]> git.karo-electronics.de Git - mv-sheeva.git/blobdiff - fs/nfsd/nfs4state.c
nfsd4: fix downgrade/lock logic
[mv-sheeva.git] / fs / nfsd / nfs4state.c
index 6a8fedaa4f55a8ad0f6236296063e385d8eb1a17..69b266db7f5cf18cfb42f4e71e1e9973bd634711 100644 (file)
 #define NFSDDBG_FACILITY                NFSDDBG_PROC
 
 /* Globals */
-static time_t lease_time = 90;     /* default lease time */
-static time_t user_lease_time = 90;
+time_t nfsd4_lease = 90;     /* default lease time */
+time_t nfsd4_grace = 90;
 static time_t boot_time;
 static u32 current_ownerid = 1;
 static u32 current_fileid = 1;
 static u32 current_delegid = 1;
-static u32 nfs4_init;
 static stateid_t zerostateid;             /* bits all 0 */
 static stateid_t onestateid;              /* bits all 1 */
 static u64 current_sessionid = 1;
@@ -163,6 +162,46 @@ static struct list_head    ownerstr_hashtbl[OWNER_HASH_SIZE];
 static struct list_head file_hashtbl[FILE_HASH_SIZE];
 static struct list_head stateid_hashtbl[STATEID_HASH_SIZE];
 
+static void __nfs4_file_get_access(struct nfs4_file *fp, int oflag)
+{
+       BUG_ON(!(fp->fi_fds[oflag] || fp->fi_fds[O_RDWR]));
+       atomic_inc(&fp->fi_access[oflag]);
+}
+
+static void nfs4_file_get_access(struct nfs4_file *fp, int oflag)
+{
+       if (oflag == O_RDWR) {
+               __nfs4_file_get_access(fp, O_RDONLY);
+               __nfs4_file_get_access(fp, O_WRONLY);
+       } else
+               __nfs4_file_get_access(fp, oflag);
+}
+
+static void nfs4_file_put_fd(struct nfs4_file *fp, int oflag)
+{
+       if (fp->fi_fds[oflag]) {
+               fput(fp->fi_fds[oflag]);
+               fp->fi_fds[oflag] = NULL;
+       }
+}
+
+static void __nfs4_file_put_access(struct nfs4_file *fp, int oflag)
+{
+       if (atomic_dec_and_test(&fp->fi_access[oflag])) {
+               nfs4_file_put_fd(fp, O_RDWR);
+               nfs4_file_put_fd(fp, oflag);
+       }
+}
+
+static void nfs4_file_put_access(struct nfs4_file *fp, int oflag)
+{
+       if (oflag == O_RDWR) {
+               __nfs4_file_put_access(fp, O_RDONLY);
+               __nfs4_file_put_access(fp, O_WRONLY);
+       } else
+               __nfs4_file_put_access(fp, oflag);
+}
+
 static struct nfs4_delegation *
 alloc_init_deleg(struct nfs4_client *clp, struct nfs4_stateid *stp, struct svc_fh *current_fh, u32 type)
 {
@@ -171,6 +210,13 @@ alloc_init_deleg(struct nfs4_client *clp, struct nfs4_stateid *stp, struct svc_f
        struct nfs4_cb_conn *cb = &stp->st_stateowner->so_client->cl_cb_conn;
 
        dprintk("NFSD alloc_init_deleg\n");
+       /*
+        * Major work on the lease subsystem (for example, to support
+        * calbacks on stat) will be required before we can support
+        * write delegations properly.
+        */
+       if (type != NFS4_OPEN_DELEGATE_READ)
+               return NULL;
        if (fp->fi_had_conflict)
                return NULL;
        if (num_delegations > max_delegations)
@@ -185,12 +231,11 @@ alloc_init_deleg(struct nfs4_client *clp, struct nfs4_stateid *stp, struct svc_f
        dp->dl_client = clp;
        get_nfs4_file(fp);
        dp->dl_file = fp;
+       nfs4_file_get_access(fp, O_RDONLY);
        dp->dl_flock = NULL;
-       get_file(stp->st_vfs_file);
-       dp->dl_vfs_file = stp->st_vfs_file;
        dp->dl_type = type;
        dp->dl_ident = cb->cb_ident;
-       dp->dl_stateid.si_boot = get_seconds();
+       dp->dl_stateid.si_boot = boot_time;
        dp->dl_stateid.si_stateownerid = current_delegid++;
        dp->dl_stateid.si_fileid = 0;
        dp->dl_stateid.si_generation = 0;
@@ -199,6 +244,7 @@ alloc_init_deleg(struct nfs4_client *clp, struct nfs4_stateid *stp, struct svc_f
        atomic_set(&dp->dl_count, 1);
        list_add(&dp->dl_perfile, &fp->fi_delegations);
        list_add(&dp->dl_perclnt, &clp->cl_delegations);
+       INIT_WORK(&dp->dl_recall.cb_work, nfsd4_do_callback_rpc);
        return dp;
 }
 
@@ -221,15 +267,12 @@ nfs4_put_delegation(struct nfs4_delegation *dp)
 static void
 nfs4_close_delegation(struct nfs4_delegation *dp)
 {
-       struct file *filp = dp->dl_vfs_file;
+       struct file *filp = find_readable_file(dp->dl_file);
 
        dprintk("NFSD: close_delegation dp %p\n",dp);
-       dp->dl_vfs_file = NULL;
-       /* The following nfsd_close may not actually close the file,
-        * but we want to remove the lease in any case. */
        if (dp->dl_flock)
                vfs_setlease(filp, F_UNLCK, &dp->dl_flock);
-       nfsd_close(filp);
+       nfs4_file_put_access(dp->dl_file, O_RDONLY);
 }
 
 /* Called under the state lock. */
@@ -249,6 +292,9 @@ unhash_delegation(struct nfs4_delegation *dp)
  * SETCLIENTID state 
  */
 
+/* client_lock protects the client lru list and session hash table */
+static DEFINE_SPINLOCK(client_lock);
+
 /* Hash tables for nfs4_clientid state */
 #define CLIENT_HASH_BITS                 4
 #define CLIENT_HASH_SIZE                (1 << CLIENT_HASH_BITS)
@@ -298,8 +344,12 @@ static void free_generic_stateid(struct nfs4_stateid *stp)
 
 static void release_lock_stateid(struct nfs4_stateid *stp)
 {
+       struct file *file;
+
        unhash_generic_stateid(stp);
-       locks_remove_posix(stp->st_vfs_file, (fl_owner_t)stp->st_stateowner);
+       file = find_any_file(stp->st_file);
+       if (file)
+               locks_remove_posix(file, (fl_owner_t)stp->st_stateowner);
        free_generic_stateid(stp);
 }
 
@@ -337,11 +387,85 @@ release_stateid_lockowners(struct nfs4_stateid *open_stp)
        }
 }
 
