;
+
+       if (status == -NFS4ERR_BAD_SEQID)
                nfs4_drop_state_owner(sp);
-       }
-       nfs_increment_seqid(status, seqid);
+       if (!nfs4_has_session(server->nfs_client))
+               nfs_increment_seqid(status, seqid);
 }
 
 /*
@@ -770,32 +857,34 @@ unlock:
        return status;
 }
 
-static int reclaimer(void *);
+static int nfs4_run_state_manager(void *);
 
-static inline void nfs4_clear_recover_bit(struct nfs_client *clp)
+static void nfs4_clear_state_manager_bit(struct nfs_client *clp)
 {
        smp_mb__before_clear_bit();
-       clear_bit(NFS4CLNT_STATE_RECOVER, &clp->cl_state);
+       clear_bit(NFS4CLNT_MANAGER_RUNNING, &clp->cl_state);
        smp_mb__after_clear_bit();
-       wake_up_bit(&clp->cl_state, NFS4CLNT_STATE_RECOVER);
+       wake_up_bit(&clp->cl_state, NFS4CLNT_MANAGER_RUNNING);
        rpc_wake_up(&clp->cl_rpcwaitq);
 }
 
 /*
- * State recovery routine
+ * Schedule the nfs_client asynchronous state management routine
  */
-static void nfs4_recover_state(struct nfs_client *clp)
+void nfs4_schedule_state_manager(struct nfs_client *clp)
 {
        struct task_struct *task;
 
+       if (test_and_set_bit(NFS4CLNT_MANAGER_RUNNING, &clp->cl_state) != 0)
+               return;
        __module_get(THIS_MODULE);
        atomic_inc(&clp->cl_count);
-       task = kthread_run(reclaimer, clp, "%s-reclaim",
+       task = kthread_run(nfs4_run_state_manager, clp, "%s-manager",
                                rpc_peeraddr2str(clp->cl_rpcclient,
                                                        RPC_DISPLAY_ADDR));
        if (!IS_ERR(task))
                return;
-       nfs4_clear_recover_bit(clp);
+       nfs4_clear_state_manager_bit(clp);
        nfs_put_client(clp);
        module_put(THIS_MODULE);
 }
@@ -807,40 +896,85 @@ void nfs4_schedule_state_recovery(struct nfs_client *clp)
 {
        if (!clp)
                return;
-       if (test_and_set_bit(NFS4CLNT_STATE_RECOVER, &clp->cl_state) == 0)
-               nfs4_recover_state(clp);
+       if (!test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state))
+               set_bit(NFS4CLNT_CHECK_LEASE, &clp->cl_state);
+       nfs4_schedule_state_manager(clp);
+}
+
+int nfs4_state_mark_reclaim_reboot(struct nfs_client *clp, struct nfs4_state *state)
+{
+
+       set_bit(NFS_STATE_RECLAIM_REBOOT, &state->flags);
+       /* Don't recover state that expired before the reboot */
+       if (test_bit(NFS_STATE_RECLAIM_NOGRACE, &state->flags)) {
+               clear_bit(NFS_STATE_RECLAIM_REBOOT, &state->flags);
+               return 0;
+       }
+       set_bit(NFS_OWNER_RECLAIM_REBOOT, &state->owner->so_flags);
+       set_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state);
+       return 1;
+}
+
+int nfs4_state_mark_reclaim_nograce(struct nfs_client *clp, struct nfs4_state *state)
+{
+       set_bit(NFS_STATE_RECLAIM_NOGRACE, &state->flags);
+       clear_bit(NFS_STATE_RECLAIM_REBOOT, &state->flags);
+       set_bit(NFS_OWNER_RECLAIM_NOGRACE, &state->owner->so_flags);
+       set_bit(NFS4CLNT_RECLAIM_NOGRACE, &clp->cl_state);
+       return 1;
 }
 
 static int nfs4_reclaim_locks(struct nfs4_state *state, const struct nfs4_state_recovery_ops *ops)
 {
        struct inode *inode = state->inode;
+       struct nfs_inode *nfsi = NFS_I(inode);
        struct file_lock *fl;
        int status = 0;
 
