Whamcloud - gitweb
LU-354 test: Change dev_set_rdonly() check to warning
[fs/lustre-release.git] / lustre / mdt / mdt_handler.c
index 8921815..55ab803 100644 (file)
  * GPL HEADER END
  */
 /*
- * Copyright  2008 Sun Microsystems, Inc. All rights reserved
+ * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
  * Use is subject to license terms.
+ *
+ * Copyright (c) 2011, 2012, Whamcloud, Inc.
  */
 /*
  * This file is part of Lustre, http://www.lustre.org/
@@ -99,7 +101,9 @@ ldlm_mode_t mdt_dlm_lock_modes[] = {
 /*
  * Initialized in mdt_mod_init().
  */
-unsigned long mdt_num_threads;
+static unsigned long mdt_num_threads;
+static unsigned long mdt_min_threads;
+static unsigned long mdt_max_threads;
 
 /* ptlrpc request handler for MDT. All handlers are
  * grouped into several slices - struct mdt_opc_slice,
@@ -208,7 +212,7 @@ void mdt_lock_pdo_init(struct mdt_lock_handle *lh, ldlm_mode_t lm,
         lh->mlh_reg_mode = lm;
         lh->mlh_type = MDT_PDO_LOCK;
 
-        if (name != NULL) {
+        if (name != NULL && (name[0] != '\0')) {
                 LASSERT(namelen > 0);
                 lh->mlh_pdo_hash = full_name_hash(name, namelen);
         } else {
@@ -339,6 +343,7 @@ static int mdt_getstatus(struct mdt_thread_info *info)
 
 static int mdt_statfs(struct mdt_thread_info *info)
 {
+        struct ptlrpc_request *req = mdt_info_req(info);
         struct md_device      *next  = info->mti_mdt->mdt_child;
         struct ptlrpc_service *svc;
         struct obd_statfs     *osfs;
@@ -365,36 +370,46 @@ static int mdt_statfs(struct mdt_thread_info *info)
                                               &info->mti_u.ksfs);
                 statfs_pack(osfs, &info->mti_u.ksfs);
         }
+
+        if (rc == 0)
+                mdt_counter_incr(req->rq_export, LPROC_MDT_STATFS);
+
         RETURN(rc);
 }
 
-void mdt_pack_size2body(struct mdt_thread_info *info, struct mdt_object *o)
+/**
+ * Pack SOM attributes into the reply.
+ * Call under a DLM UPDATE lock.
+ */
+static void mdt_pack_size2body(struct mdt_thread_info *info,
+                               struct mdt_object *mo)
 {
         struct mdt_body *b;
-        struct lu_attr *attr = &info->mti_attr.ma_attr;
+        struct md_attr *ma = &info->mti_attr;
 
+        LASSERT(ma->ma_attr.la_valid & LA_MODE);
         b = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
 
-        /* Check if Size-on-MDS is enabled. */
-        if ((mdt_conn_flags(info) & OBD_CONNECT_SOM) &&
-            S_ISREG(attr->la_mode) && mdt_sizeonmds_enabled(o)) {
-                b->valid |= (OBD_MD_FLSIZE | OBD_MD_FLBLOCKS);
-                b->size = attr->la_size;
-                b->blocks = attr->la_blocks;
-        }
+        /* Check if Size-on-MDS is supported, if this is a regular file,
+         * if SOM is enabled on the object and if SOM cache exists and valid.
+         * Otherwise do not pack Size-on-MDS attributes to the reply. */
+        if (!(mdt_conn_flags(info) & OBD_CONNECT_SOM) ||
+            !S_ISREG(ma->ma_attr.la_mode) ||
+            !mdt_object_is_som_enabled(mo) ||
+            !(ma->ma_valid & MA_SOM))
+                return;
+
+        b->valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS;
+        b->size = ma->ma_som->msd_size;
+        b->blocks = ma->ma_som->msd_blocks;
 }
 
 void mdt_pack_attr2body(struct mdt_thread_info *info, struct mdt_body *b,
                         const struct lu_attr *attr, const struct lu_fid *fid)
 {
-        /*XXX should pack the reply body according to lu_valid*/
-        b->valid |= OBD_MD_FLCTIME | OBD_MD_FLUID   |
-                    OBD_MD_FLGID   | OBD_MD_FLTYPE  |
-                    OBD_MD_FLMODE  | OBD_MD_FLNLINK | OBD_MD_FLFLAGS |
-                    OBD_MD_FLATIME | OBD_MD_FLMTIME ;
+        struct md_attr *ma = &info->mti_attr;
 
-        if (!S_ISREG(attr->la_mode))
-                b->valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS | OBD_MD_FLRDEV;
+        LASSERT(ma->ma_valid & MA_INODE);
 
         b->atime      = attr->la_atime;
         b->mtime      = attr->la_mtime;
@@ -408,6 +423,23 @@ void mdt_pack_attr2body(struct mdt_thread_info *info, struct mdt_body *b,
         b->nlink      = attr->la_nlink;
         b->rdev       = attr->la_rdev;
 
+        /*XXX should pack the reply body according to lu_valid*/
+        b->valid |= OBD_MD_FLCTIME | OBD_MD_FLUID   |
+                    OBD_MD_FLGID   | OBD_MD_FLTYPE  |
+                    OBD_MD_FLMODE  | OBD_MD_FLNLINK | OBD_MD_FLFLAGS |
+                    OBD_MD_FLATIME | OBD_MD_FLMTIME ;
+
+        if (!S_ISREG(attr->la_mode)) {
+                b->valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS | OBD_MD_FLRDEV;
+        } else if (ma->ma_need & MA_LOV && ma->ma_lmm_size == 0) {
+                /* means no objects are allocated on osts. */
+                LASSERT(!(ma->ma_valid & MA_LOV));
+                /* just ignore blocks occupied by extend attributes on MDS */
+                b->blocks = 0;
+                /* if no object is allocated on osts, the size on mds is valid. b=22272 */
+                b->valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS;
+        }
+
         if (fid) {
                 b->fid1 = *fid;
                 b->valid |= OBD_MD_FLID;
@@ -423,6 +455,10 @@ void mdt_pack_attr2body(struct mdt_thread_info *info, struct mdt_body *b,
 
         if (info)
                 mdt_body_reverse_idmap(info, b);
+
+        if (b->valid & OBD_MD_FLSIZE)
+                CDEBUG(D_VFSTRACE, DFID": returning size %llu\n",
+                       PFID(fid), (unsigned long long)b->size);
 }
 
 static inline int mdt_body_has_lov(const struct lu_attr *la,
@@ -432,8 +468,34 @@ static inline int mdt_body_has_lov(const struct lu_attr *la,
                 (S_ISDIR(la->la_mode) && (body->valid & OBD_MD_FLDIREA )) );
 }
 
+void mdt_client_compatibility(struct mdt_thread_info *info)
+{
+        struct mdt_body       *body;
+        struct ptlrpc_request *req = mdt_info_req(info);
+        struct obd_export     *exp = req->rq_export;
+        struct md_attr        *ma = &info->mti_attr;
+        struct lu_attr        *la = &ma->ma_attr;
+        ENTRY;
+
+        if (exp->exp_connect_flags & OBD_CONNECT_LAYOUTLOCK)
+                /* the client can deal with 16-bit lmm_stripe_count */
+                RETURN_EXIT;
+
+        body = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
+
+        if (!mdt_body_has_lov(la, body))
+                RETURN_EXIT;
+
+        /* now we have a reply with a lov for a client not compatible with the
+         * layout lock so we have to clean the layout generation number */
+        if (S_ISREG(la->la_mode))
+                ma->ma_lmm->lmm_layout_gen = 0;
+        EXIT;
+}
+
+
 static int mdt_getattr_internal(struct mdt_thread_info *info,
-                                struct mdt_object *o)
+                                struct mdt_object *o, int ma_need)
 {
         struct md_object        *next = mdt_object_child(o);
         const struct mdt_body   *reqbody = info->mti_body;
@@ -484,6 +546,10 @@ static int mdt_getattr_internal(struct mdt_thread_info *info,
                 /* get default stripe info for this dir. */
                 ma->ma_need |= MA_LOV_DEF;
         }
+        ma->ma_need |= ma_need;
+        if (ma->ma_need & MA_SOM)
+                ma->ma_som = &info->mti_u.som.data;
+
         rc = mo_attr_get(env, next, ma);
         if (unlikely(rc)) {
                 CERROR("getattr error for "DFID": %d\n",
@@ -511,9 +577,6 @@ static int mdt_getattr_internal(struct mdt_thread_info *info,
                         repbody->eadatasize = ma->ma_lmv_size;
                         repbody->valid |= (OBD_MD_FLDIREA|OBD_MD_MEA);
                 }
-                if (!(ma->ma_valid & MA_LOV) && !(ma->ma_valid & MA_LMV)) {
-                        repbody->valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS;
-                }
         } else if (S_ISLNK(la->la_mode) &&
                    reqbody->valid & OBD_MD_LINKNAME) {
                 buffer->lb_buf = ma->ma_lmm;
@@ -637,6 +700,7 @@ static int mdt_renew_capa(struct mdt_thread_info *info)
 
 static int mdt_getattr(struct mdt_thread_info *info)
 {
+        struct ptlrpc_request   *req = mdt_info_req(info);
         struct mdt_object       *obj = info->mti_object;
         struct req_capsule      *pill = info->mti_pill;
         struct mdt_body         *reqbody;
@@ -693,11 +757,15 @@ static int mdt_getattr(struct mdt_thread_info *info)
          * remote obj, and at that time no capability is available.
          */
         mdt_set_capainfo(info, 1, &reqbody->fid1, BYPASS_CAPA);
-        rc = mdt_getattr_internal(info, obj);
+        rc = mdt_getattr_internal(info, obj, 0);
         if (reqbody->valid & OBD_MD_FLRMTPERM)
                 mdt_exit_ucred(info);
         EXIT;
 out_shrink:
+        if (rc == 0)
+                mdt_counter_incr(req->rq_export, LPROC_MDT_GETATTR);
+
+        mdt_client_compatibility(info);
         mdt_shrink_reply(info);
         return rc;
 }
@@ -747,6 +815,7 @@ static int mdt_raw_lookup(struct mdt_thread_info *info,
         LASSERT(!info->mti_cross_ref);
 
         /* Only got the fid of this obj by name */
+        fid_zero(child_fid);
         rc = mdo_lookup(info->mti_env, next, lname, child_fid,
                         &info->mti_spec);
 #if 0
@@ -788,10 +857,11 @@ static int mdt_getattr_name_lock(struct mdt_thread_info *info,
         struct lu_name         *lname     = NULL;
         const char             *name      = NULL;
         int                     namelen   = 0;
-        struct mdt_lock_handle *lhp;
+        struct mdt_lock_handle *lhp       = NULL;
         struct ldlm_lock       *lock;
         struct ldlm_res_id     *res_id;
         int                     is_resent;
+        int                     ma_need = 0;
         int                     rc;
 
         ENTRY;
@@ -817,13 +887,17 @@ static int mdt_getattr_name_lock(struct mdt_thread_info *info,
                 if (namelen == 0) {
                         reqbody = req_capsule_client_get(info->mti_pill,
                                                          &RMF_MDT_BODY);
-                        LASSERT(fid_is_sane(&reqbody->fid2));
-                        name = NULL;
+                        if (unlikely(reqbody == NULL))
+                                RETURN(err_serious(-EFAULT));
+
+                        if (unlikely(!fid_is_sane(&reqbody->fid2)))
+                                RETURN(err_serious(-EINVAL));
 
+                        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);
+                               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, "
@@ -886,30 +960,34 @@ static int mdt_getattr_name_lock(struct mdt_thread_info *info,
                         /* Finally, we can get attr for child. */
                         mdt_set_capainfo(info, 0, mdt_object_fid(child),
                                          BYPASS_CAPA);
-                        rc = mdt_getattr_internal(info, child);
+                        rc = mdt_getattr_internal(info, child, 0);
                         if (unlikely(rc != 0))
                                 mdt_object_unlock(info, child, lhc, 1);
                 }
                 RETURN(rc);
         }
 
