Whamcloud - gitweb
Branch HEAD
[fs/lustre-release.git] / lustre / mdt / mdt_open.c
index 872bd1b..e2e8802 100644 (file)
@@ -1,29 +1,43 @@
 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
  * vim:expandtab:shiftwidth=8:tabstop=8:
  *
- *  linux/mdt/mdt_open.c
- *  Lustre Metadata Target (mdt) open/close file handling
+ * GPL HEADER START
  *
- *  Copyright (C) 2002-2006 Cluster File Systems, Inc.
- *   Author: Huang Hua <huanghua@clusterfs.com>
+ * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
  *
- *   This file is part of the Lustre file system, http://www.lustre.org
- *   Lustre is a trademark of Cluster File Systems, Inc.
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License version 2 only,
+ * as published by the Free Software Foundation.
  *
- *   You may have signed or agreed to another license before downloading
- *   this software.  If so, you are bound by the terms and conditions
- *   of that agreement, and the following does not apply to you.  See the
- *   LICENSE file included with this distribution for more information.
+ * This program is distributed in the hope that it will be useful, but
+ * WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * General Public License version 2 for more details (a copy is included
+ * in the LICENSE file that accompanied this code).
  *
- *   If you did not agree to a different license, then this copy of Lustre
- *   is open source software; you can redistribute it and/or modify it
- *   under the terms of version 2 of the GNU General Public License as
- *   published by the Free Software Foundation.
+ * You should have received a copy of the GNU General Public License
+ * version 2 along with this program; If not, see
+ * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
  *
- *   In either case, Lustre is distributed in the hope that it will be
- *   useful, but WITHOUT ANY WARRANTY; without even the implied warranty
- *   of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
- *   license text for more details.
+ * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
+ * CA 95054 USA or visit www.sun.com if you need additional information or
+ * have any questions.
+ *
+ * GPL HEADER END
+ */
+/*
+ * Copyright  2008 Sun Microsystems, Inc. All rights reserved
+ * Use is subject to license terms.
+ */
+/*
+ * This file is part of Lustre, http://www.lustre.org/
+ * Lustre is a trademark of Sun Microsystems, Inc.
+ *
+ * lustre/mdt/mdt_open.c
+ *
+ * Lustre Metadata Target (mdt) open/close file handling
+ *
+ * Author: Huang Hua <huanghua@clusterfs.com>
  */
 
 #ifndef EXPORT_SYMTAB
@@ -31,7 +45,7 @@
 #endif
 #define DEBUG_SUBSYSTEM S_MDS
 
-#include <linux/lustre_acl.h>
+#include <lustre_acl.h>
 #include <lustre_mds.h>
 #include "mdt_internal.h"
 
@@ -49,8 +63,8 @@ struct mdt_file_data *mdt_mfd_new(void)
 
         OBD_ALLOC_PTR(mfd);
         if (mfd != NULL) {
-                INIT_LIST_HEAD(&mfd->mfd_handle.h_link);
-                INIT_LIST_HEAD(&mfd->mfd_list);
+                CFS_INIT_LIST_HEAD(&mfd->mfd_handle.h_link);
+                CFS_INIT_LIST_HEAD(&mfd->mfd_list);
                 class_handle_hash(&mfd->mfd_handle, mdt_mfd_get);
         }
         RETURN(mfd);
@@ -125,7 +139,7 @@ void mdt_sizeonmds_enable(struct mdt_thread_info *info,
 {
        spin_lock(&info->mti_mdt->mdt_ioepoch_lock);
        if (info->mti_epoch->ioepoch == mo->mot_ioepoch) {
-                LASSERT(!mdt_epoch_opened(mo)); 
+                LASSERT(!mdt_epoch_opened(mo));
                 mo->mot_ioepoch = 0;
                 mo->mot_flags = 0;
        }
@@ -141,7 +155,7 @@ int mdt_epoch_open(struct mdt_thread_info *info, struct mdt_object *o)
         int rc = 0;
         ENTRY;
 
