Whamcloud - gitweb
Add lu_ref tracking to obd_device.
[fs/lustre-release.git] / lustre / mdt / mdt_handler.c
index 2a8e4f5..74b2f11 100644 (file)
@@ -1,35 +1,49 @@
 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
  * vim:expandtab:shiftwidth=8:tabstop=8:
  *
- *  lustre/mdt/mdt_handler.c
- *  Lustre Metadata Target (mdt) request handler
+ * GPL HEADER START
  *
- *  Copyright (c) 2006 Cluster File Systems, Inc.
- *   Author: Peter Braam <braam@clusterfs.com>
- *   Author: Andreas Dilger <adilger@clusterfs.com>
- *   Author: Phil Schwan <phil@clusterfs.com>
- *   Author: Mike Shaver <shaver@clusterfs.com>
- *   Author: Nikita Danilov <nikita@clusterfs.com>
- *   Author: Huang Hua <huanghua@clusterfs.com>
- *   Author: Yury Umanets <umka@clusterfs.com>
+ * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
  *
- *   This file is part of the Lustre file system, http://www.lustre.org
- *   Lustre is a trademark of Cluster File Systems, Inc.
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License version 2 only,
+ * as published by the Free Software Foundation.
  *
- *   You may have signed or agreed to another license before downloading
- *   this software.  If so, you are bound by the terms and conditions
- *   of that agreement, and the following does not apply to you.  See the
- *   LICENSE file included with this distribution for more information.
+ * This program is distributed in the hope that it will be useful, but
+ * WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * General Public License version 2 for more details (a copy is included
+ * in the LICENSE file that accompanied this code).
  *
- *   If you did not agree to a different license, then this copy of Lustre
- *   is open source software; you can redistribute it and/or modify it
- *   under the terms of version 2 of the GNU General Public License as
- *   published by the Free Software Foundation.
+ * You should have received a copy of the GNU General Public License
+ * version 2 along with this program; If not, see
+ * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
  *
- *   In either case, Lustre is distributed in the hope that it will be
- *   useful, but WITHOUT ANY WARRANTY; without even the implied warranty
- *   of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
- *   license text for more details.
+ * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
+ * CA 95054 USA or visit www.sun.com if you need additional information or
+ * have any questions.
+ *
+ * GPL HEADER END
+ */
+/*
+ * Copyright  2008 Sun Microsystems, Inc. All rights reserved
+ * Use is subject to license terms.
+ */
+/*
+ * This file is part of Lustre, http://www.lustre.org/
+ * Lustre is a trademark of Sun Microsystems, Inc.
+ *
+ * lustre/mdt/mdt_handler.c
+ *
+ * Lustre Metadata Target (mdt) request handler
+ *
+ * Author: Peter Braam <braam@clusterfs.com>
+ * Author: Andreas Dilger <adilger@clusterfs.com>
+ * Author: Phil Schwan <phil@clusterfs.com>
+ * Author: Mike Shaver <shaver@clusterfs.com>
+ * Author: Nikita Danilov <nikita@clusterfs.com>
+ * Author: Huang Hua <huanghua@clusterfs.com>
+ * Author: Yury Umanets <umka@clusterfs.com>
  */
 
 #ifndef EXPORT_SYMTAB
@@ -53,7 +67,7 @@
 #include <lustre_mds.h>
 #include <lustre_mdt.h>
 #include "mdt_internal.h"
-#include <linux/lustre_acl.h>
+#include <lustre_acl.h>
 #include <lustre_param.h>
 
 mdl_mode_t mdt_mdl_lock_modes[] = {
@@ -148,7 +162,7 @@ static struct mdt_device *mdt_dev(struct lu_device *d);
 static int mdt_regular_handle(struct ptlrpc_request *req);
 static int mdt_unpack_req_pack_rep(struct mdt_thread_info *info, __u32 flags);
 
-static struct lu_object_operations mdt_obj_ops;
+static const struct lu_object_operations mdt_obj_ops;
 
 int mdt_get_disposition(struct ldlm_reply *rep, int flag)
 {
@@ -763,7 +777,7 @@ static int mdt_getattr_name_lock(struct mdt_thread_info *info,
         struct md_object       *next      = mdt_object_child(parent);
         struct lu_fid          *child_fid = &info->mti_tmp_fid1;
         struct lu_name         *lname     = NULL;
-        const char             *name;
+        const char             *name      = NULL;
         int                     namelen   = 0;
         struct mdt_lock_handle *lhp;
         struct ldlm_lock       *lock;
@@ -784,27 +798,30 @@ static int mdt_getattr_name_lock(struct mdt_thread_info *info,
 
         namelen = req_capsule_get_size(info->mti_pill, &RMF_NAME,
                                        RCL_CLIENT) - 1;
-        LASSERT(namelen >= 0);
-
-        /* XXX: "namelen == 0" is for getattr by fid (OBD_CONNECT_ATTRFID),
-         * otherwise do not allow empty name, that is the name must contain
-         * at least one character and the terminating '\0'*/
-        if (namelen == 0) {
-                reqbody =req_capsule_client_get(info->mti_pill, &RMF_MDT_BODY);
-                LASSERT(fid_is_sane(&reqbody->fid2));
-                name = NULL;
-
-                CDEBUG(D_INODE, "getattr with lock for "DFID"/"DFID", "
-                       "ldlm_rep = %p\n",
-                       PFID(mdt_object_fid(parent)), PFID(&reqbody->fid2),
-                       ldlm_rep);
-        } else {
-                lname = mdt_name(info->mti_env, (char *)name, namelen);
-                CDEBUG(D_INODE, "getattr with lock for "DFID"/%s, "
-                       "ldlm_rep = %p\n",
-                       PFID(mdt_object_fid(parent)), name, ldlm_rep);
-        }
+        if (!info->mti_cross_ref) {
+                /* 
+                 * XXX: Check for "namelen == 0" is for getattr by fid 
+                 * (OBD_CONNECT_ATTRFID), otherwise do not allow empty name,
+                 * that is the name must contain at least one character and
+                 * the terminating '\0'
+                 */
+                if (namelen == 0) {
+                        reqbody = req_capsule_client_get(info->mti_pill, 
+                                                         &RMF_MDT_BODY);
+                        LASSERT(fid_is_sane(&reqbody->fid2));
+                        name = NULL;
 