+/*
+ * We store the NONE, READ, WRITE, and BOTH bits separately in the
+ * st_{access,deny}_bmap field of the stateid, in order to track not
+ * only what share bits are currently in force, but also what
+ * combinations of share bits previous opens have used.  This allows us
+ * to enforce the recommendation of rfc 3530 14.2.19 that the server
+ * return an error if the client attempt to downgrade to a combination
+ * of share bits not explicable by closing some of its previous opens.
+ *
+ * XXX: This enforcement is actually incomplete, since we don't keep
+ * track of access/deny bit combinations; so, e.g., we allow:
+ *
+ *     OPEN allow read, deny write
+ *     OPEN allow both, deny none
+ *     DOWNGRADE allow read, deny none
+ *
+ * which we should reject.
+ */
+static void
+set_access(unsigned int *access, unsigned long bmap) {
+       int i;
+
+       *access = 0;
+       for (i = 1; i < 4; i++) {
+               if (test_bit(i, &bmap))
+                       *access |= i;
+       }
+}
+
+static void
+set_deny(unsigned int *deny, unsigned long bmap) {
+       int i;
+
+       *deny = 0;
+       for (i = 0; i < 4; i++) {
+               if (test_bit(i, &bmap))
+                       *deny |= i ;
+       }
+}
+
+static int
+test_share(struct nfs4_stateid *stp, struct nfsd4_open *open) {
+       unsigned int access, deny;
+
+       set_access(&access, stp->st_access_bmap);
+       set_deny(&deny, stp->st_deny_bmap);
+       if ((access & open->op_share_deny) || (deny & open->op_share_access))
+               return 0;
+       return 1;
+}
+
+static int nfs4_access_to_omode(u32 access)
+{
+       switch (access) {
+       case NFS4_SHARE_ACCESS_READ:
+               return O_RDONLY;
+       case NFS4_SHARE_ACCESS_WRITE:
+               return O_WRONLY;
+       case NFS4_SHARE_ACCESS_BOTH:
+               return O_RDWR;
+       }
+       BUG();
+}
+
+static int nfs4_access_bmap_to_omode(struct nfs4_stateid *stp)
+{
+       unsigned int access;
+
+       set_access(&access, stp->st_access_bmap);
+       return nfs4_access_to_omode(access);
+}
+
 static void release_open_stateid(struct nfs4_stateid *stp)
 {
+       int oflag = nfs4_access_bmap_to_omode(stp);
+
        unhash_generic_stateid(stp);
        release_stateid_lockowners(stp);
-       nfsd_close(stp->st_vfs_file);
+       nfs4_file_put_access(stp->st_file, oflag);
        free_generic_stateid(stp);
 }
 
@@ -367,7 +491,6 @@ static void release_openowner(struct nfs4_stateowner *sop)
        nfs4_put_stateowner(sop);
 }
 
-static DEFINE_SPINLOCK(sessionid_lock);
 #define SESSION_HASH_SIZE      512
 static struct list_head sessionid_hashtbl[SESSION_HASH_SIZE];
 
@@ -454,7 +577,7 @@ static int set_forechannel_drc_size(struct nfsd4_channel_attrs *fchan)
        spin_unlock(&nfsd_drc_lock);
 
        if (fchan->maxreqs == 0)
-               return nfserr_serverfault;
+               return nfserr_jukebox;
 
        fchan->maxresp_cached = size + NFSD_MIN_HDR_SEQ_SZ;
        return 0;
@@ -539,7 +662,7 @@ alloc_init_session(struct svc_rqst *rqstp, struct nfs4_client *clp,
        BUILD_BUG_ON(NFSD_MAX_SLOTS_PER_SESSION * sizeof(struct nfsd4_slot)
                     + sizeof(struct nfsd4_session) > PAGE_SIZE);
 
-       status = nfserr_serverfault;
+       status = nfserr_jukebox;
        /* allocate struct nfsd4_session and slot table pointers in one piece */
        slotsize = tmp.se_fchannel.maxreqs * sizeof(struct nfsd4_slot *);
        new = kzalloc(sizeof(*new) + slotsize, GFP_KERNEL);
@@ -565,10 +688,10 @@ alloc_init_session(struct svc_rqst *rqstp, struct nfs4_client *clp,
 
        new->se_flags = cses->flags;
        kref_init(&new->se_ref);
-       spin_lock(&sessionid_lock);
+       spin_lock(&client_lock);
        list_add(&new->se_hash, &sessionid_hashtbl[idx]);
        list_add(&new->se_perclnt, &clp->cl_sessions);
-       spin_unlock(&sessionid_lock);
+       spin_unlock(&client_lock);
 
        status = nfs_ok;
 out:
@@ -579,7 +702,7 @@ out_free:
        goto out;
 }
 
-/* caller must hold sessionid_lock */
+/* caller must hold client_lock */
 static struct nfsd4_session *
 find_in_sessionid_hashtbl(struct nfs4_sessionid *sessionid)
 {
@@ -588,10 +711,8 @@ find_in_sessionid_hashtbl(struct nfs4_sessionid *sessionid)
 
        dump_sessionid(__func__, sessionid);
        idx = hash_sessionid(sessionid);
-       dprintk("%s: idx is %d\n", __func__, idx);
        /* Search in the appropriate list */
        list_for_each_entry(elem, &sessionid_hashtbl[idx], se_hash) {
-               dump_sessionid("list traversal", &elem->se_sessionid);
                if (!memcmp(elem->se_sessionid.data, sessionid->data,
                            NFS4_MAX_SESSIONID_LEN)) {
                        return elem;
@@ -602,7 +723,7 @@ find_in_sessionid_hashtbl(struct nfs4_sessionid *sessionid)
        return NULL;
 }
 
-/* caller must hold sessionid_lock */
+/* caller must hold client_lock */
 static void
 unhash_session(struct nfsd4_session *ses)
 {
@@ -610,15 +731,6 @@ unhash_session(struct nfsd4_session *ses)
        list_del(&ses->se_perclnt);
 }
 
-static void
-release_session(struct nfsd4_session *ses)
-{
-       spin_lock(&sessionid_lock);
-       unhash_session(ses);
-       spin_unlock(&sessionid_lock);
-       nfsd4_put_session(ses);
-}
-
 void
 free_session(struct kref *kref)
 {
@@ -634,9 +746,18 @@ free_session(struct kref *kref)
        kfree(ses);
 }
 
+/* must be called under the client_lock */
 static inline void
-renew_client(struct nfs4_client *clp)
+renew_client_locked(struct nfs4_client *clp)
 {
+       if (is_client_expired(clp)) {
+               dprintk("%s: client (clientid %08x/%08x) already expired\n",
+                       __func__,
+                       clp->cl_clientid.cl_boot,
+                       clp->cl_clientid.cl_id);
+               return;
+       }
+
        /*
        * Move client to the end to the LRU list.
        */
@@ -647,6 +768,14 @@ renew_client(struct nfs4_client *clp)
        clp->cl_time = get_seconds();
 }
 
+static inline void
+renew_client(struct nfs4_client *clp)
+{
+       spin_lock(&client_lock);
+       renew_client_locked(clp);
+       spin_unlock(&client_lock);
+}
+
 /* SETCLIENTID and SETCLIENTID_CONFIRM Helper functions */
 static int
 STALE_CLIENTID(clientid_t *clid)
@@ -680,27 +809,9 @@ static struct nfs4_client *alloc_client(struct xdr_netobj name)
        return clp;
 }
 
