Whamcloud - gitweb
LU-3544 fid: do open-by-fid by default
[fs/lustre-release.git] / lustre / mdc / mdc_locks.c
index 6c28587..2d3d171 100644 (file)
@@ -1,83 +1,71 @@
-/* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
- * vim:expandtab:shiftwidth=8:tabstop=8:
+/*
+ * GPL HEADER START
+ *
+ * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
+ *
+ * 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.
  *
- * Copyright (C) 2001-2003 Cluster File Systems, Inc.
+ * 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).
  *
- *   This file is part of the Lustre file system, http://www.lustre.org
- *   Lustre is a trademark of Cluster File Systems, Inc.
+ * 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
  *
- *   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.
+ * 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.
  *
- *   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.
+ * GPL HEADER END
+ */
+/*
+ * Copyright (c) 2003, 2010, Oracle and/or its affiliates. All rights reserved.
+ * Use is subject to license terms.
  *
- *   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.
+ * Copyright (c) 2011, 2013, Intel Corporation.
+ */
+/*
+ * This file is part of Lustre, http://www.lustre.org/
+ * Lustre is a trademark of Sun Microsystems, Inc.
  */
 
-#ifndef EXPORT_SYMTAB
-# define EXPORT_SYMTAB
-#endif
 #define DEBUG_SUBSYSTEM S_MDC
 
 #ifdef __KERNEL__
 # include <linux/module.h>
-# include <linux/pagemap.h>
-# include <linux/miscdevice.h>
-# include <linux/init.h>
+# include <linux/lustre_intent.h>
 #else
 # include <liblustre.h>
 #endif
 
-#include <linux/lustre_acl.h>
+#include <obd.h>
 #include <obd_class.h>
 #include <lustre_dlm.h>
-/* fid_res_name_eq() */
-#include <lustre_fid.h>
-#include <lprocfs_status.h>
+#include <lustre_fid.h> /* fid_res_name_eq() */
+#include <lustre_mdc.h>
+#include <lustre_net.h>
+#include <lustre_req_layout.h>
 #include "mdc_internal.h"
 
-int it_disposition(struct lookup_intent *it, int flag)
-{
-        return it->d.lustre.it_disposition & flag;
-}
-EXPORT_SYMBOL(it_disposition);
-
-void it_set_disposition(struct lookup_intent *it, int flag)
-{
-        it->d.lustre.it_disposition |= flag;
-}
-EXPORT_SYMBOL(it_set_disposition);
-
-void it_clear_disposition(struct lookup_intent *it, int flag)
-{
-        it->d.lustre.it_disposition &= ~flag;
-}
-EXPORT_SYMBOL(it_clear_disposition);
-
-static int it_to_lock_mode(struct lookup_intent *it)
-{
-        ENTRY;
-
-        /* CREAT needs to be tested before open (both could be set) */
-        if (it->it_op & IT_CREAT)
-                return LCK_PW;
-        else if (it->it_op & (IT_READDIR | IT_GETATTR | IT_OPEN | IT_LOOKUP))
-                return LCK_PR;
-
-        LBUG();
-        RETURN(-EINVAL);
-}
+struct mdc_getattr_args {
+        struct obd_export           *ga_exp;
+        struct md_enqueue_info      *ga_minfo;
+        struct ldlm_enqueue_info    *ga_einfo;
+};
 
 int it_open_error(int phase, struct lookup_intent *it)
 {
+       if (it_disposition(it, DISP_OPEN_LEASE)) {
+               if (phase >= DISP_OPEN_LEASE)
+                       return it->d.lustre.it_status;
+               else
+                       return 0;
+       }
         if (it_disposition(it, DISP_OPEN_OPEN)) {
                 if (phase >= DISP_OPEN_OPEN)
                         return it->d.lustre.it_status;
@@ -113,102 +101,134 @@ int it_open_error(int phase, struct lookup_intent *it)
 EXPORT_SYMBOL(it_open_error);
 
 /* this must be called on a lockh that is known to have a referenced lock */
-int mdc_set_lock_data(struct obd_export *exp, __u64 *lockh, void *data)
+int mdc_set_lock_data(struct obd_export *exp, __u64 *lockh, void *data,
+                      __u64 *bits)
 {
-        struct ldlm_lock *lock;
+       struct ldlm_lock *lock;
+       struct inode *new_inode = data;
         ENTRY;
 
-        if (!*lockh) {
-                EXIT;
+        if(bits)
+                *bits = 0;
+
+        if (!*lockh)
                 RETURN(0);
-        }
 
         lock = ldlm_handle2lock((struct lustre_handle *)lockh);
 
         LASSERT(lock != NULL);
         lock_res_and_lock(lock);
 #ifdef __KERNEL__
-        if (lock->l_ast_data && lock->l_ast_data != data) {
-                struct inode *new_inode = data;
-                struct inode *old_inode = lock->l_ast_data;
-                LASSERTF(old_inode->i_state & I_FREEING,
-                         "Found existing inode %p/%lu/%u state %lu in lock: "
-                         "setting data to %p/%lu/%u\n", old_inode,
-                         old_inode->i_ino, old_inode->i_generation,
-                         old_inode->i_state,
-                         new_inode, new_inode->i_ino, new_inode->i_generation);
-        }
+       if (lock->l_resource->lr_lvb_inode &&
+           lock->l_resource->lr_lvb_inode != data) {
+               struct inode *old_inode = lock->l_resource->lr_lvb_inode;
+               LASSERTF(old_inode->i_state & I_FREEING,
+                        "Found existing inode %p/%lu/%u state %lu in lock: "
+                        "setting data to %p/%lu/%u\n", old_inode,
+                        old_inode->i_ino, old_inode->i_generation,
+                        old_inode->i_state,
+                        new_inode, new_inode->i_ino, new_inode->i_generation);
+       }
 #endif
-        lock->l_ast_data = data;
+       lock->l_resource->lr_lvb_inode = new_inode;
+        if (bits)
+                *bits = lock->l_policy_data.l_inodebits.bits;
+
         unlock_res_and_lock(lock);
         LDLM_LOCK_PUT(lock);
 
         RETURN(0);
 }
 
-int mdc_lock_match(struct obd_export *exp, int flags,
-                   const struct lu_fid *fid, ldlm_type_t type,
-                   ldlm_policy_data_t *policy, ldlm_mode_t mode,
-                   struct lustre_handle *lockh)
+ldlm_mode_t mdc_lock_match(struct obd_export *exp, __u64 flags,
+                          const struct lu_fid *fid, ldlm_type_t type,
+                          ldlm_policy_data_t *policy, ldlm_mode_t mode,
+                          struct lustre_handle *lockh)
 {
-        struct ldlm_res_id res_id =
-                { .name = {fid_seq(fid),
-                           fid_oid(fid),
-                           fid_ver(fid)} };
-        struct obd_device *obd = class_exp2obd(exp);
-        int rc;
-        ENTRY;
-
-        rc = ldlm_lock_match(obd->obd_namespace, flags,
-                             &res_id, type, policy, mode, lockh);
-
-        RETURN(rc);
+       struct ldlm_res_id res_id;
+       ldlm_mode_t rc;
+       ENTRY;
+
+       fid_build_reg_res_name(fid, &res_id);
+       /* LU-4405: Clear bits not supported by server */
+       policy->l_inodebits.bits &= exp_connect_ibits(exp);
+       rc = ldlm_lock_match(class_exp2obd(exp)->obd_namespace, flags,
+                            &res_id, type, policy, mode, lockh, 0);
+       RETURN(rc);
 }
 
 int mdc_cancel_unused(struct obd_export *exp,
                       const struct lu_fid *fid,
                       ldlm_policy_data_t *policy,
-                      ldlm_mode_t mode, int flags, void *opaque)
+                      ldlm_mode_t mode,
+                      ldlm_cancel_flags_t flags,
+                      void *opaque)
 {
-        struct ldlm_res_id res_id =
-                { .name = {fid_seq(fid),
-                           fid_oid(fid),
-                           fid_ver(fid)} };
+        struct ldlm_res_id res_id;
         struct obd_device *obd = class_exp2obd(exp);
         int rc;
 
         ENTRY;
 
+        fid_build_reg_res_name(fid, &res_id);
         rc = ldlm_cli_cancel_unused_resource(obd->obd_namespace, &res_id,
                                              policy, mode, flags, opaque);
         RETURN(rc);
 }
 
-int mdc_change_cbdata(struct obd_export *exp,
-                      const struct lu_fid *fid,
-                      ldlm_iterator_t it, void *data)
+int mdc_null_inode(struct obd_export *exp,
+                  const struct lu_fid *fid)
 {
-        struct ldlm_res_id res_id = { .name = {0} };
-        ENTRY;
+       struct ldlm_res_id res_id;
+       struct ldlm_resource *res;
+       struct ldlm_namespace *ns = class_exp2obd(exp)->obd_namespace;
+       ENTRY;
 
-        res_id.name[0] = fid_seq(fid);
-        res_id.name[1] = fid_oid(fid);
-        res_id.name[2] = fid_ver(fid);
+       LASSERTF(ns != NULL, "no namespace passed\n");
 
-        ldlm_resource_iterate(class_exp2obd(exp)->obd_namespace,
-                              &res_id, it, data);
+       fid_build_reg_res_name(fid, &res_id);
 