+       if (inode->i_flock == NULL)
+               return 0;
+
+       /* Guard against delegation returns and new lock/unlock calls */
+       down_write(&nfsi->rwsem);
+       /* Protect inode->i_flock using the BKL */
+       lock_kernel();
        for (fl = inode->i_flock; fl != NULL; fl = fl->fl_next) {
                if (!(fl->fl_flags & (FL_POSIX|FL_FLOCK)))
                        continue;
                if (nfs_file_open_context(fl->fl_file)->state != state)
                        continue;
+               unlock_kernel();
                status = ops->recover_lock(state, fl);
-               if (status >= 0)
-                       continue;
                switch (status) {
+                       case 0:
+                               break;
+                       case -ESTALE:
+                       case -NFS4ERR_ADMIN_REVOKED:
+                       case -NFS4ERR_STALE_STATEID:
+                       case -NFS4ERR_BAD_STATEID:
+                       case -NFS4ERR_EXPIRED:
+                       case -NFS4ERR_NO_GRACE:
+                       case -NFS4ERR_STALE_CLIENTID:
+                       case -NFS4ERR_BADSESSION:
+                       case -NFS4ERR_BADSLOT:
+                       case -NFS4ERR_BAD_HIGH_SLOT:
+                       case -NFS4ERR_CONN_NOT_BOUND_TO_SESSION:
+                               goto out;
                        default:
                                printk(KERN_ERR "%s: unhandled error %d. Zeroing state\n",
                                                __func__, status);
-                       case -NFS4ERR_EXPIRED:
-                       case -NFS4ERR_NO_GRACE:
+                       case -ENOMEM:
+                       case -NFS4ERR_DENIED:
                        case -NFS4ERR_RECLAIM_BAD:
                        case -NFS4ERR_RECLAIM_CONFLICT:
                                /* kill_proc(fl->fl_pid, SIGLOST, 1); */
-                               break;
-                       case -NFS4ERR_STALE_CLIENTID:
-                               goto out_err;
+                               status = 0;
                }
+               lock_kernel();
        }
-       return 0;
-out_err:
+       unlock_kernel();
+out:
+       up_write(&nfsi->rwsem);
        return status;
 }
 
@@ -858,9 +992,15 @@ static int nfs4_reclaim_open_state(struct nfs4_state_owner *sp, const struct nfs
         * recovering after a network partition or a reboot from a
         * server that doesn't support a grace period.
         */
+restart:
+       spin_lock(&sp->so_lock);
        list_for_each_entry(state, &sp->so_states, open_states) {
+               if (!test_and_clear_bit(ops->state_flag_bit, &state->flags))
+                       continue;
                if (state->state == 0)
                        continue;
+               atomic_inc(&state->count);
+               spin_unlock(&sp->so_lock);
                status = ops->recover_open(sp, state);
                if (status >= 0) {
                        status = nfs4_reclaim_locks(state, ops);
@@ -870,7 +1010,8 @@ static int nfs4_reclaim_open_state(struct nfs4_state_owner *sp, const struct nfs
                                                printk("%s: Lock reclaim failed!\n",
                                                        __func__);
                                }
-                               continue;
+                               nfs4_put_open_state(state);
+                               goto restart;
                        }
                }
                switch (status) {
@@ -878,8 +1019,8 @@ static int nfs4_reclaim_open_state(struct nfs4_state_owner *sp, const struct nfs
                                printk(KERN_ERR "%s: unhandled error %d. Zeroing state\n",
                                                __func__, status);
                        case -ENOENT:
-                       case -NFS4ERR_RECLAIM_BAD:
-                       case -NFS4ERR_RECLAIM_CONFLICT:
+                       case -ENOMEM:
+                       case -ESTALE:
                                /*
                                 * Open state on this file cannot be recovered
                                 * All we can do is revert to using the zero stateid.
@@ -889,159 +1030,476 @@ static int nfs4_reclaim_open_state(struct nfs4_state_owner *sp, const struct nfs
                                /* Mark the file as being 'closed' */
                                state->state = 0;
                                break;
+                       case -NFS4ERR_ADMIN_REVOKED:
+                       case -NFS4ERR_STALE_STATEID:
+                       case -NFS4ERR_BAD_STATEID:
+                       case -NFS4ERR_RECLAIM_BAD:
+                       case -NFS4ERR_RECLAIM_CONFLICT:
+                               nfs4_state_mark_reclaim_nograce(sp->so_client, state);
+                               break;
                        case -NFS4ERR_EXPIRED:
                        case -NFS4ERR_NO_GRACE:
+                               nfs4_state_mark_reclaim_nograce(sp->so_client, state);
                        case -NFS4ERR_STALE_CLIENTID:
+                       case -NFS4ERR_BADSESSION:
+                       case -NFS4ERR_BADSLOT:
+                       case -NFS4ERR_BAD_HIGH_SLOT:
+                       case -NFS4ERR_CONN_NOT_BOUND_TO_SESSION:
                                goto out_err;
                }
+               nfs4_put_open_state(state);
+               goto restart;
        }
+       spin_unlock(&sp->so_lock);
        return 0;
 out_err:
+       nfs4_put_open_state(state);
        return status;
 }
 
