Whamcloud - gitweb
LU-1303 lod: introduce lod device
[fs/lustre-release.git] / lustre / mdt / mdt_lib.c
index 37cb8f0..b734013 100644 (file)
@@ -1,45 +1,55 @@
-/* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
- * vim:expandtab:shiftwidth=8:tabstop=8:
+/*
+ * GPL HEADER START
  *
- *  lustre/mdt/mdt_lib.c
- *  Lustre Metadata Target (mdt) request unpacking helper.
+ * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
  *
- *  Copyright (c) 2006 Cluster File Systems, Inc.
- *   Author: Peter Braam <braam@clusterfs.com>
- *   Author: Andreas Dilger <adilger@clusterfs.com>
- *   Author: Phil Schwan <phil@clusterfs.com>
- *   Author: Mike Shaver <shaver@clusterfs.com>
- *   Author: Nikita Danilov <nikita@clusterfs.com>
- *   Author: Huang Hua <huanghua@clusterfs.com>
- *   Author: Fan Yong <fanyong@clusterfs.com>
+ * 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.
  *
+ * 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) 2007, 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, 2012, Whamcloud, Inc.
+ */
+/*
+ * This file is part of Lustre, http://www.lustre.org/
+ * Lustre is a trademark of Sun Microsystems, Inc.
+ *
+ * lustre/mdt/mdt_lib.c
+ *
+ * Lustre Metadata Target (mdt) request unpacking helper.
+ *
+ * Author: Peter Braam <braam@clusterfs.com>
+ * Author: Andreas Dilger <adilger@clusterfs.com>
+ * Author: Phil Schwan <phil@clusterfs.com>
+ * Author: Mike Shaver <shaver@clusterfs.com>
+ * Author: Nikita Danilov <nikita@clusterfs.com>
+ * Author: Huang Hua <huanghua@clusterfs.com>
+ * Author: Fan Yong <fanyong@clusterfs.com>
  */
 
-
-#ifndef EXPORT_SYMTAB
-# define EXPORT_SYMTAB
-#endif
 #define DEBUG_SUBSYSTEM S_MDS
 
 #include "mdt_internal.h"
+#include <lnet/lib-lnet.h>
 
 
 typedef enum ucred_init_type {
@@ -56,7 +66,7 @@ void mdt_exit_ucred(struct mdt_thread_info *info)
         if (uc->mu_valid != UCRED_INIT) {
                 uc->mu_suppgids[0] = uc->mu_suppgids[1] = -1;
                 if (uc->mu_ginfo) {
-                        groups_free(uc->mu_ginfo);
+                        cfs_put_group_info(uc->mu_ginfo);
                         uc->mu_ginfo = NULL;
                 }
                 if (uc->mu_identity) {
@@ -68,33 +78,70 @@ void mdt_exit_ucred(struct mdt_thread_info *info)
         }
 }
 
-/* XXX: root_squash will be redesigned in Lustre 1.7.
- * Do not root_squash for inter-MDS operations */
-static int mdt_root_squash(struct mdt_thread_info *info)
+static int match_nosquash_list(cfs_rw_semaphore_t *sem,
+                               cfs_list_t *nidlist,
+                               lnet_nid_t peernid)
 {
-        return 0;
+        int rc;
+        ENTRY;
+        cfs_down_read(sem);
+        rc = cfs_match_nid(peernid, nidlist);
+        cfs_up_read(sem);
+        RETURN(rc);
+}
+
+/* root_squash for inter-MDS operations */
+static int mdt_root_squash(struct mdt_thread_info *info, lnet_nid_t peernid)
+{
+        struct md_ucred *ucred = mdt_ucred(info);
+        ENTRY;
+
+        if (!info->mti_mdt->mdt_squash_uid || ucred->mu_fsuid)
+                RETURN(0);
+
+        if (match_nosquash_list(&info->mti_mdt->mdt_squash_sem,
+                                &info->mti_mdt->mdt_nosquash_nids,
+                                peernid)) {
+                CDEBUG(D_OTHER, "%s is in nosquash_nids list\n",
+                       libcfs_nid2str(peernid));
+                RETURN(0);
+        }
+
+        CDEBUG(D_OTHER, "squash req from %s, (%d:%d/%x)=>(%d:%d/%x)\n",
+               libcfs_nid2str(peernid),
+               ucred->mu_fsuid, ucred->mu_fsgid, ucred->mu_cap,
+               info->mti_mdt->mdt_squash_uid, info->mti_mdt->mdt_squash_gid,
+               0);
+
+        ucred->mu_fsuid = info->mti_mdt->mdt_squash_uid;
+        ucred->mu_fsgid = info->mti_mdt->mdt_squash_gid;
+        ucred->mu_cap = 0;
+        ucred->mu_suppgids[0] = -1;
+        ucred->mu_suppgids[1] = -1;
+
+        RETURN(0);
 }
 
 static int new_init_ucred(struct mdt_thread_info *info, ucred_init_type_t type,
                           void *buf)
 {
         struct ptlrpc_request   *req = mdt_info_req(info);
-        struct mdt_export_data  *med = mdt_req2med(req);
         struct mdt_device       *mdt = info->mti_mdt;
         struct ptlrpc_user_desc *pud = req->rq_user_desc;
         struct md_ucred         *ucred = mdt_ucred(info);
-        lnet_nid_t              peernid = req->rq_peer.nid;
+        lnet_nid_t               peernid = req->rq_peer.nid;
         __u32                    perm = 0;
-        int                     setuid;
-        int                     setgid;
-        int                     rc = 0;
+        __u32                    remote = exp_connect_rmtclient(info->mti_exp);
+        int                      setuid;
+        int                      setgid;
+        int                      rc = 0;
 