-        EXIT;
-        return 0;
+       res = ldlm_resource_get(ns, NULL, &res_id, 0, 0);
+       if (IS_ERR(res))
+               RETURN(0);
+
+       lock_res(res);
+       res->lr_lvb_inode = NULL;
+       unlock_res(res);
+
+       ldlm_resource_putref(res);
+       RETURN(0);
+}
+
+/* find any ldlm lock of the inode in mdc
+ * return 0    not find
+ *        1    find one
+ *      < 0    error */
+int mdc_find_cbdata(struct obd_export *exp,
+                    const struct lu_fid *fid,
+                    ldlm_iterator_t it, void *data)
+{
+        struct ldlm_res_id res_id;
+        int rc = 0;
+        ENTRY;
+
+        fid_build_reg_res_name((struct lu_fid*)fid, &res_id);
+        rc = ldlm_resource_iterate(class_exp2obd(exp)->obd_namespace, &res_id,
+                                   it, data);
+        if (rc == LDLM_ITER_STOP)
+                RETURN(1);
+        else if (rc == LDLM_ITER_CONTINUE)
+                RETURN(0);
+        RETURN(rc);
 }
 
 static inline void mdc_clear_replay_flag(struct ptlrpc_request *req, int rc)
 {
-        /* Don't hold error requests for replay. */
-        if (req->rq_replay) {
-                spin_lock(&req->rq_lock);
-                req->rq_replay = 0;
-                spin_unlock(&req->rq_lock);
+       /* Don't hold error requests for replay. */
+       if (req->rq_replay) {
+               spin_lock(&req->rq_lock);
+               req->rq_replay = 0;
+               spin_unlock(&req->rq_lock);
         }
         if (rc && req->rq_transno != 0) {
                 DEBUG_REQ(D_ERROR, req, "transno returned on error rc %d", rc);
@@ -228,236 +248,349 @@ static inline void mdc_clear_replay_flag(struct ptlrpc_request *req, int rc)
  * but this is incredibly unlikely, and questionable whether the client
  * could do MDS recovery under OOM anyways... */
 static void mdc_realloc_openmsg(struct ptlrpc_request *req,
-                                struct mdt_body *body, int size[9])
+                               struct mdt_body *body)
 {
-        int     rc;
+       int     rc;
+
+       /* FIXME: remove this explicit offset. */
+       rc = sptlrpc_cli_enlarge_reqbuf(req, DLM_INTENT_REC_OFF + 4,
+                                       body->mbo_eadatasize);
+       if (rc) {
+               CERROR("Can't enlarge segment %d size to %d\n",
+                      DLM_INTENT_REC_OFF + 4, body->mbo_eadatasize);
+               body->mbo_valid &= ~OBD_MD_FLEASIZE;
+               body->mbo_eadatasize = 0;
+       }
+}
+
+static struct ptlrpc_request *
+mdc_intent_open_pack(struct obd_export *exp, struct lookup_intent *it,
+                    struct md_op_data *op_data)
+{
+       struct ptlrpc_request   *req;
+       struct obd_device       *obddev = class_exp2obd(exp);
+       struct ldlm_intent      *lit;
+       const void              *lmm = op_data->op_data;
+       int                      lmmsize = op_data->op_data_size;
+       struct list_head         cancels = LIST_HEAD_INIT(cancels);
+       int                      count = 0;
+       int                      mode;
+       int                      rc;
         ENTRY;
 
-        rc = sptlrpc_cli_enlarge_reqbuf(req, DLM_INTENT_REC_OFF + 4,
-                                        body->eadatasize);
-        if (rc) {
-                CERROR("Can't enlarge segment %d size to %d\n",
-                       DLM_INTENT_REC_OFF + 4, body->eadatasize);
-                body->valid &= ~OBD_MD_FLEASIZE;
-                body->eadatasize = 0;
+        it->it_create_mode = (it->it_create_mode & ~S_IFMT) | S_IFREG;
+
+        /* XXX: openlock is not cancelled for cross-refs. */
+        /* If inode is known, cancel conflicting OPEN locks. */
+       if (fid_is_sane(&op_data->op_fid2)) {
+               if (it->it_flags & MDS_OPEN_LEASE) { /* try to get lease */
+                       if (it->it_flags & FMODE_WRITE)
+                               mode = LCK_EX;
+                       else
+                               mode = LCK_PR;
+               } else {
+                       if (it->it_flags & (FMODE_WRITE|MDS_OPEN_TRUNC))
+                               mode = LCK_CW;
+#ifdef FMODE_EXEC
+                       else if (it->it_flags & FMODE_EXEC)
+                               mode = LCK_PR;
+#endif
+                       else
+                               mode = LCK_CR;
+               }
+               count = mdc_resource_get_unused(exp, &op_data->op_fid2,
+                                               &cancels, mode,
+                                               MDS_INODELOCK_OPEN);
+       }
+
+        /* If CREATE, cancel parent's UPDATE lock. */
+        if (it->it_op & IT_CREAT)
+                mode = LCK_EX;
+        else
+                mode = LCK_CR;
+        count += mdc_resource_get_unused(exp, &op_data->op_fid1,
+                                         &cancels, mode,
+                                         MDS_INODELOCK_UPDATE);
+
+        req = ptlrpc_request_alloc(class_exp2cliimp(exp),
+                                   &RQF_LDLM_INTENT_OPEN);
+        if (req == NULL) {
+                ldlm_lock_list_put(&cancels, l_bl_ast, count);
+                RETURN(ERR_PTR(-ENOMEM));
         }
-        EXIT;
+
+        /* parent capability */
+        mdc_set_capa_size(req, &RMF_CAPA1, op_data->op_capa1);
+        /* child capability, reserve the size according to parent capa, it will
+         * be filled after we get the reply */
+        mdc_set_capa_size(req, &RMF_CAPA2, op_data->op_capa1);
+
+        req_capsule_set_size(&req->rq_pill, &RMF_NAME, RCL_CLIENT,
+                             op_data->op_namelen + 1);
+       req_capsule_set_size(&req->rq_pill, &RMF_EADATA, RCL_CLIENT,
+                            max(lmmsize, obddev->u.cli.cl_default_mds_easize));
+
+       rc = ldlm_prep_enqueue_req(exp, req, &cancels, count);
+       if (rc < 0) {
+               ptlrpc_request_free(req);
+               RETURN(ERR_PTR(rc));
+       }
+
+       spin_lock(&req->rq_lock);
+       req->rq_replay = req->rq_import->imp_replayable;
+       spin_unlock(&req->rq_lock);
+
+        /* pack the intent */
+        lit = req_capsule_client_get(&req->rq_pill, &RMF_LDLM_INTENT);
+        lit->opc = (__u64)it->it_op;
+
+        /* pack the intended request */
+        mdc_open_pack(req, op_data, it->it_create_mode, 0, it->it_flags, lmm,
+                      lmmsize);
+
+       req_capsule_set_size(&req->rq_pill, &RMF_MDT_MD, RCL_SERVER,
+                            obddev->u.cli.cl_max_mds_easize);
+
+        /* for remote client, fetch remote perm for current user */
+        if (client_is_remote(exp))
+                req_capsule_set_size(&req->rq_pill, &RMF_ACL, RCL_SERVER,
+                                     sizeof(struct mdt_remote_perm));
+        ptlrpc_request_set_replen(req);
+        return req;
 }
 
-/* We always reserve enough space in the reply packet for a stripe MD, because
- * we don't know in advance the file type. */
-int mdc_enqueue(struct obd_export *exp, struct ldlm_enqueue_info *einfo,
-                struct lookup_intent *it, struct md_op_data *op_data,
-                struct lustre_handle *lockh, void *lmm, int lmmsize,
-                int extra_lock_flags)
+static struct ptlrpc_request *
+mdc_intent_getxattr_pack(struct obd_export *exp,
+                        struct lookup_intent *it,
+                        struct md_op_data *op_data)
+{
+       struct ptlrpc_request   *req;
+       struct ldlm_intent      *lit;
+       int                     rc, count = 0, maxdata;
+       struct list_head        cancels = LIST_HEAD_INIT(cancels);
+
+       ENTRY;
+
+       req = ptlrpc_request_alloc(class_exp2cliimp(exp),
+                                       &RQF_LDLM_INTENT_GETXATTR);
+       if (req == NULL)
+               RETURN(ERR_PTR(-ENOMEM));
+
+       mdc_set_capa_size(req, &RMF_CAPA1, op_data->op_capa1);
+
+       rc = ldlm_prep_enqueue_req(exp, req, &cancels, count);
+       if (rc) {
+               ptlrpc_request_free(req);
+               RETURN(ERR_PTR(rc));
+       }
+
+       /* pack the intent */
+       lit = req_capsule_client_get(&req->rq_pill, &RMF_LDLM_INTENT);
+       lit->opc = IT_GETXATTR;
+
+       maxdata = class_exp2cliimp(exp)->imp_connect_data.ocd_max_easize;
+
+       /* pack the intended request */
+       mdc_pack_body(req, &op_data->op_fid1, op_data->op_capa1,
+                       op_data->op_valid, maxdata, -1, 0);
+
+       req_capsule_set_size(&req->rq_pill, &RMF_EADATA,
+                               RCL_SERVER, maxdata);
+
+       req_capsule_set_size(&req->rq_pill, &RMF_EAVALS,
+                               RCL_SERVER, maxdata);
+
+       req_capsule_set_size(&req->rq_pill, &RMF_EAVALS_LENS,
+                               RCL_SERVER, maxdata);
+
+       ptlrpc_request_set_replen(req);
+
+       RETURN(req);
+}
+
+static struct ptlrpc_request *mdc_intent_unlink_pack(struct obd_export *exp,
+                                                     struct lookup_intent *it,
+                                                     struct md_op_data *op_data)
 {
         struct ptlrpc_request *req;
-        struct obd_device *obddev = class_exp2obd(exp);
-        struct ldlm_res_id res_id =
-                { .name = {fid_seq(&op_data->op_fid1),
-                           fid_oid(&op_data->op_fid1),
-                           fid_ver(&op_data->op_fid1)} };
-        ldlm_policy_data_t policy = { .l_inodebits = { MDS_INODELOCK_LOOKUP } };
-        struct ldlm_request *lockreq;
-        struct ldlm_intent *lit;
-        struct ldlm_reply *lockrep;
-        int size[9] = { [MSG_PTLRPC_BODY_OFF] = sizeof(struct ptlrpc_body),
-                        [DLM_LOCKREQ_OFF]     = sizeof(*lockreq),
-                        [DLM_INTENT_IT_OFF]   = sizeof(*lit),
-                        0, 0, 0, 0, 0, 0 };
-        int repsize[7] = { [MSG_PTLRPC_BODY_OFF] = sizeof(struct ptlrpc_body),
-                           [DLM_LOCKREPLY_OFF]   = sizeof(*lockrep),
-                           [DLM_REPLY_REC_OFF]   = sizeof(struct mdt_body),
-                           [DLM_REPLY_REC_OFF+1] = obddev->u.cli.
-                                                   cl_max_mds_easize,
-                           0, 0, 0 };
-        int flags = extra_lock_flags | LDLM_FL_HAS_INTENT;
-        int repbufcnt = 4, rc;
+        struct obd_device     *obddev = class_exp2obd(exp);
+        struct ldlm_intent    *lit;
+        int                    rc;
         ENTRY;
 