-        if (!(mdt_conn_flags(info) & OBD_CONNECT_SOM) || 
+        if (!(mdt_conn_flags(info) & OBD_CONNECT_SOM) ||
             !S_ISREG(lu_object_attr(&o->mot_obj.mo_lu)))
                 RETURN(0);
 
@@ -194,7 +208,7 @@ static int mdt_sizeonmds_update(struct mdt_thread_info *info,
                  * and "close", maybe someone has changed the file mode
                  * or flags, or the file created mode do not permit wirte,
                  * and so on. Just set MDS_PERM_BYPASS for all the cases. */
-                info->mti_attr.ma_attr_flags |= MDS_PERM_BYPASS;
+                info->mti_attr.ma_attr_flags |= MDS_PERM_BYPASS | MDS_SOM;
                 info->mti_attr.ma_attr.la_valid &= LA_SIZE | LA_BLOCKS |
                                                 LA_ATIME | LA_MTIME | LA_CTIME;
                 RETURN(mdt_attr_set(info, o, 0));
@@ -377,17 +391,17 @@ static int mdt_mfd_open(struct mdt_thread_info *info, struct mdt_object *p,
         int                      rc = 0, isdir, isreg;
         ENTRY;
 
-        repbody = req_capsule_server_get(&info->mti_pill, &RMF_MDT_BODY);
+        repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
 
         isreg = S_ISREG(la->la_mode);
         isdir = S_ISDIR(la->la_mode);
-        if ((isreg && !(ma->ma_valid & MA_LOV))) {
+        if (isreg && !(ma->ma_valid & MA_LOV)) {
                 /*
                  * No EA, check whether it is will set regEA and dirEA since in
                  * above attr get, these size might be zero, so reset it, to
                  * retrieve the MD after create obj.
                  */
-                ma->ma_lmm_size = req_capsule_get_size(&info->mti_pill,
+                ma->ma_lmm_size = req_capsule_get_size(info->mti_pill,
                                                        &RMF_MDT_MD,
                                                        RCL_SERVER);
                 /* in replay case, p == NULL */
@@ -451,7 +465,7 @@ static int mdt_mfd_open(struct mdt_thread_info *info, struct mdt_object *p,
                         /* Check wheather old cookie already exist in
                          * the list, becasue when do recovery, client
                          * might be disconnected from server, and
-                         * restart replay, so there maybe some orphan 
+                         * restart replay, so there maybe some orphan
                          * mfd here, we should remove them */
                         LASSERT(info->mti_rr.rr_handle != NULL);
                         old_mfd = mdt_handle2mfd(info, info->mti_rr.rr_handle);
@@ -487,8 +501,8 @@ static int mdt_finish_open(struct mdt_thread_info *info,
                            int flags, int created, struct ldlm_reply *rep)
 {
         struct ptlrpc_request   *req = mdt_info_req(info);
+        struct obd_export       *exp = req->rq_export;
         struct mdt_export_data  *med = &req->rq_export->exp_mdt_data;
-        struct mdt_device       *mdt = info->mti_mdt;
         struct md_attr          *ma  = &info->mti_attr;
         struct lu_attr          *la  = &ma->ma_attr;
         struct mdt_file_data    *mfd;
@@ -500,15 +514,15 @@ static int mdt_finish_open(struct mdt_thread_info *info,
 
         LASSERT(ma->ma_valid & MA_INODE);
 
-        repbody = req_capsule_server_get(&info->mti_pill, &RMF_MDT_BODY);
+        repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
 
         isreg = S_ISREG(la->la_mode);
         isdir = S_ISDIR(la->la_mode);
         islnk = S_ISLNK(la->la_mode);
         mdt_pack_attr2body(info, repbody, la, mdt_object_fid(o));
 
