Whamcloud - gitweb
LU-5319 mdc: manage number of modify RPCs in flight
[fs/lustre-release.git] / lustre / mdc / mdc_reint.c
index 6f8fc77..31985a2 100644 (file)
@@ -1,63 +1,61 @@
-/* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
- * vim:expandtab:shiftwidth=8:tabstop=8:
+/*
+ * GPL HEADER START
  *
- * Copyright (C) 2001-2003 Cluster File Systems, Inc.
+ * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
  *
- *   This file is part of the Lustre file system, http://www.lustre.org
- *   Lustre is a trademark of Cluster File Systems, Inc.
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License version 2 only,
+ * as published by the Free Software Foundation.
  *
- *   You may have signed or agreed to another license before downloading
- *   this software.  If so, you are bound by the terms and conditions
- *   of that agreement, and the following does not apply to you.  See the
- *   LICENSE file included with this distribution for more information.
+ * This program is distributed in the hope that it will be useful, but
+ * WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * General Public License version 2 for more details (a copy is included
+ * in the LICENSE file that accompanied this code).
  *
- *   If you did not agree to a different license, then this copy of Lustre
- *   is open source software; you can redistribute it and/or modify it
- *   under the terms of version 2 of the GNU General Public License as
- *   published by the Free Software Foundation.
+ * You should have received a copy of the GNU General Public License
+ * version 2 along with this program; If not, see
+ * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
  *
- *   In either case, Lustre is distributed in the hope that it will be
- *   useful, but WITHOUT ANY WARRANTY; without even the implied warranty
- *   of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
- *   license text for more details.
+ * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
+ * CA 95054 USA or visit www.sun.com if you need additional information or
+ * have any questions.
+ *
+ * GPL HEADER END
+ */
+/*
+ * Copyright (c) 2002, 2010, Oracle and/or its affiliates. All rights reserved.
+ * Use is subject to license terms.
+ *
+ * Copyright (c) 2012, 2014, 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__
-#ifdef HAVE_KERNEL_CONFIG_H
-# include <linux/config.h>
-#endif
-# include <linux/module.h>
-# include <linux/kernel.h>
-#else
-# include <liblustre.h>
-#endif
+#include <linux/module.h>
+#include <linux/kernel.h>
 
 #include <obd_class.h>
 #include "mdc_internal.h"
 #include <lustre_fid.h>
 
 /* mdc_setattr does its own semaphore handling */
-static int mdc_reint(struct ptlrpc_request *request,
-                     struct mdc_rpc_lock *rpc_lock, int level)
+static int mdc_reint(struct ptlrpc_request *request, int level)
 {
         int rc;
 
         request->rq_send_state = level;
 