-static void nfs4_state_mark_reclaim(struct nfs_client *clp)
+static void nfs4_clear_open_state(struct nfs4_state *state)
+{
+       struct nfs4_lock_state *lock;
+
+       clear_bit(NFS_DELEGATED_STATE, &state->flags);
+       clear_bit(NFS_O_RDONLY_STATE, &state->flags);
+       clear_bit(NFS_O_WRONLY_STATE, &state->flags);
+       clear_bit(NFS_O_RDWR_STATE, &state->flags);
+       list_for_each_entry(lock, &state->lock_states, ls_locks) {
+               lock->ls_seqid.flags = 0;
+               lock->ls_flags &= ~NFS_LOCK_INITIALIZED;
+       }
+}
+
+static void nfs4_state_mark_reclaim_helper(struct nfs_client *clp, int (*mark_reclaim)(struct nfs_client *clp, struct nfs4_state *state))
 {
        struct nfs4_state_owner *sp;
        struct rb_node *pos;
        struct nfs4_state *state;
-       struct nfs4_lock_state *lock;
 
        /* Reset all sequence ids to zero */
        for (pos = rb_first(&clp->cl_state_owners); pos != NULL; pos = rb_next(pos)) {
                sp = rb_entry(pos, struct nfs4_state_owner, so_client_node);
-               sp->so_seqid.counter = 0;
                sp->so_seqid.flags = 0;
                spin_lock(&sp->so_lock);
                list_for_each_entry(state, &sp->so_states, open_states) {
-                       clear_bit(NFS_DELEGATED_STATE, &state->flags);
-                       clear_bit(NFS_O_RDONLY_STATE, &state->flags);
-                       clear_bit(NFS_O_WRONLY_STATE, &state->flags);
-                       clear_bit(NFS_O_RDWR_STATE, &state->flags);
-                       list_for_each_entry(lock, &state->lock_states, ls_locks) {
-                               lock->ls_seqid.counter = 0;
-                               lock->ls_seqid.flags = 0;
-                               lock->ls_flags &= ~NFS_LOCK_INITIALIZED;
-                       }
+                       if (mark_reclaim(clp, state))
+                               nfs4_clear_open_state(state);
                }
                spin_unlock(&sp->so_lock);
        }
 }
 