+                        CDEBUG(D_INODE, "getattr with lock for "DFID"/"DFID", "
+                               "ldlm_rep = %p\n",
+                               PFID(mdt_object_fid(parent)), PFID(&reqbody->fid2),
+                               ldlm_rep);
+                } else {
+                        lname = mdt_name(info->mti_env, (char *)name, namelen);
+                        CDEBUG(D_INODE, "getattr with lock for "DFID"/%s, "
+                               "ldlm_rep = %p\n", PFID(mdt_object_fid(parent)), 
+                               name, ldlm_rep);
+                }
+        }
         mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_EXECD);
 
         rc = mdt_object_exists(parent);
@@ -813,10 +830,10 @@ static int mdt_getattr_name_lock(struct mdt_thread_info *info,
                                 &parent->mot_obj.mo_lu,
                                 "Parent doesn't exist!\n");
                 RETURN(-ESTALE);
-        } else
+        } else if (!info->mti_cross_ref) {
                 LASSERTF(rc > 0, "Parent "DFID" is on remote server\n",
                          PFID(mdt_object_fid(parent)));
-
+        }
         if (lname) {
                 rc = mdt_raw_lookup(info, parent, lname, ldlm_rep);
                 if (rc != 0) {
@@ -938,6 +955,7 @@ relock:
                         LU_OBJECT_DEBUG(D_WARNING, info->mti_env,
                                         &child->mot_obj.mo_lu,
                                         "Object doesn't exist!\n");
+                        GOTO(out_child, rc = -ESTALE);
                 }
 
                 ma->ma_valid = 0;
@@ -1038,7 +1056,7 @@ out_shrink:
         return rc;
 }
 
-static struct lu_device_operations mdt_lu_ops;
+static const struct lu_device_operations mdt_lu_ops;
 
 static int lu_device_is_mdt(struct lu_device *d)
 {
@@ -1229,7 +1247,7 @@ static int mdt_write_dir_page(struct mdt_thread_info *info, struct page *page,
                         continue;
 
                 fid_le_to_cpu(lf, &ent->lde_fid);
-                if (le32_to_cpu(ent->lde_hash) & MAX_HASH_HIGHEST_BIT)
+                if (le64_to_cpu(ent->lde_hash) & MAX_HASH_HIGHEST_BIT)
                         ma->ma_attr.la_mode = S_IFDIR;
                 else
                         ma->ma_attr.la_mode = 0;
@@ -1239,7 +1257,7 @@ static int mdt_write_dir_page(struct mdt_thread_info *info, struct page *page,
 
                 memcpy(name, ent->lde_name, le16_to_cpu(ent->lde_namelen));
                 lname = mdt_name(info->mti_env, name,
-                                 le16_to_cpu(ent->lde_namelen) + 1);
+                                 le16_to_cpu(ent->lde_namelen));
                 ma->ma_attr_flags |= MDS_PERM_BYPASS;
                 rc = mdo_name_insert(info->mti_env,
                                      md_object_next(&object->mot_obj),
@@ -1377,9 +1395,9 @@ static int mdt_readpage(struct mdt_thread_info *info)
          * reqbody->nlink contains number bytes to read.
          */
         rdpg->rp_hash = reqbody->size;
-        if ((__u64)rdpg->rp_hash != reqbody->size) {
-                CERROR("Invalid hash: %#llx != %#llx\n",
-                       (__u64)rdpg->rp_hash, reqbody->size);
+        if (rdpg->rp_hash != reqbody->size) {
+                CERROR("Invalid hash: "LPX64" != "LPX64"\n",
+                       rdpg->rp_hash, reqbody->size);
                 RETURN(-EFAULT);
         }
         rdpg->rp_count  = reqbody->nlink;
@@ -1530,18 +1548,33 @@ static int mdt_reint(struct mdt_thread_info *info)
         RETURN(rc);
 }
 
-/* TODO these two methods not available now. */
-
 /* this should sync the whole device */
-static int mdt_device_sync(struct mdt_thread_info *info)
+static int mdt_device_sync(const struct lu_env *env, struct mdt_device *mdt)
 {
-        return 0;
+        struct dt_device *dt = mdt->mdt_bottom;
+        int rc;
+        ENTRY;
+
+        rc = dt->dd_ops->dt_sync(env, dt);
+        RETURN(rc);
 }
 
 /* this should sync this object */
 static int mdt_object_sync(struct mdt_thread_info *info)
 {
-        return 0;
+        struct md_object *next;
+        int rc;
+        ENTRY;
+
+        if (!mdt_object_exists(info->mti_object)) {
+                CWARN("Non existing object  "DFID"!\n",
+                      PFID(mdt_object_fid(info->mti_object)));
+                RETURN(-ESTALE);
+        }
+        next = mdt_object_child(info->mti_object);
+        rc = mo_object_sync(info->mti_env, next);
+
+        RETURN(rc);
 }
 
 static int mdt_sync(struct mdt_thread_info *info)
@@ -1565,7 +1598,7 @@ static int mdt_sync(struct mdt_thread_info *info)
                 /* sync the whole device */
                 rc = req_capsule_server_pack(pill);
                 if (rc == 0)
-                        rc = mdt_device_sync(info);
+                        rc = mdt_device_sync(info->mti_env, info->mti_mdt);
                 else
                         rc = err_serious(rc);
         } else {
@@ -1735,7 +1768,7 @@ struct mdt_object *mdt_object_find(const struct lu_env *env,
         ENTRY;
 