-        if (med->med_rmtclient) {
-                void *buf = req_capsule_server_get(&info->mti_pill, &RMF_ACL);
+        if (exp_connect_rmtclient(exp)) {
+                void *buf = req_capsule_server_get(info->mti_pill, &RMF_ACL);
 
                 rc = mdt_pack_remote_perm(info, o, buf);
                 if (rc) {
@@ -520,13 +534,13 @@ static int mdt_finish_open(struct mdt_thread_info *info,
                 }
         }
 #ifdef CONFIG_FS_POSIX_ACL
-        else if (req->rq_export->exp_connect_flags & OBD_CONNECT_ACL) {
+        else if (exp->exp_connect_flags & OBD_CONNECT_ACL) {
                 const struct lu_env *env = info->mti_env;
                 struct md_object *next = mdt_object_child(o);
                 struct lu_buf *buf = &info->mti_buf;
 
-                buf->lb_buf = req_capsule_server_get(&info->mti_pill, &RMF_ACL);
-                buf->lb_len = req_capsule_get_size(&info->mti_pill, &RMF_ACL,
+                buf->lb_buf = req_capsule_server_get(info->mti_pill, &RMF_ACL);
+                buf->lb_len = req_capsule_get_size(info->mti_pill, &RMF_ACL,
                                                    RCL_SERVER);
                 if (buf->lb_len > 0) {
                         rc = mo_xattr_get(env, next, buf,
@@ -550,26 +564,26 @@ static int mdt_finish_open(struct mdt_thread_info *info,
         }
 #endif
 
-        if (mdt->mdt_opts.mo_mds_capa) {
+        if (info->mti_mdt->mdt_opts.mo_mds_capa &&
+            exp->exp_connect_flags & OBD_CONNECT_MDS_CAPA) {
                 struct lustre_capa *capa;
 
-                capa = req_capsule_server_get(&info->mti_pill, &RMF_CAPA1);
+                capa = req_capsule_server_get(info->mti_pill, &RMF_CAPA1);
                 LASSERT(capa);
                 capa->lc_opc = CAPA_OPC_MDS_DEFAULT;
-                capa->lc_uid = 0;
                 rc = mo_capa_get(info->mti_env, mdt_object_child(o), capa, 0);
                 if (rc)
                         RETURN(rc);
                 repbody->valid |= OBD_MD_FLMDSCAPA;
         }
-        if (mdt->mdt_opts.mo_oss_capa &&
+        if (info->mti_mdt->mdt_opts.mo_oss_capa &&
+            exp->exp_connect_flags & OBD_CONNECT_OSS_CAPA &&
             S_ISREG(lu_object_attr(&o->mot_obj.mo_lu))) {
                 struct lustre_capa *capa;
 
-                capa = req_capsule_server_get(&info->mti_pill, &RMF_CAPA2);
+                capa = req_capsule_server_get(info->mti_pill, &RMF_CAPA2);
                 LASSERT(capa);
                 capa->lc_opc = CAPA_OPC_OSS_DEFAULT | capa_open_opc(flags);
-                capa->lc_uid = 0;
                 rc = mo_capa_get(info->mti_env, mdt_object_child(o), capa, 0);
                 if (rc)
                         RETURN(rc);
@@ -604,8 +618,8 @@ static int mdt_finish_open(struct mdt_thread_info *info,
         } else if (flags & MDS_OPEN_DIRECTORY)
                 RETURN(-ENOTDIR);
 
-        if (MDT_FAIL_CHECK(OBD_FAIL_MDS_OPEN_CREATE)) {
-                obd_fail_loc = OBD_FAIL_LDLM_REPLY | OBD_FAIL_ONCE;
+        if (OBD_FAIL_CHECK_RESET(OBD_FAIL_MDS_OPEN_CREATE,
+                                 OBD_FAIL_LDLM_REPLY | OBD_FAIL_ONCE)) {
                 RETURN(-EAGAIN);
         }
 