-static void
-shutdown_callback_client(struct nfs4_client *clp)
-{
-       struct rpc_clnt *clnt = clp->cl_cb_conn.cb_client;
-
-       if (clnt) {
-               /*
-                * Callback threads take a reference on the client, so there
-                * should be no outstanding callbacks at this point.
-                */
-               clp->cl_cb_conn.cb_client = NULL;
-               rpc_shutdown_client(clnt);
-       }
-}
-
 static inline void
 free_client(struct nfs4_client *clp)
 {
-       shutdown_callback_client(clp);
-       if (clp->cl_cb_xprt)
-               svc_xprt_put(clp->cl_cb_xprt);
        if (clp->cl_cred.cr_group_info)
                put_group_info(clp->cl_cred.cr_group_info);
        kfree(clp->cl_principal);
@@ -709,10 +820,33 @@ free_client(struct nfs4_client *clp)
 }
 
 void
-put_nfs4_client(struct nfs4_client *clp)
+release_session_client(struct nfsd4_session *session)
 {
-       if (atomic_dec_and_test(&clp->cl_count))
+       struct nfs4_client *clp = session->se_client;
+
+       if (!atomic_dec_and_lock(&clp->cl_refcount, &client_lock))
+               return;
+       if (is_client_expired(clp)) {
                free_client(clp);
+               session->se_client = NULL;
+       } else
+               renew_client_locked(clp);
+       spin_unlock(&client_lock);
+}
+
+/* must be called under the client_lock */
+static inline void
+unhash_client_locked(struct nfs4_client *clp)
+{
+       mark_client_expired(clp);
+       list_del(&clp->cl_lru);
+       while (!list_empty(&clp->cl_sessions)) {
+               struct nfsd4_session  *ses;
+               ses = list_entry(clp->cl_sessions.next, struct nfsd4_session,
+                                se_perclnt);
+               unhash_session(ses);
+               nfsd4_put_session(ses);
+       }
 }
 
 static void
@@ -722,9 +856,6 @@ expire_client(struct nfs4_client *clp)
        struct nfs4_delegation *dp;
        struct list_head reaplist;
 
-       dprintk("NFSD: expire_client cl_count %d\n",
-                           atomic_read(&clp->cl_count));
-
        INIT_LIST_HEAD(&reaplist);
        spin_lock(&recall_lock);
        while (!list_empty(&clp->cl_delegations)) {
@@ -740,20 +871,20 @@ expire_client(struct nfs4_client *clp)
                list_del_init(&dp->dl_recall_lru);
                unhash_delegation(dp);
        }
-       list_del(&clp->cl_idhash);
-       list_del(&clp->cl_strhash);
-       list_del(&clp->cl_lru);
        while (!list_empty(&clp->cl_openowners)) {
                sop = list_entry(clp->cl_openowners.next, struct nfs4_stateowner, so_perclient);
                release_openowner(sop);
        }
-       while (!list_empty(&clp->cl_sessions)) {
-               struct nfsd4_session  *ses;
-               ses = list_entry(clp->cl_sessions.next, struct nfsd4_session,
-                                se_perclnt);
-               release_session(ses);
-       }
-       put_nfs4_client(clp);
+       nfsd4_set_callback_client(clp, NULL);
+       if (clp->cl_cb_conn.cb_xprt)
+               svc_xprt_put(clp->cl_cb_conn.cb_xprt);
+       list_del(&clp->cl_idhash);
+       list_del(&clp->cl_strhash);
+       spin_lock(&client_lock);
+       unhash_client_locked(clp);
+       if (atomic_read(&clp->cl_refcount) == 0)
+               free_client(clp);
+       spin_unlock(&client_lock);
 }
 
 static void copy_verf(struct nfs4_client *target, nfs4_verifier *source)
@@ -839,14 +970,15 @@ static struct nfs4_client *create_client(struct xdr_netobj name, char *recdir,
        }
 
        memcpy(clp->cl_recdir, recdir, HEXDIR_LEN);
-       atomic_set(&clp->cl_count, 1);
-       atomic_set(&clp->cl_cb_conn.cb_set, 0);
+       atomic_set(&clp->cl_refcount, 0);
+       atomic_set(&clp->cl_cb_set, 0);
        INIT_LIST_HEAD(&clp->cl_idhash);
        INIT_LIST_HEAD(&clp->cl_strhash);
        INIT_LIST_HEAD(&clp->cl_openowners);
        INIT_LIST_HEAD(&clp->cl_delegations);
        INIT_LIST_HEAD(&clp->cl_sessions);
        INIT_LIST_HEAD(&clp->cl_lru);
+       clp->cl_time = get_seconds();
        clear_bit(0, &clp->cl_cb_slot_busy);
        rpc_init_wait_queue(&clp->cl_cb_waitq, "Backchannel slot table");
        copy_verf(clp, verf);
@@ -877,8 +1009,7 @@ add_to_unconfirmed(struct nfs4_client *clp, unsigned int strhashval)
        list_add(&clp->cl_strhash, &unconf_str_hashtbl[strhashval]);
        idhashval = clientid_hashval(clp->cl_clientid.cl_id);
        list_add(&clp->cl_idhash, &unconf_id_hashtbl[idhashval]);
-       list_add_tail(&clp->cl_lru, &client_lru);
-       clp->cl_time = get_seconds();
+       renew_client(clp);
 }
 
 static void
@@ -888,10 +1019,9 @@ move_to_confirmed(struct nfs4_client *clp)
        unsigned int strhashval;
 
        dprintk("NFSD: move_to_confirm nfs4_client %p\n", clp);
-       list_del_init(&clp->cl_strhash);
        list_move(&clp->cl_idhash, &conf_id_hashtbl[idhashval]);
        strhashval = clientstr_hashval(clp->cl_recdir);
-       list_add(&clp->cl_strhash, &conf_str_hashtbl[strhashval]);
+       list_move(&clp->cl_strhash, &conf_str_hashtbl[strhashval]);
        renew_client(clp);
 }
 
@@ -1207,7 +1337,7 @@ out_new:
        /* Normal case */
        new = create_client(exid->clname, dname, rqstp, &verf);
        if (new == NULL) {
-               status = nfserr_serverfault;
+               status = nfserr_jukebox;
                goto out;
        }
 
@@ -1327,15 +1457,9 @@ nfsd4_create_session(struct svc_rqst *rqstp,
                cs_slot->sl_seqid++; /* from 0 to 1 */
                move_to_confirmed(unconf);
 
-               /*
-                * We do not support RDMA or persistent sessions
-                */
-               cr_ses->flags &= ~SESSION4_PERSIST;
-               cr_ses->flags &= ~SESSION4_RDMA;
-
                if (cr_ses->flags & SESSION4_BACK_CHAN) {
-                       unconf->cl_cb_xprt = rqstp->rq_xprt;
-                       svc_xprt_get(unconf->cl_cb_xprt);
+                       unconf->cl_cb_conn.cb_xprt = rqstp->rq_xprt;
+                       svc_xprt_get(rqstp->rq_xprt);
                        rpc_copy_addr(
                                (struct sockaddr *)&unconf->cl_cb_conn.cb_addr,
                                sa);
@@ -1344,7 +1468,7 @@ nfsd4_create_session(struct svc_rqst *rqstp,
                                cstate->minorversion;
                        unconf->cl_cb_conn.cb_prog = cr_ses->callback_prog;
                        unconf->cl_cb_seq_nr = 1;
-                       nfsd4_probe_callback(unconf);
+                       nfsd4_probe_callback(unconf, &unconf->cl_cb_conn);
                }
                conf = unconf;
        } else {
@@ -1352,6 +1476,12 @@ nfsd4_create_session(struct svc_rqst *rqstp,
                goto out;
        }
 
+       /*
+        * We do not support RDMA or persistent sessions
+        */
+       cr_ses->flags &= ~SESSION4_PERSIST;
+       cr_ses->flags &= ~SESSION4_RDMA;
+
        status = alloc_init_session(rqstp, conf, cr_ses);
        if (status)
                goto out;
@@ -1369,6 +1499,21 @@ out:
        return status;
 }
 