         CDEBUG(D_INFO, "Find object for "DFID"\n", PFID(f));
-        o = lu_object_find(env, d->mdt_md_dev.md_lu_dev.ld_site, f);
+        o = lu_object_find(env, &d->mdt_md_dev.md_lu_dev, f, NULL);
         if (unlikely(IS_ERR(o)))
                 m = (struct mdt_object *)o;
         else
@@ -1770,6 +1803,17 @@ int mdt_object_lock(struct mdt_thread_info *info, struct mdt_object *o,
                 LASSERT(lh->mlh_type != MDT_PDO_LOCK);
         }
 
+        if (lh->mlh_type == MDT_PDO_LOCK) {
+                /* check for exists after object is locked */
+                if (mdt_object_exists(o) == 0) {
+                        /* Non-existent object shouldn't have PDO lock */
+                        RETURN(-ESTALE);
+                } else {
+                        /* Non-dir object shouldn't have PDO lock */
+                        LASSERT(S_ISDIR(lu_object_attr(&o->mot_obj.mo_lu)));
+                }
+        }
+
         memset(policy, 0, sizeof(*policy));
         fid_build_reg_res_name(mdt_object_fid(o), res_id);
 
@@ -1794,7 +1838,7 @@ int mdt_object_lock(struct mdt_thread_info *info, struct mdt_object *o,
                 }
 
                 /*
-                 * Finish res_id initializing by name hash marking patr of
+                 * Finish res_id initializing by name hash marking part of
                  * directory which is taking modification.
                  */
                 res_id->name[LUSTRE_RES_ID_HSH_OFF] = lh->mlh_pdo_hash;
@@ -1805,7 +1849,7 @@ int mdt_object_lock(struct mdt_thread_info *info, struct mdt_object *o,
         /*
          * Use LDLM_FL_LOCAL_ONLY for this lock. We do not know yet if it is
          * going to be sent to client. If it is - mdt_intent_policy() path will
-         * fix it up and turns FL_LOCAL flag off.
+         * fix it up and turn FL_LOCAL flag off.
          */
         rc = mdt_fid_lock(ns, &lh->mlh_reg_lh, lh->mlh_reg_mode, policy,
                           res_id, LDLM_FL_LOCAL_ONLY | LDLM_FL_ATOMIC_CB);
@@ -1813,16 +1857,6 @@ int mdt_object_lock(struct mdt_thread_info *info, struct mdt_object *o,
         if (rc)
                 GOTO(out, rc);
 
-        if (lh->mlh_type == MDT_PDO_LOCK) {
-                /* check for exists after object is locked */
-                if (mdt_object_exists(o) == 0) {
-                        /* Non-existent object shouldn't have PDO lock */
-                        rc = -ESTALE;
-                } else {
-                        /* Non-dir object shouldn't have PDO lock */
-                        LASSERT(S_ISDIR(lu_object_attr(&o->mot_obj.mo_lu)));
-                }
-        }
 out:
         if (rc)
                 mdt_object_unlock(info, o, lh, 1);
@@ -1831,6 +1865,23 @@ out:
         RETURN(rc);
 }
 
+static inline
+void mdt_save_lock(struct ptlrpc_request *req, struct lustre_handle *h,
+                   ldlm_mode_t mode, int decref)
+{
+        ENTRY;
+
+        if (lustre_handle_is_used(h)) {
+                if (decref)
+                        mdt_fid_unlock(h, mode);
+                else
+                        ptlrpc_save_lock(req, h, mode);
+                h->cookie = 0ull;
+        }
+
+        EXIT;
+}
+
 /*
  * Just call ldlm_lock_decref() if decref, else we only call ptlrpc_save_lock()
  * to save this lock in req.  when transaction committed, req will be released,
@@ -1842,23 +1893,8 @@ void mdt_object_unlock(struct mdt_thread_info *info, struct mdt_object *o,
         struct ptlrpc_request *req = mdt_info_req(info);
         ENTRY;
 
-        if (lustre_handle_is_used(&lh->mlh_pdo_lh)) {
-                /* Do not save PDO locks to request, just decref. */
-                mdt_fid_unlock(&lh->mlh_pdo_lh,
-                               lh->mlh_pdo_mode);
-                lh->mlh_pdo_lh.cookie = 0ull;
-        }
-
-        if (lustre_handle_is_used(&lh->mlh_reg_lh)) {
-                if (decref) {
-                        mdt_fid_unlock(&lh->mlh_reg_lh,
-                                       lh->mlh_reg_mode);
-                } else {
-                        ptlrpc_save_lock(req, &lh->mlh_reg_lh,
-                                         lh->mlh_reg_mode);
-                }
-                lh->mlh_reg_lh.cookie = 0ull;
-        }
+        mdt_save_lock(req, &lh->mlh_pdo_lh, lh->mlh_pdo_mode, decref);
+        mdt_save_lock(req, &lh->mlh_reg_lh, lh->mlh_reg_mode, decref);
 
         EXIT;
 }
@@ -2049,8 +2085,8 @@ static int mdt_req_handle(struct mdt_thread_info *info,
         LASSERT(current->journal_info == NULL);
 