@@ -640,18 +654,18 @@ static int mdt_finish_open(struct mdt_thread_info *info,
         RETURN(rc);
 }
 
-extern void mdt_req_from_mcd(struct ptlrpc_request *req,
-                             struct mdt_client_data *mcd);
+extern void mdt_req_from_lcd(struct ptlrpc_request *req,
+                             struct lsd_client_data *lcd);
 
 void mdt_reconstruct_open(struct mdt_thread_info *info,
                           struct mdt_lock_handle *lhc)
 {
         const struct lu_env *env = info->mti_env;
         struct mdt_device       *mdt  = info->mti_mdt;
-        struct req_capsule      *pill = &info->mti_pill;
+        struct req_capsule      *pill = info->mti_pill;
         struct ptlrpc_request   *req  = mdt_info_req(info);
         struct mdt_export_data  *med  = &req->rq_export->exp_mdt_data;
-        struct mdt_client_data  *mcd  = med->med_mcd;
+        struct lsd_client_data  *lcd  = med->med_lcd;
         struct md_attr          *ma   = &info->mti_attr;
         struct mdt_reint_record *rr   = &info->mti_rr;
         __u32                   flags = info->mti_spec.sp_cr_flags;
@@ -669,33 +683,53 @@ void mdt_reconstruct_open(struct mdt_thread_info *info,
         ma->ma_lmm = req_capsule_server_get(pill, &RMF_MDT_MD);
         ma->ma_lmm_size = req_capsule_get_size(pill, &RMF_MDT_MD,
                                                RCL_SERVER);
-        ma->ma_need = MA_INODE | MA_LOV;
+        ma->ma_need = MA_INODE;
+        if (ma->ma_lmm_size > 0)
+                ma->ma_need |= MA_LOV;
+
         ma->ma_valid = 0;
 
-        mdt_req_from_mcd(req, med->med_mcd);
-        mdt_set_disposition(info, ldlm_rep, mcd->mcd_last_data);
+        mdt_req_from_lcd(req, med->med_lcd);
+        mdt_set_disposition(info, ldlm_rep, lcd->lcd_last_data);
 
         CERROR("This is reconstruct open: disp="LPX64", result=%d\n",
                 ldlm_rep->lock_policy_res1, req->rq_status);
 
         if (mdt_get_disposition(ldlm_rep, DISP_OPEN_CREATE) &&
-            req->rq_status != 0) {
+            req->rq_status != 0)
                 /* We did not create successfully, return error to client. */
-                mdt_shrink_reply(info);
                 GOTO(out, rc = req->rq_status);
-        }
 
         if (mdt_get_disposition(ldlm_rep, DISP_OPEN_CREATE)) {
+                struct obd_export *exp = req->rq_export;
                 /*
                  * We failed after creation, but we do not know in which step
                  * we failed. So try to check the child object.
                  */
                 parent = mdt_object_find(env, mdt, rr->rr_fid1);
-                LASSERT(!IS_ERR(parent));
-
+                if (IS_ERR(parent)) {
+                        rc = PTR_ERR(parent);
+                        LCONSOLE_WARN("Parent "DFID" lookup error %d."
+                                      " Evicting client %s with export %s.\n",
+                                      PFID(mdt_object_fid(parent)), rc,
+                                      obd_uuid2str(&exp->exp_client_uuid),
+                                      obd_export_nid2str(exp));
+                        mdt_export_evict(exp);
+                        EXIT;
+                        return;
+                }
                 child = mdt_object_find(env, mdt, rr->rr_fid2);
-                LASSERT(!IS_ERR(child));
-
+                if (IS_ERR(child)) {
+                        rc = PTR_ERR(child);
+                        LCONSOLE_WARN("Child "DFID" lookup error %d."
+                                      " Evicting client %s with export %s.\n",
+                                      PFID(mdt_object_fid(child)), rc,
+                                      obd_uuid2str(&exp->exp_client_uuid),
+                                      obd_export_nid2str(exp));
+                        mdt_export_evict(exp);
+                        EXIT;
+                        return;
+                }
                 rc = mdt_object_exists(child);
                 if (rc > 0) {
                         struct md_object *next;
@@ -719,7 +753,6 @@ void mdt_reconstruct_open(struct mdt_thread_info *info,
                 }
                 mdt_object_put(env, parent);
                 mdt_object_put(env, child);
-                mdt_shrink_reply(info);
                 GOTO(out, rc);
         } else {
 regular_open:
@@ -745,7 +778,6 @@ static int mdt_open_by_fid(struct mdt_thread_info* info,
         int                      rc;
         ENTRY;
 