-        LASSERTF(einfo->ei_type == LDLM_IBITS,"lock type %d\n", einfo->ei_type);
-
-        if (it->it_op & IT_OPEN) {
-                int do_join = !!(it->it_flags & O_JOIN_FILE);
-                CFS_LIST_HEAD(cancels);
-                int count = 0;
-                int mode;
-
-                it->it_create_mode = (it->it_create_mode & ~S_IFMT) | S_IFREG;
-
-                size[DLM_INTENT_REC_OFF] = sizeof(struct mdt_rec_create);
-                /* parent capability */
-                size[DLM_INTENT_REC_OFF + 1] = op_data->op_capa1 ?
-                                               sizeof(struct lustre_capa) : 0;
-                /* child capability, used for replay only */
-                size[DLM_INTENT_REC_OFF + 2] = sizeof(struct lustre_capa);
-                size[DLM_INTENT_REC_OFF + 3] = op_data->op_namelen + 1;
-                /* As an optimization, we allocate an RPC request buffer for
-                 * at least a default-sized LOV EA even if we aren't sending
-                 * one.
-                 */
-                size[DLM_INTENT_REC_OFF + 4] = max(lmmsize,
-                                           obddev->u.cli.cl_default_mds_easize);
-
-                /* XXX: openlock is not cancelled for cross-refs. */
-                /* If inode is known, cancel conflicting OPEN locks. */
-                if (fid_is_sane(&op_data->op_fid2)) {
-                        if (it->it_flags & (FMODE_WRITE|MDS_OPEN_TRUNC))
-                                mode = LCK_CW;
-#ifdef FMODE_EXEC
-                        else if (it->it_flags & FMODE_EXEC)
-                                mode = LCK_PR;
-#endif
-                        else 
-                                mode = LCK_CR;
-                        count = mdc_resource_get_unused(exp, &op_data->op_fid2,
-                                                        &cancels, mode,
-                                                        MDS_INODELOCK_OPEN);
-                }
+        req = ptlrpc_request_alloc(class_exp2cliimp(exp),
+                                   &RQF_LDLM_INTENT_UNLINK);
+        if (req == NULL)
+                RETURN(ERR_PTR(-ENOMEM));
 
-                /* If CREATE or JOIN_FILE, cancel parent's UPDATE lock. */
-                if (it->it_op & IT_CREAT || it->it_flags & O_JOIN_FILE)
-                        mode = LCK_EX;
-                else
-                        mode = LCK_CR;
-                count += mdc_resource_get_unused(exp, &op_data->op_fid1,
-                                                 &cancels, mode,
-                                                 MDS_INODELOCK_UPDATE);
-
-                if (do_join)
-                        size[DLM_INTENT_REC_OFF + 5] =
-                                                sizeof(struct mdt_rec_join);
-
-                req = ldlm_prep_enqueue_req(exp, 8 + do_join, size, &cancels,
-                                            count);
-                if (!req)
-                        RETURN(-ENOMEM);
-
-                if (do_join) {
-                        /* join is like an unlink of the tail */
-                        policy.l_inodebits.bits = MDS_INODELOCK_UPDATE;
-                        mdc_join_pack(req, DLM_INTENT_REC_OFF + 5, op_data,
-                                      (*(__u64 *)op_data->op_data));
-                }
+        mdc_set_capa_size(req, &RMF_CAPA1, op_data->op_capa1);
+        req_capsule_set_size(&req->rq_pill, &RMF_NAME, RCL_CLIENT,
+                             op_data->op_namelen + 1);
 
-                spin_lock(&req->rq_lock);
-                req->rq_replay = 1;
-                spin_unlock(&req->rq_lock);
-
-                /* pack the intent */
-                lit = lustre_msg_buf(req->rq_reqmsg, DLM_INTENT_IT_OFF,
-                                     sizeof(*lit));
-                lit->opc = (__u64)it->it_op;
-
-                /* pack the intended request */
-                mdc_open_pack(req, DLM_INTENT_REC_OFF, op_data,
-                              it->it_create_mode, 0, it->it_flags,
-                              lmm, lmmsize);
-
-                /* for remote client, fetch remote perm for current user */
-                repsize[repbufcnt++] = client_is_remote(exp) ?
-                                       sizeof(struct mdt_remote_perm) :
-                                       LUSTRE_POSIX_ACL_MAX_SIZE;
-                repsize[repbufcnt++] = sizeof(struct lustre_capa);
-                repsize[repbufcnt++] = sizeof(struct lustre_capa);
-        } else if (it->it_op & IT_UNLINK) {
-                size[DLM_INTENT_REC_OFF] = sizeof(struct mdt_rec_unlink);
-                size[DLM_INTENT_REC_OFF + 1] = op_data->op_capa1 ?
-                                               sizeof(struct lustre_capa) : 0;
-                size[DLM_INTENT_REC_OFF + 2] = op_data->op_namelen + 1;
-                policy.l_inodebits.bits = MDS_INODELOCK_UPDATE;
-                req = ldlm_prep_enqueue_req(exp, 6, size, NULL, 0);
-                if (!req)
-                        RETURN(-ENOMEM);
-
-                /* pack the intent */
-                lit = lustre_msg_buf(req->rq_reqmsg, DLM_INTENT_IT_OFF,
-                                     sizeof(*lit));
-                lit->opc = (__u64)it->it_op;
-
-                /* pack the intended request */
-                mdc_unlink_pack(req, DLM_INTENT_REC_OFF, op_data);
-
-                repsize[repbufcnt++] = obddev->u.cli.cl_max_mds_cookiesize;
-        } else if (it->it_op & (IT_GETATTR | IT_LOOKUP)) {
-                obd_valid valid = OBD_MD_FLGETATTR | OBD_MD_FLEASIZE |
-                        OBD_MD_FLMODEASIZE | OBD_MD_FLDIREA |
-                        OBD_MD_FLMDSCAPA | OBD_MD_MEA;
-                valid |= client_is_remote(exp) ? OBD_MD_FLRMTPERM :
-                                                 OBD_MD_FLACL;
-                size[DLM_INTENT_REC_OFF] = sizeof(struct mdt_body);
-                size[DLM_INTENT_REC_OFF + 1] = op_data->op_capa1 ?
-                                               sizeof(struct lustre_capa) : 0;
-                size[DLM_INTENT_REC_OFF + 2] = op_data->op_namelen + 1;
-
-                if (it->it_op & IT_GETATTR)
-                        policy.l_inodebits.bits = MDS_INODELOCK_UPDATE;
-
-                req = ldlm_prep_enqueue_req(exp, 6, size, NULL, 0);
-                if (!req)
-                        RETURN(-ENOMEM);
-
-                /* pack the intent */
-                lit = lustre_msg_buf(req->rq_reqmsg, DLM_INTENT_IT_OFF,
-                                     sizeof(*lit));
-                lit->opc = (__u64)it->it_op;
-
-                /* pack the intended request */
-                mdc_getattr_pack(req, DLM_INTENT_REC_OFF, valid,
-                                 it->it_flags, op_data);
-
-                repsize[repbufcnt++] = client_is_remote(exp) ?
-                                       sizeof(struct mdt_remote_perm) :
-                                       LUSTRE_POSIX_ACL_MAX_SIZE;
-                repsize[repbufcnt++] = sizeof(struct lustre_capa);
-        } else if (it->it_op == IT_READDIR) {
-                policy.l_inodebits.bits = MDS_INODELOCK_UPDATE;
-                req = ldlm_prep_enqueue_req(exp, 2, size, NULL, 0);
-                if (!req)
-                        RETURN(-ENOMEM);
-
-                repbufcnt = 2;
-        } else {
-                LBUG();
-                RETURN(-EINVAL);
+        rc = ldlm_prep_enqueue_req(exp, req, NULL, 0);
+        if (rc) {
+                ptlrpc_request_free(req);
+                RETURN(ERR_PTR(rc));
         }
 
-        /* get ready for the reply */
-        ptlrpc_req_set_repsize(req, repbufcnt, repsize);
+        /* pack the intent */
+        lit = req_capsule_client_get(&req->rq_pill, &RMF_LDLM_INTENT);
+        lit->opc = (__u64)it->it_op;
 