         /*
-         * Checking for various OBD_FAIL_$PREF_$OPC_NET codes. _Do_ not try 
-         * to put same checks into handlers like mdt_close(), mdt_reint(), 
+         * Checking for various OBD_FAIL_$PREF_$OPC_NET codes. _Do_ not try
+         * to put same checks into handlers like mdt_close(), mdt_reint(),
          * etc., without talking to mdt authors first. Checking same thing
          * there again is useless and returning 0 error wihtout packing reply
          * is buggy! Handlers either pack reply or return error.
@@ -2644,7 +2680,7 @@ int mdt_intent_lock_replace(struct mdt_thread_info *info,
          * lock.
          */
         if (new_lock == NULL)
-                new_lock = ldlm_handle2lock(&lh->mlh_reg_lh);
+                new_lock = ldlm_handle2lock_long(&lh->mlh_reg_lh, 0);
 
         if (new_lock == NULL && (flags & LDLM_FL_INTENT_ONLY)) {
                 lh->mlh_reg_lh.cookie = 0;
@@ -2680,29 +2716,35 @@ int mdt_intent_lock_replace(struct mdt_thread_info *info,
                 RETURN(ELDLM_LOCK_REPLACED);
         }
 
-        /* This lock might already be given to the client by an resent req,
-         * in this case we should return ELDLM_LOCK_ABORTED,
-         * so we should check led_held_locks here, but it will affect
-         * performance, FIXME
+        /*
+         * Fixup the lock to be given to the client.
          */
-        /* Fixup the lock to be given to the client */
         lock_res_and_lock(new_lock);
-        new_lock->l_readers = 0;
-        new_lock->l_writers = 0;
+        /* Zero new_lock->l_readers and new_lock->l_writers without triggering
+         * possible blocking AST. */
+        while (new_lock->l_readers > 0) {
+                lu_ref_del(&new_lock->l_reference, "reader", new_lock);
+                lu_ref_del(&new_lock->l_reference, "user", new_lock);
+                new_lock->l_readers--;
+        }
+        while (new_lock->l_writers > 0) {
+                lu_ref_del(&new_lock->l_reference, "writer", new_lock);
+                lu_ref_del(&new_lock->l_reference, "user", new_lock);
+                new_lock->l_writers--;
+        }
 
         new_lock->l_export = class_export_get(req->rq_export);
-        spin_lock(&req->rq_export->exp_ldlm_data.led_lock);
-        list_add(&new_lock->l_export_chain,
-                 &new_lock->l_export->exp_ldlm_data.led_held_locks);
-        spin_unlock(&req->rq_export->exp_ldlm_data.led_lock);
-
         new_lock->l_blocking_ast = lock->l_blocking_ast;
         new_lock->l_completion_ast = lock->l_completion_ast;
         new_lock->l_remote_handle = lock->l_remote_handle;
         new_lock->l_flags &= ~LDLM_FL_LOCAL;
 
+        lustre_hash_add(new_lock->l_export->exp_lock_hash,
+                        &new_lock->l_remote_handle,
+                        &new_lock->l_exp_hash);
+
         unlock_res_and_lock(new_lock);
-        LDLM_LOCK_PUT(new_lock);
+        LDLM_LOCK_RELEASE(new_lock);
         lh->mlh_reg_lh.cookie = 0;
 
         RETURN(ELDLM_LOCK_REPLACED);
@@ -2717,7 +2759,7 @@ static void mdt_intent_fixup_resent(struct mdt_thread_info *info,
         struct obd_export      *exp = req->rq_export;
         struct lustre_handle    remote_hdl;
         struct ldlm_request    *dlmreq;
-        struct list_head       *iter;
+        struct ldlm_lock       *lock;
 
         if (!(lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT))
                 return;
@@ -2725,27 +2767,24 @@ static void mdt_intent_fixup_resent(struct mdt_thread_info *info,
         dlmreq = req_capsule_client_get(info->mti_pill, &RMF_DLM_REQ);
         remote_hdl = dlmreq->lock_handle[0];
 
-        spin_lock(&exp->exp_ldlm_data.led_lock);
-        list_for_each(iter, &exp->exp_ldlm_data.led_held_locks) {
-                struct ldlm_lock *lock;
-                lock = list_entry(iter, struct ldlm_lock, l_export_chain);
-                if (lock == new_lock)
-                        continue;
-                if (lock->l_remote_handle.cookie == remote_hdl.cookie) {
+        lock = lustre_hash_lookup(exp->exp_lock_hash, &remote_hdl);
+        if (lock) {
+                if (lock != new_lock) {
                         lh->mlh_reg_lh.cookie = lock->l_handle.h_cookie;
                         lh->mlh_reg_mode = lock->l_granted_mode;
 
-                        LDLM_DEBUG(lock, "restoring lock cookie");
+                        LDLM_DEBUG(lock, "Restoring lock cookie");
                         DEBUG_REQ(D_DLMTRACE, req,
                                   "restoring lock cookie "LPX64,
                                   lh->mlh_reg_lh.cookie);
                         if (old_lock)
                                 *old_lock = LDLM_LOCK_GET(lock);
-                        spin_unlock(&exp->exp_ldlm_data.led_lock);
+                        lh_put(exp->exp_lock_hash, &lock->l_exp_hash);
                         return;
                 }
+
+                lh_put(exp->exp_lock_hash, &lock->l_exp_hash);
         }
-        spin_unlock(&exp->exp_ldlm_data.led_lock);
 