+static bool nfsd4_last_compound_op(struct svc_rqst *rqstp)
+{
+       struct nfsd4_compoundres *resp = rqstp->rq_resp;
+       struct nfsd4_compoundargs *argp = rqstp->rq_argp;
+
+       return argp->opcnt == resp->opcnt;
+}
+
+static bool nfsd4_compound_in_session(struct nfsd4_session *session, struct nfs4_sessionid *sid)
+{
+       if (!session)
+               return 0;
+       return !memcmp(sid, &session->se_sessionid, sizeof(*sid));
+}
+
 __be32
 nfsd4_destroy_session(struct svc_rqst *r,
                      struct nfsd4_compound_state *cstate,
@@ -1384,19 +1529,25 @@ nfsd4_destroy_session(struct svc_rqst *r,
         * - Do we need to clear any callback info from previous session?
         */
 
+       if (nfsd4_compound_in_session(cstate->session, &sessionid->sessionid)) {
+               if (!nfsd4_last_compound_op(r))
+                       return nfserr_not_only_op;
+       }
        dump_sessionid(__func__, &sessionid->sessionid);
-       spin_lock(&sessionid_lock);
+       spin_lock(&client_lock);
        ses = find_in_sessionid_hashtbl(&sessionid->sessionid);
        if (!ses) {
-               spin_unlock(&sessionid_lock);
+               spin_unlock(&client_lock);
                goto out;
        }
 
        unhash_session(ses);
-       spin_unlock(&sessionid_lock);
+       spin_unlock(&client_lock);
 
+       nfs4_lock_state();
        /* wait for callbacks */
-       shutdown_callback_client(ses->se_client);
+       nfsd4_set_callback_client(ses->se_client, NULL);
+       nfs4_unlock_state();
        nfsd4_put_session(ses);
        status = nfs_ok;
 out:
@@ -1417,7 +1568,7 @@ nfsd4_sequence(struct svc_rqst *rqstp,
        if (resp->opcnt != 1)
                return nfserr_sequence_pos;
 
-       spin_lock(&sessionid_lock);
+       spin_lock(&client_lock);
        status = nfserr_badsession;
        session = find_in_sessionid_hashtbl(&seq->sessionid);
        if (!session)
@@ -1456,22 +1607,46 @@ nfsd4_sequence(struct svc_rqst *rqstp,
        cstate->slot = slot;
        cstate->session = session;
 
-       /* Hold a session reference until done processing the compound:
-        * nfsd4_put_session called only if the cstate slot is set.
-        */
-       nfsd4_get_session(session);
 out:
-       spin_unlock(&sessionid_lock);
-       /* Renew the clientid on success and on replay */
+       /* Hold a session reference until done processing the compound. */
        if (cstate->session) {
-               nfs4_lock_state();
-               renew_client(session->se_client);
-               nfs4_unlock_state();
+               nfsd4_get_session(cstate->session);
+               atomic_inc(&session->se_client->cl_refcount);
        }
+       spin_unlock(&client_lock);
        dprintk("%s: return %d\n", __func__, ntohl(status));
        return status;
 }
 
+__be32
+nfsd4_reclaim_complete(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate, struct nfsd4_reclaim_complete *rc)
+{
+       if (rc->rca_one_fs) {
+               if (!cstate->current_fh.fh_dentry)
+                       return nfserr_nofilehandle;
+               /*
+                * We don't take advantage of the rca_one_fs case.
+                * That's OK, it's optional, we can safely ignore it.
+                */
+                return nfs_ok;
+       }
+       nfs4_lock_state();
+       if (is_client_expired(cstate->session->se_client)) {
+               nfs4_unlock_state();
+               /*
+                * The following error isn't really legal.
+                * But we only get here if the client just explicitly
+                * destroyed the client.  Surely it no longer cares what
+                * error it gets back on an operation for the dead
+                * client.
+                */
+               return nfserr_stale_clientid;
+       }
+       nfsd4_create_clid_dir(cstate->session->se_client);
+       nfs4_unlock_state();
+       return nfs_ok;
+}
+
 __be32
 nfsd4_setclientid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
                  struct nfsd4_setclientid *setclid)
@@ -1631,9 +1806,8 @@ nfsd4_setclientid_confirm(struct svc_rqst *rqstp,
                if (!same_creds(&conf->cl_cred, &unconf->cl_cred))
                        status = nfserr_clid_inuse;
                else {
-                       /* XXX: We just turn off callbacks until we can handle
-                         * change request correctly. */
-                       atomic_set(&conf->cl_cb_conn.cb_set, 0);
+                       atomic_set(&conf->cl_cb_set, 0);
+                       nfsd4_probe_callback(conf, &unconf->cl_cb_conn);
                        expire_client(unconf);
                        status = nfs_ok;
 
@@ -1667,7 +1841,7 @@ nfsd4_setclientid_confirm(struct svc_rqst *rqstp,
                        }
                        move_to_confirmed(unconf);
                        conf = unconf;
-                       nfsd4_probe_callback(conf);
+                       nfsd4_probe_callback(conf, &conf->cl_cb_conn);
                        status = nfs_ok;
                }
        } else if ((!conf || (conf && !same_verf(&conf->cl_confirm, &confirm)))
@@ -1700,12 +1874,14 @@ alloc_init_file(struct inode *ino)
                INIT_LIST_HEAD(&fp->fi_hash);
                INIT_LIST_HEAD(&fp->fi_stateids);
                INIT_LIST_HEAD(&fp->fi_delegations);
-               spin_lock(&recall_lock);
-               list_add(&fp->fi_hash, &file_hashtbl[hashval]);
-               spin_unlock(&recall_lock);
                fp->fi_inode = igrab(ino);
                fp->fi_id = current_fileid++;
                fp->fi_had_conflict = false;
+               memset(fp->fi_fds, 0, sizeof(fp->fi_fds));
+               memset(fp->fi_access, 0, sizeof(fp->fi_access));
+               spin_lock(&recall_lock);
+               list_add(&fp->fi_hash, &file_hashtbl[hashval]);
+               spin_unlock(&recall_lock);
                return fp;
        }
        return NULL;
@@ -1827,7 +2003,7 @@ init_stateid(struct nfs4_stateid *stp, struct nfs4_file *fp, struct nfsd4_open *
        stp->st_stateowner = sop;
        get_nfs4_file(fp);
        stp->st_file = fp;
-       stp->st_stateid.si_boot = get_seconds();
+       stp->st_stateid.si_boot = boot_time;
        stp->st_stateid.si_stateownerid = sop->so_id;
        stp->st_stateid.si_fileid = fp->fi_id;
        stp->st_stateid.si_generation = 0;
@@ -1913,57 +2089,6 @@ static inline int deny_valid(u32 x)
        return x <= NFS4_SHARE_DENY_BOTH;
 }
 