-        LASSERT(info->mti_spec.u.sp_ea.no_lov_create);
         o = mdt_object_find(info->mti_env, info->mti_mdt, rr->rr_fid2);
         if (IS_ERR(o))
                 RETURN(rc = PTR_ERR(o));
@@ -758,13 +790,13 @@ static int mdt_open_by_fid(struct mdt_thread_info* info,
 
                 rc = mo_attr_get(env, mdt_object_child(o), ma);
                 if (rc == 0)
-                        rc = mdt_mfd_open(info, NULL, o, flags, 0);
+                        rc = mdt_finish_open(info, NULL, o, flags, 0, rep);
         } else if (rc == 0) {
                 rc = -ENOENT;
         } else  {
                 /* the child object was created on remote server */
                 struct mdt_body *repbody;
-                repbody = req_capsule_server_get(&info->mti_pill, &RMF_MDT_BODY);
+                repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
                 repbody->fid1 = *rr->rr_fid2;
                 repbody->valid |= (OBD_MD_FLID | OBD_MD_MDS);
                 rc = 0;
@@ -842,43 +874,52 @@ int mdt_reint_open(struct mdt_thread_info *info, struct mdt_lock_handle *lhc)
         __u32                    create_flags = info->mti_spec.sp_cr_flags;
         struct mdt_reint_record *rr = &info->mti_rr;
         struct lu_name          *lname;
-        int                      result;
+        int                      result, rc;
         int                      created = 0;
+        __u32                    msg_flags;
         ENTRY;
 
-        OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_PAUSE_OPEN | OBD_FAIL_ONCE,
-                         (obd_timeout + 1) / 4);
+        OBD_FAIL_TIMEOUT_ORSET(OBD_FAIL_MDS_PAUSE_OPEN, OBD_FAIL_ONCE,
+                               (obd_timeout + 1) / 4);
 
-        repbody = req_capsule_server_get(&info->mti_pill, &RMF_MDT_BODY);
+        repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
 
-        ma->ma_lmm = req_capsule_server_get(&info->mti_pill, &RMF_MDT_MD);
-        ma->ma_lmm_size = req_capsule_get_size(&info->mti_pill, &RMF_MDT_MD,
+        ma->ma_lmm = req_capsule_server_get(info->mti_pill, &RMF_MDT_MD);
+        ma->ma_lmm_size = req_capsule_get_size(info->mti_pill, &RMF_MDT_MD,
                                                RCL_SERVER);
-        ma->ma_need = MA_INODE | MA_LOV;
+        ma->ma_need = MA_INODE;
+        if (ma->ma_lmm_size > 0)
+                ma->ma_need |= MA_LOV;
+
         ma->ma_valid = 0;
 
-        LASSERT(info->mti_pill.rc_fmt == &RQF_LDLM_INTENT_OPEN);
-        ldlm_rep = req_capsule_server_get(&info->mti_pill, &RMF_DLM_REP);
+        LASSERT(info->mti_pill->rc_fmt == &RQF_LDLM_INTENT_OPEN);
+        ldlm_rep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP);
 