         /*
          * If the xid matches, then we know this is a resent request, and allow
@@ -2886,8 +2925,25 @@ static int mdt_intent_reint(enum mdt_it_code opcode,
         rep->lock_policy_res2 = clear_serious(rc);
 
         lhc->mlh_reg_lh.cookie = 0ull;
-        rc = ELDLM_LOCK_ABORTED;
-        RETURN(rc);
+        if (rc == -ENOTCONN || rc == -ENODEV) {
+                /*
+                 * If it is the disconnect error (ENODEV & ENOCONN), the error
+                 * will be returned by rq_status, and client at ptlrpc layer
+                 * will detect this, then disconnect, reconnect the import
+                 * immediately, instead of impacting the following the rpc.
+                 */
+                RETURN(rc);
+        } else {
+                /*
+                 * For other cases, the error will be returned by intent.
+                 * and client will retrieve the result from intent.
+                 */
+                 /*
+                  * FIXME: when open lock is finished, that should be
+                  * checked here.
+                  */
+                RETURN(ELDLM_LOCK_ABORTED);
+        }
 }
 
 static int mdt_intent_code(long itcode)
@@ -2991,12 +3047,6 @@ static int mdt_intent_policy(struct ldlm_namespace *ns,
                 if (it != NULL) {
                         const struct ldlm_request *dlmreq;
                         __u64 req_bits;
-#if 0
-                        struct ldlm_lock       *lock = *lockp;
-
-                        LDLM_DEBUG(lock, "intent policy opc: %s\n",
-                                   ldlm_it2str(it->opc));
-#endif
 
                         rc = mdt_intent_opc(it->opc, info, lockp, flags);
                         if (rc == 0)
@@ -3030,15 +3080,15 @@ static int mdt_intent_policy(struct ldlm_namespace *ns,
 static void mdt_seq_adjust(const struct lu_env *env,
                           struct mdt_device *m, int lost)
 {
-        struct lu_site *ls = m->mdt_md_dev.md_lu_dev.ld_site;
+        struct md_site *ms = mdt_md_site(m);
         struct lu_range out;
         ENTRY;
 
-        LASSERT(ls && ls->ls_server_seq);
+        LASSERT(ms && ms->ms_server_seq);
         LASSERT(lost >= 0);
         /* get extra seq from seq_server, moving it's range up */
         while (lost-- > 0) {
-                seq_server_alloc_meta(ls->ls_server_seq, NULL, &out, env);
+                seq_server_alloc_meta(ms->ms_server_seq, NULL, &out, env);
         }
         EXIT;
 }
@@ -3046,25 +3096,27 @@ static void mdt_seq_adjust(const struct lu_env *env,
 static int mdt_seq_fini(const struct lu_env *env,
                         struct mdt_device *m)
 {
-        struct lu_site *ls = m->mdt_md_dev.md_lu_dev.ld_site;
+        struct md_site *ms = mdt_md_site(m);
         ENTRY;
 
-        if (ls && ls->ls_server_seq) {
-                seq_server_fini(ls->ls_server_seq, env);
-                OBD_FREE_PTR(ls->ls_server_seq);
-                ls->ls_server_seq = NULL;
+        if (ms != NULL) {
+                if (ms->ms_server_seq) {
+                        seq_server_fini(ms->ms_server_seq, env);
+                        OBD_FREE_PTR(ms->ms_server_seq);
+                        ms->ms_server_seq = NULL;
         }
 
-        if (ls && ls->ls_control_seq) {
-                seq_server_fini(ls->ls_control_seq, env);
-                OBD_FREE_PTR(ls->ls_control_seq);
-                ls->ls_control_seq = NULL;
+                if (ms->ms_control_seq) {
+                        seq_server_fini(ms->ms_control_seq, env);
+                        OBD_FREE_PTR(ms->ms_control_seq);
+                        ms->ms_control_seq = NULL;
         }
 
-        if (ls && ls->ls_client_seq) {
-                seq_client_fini(ls->ls_client_seq);
-                OBD_FREE_PTR(ls->ls_client_seq);
-                ls->ls_client_seq = NULL;
+                if (ms->ms_client_seq) {
+                        seq_client_fini(ms->ms_client_seq);
+                        OBD_FREE_PTR(ms->ms_client_seq);
+                        ms->ms_client_seq = NULL;
+                }
         }
 
         RETURN(0);
@@ -3074,25 +3126,25 @@ static int mdt_seq_init(const struct lu_env *env,
                         const char *uuid,
                         struct mdt_device *m)
 {
-        struct lu_site *ls;
+        struct md_site *ms;
         char *prefix;
         int rc;
         ENTRY;
 
-        ls = m->mdt_md_dev.md_lu_dev.ld_site;
+        ms = mdt_md_site(m);
 
         /*
          * This is sequence-controller node. Init seq-controller server on local
          * MDT.
          */
-        if (ls->ls_node_id == 0) {
-                LASSERT(ls->ls_control_seq == NULL);
+        if (ms->ms_node_id == 0) {
+                LASSERT(ms->ms_control_seq == NULL);
 
-                OBD_ALLOC_PTR(ls->ls_control_seq);
-                if (ls->ls_control_seq == NULL)
+                OBD_ALLOC_PTR(ms->ms_control_seq);
+                if (ms->ms_control_seq == NULL)
                         RETURN(-ENOMEM);
 
-                rc = seq_server_init(ls->ls_control_seq,
+                rc = seq_server_init(ms->ms_control_seq,
                                      m->mdt_bottom, uuid,
                                      LUSTRE_SEQ_CONTROLLER,
                                      env);
@@ -3100,13 +3152,13 @@ static int mdt_seq_init(const struct lu_env *env,
                 if (rc)
                         GOTO(out_seq_fini, rc);
 
-                OBD_ALLOC_PTR(ls->ls_client_seq);
-                if (ls->ls_client_seq == NULL)
+                OBD_ALLOC_PTR(ms->ms_client_seq);
+                if (ms->ms_client_seq == NULL)
                         GOTO(out_seq_fini, rc = -ENOMEM);
 