-        /* step 1: lock parent */
-        lhp = &info->mti_lh[MDT_LH_PARENT];
-        mdt_lock_pdo_init(lhp, LCK_PR, name, namelen);
-        rc = mdt_object_lock(info, parent, lhp, MDS_INODELOCK_UPDATE,
-                             MDT_LOCAL_LOCK);
-
-        if (unlikely(rc != 0))
-                RETURN(rc);
-
         if (lname) {
+                /* step 1: lock parent only if parent is a directory */
+                if (S_ISDIR(lu_object_attr(&parent->mot_obj.mo_lu))) {
+                        lhp = &info->mti_lh[MDT_LH_PARENT];
+                        mdt_lock_pdo_init(lhp, LCK_PR, name, namelen);
+                        rc = mdt_object_lock(info, parent, lhp,
+                                             MDS_INODELOCK_UPDATE,
+                                             MDT_LOCAL_LOCK);
+                        if (unlikely(rc != 0))
+                                RETURN(rc);
+                }
+
                 /* step 2: lookup child's fid by name */
+                fid_zero(child_fid);
                 rc = mdo_lookup(info->mti_env, next, lname, child_fid,
                                 &info->mti_spec);
 
                 if (rc != 0) {
                         if (rc == -ENOENT)
-                                mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_NEG);
+                                mdt_set_disposition(info, ldlm_rep,
+                                                    DISP_LOOKUP_NEG);
                         GOTO(out_parent, rc);
                 } else
                         mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_POS);
@@ -953,36 +1031,48 @@ static int mdt_getattr_name_lock(struct mdt_thread_info *info,
                 LDLM_LOCK_PUT(lock);
                 rc = 0;
         } else {
-                struct md_attr *ma;
 relock:
-                ma = &info->mti_attr;
-
                 OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_RESEND, obd_timeout*2);
                 mdt_lock_handle_init(lhc);
-                mdt_lock_reg_init(lhc, LCK_PR);
+                if (child_bits == MDS_INODELOCK_LAYOUT)
+                        mdt_lock_reg_init(lhc, LCK_CR);
+                else
+                        mdt_lock_reg_init(lhc, LCK_PR);
 
                 if (mdt_object_exists(child) == 0) {
-                        LU_OBJECT_DEBUG(D_WARNING, info->mti_env,
+                        LU_OBJECT_DEBUG(D_INODE, info->mti_env,
                                         &child->mot_obj.mo_lu,
                                         "Object doesn't exist!\n");
-                        GOTO(out_child, rc = -ESTALE);
+                        GOTO(out_child, rc = -ENOENT);
                 }
 
-                ma->ma_valid = 0;
-                ma->ma_need = MA_INODE;
-                rc = mo_attr_get(info->mti_env, next, ma);
-                if (unlikely(rc != 0))
-                        GOTO(out_child, rc);
+                if (!(child_bits & MDS_INODELOCK_UPDATE)) {
+                        struct md_attr *ma = &info->mti_attr;
 
-                /* If the file has not been changed for some time, we return
-                 * not only a LOOKUP lock, but also an UPDATE lock and this
-                 * might save us RPC on later STAT. For directories, it also
-                 * let negative dentry starts working for this dir. */
-                if (ma->ma_valid & MA_INODE &&
-                    ma->ma_attr.la_valid & LA_CTIME &&
-                    info->mti_mdt->mdt_namespace->ns_ctime_age_limit +
-                    ma->ma_attr.la_ctime < cfs_time_current_sec())
-                        child_bits |= MDS_INODELOCK_UPDATE;
+                        ma->ma_valid = 0;
+                        ma->ma_need = MA_INODE;
+                        rc = mo_attr_get(info->mti_env,
+                                         mdt_object_child(child), ma);
+                        if (unlikely(rc != 0))
+                                GOTO(out_child, rc);
+
+                        /* layout lock is used only on regular files */
+                        if ((ma->ma_valid & MA_INODE) &&
+                            (ma->ma_attr.la_valid & LA_MODE) &&
+                            !S_ISREG(ma->ma_attr.la_mode))
+                                child_bits &= ~MDS_INODELOCK_LAYOUT;
+
+                        /* If the file has not been changed for some time, we
+                         * return not only a LOOKUP lock, but also an UPDATE
+                         * lock and this might save us RPC on later STAT. For
+                         * directories, it also let negative dentry starts
+                         * working for this dir. */
+                        if (ma->ma_valid & MA_INODE &&
+                            ma->ma_attr.la_valid & LA_CTIME &&
+                            info->mti_mdt->mdt_namespace->ns_ctime_age_limit +
+                                ma->ma_attr.la_ctime < cfs_time_current_sec())
+                                child_bits |= MDS_INODELOCK_UPDATE;
+                }
 
                 rc = mdt_object_lock(info, child, lhc, child_bits,
                                      MDT_CROSS_LOCK);
@@ -991,43 +1081,40 @@ relock:
                         GOTO(out_child, rc);
         }
 
+        lock = ldlm_handle2lock(&lhc->mlh_reg_lh);
+        /* Get MA_SOM attributes if update lock is given. */
+        if (lock &&
+            lock->l_policy_data.l_inodebits.bits & MDS_INODELOCK_UPDATE &&
+            S_ISREG(lu_object_attr(&mdt_object_child(child)->mo_lu)))
+                ma_need = MA_SOM;
+
         /* finally, we can get attr for child. */
         mdt_set_capainfo(info, 1, child_fid, BYPASS_CAPA);
-        rc = mdt_getattr_internal(info, child);
+        rc = mdt_getattr_internal(info, child, ma_need);
         if (unlikely(rc != 0)) {
                 mdt_object_unlock(info, child, lhc, 1);
-        } else {
-                lock = ldlm_handle2lock(&lhc->mlh_reg_lh);
-                if (lock) {
-                        struct mdt_body *repbody;
+        } else if (lock) {
+                /* Debugging code. */
+                res_id = &lock->l_resource->lr_name;
+                LDLM_DEBUG(lock, "Returning lock to client");
+                LASSERTF(fid_res_name_eq(mdt_object_fid(child),
+                                         &lock->l_resource->lr_name),
+                         "Lock res_id: %lu/%lu/%lu, Fid: "DFID".\n",
+                         (unsigned long)res_id->name[0],
+                         (unsigned long)res_id->name[1],
+                         (unsigned long)res_id->name[2],
+                         PFID(mdt_object_fid(child)));
+                mdt_pack_size2body(info, child);
+        }
+        if (lock)
+                LDLM_LOCK_PUT(lock);
 
-                        /* Debugging code. */
-                        res_id = &lock->l_resource->lr_name;
-                        LDLM_DEBUG(lock, "Returning lock to client");
-                        LASSERTF(fid_res_name_eq(mdt_object_fid(child),
-                                                 &lock->l_resource->lr_name),
-                                 "Lock res_id: %lu/%lu/%lu, Fid: "DFID".\n",
-                                 (unsigned long)res_id->name[0],
-                                 (unsigned long)res_id->name[1],
-                                 (unsigned long)res_id->name[2],
-                                 PFID(mdt_object_fid(child)));
-                        /*
-                         * Pack Size-on-MDS inode attributes to the body if
-                         * update lock is given.
-                         */
-                        repbody = req_capsule_server_get(info->mti_pill,
-                                                         &RMF_MDT_BODY);
-                        if (lock->l_policy_data.l_inodebits.bits &
-                            MDS_INODELOCK_UPDATE)
-                                mdt_pack_size2body(info, child);
-                        LDLM_LOCK_PUT(lock);
-                }
-        }
         EXIT;
 out_child:
         mdt_object_put(info->mti_env, child);
 out_parent:
-        mdt_object_unlock(info, parent, lhp, 1);
+        if (lhp)
+                mdt_object_unlock(info, parent, lhp, 1);
         return rc;
 }
 
@@ -1062,6 +1149,7 @@ static int mdt_getattr_name(struct mdt_thread_info *info)
         mdt_exit_ucred(info);
         EXIT;
 out_shrink:
+        mdt_client_compatibility(info);
         mdt_shrink_reply(info);
         return rc;
 }
@@ -1111,12 +1199,12 @@ static int mdt_set_info(struct mdt_thread_info *info)
                 req->rq_status = 0;
                 lustre_msg_set_status(req->rq_repmsg, 0);
 
-                spin_lock(&req->rq_export->exp_lock);
+                cfs_spin_lock(&req->rq_export->exp_lock);
                 if (*(__u32 *)val)
                         req->rq_export->exp_connect_flags |= OBD_CONNECT_RDONLY;
                 else
                         req->rq_export->exp_connect_flags &=~OBD_CONNECT_RDONLY;
-                spin_unlock(&req->rq_export->exp_lock);
+                cfs_spin_unlock(&req->rq_export->exp_lock);
 
         } else if (KEY_IS(KEY_CHANGELOG_CLEAR)) {
                 struct changelog_setinfo *cs =
@@ -1173,7 +1261,7 @@ static int mdt_disconnect(struct mdt_thread_info *info)
 }
 
 static int mdt_sendpage(struct mdt_thread_info *info,
-                        struct lu_rdpg *rdpg)
+                        struct lu_rdpg *rdpg, int nob)
 {
         struct ptlrpc_request   *req = mdt_info_req(info);
         struct obd_export       *exp = req->rq_export;
@@ -1181,7 +1269,6 @@ static int mdt_sendpage(struct mdt_thread_info *info,
         struct l_wait_info      *lwi = &info->mti_u.rdpg.mti_wait_info;
         int                      tmpcount;
         int                      tmpsize;
-        int                      timeout;
         int                      i;
         int                      rc;
         ENTRY;
@@ -1191,58 +1278,16 @@ static int mdt_sendpage(struct mdt_thread_info *info,
         if (desc == NULL)
                 RETURN(-ENOMEM);
 
-        for (i = 0, tmpcount = rdpg->rp_count;
-                i < rdpg->rp_npages; i++, tmpcount -= tmpsize) {
+        for (i = 0, tmpcount = nob;
+                i < rdpg->rp_npages && tmpcount > 0; i++, tmpcount -= tmpsize) {
                 tmpsize = min_t(int, tmpcount, CFS_PAGE_SIZE);
                 ptlrpc_prep_bulk_page(desc, rdpg->rp_pages[i], 0, tmpsize);
         }
 
-        LASSERT(desc->bd_nob == rdpg->rp_count);
-        rc = sptlrpc_svc_wrap_bulk(req, desc);
-        if (rc)
-                GOTO(free_desc, rc);
-
-        rc = ptlrpc_start_bulk_transfer(desc);
-        if (rc)
-                GOTO(free_desc, rc);
-
-        if (OBD_FAIL_CHECK(OBD_FAIL_MDS_SENDPAGE))
-                GOTO(abort_bulk, rc = 0);
-
-        timeout = (int) req->rq_deadline - cfs_time_current_sec();
-        if (timeout < 0)
-                CERROR("Req deadline already passed %lu (now: %lu)\n",
-                       req->rq_deadline, cfs_time_current_sec());
-        *lwi = LWI_TIMEOUT_INTERVAL(cfs_time_seconds(max(timeout, 1)),
-                                    cfs_time_seconds(1), NULL, NULL);
-        rc = l_wait_event(desc->bd_waitq, !ptlrpc_server_bulk_active(desc) ||
-                          exp->exp_failed || exp->exp_abort_active_req, lwi);
-        LASSERT (rc == 0 || rc == -ETIMEDOUT);
-
-        if (rc == 0) {
-                if (desc->bd_success &&
-                    desc->bd_nob_transferred == rdpg->rp_count)
-                        GOTO(free_desc, rc);
-
-                rc = -ETIMEDOUT;
-                if (exp->exp_abort_active_req || exp->exp_failed)
-                        GOTO(abort_bulk, rc);
-        }
-
-        DEBUG_REQ(D_ERROR, req, "bulk failed: %s %d(%d), evicting %s@%s",
-                  (rc == -ETIMEDOUT) ? "timeout" : "network error",
-                  desc->bd_nob_transferred, rdpg->rp_count,
-                  exp->exp_client_uuid.uuid,
-                  exp->exp_connection->c_remote_uuid.uuid);
-
-        class_fail_export(exp);
-
-        EXIT;
-abort_bulk:
-        ptlrpc_abort_bulk(desc);
-free_desc:
+        LASSERT(desc->bd_nob == nob);
+        rc = target_bulk_io(exp, desc, lwi);
         ptlrpc_free_bulk(desc);
-        return rc;
+        RETURN(rc);
 }
 