         /* TODO: JOIN file */
         if (create_flags & MDS_OPEN_JOIN_FILE) {
                 CERROR("JOIN file will be supported soon\n");
                 GOTO(out, result = err_serious(-EOPNOTSUPP));
         }
+        msg_flags = lustre_msg_get_flags(req->rq_reqmsg);
 
         CDEBUG(D_INODE, "I am going to open "DFID"/(%s->"DFID") "
                "cr_flag=0%o mode=0%06o msg_flag=0x%x\n",
                PFID(rr->rr_fid1), rr->rr_name,
                PFID(rr->rr_fid2), create_flags,
-               ma->ma_attr.la_mode, lustre_msg_get_flags(req->rq_reqmsg));
+               ma->ma_attr.la_mode, msg_flags);
 
-        if (lustre_msg_get_flags(req->rq_reqmsg) & MSG_REPLAY) {
-                /* This is a replay request. */
+        if (msg_flags & MSG_REPLAY ||
+            (req->rq_export->exp_libclient && create_flags&MDS_OPEN_HAS_EA)) {
+                /* This is a replay request or from liblustre with ea. */
                 result = mdt_open_by_fid(info, ldlm_rep);
 
-                if (result != -ENOENT)
+                if (result != -ENOENT) {
+                        if (req->rq_export->exp_libclient &&
+                            create_flags&MDS_OPEN_HAS_EA)
+                                GOTO(out, result = 0);
                         GOTO(out, result);
-
+                }
                 /*
                  * We didn't find the correct object, so we need to re-create it
                  * via a regular replay.
@@ -891,7 +932,7 @@ int mdt_reint_open(struct mdt_thread_info *info, struct mdt_lock_handle *lhc)
                        "regular open\n");
         }
 
-        if (MDT_FAIL_CHECK(OBD_FAIL_MDS_OPEN_PACK))
+        if (OBD_FAIL_CHECK(OBD_FAIL_MDS_OPEN_PACK))
                 GOTO(out, result = err_serious(-ENOMEM));
 
         mdt_set_disposition(info, ldlm_rep,
@@ -968,6 +1009,7 @@ int mdt_reint_open(struct mdt_thread_info *info, struct mdt_lock_handle *lhc)
                  * not exist.
                  */
                 info->mti_spec.sp_cr_lookup = 0;
+                info->mti_spec.sp_feat = &dt_directory_features;
 
                 result = mdo_create(info->mti_env,
                                     mdt_object_child(parent),
@@ -991,8 +1033,6 @@ int mdt_reint_open(struct mdt_thread_info *info, struct mdt_lock_handle *lhc)
                  * The object is on remote node, return its FID for remote open.
                  */
                 if (result == -EREMOTE) {
-                        int rc;
-
                         /*
                          * Check if this lock already was sent to client and
                          * this is resent case. For resent case do not take lock
@@ -1003,8 +1043,7 @@ int mdt_reint_open(struct mdt_thread_info *info, struct mdt_lock_handle *lhc)
                         if (lustre_handle_is_used(&lhc->mlh_reg_lh)) {
                                 struct ldlm_lock *lock;
 
-                                LASSERT(lustre_msg_get_flags(req->rq_reqmsg) &
-                                        MSG_RESENT);
+                                LASSERT(msg_flags & MSG_RESENT);
 
                                 lock = ldlm_handle2lock(&lhc->mlh_reg_lh);
                                 if (!lock) {
@@ -1032,23 +1071,55 @@ int mdt_reint_open(struct mdt_thread_info *info, struct mdt_lock_handle *lhc)
                 }
         }
 