-         /* It is important to obtain rpc_lock first (if applicable), so that
-          * threads that are serialised with rpc_lock are not polluting our
-          * rpcs in flight counter */
-        mdc_get_rpc_lock(obddev->u.cli.cl_rpc_lock, it);
-        mdc_enter_request(&obddev->u.cli);
-        rc = ldlm_cli_enqueue(exp, &req, einfo, &res_id, &policy, &flags, NULL,
-                              0, NULL, lockh, 0);
-        mdc_exit_request(&obddev->u.cli);
-        mdc_put_rpc_lock(obddev->u.cli.cl_rpc_lock, it);
+        /* pack the intended request */
+        mdc_unlink_pack(req, op_data);
+
+       req_capsule_set_size(&req->rq_pill, &RMF_MDT_MD, RCL_SERVER,
+                            obddev->u.cli.cl_default_mds_easize);
+       req_capsule_set_size(&req->rq_pill, &RMF_ACL, RCL_SERVER,
+                            obddev->u.cli.cl_default_mds_cookiesize);
+       ptlrpc_request_set_replen(req);
+       RETURN(req);
+}
+
+static struct ptlrpc_request *mdc_intent_getattr_pack(struct obd_export *exp,
+                                                      struct lookup_intent *it,
+                                                      struct md_op_data *op_data)
+{
+       struct ptlrpc_request *req;
+       struct obd_device     *obddev = class_exp2obd(exp);
+       obd_valid              valid = OBD_MD_FLGETATTR | OBD_MD_FLEASIZE |
+                                      OBD_MD_FLMODEASIZE | OBD_MD_FLDIREA |
+                                      OBD_MD_FLMDSCAPA | OBD_MD_MEA |
+                                      (client_is_remote(exp) ?
+                                              OBD_MD_FLRMTPERM : OBD_MD_FLACL);
+       struct ldlm_intent    *lit;
+       int                    rc;
+       int                     easize;
+       ENTRY;
+
+        req = ptlrpc_request_alloc(class_exp2cliimp(exp),
+                                   &RQF_LDLM_INTENT_GETATTR);
+        if (req == NULL)
+                RETURN(ERR_PTR(-ENOMEM));
+
+        mdc_set_capa_size(req, &RMF_CAPA1, op_data->op_capa1);
+        req_capsule_set_size(&req->rq_pill, &RMF_NAME, RCL_CLIENT,
+                             op_data->op_namelen + 1);
+
+        rc = ldlm_prep_enqueue_req(exp, req, NULL, 0);
+        if (rc) {
+                ptlrpc_request_free(req);
+                RETURN(ERR_PTR(rc));
+        }
+
+        /* pack the intent */
+        lit = req_capsule_client_get(&req->rq_pill, &RMF_LDLM_INTENT);
+        lit->opc = (__u64)it->it_op;
+
+       if (obddev->u.cli.cl_default_mds_easize > 0)
+               easize = obddev->u.cli.cl_default_mds_easize;
+       else
+               easize = obddev->u.cli.cl_max_mds_easize;
+
+       /* pack the intended request */
+       mdc_getattr_pack(req, valid, it->it_flags, op_data, easize);
+
+       req_capsule_set_size(&req->rq_pill, &RMF_MDT_MD, RCL_SERVER, easize);
+       if (client_is_remote(exp))
+               req_capsule_set_size(&req->rq_pill, &RMF_ACL, RCL_SERVER,
+                                    sizeof(struct mdt_remote_perm));
+       ptlrpc_request_set_replen(req);
+       RETURN(req);
+}
+
+static struct ptlrpc_request *mdc_intent_layout_pack(struct obd_export *exp,
+                                                    struct lookup_intent *it,
+                                                    struct md_op_data *unused)
+{
+       struct obd_device     *obd = class_exp2obd(exp);
+       struct ptlrpc_request *req;
+       struct ldlm_intent    *lit;
+       struct layout_intent  *layout;
+       int rc;
+       ENTRY;
+
+       req = ptlrpc_request_alloc(class_exp2cliimp(exp),
+                               &RQF_LDLM_INTENT_LAYOUT);
+       if (req == NULL)
+               RETURN(ERR_PTR(-ENOMEM));
+
+       req_capsule_set_size(&req->rq_pill, &RMF_EADATA, RCL_CLIENT, 0);
+       rc = ldlm_prep_enqueue_req(exp, req, NULL, 0);
+       if (rc) {
+               ptlrpc_request_free(req);
+               RETURN(ERR_PTR(rc));
+       }
+
+       /* pack the intent */
+       lit = req_capsule_client_get(&req->rq_pill, &RMF_LDLM_INTENT);
+       lit->opc = (__u64)it->it_op;
+
+       /* pack the layout intent request */
+       layout = req_capsule_client_get(&req->rq_pill, &RMF_LAYOUT_INTENT);
+       /* LAYOUT_INTENT_ACCESS is generic, specific operation will be
+        * set for replication */
+       layout->li_opc = LAYOUT_INTENT_ACCESS;
+
+       req_capsule_set_size(&req->rq_pill, &RMF_DLM_LVB, RCL_SERVER,
+                            obd->u.cli.cl_default_mds_easize);
+       ptlrpc_request_set_replen(req);
+       RETURN(req);
+}
+
+static struct ptlrpc_request *
+mdc_enqueue_pack(struct obd_export *exp, int lvb_len)
+{
+        struct ptlrpc_request *req;
+        int rc;
+        ENTRY;
+
+        req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_LDLM_ENQUEUE);
+        if (req == NULL)
+                RETURN(ERR_PTR(-ENOMEM));
 
+        rc = ldlm_prep_enqueue_req(exp, req, NULL, 0);
+        if (rc) {
+                ptlrpc_request_free(req);
+                RETURN(ERR_PTR(rc));
+        }
+
+       req_capsule_set_size(&req->rq_pill, &RMF_DLM_LVB, RCL_SERVER, lvb_len);
+        ptlrpc_request_set_replen(req);
+        RETURN(req);
+}
+
+static int mdc_finish_enqueue(struct obd_export *exp,
+                              struct ptlrpc_request *req,
+                              struct ldlm_enqueue_info *einfo,
+                              struct lookup_intent *it,
+                              struct lustre_handle *lockh,
+                              int rc)
+{
+       struct req_capsule  *pill = &req->rq_pill;
+       struct ldlm_request *lockreq;
+       struct ldlm_reply   *lockrep;
+       struct lustre_intent_data *intent = &it->d.lustre;
+       struct ldlm_lock    *lock;
+       void                *lvb_data = NULL;
+       int                  lvb_len = 0;
+        ENTRY;
+
+        LASSERT(rc >= 0);
         /* Similarly, if we're going to replay this request, we don't want to
          * actually get a lock, just perform the intent. */
         if (req->rq_transno || req->rq_replay) {
-                lockreq = lustre_msg_buf(req->rq_reqmsg, DLM_LOCKREQ_OFF,
-                                         sizeof(*lockreq));
-                lockreq->lock_flags |= LDLM_FL_INTENT_ONLY;
+                lockreq = req_capsule_client_get(pill, &RMF_DLM_REQ);
+               lockreq->lock_flags |= ldlm_flags_to_wire(LDLM_FL_INTENT_ONLY);
         }
 
-        /* This can go when we're sure that this can never happen */
-        LASSERT(rc != -ENOENT);
         if (rc == ELDLM_LOCK_ABORTED) {
                 einfo->ei_mode = 0;
                 memset(lockh, 0, sizeof(*lockh));
                 rc = 0;
-        } else if (rc != 0) {
-                CERROR("ldlm_cli_enqueue: %d\n", rc);
-                LASSERTF(rc < 0, "rc %d\n", rc);
-                mdc_clear_replay_flag(req, rc);
-                ptlrpc_req_finished(req);
-                RETURN(rc);
         } else { /* rc = 0 */
-                struct ldlm_lock *lock = ldlm_handle2lock(lockh);
-                LASSERT(lock);
+               lock = ldlm_handle2lock(lockh);
+               LASSERT(lock != NULL);
 
                 /* If the server gave us back a different lock mode, we should
                  * fix up our variables. */
@@ -466,21 +599,22 @@ int mdc_enqueue(struct obd_export *exp, struct ldlm_enqueue_info *einfo,
                         ldlm_lock_decref(lockh, einfo->ei_mode);
                         einfo->ei_mode = lock->l_req_mode;
                 }
-                LDLM_LOCK_PUT(lock);
-        }
+               LDLM_LOCK_PUT(lock);
+       }
 
-        lockrep = lustre_msg_buf(req->rq_repmsg, DLM_LOCKREPLY_OFF,
-                                 sizeof(*lockrep));
-        LASSERT(lockrep != NULL);                 /* checked by ldlm_cli_enqueue() */
-        LASSERT_REPSWABBED(req, DLM_LOCKREPLY_OFF); /* swabbed by ldlm_cli_enqueue() */
+       lockrep = req_capsule_server_get(pill, &RMF_DLM_REP);
+       LASSERT(lockrep != NULL); /* checked by ldlm_cli_enqueue() */
 
-        it->d.lustre.it_disposition = (int)lockrep->lock_policy_res1;
-        it->d.lustre.it_status = (int)lockrep->lock_policy_res2;
-        it->d.lustre.it_lock_mode = einfo->ei_mode;
-        it->d.lustre.it_data = req;
+       intent->it_disposition = (int)lockrep->lock_policy_res1;
+       intent->it_status = (int)lockrep->lock_policy_res2;
+       intent->it_lock_mode = einfo->ei_mode;
+       intent->it_lock_handle = lockh->cookie;
+       intent->it_data = req;
 
-        if (it->d.lustre.it_status < 0 && req->rq_replay)
-                mdc_clear_replay_flag(req, it->d.lustre.it_status);
+       /* Technically speaking rq_transno must already be zero if
+        * it_status is in error, so the check is a bit redundant */
+       if ((!req->rq_transno || intent->it_status < 0) && req->rq_replay)
+               mdc_clear_replay_flag(req, intent->it_status);
 
         /* If we're doing an IT_OPEN which did not result in an actual
          * successful open, then we need to remove the bit which saves
@@ -490,26 +624,23 @@ int mdc_enqueue(struct obd_export *exp, struct ldlm_enqueue_info *einfo,
          * function without doing so, and try to replay a failed create
          * (bug 3440) */
         if (it->it_op & IT_OPEN && req->rq_replay &&
-            (!it_disposition(it, DISP_OPEN_OPEN) ||it->d.lustre.it_status != 0))
-                mdc_clear_replay_flag(req, it->d.lustre.it_status);
+           (!it_disposition(it, DISP_OPEN_OPEN) ||intent->it_status != 0))
+               mdc_clear_replay_flag(req, intent->it_status);
 