-        mdc_get_rpc_lock(rpc_lock, NULL);
-        rc = ptlrpc_queue_wait(request);
-        mdc_put_rpc_lock(rpc_lock, NULL);
+       mdc_get_mod_rpc_slot(request, NULL);
+       rc = ptlrpc_queue_wait(request);
+       mdc_put_mod_rpc_slot(request, NULL);
         if (rc)
                 CDEBUG(D_INFO, "error in handling %d\n", rc);
-        else if (!lustre_swab_repbuf(request, REPLY_REC_OFF,
-                                     sizeof(struct mdt_body),
-                                     lustre_swab_mdt_body)) {
-                CERROR ("Can't unpack mdt_body\n");
+        else if (!req_capsule_server_get(&request->rq_pill, &RMF_MDT_BODY)) {
                 rc = -EPROTO;
         }
         return rc;
@@ -66,198 +64,186 @@ static int mdc_reint(struct ptlrpc_request *request,
 /* Find and cancel locally locks matched by inode @bits & @mode in the resource
  * found by @fid. Found locks are added into @cancel list. Returns the amount of
  * locks added to @cancels list. */
-int mdc_resource_get_unused(struct obd_export *exp, struct lu_fid *fid,
-                            struct list_head *cancels, ldlm_mode_t mode,
+int mdc_resource_get_unused(struct obd_export *exp, const struct lu_fid *fid,
+                           struct list_head *cancels, ldlm_mode_t mode,
                             __u64 bits)
 {
-        ldlm_policy_data_t policy = {{0}};
-        struct ldlm_res_id res_id;
-        struct ldlm_resource *res;
-        int count;
-        ENTRY;
-
-        fid_build_reg_res_name(fid, &res_id);
-        res = ldlm_resource_get(exp->exp_obd->obd_namespace,
-                                NULL, &res_id, 0, 0);
-        if (res == NULL)
-                RETURN(0);
-
-        /* Initialize ibits lock policy. */
-        policy.l_inodebits.bits = bits;
-        count = ldlm_cancel_resource_local(res, cancels, &policy,
-                                           mode, 0, 0, NULL);
-        ldlm_resource_putref(res);
-        RETURN(count);
+       struct ldlm_namespace *ns = exp->exp_obd->obd_namespace;
+       ldlm_policy_data_t policy = { {0} };
+       struct ldlm_res_id res_id;
+       struct ldlm_resource *res;
+       int count;
+       ENTRY;
+
+       /* Return, i.e. cancel nothing, only if ELC is supported (flag in
+        * export) but disabled through procfs (flag in NS).
+        *
+        * This distinguishes from a case when ELC is not supported originally,
+        * when we still want to cancel locks in advance and just cancel them
+        * locally, without sending any RPC. */
+       if (exp_connect_cancelset(exp) && !ns_connect_cancelset(ns))
+               RETURN(0);
+
+       fid_build_reg_res_name(fid, &res_id);
+       res = ldlm_resource_get(exp->exp_obd->obd_namespace,
+                               NULL, &res_id, 0, 0);
+       if (IS_ERR(res))
+               RETURN(0);
+       LDLM_RESOURCE_ADDREF(res);
+       /* Initialize ibits lock policy. */
+       policy.l_inodebits.bits = bits;
+       count = ldlm_cancel_resource_local(res, cancels, &policy,
+                                          mode, 0, 0, NULL);
+       LDLM_RESOURCE_DELREF(res);
+       ldlm_resource_putref(res);
+       RETURN(count);
 }
 
-/* If mdc_setattr is called with an 'iattr', then it is a normal RPC that
- * should take the normal semaphore and go to the normal portal.
- *
- * If it is called with iattr->ia_valid & ATTR_FROM_OPEN, then it is a
- * magic open-path setattr that should take the setattr semaphore and
- * go to the setattr portal. */
 int mdc_setattr(struct obd_export *exp, struct md_op_data *op_data,
-                void *ea, int ealen, void *ea2, int ea2len,
-                struct ptlrpc_request **request)
+               void *ea, size_t ealen, struct ptlrpc_request **request)
 {
-        CFS_LIST_HEAD(cancels);
+       struct list_head cancels = LIST_HEAD_INIT(cancels);
         struct ptlrpc_request *req;
-        struct mdt_rec_setattr *rec;
-        struct mdc_rpc_lock *rpc_lock;
-        struct obd_device *obd = exp->exp_obd;
-        int size[7] = { sizeof(struct ptlrpc_body),
-                        sizeof(*rec), 0, 0, ealen, ea2len, 0 };
-        int count = 0, bufcount = 4, rc;
+        int count = 0, rc;
         __u64 bits;
         ENTRY;
 
         LASSERT(op_data != NULL);
 
-        size[REQ_REC_OFF + 1] = op_data->op_capa1 ?
-                sizeof(struct lustre_capa) : 0;
-
-        if (op_data->op_flags & (MF_SOM_CHANGE | MF_EPOCH_OPEN))
-                size[REQ_REC_OFF + 2] = sizeof(struct mdt_epoch);
-
-        if (ealen > 0) {
-                bufcount++;
-                if (ea2len > 0)
-                        bufcount++;
-        }
-
         bits = MDS_INODELOCK_UPDATE;
         if (op_data->op_attr.ia_valid & (ATTR_MODE|ATTR_UID|ATTR_GID))
                 bits |= MDS_INODELOCK_LOOKUP;
-        if ((op_data->op_flags & MF_MDC_CANCEL_FID1) && 
-            (fid_is_sane(&op_data->op_fid1)))
-                count = mdc_resource_get_unused(exp, &op_data->op_fid1,
-                                                &cancels, LCK_EX, bits);
-        if (exp_connect_cancelset(exp) && count) {
-                bufcount = 7;
-                size[REQ_REC_OFF + 5] = ldlm_request_bufsize(count, MDS_REINT);
-        }
-        req = ptlrpc_prep_req(class_exp2cliimp(exp), LUSTRE_MDS_VERSION,
-                              MDS_REINT, bufcount, size, NULL);
-        if (exp_connect_cancelset(exp) && req)
-                ldlm_cli_cancel_list(&cancels, count, req, REQ_REC_OFF + 5, 0);
-        else
+       if ((op_data->op_flags & MF_MDC_CANCEL_FID1) &&
+           (fid_is_sane(&op_data->op_fid1)))
+               count = mdc_resource_get_unused(exp, &op_data->op_fid1,
+                                               &cancels, LCK_EX, bits);
+        req = ptlrpc_request_alloc(class_exp2cliimp(exp),
+                                   &RQF_MDS_REINT_SETATTR);
+        if (req == NULL) {
                 ldlm_lock_list_put(&cancels, l_bl_ast, count);
-
-        if (req == NULL)
                 RETURN(-ENOMEM);
-
-        if (op_data->op_attr.ia_valid & ATTR_FROM_OPEN) {
-                req->rq_request_portal = MDS_SETATTR_PORTAL; //XXX FIXME bug 249
-                rpc_lock = obd->u.cli.cl_setattr_lock;
-        } else {
-                rpc_lock = obd->u.cli.cl_rpc_lock;
         }
 