+static void nfs4_state_start_reclaim_reboot(struct nfs_client *clp)
+{
+       /* Mark all delegations for reclaim */
+       nfs_delegation_mark_reclaim(clp);
+       nfs4_state_mark_reclaim_helper(clp, nfs4_state_mark_reclaim_reboot);
+}
+
+static void nfs4_reclaim_complete(struct nfs_client *clp,
+                                const struct nfs4_state_recovery_ops *ops)
+{
+       /* Notify the server we're done reclaiming our state */
+       if (ops->reclaim_complete)
+               (void)ops->reclaim_complete(clp);
+}
+
+static void nfs4_state_end_reclaim_reboot(struct nfs_client *clp)
+{
+       struct nfs4_state_owner *sp;
+       struct rb_node *pos;
+       struct nfs4_state *state;
+
+       if (!test_and_clear_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state))
+               return;
+
+       nfs4_reclaim_complete(clp,
+               nfs4_reboot_recovery_ops[clp->cl_minorversion]);
+
+       for (pos = rb_first(&clp->cl_state_owners); pos != NULL; pos = rb_next(pos)) {
+               sp = rb_entry(pos, struct nfs4_state_owner, so_client_node);
+               spin_lock(&sp->so_lock);
+               list_for_each_entry(state, &sp->so_states, open_states) {
+                       if (!test_and_clear_bit(NFS_STATE_RECLAIM_REBOOT, &state->flags))
+                               continue;
+                       nfs4_state_mark_reclaim_nograce(clp, state);
+               }
+               spin_unlock(&sp->so_lock);
+       }
+
+       nfs_delegation_reap_unclaimed(clp);
+}
+
+static void nfs_delegation_clear_all(struct nfs_client *clp)
+{
+       nfs_delegation_mark_reclaim(clp);
+       nfs_delegation_reap_unclaimed(clp);
+}
+
+static void nfs4_state_start_reclaim_nograce(struct nfs_client *clp)
+{
+       nfs_delegation_clear_all(clp);
+       nfs4_state_mark_reclaim_helper(clp, nfs4_state_mark_reclaim_nograce);
+}
+
+static int nfs4_recovery_handle_error(struct nfs_client *clp, int error)
+{
+       switch (error) {
+               case -NFS4ERR_CB_PATH_DOWN:
+                       nfs_handle_cb_pathdown(clp);
+                       return 0;
+               case -NFS4ERR_NO_GRACE:
+                       nfs4_state_end_reclaim_reboot(clp);
+                       return 0;
+               case -NFS4ERR_STALE_CLIENTID:
+               case -NFS4ERR_LEASE_MOVED:
+                       set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
+                       nfs4_state_end_reclaim_reboot(clp);
+                       nfs4_state_start_reclaim_reboot(clp);
+                       break;
+               case -NFS4ERR_EXPIRED:
+                       set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
+                       nfs4_state_start_reclaim_nograce(clp);
+                       break;
+               case -NFS4ERR_BADSESSION:
+               case -NFS4ERR_BADSLOT:
+               case -NFS4ERR_BAD_HIGH_SLOT:
+               case -NFS4ERR_DEADSESSION:
+               case -NFS4ERR_CONN_NOT_BOUND_TO_SESSION:
+               case -NFS4ERR_SEQ_FALSE_RETRY:
+               case -NFS4ERR_SEQ_MISORDERED:
+                       set_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state);
+                       /* Zero session reset errors */
+                       return 0;
+       }
+       return error;
+}
+
 static int nfs4_do_reclaim(struct nfs_client *clp, const struct nfs4_state_recovery_ops *ops)
 {
        struct rb_node *pos;
        int status = 0;
 
-       /* Note: list is protected by exclusive lock on cl->cl_sem */
+restart:
+       spin_lock(&clp->cl_lock);
        for (pos = rb_first(&clp->cl_state_owners); pos != NULL; pos = rb_next(pos)) {
                struct nfs4_state_owner *sp = rb_entry(pos, struct nfs4_state_owner, so_client_node);
+               if (!test_and_clear_bit(ops->owner_flag_bit, &sp->so_flags))
+                       continue;
+               atomic_inc(&sp->so_count);
+               spin_unlock(&clp->cl_lock);
                status = nfs4_reclaim_open_state(sp, ops);
-               if (status < 0)
-                       break;
+               if (status < 0) {
+                       set_bit(ops->owner_flag_bit, &sp->so_flags);
+                       nfs4_put_state_owner(sp);
+                       return nfs4_recovery_handle_error(clp, status);
+               }
+               nfs4_put_state_owner(sp);
+               goto restart;
        }
+       spin_unlock(&clp->cl_lock);
        return status;
 }
 
 static int nfs4_check_lease(struct nfs_client *clp)
 {
        struct rpc_cred *cred;
+       struct nfs4_state_maintenance_ops *ops =
+               nfs4_state_renewal_ops[clp->cl_minorversion];
        int status = -NFS4ERR_EXPIRED;
 
-       /* Are there any open files on this volume? */
-       cred = nfs4_get_renew_cred(clp);
-       if (cred != NULL) {
-               /* Yes there are: try to renew the old lease */
-               status = nfs4_proc_renew(clp, cred);
-               put_rpccred(cred);
-               return status;
+       /* Is the client already known to have an expired lease? */
+       if (test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state))
+               return 0;
+       spin_lock(&clp->cl_lock);
+       cred = ops->get_state_renewal_cred_locked(clp);
+       spin_unlock(&clp->cl_lock);
+       if (cred == NULL) {
+               cred = nfs4_get_setclientid_cred(clp);
+               if (cred == NULL)
+                       goto out;
        }
-
-       /* "reboot" to ensure we clear all state on the server */
-       clp->cl_boot_time = CURRENT_TIME;
-       return status;
+       status = ops->renew_lease(clp, cred);
+       put_rpccred(cred);
+out:
+       return nfs4_recovery_handle_error(clp, status);
 }
 
 static int nfs4_reclaim_lease(struct nfs_client *clp)
 {
        struct rpc_cred *cred;
+       struct nfs4_state_recovery_ops *ops =
+               nfs4_reboot_recovery_ops[clp->cl_minorversion];
        int status = -ENOENT;
 