 #ifdef HAVE_SPLIT_SUPPORT
@@ -1381,7 +1426,7 @@ static int mdt_writepage(struct mdt_thread_info *info)
         else
                 rc = ptlrpc_start_bulk_transfer (desc);
         if (rc == 0) {
-                *lwi = LWI_TIMEOUT_INTERVAL(obd_timeout * HZ / 4, HZ,
+                *lwi = LWI_TIMEOUT_INTERVAL(obd_timeout * CFS_HZ / 4, CFS_HZ,
                                             mdt_bulk_timeout, desc);
                 rc = l_wait_event(desc->bd_waitq, !ptlrpc_bulk_active(desc) ||
                                   desc->bd_export->exp_failed, lwi);
@@ -1450,8 +1495,12 @@ static int mdt_readpage(struct mdt_thread_info *info)
         }
 
         rdpg->rp_attrs = reqbody->mode;
-        rdpg->rp_count  = reqbody->nlink;
-        rdpg->rp_npages = (rdpg->rp_count + CFS_PAGE_SIZE - 1)>>CFS_PAGE_SHIFT;
+        if (info->mti_exp->exp_connect_flags & OBD_CONNECT_64BITHASH)
+                rdpg->rp_attrs |= LUDA_64BITHASH;
+        rdpg->rp_count  = min_t(unsigned int, reqbody->nlink,
+                                PTLRPC_MAX_BRW_SIZE);
+        rdpg->rp_npages = (rdpg->rp_count + CFS_PAGE_SIZE - 1) >>
+                          CFS_PAGE_SHIFT;
         OBD_ALLOC(rdpg->rp_pages, rdpg->rp_npages * sizeof rdpg->rp_pages[0]);
         if (rdpg->rp_pages == NULL)
                 RETURN(-ENOMEM);
@@ -1464,11 +1513,11 @@ static int mdt_readpage(struct mdt_thread_info *info)
 
         /* call lower layers to fill allocated pages with directory data */
         rc = mo_readpage(info->mti_env, mdt_object_child(object), rdpg);
-        if (rc)
+        if (rc < 0)
                 GOTO(free_rdpg, rc);
 
         /* send pages to client */
-        rc = mdt_sendpage(info, rdpg);
+        rc = mdt_sendpage(info, rdpg, rc);
 
         EXIT;
 free_rdpg:
@@ -1490,6 +1539,7 @@ static int mdt_reint_internal(struct mdt_thread_info *info,
 {
         struct req_capsule      *pill = info->mti_pill;
         struct mdt_device       *mdt = info->mti_mdt;
+        struct md_quota         *mq = md_quota(info->mti_env);
         struct mdt_body         *repbody;
         int                      rc = 0;
         ENTRY;
@@ -1548,11 +1598,13 @@ static int mdt_reint_internal(struct mdt_thread_info *info,
                 rc = lustre_msg_get_status(mdt_info_req(info)->rq_repmsg);
                 GOTO(out_ucred, rc);
         }
+        mq->mq_exp = info->mti_exp;
         rc = mdt_reint_rec(info, lhc);
         EXIT;
 out_ucred:
         mdt_exit_ucred(info);
 out_shrink:
+        mdt_client_compatibility(info);
         mdt_shrink_reply(info);
         return rc;
 }
@@ -1641,6 +1693,7 @@ static int mdt_object_sync(struct mdt_thread_info *info)
 
 static int mdt_sync(struct mdt_thread_info *info)
 {
+        struct ptlrpc_request *req = mdt_info_req(info);
         struct req_capsule *pill = info->mti_pill;
         struct mdt_body *body;
         int rc;
@@ -1688,6 +1741,9 @@ static int mdt_sync(struct mdt_thread_info *info)
                 } else
                         rc = err_serious(rc);
         }
+        if (rc == 0)
+                mdt_counter_incr(req->rq_export, LPROC_MDT_SYNC);
+
         RETURN(rc);
 }
 
@@ -1697,6 +1753,7 @@ static int mdt_quotacheck_handle(struct mdt_thread_info *info)
         struct obd_quotactl *oqctl;
         struct req_capsule *pill = info->mti_pill;
         struct obd_export *exp = info->mti_exp;
+        struct md_quota *mq = md_quota(info->mti_env);
         struct md_device *next = info->mti_mdt->mdt_child;
         int rc;
         ENTRY;
@@ -1713,7 +1770,8 @@ static int mdt_quotacheck_handle(struct mdt_thread_info *info)
         if (rc)
                 RETURN(rc);
 
-        rc = next->md_ops->mdo_quota.mqo_check(info->mti_env, next, exp,
+        mq->mq_exp = exp;
+        rc = next->md_ops->mdo_quota.mqo_check(info->mti_env, next,
                                                oqctl->qc_type);
         RETURN(rc);
 }
@@ -1723,6 +1781,7 @@ static int mdt_quotactl_handle(struct mdt_thread_info *info)
         struct obd_quotactl *oqctl, *repoqc;
         struct req_capsule *pill = info->mti_pill;
         struct obd_export *exp = info->mti_exp;
+        struct md_quota *mq = md_quota(info->mti_env);
         struct md_device *next = info->mti_mdt->mdt_child;
         const struct md_quota_operations *mqo = &next->md_ops->mdo_quota;
         int id, rc;
@@ -1766,6 +1825,7 @@ static int mdt_quotactl_handle(struct mdt_thread_info *info)
         repoqc = req_capsule_server_get(pill, &RMF_OBD_QUOTACTL);
         LASSERT(repoqc != NULL);
 
+        mq->mq_exp = exp;
         switch (oqctl->qc_cmd) {
         case Q_QUOTAON:
                 rc = mqo->mqo_on(info->mti_env, next, oqctl->qc_type);
@@ -1949,7 +2009,6 @@ static struct ldlm_callback_suite cbs = {
 static int mdt_enqueue(struct mdt_thread_info *info)
 {
         struct ptlrpc_request *req;
-        __u64 req_bits;
         int rc;
 
         /*
@@ -1959,17 +2018,6 @@ static int mdt_enqueue(struct mdt_thread_info *info)
         LASSERT(info->mti_dlm_req != NULL);
 
         req = mdt_info_req(info);
-
-        /*
-         * Lock without inodebits makes no sense and will oops later in
-         * ldlm. Let's check it now to see if we have wrong lock from client or
-         * bits get corrupted somewhere in mdt_intent_policy().
-         */
-        req_bits = info->mti_dlm_req->lock_desc.l_policy_data.l_inodebits.bits;
-        /* This is disabled because we need to support liblustre flock.
-         * LASSERT(req_bits != 0);
-         */
-
         rc = ldlm_handle_enqueue0(info->mti_mdt->mdt_namespace,
                                   req, info->mti_dlm_req, &cbs);
         info->mti_fail_id = OBD_FAIL_LDLM_REPLY;
@@ -2019,7 +2067,7 @@ static int mdt_sec_ctx_handle(struct mdt_thread_info *info)
                         sptlrpc_svc_ctx_invalidate(req);
         }
 
-        OBD_FAIL_TIMEOUT(OBD_FAIL_SEC_CTX_HDL_PAUSE, obd_fail_val);
+        CFS_FAIL_TIMEOUT(OBD_FAIL_SEC_CTX_HDL_PAUSE, cfs_fail_val);
 
         return rc;
 }
@@ -2112,7 +2160,7 @@ static inline int mdt_is_lock_sync(struct ldlm_lock *lock)
 int mdt_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
                      void *data, int flag)
 {
-        struct obd_device *obd = lock->l_resource->lr_namespace->ns_obd;
+        struct obd_device *obd = ldlm_lock_to_ns(lock)->ns_obd;
         struct mdt_device *mdt = mdt_dev(obd->obd_lu_dev);
         int rc;
         ENTRY;
@@ -2183,7 +2231,8 @@ int mdt_object_lock(struct mdt_thread_info *info, struct mdt_object *o,
                         RETURN(-ESTALE);
                 } else {
                         /* Non-dir object shouldn't have PDO lock */
-                        LASSERT(S_ISDIR(lu_object_attr(&o->mot_obj.mo_lu)));
+                        if (!S_ISDIR(lu_object_attr(&o->mot_obj.mo_lu)))
+                                RETURN(-ENOTDIR);
                 }
         }
 
@@ -2229,12 +2278,12 @@ int mdt_object_lock(struct mdt_thread_info *info, struct mdt_object *o,
                           res_id, LDLM_FL_LOCAL_ONLY | LDLM_FL_ATOMIC_CB,
                           &info->mti_exp->exp_handle.h_cookie);
         if (rc)
-                GOTO(out, rc);
-
-out:
-        if (rc)
                 mdt_object_unlock(info, o, lh, 1);
-
+        else if (unlikely(OBD_FAIL_PRECHECK(OBD_FAIL_MDS_PDO_LOCK)) &&
+                 lh->mlh_pdo_hash != 0 &&
+                 (lh->mlh_reg_mode == LCK_PW || lh->mlh_reg_mode == LCK_EX)) {
+                OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_PDO_LOCK, 15);
+        }
 
         RETURN(rc);
 }
@@ -2539,10 +2588,25 @@ static int mdt_req_handle(struct mdt_thread_info *info,
 
                 dlm_req = req_capsule_client_get(info->mti_pill, &RMF_DLM_REQ);
                 if (dlm_req != NULL) {
-                        if (info->mti_mdt->mdt_opts.mo_compat_resname)
-                                rc = mdt_lock_resname_compat(info->mti_mdt,
-                                                             dlm_req);
-                        info->mti_dlm_req = dlm_req;
+                        if (unlikely(dlm_req->lock_desc.l_resource.lr_type ==
+                                        LDLM_IBITS &&
+                                     dlm_req->lock_desc.l_policy_data.\
+                                        l_inodebits.bits == 0)) {
+                                /*
+                                 * Lock without inodebits makes no sense and
+                                 * will oops later in ldlm. If client miss to
+                                 * set such bits, do not trigger ASSERTION.
+                                 *
+                                 * For liblustre flock case, it maybe zero.
+                                 */
+                                rc = -EPROTO;
+                        } else {
+                                if (info->mti_mdt->mdt_opts.mo_compat_resname)
+                                        rc = mdt_lock_resname_compat(
+                                                                info->mti_mdt,
+                                                                dlm_req);
+                                info->mti_dlm_req = dlm_req;
+                        }
                 } else {
                         rc = -EFAULT;
                 }
@@ -2599,7 +2663,7 @@ static int mdt_req_handle(struct mdt_thread_info *info,
         if (likely(rc == 0 && req->rq_export && h->mh_opc != MDS_DISCONNECT))
                 target_committed_to_req(req);
 
-        if (unlikely((lustre_msg_get_flags(req->rq_reqmsg) & MSG_REPLAY) &&
+        if (unlikely(req_is_replay(req) &&
                      lustre_msg_get_transno(req->rq_reqmsg) == 0)) {
                 DEBUG_REQ(D_ERROR, req, "transno is 0 during REPLAY");
                 LBUG();
@@ -2663,10 +2727,7 @@ static void mdt_thread_info_init(struct ptlrpc_request *req,
 
         info->mti_fail_id = OBD_FAIL_MDS_ALL_REPLY_NET;
         info->mti_transno = lustre_msg_get_transno(req->rq_reqmsg);
-        info->mti_mos[0] = NULL;
-        info->mti_mos[1] = NULL;
-        info->mti_mos[2] = NULL;
-        info->mti_mos[3] = NULL;
+        info->mti_mos = NULL;
 
         memset(&info->mti_attr, 0, sizeof(info->mti_attr));
         info->mti_body = NULL;
@@ -2739,7 +2800,6 @@ static int mdt_filter_recovery_request(struct ptlrpc_request *req,
 static int mdt_recovery(struct mdt_thread_info *info)
 {
         struct ptlrpc_request *req = mdt_info_req(info);
-        int recovering;
         struct obd_device *obd;
 
         ENTRY;
@@ -2762,7 +2822,7 @@ static int mdt_recovery(struct mdt_thread_info *info)
                 }
         }
 
-        if (unlikely(req->rq_export == NULL)) {
+        if (unlikely(!class_connected_export(req->rq_export))) {
                 CERROR("operation %d on unconnected MDS from %s\n",
                        lustre_msg_get_opc(req->rq_reqmsg),
                        libcfs_id2str(req->rq_peer));
@@ -2800,10 +2860,7 @@ static int mdt_recovery(struct mdt_thread_info *info)
         obd = req->rq_export->exp_obd;
 