+       req_capsule_set_size(&req->rq_pill, &RMF_MDT_EPOCH, RCL_CLIENT, 0);
+       req_capsule_set_size(&req->rq_pill, &RMF_EADATA, RCL_CLIENT, ealen);
+       req_capsule_set_size(&req->rq_pill, &RMF_LOGCOOKIES, RCL_CLIENT, 0);
+
+       rc = mdc_prep_elc_req(exp, req, MDS_REINT, &cancels, count);
+       if (rc) {
+               ptlrpc_request_free(req);
+               RETURN(rc);
+       }
+
         if (op_data->op_attr.ia_valid & (ATTR_MTIME | ATTR_CTIME))
-                CDEBUG(D_INODE, "setting mtime %lu, ctime %lu\n",
+                CDEBUG(D_INODE, "setting mtime "CFS_TIME_T
+                       ", ctime "CFS_TIME_T"\n",
                        LTIME_S(op_data->op_attr.ia_mtime),
                        LTIME_S(op_data->op_attr.ia_ctime));
-        mdc_setattr_pack(req, REQ_REC_OFF, op_data, ea, ealen, ea2, ea2len);
+       mdc_setattr_pack(req, op_data, ea, ealen);
 
-        size[REPLY_REC_OFF] = sizeof(struct mdt_body);
-        size[REPLY_REC_OFF + 1] = sizeof(struct lustre_capa);
-        ptlrpc_req_set_repsize(req, 3, size);
+        ptlrpc_request_set_replen(req);
 
-        rc = mdc_reint(req, rpc_lock, LUSTRE_IMP_FULL);
-        *request = req;
-        if (rc == -ERESTARTSYS)
+       rc = mdc_reint(req, LUSTRE_IMP_FULL);
+       if (rc == -ERESTARTSYS)
                 rc = 0;
 
-        RETURN(rc);
+        *request = req;
+
+       RETURN(rc);
 }
 
 int mdc_create(struct obd_export *exp, struct md_op_data *op_data,
-               const void *data, int datalen, int mode, __u32 uid, __u32 gid,
-               __u32 cap_effective, __u64 rdev, struct ptlrpc_request **request)
+               const void *data, size_t datalen,
+               umode_t mode, uid_t uid, gid_t gid,
+               cfs_cap_t cap_effective, __u64 rdev,
+               struct ptlrpc_request **request)
 {
-        int size[6] = { sizeof(struct ptlrpc_body),
-                        sizeof(struct mdt_rec_create),
-                        0, op_data->op_namelen + 1, 0, 0 };
-        struct obd_device *obd = exp->exp_obd;
-        int level, bufcount = 4, rc;
         struct ptlrpc_request *req;
-        int count = 0;
-        CFS_LIST_HEAD(cancels);
+        int level, rc;
+        int count, resends = 0;
+        struct obd_import *import = exp->exp_obd->u.cli.cl_import;
+        int generation = import->imp_generation;
+       struct list_head cancels = LIST_HEAD_INIT(cancels);
         ENTRY;
 