-       cred = nfs4_get_setclientid_cred(clp);
+       cred = ops->get_clid_cred(clp);
        if (cred != NULL) {
-               status = nfs4_init_client(clp, cred);
+               status = ops->establish_clid(clp, cred);
                put_rpccred(cred);
                /* Handle case where the user hasn't set up machine creds */
                if (status == -EACCES && cred == clp->cl_machine_cred) {
                        nfs4_clear_machine_cred(clp);
                        status = -EAGAIN;
                }
+               if (status == -NFS4ERR_MINOR_VERS_MISMATCH)
+                       status = -EPROTONOSUPPORT;
        }
        return status;
 }
 
-static int reclaimer(void *ptr)
+#ifdef CONFIG_NFS_V4_1
+void nfs41_handle_recall_slot(struct nfs_client *clp)
 {
-       struct nfs_client *clp = ptr;
-       const struct nfs4_state_recovery_ops *ops;
-       int status = 0;
+       set_bit(NFS4CLNT_RECALL_SLOT, &clp->cl_state);
+       nfs4_schedule_state_recovery(clp);
+}
 
-       allow_signal(SIGKILL);
+static void nfs4_reset_all_state(struct nfs_client *clp)
+{
+       if (test_and_set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state) == 0) {
+               clp->cl_boot_time = CURRENT_TIME;
+               nfs4_state_start_reclaim_nograce(clp);
+               nfs4_schedule_state_recovery(clp);
+       }
+}
 