                 OBD_ALLOC(prefix, MAX_OBD_NAME + 5);
                 if (prefix == NULL) {
-                        OBD_FREE_PTR(ls->ls_client_seq);
+                        OBD_FREE_PTR(ms->ms_client_seq);
                         GOTO(out_seq_fini, rc = -ENOMEM);
                 }
 
@@ -3115,11 +3167,11 @@ static int mdt_seq_init(const struct lu_env *env,
 
                 /*
                  * Init seq-controller client after seq-controller server is
-                 * ready. Pass ls->ls_control_seq to it for direct talking.
+                 * ready. Pass ms->ms_control_seq to it for direct talking.
                  */
-                rc = seq_client_init(ls->ls_client_seq, NULL,
+                rc = seq_client_init(ms->ms_client_seq, NULL,
                                      LUSTRE_SEQ_METADATA, prefix,
-                                     ls->ls_control_seq);
+                                     ms->ms_control_seq);
                 OBD_FREE(prefix, MAX_OBD_NAME + 5);
 
                 if (rc)
@@ -3127,13 +3179,13 @@ static int mdt_seq_init(const struct lu_env *env,
         }
 
         /* Init seq-server on local MDT */
-        LASSERT(ls->ls_server_seq == NULL);
+        LASSERT(ms->ms_server_seq == NULL);
 
-        OBD_ALLOC_PTR(ls->ls_server_seq);
-        if (ls->ls_server_seq == NULL)
+        OBD_ALLOC_PTR(ms->ms_server_seq);
+        if (ms->ms_server_seq == NULL)
                 GOTO(out_seq_fini, rc = -ENOMEM);
 
-        rc = seq_server_init(ls->ls_server_seq,
+        rc = seq_server_init(ms->ms_server_seq,
                              m->mdt_bottom, uuid,
                              LUSTRE_SEQ_SERVER,
                              env);
@@ -3141,11 +3193,11 @@ static int mdt_seq_init(const struct lu_env *env,
                 GOTO(out_seq_fini, rc = -ENOMEM);
 
         /* Assign seq-controller client to local seq-server. */
-        if (ls->ls_node_id == 0) {
-                LASSERT(ls->ls_client_seq != NULL);
+        if (ms->ms_node_id == 0) {
+                LASSERT(ms->ms_client_seq != NULL);
 
-                rc = seq_server_set_cli(ls->ls_server_seq,
-                                        ls->ls_client_seq,
+                rc = seq_server_set_cli(ms->ms_server_seq,
+                                        ms->ms_client_seq,
                                         env);
         }
 
@@ -3164,7 +3216,7 @@ static int mdt_seq_init_cli(const struct lu_env *env,
                             struct mdt_device *m,
                             struct lustre_cfg *cfg)
 {
-        struct lu_site    *ls = m->mdt_md_dev.md_lu_dev.ld_site;
+        struct md_site    *ms = mdt_md_site(m);
         struct obd_device *mdc;
         struct obd_uuid   *uuidp, *mdcuuidp;
         char              *uuid_str, *mdc_uuid_str;
@@ -3188,7 +3240,7 @@ static int mdt_seq_init_cli(const struct lu_env *env,
 
         /* check if this is adding the first MDC and controller is not yet
          * initialized. */
-        if (index != 0 || ls->ls_client_seq)
+        if (index != 0 || ms->ms_client_seq)
                 RETURN(0);
 
         uuid_str = lustre_cfg_string(cfg, 1);
@@ -3205,9 +3257,9 @@ static int mdt_seq_init_cli(const struct lu_env *env,
                 CERROR("target %s not set up\n", mdc->obd_name);
                 rc = -EINVAL;
         } else {
-                LASSERT(ls->ls_control_exp);
-                OBD_ALLOC_PTR(ls->ls_client_seq);
-                if (ls->ls_client_seq != NULL) {
+                LASSERT(ms->ms_control_exp);
+                OBD_ALLOC_PTR(ms->ms_client_seq);
+                if (ms->ms_client_seq != NULL) {
                         char *prefix;
 
                         OBD_ALLOC(prefix, MAX_OBD_NAME + 5);
@@ -3217,8 +3269,8 @@ static int mdt_seq_init_cli(const struct lu_env *env,
                         snprintf(prefix, MAX_OBD_NAME + 5, "ctl-%s",
                                  mdc->obd_name);
 
-                        rc = seq_client_init(ls->ls_client_seq,
-                                             ls->ls_control_exp,
+                        rc = seq_client_init(ms->ms_client_seq,
+                                             ms->ms_control_exp,
                                              LUSTRE_SEQ_METADATA,
                                              prefix, NULL);
                         OBD_FREE(prefix, MAX_OBD_NAME + 5);
@@ -3228,8 +3280,8 @@ static int mdt_seq_init_cli(const struct lu_env *env,
                 if (rc)
                         RETURN(rc);
 
-                LASSERT(ls->ls_server_seq != NULL);
-                rc = seq_server_set_cli(ls->ls_server_seq, ls->ls_client_seq,
+                LASSERT(ms->ms_server_seq != NULL);
+                rc = seq_server_set_cli(ms->ms_server_seq, ms->ms_client_seq,
                                         env);
         }
 
@@ -3238,19 +3290,21 @@ static int mdt_seq_init_cli(const struct lu_env *env,
 
 static void mdt_seq_fini_cli(struct mdt_device *m)
 {
-        struct lu_site *ls;
+        struct md_site *ms;
 
         ENTRY;
 
-        ls = m->mdt_md_dev.md_lu_dev.ld_site;
+        ms = mdt_md_site(m);
 
-        if (ls && ls->ls_server_seq)
-                seq_server_set_cli(ls->ls_server_seq,
+        if (ms != NULL) {
+                if (ms->ms_server_seq)
+                        seq_server_set_cli(ms->ms_server_seq,
                                    NULL, NULL);
 