-/*
- * We store the NONE, READ, WRITE, and BOTH bits separately in the
- * st_{access,deny}_bmap field of the stateid, in order to track not
- * only what share bits are currently in force, but also what
- * combinations of share bits previous opens have used.  This allows us
- * to enforce the recommendation of rfc 3530 14.2.19 that the server
- * return an error if the client attempt to downgrade to a combination
- * of share bits not explicable by closing some of its previous opens.
- *
- * XXX: This enforcement is actually incomplete, since we don't keep
- * track of access/deny bit combinations; so, e.g., we allow:
- *
- *     OPEN allow read, deny write
- *     OPEN allow both, deny none
- *     DOWNGRADE allow read, deny none
- *
- * which we should reject.
- */
-static void
-set_access(unsigned int *access, unsigned long bmap) {
-       int i;
-
-       *access = 0;
-       for (i = 1; i < 4; i++) {
-               if (test_bit(i, &bmap))
-                       *access |= i;
-       }
-}
-
-static void
-set_deny(unsigned int *deny, unsigned long bmap) {
-       int i;
-
-       *deny = 0;
-       for (i = 0; i < 4; i++) {
-               if (test_bit(i, &bmap))
-                       *deny |= i ;
-       }
-}
-
-static int
-test_share(struct nfs4_stateid *stp, struct nfsd4_open *open) {
-       unsigned int access, deny;
-
-       set_access(&access, stp->st_access_bmap);
-       set_deny(&deny, stp->st_deny_bmap);
-       if ((access & open->op_share_deny) || (deny & open->op_share_access))
-               return 0;
-       return 1;
-}
-
 /*
  * Called to check deny when READ with all zero stateid or
  * WRITE with all zero or all one stateid
@@ -1995,14 +2120,12 @@ out:
 }
 
 static inline void
-nfs4_file_downgrade(struct file *filp, unsigned int share_access)
+nfs4_file_downgrade(struct nfs4_file *fp, unsigned int share_access)
 {
-       if (share_access & NFS4_SHARE_ACCESS_WRITE) {
-               drop_file_write_access(filp);
-               spin_lock(&filp->f_lock);
-               filp->f_mode = (filp->f_mode | FMODE_READ) & ~FMODE_WRITE;
-               spin_unlock(&filp->f_lock);
-       }
+       if (share_access & NFS4_SHARE_ACCESS_WRITE)
+               nfs4_file_put_access(fp, O_WRONLY);
+       if (share_access & NFS4_SHARE_ACCESS_READ)
+               nfs4_file_put_access(fp, O_RDONLY);
 }
 
 /*
@@ -2028,7 +2151,6 @@ void nfsd_break_deleg_cb(struct file_lock *fl)
         * lock) we know the server hasn't removed the lease yet, we know
         * it's safe to take a reference: */
        atomic_inc(&dp->dl_count);
-       atomic_inc(&dp->dl_client->cl_count);
 
        spin_lock(&recall_lock);
        list_add_tail(&dp->dl_recall_lru, &del_recall_lru);
@@ -2199,6 +2321,13 @@ find_delegation_file(struct nfs4_file *fp, stateid_t *stid)
        return NULL;
 }
 
+int share_access_to_flags(u32 share_access)
+{
+       share_access &= ~NFS4_SHARE_WANT_MASK;
+
+       return share_access == NFS4_SHARE_ACCESS_READ ? RD_STATE : WR_STATE;
+}
+
 static __be32
 nfs4_check_deleg(struct nfs4_file *fp, struct nfsd4_open *open,
                struct nfs4_delegation **dp)
@@ -2209,8 +2338,7 @@ nfs4_check_deleg(struct nfs4_file *fp, struct nfsd4_open *open,
        *dp = find_delegation_file(fp, &open->op_delegate_stateid);
        if (*dp == NULL)
                goto out;
-       flags = open->op_share_access == NFS4_SHARE_ACCESS_READ ?
-                                               RD_STATE : WR_STATE;
+       flags = share_access_to_flags(open->op_share_access);
        status = nfs4_check_delegmode(*dp, flags);
        if (status)
                *dp = NULL;
@@ -2252,30 +2380,53 @@ nfs4_alloc_stateid(void)
        return kmem_cache_alloc(stateid_slab, GFP_KERNEL);
 }
 
+static inline int nfs4_access_to_access(u32 nfs4_access)
+{
+       int flags = 0;
+
+       if (nfs4_access & NFS4_SHARE_ACCESS_READ)
+               flags |= NFSD_MAY_READ;
+       if (nfs4_access & NFS4_SHARE_ACCESS_WRITE)
+               flags |= NFSD_MAY_WRITE;
+       return flags;
+}
+
+static __be32 nfs4_get_vfs_file(struct svc_rqst *rqstp, struct nfs4_file
+*fp, struct svc_fh *cur_fh, u32 nfs4_access)
+{
+       __be32 status;
+       int oflag = nfs4_access_to_omode(nfs4_access);
+       int access = nfs4_access_to_access(nfs4_access);
+
+       if (!fp->fi_fds[oflag]) {
+               status = nfsd_open(rqstp, cur_fh, S_IFREG, access,
+                       &fp->fi_fds[oflag]);
+               if (status == nfserr_dropit)
+                       status = nfserr_jukebox;
+               if (status)
+                       return status;
+       }
+       nfs4_file_get_access(fp, oflag);
+
+       return nfs_ok;
+}
+
 static __be32
 nfs4_new_open(struct svc_rqst *rqstp, struct nfs4_stateid **stpp,
-               struct nfs4_delegation *dp,
-               struct svc_fh *cur_fh, int flags)
+               struct nfs4_file *fp, struct svc_fh *cur_fh,
+               struct nfsd4_open *open)
 {
        struct nfs4_stateid *stp;
+       __be32 status;
 
        stp = nfs4_alloc_stateid();
        if (stp == NULL)
                return nfserr_resource;
 
-       if (dp) {
-               get_file(dp->dl_vfs_file);
-               stp->st_vfs_file = dp->dl_vfs_file;
-       } else {
-               __be32 status;
-               status = nfsd_open(rqstp, cur_fh, S_IFREG, flags,
-                               &stp->st_vfs_file);
-               if (status) {
-                       if (status == nfserr_dropit)
-                               status = nfserr_jukebox;
-                       kmem_cache_free(stateid_slab, stp);
-                       return status;
-               }
+       status = nfs4_get_vfs_file(rqstp, fp, cur_fh, open->op_share_access);
+       if (status) {
+               kmem_cache_free(stateid_slab, stp);
+               return status;
        }
        *stpp = stp;
        return 0;
@@ -2297,35 +2448,28 @@ nfsd4_truncate(struct svc_rqst *rqstp, struct svc_fh *fh,
 }
 
 static __be32
-nfs4_upgrade_open(struct svc_rqst *rqstp, struct svc_fh *cur_fh, struct nfs4_stateid *stp, struct nfsd4_open *open)
+nfs4_upgrade_open(struct svc_rqst *rqstp, struct nfs4_file *fp, struct svc_fh *cur_fh, struct nfs4_stateid *stp, struct nfsd4_open *open)
 {
-       struct file *filp = stp->st_vfs_file;
-       struct inode *inode = filp->f_path.dentry->d_inode;
-       unsigned int share_access, new_writer;
+       u32 op_share_access = open->op_share_access & ~NFS4_SHARE_WANT_MASK;
+       bool new_access;
        __be32 status;
 