         /* Check for aborted recovery... */
-        spin_lock_bh(&obd->obd_processing_task_lock);
-        recovering = obd->obd_recovering;
-        spin_unlock_bh(&obd->obd_processing_task_lock);
-        if (unlikely(recovering)) {
+        if (unlikely(obd->obd_recovering)) {
                 int rc;
                 int should_process;
                 DEBUG_REQ(D_INFO, req, "Got new replay");
@@ -2826,7 +2883,6 @@ static int mdt_msg_check_version(struct lustre_msg *msg)
         switch (lustre_msg_get_opc(msg)) {
         case MDS_CONNECT:
         case MDS_DISCONNECT:
-        case MDS_SET_INFO:
         case OBD_PING:
         case SEC_CTX_INIT:
         case SEC_CTX_INIT_CONT:
@@ -2852,6 +2908,7 @@ static int mdt_msg_check_version(struct lustre_msg *msg)
         case MDS_SYNC:
         case MDS_GETXATTR:
         case MDS_SETXATTR:
+        case MDS_SET_INFO:
         case MDS_GET_INFO:
         case MDS_QUOTACHECK:
         case MDS_QUOTACTL:
@@ -3034,6 +3091,7 @@ enum mdt_it_code {
         MDT_IT_UNLINK,
         MDT_IT_TRUNC,
         MDT_IT_GETXATTR,
+        MDT_IT_LAYOUT,
         MDT_IT_NR
 };
 
@@ -3104,6 +3162,11 @@ static struct mdt_it_flavor {
                 .it_fmt   = NULL,
                 .it_flags = 0,
                 .it_act   = NULL
+        },
+        [MDT_IT_LAYOUT] = {
+                .it_fmt   = &RQF_LDLM_INTENT_GETATTR,
+                .it_flags = HABEO_REFERO,
+                .it_act   = mdt_intent_getattr
         }
 };
 
@@ -3174,7 +3237,7 @@ int mdt_intent_lock_replace(struct mdt_thread_info *info,
                 new_lock->l_writers--;
         }
 
-        new_lock->l_export = class_export_lock_get(req->rq_export);
+        new_lock->l_export = class_export_lock_get(req->rq_export, new_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;
@@ -3182,9 +3245,9 @@ int mdt_intent_lock_replace(struct mdt_thread_info *info,
 
         unlock_res_and_lock(new_lock);
 
-        lustre_hash_add(new_lock->l_export->exp_lock_hash,
-                        &new_lock->l_remote_handle,
-                        &new_lock->l_exp_hash);
+        cfs_hash_add(new_lock->l_export->exp_lock_hash,
+                     &new_lock->l_remote_handle,
+                     &new_lock->l_exp_hash);
 
         LDLM_LOCK_RELEASE(new_lock);
         lh->mlh_reg_lh.cookie = 0;
@@ -3209,7 +3272,7 @@ 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];
 
-        lock = lustre_hash_lookup(exp->exp_lock_hash, &remote_hdl);
+        lock = cfs_hash_lookup(exp->exp_lock_hash, &remote_hdl);
         if (lock) {
                 if (lock != new_lock) {
                         lh->mlh_reg_lh.cookie = lock->l_handle.h_cookie;
@@ -3221,11 +3284,11 @@ static void mdt_intent_fixup_resent(struct mdt_thread_info *info,
                                   lh->mlh_reg_lh.cookie);
                         if (old_lock)
                                 *old_lock = LDLM_LOCK_GET(lock);
-                        lh_put(exp->exp_lock_hash, &lock->l_exp_hash);
+                        cfs_hash_put(exp->exp_lock_hash, &lock->l_exp_hash);
                         return;
                 }
 
-                lh_put(exp->exp_lock_hash, &lock->l_exp_hash);
+                cfs_hash_put(exp->exp_lock_hash, &lock->l_exp_hash);
         }
 
         /*
@@ -3279,8 +3342,18 @@ static int mdt_intent_getattr(enum mdt_it_code opcode,
         case MDT_IT_GETATTR:
                 child_bits = MDS_INODELOCK_LOOKUP | MDS_INODELOCK_UPDATE;
                 break;
+        case MDT_IT_LAYOUT: {
+                static int printed = 0;
+
+                if (!printed) {
+                        CERROR("layout lock not supported by this version\n");
+                        printed = 1;
+                }
+                GOTO(out_shrink, rc = -EINVAL);
+                break;
+        }
         default:
-                CERROR("Unhandled till now");
+                CERROR("Unsupported intent (%d)\n", opcode);
                 GOTO(out_shrink, rc = -EINVAL);
         }
 
@@ -3311,6 +3384,7 @@ static int mdt_intent_getattr(enum mdt_it_code opcode,
 out_ucred:
         mdt_exit_ucred(info);
 out_shrink:
+        mdt_client_compatibility(info);
         mdt_shrink_reply(info);
         return rc;
 }
@@ -3366,6 +3440,10 @@ static int mdt_intent_reint(enum mdt_it_code opcode,
         }
         rep->lock_policy_res2 = clear_serious(rc);
 
+        if (rep->lock_policy_res2 == -ENOENT &&
+            mdt_get_disposition(rep, DISP_LOOKUP_NEG))
+                rep->lock_policy_res2 = 0;
+
         if (rc == -ENOTCONN || rc == -ENODEV ||
             rc == -EOVERFLOW) { /**< if VBR failure then return error */
                 /*
@@ -3386,6 +3464,8 @@ static int mdt_intent_reint(enum mdt_it_code opcode,
                   * checked here.
                   */
                 if (lustre_handle_is_used(&lhc->mlh_reg_lh)) {
+                        LASSERTF(rc == 0, "Error occurred but lock handle "
+                                 "is still in use\n");
                         rep->lock_policy_res2 = 0;
                         rc = mdt_intent_lock_replace(info, lockp, NULL, lhc, flags);
                         RETURN(rc);
@@ -3428,6 +3508,9 @@ static int mdt_intent_code(long itcode)
         case IT_GETXATTR:
                 rc = MDT_IT_GETXATTR;
                 break;
+        case IT_LAYOUT:
+                rc = MDT_IT_LAYOUT;
+                break;
         default:
                 CERROR("Unknown intent opcode: %ld\n", itcode);
                 rc = -EINVAL;
@@ -3495,23 +3578,19 @@ static int mdt_intent_policy(struct ldlm_namespace *ns,
                 req_capsule_extend(pill, &RQF_LDLM_INTENT);
                 it = req_capsule_client_get(pill, &RMF_LDLM_INTENT);
                 if (it != NULL) {
-                        const struct ldlm_request *dlmreq;
-                        __u64 req_bits;
-
                         rc = mdt_intent_opc(it->opc, info, lockp, flags);
                         if (rc == 0)
                                 rc = ELDLM_OK;
 
-                        /*
-                         * Lock without inodebits makes no sense and will oops
+                        /* Lock without inodebits makes no sense and will oops
                          * later in ldlm. Let's check it now to see if we have
-                         * wrong lock from client or bits get corrupted
-                         * somewhere in mdt_intent_opc().
-                         */
-                        dlmreq = info->mti_dlm_req;
-                        req_bits = dlmreq->lock_desc.l_policy_data.l_inodebits.bits;
-                        LASSERT(req_bits != 0);
-
+                         * ibits corrupted somewhere in mdt_intent_opc().
+                         * The case for client miss to set ibits has been
+                         * processed by others. */
+                        LASSERT(ergo(info->mti_dlm_req->lock_desc.l_resource.\
+                                        lr_type == LDLM_IBITS,
+                                     info->mti_dlm_req->lock_desc.\
+                                        l_policy_data.l_inodebits.bits != 0));
                 } else
                         rc = err_serious(-EFAULT);
         } else {
@@ -3524,25 +3603,6 @@ static int mdt_intent_policy(struct ldlm_namespace *ns,
         RETURN(rc);
 }
 
-/*
- * Seq wrappers
- */
-static void mdt_seq_adjust(const struct lu_env *env,
-                          struct mdt_device *m, int lost)
-{
-        struct md_site *ms = mdt_md_site(m);
-        struct lu_seq_range out;
-        ENTRY;
-
-        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(ms->ms_server_seq, NULL, &out, env);
-        }
-        EXIT;
-}
-
 static int mdt_seq_fini(const struct lu_env *env,
                         struct mdt_device *m)
 {
@@ -3865,9 +3925,8 @@ static int mdt_start_ptlrpc_service(struct mdt_device *m)
                  * We'd like to have a mechanism to set this on a per-device
                  * basis, but alas...
                  */
-                .psc_min_threads    = min(max(mdt_num_threads, MDT_MIN_THREADS),
-                                          MDT_MAX_THREADS),
-                .psc_max_threads     = MDT_MAX_THREADS,
+                .psc_min_threads     = mdt_min_threads,
+                .psc_max_threads     = mdt_max_threads,
                 .psc_ctx_tags        = LCT_MD_THREAD
         };
 
@@ -3882,7 +3941,7 @@ static int mdt_start_ptlrpc_service(struct mdt_device *m)
         if (m->mdt_regular_service == NULL)
                 RETURN(-ENOMEM);
 
-        rc = ptlrpc_start_threads(NULL, m->mdt_regular_service);
+        rc = ptlrpc_start_threads(m->mdt_regular_service);
         if (rc)
                 GOTO(err_mdt_svc, rc);
 
@@ -3898,9 +3957,8 @@ static int mdt_start_ptlrpc_service(struct mdt_device *m)
                 .psc_req_portal      = MDS_READPAGE_PORTAL,
                 .psc_rep_portal      = MDC_REPLY_PORTAL,
                 .psc_watchdog_factor = MDT_SERVICE_WATCHDOG_FACTOR,
-                .psc_min_threads    = min(max(mdt_num_threads, MDT_MIN_THREADS),
-                                          MDT_MAX_THREADS),
-                .psc_max_threads     = MDT_MAX_THREADS,
+                .psc_min_threads     = mdt_min_threads,
+                .psc_max_threads     = mdt_max_threads,
                 .psc_ctx_tags        = LCT_MD_THREAD
         };
         m->mdt_readpage_service =
@@ -3913,10 +3971,14 @@ static int mdt_start_ptlrpc_service(struct mdt_device *m)
                 GOTO(err_mdt_svc, rc = -ENOMEM);
         }
 
-        rc = ptlrpc_start_threads(NULL, m->mdt_readpage_service);
+        rc = ptlrpc_start_threads(m->mdt_readpage_service);
 
         /*
          * setattr service configuration.
+         *
+         * XXX To keep the compatibility with old client(< 2.2), we need to
+         * preserve this portal for a certain time, it should be removed
+         * eventually. LU-617.
          */
         conf = (typeof(conf)) {
                 .psc_nbufs           = MDS_NBUFS,
@@ -3926,9 +3988,8 @@ static int mdt_start_ptlrpc_service(struct mdt_device *m)
                 .psc_req_portal      = MDS_SETATTR_PORTAL,
                 .psc_rep_portal      = MDC_REPLY_PORTAL,
                 .psc_watchdog_factor = MDT_SERVICE_WATCHDOG_FACTOR,
-                .psc_min_threads   = min(max(mdt_num_threads, MDT_MIN_THREADS),
-                                         MDT_MAX_THREADS),
-                .psc_max_threads     = MDT_MAX_THREADS,
+                .psc_min_threads     = mdt_min_threads,
+                .psc_max_threads     = mdt_max_threads,
                 .psc_ctx_tags        = LCT_MD_THREAD
         };
 
@@ -3942,7 +4003,7 @@ static int mdt_start_ptlrpc_service(struct mdt_device *m)
                 GOTO(err_mdt_svc, rc = -ENOMEM);
         }
 
-        rc = ptlrpc_start_threads(NULL, m->mdt_setattr_service);
+        rc = ptlrpc_start_threads(m->mdt_setattr_service);
         if (rc)
                 GOTO(err_mdt_svc, rc);
 