-        if (ls && ls->ls_control_exp) {
-                class_export_put(ls->ls_control_exp);
-                ls->ls_control_exp = NULL;
+                if (ms->ms_control_exp) {
+                        class_export_put(ms->ms_control_exp);
+                        ms->ms_control_exp = NULL;
+                }
         }
         EXIT;
 }
@@ -3261,13 +3315,13 @@ static void mdt_seq_fini_cli(struct mdt_device *m)
 static int mdt_fld_fini(const struct lu_env *env,
                         struct mdt_device *m)
 {
-        struct lu_site *ls = m->mdt_md_dev.md_lu_dev.ld_site;
+        struct md_site *ms = mdt_md_site(m);
         ENTRY;
 
-        if (ls && ls->ls_server_fld) {
-                fld_server_fini(ls->ls_server_fld, env);
-                OBD_FREE_PTR(ls->ls_server_fld);
-                ls->ls_server_fld = NULL;
+        if (ms && ms->ms_server_fld) {
+                fld_server_fini(ms->ms_server_fld, env);
+                OBD_FREE_PTR(ms->ms_server_fld);
+                ms->ms_server_fld = NULL;
         }
 
         RETURN(0);
@@ -3277,21 +3331,21 @@ static int mdt_fld_init(const struct lu_env *env,
                         const char *uuid,
                         struct mdt_device *m)
 {
-        struct lu_site *ls;
+        struct md_site *ms;
         int rc;
         ENTRY;
 
-        ls = m->mdt_md_dev.md_lu_dev.ld_site;
+        ms = mdt_md_site(m);
 
-        OBD_ALLOC_PTR(ls->ls_server_fld);
-        if (ls->ls_server_fld == NULL)
+        OBD_ALLOC_PTR(ms->ms_server_fld);
+        if (ms->ms_server_fld == NULL)
                 RETURN(rc = -ENOMEM);
 
-        rc = fld_server_init(ls->ls_server_fld,
+        rc = fld_server_init(ms->ms_server_fld,
                              m->mdt_bottom, uuid, env);
         if (rc) {
-                OBD_FREE_PTR(ls->ls_server_fld);
-                ls->ls_server_fld = NULL;
+                OBD_FREE_PTR(ms->ms_server_fld);
+                ms->ms_server_fld = NULL;
                 RETURN(rc);
         }
 
@@ -3631,7 +3685,7 @@ static void mdt_stack_fini(const struct lu_env *env,
         m->mdt_bottom = NULL;
 }
 
-static struct lu_device *mdt_layer_setup(const struct lu_env *env,
+static struct lu_device *mdt_layer_setup(struct lu_env *env,
                                          const char *typename,
                                          struct lu_device *child,
                                          struct lustre_cfg *cfg)
@@ -3687,6 +3741,7 @@ static struct lu_device *mdt_layer_setup(const struct lu_env *env,
                 GOTO(out_alloc, rc);
         }
         lu_device_get(d);
+        lu_ref_add(&d->ld_reference, "lu-stack", &lu_site_init);
 
         RETURN(d);
 
@@ -3699,7 +3754,7 @@ out:
         return ERR_PTR(rc);
 }
 
-static int mdt_stack_init(const struct lu_env *env,
+static int mdt_stack_init(struct lu_env *env,
                           struct mdt_device *m, struct lustre_cfg *cfg)
 {
         struct lu_device  *d = &m->mdt_md_dev.md_lu_dev;
@@ -3788,8 +3843,11 @@ static void mdt_fini(const struct lu_env *env, struct mdt_device *m)
         mdt_stack_fini(env, m, md2lu_dev(m->mdt_child));
 
         if (ls) {
+                struct md_site *mite;
+
                 lu_site_fini(ls);
-                OBD_FREE_PTR(ls);
+                mite = lu_site2md(ls);
+                OBD_FREE_PTR(mite);
                 d->ld_site = NULL;
         }
         LASSERT(atomic_read(&d->ld_ref) == 0);
@@ -3852,6 +3910,7 @@ static int mdt_init0(const struct lu_env *env, struct mdt_device *m,
         struct lustre_mount_info  *lmi;
         struct lustre_sb_info     *lsi;
         struct lu_site            *s;
+        struct md_site            *mite;
         const char                *identity_upcall = "NONE";
         int                        rc;
         ENTRY;
@@ -3890,11 +3949,13 @@ static int mdt_init0(const struct lu_env *env, struct mdt_device *m,
 
         spin_lock_init(&m->mdt_client_bitmap_lock);
 
-        OBD_ALLOC_PTR(s);
-        if (s == NULL)
+        OBD_ALLOC_PTR(mite);
+        if (mite == NULL)
                 RETURN(-ENOMEM);
 
         md_device_init(&m->mdt_md_dev, ldt);
+        s = &mite->ms_lu;
+
         m->mdt_md_dev.md_lu_dev.ld_ops = &mdt_lu_ops;
         m->mdt_md_dev.md_lu_dev.ld_obd = obd;
         /* set this lu_device to obd, because error handling need it */
@@ -3922,7 +3983,7 @@ static int mdt_init0(const struct lu_env *env, struct mdt_device *m,
 
         /* set server index */
         LASSERT(num);
-        s->ls_node_id = simple_strtol(num, NULL, 10);
+        lu_site2md(s)->ms_node_id = simple_strtol(num, NULL, 10);
 
         /* failover is the default
          * FIXME: we do not failout mds0/mgs, which may cause some problems.
@@ -3941,7 +4002,7 @@ static int mdt_init0(const struct lu_env *env, struct mdt_device *m,
         }
 