-        DEBUG_REQ(D_RPCTRACE, req, "op: %d disposition: %x, status: %d",
-                  it->it_op,it->d.lustre.it_disposition,it->d.lustre.it_status);
+       DEBUG_REQ(D_RPCTRACE, req, "op: %d disposition: %x, status: %d",
+                 it->it_op, intent->it_disposition, intent->it_status);
 
         /* We know what to expect, so we do any byte flipping required here */
-        LASSERT(repbufcnt == 7 || repbufcnt == 6 || repbufcnt == 2);
-        if (repbufcnt >= 6) {
-                int reply_off = DLM_REPLY_REC_OFF;
+        if (it->it_op & (IT_OPEN | IT_UNLINK | IT_LOOKUP | IT_GETATTR)) {
                 struct mdt_body *body;
 
-                body = lustre_swab_repbuf(req, reply_off++, sizeof(*body),
-                                         lustre_swab_mdt_body);
+                body = req_capsule_server_get(pill, &RMF_MDT_BODY);
                 if (body == NULL) {
                         CERROR ("Can't swab mdt_body\n");
                         RETURN (-EPROTO);
                 }
 
-                if (req->rq_replay && it_disposition(it, DISP_OPEN_OPEN) &&
+                if (it_disposition(it, DISP_OPEN_OPEN) &&
                     !it_open_error(DISP_OPEN_OPEN, it)) {
                         /*
                          * If this is a successful OPEN request, we need to set
@@ -517,38 +648,27 @@ int mdc_enqueue(struct obd_export *exp, struct ldlm_enqueue_info *einfo,
                          * happens immediately after swabbing below, new reply
                          * is swabbed by that handler correctly.
                          */
-                        mdc_set_open_replay_data(NULL, NULL, req);
-                }
+                       mdc_set_open_replay_data(NULL, NULL, it);
+               }
 
-                if ((body->valid & (OBD_MD_FLDIREA | OBD_MD_FLEASIZE)) != 0) {
+               if (body->mbo_valid & (OBD_MD_FLDIREA | OBD_MD_FLEASIZE)) {
                         void *eadata;
 
+                       mdc_update_max_ea_from_body(exp, body);
+
                         /*
                          * The eadata is opaque; just check that it is there.
                          * Eventually, obd_unpackmd() will check the contents.
                          */
-                        eadata = lustre_swab_repbuf(req, reply_off++,
-                                                    body->eadatasize, NULL);
-                        if (eadata == NULL) {
-                                CERROR("Missing/short eadata\n");
-                                RETURN(-EPROTO);
-                        }
-                        if (body->valid & OBD_MD_FLMODEASIZE) {
-                                if (obddev->u.cli.cl_max_mds_easize <
-                                    body->max_mdsize) {
-                                        obddev->u.cli.cl_max_mds_easize =
-                                                body->max_mdsize;
-                                        CDEBUG(D_INFO, "maxeasize become %d\n",
-                                               body->max_mdsize);
-                                }
-                                if (obddev->u.cli.cl_max_mds_cookiesize <
-                                    body->max_cookiesize) {
-                                        obddev->u.cli.cl_max_mds_cookiesize =
-                                                body->max_cookiesize;
-                                        CDEBUG(D_INFO, "cookiesize become %d\n",
-                                               body->max_cookiesize);
-                                }
-                        }
+                        eadata = req_capsule_server_sized_get(pill, &RMF_MDT_MD,
+                                                       body->mbo_eadatasize);
+                       if (eadata == NULL)
+                               RETURN(-EPROTO);
+
+                       /* save lvb data and length in case this is for layout
+                        * lock */
+                       lvb_data = eadata;
+                       lvb_len = body->mbo_eadatasize;
 
                         /*
                          * We save the reply LOV EA in case we have to replay a
@@ -560,207 +680,288 @@ int mdc_enqueue(struct obd_export *exp, struct ldlm_enqueue_info *einfo,
                          * (for example error one).
                          */
                         if ((it->it_op & IT_OPEN) && req->rq_replay) {
-                                if (lustre_msg_buflen(req->rq_reqmsg,
-                                                      DLM_INTENT_REC_OFF + 4) <
-                                    body->eadatasize)
-                                        mdc_realloc_openmsg(req, body, size);
-
-                                lmm = lustre_msg_buf(req->rq_reqmsg,
-                                                     DLM_INTENT_REC_OFF + 4,
-                                                     body->eadatasize);
-                                if (lmm)
-                                        memcpy(lmm, eadata, body->eadatasize);
-                        }
-                }
-                if (body->valid & OBD_MD_FLRMTPERM) {
+                                void *lmm;
+                                if (req_capsule_get_size(pill, &RMF_EADATA,
+                                                         RCL_CLIENT) <
+                                   body->mbo_eadatasize)
+                                       mdc_realloc_openmsg(req, body);
+                               else
+                                       req_capsule_shrink(pill, &RMF_EADATA,
+                                                          body->mbo_eadatasize,
+                                                          RCL_CLIENT);
+
+                               req_capsule_set_size(pill, &RMF_EADATA,
+                                                    RCL_CLIENT,
+                                                    body->mbo_eadatasize);
+
+                               lmm = req_capsule_client_get(pill, &RMF_EADATA);
+                               if (lmm)
+                                       memcpy(lmm, eadata,
+                                              body->mbo_eadatasize);
+                       }
+               }
+
+               if (body->mbo_valid & OBD_MD_FLRMTPERM) {
                         struct mdt_remote_perm *perm;
 
                         LASSERT(client_is_remote(exp));
-                        perm = lustre_swab_repbuf(req, reply_off++,
-                                                  sizeof(*perm),
-                                                  lustre_swab_mdt_remote_perm);
-                        if (perm == NULL) {
-                                CERROR("missing remote permission!\n");
+                        perm = req_capsule_server_swab_get(pill, &RMF_ACL,
+                                                lustre_swab_mdt_remote_perm);
+                        if (perm == NULL)
                                 RETURN(-EPROTO);
-                        }
-                } else if ((body->valid & OBD_MD_FLACL) && body->aclsize) {
-                        reply_off++;
                 }
-                if (body->valid & OBD_MD_FLMDSCAPA) {
+               if (body->mbo_valid & OBD_MD_FLMDSCAPA) {
                         struct lustre_capa *capa, *p;
 
-                        capa = lustre_unpack_capa(req->rq_repmsg, reply_off++);
-                        if (capa == NULL) {
-                                CERROR("Missing/short MDS capability\n");
+                        capa = req_capsule_server_get(pill, &RMF_CAPA1);
+                        if (capa == NULL)
                                 RETURN(-EPROTO);
-                        }
 
                         if (it->it_op & IT_OPEN) {
                                 /* client fid capa will be checked in replay */
-                                p = lustre_msg_buf(req->rq_reqmsg,
-                                                   DLM_INTENT_REC_OFF + 2,
-                                                   sizeof(*p));
+                                p = req_capsule_client_get(pill, &RMF_CAPA2);
                                 LASSERT(p);
                                 *p = *capa;
                         }
                 }
-                if (body->valid & OBD_MD_FLOSSCAPA) {
+               if (body->mbo_valid & OBD_MD_FLOSSCAPA) {
                         struct lustre_capa *capa;
 