-        /* For case if upper layer did not alloc fid, do it now. */
-        if (!fid_is_sane(&op_data->op_fid2)) {
-                /*
-                 * mdc_fid_alloc() may return errno 1 in case of switch to new
-                 * sequence, handle this.
-                 */
-                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);
-                }
-        }
-
-        size[REQ_REC_OFF + 1] = op_data->op_capa1 ?
-                sizeof(struct lustre_capa) : 0;
-        
-        if (data && datalen) {
-                size[bufcount] = datalen;
-                bufcount++;
-        }
-
-        if ((op_data->op_flags & MF_MDC_CANCEL_FID1) && 
+       /* For case if upper layer did not alloc fid, do it now. */
+       if (!fid_is_sane(&op_data->op_fid2)) {
+               /*
+                * mdc_fid_alloc() may return errno 1 in case of switch to new
+                * sequence, handle this.
+                */
+               rc = mdc_fid_alloc(NULL, exp, &op_data->op_fid2, op_data);
+               if (rc < 0)
+                       RETURN(rc);
+       }
+
+rebuild:
+        count = 0;
+        if ((op_data->op_flags & MF_MDC_CANCEL_FID1) &&
             (fid_is_sane(&op_data->op_fid1)))
                 count = mdc_resource_get_unused(exp, &op_data->op_fid1,
                                                 &cancels, LCK_EX,
                                                 MDS_INODELOCK_UPDATE);
-        if (exp_connect_cancelset(exp) && count) {
-                bufcount = 6;
-                size[REQ_REC_OFF + 4] = ldlm_request_bufsize(count, MDS_REINT);
-        }
-        req = ptlrpc_prep_req(class_exp2cliimp(exp), LUSTRE_MDS_VERSION,
-                              MDS_REINT, bufcount, size, NULL);
-        if (exp_connect_cancelset(exp) && req)
-                ldlm_cli_cancel_list(&cancels, count, req, REQ_REC_OFF + 4, 0);
-        else
-                ldlm_lock_list_put(&cancels, l_bl_ast, count);
 
-        if (req == NULL)
+        req = ptlrpc_request_alloc(class_exp2cliimp(exp),
+                                   &RQF_MDS_REINT_CREATE_RMT_ACL);
+        if (req == NULL) {
+                ldlm_lock_list_put(&cancels, l_bl_ast, count);
                 RETURN(-ENOMEM);
+        }
+
+        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,
+                             data && datalen ? datalen : 0);
+
+       rc = mdc_prep_elc_req(exp, req, MDS_REINT, &cancels, count);
+       if (rc) {
+               ptlrpc_request_free(req);
+               RETURN(rc);
+       }
 
         /*
          * mdc_create_pack() fills msg->bufs[1] with name and msg->bufs[2] with
          * tgt, for symlinks or lov MD data.
          */
-        mdc_create_pack(req, REQ_REC_OFF, op_data, data, datalen, mode, uid,
+        mdc_create_pack(req, op_data, data, datalen, mode, uid,
                         gid, cap_effective, rdev);
 
-        size[REPLY_REC_OFF] = sizeof(struct mdt_body);
-        size[REPLY_REC_OFF + 1] = sizeof(struct lustre_capa);
-        ptlrpc_req_set_repsize(req, 3, size);
+        ptlrpc_request_set_replen(req);
 
+       /* ask ptlrpc not to resend on EINPROGRESS since we have our own retry
+        * logic here */
+       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;
+        }
         level = LUSTRE_IMP_FULL;
  resend:
-        rc = mdc_reint(req, obd->u.cli.cl_rpc_lock, level);
-        
+       rc = mdc_reint(req, level);
+
         /* Resend if we were told to. */
         if (rc == -ERESTARTSYS) {
                 level = LUSTRE_IMP_RECOVER;
                 goto resend;
-        } else if (rc == 0) {
-                struct mdt_body *body;
-                struct lustre_capa *capa;
-
-                body = lustre_msg_buf(req->rq_repmsg, REPLY_REC_OFF,
-                                      sizeof(*body));
-                LASSERT(body);
-                if (body->valid & OBD_MD_FLMDSCAPA) {
-                        capa = lustre_unpack_capa(req->rq_repmsg,
-                                                  REPLY_REC_OFF + 1);
-                        if (capa == NULL) {
-                                CERROR("Missing/short MDS capability\n");
-                                rc = -EPROTO;
-                        }
+        } else if (rc == -EINPROGRESS) {
+                /* Retry create infinitely until succeed or get other
+                 * error code. */
+                ptlrpc_req_finished(req);
+                resends++;
+
+                CDEBUG(D_HA, "%s: resend:%d create on "DFID"/"DFID"\n",
+                       exp->exp_obd->obd_name, resends,
+                       PFID(&op_data->op_fid1), PFID(&op_data->op_fid2));
+
+                if (generation == import->imp_generation) {
+                        goto rebuild;
+                } else {
+                        CDEBUG(D_HA, "resend cross eviction\n");
+                        RETURN(-EIO);
                 }
         }
 