+        LASSERT(!lustre_handle_is_used(&lhc->mlh_reg_lh));
+
+        /* get openlock if this is not replay and if a client requested it */
+        if (!(msg_flags & MSG_REPLAY) && create_flags & MDS_OPEN_LOCK) {
+                ldlm_mode_t lm;
+
+                LASSERT(!created);
+                if (create_flags & FMODE_WRITE)
+                        lm = LCK_CW;
+                else if (create_flags & MDS_FMODE_EXEC)
+                        lm = LCK_PR;
+                else
+                        lm = LCK_CR;
+                mdt_lock_handle_init(lhc);
+                mdt_lock_reg_init(lhc, lm);
+                rc = mdt_object_lock(info, child, lhc,
+                                     MDS_INODELOCK_LOOKUP | MDS_INODELOCK_OPEN,
+                                     MDT_CROSS_LOCK);
+                if (rc) {
+                        result = rc;
+                        GOTO(out_child, result);
+                } else {
+                        result = -EREMOTE;
+                        mdt_set_disposition(info, ldlm_rep, DISP_OPEN_LOCK);
+                }
+        }
+
         /* Try to open it now. */
-        result = mdt_finish_open(info, parent, child, create_flags,
-                                 created, ldlm_rep);
-
-        if (result != 0 && created) {
-                int rc2;
-                ma->ma_need = 0;
-                ma->ma_valid = 0;
-                ma->ma_cookie_size = 0;
-                info->mti_no_need_trans = 1;
-                rc2 = mdo_unlink(info->mti_env,
-                                 mdt_object_child(parent),
-                                 mdt_object_child(child),
-                                 lname,
-                                 &info->mti_attr);
-                if (rc2 != 0)
-                        CERROR("Error in cleanup of open\n");
+        rc = mdt_finish_open(info, parent, child, create_flags,
+                             created, ldlm_rep);
+        if (rc) {
+                result = rc;
+                if (lustre_handle_is_used(&lhc->mlh_reg_lh))
+                        /* openlock was acquired and mdt_finish_open failed -
+                           drop the openlock */
+                        mdt_object_unlock(info, child, lhc, 1);
+                if (created) {
+                        ma->ma_need = 0;
+                        ma->ma_valid = 0;
+                        ma->ma_cookie_size = 0;
+                        info->mti_no_need_trans = 1;
+                        rc = mdo_unlink(info->mti_env,
+                                        mdt_object_child(parent),
+                                        mdt_object_child(child),
+                                        lname,
+                                        &info->mti_attr);
+                        if (rc != 0)
+                                CERROR("Error in cleanup of open\n");
+                }
         }
         EXIT;
 out_child:
@@ -1056,8 +1127,7 @@ out_child:
 out_parent:
         mdt_object_unlock_put(info, parent, lh, result);
 out:
-        mdt_shrink_reply(info);
-        if (result)
+        if (result && result != -EREMOTE)
                 lustre_msg_set_transno(req->rq_repmsg, 0);
         return result;
 }
@@ -1088,8 +1158,8 @@ int mdt_mfd_close(struct mdt_thread_info *info, struct mdt_file_data *mfd)
                 mdt_write_allow(info->mti_mdt, o);
         } else if (mode & FMODE_EPOCH) {
                 ret = mdt_epoch_close(info, o);
-        } 
-        
+        }
+
         /* Update atime on close only. */
         if ((mode & MDS_FMODE_EXEC || mode & FMODE_READ || mode & FMODE_WRITE)
             && (ma->ma_valid & MA_INODE) && (ma->ma_attr.la_valid & LA_ATIME)) {
@@ -1159,33 +1229,34 @@ int mdt_close(struct mdt_thread_info *info)
 
         LASSERT(info->mti_epoch);
 
-        req_capsule_set_size(&info->mti_pill, &RMF_MDT_MD, RCL_SERVER,
+        req_capsule_set_size(info->mti_pill, &RMF_MDT_MD, RCL_SERVER,
                              info->mti_mdt->mdt_max_mdsize);
-        req_capsule_set_size(&info->mti_pill, &RMF_LOGCOOKIES, RCL_SERVER,
+        req_capsule_set_size(info->mti_pill, &RMF_LOGCOOKIES, RCL_SERVER,
                              info->mti_mdt->mdt_max_cookiesize);
-        rc = req_capsule_pack(&info->mti_pill);
+        rc = req_capsule_server_pack(info->mti_pill);
         if (mdt_check_resent(info, mdt_reconstruct_generic, NULL))
                 RETURN(lustre_msg_get_status(req->rq_repmsg));
 