-                        capa = lustre_unpack_capa(req->rq_repmsg, reply_off++);
-                        if (capa == NULL) {
-                                CERROR("Missing/short OSS capability\n");
+                        capa = req_capsule_server_get(pill, &RMF_CAPA2);
+                        if (capa == NULL)
                                 RETURN(-EPROTO);
-                        }
                 }
-        }
-
-        RETURN(rc);
+        } else if (it->it_op & IT_LAYOUT) {
+               /* maybe the lock was granted right away and layout
+                * is packed into RMF_DLM_LVB of req */
+               lvb_len = req_capsule_get_size(pill, &RMF_DLM_LVB, RCL_SERVER);
+               if (lvb_len > 0) {
+                       lvb_data = req_capsule_server_sized_get(pill,
+                                                       &RMF_DLM_LVB, lvb_len);
+                       if (lvb_data == NULL)
+                               RETURN(-EPROTO);
+               }
+       }
+
+       /* fill in stripe data for layout lock */
+       lock = ldlm_handle2lock(lockh);
+       if (lock != NULL && ldlm_has_layout(lock) && lvb_data != NULL) {
+               void *lmm;
+
+               LDLM_DEBUG(lock, "layout lock returned by: %s, lvb_len: %d\n",
+                       ldlm_it2str(it->it_op), lvb_len);
+
+               OBD_ALLOC_LARGE(lmm, lvb_len);
+               if (lmm == NULL) {
+                       LDLM_LOCK_PUT(lock);
+                       RETURN(-ENOMEM);
+               }
+               memcpy(lmm, lvb_data, lvb_len);
+
+               /* install lvb_data */
+               lock_res_and_lock(lock);
+               if (lock->l_lvb_data == NULL) {
+                       lock->l_lvb_type = LVB_T_LAYOUT;
+                       lock->l_lvb_data = lmm;
+                       lock->l_lvb_len = lvb_len;
+                       lmm = NULL;
+               }
+               unlock_res_and_lock(lock);
+               if (lmm != NULL)
+                       OBD_FREE_LARGE(lmm, lvb_len);
+       }
+       if (lock != NULL)
+               LDLM_LOCK_PUT(lock);
+
+       RETURN(rc);
 }
-/*
- * This long block is all about fixing up the lock and request state
- * so that it is correct as of the moment _before_ the operation was
- * applied; that way, the VFS will think that everything is normal and
- * call Lustre's regular VFS methods.
- *
- * If we're performing a creation, that means that unless the creation
- * failed with EEXIST, we should fake up a negative dentry.
- *
- * For everything else, we want to lookup to succeed.
- *
- * One additional note: if CREATE or OPEN succeeded, we add an extra
- * reference to the request because we need to keep it around until
- * ll_create/ll_open gets called.
- *
- * The server will return to us, in it_disposition, an indication of
- * exactly what d.lustre.it_status refers to.
- *
- * If DISP_OPEN_OPEN is set, then d.lustre.it_status refers to the open() call,
- * otherwise if DISP_OPEN_CREATE is set, then it status is the
- * creation failure mode.  In either case, one of DISP_LOOKUP_NEG or
- * DISP_LOOKUP_POS will be set, indicating whether the child lookup
- * was successful.
- *
- * Else, if DISP_LOOKUP_EXECD then d.lustre.it_status is the rc of the
- * child lookup.
- */
-int mdc_intent_lock(struct obd_export *exp, struct md_op_data *op_data,
-                    void *lmm, int lmmsize, struct lookup_intent *it,
-                    int lookup_flags, struct ptlrpc_request **reqp,
-                    ldlm_blocking_callback cb_blocking,
-                    int extra_lock_flags)
+
+/* We always reserve enough space in the reply packet for a stripe MD, because
+ * we don't know in advance the file type. */
+int mdc_enqueue(struct obd_export *exp,
+               struct ldlm_enqueue_info *einfo,
+               const union ldlm_policy_data *policy,
+               struct lookup_intent *it, struct md_op_data *op_data,
+               struct lustre_handle *lockh, __u64 extra_lock_flags)
 {
-        struct ptlrpc_request *request;
-        struct lustre_handle old_lock;
-        struct lustre_handle lockh;
-        struct mdt_body *mdt_body;
-        struct ldlm_lock *lock;
-        int rc = 0;
+        struct obd_device     *obddev = class_exp2obd(exp);
+        struct ptlrpc_request *req = NULL;
+       __u64                  flags, saved_flags = extra_lock_flags;
+        int                    rc;
+        struct ldlm_res_id res_id;
+        static const ldlm_policy_data_t lookup_policy =
+                            { .l_inodebits = { MDS_INODELOCK_LOOKUP } };
+        static const ldlm_policy_data_t update_policy =
+                            { .l_inodebits = { MDS_INODELOCK_UPDATE } };
+       static const ldlm_policy_data_t layout_policy =
+                           { .l_inodebits = { MDS_INODELOCK_LAYOUT } };
+       static const ldlm_policy_data_t getxattr_policy = {
+                             .l_inodebits = { MDS_INODELOCK_XATTR } };
+        int                    generation, resends = 0;
+        struct ldlm_reply     *lockrep;
+       enum lvb_type          lvb_type = 0;
         ENTRY;
-        LASSERT(it);
-
-        CDEBUG(D_DLMTRACE, "(name: %.*s,"DFID") in obj "DFID
-               ", intent: %s flags %#o\n", op_data->op_namelen,
-               op_data->op_name, PFID(&op_data->op_fid2),
-               PFID(&op_data->op_fid1), ldlm_it2str(it->it_op),
-               it->it_flags);
-
-        if (fid_is_sane(&op_data->op_fid2) &&
-            (it->it_op & (IT_LOOKUP | IT_GETATTR))) {
-                /* We could just return 1 immediately, but since we should only
-                 * be called in revalidate_it if we already have a lock, let's
-                 * verify that. */
-                struct ldlm_res_id res_id = { .name = { fid_seq(&op_data->op_fid2),
-                                                        fid_oid(&op_data->op_fid2),
-                                                        fid_ver(&op_data->op_fid2) } };
-                ldlm_policy_data_t policy;
-                ldlm_mode_t mode = LCK_CR;
-
-                /* As not all attributes are kept under update lock, e.g.
-                   owner/group/acls are under lookup lock, we need both
-                   ibits for GETATTR. */
-
-                /* For CMD, UPDATE lock and LOOKUP lock can not be got
-                 * at the same for cross-object, so we can not match
-                 * the 2 lock at the same time FIXME: but how to handle
-                 * the above situation */
-                policy.l_inodebits.bits = (it->it_op == IT_GETATTR) ?
-                        MDS_INODELOCK_UPDATE : MDS_INODELOCK_LOOKUP;
-
-                rc = ldlm_lock_match(exp->exp_obd->obd_namespace,
-                                     LDLM_FL_BLOCK_GRANTED, &res_id,
-                                     LDLM_IBITS, &policy, mode, &lockh);
-                if (!rc) {
-                        mode = LCK_CW;
-                        rc = ldlm_lock_match(exp->exp_obd->obd_namespace,
-                                             LDLM_FL_BLOCK_GRANTED, &res_id,
-                                             LDLM_IBITS, &policy, mode, &lockh);
-                }
-                if (!rc) {
-                        mode = LCK_PR;
-                        rc = ldlm_lock_match(exp->exp_obd->obd_namespace,
-                                             LDLM_FL_BLOCK_GRANTED, &res_id,
-                                             LDLM_IBITS, &policy, mode, &lockh);
-                }
 
-                if (!rc) {
-                        mode = LCK_PW;
-                        rc = ldlm_lock_match(exp->exp_obd->obd_namespace,
-                                             LDLM_FL_BLOCK_GRANTED, &res_id,
-                                             LDLM_IBITS, &policy, mode, &lockh);
-                }
+        LASSERTF(!it || einfo->ei_type == LDLM_IBITS, "lock type %d\n",
+                 einfo->ei_type);
+        fid_build_reg_res_name(&op_data->op_fid1, &res_id);
+
+       if (it != NULL) {
+               LASSERT(policy == NULL);
+
+               saved_flags |= LDLM_FL_HAS_INTENT;
+               if (it->it_op & (IT_OPEN | IT_UNLINK | IT_GETATTR | IT_READDIR))
+                       policy = &update_policy;
+               else if (it->it_op & IT_LAYOUT)
+                       policy = &layout_policy;
+               else if (it->it_op & (IT_GETXATTR | IT_SETXATTR))
+                       policy = &getxattr_policy;
+               else
+                       policy = &lookup_policy;
+       }
+
+        generation = obddev->u.cli.cl_import->imp_generation;
+resend:
+        flags = saved_flags;
+       if (it == NULL) {
+               /* The only way right now is FLOCK. */
+               LASSERTF(einfo->ei_type == LDLM_FLOCK, "lock type %d\n",
+                        einfo->ei_type);
+               res_id.name[3] = LDLM_FLOCK;
+       } else if (it->it_op & IT_OPEN) {
+               req = mdc_intent_open_pack(exp, it, op_data);
+       } else if (it->it_op & IT_UNLINK) {
+               req = mdc_intent_unlink_pack(exp, it, op_data);
+       } else if (it->it_op & (IT_GETATTR | IT_LOOKUP)) {
+               req = mdc_intent_getattr_pack(exp, it, op_data);
+       } else if (it->it_op & IT_READDIR) {
+               req = mdc_enqueue_pack(exp, 0);
+       } else if (it->it_op & IT_LAYOUT) {
+               if (!imp_connect_lvb_type(class_exp2cliimp(exp)))
+                       RETURN(-EOPNOTSUPP);
+               req = mdc_intent_layout_pack(exp, it, op_data);
+               lvb_type = LVB_T_LAYOUT;
+       } else if (it->it_op & IT_GETXATTR) {
+               req = mdc_intent_getxattr_pack(exp, it, op_data);
+       } else {
+                LBUG();
+                RETURN(-EINVAL);
+        }
 
-                if (rc) {
-                        memcpy(&it->d.lustre.it_lock_handle, &lockh,
-                               sizeof(lockh));
-                        it->d.lustre.it_lock_mode = mode;
-                }
+        if (IS_ERR(req))
+                RETURN(PTR_ERR(req));
 
-                /* Only return failure if it was not GETATTR by cfid
-                   (from inode_revalidate) */
-                if (rc || op_data->op_namelen != 0)
+       if (req != NULL && it && it->it_op & IT_CREAT)
+               /* ask ptlrpc not to resend on EINPROGRESS since we have our own
+                * retry logic */
+               req->rq_no_retry_einprogress = 1;
+
+        if (resends) {
+                req->rq_generation_set = 1;
+                req->rq_import_generation = generation;
+                req->rq_sent = cfs_time_current_sec() + resends;
+        }
+
+        /* It is important to obtain rpc_lock first (if applicable), so that
+         * threads that are serialised with rpc_lock are not polluting our
+         * rpcs in flight counter. We do not do flock request limiting, though*/
+        if (it) {
+                mdc_get_rpc_lock(obddev->u.cli.cl_rpc_lock, it);
+               rc = obd_get_request_slot(&obddev->u.cli);
+                if (rc != 0) {
+                        mdc_put_rpc_lock(obddev->u.cli.cl_rpc_lock, it);
+                        mdc_clear_replay_flag(req, 0);
+                        ptlrpc_req_finished(req);
                         RETURN(rc);
+                }
         }
 