@@ -3957,8 +4018,8 @@ static int mdt_start_ptlrpc_service(struct mdt_device *m)
                 .psc_req_portal      = SEQ_CONTROLLER_PORTAL,
                 .psc_rep_portal      = MDC_REPLY_PORTAL,
                 .psc_watchdog_factor = MDT_SERVICE_WATCHDOG_FACTOR,
-                .psc_min_threads     = SEQ_NUM_THREADS,
-                .psc_max_threads     = SEQ_NUM_THREADS,
+                .psc_min_threads     = mdt_min_threads,
+                .psc_max_threads     = mdt_max_threads,
                 .psc_ctx_tags        = LCT_MD_THREAD|LCT_DT_THREAD
         };
 
@@ -3971,7 +4032,7 @@ static int mdt_start_ptlrpc_service(struct mdt_device *m)
                 GOTO(err_mdt_svc, rc = -ENOMEM);
         }
 
-        rc = ptlrpc_start_threads(NULL, m->mdt_mdsc_service);
+        rc = ptlrpc_start_threads(m->mdt_mdsc_service);
         if (rc)
                 GOTO(err_mdt_svc, rc);
 
@@ -3986,8 +4047,8 @@ static int mdt_start_ptlrpc_service(struct mdt_device *m)
                 .psc_req_portal      = SEQ_METADATA_PORTAL,
                 .psc_rep_portal      = MDC_REPLY_PORTAL,
                 .psc_watchdog_factor = MDT_SERVICE_WATCHDOG_FACTOR,
-                .psc_min_threads     = SEQ_NUM_THREADS,
-                .psc_max_threads     = SEQ_NUM_THREADS,
+                .psc_min_threads     = mdt_min_threads,
+                .psc_max_threads     = mdt_max_threads,
                 .psc_ctx_tags        = LCT_MD_THREAD|LCT_DT_THREAD
         };
 
@@ -4000,7 +4061,7 @@ static int mdt_start_ptlrpc_service(struct mdt_device *m)
                 GOTO(err_mdt_svc, rc = -ENOMEM);
         }
 
-        rc = ptlrpc_start_threads(NULL, m->mdt_mdss_service);
+        rc = ptlrpc_start_threads(m->mdt_mdss_service);
         if (rc)
                 GOTO(err_mdt_svc, rc);
 
@@ -4018,8 +4079,8 @@ static int mdt_start_ptlrpc_service(struct mdt_device *m)
                 .psc_req_portal      = SEQ_DATA_PORTAL,
                 .psc_rep_portal      = OSC_REPLY_PORTAL,
                 .psc_watchdog_factor = MDT_SERVICE_WATCHDOG_FACTOR,
-                .psc_min_threads     = SEQ_NUM_THREADS,
-                .psc_max_threads     = SEQ_NUM_THREADS,
+                .psc_min_threads     = mdt_min_threads,
+                .psc_max_threads     = mdt_max_threads,
                 .psc_ctx_tags        = LCT_MD_THREAD|LCT_DT_THREAD
         };
 
@@ -4032,7 +4093,7 @@ static int mdt_start_ptlrpc_service(struct mdt_device *m)
                 GOTO(err_mdt_svc, rc = -ENOMEM);
         }
 
-        rc = ptlrpc_start_threads(NULL, m->mdt_dtss_service);
+        rc = ptlrpc_start_threads(m->mdt_dtss_service);
         if (rc)
                 GOTO(err_mdt_svc, rc);
 
@@ -4045,8 +4106,8 @@ static int mdt_start_ptlrpc_service(struct mdt_device *m)
                 .psc_req_portal      = FLD_REQUEST_PORTAL,
                 .psc_rep_portal      = MDC_REPLY_PORTAL,
                 .psc_watchdog_factor = MDT_SERVICE_WATCHDOG_FACTOR,
-                .psc_min_threads     = FLD_NUM_THREADS,
-                .psc_max_threads     = FLD_NUM_THREADS,
+                .psc_min_threads     = mdt_min_threads,
+                .psc_max_threads     = mdt_max_threads,
                 .psc_ctx_tags        = LCT_DT_THREAD|LCT_MD_THREAD
         };
 
@@ -4059,7 +4120,7 @@ static int mdt_start_ptlrpc_service(struct mdt_device *m)
                 GOTO(err_mdt_svc, rc = -ENOMEM);
         }
 
-        rc = ptlrpc_start_threads(NULL, m->mdt_fld_service);
+        rc = ptlrpc_start_threads(m->mdt_fld_service);
         if (rc)
                 GOTO(err_mdt_svc, rc);
 
@@ -4075,9 +4136,8 @@ static int mdt_start_ptlrpc_service(struct mdt_device *m)
                 .psc_req_portal      = MDS_MDS_PORTAL,
                 .psc_rep_portal      = MDC_REPLY_PORTAL,
                 .psc_watchdog_factor = MDT_SERVICE_WATCHDOG_FACTOR,
-                .psc_min_threads    = min(max(mdt_num_threads, MDT_MIN_THREADS),
-                                          MDT_MAX_THREADS),
-                .psc_max_threads     = MDT_MAX_THREADS,
+                .psc_min_threads     = mdt_min_threads,
+                .psc_max_threads     = mdt_max_threads,
                 .psc_ctx_tags        = LCT_MD_THREAD
         };
         m->mdt_xmds_service =
@@ -4086,11 +4146,11 @@ static int mdt_start_ptlrpc_service(struct mdt_device *m)
                                      procfs_entry, target_print_req,"mdt_xmds");
 
         if (m->mdt_xmds_service == NULL) {
-                CERROR("failed to start readpage service\n");
+                CERROR("failed to start xmds service\n");
                 GOTO(err_mdt_svc, rc = -ENOMEM);
         }
 
-        rc = ptlrpc_start_threads(NULL, m->mdt_xmds_service);
+        rc = ptlrpc_start_threads(m->mdt_xmds_service);
         if (rc)
                 GOTO(err_mdt_svc, rc);
 
@@ -4188,8 +4248,11 @@ static struct lu_device *mdt_layer_setup(struct lu_env *env,
         lu_device_get(d);
         lu_ref_add(&d->ld_reference, "lu-stack", &lu_site_init);
 
-        RETURN(d);
+        cfs_spin_lock(&d->ld_site->ls_ld_lock);
+        cfs_list_add_tail(&d->ld_linkage, &d->ld_site->ls_ld_linkage);
+        cfs_spin_unlock(&d->ld_site->ls_ld_lock);
 
+        RETURN(d);
 out_alloc:
         ldt->ldt_ops->ldto_device_free(env, d);
         type->typ_refcnt--;
@@ -4326,6 +4389,9 @@ static void mdt_fini(const struct lu_env *env, struct mdt_device *m)
         mdt_obd_llog_cleanup(obd);
         obd_exports_barrier(obd);
         obd_zombie_barrier();
+
+        mdt_procfs_fini(m);
+
 #ifdef HAVE_QUOTA_SUPPORT
         next->md_ops->mdo_quota.mqo_cleanup(env, next);
 #endif
@@ -4361,16 +4427,6 @@ static void mdt_fini(const struct lu_env *env, struct mdt_device *m)
          */
         mdt_stack_fini(env, m, md2lu_dev(m->mdt_child));
 
-        mdt_procfs_fini(m);
-        if (obd->obd_proc_exports_entry) {
-                lprocfs_remove_proc_entry("clear", obd->obd_proc_exports_entry);
-                obd->obd_proc_exports_entry = NULL;
-        }
-        lprocfs_free_per_client_stats(obd);
-        lprocfs_free_obd_stats(obd);
-        ptlrpc_lprocfs_unregister_obd(obd);
-        lprocfs_obd_cleanup(obd);
-
         if (ls) {
                 struct md_site *mite;
 
@@ -4379,7 +4435,7 @@ static void mdt_fini(const struct lu_env *env, struct mdt_device *m)
                 OBD_FREE_PTR(mite);
                 d->ld_site = NULL;
         }
-        LASSERT(atomic_read(&d->ld_ref) == 0);
+        LASSERT(cfs_atomic_read(&d->ld_ref) == 0);
 
         EXIT;
 }
@@ -4400,63 +4456,19 @@ static int mdt_adapt_sptlrpc_conf(struct obd_device *obd, int initial)
 
         sptlrpc_target_update_exp_flavor(obd, &tmp_rset);
 
-        write_lock(&m->mdt_sptlrpc_lock);
+        cfs_write_lock(&m->mdt_sptlrpc_lock);
         sptlrpc_rule_set_free(&m->mdt_sptlrpc_rset);
         m->mdt_sptlrpc_rset = tmp_rset;
-        write_unlock(&m->mdt_sptlrpc_lock);
+        cfs_write_unlock(&m->mdt_sptlrpc_lock);
 
         return 0;
 }
 
-static void fsoptions_to_mdt_flags(struct mdt_device *m, char *options)
-{
-        char *p = options;
-
-        m->mdt_opts.mo_mds_capa = 1;
-        m->mdt_opts.mo_oss_capa = 1;
-#ifdef CONFIG_FS_POSIX_ACL
-        /* ACLs should be enabled by default (b=13829) */
-        m->mdt_opts.mo_acl = 1;
-        LCONSOLE_INFO("Enabling ACL\n");
-#else
-        m->mdt_opts.mo_acl = 0;
-        LCONSOLE_INFO("Disabling ACL\n");
-#endif
-
-        if (!options)
-                return;
-
-        while (*options) {
-                int len;
-
-                while (*p && *p != ',')
-                        p++;
-
-                len = p - options;
-                if ((len == sizeof("user_xattr") - 1) &&
-                    (memcmp(options, "user_xattr", len) == 0)) {
-                        m->mdt_opts.mo_user_xattr = 1;
-                        LCONSOLE_INFO("Enabling user_xattr\n");
-                } else if ((len == sizeof("nouser_xattr") - 1) &&
-                           (memcmp(options, "nouser_xattr", len) == 0)) {
-                        m->mdt_opts.mo_user_xattr = 0;
-                        LCONSOLE_INFO("Disabling user_xattr\n");
-                } else if ((len == sizeof("noacl") - 1) &&
-                           (memcmp(options, "noacl", len) == 0)) {
-                        m->mdt_opts.mo_acl = 0;
-                        LCONSOLE_INFO("Disabling ACL\n");
-                }
-
-                options = ++p;
-        }
-}
-
 int mdt_postrecov(const struct lu_env *, struct mdt_device *);
 
 static int mdt_init0(const struct lu_env *env, struct mdt_device *m,
                      struct lu_device_type *ldt, struct lustre_cfg *cfg)
 {
-        struct lprocfs_static_vars lvars;
         struct mdt_thread_info    *info;
         struct obd_device         *obd;
         const char                *dev = lustre_cfg_string(cfg, 0);
@@ -4467,11 +4479,10 @@ static int mdt_init0(const struct lu_env *env, struct mdt_device *m,
         struct lu_site            *s;
         struct md_site            *mite;
         const char                *identity_upcall = "NONE";
-#ifdef HAVE_QUOTA_SUPPORT
         struct md_device          *next;
-#endif
         int                        rc;
         int                        node_id;
+        mntopt_t                   mntopts;
         ENTRY;
 
         md_device_init(&m->mdt_md_dev, ldt);
@@ -4493,14 +4504,10 @@ static int mdt_init0(const struct lu_env *env, struct mdt_device *m,
         obd = class_name2obd(dev);
         LASSERT(obd != NULL);
 
-        spin_lock_init(&m->mdt_transno_lock);
-
         m->mdt_max_mdsize = MAX_MD_SIZE;
         m->mdt_max_cookiesize = sizeof(struct llog_cookie);
         m->mdt_som_conf = 0;
 
-        m->mdt_opts.mo_user_xattr = 0;
-        m->mdt_opts.mo_acl = 0;
         m->mdt_opts.mo_cos = MDT_COS_DEFAULT;
         lmi = server_get_mount_2(dev);
         if (lmi == NULL) {
@@ -4508,7 +4515,6 @@ static int mdt_init0(const struct lu_env *env, struct mdt_device *m,
                 RETURN(-EFAULT);
         } else {
                 lsi = s2lsi(lmi->lmi_sb);
-                fsoptions_to_mdt_flags(m, lsi->lsi_lmd->lmd_opts);
                 /* CMD is supported only in IAM mode */
                 ldd = lsi->lsi_ldd;
                 LASSERT(num);
@@ -4517,13 +4523,17 @@ static int mdt_init0(const struct lu_env *env, struct mdt_device *m,
                         CERROR("CMD Operation not allowed in IOP mode\n");
                         GOTO(err_lmi, rc = -EINVAL);
                 }
+
+                obd->u.obt.obt_magic = OBT_MAGIC;
         }
 