         /* init the stack */
-        rc = mdt_stack_init(env, m, cfg);
+        rc = mdt_stack_init((struct lu_env *)env, m, cfg);
         if (rc) {
                 CERROR("Can't init device stack, rc %d\n", rc);
                 GOTO(err_fini_proc, rc);
@@ -4037,6 +4098,7 @@ err_fini_stack:
         mdt_stack_fini(env, m, md2lu_dev(m->mdt_child));
 err_fini_proc:
         mdt_procfs_fini(m);
+        ptlrpc_lprocfs_unregister_obd(obd);
         lprocfs_obd_cleanup(obd);
 err_fini_site:
         lu_site_fini(s);
@@ -4091,9 +4153,10 @@ static int mdt_process_config(const struct lu_env *env,
                 struct obd_device *obd = d->ld_obd;
 
                 lprocfs_mdt_init_vars(&lvars);
-                rc = class_process_proc_param(PARAM_MDT, lvars.obd_vars, cfg, obd);
-                if (rc)
-                        /* others are passed further */
+                rc = class_process_proc_param(PARAM_MDT, lvars.obd_vars,
+                                              cfg, obd);
+                if (rc == -ENOSYS)
+                        /* we don't understand; pass it on */
                         rc = next->ld_ops->ldo_process_config(env, next, cfg);
                 break;
         }
@@ -4140,7 +4203,8 @@ static struct lu_object *mdt_object_alloc(const struct lu_env *env,
                 RETURN(NULL);
 }
 
-static int mdt_object_init(const struct lu_env *env, struct lu_object *o)
+static int mdt_object_init(const struct lu_env *env, struct lu_object *o,
+                           const struct lu_object_conf *_)
 {
         struct mdt_device *d = mdt_dev(o->lo_dev);
         struct lu_device  *under;
@@ -4177,21 +4241,14 @@ static void mdt_object_free(const struct lu_env *env, struct lu_object *o)
         EXIT;
 }
 
-static int mdt_object_print(const struct lu_env *env, void *cookie,
-                            lu_printer_t p, const struct lu_object *o)
-{
-        return (*p)(env, cookie, LUSTRE_MDT_NAME"-object@%p", o);
-}
-
-static struct lu_device_operations mdt_lu_ops = {
+static const struct lu_device_operations mdt_lu_ops = {
         .ldo_object_alloc   = mdt_object_alloc,
         .ldo_process_config = mdt_process_config
 };
 
-static struct lu_object_operations mdt_obj_ops = {
+static const struct lu_object_operations mdt_obj_ops = {
         .loo_object_init    = mdt_object_init,
-        .loo_object_free    = mdt_object_free,
-        .loo_object_print   = mdt_object_print
+        .loo_object_free    = mdt_object_free
 };
 
 /* mds_connect_internal */
@@ -4409,6 +4466,9 @@ static int mdt_obd_disconnect(struct obd_export *exp)
         if (mdt->mdt_namespace != NULL || exp->exp_obd->obd_namespace != NULL)
                 ldlm_cancel_locks_for_export(exp);
 
+        /* release nid stat refererence */
+        lprocfs_exp_cleanup(exp);
+
         /* complete all outstanding replies */
         spin_lock(&exp->exp_lock);
         while (!list_empty(&exp->exp_outstanding_replies)) {
@@ -4432,6 +4492,7 @@ static int mdt_obd_disconnect(struct obd_export *exp)
 static int mdt_init_export(struct obd_export *exp)
 {
         struct mdt_export_data *med = &exp->exp_mdt_data;
+        int                     rc;
         ENTRY;
 
         CFS_INIT_LIST_HEAD(&med->med_open_head);
@@ -4441,7 +4502,10 @@ static int mdt_init_export(struct obd_export *exp)
         spin_lock(&exp->exp_lock);
         exp->exp_connecting = 1;
         spin_unlock(&exp->exp_lock);
-        RETURN(0);
+        rc = ldlm_init_export(exp);
+        if (rc)
+                CERROR("Error %d while initializing export\n", rc);
+        RETURN(rc);
 }
 
 static int mdt_destroy_export(struct obd_export *export)
@@ -4462,6 +4526,7 @@ static int mdt_destroy_export(struct obd_export *export)
                 mdt_cleanup_idmap(med);
 
         target_destroy_export(export);
+        ldlm_destroy_export(export);
 
         if (obd_uuid_equals(&export->exp_client_uuid, &obd->obd_uuid))
                 RETURN(0);
@@ -4611,7 +4676,7 @@ static int mdt_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
 
         switch (cmd) {
         case OBD_IOC_SYNC:
-                rc = dt->dd_ops->dt_sync(&env, dt);
+                rc = mdt_device_sync(&env, mdt);
                 break;
         case OBD_IOC_SET_READONLY:
                 rc = dt->dd_ops->dt_sync(&env, dt);
@@ -4743,6 +4808,9 @@ static struct lu_device_type_operations mdt_device_type_ops = {
         .ldto_init = mdt_type_init,
         .ldto_fini = mdt_type_fini,
 
+        .ldto_start = mdt_type_start,
+        .ldto_stop  = mdt_type_stop,
+
         .ldto_device_alloc = mdt_device_alloc,
         .ldto_device_free  = mdt_device_free,
         .ldto_device_fini  = mdt_device_fini
@@ -4978,7 +5046,7 @@ static struct mdt_opc_slice mdt_fld_handlers[] = {
         }
 };
 
-MODULE_AUTHOR("Cluster File Systems, Inc. <info@clusterfs.com>");
+MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
 MODULE_DESCRIPTION("Lustre Meta-data Target ("LUSTRE_MDT_NAME")");
 MODULE_LICENSE("GPL");