-       /* Ensure exclusive access to NFSv4 state */
-       down_write(&clp->cl_sem);
-       while (!list_empty(&clp->cl_superblocks)) {
-               ops = &nfs4_network_partition_recovery_ops;
-               status = nfs4_check_lease(clp);
+static void nfs41_handle_server_reboot(struct nfs_client *clp)
+{
+       if (test_and_set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state) == 0) {
+               nfs4_state_start_reclaim_reboot(clp);
+               nfs4_schedule_state_recovery(clp);
+       }
+}
+
+static void nfs41_handle_state_revoked(struct nfs_client *clp)
+{
+       /* Temporary */
+       nfs4_reset_all_state(clp);
+}
+
+static void nfs41_handle_recallable_state_revoked(struct nfs_client *clp)
+{
+       /* This will need to handle layouts too */
+       nfs_expire_all_delegations(clp);
+}
+
+static void nfs41_handle_cb_path_down(struct nfs_client *clp)
+{
+       nfs_expire_all_delegations(clp);
+       if (test_and_set_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state) == 0)
+               nfs4_schedule_state_recovery(clp);
+}
+
+void nfs41_handle_sequence_flag_errors(struct nfs_client *clp, u32 flags)
+{
+       if (!flags)
+               return;
+       else if (flags & SEQ4_STATUS_RESTART_RECLAIM_NEEDED)
+               nfs41_handle_server_reboot(clp);
+       else if (flags & (SEQ4_STATUS_EXPIRED_ALL_STATE_REVOKED |
+                           SEQ4_STATUS_EXPIRED_SOME_STATE_REVOKED |
+                           SEQ4_STATUS_ADMIN_STATE_REVOKED |
+                           SEQ4_STATUS_LEASE_MOVED))
+               nfs41_handle_state_revoked(clp);
+       else if (flags & SEQ4_STATUS_RECALLABLE_STATE_REVOKED)
+               nfs41_handle_recallable_state_revoked(clp);
+       else if (flags & (SEQ4_STATUS_CB_PATH_DOWN |
+                           SEQ4_STATUS_BACKCHANNEL_FAULT |
+                           SEQ4_STATUS_CB_PATH_DOWN_SESSION))
+               nfs41_handle_cb_path_down(clp);
+}
+
+static int nfs4_reset_session(struct nfs_client *clp)
+{
+       int status;
+
+       nfs4_begin_drain_session(clp);
+       status = nfs4_proc_destroy_session(clp->cl_session);
+       if (status && status != -NFS4ERR_BADSESSION &&
+           status != -NFS4ERR_DEADSESSION) {
+               status = nfs4_recovery_handle_error(clp, status);
+               goto out;
+       }
+
+       memset(clp->cl_session->sess_id.data, 0, NFS4_MAX_SESSIONID_LEN);
+       status = nfs4_proc_create_session(clp);
+       if (status) {
+               status = nfs4_recovery_handle_error(clp, status);
+               goto out;
+       }
+       /* create_session negotiated new slot table */
+       clear_bit(NFS4CLNT_RECALL_SLOT, &clp->cl_state);
+
+        /* Let the state manager reestablish state */
+       if (!test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state))
+               nfs41_setup_state_renewal(clp);
+out:
+       return status;
+}
+
+static int nfs4_recall_slot(struct nfs_client *clp)
+{
+       struct nfs4_slot_table *fc_tbl = &clp->cl_session->fc_slot_table;
+       struct nfs4_channel_attrs *fc_attrs = &clp->cl_session->fc_attrs;
+       struct nfs4_slot *new, *old;
+       int i;
+
+       nfs4_begin_drain_session(clp);
+       new = kmalloc(fc_tbl->target_max_slots * sizeof(struct nfs4_slot),
+                     GFP_KERNEL);
+        if (!new)
+               return -ENOMEM;
+
+       spin_lock(&fc_tbl->slot_tbl_lock);
+       for (i = 0; i < fc_tbl->target_max_slots; i++)
+               new[i].seq_nr = fc_tbl->slots[i].seq_nr;
+       old = fc_tbl->slots;
+       fc_tbl->slots = new;
+       fc_tbl->max_slots = fc_tbl->target_max_slots;
+       fc_tbl->target_max_slots = 0;
+       fc_attrs->max_reqs = fc_tbl->max_slots;
+       spin_unlock(&fc_tbl->slot_tbl_lock);
+
+       kfree(old);
+       nfs4_end_drain_session(clp);
+       return 0;
+}
+
+#else /* CONFIG_NFS_V4_1 */
+static int nfs4_reset_session(struct nfs_client *clp) { return 0; }
+static int nfs4_end_drain_session(struct nfs_client *clp) { return 0; }
+static int nfs4_recall_slot(struct nfs_client *clp) { return 0; }
+#endif /* CONFIG_NFS_V4_1 */
+
+/* Set NFS4CLNT_LEASE_EXPIRED for all v4.0 errors and for recoverable errors
+ * on EXCHANGE_ID for v4.1
+ */
+static void nfs4_set_lease_expired(struct nfs_client *clp, int status)
+{
+       if (nfs4_has_session(clp)) {
                switch (status) {
-                       case 0:
-                       case -NFS4ERR_CB_PATH_DOWN:
-                               goto out;
-                       case -NFS4ERR_STALE_CLIENTID:
-                       case -NFS4ERR_LEASE_MOVED:
-                               ops = &nfs4_reboot_recovery_ops;
+               case -NFS4ERR_DELAY:
+               case -NFS4ERR_CLID_INUSE:
+               case -EAGAIN:
+               case -EKEYEXPIRED:
+                       break;
+
+               case -NFS4ERR_NOT_SAME: /* FixMe: implement recovery
+                                        * in nfs4_exchange_id */
+               default:
+                       return;
                }
+       }
+       set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
+}
+
+static void nfs4_state_manager(struct nfs_client *clp)
+{
+       int status = 0;
 
-               /* We're going to have to re-establish a clientid */
-               nfs4_state_mark_reclaim(clp);
+       /* Ensure exclusive access to NFSv4 state */
+       for(;;) {
+               if (test_and_clear_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state)) {
+                       /* We're going to have to re-establish a clientid */
+                       status = nfs4_reclaim_lease(clp);
+                       if (status) {
+                               nfs4_set_lease_expired(clp, status);
+                               if (test_bit(NFS4CLNT_LEASE_EXPIRED,
+                                                       &clp->cl_state))
+                                       continue;
+                               if (clp->cl_cons_state ==
+                                                       NFS_CS_SESSION_INITING)
+                                       nfs_mark_client_ready(clp, status);
+                               goto out_error;
+                       }
+                       clear_bit(NFS4CLNT_CHECK_LEASE, &clp->cl_state);
+                       set_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state);
+               }
 