-        rwlock_init(&m->mdt_sptlrpc_lock);
+        cfs_rwlock_init(&m->mdt_sptlrpc_lock);
         sptlrpc_rule_set_init(&m->mdt_sptlrpc_rset);
 
-        spin_lock_init(&m->mdt_ioepoch_lock);
+        cfs_spin_lock_init(&m->mdt_ioepoch_lock);
         m->mdt_opts.mo_compat_resname = 0;
+        m->mdt_opts.mo_mds_capa = 1;
+        m->mdt_opts.mo_oss_capa = 1;
         m->mdt_capa_timeout = CAPA_TIMEOUT;
         m->mdt_capa_alg = CAPA_HMAC_ALG_SHA1;
         m->mdt_ck_timeout = CAPA_KEY_TIMEOUT;
@@ -4532,9 +4542,7 @@ static int mdt_init0(const struct lu_env *env, struct mdt_device *m,
         CFS_INIT_LIST_HEAD(&m->mdt_nosquash_nids);
         m->mdt_nosquash_str = NULL;
         m->mdt_nosquash_strlen = 0;
-        init_rwsem(&m->mdt_squash_sem);
-
-        spin_lock_init(&m->mdt_client_bitmap_lock);
+        cfs_init_rwsem(&m->mdt_squash_sem);
 
         OBD_ALLOC_PTR(mite);
         if (mite == NULL)
@@ -4553,27 +4561,6 @@ static int mdt_init0(const struct lu_env *env, struct mdt_device *m,
                 GOTO(err_free_site, rc);
         }
 
-        lprocfs_mdt_init_vars(&lvars);
-        rc = lprocfs_obd_setup(obd, lvars.obd_vars);
-        if (rc) {
-                CERROR("Can't init lprocfs, rc %d\n", rc);
-                GOTO(err_fini_site, rc);
-        }
-        ptlrpc_lprocfs_register_obd(obd);
-
-        rc = mdt_procfs_init(m, dev);
-        if (rc) {
-                CERROR("Can't init MDT lprocfs, rc %d\n", rc);
-                GOTO(err_fini_proc, rc);
-        }
-
-        obd->obd_proc_exports_entry = proc_mkdir("exports",
-                                                 obd->obd_proc_entry);
-        if (obd->obd_proc_exports_entry)
-                lprocfs_add_simple(obd->obd_proc_exports_entry,
-                                   "clear", lprocfs_nid_stats_clear_read,
-                                   lprocfs_nid_stats_clear_write, obd, NULL);
-
         /* set server index */
         lu_site2md(s)->ms_node_id = node_id;
 
@@ -4597,7 +4584,7 @@ static int mdt_init0(const struct lu_env *env, struct mdt_device *m,
         rc = mdt_stack_init((struct lu_env *)env, m, cfg, lmi);
         if (rc) {
                 CERROR("Can't init device stack, rc %d\n", rc);
-                GOTO(err_fini_proc, rc);
+                GOTO(err_lu_site, rc);
         }
 
         rc = lut_init(env, &m->mdt_lut, obd, m->mdt_bottom);
@@ -4616,7 +4603,8 @@ static int mdt_init0(const struct lu_env *env, struct mdt_device *m,
                  LUSTRE_MDT_NAME"-%p", m);
         m->mdt_namespace = ldlm_namespace_new(obd, info->mti_u.ns_name,
                                               LDLM_NAMESPACE_SERVER,
-                                              LDLM_NAMESPACE_GREEDY);
+                                              LDLM_NAMESPACE_GREEDY,
+                                              LDLM_NS_TYPE_MDT);
         if (m->mdt_namespace == NULL)
                 GOTO(err_fini_seq, rc = -ENOMEM);
 
@@ -4624,19 +4612,6 @@ static int mdt_init0(const struct lu_env *env, struct mdt_device *m,
         /* set obd_namespace for compatibility with old code */
         obd->obd_namespace = m->mdt_namespace;
 
-        /* XXX: to support suppgid for ACL, we enable identity_upcall
-         * by default, otherwise, maybe got unexpected -EACCESS. */
-        if (m->mdt_opts.mo_acl)
-                identity_upcall = MDT_IDENTITY_UPCALL_PATH;
-
-        m->mdt_identity_cache = upcall_cache_init(obd->obd_name, identity_upcall,
-                                                  &mdt_identity_upcall_cache_ops);
-        if (IS_ERR(m->mdt_identity_cache)) {
-                rc = PTR_ERR(m->mdt_identity_cache);
-                m->mdt_identity_cache = NULL;
-                GOTO(err_free_ns, rc);
-        }
-
         cfs_timer_init(&m->mdt_ck_timer, mdt_ck_timer_callback, m);
 
         rc = mdt_ck_thread_start(m);
@@ -4657,8 +4632,8 @@ static int mdt_init0(const struct lu_env *env, struct mdt_device *m,
 
         mdt_adapt_sptlrpc_conf(obd, 1);
 
-#ifdef HAVE_QUOTA_SUPPORT
         next = m->mdt_child;
+#ifdef HAVE_QUOTA_SUPPORT
         rc = next->md_ops->mdo_quota.mqo_setup(env, next, lmi->lmi_mnt);
         if (rc)
                 GOTO(err_llog_cleanup, rc);
@@ -4667,11 +4642,45 @@ static int mdt_init0(const struct lu_env *env, struct mdt_device *m,
         server_put_mount_2(dev, lmi->lmi_mnt);
         lmi = NULL;
 
+        rc = next->md_ops->mdo_iocontrol(env, next, OBD_IOC_GET_MNTOPT, 0,
+                                         &mntopts);
+        if (rc)
+                GOTO(err_quota, rc);
+
+        if (mntopts & MNTOPT_USERXATTR)
+                m->mdt_opts.mo_user_xattr = 1;
+        else
+                m->mdt_opts.mo_user_xattr = 0;
+
+        if (mntopts & MNTOPT_ACL)
+                m->mdt_opts.mo_acl = 1;
+        else
+                m->mdt_opts.mo_acl = 0;
+
+        /* XXX: to support suppgid for ACL, we enable identity_upcall
+         * by default, otherwise, maybe got unexpected -EACCESS. */
+        if (m->mdt_opts.mo_acl)
+                identity_upcall = MDT_IDENTITY_UPCALL_PATH;
+
+        m->mdt_identity_cache = upcall_cache_init(obd->obd_name,identity_upcall,
+                                                &mdt_identity_upcall_cache_ops);
+        if (IS_ERR(m->mdt_identity_cache)) {
+                rc = PTR_ERR(m->mdt_identity_cache);
+                m->mdt_identity_cache = NULL;
+                GOTO(err_quota, rc);
+        }
+
         target_recovery_init(&m->mdt_lut, mdt_recovery_handle);
 
+        rc = mdt_procfs_init(m, dev);
+        if (rc) {
+                CERROR("Can't init MDT lprocfs, rc %d\n", rc);
+                GOTO(err_recovery, rc);
+        }
+
         rc = mdt_start_ptlrpc_service(m);
         if (rc)
-                GOTO(err_recovery, rc);
+                GOTO(err_procfs, rc);
 
         ping_evictor_start();
 
@@ -4695,8 +4704,13 @@ static int mdt_init0(const struct lu_env *env, struct mdt_device *m,
 err_stop_service:
         ping_evictor_stop();
         mdt_stop_ptlrpc_service(m);
+err_procfs:
+        mdt_procfs_fini(m);
 err_recovery:
         target_recovery_fini(obd);
+        upcall_cache_cleanup(m->mdt_identity_cache);
+        m->mdt_identity_cache = NULL;
+err_quota:
 #ifdef HAVE_QUOTA_SUPPORT
         next->md_ops->mdo_quota.mqo_cleanup(env, next);
 #endif
@@ -4709,8 +4723,6 @@ err_capa:
         cfs_timer_disarm(&m->mdt_ck_timer);
         mdt_ck_thread_stop(m);
 err_free_ns:
-        upcall_cache_cleanup(m->mdt_identity_cache);
-        m->mdt_identity_cache = NULL;
         ldlm_namespace_free(m->mdt_namespace, NULL, 0);
         obd->obd_namespace = m->mdt_namespace = NULL;
 err_fini_seq:
@@ -4721,15 +4733,7 @@ err_lut:
         lut_fini(env, &m->mdt_lut);
 err_fini_stack:
         mdt_stack_fini(env, m, md2lu_dev(m->mdt_child));
-err_fini_proc:
-        mdt_procfs_fini(m);
-        if (obd->obd_proc_exports_entry) {
-                lprocfs_remove_proc_entry("clear", obd->obd_proc_exports_entry);
-                obd->obd_proc_exports_entry = NULL;
-        }
-        ptlrpc_lprocfs_unregister_obd(obd);
-        lprocfs_obd_cleanup(obd);
-err_fini_site:
+err_lu_site:
         lu_site_fini(s);
 err_free_site:
         OBD_FREE_PTR(mite);
@@ -4756,15 +4760,21 @@ static int mdt_process_config(const struct lu_env *env,
 
                 /*
                  * For interoperability between 1.8 and 2.0,
-                 * skip old "mdt.group_upcall" param.
                  */
                 {
+                        /* Skip old "mdt.group_upcall" param. */
                         char *param = lustre_cfg_string(cfg, 1);
                         if (param && !strncmp("mdt.group_upcall", param, 16)) {
                                 CWARN("For 1.8 interoperability, skip this"
                                        " mdt.group_upcall. It is obsolete\n");
                                 break;
                         }
+                        /* Rename old "mdt.quota_type" to "mdd.quota_type. */
+                        if (param && !strncmp("mdt.quota_type", param, 14)) {
+                                CWARN("Found old param mdt.quota_type, changed"
+                                      " it to mdd.quota_type.\n");
+                                param[2] = 'd';
+                        }
                 }
 
                 lprocfs_mdt_init_vars(&lvars);
@@ -4813,6 +4823,8 @@ static struct lu_object *mdt_object_alloc(const struct lu_env *env,
                 lu_object_init(o, h, d);
                 lu_object_add_top(h, o);
                 o->lo_ops = &mdt_obj_ops;
+                cfs_mutex_init(&mo->mot_ioepoch_mutex);
+                cfs_mutex_init(&mo->mot_lov_mutex);
                 RETURN(o);
         } else
                 RETURN(NULL);
@@ -4860,10 +4872,10 @@ static int mdt_object_print(const struct lu_env *env, void *cookie,
                             lu_printer_t p, const struct lu_object *o)
 {
         struct mdt_object *mdto = mdt_obj((struct lu_object *)o);
-        return (*p)(env, cookie, LUSTRE_MDT_NAME"-object@%p(ioepoch=%llu "
-                    "flags=%llx, epochcount=%d, writecount=%d)",
+        return (*p)(env, cookie, LUSTRE_MDT_NAME"-object@%p(ioepoch="LPU64" "
+                    "flags="LPX64", epochcount=%d, writecount=%d)",
                     mdto, mdto->mot_ioepoch, mdto->mot_flags,
-                    mdto->mot_epochcount, mdto->mot_writecount);
+                    mdto->mot_ioepoch_count, mdto->mot_writecount);
 }
 
 static const struct lu_device_operations mdt_lu_ops = {
@@ -4921,9 +4933,27 @@ static int mdt_connect_internal(struct obd_export *exp,
                 if (!mdt->mdt_som_conf)
                         data->ocd_connect_flags &= ~OBD_CONNECT_SOM;
 
-                spin_lock(&exp->exp_lock);
+                if (data->ocd_connect_flags & OBD_CONNECT_BRW_SIZE) {
+                        data->ocd_brw_size = min(data->ocd_brw_size,
+                               (__u32)(PTLRPC_MAX_BRW_PAGES << CFS_PAGE_SHIFT));
+                        if (data->ocd_brw_size == 0) {
+                                CERROR("%s: cli %s/%p ocd_connect_flags: "LPX64
+                                       " ocd_version: %x ocd_grant: %d "
+                                       "ocd_index: %u ocd_brw_size is "
+                                       "unexpectedly zero, network data "
+                                       "corruption? Refusing connection of this"
+                                       " client\n",
+                                       exp->exp_obd->obd_name,
+                                       exp->exp_client_uuid.uuid,
+                                       exp, data->ocd_connect_flags, data->ocd_version,
+                                       data->ocd_grant, data->ocd_index);
+                                return -EPROTO;
+                        }
+                }
+
+                cfs_spin_lock(&exp->exp_lock);
                 exp->exp_connect_flags = data->ocd_connect_flags;
-                spin_unlock(&exp->exp_lock);
+                cfs_spin_unlock(&exp->exp_lock);
                 data->ocd_version = LUSTRE_VERSION_CODE;
                 exp->exp_mdt_data.med_ibits_known = data->ocd_ibits_known;
         }
@@ -4943,9 +4973,8 @@ static int mdt_connect_internal(struct obd_export *exp,
                 return -EBADE;
         }
 