@@ -268,53 +254,49 @@ int mdc_create(struct obd_export *exp, struct md_op_data *op_data,
 int mdc_unlink(struct obd_export *exp, struct md_op_data *op_data,
                struct ptlrpc_request **request)
 {
-        CFS_LIST_HEAD(cancels);
+       struct list_head cancels = LIST_HEAD_INIT(cancels);
         struct obd_device *obd = class_exp2obd(exp);
         struct ptlrpc_request *req = *request;
-        int size[5] = { sizeof(struct ptlrpc_body),
-                        sizeof(struct mdt_rec_unlink),
-                        0, op_data->op_namelen + 1, 0 };
-        int count = 0, rc, bufcount = 4;
+        int count = 0, rc;
         ENTRY;
 
         LASSERT(req == NULL);
 
-        size[REQ_REC_OFF + 1] = op_data->op_capa1 ?
-                sizeof(struct lustre_capa) : 0;
-
-        if ((op_data->op_flags & MF_MDC_CANCEL_FID1) && 
-            (fid_is_sane(&op_data->op_fid1)))
-                count = mdc_resource_get_unused(exp, &op_data->op_fid1,
-                                                &cancels, LCK_EX,
-                                                MDS_INODELOCK_UPDATE);
-        if ((op_data->op_flags & MF_MDC_CANCEL_FID3) && 
-            (fid_is_sane(&op_data->op_fid3)))
-                count += mdc_resource_get_unused(exp, &op_data->op_fid3,
-                                                 &cancels, LCK_EX,
-                                                 MDS_INODELOCK_FULL);
-        if (exp_connect_cancelset(exp) && count) {
-                bufcount = 5;
-                size[REQ_REC_OFF + 3] = ldlm_request_bufsize(count, MDS_REINT);
-        }
-        req = ptlrpc_prep_req(class_exp2cliimp(exp), LUSTRE_MDS_VERSION,
-                              MDS_REINT, bufcount, size, NULL);
-        if (exp_connect_cancelset(exp) && req)
-                ldlm_cli_cancel_list(&cancels, count, req, REQ_REC_OFF + 3, 0);
-        else
+       if ((op_data->op_flags & MF_MDC_CANCEL_FID1) &&
+           (fid_is_sane(&op_data->op_fid1)))
+               count = mdc_resource_get_unused(exp, &op_data->op_fid1,
+                                               &cancels, LCK_EX,
+                                               MDS_INODELOCK_UPDATE);
+       if ((op_data->op_flags & MF_MDC_CANCEL_FID3) &&
+           (fid_is_sane(&op_data->op_fid3)))
+               count += mdc_resource_get_unused(exp, &op_data->op_fid3,
+                                                &cancels, LCK_EX,
+                                                MDS_INODELOCK_FULL);
+        req = ptlrpc_request_alloc(class_exp2cliimp(exp),
+                                   &RQF_MDS_REINT_UNLINK);
+        if (req == NULL) {
                 ldlm_lock_list_put(&cancels, l_bl_ast, count);
-
-        if (req == NULL)
                 RETURN(-ENOMEM);
-        *request = req;
+        }
+
+        req_capsule_set_size(&req->rq_pill, &RMF_NAME, RCL_CLIENT,
+                             op_data->op_namelen + 1);
 
-        size[REPLY_REC_OFF] = sizeof(struct mdt_body);
-        size[REPLY_REC_OFF + 1] = obd->u.cli.cl_max_mds_easize;
-        size[REPLY_REC_OFF + 2] = obd->u.cli.cl_max_mds_cookiesize;
-        ptlrpc_req_set_repsize(req, 4, size);
+       rc = mdc_prep_elc_req(exp, req, MDS_REINT, &cancels, count);
+       if (rc) {
+               ptlrpc_request_free(req);
+               RETURN(rc);
+       }
 
-        mdc_unlink_pack(req, REQ_REC_OFF, op_data);
+       mdc_unlink_pack(req, op_data);
 