-       set_access(&share_access, stp->st_access_bmap);
-       new_writer = (~share_access) & open->op_share_access
-                       & NFS4_SHARE_ACCESS_WRITE;
-
-       if (new_writer) {
-               int err = get_write_access(inode);
-               if (err)
-                       return nfserrno(err);
-               err = mnt_want_write(cur_fh->fh_export->ex_path.mnt);
-               if (err)
-                       return nfserrno(err);
-               file_take_write(filp);
+       new_access = !test_bit(op_share_access, &stp->st_access_bmap);
+       if (new_access) {
+               status = nfs4_get_vfs_file(rqstp, fp, cur_fh, op_share_access);
+               if (status)
+                       return status;
        }
        status = nfsd4_truncate(rqstp, cur_fh, open);
        if (status) {
-               if (new_writer)
-                       put_write_access(inode);
+               if (new_access) {
+                       int oflag = nfs4_access_to_omode(new_access);
+                       nfs4_file_put_access(fp, oflag);
+               }
                return status;
        }
        /* remember the open */
-       filp->f_mode |= open->op_share_access;
-       __set_bit(open->op_share_access, &stp->st_access_bmap);
+       __set_bit(op_share_access, &stp->st_access_bmap);
        __set_bit(open->op_share_deny, &stp->st_deny_bmap);
 
        return nfs_ok;
@@ -2347,7 +2491,7 @@ nfs4_open_delegation(struct svc_fh *fh, struct nfsd4_open *open, struct nfs4_sta
 {
        struct nfs4_delegation *dp;
        struct nfs4_stateowner *sop = stp->st_stateowner;
-       struct nfs4_cb_conn *cb = &sop->so_client->cl_cb_conn;
+       int cb_up = atomic_read(&sop->so_client->cl_cb_set);
        struct file_lock fl, *flp = &fl;
        int status, flag = 0;
 
@@ -2355,7 +2499,7 @@ nfs4_open_delegation(struct svc_fh *fh, struct nfsd4_open *open, struct nfs4_sta
        open->op_recall = 0;
        switch (open->op_claim_type) {
                case NFS4_OPEN_CLAIM_PREVIOUS:
-                       if (!atomic_read(&cb->cb_set))
+                       if (!cb_up)
                                open->op_recall = 1;
                        flag = open->op_delegate_type;
                        if (flag == NFS4_OPEN_DELEGATE_NONE)
@@ -2366,7 +2510,7 @@ nfs4_open_delegation(struct svc_fh *fh, struct nfsd4_open *open, struct nfs4_sta
                         * had the chance to reclaim theirs.... */
                        if (locks_in_grace())
                                goto out;
-                       if (!atomic_read(&cb->cb_set) || !sop->so_confirmed)
+                       if (!cb_up || !sop->so_confirmed)
                                goto out;
                        if (open->op_share_access & NFS4_SHARE_ACCESS_WRITE)
                                flag = NFS4_OPEN_DELEGATE_WRITE;
@@ -2388,13 +2532,14 @@ nfs4_open_delegation(struct svc_fh *fh, struct nfsd4_open *open, struct nfs4_sta
        fl.fl_type = flag == NFS4_OPEN_DELEGATE_READ? F_RDLCK: F_WRLCK;
        fl.fl_end = OFFSET_MAX;
        fl.fl_owner =  (fl_owner_t)dp;
-       fl.fl_file = stp->st_vfs_file;
+       fl.fl_file = find_readable_file(stp->st_file);
+       BUG_ON(!fl.fl_file);
        fl.fl_pid = current->tgid;
 
        /* vfs_setlease checks to see if delegation should be handed out.
         * the lock_manager callbacks fl_mylease and fl_change are used
         */
-       if ((status = vfs_setlease(stp->st_vfs_file, fl.fl_type, &flp))) {
+       if ((status = vfs_setlease(fl.fl_file, fl.fl_type, &flp))) {
                dprintk("NFSD: setlease failed [%d], no delegation\n", status);
                unhash_delegation(dp);
                flag = NFS4_OPEN_DELEGATE_NONE;
@@ -2458,18 +2603,12 @@ nfsd4_process_open2(struct svc_rqst *rqstp, struct svc_fh *current_fh, struct nf
         */
        if (stp) {
                /* Stateid was found, this is an OPEN upgrade */
-               status = nfs4_upgrade_open(rqstp, current_fh, stp, open);
+               status = nfs4_upgrade_open(rqstp, fp, current_fh, stp, open);
                if (status)
                        goto out;
                update_stateid(&stp->st_stateid);
        } else {
-               /* Stateid was not found, this is a new OPEN */
-               int flags = 0;
-               if (open->op_share_access & NFS4_SHARE_ACCESS_READ)
-                       flags |= NFSD_MAY_READ;
-               if (open->op_share_access & NFS4_SHARE_ACCESS_WRITE)
-                       flags |= NFSD_MAY_WRITE;
-               status = nfs4_new_open(rqstp, &stp, dp, current_fh, flags);
+               status = nfs4_new_open(rqstp, &stp, fp, current_fh, open);
                if (status)
                        goto out;
                init_stateid(stp, fp, open);
@@ -2483,10 +2622,8 @@ nfsd4_process_open2(struct svc_rqst *rqstp, struct svc_fh *current_fh, struct nf
        }
        memcpy(&open->op_stateid, &stp->st_stateid, sizeof(stateid_t));
 
-       if (nfsd4_has_session(&resp->cstate)) {
+       if (nfsd4_has_session(&resp->cstate))
                open->op_stateowner->so_confirmed = 1;
-               nfsd4_create_clid_dir(open->op_stateowner->so_client);
-       }
 
        /*
        * Attempt to hand out a delegation. No error return, because the
@@ -2537,7 +2674,7 @@ nfsd4_renew(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
        renew_client(clp);
        status = nfserr_cb_path_down;
        if (!list_empty(&clp->cl_delegations)
-                       && !atomic_read(&clp->cl_cb_conn.cb_set))
+                       && !atomic_read(&clp->cl_cb_set))
                goto out;
        status = nfs_ok;
 out:
@@ -2554,6 +2691,12 @@ nfsd4_end_grace(void)
        dprintk("NFSD: end of grace period\n");
        nfsd4_recdir_purge_old();
        locks_end_grace(&nfsd4_manager);
+       /*
+        * Now that every NFSv4 client has had the chance to recover and
+        * to see the (possibly new, possibly shorter) lease time, we
+        * can safely set the next grace time to the current lease time:
+        */
+       nfsd4_grace = nfsd4_lease;
 }
 
 static time_t
@@ -2563,15 +2706,17 @@ nfs4_laundromat(void)
        struct nfs4_stateowner *sop;
        struct nfs4_delegation *dp;
        struct list_head *pos, *next, reaplist;
-       time_t cutoff = get_seconds() - NFSD_LEASE_TIME;
-       time_t t, clientid_val = NFSD_LEASE_TIME;
-       time_t u, test_val = NFSD_LEASE_TIME;
+       time_t cutoff = get_seconds() - nfsd4_lease;
+       time_t t, clientid_val = nfsd4_lease;
+       time_t u, test_val = nfsd4_lease;
 
        nfs4_lock_state();
 