-        if (mdt->mdt_som_conf &&
-            !(exp->exp_connect_flags & OBD_CONNECT_MDS_MDS) &&
-            !(exp->exp_connect_flags & OBD_CONNECT_SOM)) {
+        if (mdt->mdt_som_conf && !exp_connect_som(exp) &&
+            !(exp->exp_connect_flags & OBD_CONNECT_MDS_MDS)) {
                 CWARN("%s: MDS has SOM enabled, but client does not support "
                       "it\n", mdt->mdt_md_dev.md_lu_dev.ld_obd->obd_name);
                 return -EBADE;
@@ -4962,14 +4991,14 @@ static int mdt_connect_check_sptlrpc(struct mdt_device *mdt,
         int                     rc = 0;
 
         if (exp->exp_flvr.sf_rpc == SPTLRPC_FLVR_INVALID) {
-                read_lock(&mdt->mdt_sptlrpc_lock);
+                cfs_read_lock(&mdt->mdt_sptlrpc_lock);
                 sptlrpc_target_choose_flavor(&mdt->mdt_sptlrpc_rset,
                                              req->rq_sp_from,
                                              req->rq_peer.nid,
                                              &flvr);
-                read_unlock(&mdt->mdt_sptlrpc_lock);
+                cfs_read_unlock(&mdt->mdt_sptlrpc_lock);
 
-                spin_lock(&exp->exp_lock);
+                cfs_spin_lock(&exp->exp_lock);
 
                 exp->exp_sp_peer = req->rq_sp_from;
                 exp->exp_flvr = flvr;
@@ -4983,7 +5012,7 @@ static int mdt_connect_check_sptlrpc(struct mdt_device *mdt,
                         rc = -EACCES;
                 }
 
-                spin_unlock(&exp->exp_lock);
+                cfs_spin_unlock(&exp->exp_lock);
         } else {
                 if (exp->exp_sp_peer != req->rq_sp_from) {
                         CERROR("RPC source %s doesn't match %s\n",
@@ -5006,7 +5035,6 @@ static int mdt_obd_connect(const struct lu_env *env,
                            void *localdata)
 {
         struct mdt_thread_info *info;
-        struct lsd_client_data *lcd;
         struct obd_export      *lexp;
         struct lustre_handle    conn = { 0 };
         struct mdt_device      *mdt;
@@ -5033,33 +5061,30 @@ static int mdt_obd_connect(const struct lu_env *env,
         if (rc)
                 GOTO(out, rc);
 
+        if (OBD_FAIL_CHECK(OBD_FAIL_TGT_RCVG_FLAG))
+                lustre_msg_add_op_flags(req->rq_repmsg, MSG_CONNECT_RECOVERING);
+
         rc = mdt_connect_internal(lexp, mdt, data);
         if (rc == 0) {
-                OBD_ALLOC_PTR(lcd);
-                if (lcd != NULL) {
-                        struct mdt_thread_info *mti;
-                        mti = lu_context_key_get(&env->le_ctx,
-                                                 &mdt_thread_key);
-                        LASSERT(mti != NULL);
-                        mti->mti_exp = lexp;
-                        memcpy(lcd->lcd_uuid, cluuid, sizeof lcd->lcd_uuid);
-                        lexp->exp_mdt_data.med_lcd = lcd;
-                        rc = mdt_client_new(env, mdt);
-                        if (rc != 0) {
-                                OBD_FREE_PTR(lcd);
-                                lexp->exp_mdt_data.med_lcd = NULL;
-                        } else {
-                                mdt_export_stats_init(obd, lexp, localdata);
-                        }
-                } else
-                        rc = -ENOMEM;
+                struct mdt_thread_info *mti;
+                struct lsd_client_data *lcd = lexp->exp_target_data.ted_lcd;
+                LASSERT(lcd);
+                mti = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
+                LASSERT(mti != NULL);
+                mti->mti_exp = lexp;
+                memcpy(lcd->lcd_uuid, cluuid, sizeof lcd->lcd_uuid);
+                rc = mdt_client_new(env, mdt);
+                if (rc == 0)
+                        mdt_export_stats_init(obd, lexp, localdata);
         }
 
 out:
-        if (rc != 0)
+        if (rc != 0) {
                 class_disconnect(lexp);
-        else
+                *exp = NULL;
+        } else {
                 *exp = lexp;
+        }
 
         RETURN(rc);
 }
@@ -5093,7 +5118,7 @@ static int mdt_obd_reconnect(const struct lu_env *env,
 
         RETURN(rc);
 }
-static int mdt_mfd_cleanup(struct obd_export *exp)
+static int mdt_export_cleanup(struct obd_export *exp)
 {
         struct mdt_export_data *med = &exp->exp_mdt_data;
         struct obd_device      *obd = exp->exp_obd;
@@ -5105,17 +5130,17 @@ static int mdt_mfd_cleanup(struct obd_export *exp)
         int rc = 0;
         ENTRY;
 
-        spin_lock(&med->med_open_lock);
-        while (!list_empty(&med->med_open_head)) {
-                struct list_head *tmp = med->med_open_head.next;
-                mfd = list_entry(tmp, struct mdt_file_data, mfd_list);
+        cfs_spin_lock(&med->med_open_lock);
+        while (!cfs_list_empty(&med->med_open_head)) {
+                cfs_list_t *tmp = med->med_open_head.next;
+                mfd = cfs_list_entry(tmp, struct mdt_file_data, mfd_list);
 
                 /* Remove mfd handle so it can't be found again.
                  * We are consuming the mfd_list reference here. */
                 class_handle_unhash(&mfd->mfd_handle);
-                list_move_tail(&mfd->mfd_list, &closing_list);
+                cfs_list_move_tail(&mfd->mfd_list, &closing_list);
         }
-        spin_unlock(&med->med_open_lock);
+        cfs_spin_unlock(&med->med_open_lock);
         mdt = mdt_dev(obd->obd_lu_dev);
         LASSERT(mdt != NULL);
 
@@ -5130,42 +5155,49 @@ static int mdt_mfd_cleanup(struct obd_export *exp)
         info->mti_mdt = mdt;
         info->mti_exp = exp;
 
-        if (!list_empty(&closing_list)) {
+        if (!cfs_list_empty(&closing_list)) {
                 struct md_attr *ma = &info->mti_attr;
                 int lmm_size;
                 int cookie_size;
 
                 lmm_size = mdt->mdt_max_mdsize;
-                OBD_ALLOC(ma->ma_lmm, lmm_size);
+                OBD_ALLOC_LARGE(ma->ma_lmm, lmm_size);
                 if (ma->ma_lmm == NULL)
                         GOTO(out_lmm, rc = -ENOMEM);
 
                 cookie_size = mdt->mdt_max_cookiesize;
-                OBD_ALLOC(ma->ma_cookie, cookie_size);
+                OBD_ALLOC_LARGE(ma->ma_cookie, cookie_size);
                 if (ma->ma_cookie == NULL)
                         GOTO(out_cookie, rc = -ENOMEM);
 
                 /* Close any open files (which may also cause orphan unlinking). */
-                list_for_each_entry_safe(mfd, n, &closing_list, mfd_list) {
-                        list_del_init(&mfd->mfd_list);
+                cfs_list_for_each_entry_safe(mfd, n, &closing_list, mfd_list) {
+                        cfs_list_del_init(&mfd->mfd_list);
                         memset(&ma->ma_attr, 0, sizeof(ma->ma_attr));
                         ma->ma_lmm_size = lmm_size;
                         ma->ma_cookie_size = cookie_size;
                         ma->ma_need = 0;
                         /* It is not for setattr, just tell MDD to send
                          * DESTROY RPC to OSS if needed */
-                        ma->ma_attr_flags = MDS_CLOSE_CLEANUP;
                         ma->ma_valid = MA_FLAGS;
+                        ma->ma_attr_flags = MDS_CLOSE_CLEANUP;
+                        /* Don't unlink orphan on failover umount, LU-184 */
+                        if (exp->exp_flags & OBD_OPT_FAILOVER)
+                                ma->ma_attr_flags |= MDS_KEEP_ORPHAN;
                         mdt_mfd_close(info, mfd);
                 }
-                info->mti_mdt = NULL;
-                OBD_FREE(ma->ma_cookie, cookie_size);
+                OBD_FREE_LARGE(ma->ma_cookie, cookie_size);
                 ma->ma_cookie = NULL;
 out_cookie:
-                OBD_FREE(ma->ma_lmm, lmm_size);
+                OBD_FREE_LARGE(ma->ma_lmm, lmm_size);
                 ma->ma_lmm = NULL;
         }
 out_lmm:
+        info->mti_mdt = NULL;
+        /* cleanup client slot early */
+        /* Do not erase record for recoverable client. */
+        if (!(exp->exp_flags & OBD_OPT_FAILOVER) || exp->exp_failed)
+                mdt_client_del(&env, mdt);
         lu_env_fini(&env);
 
         RETURN(rc);
@@ -5183,7 +5215,7 @@ static int mdt_obd_disconnect(struct obd_export *exp)
         if (rc != 0)
                 CDEBUG(D_IOCTL, "server disconnect error: %d\n", rc);
 
-        rc = mdt_mfd_cleanup(exp);
+        rc = mdt_export_cleanup(exp);
         class_export_put(exp);
         RETURN(rc);
 }
@@ -5196,71 +5228,70 @@ static int mdt_init_export(struct obd_export *exp)
         ENTRY;
 
         CFS_INIT_LIST_HEAD(&med->med_open_head);
-        spin_lock_init(&med->med_open_lock);
-        sema_init(&med->med_idmap_sem, 1);
+        cfs_spin_lock_init(&med->med_open_lock);
+        cfs_mutex_init(&med->med_idmap_mutex);
         med->med_idmap = NULL;
-        spin_lock(&exp->exp_lock);
+        cfs_spin_lock(&exp->exp_lock);
         exp->exp_connecting = 1;
-        spin_unlock(&exp->exp_lock);
-        rc = ldlm_init_export(exp);
+        cfs_spin_unlock(&exp->exp_lock);
+
+        /* self-export doesn't need client data and ldlm initialization */
+        if (unlikely(obd_uuid_equals(&exp->exp_obd->obd_uuid,
+                                     &exp->exp_client_uuid)))
+                RETURN(0);
+
+        rc = lut_client_alloc(exp);
+        if (rc == 0)
+                rc = ldlm_init_export(exp);
         if (rc)
-                CERROR("Error %d while initializing export\n", rc);
+                CERROR("%s: Error %d while initializing export\n",
+                       exp->exp_obd->obd_name, rc);
         RETURN(rc);
 }
 
-static int mdt_destroy_export(struct obd_export *export)
+static int mdt_destroy_export(struct obd_export *exp)
 {
-        struct mdt_export_data *med;
-        struct obd_device      *obd = export->exp_obd;
-        struct mdt_device      *mdt;
-        struct mdt_thread_info *info;
-        struct lu_env           env;
-        int rc = 0;
         ENTRY;
 
-        med = &export->exp_mdt_data;
-        if (exp_connect_rmtclient(export))
-                mdt_cleanup_idmap(med);
+        if (exp_connect_rmtclient(exp))
+                mdt_cleanup_idmap(&exp->exp_mdt_data);
 
-        target_destroy_export(export);
-        ldlm_destroy_export(export);
-
-        LASSERT(list_empty(&export->exp_outstanding_replies));
-        LASSERT(list_empty(&med->med_open_head));
-        if (obd_uuid_equals(&export->exp_client_uuid, &obd->obd_uuid))
+        target_destroy_export(exp);
+        /* destroy can be called from failed obd_setup, so
+         * checking uuid is safer than obd_self_export */
+        if (unlikely(obd_uuid_equals(&exp->exp_obd->obd_uuid,
+                                     &exp->exp_client_uuid)))
                 RETURN(0);
 