-        rc = mdc_reint(req, obd->u.cli.cl_rpc_lock, LUSTRE_IMP_FULL);
+       req_capsule_set_size(&req->rq_pill, &RMF_MDT_MD, RCL_SERVER,
+                            obd->u.cli.cl_default_mds_easize);
+       ptlrpc_request_set_replen(req);
+
+        *request = req;
+
+       rc = mdc_reint(req, LUSTRE_IMP_FULL);
         if (rc == -ERESTARTSYS)
                 rc = 0;
         RETURN(rc);
@@ -323,20 +305,11 @@ int mdc_unlink(struct obd_export *exp, struct md_op_data *op_data,
 int mdc_link(struct obd_export *exp, struct md_op_data *op_data,
              struct ptlrpc_request **request)
 {
-        CFS_LIST_HEAD(cancels);
-        struct obd_device *obd = exp->exp_obd;
+       struct list_head cancels = LIST_HEAD_INIT(cancels);
         struct ptlrpc_request *req;
-        int size[6] = { sizeof(struct ptlrpc_body),
-                        sizeof(struct mdt_rec_link),
-                        0, 0, op_data->op_namelen + 1, 0 };
-        int count = 0, rc, bufcount = 5;
+        int count = 0, rc;
         ENTRY;
 
-        size[REQ_REC_OFF + 1] = op_data->op_capa1 ?
-                sizeof(struct lustre_capa) : 0;
-        size[REQ_REC_OFF + 2] = op_data->op_capa2 ?
-                sizeof(struct lustre_capa) : 0;
-
         if ((op_data->op_flags & MF_MDC_CANCEL_FID2) &&
             (fid_is_sane(&op_data->op_fid2)))
                 count = mdc_resource_get_unused(exp, &op_data->op_fid2,
@@ -347,27 +320,26 @@ int mdc_link(struct obd_export *exp, struct md_op_data *op_data,
                 count += mdc_resource_get_unused(exp, &op_data->op_fid1,
                                                  &cancels, LCK_EX,
                                                  MDS_INODELOCK_UPDATE);
-        if (exp_connect_cancelset(exp) && count) {
-                bufcount = 6;
-                size[REQ_REC_OFF + 4] = ldlm_request_bufsize(count, MDS_REINT);
-        }
 
-        req = ptlrpc_prep_req(class_exp2cliimp(exp), LUSTRE_MDS_VERSION,
-                              MDS_REINT, bufcount, size, NULL);
-        if (exp_connect_cancelset(exp) && req)
-                ldlm_cli_cancel_list(&cancels, count, req, REQ_REC_OFF + 4, 0);
-        else
+        req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_MDS_REINT_LINK);
+        if (req == NULL) {
                 ldlm_lock_list_put(&cancels, l_bl_ast, count);
-
-        if (req == NULL)
                 RETURN(-ENOMEM);
+        }
+
+        req_capsule_set_size(&req->rq_pill, &RMF_NAME, RCL_CLIENT,
+                             op_data->op_namelen + 1);
 
-        mdc_link_pack(req, REQ_REC_OFF, op_data);
+       rc = mdc_prep_elc_req(exp, req, MDS_REINT, &cancels, count);
+       if (rc) {
+               ptlrpc_request_free(req);
+               RETURN(rc);
+       }
 
-        size[REPLY_REC_OFF] = sizeof(struct mdt_body);
-        ptlrpc_req_set_repsize(req, 2, size);
+        mdc_link_pack(req, op_data);
+        ptlrpc_request_set_replen(req);
 
-        rc = mdc_reint(req, obd->u.cli.cl_rpc_lock, LUSTRE_IMP_FULL);
+       rc = mdc_reint(req, LUSTRE_IMP_FULL);
         *request = req;
         if (rc == -ERESTARTSYS)
                 rc = 0;
@@ -376,23 +348,15 @@ int mdc_link(struct obd_export *exp, struct md_op_data *op_data,
 }
 
 int mdc_rename(struct obd_export *exp, struct md_op_data *op_data,
-               const char *old, int oldlen, const char *new, int newlen,
-               struct ptlrpc_request **request)
+               const char *old, size_t oldlen, const char *new, size_t newlen,
+               struct ptlrpc_request **request)
 {
-        CFS_LIST_HEAD(cancels);
+       struct list_head cancels = LIST_HEAD_INIT(cancels);
         struct obd_device *obd = exp->exp_obd;
         struct ptlrpc_request *req;
-        int size[7] = { sizeof(struct ptlrpc_body),
-                        sizeof(struct mdt_rec_rename),
-                        0, 0, oldlen + 1, newlen + 1, 0 };
-        int count = 0, rc, bufcount = 6;
+        int count = 0, rc;
         ENTRY;
 