-        /* lookup_it may be called only after revalidate_it has run, because
-         * revalidate_it cannot return errors, only zero.  Returning zero causes
-         * this call to lookup, which *can* return an error.
-         *
-         * We only want to execute the request associated with the intent one
-         * time, however, so don't send the request again.  Instead, skip past
-         * this and use the request from revalidate.  In this case, revalidate
-         * never dropped its reference, so the refcounts are all OK */
-        if (!it_disposition(it, DISP_ENQ_COMPLETE)) {
-                struct ldlm_enqueue_info einfo =
-                        { LDLM_IBITS, it_to_lock_mode(it), cb_blocking,
-                          ldlm_completion_ast, NULL, NULL };
-
-                /* For case if upper layer did not alloc fid, do it now. */
-                if (!fid_is_sane(&op_data->op_fid2) && it->it_op & IT_CREAT) {
-                        rc = mdc_fid_alloc(exp, &op_data->op_fid2, op_data);
-                        if (rc < 0) {
-                                CERROR("Can't alloc new fid, rc %d\n", rc);
-                                RETURN(rc);
-                        }
+        rc = ldlm_cli_enqueue(exp, &req, einfo, &res_id, policy, &flags, NULL,
+                             0, lvb_type, lockh, 0);
+        if (!it) {
+                /* For flock requests we immediatelly return without further
+                   delay and let caller deal with the rest, since rest of
+                   this function metadata processing makes no sense for flock
+                  requests anyway. But in case of problem during comms with
+                  Server (ETIMEDOUT) or any signal/kill attempt (EINTR), we
+                  can not rely on caller and this mainly for F_UNLCKs
+                  (explicits or automatically generated by Kernel to clean
+                  current FLocks upon exit) that can't be trashed */
+               if (((rc == -EINTR) || (rc == -ETIMEDOUT)) &&
+                   (einfo->ei_type == LDLM_FLOCK) &&
+                   (einfo->ei_mode == LCK_NL))
+                       goto resend;
+               RETURN(rc);
+       }
+
+       obd_put_request_slot(&obddev->u.cli);
+       mdc_put_rpc_lock(obddev->u.cli.cl_rpc_lock, it);
+
+       if (rc < 0) {
+               CDEBUG(D_INFO, "%s: ldlm_cli_enqueue failed: rc = %d\n",
+                      obddev->obd_name, rc);
+
+               mdc_clear_replay_flag(req, rc);
+               ptlrpc_req_finished(req);
+               RETURN(rc);
+       }
+
+       lockrep = req_capsule_server_get(&req->rq_pill, &RMF_DLM_REP);
+       LASSERT(lockrep != NULL);
+
+       lockrep->lock_policy_res2 =
+               ptlrpc_status_ntoh(lockrep->lock_policy_res2);
+
+        /* Retry the create infinitely when we get -EINPROGRESS from
+         * server. This is required by the new quota design. */
+        if (it && it->it_op & IT_CREAT &&
+            (int)lockrep->lock_policy_res2 == -EINPROGRESS) {
+                mdc_clear_replay_flag(req, rc);
+                ptlrpc_req_finished(req);
+                resends++;
+
+                CDEBUG(D_HA, "%s: resend:%d op:%d "DFID"/"DFID"\n",
+                       obddev->obd_name, resends, it->it_op,
+                       PFID(&op_data->op_fid1), PFID(&op_data->op_fid2));
+
+                if (generation == obddev->u.cli.cl_import->imp_generation) {
+                        goto resend;
+                } else {
+                       CDEBUG(D_HA, "resend cross eviction\n");
+                        RETURN(-EIO);
                 }
-                rc = mdc_enqueue(exp, &einfo, it, op_data, &lockh,
-                                 lmm, lmmsize, extra_lock_flags);
-                if (rc < 0)
-                        RETURN(rc);
-                memcpy(&it->d.lustre.it_lock_handle, &lockh, sizeof(lockh));
-        } else if (!fid_is_sane(&op_data->op_fid2) ||
-                   !(it->it_flags & O_CHECK_STALE)) {
-                /* DISP_ENQ_COMPLETE set means there is extra reference on
-                 * request referenced from this intent, saved for subsequent
-                 * lookup.  This path is executed when we proceed to this
-                 * lookup, so we clear DISP_ENQ_COMPLETE */
-                it_clear_disposition(it, DISP_ENQ_COMPLETE);
         }
-        request = *reqp = it->d.lustre.it_data;
+
+       rc = mdc_finish_enqueue(exp, req, einfo, it, lockh, rc);
+       if (rc < 0) {
+               if (lustre_handle_is_used(lockh)) {
+                       ldlm_lock_decref(lockh, einfo->ei_mode);
+                       memset(lockh, 0, sizeof(*lockh));
+               }
+               ptlrpc_req_finished(req);
+       }
+       RETURN(rc);
+}
+
+static int mdc_finish_intent_lock(struct obd_export *exp,
+                                  struct ptlrpc_request *request,
+                                  struct md_op_data *op_data,
+                                  struct lookup_intent *it,
+                                  struct lustre_handle *lockh)
+{
+        struct lustre_handle old_lock;
+        struct mdt_body *mdt_body;
+        struct ldlm_lock *lock;
+        int rc;
+       ENTRY;
+
         LASSERT(request != NULL);
         LASSERT(request != LP_POISON);
         LASSERT(request->rq_repmsg != LP_POISON);
 
+       if (it->it_op & IT_READDIR)
+               RETURN(0);
+
         if (!it_disposition(it, DISP_IT_EXECD)) {
                 /* The server failed before it even started executing the
                  * intent, i.e. because it couldn't unpack the request. */
@@ -771,21 +972,8 @@ int mdc_intent_lock(struct obd_export *exp, struct md_op_data *op_data,
         if (rc)
                 RETURN(rc);
 
-        mdt_body = lustre_msg_buf(request->rq_repmsg, DLM_REPLY_REC_OFF,
-                                  sizeof(*mdt_body));
+        mdt_body = req_capsule_server_get(&request->rq_pill, &RMF_MDT_BODY);
         LASSERT(mdt_body != NULL);      /* mdc_enqueue checked */
-        LASSERT_REPSWABBED(request, 1); /* mdc_enqueue swabbed */
-
-        /* If we were revalidating a fid/name pair, mark the intent in
-         * case we fail and get called again from lookup */
-        if (fid_is_sane(&op_data->op_fid2) && (it->it_flags & O_CHECK_STALE) &&
-            (it->it_op != IT_GETATTR)) {
-                it_set_disposition(it, DISP_ENQ_COMPLETE);
-
-                /* Also: did we find the same inode? */
-                if (!lu_fid_eq(&op_data->op_fid2, &mdt_body->fid1))
-                        RETURN(-ESTALE);
-        }
 
         rc = it_open_error(DISP_LOOKUP_EXECD, it);
         if (rc)
@@ -814,7 +1002,7 @@ int mdc_intent_lock(struct obd_export *exp, struct md_op_data *op_data,
         } else if (it->it_op == IT_OPEN) {
                 LASSERT(!it_disposition(it, DISP_OPEN_CREATE));
         } else {
-                LASSERT(it->it_op & (IT_GETATTR | IT_LOOKUP));
+                LASSERT(it->it_op & (IT_GETATTR | IT_LOOKUP | IT_LAYOUT));
         }
 
         /* If we already have a matching lock, then cancel the new
@@ -822,35 +1010,282 @@ int mdc_intent_lock(struct obd_export *exp, struct md_op_data *op_data,
          * mdc_enqueue, because we need to use the child's inode as
          * the l_ast_data to match, and that's not available until
          * intent_finish has performed the iget().) */
-        lock = ldlm_handle2lock(&lockh);
+        lock = ldlm_handle2lock(lockh);
         if (lock) {
                 ldlm_policy_data_t policy = lock->l_policy_data;
                 LDLM_DEBUG(lock, "matching against this");
 
-                LASSERTF(fid_res_name_eq(&mdt_body->fid1,
-                                         &lock->l_resource->lr_name),
-                         "Lock res_id: %lu/%lu/%lu, fid: %lu/%lu/%lu.\n",
-                         (unsigned long)lock->l_resource->lr_name.name[0],
-                         (unsigned long)lock->l_resource->lr_name.name[1],
-                         (unsigned long)lock->l_resource->lr_name.name[2],
-                         (unsigned long)fid_seq(&mdt_body->fid1),
-                         (unsigned long)fid_oid(&mdt_body->fid1),
-                         (unsigned long)fid_ver(&mdt_body->fid1));
-                LDLM_LOCK_PUT(lock);
-
-                memcpy(&old_lock, &lockh, sizeof(lockh));
+               LASSERTF(fid_res_name_eq(&mdt_body->mbo_fid1,
+                                        &lock->l_resource->lr_name),
+                        "Lock res_id: "DLDLMRES", fid: "DFID"\n",
+                        PLDLMRES(lock->l_resource), PFID(&mdt_body->mbo_fid1));
+               LDLM_LOCK_PUT(lock);
+
+                memcpy(&old_lock, lockh, sizeof(*lockh));
                 if (ldlm_lock_match(NULL, LDLM_FL_BLOCK_GRANTED, NULL,
-                                    LDLM_IBITS, &policy, LCK_NL, &old_lock)) {
-                        ldlm_lock_decref_and_cancel(&lockh,
+                                    LDLM_IBITS, &policy, LCK_NL, &old_lock, 0)) {
+                        ldlm_lock_decref_and_cancel(lockh,
                                                     it->d.lustre.it_lock_mode);
-                        memcpy(&lockh, &old_lock, sizeof(old_lock));
-                        memcpy(&it->d.lustre.it_lock_handle, &lockh,
-                               sizeof(lockh));
+                        memcpy(lockh, &old_lock, sizeof(old_lock));
+                        it->d.lustre.it_lock_handle = lockh->cookie;
                 }
         }
         CDEBUG(D_DENTRY,"D_IT dentry %.*s intent: %s status %d disp %x rc %d\n",
                op_data->op_namelen, op_data->op_name, ldlm_it2str(it->it_op),
                it->d.lustre.it_status, it->d.lustre.it_disposition, rc);
+        RETURN(rc);
+}
+
+int mdc_revalidate_lock(struct obd_export *exp, struct lookup_intent *it,
+                        struct lu_fid *fid, __u64 *bits)
+{
+        /* We could just return 1 immediately, but since we should only
+         * be called in revalidate_it if we already have a lock, let's
+         * verify that. */
+        struct ldlm_res_id res_id;
+        struct lustre_handle lockh;
+        ldlm_policy_data_t policy;
+        ldlm_mode_t mode;
+        ENTRY;
+
+        if (it->d.lustre.it_lock_handle) {
+                lockh.cookie = it->d.lustre.it_lock_handle;
+                mode = ldlm_revalidate_lock_handle(&lockh, bits);
+        } else {
+                fid_build_reg_res_name(fid, &res_id);
+                switch (it->it_op) {
+               case IT_GETATTR:
+                       /* File attributes are held under multiple bits:
+                        * nlink is under lookup lock, size and times are
+                        * under UPDATE lock and recently we've also got
+                        * a separate permissions lock for owner/group/acl that
+                        * were protected by lookup lock before.
+                        * Getattr must provide all of that information,
+                        * so we need to ensure we have all of those locks.
+                        * Unfortunately, if the bits are split across multiple
+                        * locks, there's no easy way to match all of them here,
+                        * so an extra RPC would be performed to fetch all
+                        * of those bits at once for now. */
+                       /* For new MDTs(> 2.4), UPDATE|PERM should be enough,
+                        * but for old MDTs (< 2.4), permission is covered
+                        * by LOOKUP lock, so it needs to match all bits here.*/
+                       policy.l_inodebits.bits = MDS_INODELOCK_UPDATE |
+                                                 MDS_INODELOCK_LOOKUP |
+                                                 MDS_INODELOCK_PERM;
+                       break;
+               case IT_READDIR:
+                       policy.l_inodebits.bits = MDS_INODELOCK_UPDATE;
+                       break;
+                case IT_LAYOUT:
+                        policy.l_inodebits.bits = MDS_INODELOCK_LAYOUT;
+                        break;
+                default:
+                        policy.l_inodebits.bits = MDS_INODELOCK_LOOKUP;
+                        break;
+                }
 