-        mdt = mdt_dev(obd->obd_lu_dev);
-        LASSERT(mdt != NULL);
-
-        rc = lu_env_init(&env, LCT_MD_THREAD);
-        if (rc)
-                RETURN(rc);
+        ldlm_destroy_export(exp);
+        lut_client_free(exp);
 
-        info = lu_context_key_get(&env.le_ctx, &mdt_thread_key);
-        LASSERT(info != NULL);
-        memset(info, 0, sizeof *info);
-        info->mti_env = &env;
-        info->mti_exp = export;
-        info->mti_mdt = NULL;
-        mdt_client_del(&env, mdt);
+        LASSERT(cfs_list_empty(&exp->exp_outstanding_replies));
+        LASSERT(cfs_list_empty(&exp->exp_mdt_data.med_open_head));
 
-        lu_env_fini(&env);
-        RETURN(rc);
+        RETURN(0);
 }
 
 static void mdt_allow_cli(struct mdt_device *m, unsigned int flag)
 {
         if (flag & CONFIG_LOG)
-                m->mdt_fl_cfglog = 1;
+                cfs_set_bit(MDT_FL_CFGLOG, &m->mdt_state);
 
         /* also notify active event */
         if (flag & CONFIG_SYNC)
-                m->mdt_fl_synced = 1;
+                cfs_set_bit(MDT_FL_SYNCED, &m->mdt_state);
+
+        if (cfs_test_bit(MDT_FL_CFGLOG, &m->mdt_state) &&
+            cfs_test_bit(MDT_FL_SYNCED, &m->mdt_state)) {
+                struct obd_device *obd = m->mdt_md_dev.md_lu_dev.ld_obd;
 
-        if (m->mdt_fl_cfglog && m->mdt_fl_synced)
                 /* Open for clients */
-                m->mdt_md_dev.md_lu_dev.ld_obd->obd_no_conn = 0;
+                if (obd->obd_no_conn) {
+                        cfs_spin_lock(&obd->obd_dev_lock);
+                        obd->obd_no_conn = 0;
+                        cfs_spin_unlock(&obd->obd_dev_lock);
+                }
+        }
 }
 
 static int mdt_upcall(const struct lu_env *env, struct md_device *md,
@@ -5281,7 +5312,8 @@ static int mdt_upcall(const struct lu_env *env, struct md_device *md,
                                      m->mdt_max_mdsize, m->mdt_max_cookiesize);
                         mdt_allow_cli(m, CONFIG_SYNC);
                         if (data)
-                                (*(__u64 *)data) = m->mdt_mount_count;
+                                (*(__u64 *)data) =
+                                      m->mdt_lut.lut_obd->u.obt.obt_mount_count;
                         break;
                 case MD_NO_TRANS:
                         mti = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
@@ -5290,7 +5322,7 @@ static int mdt_upcall(const struct lu_env *env, struct md_device *md,
                         break;
                 case MD_LOV_CONFIG:
                         /* Check that MDT is not yet configured */
-                        LASSERT(!m->mdt_fl_cfglog);
+                        LASSERT(!cfs_test_bit(MDT_FL_CFGLOG, &m->mdt_state));
                         break;
 #ifdef HAVE_QUOTA_SUPPORT
                 case MD_LOV_QUOTA:
@@ -5307,11 +5339,11 @@ static int mdt_upcall(const struct lu_env *env, struct md_device *md,
         RETURN(rc);
 }
 
-static int mdt_obd_notify(struct obd_device *host,
+static int mdt_obd_notify(struct obd_device *obd,
                           struct obd_device *watched,
                           enum obd_notify_event ev, void *data)
 {
-        struct mdt_device *mdt = mdt_dev(host->obd_lu_dev);
+        struct mdt_device *mdt = mdt_dev(obd->obd_lu_dev);
 #ifdef HAVE_QUOTA_SUPPORT
         struct md_device *next = mdt->mdt_child;
 #endif
@@ -5340,7 +5372,7 @@ static int mdt_rpc_fid2path(struct mdt_thread_info *info, void *key,
         struct getinfo_fid2path *fpout, *fpin;
         int rc = 0;
 
-        fpin = key + size_round(sizeof(KEY_FID2PATH));
+        fpin = key + cfs_size_round(sizeof(KEY_FID2PATH));
         fpout = val;
 
         if (ptlrpc_req_need_swab(info->mti_pill->rc_req))
@@ -5369,7 +5401,7 @@ static int mdt_fid2path(const struct lu_env *env, struct mdt_device *mdt,
 
         obj = mdt_object_find(env, mdt, &fp->gf_fid);
         if (obj == NULL || IS_ERR(obj)) {
-                CDEBUG(D_IOCTL, "no object "DFID": %ld\n",PFID(&fp->gf_fid),
+                CDEBUG(D_IOCTL, "no object "DFID": %ld\n", PFID(&fp->gf_fid),
                        PTR_ERR(obj));
                 RETURN(-EINVAL);
         }
@@ -5469,6 +5501,7 @@ static int mdt_ioc_version_get(struct mdt_thread_info *mti, void *karg)
         struct mdt_lock_handle  *lh;
         int rc;
         ENTRY;
+
         CDEBUG(D_IOCTL, "getting version for "DFID"\n", PFID(fid));
         if (!fid_is_sane(fid))
                 RETURN(-EINVAL);
@@ -5489,16 +5522,12 @@ static int mdt_ioc_version_get(struct mdt_thread_info *mti, void *karg)
                  */
                 CERROR("nonlocal object "DFID"\n", PFID(fid));
         } else if (rc == 0) {
+                *(__u64 *)data->ioc_inlbuf2 = ENOENT_VERSION;
                 rc = -ENOENT;
-                CDEBUG(D_IOCTL, "no such object: "DFID"\n", PFID(fid));
         } else {
-                version = mo_version_get(mti->mti_env, mdt_object_child(obj));
-                if (version < 0) {
-                        rc = (int)version;
-                } else {
-                        *(__u64 *)data->ioc_inlbuf2 = version;
-                        rc = 0;
-                }
+                version = dt_version_get(mti->mti_env, mdt_obj2dt(obj));
+               *(__u64 *)data->ioc_inlbuf2 = version;
+                rc = 0;
         }
         mdt_object_unlock_put(mti, obj, lh, 1);
         RETURN(rc);
@@ -5525,7 +5554,7 @@ static int mdt_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
                 rc = mdt_device_sync(&env, mdt);
                 break;
         case OBD_IOC_SET_READONLY:
-                dt->dd_ops->dt_ro(&env, dt);
+                rc = dt->dd_ops->dt_ro(&env, dt);
                 break;
         case OBD_IOC_ABORT_RECOVERY:
                 CERROR("Aborting recovery for device %s\n", obd->obd_name);
@@ -5561,16 +5590,12 @@ static int mdt_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
 int mdt_postrecov(const struct lu_env *env, struct mdt_device *mdt)
 {
         struct lu_device *ld = md2lu_dev(mdt->mdt_child);
-        struct obd_device *obd = mdt2obd_dev(mdt);
 #ifdef HAVE_QUOTA_SUPPORT
+        struct obd_device *obd = mdt2obd_dev(mdt);
         struct md_device *next = mdt->mdt_child;
 #endif
-        int rc, lost;
+        int rc;
         ENTRY;
-        /* if some clients didn't participate in recovery then we can possibly
-         * lost sequence. Now we should increase sequence for safe value */
-        lost = obd->obd_max_recoverable_clients - obd->obd_connected_clients;
-        mdt_seq_adjust(env, mdt, lost);
 
         rc = ld->ld_ops->ldo_recovery_complete(env, ld);
 #ifdef HAVE_QUOTA_SUPPORT
@@ -5600,7 +5625,7 @@ int mdt_obd_postrecov(struct obd_device *obd)
  */
 int mdt_hsm_copytool_send(struct obd_export *exp)
 {
-        struct lnl_hdr *lh;
+        struct kuc_hdr *lh;
         struct hsm_action_list *hal;
         struct hsm_action_item *hai;
         int rc, len;
@@ -5610,15 +5635,15 @@ int mdt_hsm_copytool_send(struct obd_export *exp)
               libcfs_nid2str(exp->exp_connection->c_peer.nid));
 
         len = sizeof(*lh) + sizeof(*hal) + MTI_NAME_MAXLEN +
-                /* for mockup below */ 2 * size_round(sizeof(*hai));
+                /* for mockup below */ 2 * cfs_size_round(sizeof(*hai));
         OBD_ALLOC(lh, len);
         if (lh == NULL)
                 RETURN(-ENOMEM);
 
-        lh->lnl_magic = LNL_MAGIC;
-        lh->lnl_transport = LNL_TRANSPORT_HSM;
-        lh->lnl_msgtype = HMT_ACTION_LIST;
-        lh->lnl_msglen = len;
+        lh->kuc_magic = KUC_MAGIC;
+        lh->kuc_transport = KUC_TRANSPORT_HSM;
+        lh->kuc_msgtype = HMT_ACTION_LIST;
+        lh->kuc_msglen = len;
 
         hal = (struct hsm_action_list *)(lh + 1);
         hal->hal_version = HAL_VERSION;
@@ -5639,10 +5664,11 @@ int mdt_hsm_copytool_send(struct obd_export *exp)
 
         /* Uses the ldlm reverse import; this rpc will be seen by
           the ldlm_callback_handler */
-        rc = target_set_info_rpc(exp->exp_imp_reverse, LDLM_SET_INFO,
-                                 sizeof(KEY_HSM_COPYTOOL_SEND),
-                                 KEY_HSM_COPYTOOL_SEND,
-                                 len, lh, NULL);
+        rc = do_set_info_async(exp->exp_imp_reverse,
+                               LDLM_SET_INFO, LUSTRE_OBD_VERSION,
+                               sizeof(KEY_HSM_COPYTOOL_SEND),
+                               KEY_HSM_COPYTOOL_SEND,
+                               len, lh, NULL);
 
         OBD_FREE(lh, len);
 
@@ -5713,15 +5739,6 @@ LU_KEY_INIT_FINI(mdt, struct mdt_thread_info);
 /* context key: mdt_thread_key */
 LU_CONTEXT_KEY_DEFINE(mdt, LCT_MD_THREAD);
 
-/* context key constructor/destructor: mdt_txn_key_init, mdt_txn_key_fini */
-LU_KEY_INIT_FINI(mdt_txn, struct mdt_txn_info);
-
-struct lu_context_key mdt_txn_key = {
-        .lct_tags = LCT_TX_HANDLE,
-        .lct_init = mdt_txn_key_init,
-        .lct_fini = mdt_txn_key_fini
-};
-
 struct md_ucred *mdt_ucred(const struct mdt_thread_info *info)
 {
         return md_ucred(info->mti_env);
@@ -5764,7 +5781,7 @@ int mdt_cos_is_enabled(struct mdt_device *mdt)
 }
 
 /* type constructor/destructor: mdt_type_init, mdt_type_fini */
-LU_TYPE_INIT_FINI(mdt, &mdt_thread_key, &mdt_txn_key);
+LU_TYPE_INIT_FINI(mdt, &mdt_thread_key);
 
 static struct lu_device_type_operations mdt_device_type_ops = {
         .ldto_init = mdt_type_init,
@@ -5798,7 +5815,17 @@ static int __init mdt_mod_init(void)
 
         llo_local_obj_register(&mdt_last_recv);
 
-        mdt_num_threads = MDT_NUM_THREADS;
+        if (mdt_num_threads > 0) {
+                if (mdt_num_threads > MDT_MAX_THREADS)
+                        mdt_num_threads = MDT_MAX_THREADS;
+                if (mdt_num_threads < MDT_MIN_THREADS)
+                        mdt_num_threads = MDT_MIN_THREADS;
+                mdt_max_threads = mdt_min_threads = mdt_num_threads;
+        } else {
+                mdt_max_threads = MDT_MAX_THREADS;
+                mdt_min_threads = MDT_MIN_THREADS;
+        }
+
         lprocfs_mdt_init_vars(&lvars);
         rc = class_register_type(&mdt_obd_device_ops, NULL,
                                  lvars.module_vars, LUSTRE_MDT_NAME,