        dprintk("NFSD: laundromat service - starting\n");
        if (locks_in_grace())
                nfsd4_end_grace();
+       INIT_LIST_HEAD(&reaplist);
+       spin_lock(&client_lock);
        list_for_each_safe(pos, next, &client_lru) {
                clp = list_entry(pos, struct nfs4_client, cl_lru);
                if (time_after((unsigned long)clp->cl_time, (unsigned long)cutoff)) {
@@ -2580,12 +2725,22 @@ nfs4_laundromat(void)
                                clientid_val = t;
                        break;
                }
+               if (atomic_read(&clp->cl_refcount)) {
+                       dprintk("NFSD: client in use (clientid %08x)\n",
+                               clp->cl_clientid.cl_id);
+                       continue;
+               }
+               unhash_client_locked(clp);
+               list_add(&clp->cl_lru, &reaplist);
+       }
+       spin_unlock(&client_lock);
+       list_for_each_safe(pos, next, &reaplist) {
+               clp = list_entry(pos, struct nfs4_client, cl_lru);
                dprintk("NFSD: purging unused client (clientid %08x)\n",
                        clp->cl_clientid.cl_id);
                nfsd4_remove_clid_dir(clp);
                expire_client(clp);
        }
-       INIT_LIST_HEAD(&reaplist);
        spin_lock(&recall_lock);
        list_for_each_safe(pos, next, &del_recall_lru) {
                dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
@@ -2605,7 +2760,7 @@ nfs4_laundromat(void)
                list_del_init(&dp->dl_recall_lru);
                unhash_delegation(dp);
        }
-       test_val = NFSD_LEASE_TIME;
+       test_val = nfsd4_lease;
        list_for_each_safe(pos, next, &close_lru) {
                sop = list_entry(pos, struct nfs4_stateowner, so_close_lru);
                if (time_after((unsigned long)sop->so_time, (unsigned long)cutoff)) {
@@ -2655,45 +2810,17 @@ search_close_lru(u32 st_id, int flags)
 static inline int
 nfs4_check_fh(struct svc_fh *fhp, struct nfs4_stateid *stp)
 {
-       return fhp->fh_dentry->d_inode != stp->st_vfs_file->f_path.dentry->d_inode;
+       return fhp->fh_dentry->d_inode != stp->st_file->fi_inode;
 }
 
 static int
 STALE_STATEID(stateid_t *stateid)
 {
-       if (time_after((unsigned long)boot_time,
-                       (unsigned long)stateid->si_boot)) {
-               dprintk("NFSD: stale stateid " STATEID_FMT "!\n",
-                       STATEID_VAL(stateid));
-               return 1;
-       }
-       return 0;
-}
-
-static int
-EXPIRED_STATEID(stateid_t *stateid)
-{
-       if (time_before((unsigned long)boot_time,
-                       ((unsigned long)stateid->si_boot)) &&
-           time_before((unsigned long)(stateid->si_boot + lease_time), get_seconds())) {
-               dprintk("NFSD: expired stateid " STATEID_FMT "!\n",
-                       STATEID_VAL(stateid));
-               return 1;
-       }
-       return 0;
-}
-
-static __be32
-stateid_error_map(stateid_t *stateid)
-{
-       if (STALE_STATEID(stateid))
-               return nfserr_stale_stateid;
-       if (EXPIRED_STATEID(stateid))
-               return nfserr_expired;
-
-       dprintk("NFSD: bad stateid " STATEID_FMT "!\n",
+       if (stateid->si_boot == boot_time)
+               return 0;
+       dprintk("NFSD: stale stateid " STATEID_FMT "!\n",
                STATEID_VAL(stateid));
-       return nfserr_bad_stateid;
+       return 1;
 }
 
 static inline int
@@ -2716,6 +2843,9 @@ __be32 nfs4_check_openmode(struct nfs4_stateid *stp, int flags)
 {
         __be32 status = nfserr_openmode;
 
+       /* For lock stateid's, we test the parent open, not the lock: */
+       if (stp->st_openstp)
+               stp = stp->st_openstp;
        if ((flags & WR_STATE) && (!access_permit_write(stp->st_access_bmap)))
                 goto out;
        if ((flags & RD_STATE) && (!access_permit_read(stp->st_access_bmap)))
@@ -2817,10 +2947,8 @@ nfs4_preprocess_stateid_op(struct nfsd4_compound_state *cstate,
        status = nfserr_bad_stateid;
        if (is_delegation_stateid(stateid)) {
                dp = find_delegation_stateid(ino, stateid);
-               if (!dp) {
-                       status = stateid_error_map(stateid);
+               if (!dp)
                        goto out;
-               }
                status = check_stateid_generation(stateid, &dp->dl_stateid,
                                                  flags);
                if (status)
@@ -2830,13 +2958,12 @@ nfs4_preprocess_stateid_op(struct nfsd4_compound_state *cstate,
                        goto out;
                renew_client(dp->dl_client);
                if (filpp)
-                       *filpp = dp->dl_vfs_file;
+                       *filpp = find_readable_file(dp->dl_file);
+               BUG_ON(!*filpp);
        } else { /* open or lock stateid */
                stp = find_stateid(stateid, flags);
-               if (!stp) {
-                       status = stateid_error_map(stateid);
+               if (!stp)
                        goto out;
-               }
                if (nfs4_check_fh(current_fh, stp))
                        goto out;
                if (!stp->st_stateowner->so_confirmed)
@@ -2849,8 +2976,12 @@ nfs4_preprocess_stateid_op(struct nfsd4_compound_state *cstate,
                if (status)
                        goto out;
                renew_client(stp->st_stateowner->so_client);
-               if (filpp)
-                       *filpp = stp->st_vfs_file;
+               if (filpp) {
+                       if (flags & RD_STATE)
+                               *filpp = find_readable_file(stp->st_file);
+                       else
+                               *filpp = find_writeable_file(stp->st_file);
+               }
        }
        status = nfs_ok;
 out:
@@ -2908,7 +3039,7 @@ nfs4_preprocess_seqid_op(struct nfsd4_compound_state *cstate, u32 seqid,
                 */
                sop = search_close_lru(stateid->si_stateownerid, flags);
                if (sop == NULL)
-                       return stateid_error_map(stateid);
+                       return nfserr_bad_stateid;
                *sopp = sop;
                goto check_replay;
        }
@@ -3086,8 +3217,7 @@ nfsd4_open_downgrade(struct svc_rqst *rqstp,
                goto out;
        }
        set_access(&share_access, stp->st_access_bmap);
-       nfs4_file_downgrade(stp->st_vfs_file,
-                           share_access & ~od->od_share_access);
+       nfs4_file_downgrade(stp->st_file, share_access & ~od->od_share_access);
 
        reset_union_bmap_access(od->od_share_access, &stp->st_access_bmap);
        reset_union_bmap_deny(od->od_share_deny, &stp->st_deny_bmap);
@@ -3175,10 +3305,8 @@ nfsd4_delegreturn(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
        if (!is_delegation_stateid(stateid))
                goto out;
        dp = find_delegation_stateid(inode, stateid);
-       if (!dp) {
-               status = stateid_error_map(stateid);
+       if (!dp)
                goto out;
-       }
        status = check_stateid_generation(stateid, &dp->dl_stateid, flags);
        if (status)
                goto out;
@@ -3308,11 +3436,9 @@ static inline void
 nfs4_set_lock_denied(struct file_lock *fl, struct nfsd4_lock_denied *deny)
 {
        struct nfs4_stateowner *sop;
-       unsigned int hval;
 
        if (fl->fl_lmops == &nfsd_posix_mng_ops) {
                sop = (struct nfs4_stateowner *) fl->fl_owner;
-               hval = lockownerid_hashval(sop->so_id);
                kref_get(&sop->so_ref);
                deny->ld_sop = sop;
                deny->ld_clientid = sop->so_client->cl_clientid;
@@ -3404,12 +3530,10 @@ alloc_init_lock_stateid(struct nfs4_stateowner *sop, struct nfs4_file *fp, struc
        stp->st_stateowner = sop;
        get_nfs4_file(fp);
        stp->st_file = fp;
-       stp->st_stateid.si_boot = get_seconds();
+       stp->st_stateid.si_boot = boot_time;
        stp->st_stateid.si_stateownerid = sop->so_id;
        stp->st_stateid.si_fileid = fp->fi_id;
        stp->st_stateid.si_generation = 0;
-       stp->st_vfs_file = open_stp->st_vfs_file; /* FIXME refcount?? */
-       stp->st_access_bmap = open_stp->st_access_bmap;
        stp->st_deny_bmap = open_stp->st_deny_bmap;
        stp->st_openstp = open_stp;
 