+               mode = mdc_lock_match(exp, LDLM_FL_BLOCK_GRANTED, fid,
+                                     LDLM_IBITS, &policy,
+                                     LCK_CR | LCK_CW | LCK_PR | LCK_PW,
+                                     &lockh);
+        }
+
+        if (mode) {
+                it->d.lustre.it_lock_handle = lockh.cookie;
+                it->d.lustre.it_lock_mode = mode;
+        } else {
+                it->d.lustre.it_lock_handle = 0;
+                it->d.lustre.it_lock_mode = 0;
+        }
+
+        RETURN(!!mode);
+}
+
+/*
+ * This long block is all about fixing up the lock and request state
+ * so that it is correct as of the moment _before_ the operation was
+ * applied; that way, the VFS will think that everything is normal and
+ * call Lustre's regular VFS methods.
+ *
+ * If we're performing a creation, that means that unless the creation
+ * failed with EEXIST, we should fake up a negative dentry.
+ *
+ * For everything else, we want to lookup to succeed.
+ *
+ * One additional note: if CREATE or OPEN succeeded, we add an extra
+ * reference to the request because we need to keep it around until
+ * ll_create/ll_open gets called.
+ *
+ * The server will return to us, in it_disposition, an indication of
+ * exactly what d.lustre.it_status refers to.
+ *
+ * If DISP_OPEN_OPEN is set, then d.lustre.it_status refers to the open() call,
+ * otherwise if DISP_OPEN_CREATE is set, then it status is the
+ * creation failure mode.  In either case, one of DISP_LOOKUP_NEG or
+ * DISP_LOOKUP_POS will be set, indicating whether the child lookup
+ * was successful.
+ *
+ * Else, if DISP_LOOKUP_EXECD then d.lustre.it_status is the rc of the
+ * child lookup.
+ */
+int mdc_intent_lock(struct obd_export *exp, struct md_op_data *op_data,
+                   struct lookup_intent *it, struct ptlrpc_request **reqp,
+                   ldlm_blocking_callback cb_blocking, __u64 extra_lock_flags)
+{
+       struct ldlm_enqueue_info einfo = {
+               .ei_type        = LDLM_IBITS,
+               .ei_mode        = it_to_lock_mode(it),
+               .ei_cb_bl       = cb_blocking,
+               .ei_cb_cp       = ldlm_completion_ast,
+       };
+       struct lustre_handle lockh;
+       int rc = 0;
+       ENTRY;
+       LASSERT(it);
+
+       CDEBUG(D_DLMTRACE, "(name: %.*s,"DFID") in obj "DFID
+               ", intent: %s flags %#"LPF64"o\n", op_data->op_namelen,
+               op_data->op_name, PFID(&op_data->op_fid2),
+               PFID(&op_data->op_fid1), ldlm_it2str(it->it_op),
+               it->it_flags);
+
+       lockh.cookie = 0;
+       if (fid_is_sane(&op_data->op_fid2) &&
+           (it->it_op & (IT_LOOKUP | IT_GETATTR | IT_READDIR))) {
+               /* We could just return 1 immediately, but since we should only
+                * be called in revalidate_it if we already have a lock, let's
+                * verify that. */
+               it->d.lustre.it_lock_handle = 0;
+               rc = mdc_revalidate_lock(exp, it, &op_data->op_fid2, NULL);
+               /* Only return failure if it was not GETATTR by cfid
+                  (from inode_revalidate) */
+               if (rc || op_data->op_namelen != 0)
+                       RETURN(rc);
+       }
+
+       /* For case if upper layer did not alloc fid, do it now. */
+       if (!fid_is_sane(&op_data->op_fid2) && it->it_op & IT_CREAT) {
+               rc = mdc_fid_alloc(NULL, exp, &op_data->op_fid2, op_data);
+               if (rc < 0) {
+                       CERROR("Can't alloc new fid, rc %d\n", rc);
+                       RETURN(rc);
+               }
+       }
+
+       rc = mdc_enqueue(exp, &einfo, NULL, it, op_data, &lockh,
+                        extra_lock_flags);
+       if (rc < 0)
+               RETURN(rc);
+
+       *reqp = it->d.lustre.it_data;
+        rc = mdc_finish_intent_lock(exp, *reqp, op_data, it, &lockh);
         RETURN(rc);
 }
+
+static int mdc_intent_getattr_async_interpret(const struct lu_env *env,
+                                              struct ptlrpc_request *req,
+                                              void *args, int rc)
+{
+        struct mdc_getattr_args  *ga = args;
+        struct obd_export        *exp = ga->ga_exp;
+        struct md_enqueue_info   *minfo = ga->ga_minfo;
+        struct ldlm_enqueue_info *einfo = ga->ga_einfo;
+        struct lookup_intent     *it;
+        struct lustre_handle     *lockh;
+        struct obd_device        *obddev;
+       struct ldlm_reply        *lockrep;
+       __u64                     flags = LDLM_FL_HAS_INTENT;
+        ENTRY;
+
+        it    = &minfo->mi_it;
+        lockh = &minfo->mi_lockh;
+
+        obddev = class_exp2obd(exp);
+
+       obd_put_request_slot(&obddev->u.cli);
+        if (OBD_FAIL_CHECK(OBD_FAIL_MDC_GETATTR_ENQUEUE))
+                rc = -ETIMEDOUT;
+
+        rc = ldlm_cli_enqueue_fini(exp, req, einfo->ei_type, 1, einfo->ei_mode,
+                                   &flags, NULL, 0, lockh, rc);
+        if (rc < 0) {
+                CERROR("ldlm_cli_enqueue_fini: %d\n", rc);
+                mdc_clear_replay_flag(req, rc);
+                GOTO(out, rc);
+        }
+
+       lockrep = req_capsule_server_get(&req->rq_pill, &RMF_DLM_REP);
+       LASSERT(lockrep != NULL);
+
+       lockrep->lock_policy_res2 =
+               ptlrpc_status_ntoh(lockrep->lock_policy_res2);
+
+        rc = mdc_finish_enqueue(exp, req, einfo, it, lockh, rc);
+        if (rc)
+                GOTO(out, rc);
+
+        rc = mdc_finish_intent_lock(exp, req, &minfo->mi_data, it, lockh);
+        EXIT;
+
+out:
+        OBD_FREE_PTR(einfo);
+        minfo->mi_cb(req, minfo, rc);
+        return 0;
+}
+
+int mdc_intent_getattr_async(struct obd_export *exp,
+                             struct md_enqueue_info *minfo,
+                             struct ldlm_enqueue_info *einfo)
+{
+        struct md_op_data       *op_data = &minfo->mi_data;
+        struct lookup_intent    *it = &minfo->mi_it;
+        struct ptlrpc_request   *req;
+        struct mdc_getattr_args *ga;
+        struct obd_device       *obddev = class_exp2obd(exp);
+        struct ldlm_res_id       res_id;
+        /*XXX: Both MDS_INODELOCK_LOOKUP and MDS_INODELOCK_UPDATE are needed
+         *     for statahead currently. Consider CMD in future, such two bits
+         *     maybe managed by different MDS, should be adjusted then. */
+        ldlm_policy_data_t       policy = {
+                                        .l_inodebits = { MDS_INODELOCK_LOOKUP | 
+                                                         MDS_INODELOCK_UPDATE }
+                                 };
+        int                      rc = 0;
+       __u64                    flags = LDLM_FL_HAS_INTENT;
+       ENTRY;
+
+       CDEBUG(D_DLMTRACE, "name: %.*s in inode "DFID", intent: %s flags %#"
+               LPF64"o\n",
+               op_data->op_namelen, op_data->op_name, PFID(&op_data->op_fid1),
+               ldlm_it2str(it->it_op), it->it_flags);
+
+       fid_build_reg_res_name(&op_data->op_fid1, &res_id);
+       req = mdc_intent_getattr_pack(exp, it, op_data);
+       if (IS_ERR(req))
+               RETURN(PTR_ERR(req));
+
+       rc = obd_get_request_slot(&obddev->u.cli);
+        if (rc != 0) {
+                ptlrpc_req_finished(req);
+                RETURN(rc);
+        }
+
+        rc = ldlm_cli_enqueue(exp, &req, einfo, &res_id, &policy, &flags, NULL,
+                             0, LVB_T_NONE, &minfo->mi_lockh, 1);
+        if (rc < 0) {
+               obd_put_request_slot(&obddev->u.cli);
+                ptlrpc_req_finished(req);
+                RETURN(rc);
+        }
+
+        CLASSERT(sizeof(*ga) <= sizeof(req->rq_async_args));
+        ga = ptlrpc_req_async_args(req);
+        ga->ga_exp = exp;
+        ga->ga_minfo = minfo;
+        ga->ga_einfo = einfo;
+
+        req->rq_interpret_reply = mdc_intent_getattr_async_interpret;
+        ptlrpcd_add_req(req, PDL_POLICY_LOCAL, -1);
+
+        RETURN(0);
+}