-        size[REQ_REC_OFF + 1] = op_data->op_capa1 ?
-                                        sizeof(struct lustre_capa) : 0;
-        size[REQ_REC_OFF + 2] = op_data->op_capa2 ?
-                                        sizeof(struct lustre_capa) : 0;
-
         if ((op_data->op_flags & MF_MDC_CANCEL_FID1) &&
             (fid_is_sane(&op_data->op_fid1)))
                 count = mdc_resource_get_unused(exp, &op_data->op_fid1,
@@ -403,7 +367,7 @@ int mdc_rename(struct obd_export *exp, struct md_op_data *op_data,
                 count += mdc_resource_get_unused(exp, &op_data->op_fid2,
                                                  &cancels, LCK_EX,
                                                  MDS_INODELOCK_UPDATE);
-        if ((op_data->op_flags & MF_MDC_CANCEL_FID3) && 
+        if ((op_data->op_flags & MF_MDC_CANCEL_FID3) &&
             (fid_is_sane(&op_data->op_fid3)))
                 count += mdc_resource_get_unused(exp, &op_data->op_fid3,
                                                  &cancels, LCK_EX,
@@ -413,28 +377,33 @@ int mdc_rename(struct obd_export *exp, struct md_op_data *op_data,
                 count += mdc_resource_get_unused(exp, &op_data->op_fid4,
                                                  &cancels, LCK_EX,
                                                  MDS_INODELOCK_FULL);
-        if (exp_connect_cancelset(exp) && count) {
-                bufcount = 7;
-                size[REQ_REC_OFF + 5] = ldlm_request_bufsize(count, MDS_REINT);
-        }
-        req = ptlrpc_prep_req(class_exp2cliimp(exp), LUSTRE_MDS_VERSION,
-                              MDS_REINT, bufcount, size, NULL);
-        if (exp_connect_cancelset(exp) && req)
-                ldlm_cli_cancel_list(&cancels, count, req, REQ_REC_OFF + 5, 0);
-        else
-                ldlm_lock_list_put(&cancels, l_bl_ast, count);
 
-        if (req == NULL)
+        req = ptlrpc_request_alloc(class_exp2cliimp(exp),
+                                   &RQF_MDS_REINT_RENAME);
+        if (req == NULL) {
+                ldlm_lock_list_put(&cancels, l_bl_ast, count);
                 RETURN(-ENOMEM);
+        }
+
+        req_capsule_set_size(&req->rq_pill, &RMF_NAME, RCL_CLIENT, oldlen + 1);
+        req_capsule_set_size(&req->rq_pill, &RMF_SYMTGT, RCL_CLIENT, newlen+1);
+
+       rc = mdc_prep_elc_req(exp, req, MDS_REINT, &cancels, count);
+       if (rc) {
+               ptlrpc_request_free(req);
+               RETURN(rc);
+       }
+
+        if (exp_connect_cancelset(exp) && req)
+                ldlm_cli_cancel_list(&cancels, count, req, 0);
 
-        mdc_rename_pack(req, REQ_REC_OFF, op_data, old, oldlen, new, newlen);
+        mdc_rename_pack(req, op_data, old, oldlen, new, newlen);
 
-        size[REPLY_REC_OFF] = sizeof(struct mdt_body);
-        size[REPLY_REC_OFF + 1] = obd->u.cli.cl_max_mds_easize;
-        size[REPLY_REC_OFF + 2] = obd->u.cli.cl_max_mds_cookiesize;
-        ptlrpc_req_set_repsize(req, 4, size);
+       req_capsule_set_size(&req->rq_pill, &RMF_MDT_MD, RCL_SERVER,
+                            obd->u.cli.cl_default_mds_easize);
+       ptlrpc_request_set_replen(req);
 
-        rc = mdc_reint(req, obd->u.cli.cl_rpc_lock, LUSTRE_IMP_FULL);
+       rc = mdc_reint(req, LUSTRE_IMP_FULL);
         *request = req;
         if (rc == -ERESTARTSYS)
                 rc = 0;