@@ -3434,7 +3558,8 @@ nfsd4_lock(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
        struct nfs4_stateowner *open_sop = NULL;
        struct nfs4_stateowner *lock_sop = NULL;
        struct nfs4_stateid *lock_stp;
-       struct file *filp;
+       struct nfs4_file *fp;
+       struct file *filp = NULL;
        struct file_lock file_lock;
        struct file_lock conflock;
        __be32 status = 0;
@@ -3464,7 +3589,6 @@ nfsd4_lock(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
                 * lock stateid.
                 */
                struct nfs4_stateid *open_stp = NULL;
-               struct nfs4_file *fp;
                
                status = nfserr_stale_clientid;
                if (!nfsd4_has_session(cstate) &&
@@ -3507,9 +3631,9 @@ nfsd4_lock(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
                if (status)
                        goto out;
                lock_sop = lock->lk_replay_owner;
+               fp = lock_stp->st_file;
        }
        /* lock->lk_replay_owner and lock_stp have been created or found */
-       filp = lock_stp->st_vfs_file;
 
        status = nfserr_grace;
        if (locks_in_grace() && !lock->lk_reclaim)
@@ -3522,11 +3646,19 @@ nfsd4_lock(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
        switch (lock->lk_type) {
                case NFS4_READ_LT:
                case NFS4_READW_LT:
+                       if (find_readable_file(lock_stp->st_file)) {
+                               nfs4_get_vfs_file(rqstp, fp, &cstate->current_fh, NFS4_SHARE_ACCESS_READ);
+                               filp = find_readable_file(lock_stp->st_file);
+                       }
                        file_lock.fl_type = F_RDLCK;
                        cmd = F_SETLK;
                break;
                case NFS4_WRITE_LT:
                case NFS4_WRITEW_LT:
+                       if (find_writeable_file(lock_stp->st_file)) {
+                               nfs4_get_vfs_file(rqstp, fp, &cstate->current_fh, NFS4_SHARE_ACCESS_WRITE);
+                               filp = find_writeable_file(lock_stp->st_file);
+                       }
                        file_lock.fl_type = F_WRLCK;
                        cmd = F_SETLK;
                break;
@@ -3534,6 +3666,10 @@ nfsd4_lock(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
                        status = nfserr_inval;
                goto out;
        }
+       if (!filp) {
+               status = nfserr_openmode;
+               goto out;
+       }
        file_lock.fl_owner = (fl_owner_t)lock_sop;
        file_lock.fl_pid = current->tgid;
        file_lock.fl_file = filp;
@@ -3702,7 +3838,11 @@ nfsd4_locku(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
                                        &locku->lu_stateowner, &stp, NULL)))
                goto out;
 
-       filp = stp->st_vfs_file;
+       filp = find_any_file(stp->st_file);
+       if (!filp) {
+               status = nfserr_lock_range;
+               goto out;
+       }
        BUG_ON(!filp);
        locks_init_lock(&file_lock);
        file_lock.fl_type = F_UNLCK;
@@ -3749,10 +3889,10 @@ out_nfserr:
  *     0: no locks held by lockowner
  */
 static int
-check_for_locks(struct file *filp, struct nfs4_stateowner *lowner)
+check_for_locks(struct nfs4_file *filp, struct nfs4_stateowner *lowner)
 {
        struct file_lock **flpp;
-       struct inode *inode = filp->f_path.dentry->d_inode;
+       struct inode *inode = filp->fi_inode;
        int status = 0;
 
        lock_kernel();
@@ -3803,7 +3943,7 @@ nfsd4_release_lockowner(struct svc_rqst *rqstp,
                                continue;
                        list_for_each_entry(stp, &sop->so_stateids,
                                        st_perstateowner) {
-                               if (check_for_locks(stp->st_vfs_file, sop))
+                               if (check_for_locks(stp->st_file, sop))
                                        goto out;
                                /* Note: so_perclient unused for lockowners,
                                 * so it's OK to fool with here. */
@@ -3976,12 +4116,6 @@ nfsd4_load_reboot_recovery_data(void)
                printk("NFSD: Failure reading reboot recovery data\n");
 }
 
-unsigned long
-get_nfs4_grace_period(void)
-{
-       return max(user_lease_time, lease_time) * HZ;
-}
-
 /*
  * Since the lifetime of a delegation isn't limited to that of an open, a
  * client may quite reasonably hang on to a delegation as long as it has
@@ -4008,41 +4142,34 @@ set_max_delegations(void)
 static int
 __nfs4_state_start(void)
 {
-       unsigned long grace_time;
+       int ret;
 
        boot_time = get_seconds();
-       grace_time = get_nfs4_grace_period();
-       lease_time = user_lease_time;
        locks_start_grace(&nfsd4_manager);
        printk(KERN_INFO "NFSD: starting %ld-second grace period\n",
-              grace_time/HZ);
+              nfsd4_grace);
+       ret = set_callback_cred();
+       if (ret)
+               return -ENOMEM;
        laundry_wq = create_singlethread_workqueue("nfsd4");
        if (laundry_wq == NULL)
                return -ENOMEM;
-       queue_delayed_work(laundry_wq, &laundromat_work, grace_time);
+       ret = nfsd4_create_callback_queue();
+       if (ret)
+               goto out_free_laundry;
+       queue_delayed_work(laundry_wq, &laundromat_work, nfsd4_grace * HZ);
        set_max_delegations();
-       return set_callback_cred();
+       return 0;
+out_free_laundry:
+       destroy_workqueue(laundry_wq);
+       return ret;
 }
 
 int
 nfs4_state_start(void)
 {
-       int ret;
-
-       if (nfs4_init)
-               return 0;
        nfsd4_load_reboot_recovery_data();
-       ret = __nfs4_state_start();
-       if (ret)
-               return ret;
-       nfs4_init = 1;
-       return 0;
-}
-
-time_t
-nfs4_lease_time(void)
-{
-       return lease_time;
+       return __nfs4_state_start();
 }
 
 static void
@@ -4077,7 +4204,6 @@ __nfs4_state_shutdown(void)
        }
 
        nfsd4_shutdown_recdir();
-       nfs4_init = 0;
 }
 
 void
@@ -4089,6 +4215,7 @@ nfs4_state_shutdown(void)
        nfs4_lock_state();
        nfs4_release_reclaim();
        __nfs4_state_shutdown();
+       nfsd4_destroy_callback_queue();
        nfs4_unlock_state();
 }
 
@@ -4128,21 +4255,3 @@ nfs4_recoverydir(void)
 {
        return user_recovery_dirname;
 }
-
-/*
- * Called when leasetime is changed.
- *
- * The only way the protocol gives us to handle on-the-fly lease changes is to
- * simulate a reboot.  Instead of doing that, we just wait till the next time
- * we start to register any changes in lease time.  If the administrator
- * really wants to change the lease time *now*, they can go ahead and bring
- * nfsd down and then back up again after changing the lease time.
- *
- * user_lease_time is protected by nfsd_mutex since it's only really accessed
- * when nfsd is starting
- */
-void
-nfs4_reset_lease(time_t leasetime)
-{
-       user_lease_time = leasetime;
-}