         /* Continue to close handle even if we can not pack reply */
         if (rc == 0) {
-                repbody = req_capsule_server_get(&info->mti_pill,
+                repbody = req_capsule_server_get(info->mti_pill,
                                                  &RMF_MDT_BODY);
-                ma->ma_lmm = req_capsule_server_get(&info->mti_pill,
+                ma->ma_lmm = req_capsule_server_get(info->mti_pill,
                                                     &RMF_MDT_MD);
-                ma->ma_lmm_size = req_capsule_get_size(&info->mti_pill,
+                ma->ma_lmm_size = req_capsule_get_size(info->mti_pill,
                                                        &RMF_MDT_MD,
                                                        RCL_SERVER);
-                ma->ma_cookie = req_capsule_server_get(&info->mti_pill,
+                ma->ma_cookie = req_capsule_server_get(info->mti_pill,
                                                        &RMF_LOGCOOKIES);
-                ma->ma_cookie_size = req_capsule_get_size(&info->mti_pill,
+                ma->ma_cookie_size = req_capsule_get_size(info->mti_pill,
                                                           &RMF_LOGCOOKIES,
                                                           RCL_SERVER);
                 ma->ma_need = MA_INODE | MA_LOV | MA_COOKIE;
                 repbody->eadatasize = 0;
                 repbody->aclsize = 0;
-        } else
+        } else {
                 rc = err_serious(rc);
+        }
 
         med = &req->rq_export->exp_mdt_data;
         spin_lock(&med->med_open_lock);
@@ -1213,9 +1284,12 @@ int mdt_close(struct mdt_thread_info *info)
         if (repbody != NULL)
                 mdt_shrink_reply(info);
 
-        if (MDT_FAIL_CHECK(OBD_FAIL_MDS_CLOSE_PACK))
+        if (OBD_FAIL_CHECK(OBD_FAIL_MDS_CLOSE_PACK))
                 RETURN(err_serious(-ENOMEM));
 
+        if (OBD_FAIL_CHECK_RESET(OBD_FAIL_MDS_CLOSE_NET_REP,
+                                 OBD_FAIL_MDS_CLOSE_NET_REP))
+                info->mti_fail_id = OBD_FAIL_MDS_CLOSE_NET_REP;
         RETURN(rc ? rc : ret);
 }
 
@@ -1227,11 +1301,11 @@ int mdt_done_writing(struct mdt_thread_info *info)
         int rc;
         ENTRY;
 
-        rc = req_capsule_pack(&info->mti_pill);
+        rc = req_capsule_server_pack(info->mti_pill);
         if (rc)
                 RETURN(err_serious(rc));
 
-        repbody = req_capsule_server_get(&info->mti_pill,
+        repbody = req_capsule_server_get(info->mti_pill,
                                          &RMF_MDT_BODY);
         repbody->eadatasize = 0;
         repbody->aclsize = 0;
@@ -1253,8 +1327,8 @@ int mdt_done_writing(struct mdt_thread_info *info)
                        ": cookie = "LPX64"\n", PFID(info->mti_rr.rr_fid1),
                        info->mti_epoch->handle.cookie);
                 RETURN(-ESTALE);
-        } 
+        }
+
         LASSERT(mfd->mfd_mode == FMODE_EPOCH ||
                 mfd->mfd_mode == FMODE_EPOCHLCK);
         class_handle_unhash(&mfd->mfd_handle);
@@ -1268,4 +1342,3 @@ int mdt_done_writing(struct mdt_thread_info *info)
         mdt_empty_transno(info);
         RETURN(rc);
 }
-