-               status = nfs4_reclaim_lease(clp);
-               if (status) {
-                       if (status == -EAGAIN)
+               if (test_and_clear_bit(NFS4CLNT_CHECK_LEASE, &clp->cl_state)) {
+                       status = nfs4_check_lease(clp);
+                       if (test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state))
                                continue;
-                       goto out_error;
+                       if (status < 0 && status != -NFS4ERR_CB_PATH_DOWN)
+                               goto out_error;
                }
 
-               /* Mark all delegations for reclaim */
-               nfs_delegation_mark_reclaim(clp);
-               /* Note: list is protected by exclusive lock on cl->cl_sem */
-               status = nfs4_do_reclaim(clp, ops);
-               if (status < 0) {
-                       if (status == -NFS4ERR_NO_GRACE) {
-                               ops = &nfs4_network_partition_recovery_ops;
-                               status = nfs4_do_reclaim(clp, ops);
-                       }
-                       if (status == -NFS4ERR_STALE_CLIENTID)
+               /* Initialize or reset the session */
+               if (test_and_clear_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state)
+                  && nfs4_has_session(clp)) {
+                       status = nfs4_reset_session(clp);
+                       if (test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state))
+                               continue;
+                       if (status < 0)
+                               goto out_error;
+               }
+
+               /* First recover reboot state... */
+               if (test_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state)) {
+                       status = nfs4_do_reclaim(clp,
+                               nfs4_reboot_recovery_ops[clp->cl_minorversion]);
+                       if (test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state) ||
+                           test_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state))
                                continue;
-                       if (status == -NFS4ERR_EXPIRED)
+                       nfs4_state_end_reclaim_reboot(clp);
+                       if (test_bit(NFS4CLNT_RECLAIM_NOGRACE, &clp->cl_state))
                                continue;
+                       if (status < 0)
+                               goto out_error;
                }
-               nfs_delegation_reap_unclaimed(clp);
-               break;
+
+               /* Now recover expired state... */
+               if (test_and_clear_bit(NFS4CLNT_RECLAIM_NOGRACE, &clp->cl_state)) {
+                       status = nfs4_do_reclaim(clp,
+                               nfs4_nograce_recovery_ops[clp->cl_minorversion]);
+                       if (test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state) ||
+                           test_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state) ||
+                           test_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state))
+                               continue;
+                       if (status < 0)
+                               goto out_error;
+               }
+
+               nfs4_end_drain_session(clp);
+               if (test_and_clear_bit(NFS4CLNT_DELEGRETURN, &clp->cl_state)) {
+                       nfs_client_return_marked_delegations(clp);
+                       continue;
+               }
+               /* Recall session slots */
+               if (test_and_clear_bit(NFS4CLNT_RECALL_SLOT, &clp->cl_state)
+                  && nfs4_has_session(clp)) {
+                       status = nfs4_recall_slot(clp);
+                       if (status < 0)
+                               goto out_error;
+                       continue;
+               }
+
+
+               nfs4_clear_state_manager_bit(clp);
+               /* Did we race with an attempt to give us more work? */
+               if (clp->cl_state == 0)
+                       break;
+               if (test_and_set_bit(NFS4CLNT_MANAGER_RUNNING, &clp->cl_state) != 0)
+                       break;
        }
-out:
-       up_write(&clp->cl_sem);
-       if (status == -NFS4ERR_CB_PATH_DOWN)
-               nfs_handle_cb_pathdown(clp);
-       nfs4_clear_recover_bit(clp);
+       return;
+out_error:
+       printk(KERN_WARNING "Error: state manager failed on NFSv4 server %s"
+                       " with error %d\n", clp->cl_hostname, -status);
+       nfs4_end_drain_session(clp);
+       nfs4_clear_state_manager_bit(clp);
+}
+
+static int nfs4_run_state_manager(void *ptr)
+{
+       struct nfs_client *clp = ptr;
+
+       allow_signal(SIGKILL);
+       nfs4_state_manager(clp);
        nfs_put_client(clp);
        module_put_and_exit(0);
        return 0;
-out_error:
-       printk(KERN_WARNING "Error: state recovery failed on NFSv4 server %s"
-                       " with error %d\n", clp->cl_hostname, -status);
-       set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
-       goto out;
 }
 
 /*