         ENTRY;
 
         LASSERT(req->rq_auth_gss);
         LASSERT(!req->rq_auth_usr_mdt);
         LASSERT(req->rq_user_desc);
-        
+
         ucred->mu_valid = UCRED_INVALID;
 
         ucred->mu_o_uid   = pud->pud_uid;
@@ -110,9 +157,9 @@ static int new_init_ucred(struct mdt_thread_info *info, ucred_init_type_t type,
         }
 
         /* sanity check: we expect the uid which client claimed is true */
-        if (med->med_rmtclient) {
+        if (remote) {
                 if (req->rq_auth_mapped_uid == INVALID_UID) {
-                        CWARN("remote user not mapped, deny access!\n");
+                        CDEBUG(D_SEC, "remote user not mapped, deny access!\n");
                         RETURN(-EACCES);
                 }
 
@@ -120,9 +167,9 @@ static int new_init_ucred(struct mdt_thread_info *info, ucred_init_type_t type,
                         RETURN(-EACCES);
 
                 if (req->rq_auth_mapped_uid != pud->pud_uid) {
-                        CERROR("remote client "LPU64": auth/mapped uid %u/%u "
-                               "while client claim %u:%u/%u:%u\n",
-                               peernid, req->rq_auth_uid,
+                        CDEBUG(D_SEC, "remote client %s: auth/mapped uid %u/%u "
+                               "while client claims %u:%u/%u:%u\n",
+                               libcfs_nid2str(peernid), req->rq_auth_uid,
                                req->rq_auth_mapped_uid,
                                pud->pud_uid, pud->pud_gid,
                                pud->pud_fsuid, pud->pud_fsgid);
@@ -130,17 +177,18 @@ static int new_init_ucred(struct mdt_thread_info *info, ucred_init_type_t type,
                 }
         } else {
                 if (req->rq_auth_uid != pud->pud_uid) {
-                        CERROR("local client "LPU64": auth uid %u "
-                               "while client claim %u:%u/%u:%u\n",
-                               peernid, req->rq_auth_uid, pud->pud_uid,
-                               pud->pud_gid, pud->pud_fsuid, pud->pud_fsgid);
+                        CDEBUG(D_SEC, "local client %s: auth uid %u "
+                               "while client claims %u:%u/%u:%u\n",
+                               libcfs_nid2str(peernid), req->rq_auth_uid,
+                               pud->pud_uid, pud->pud_gid,
+                               pud->pud_fsuid, pud->pud_fsgid);
                         RETURN(-EACCES);
                 }
         }
 
         if (is_identity_get_disabled(mdt->mdt_identity_cache)) {
-                if (med->med_rmtclient) {
-                        CERROR("remote client must run with identity_get "
+                if (remote) {
+                        CDEBUG(D_SEC, "remote client must run with identity_get "
                                "enabled!\n");
                         RETURN(-EACCES);
                 } else {
@@ -149,16 +197,25 @@ static int new_init_ucred(struct mdt_thread_info *info, ucred_init_type_t type,
                                CFS_SETGRP_PERM;
                 }
         } else {
-                ucred->mu_identity = mdt_identity_get(mdt->mdt_identity_cache,
-                                                      pud->pud_uid);
-                if (!ucred->mu_identity) {
-                        CERROR("Deny access without identity: uid %d\n",
-                               pud->pud_uid);
-                RETURN(-EACCES);
+                struct md_identity *identity;
+
+                identity = mdt_identity_get(mdt->mdt_identity_cache,
+                                            pud->pud_uid);
+                if (IS_ERR(identity)) {
+                        if (unlikely(PTR_ERR(identity) == -EREMCHG &&
+                                     !remote)) {
+                                ucred->mu_identity = NULL;
+                                perm = CFS_SETUID_PERM | CFS_SETGID_PERM |
+                                       CFS_SETGRP_PERM;
+                        } else {
+                                CDEBUG(D_SEC, "Deny access without identity: uid %u\n",
+                                       pud->pud_uid);
+                                RETURN(-EACCES);
+                        }
                 } else {
+                        ucred->mu_identity = identity;
                         perm = mdt_identity_get_perm(ucred->mu_identity,
-                                                   med->med_rmtclient,
-                                                   peernid);
+                                                     remote, peernid);
                 }
         }
 
@@ -170,39 +227,39 @@ static int new_init_ucred(struct mdt_thread_info *info, ucred_init_type_t type,
 
         /* check permission of setuid */
         if (setuid && !(perm & CFS_SETUID_PERM)) {
-                CWARN("mdt blocked setuid attempt (%u -> %u) from "
-                      LPX64"\n", pud->pud_uid, pud->pud_fsuid, peernid);
+                CDEBUG(D_SEC, "mdt blocked setuid attempt (%u -> %u) from %s\n",
+                       pud->pud_uid, pud->pud_fsuid, libcfs_nid2str(peernid));
                 GOTO(out, rc = -EACCES);
         }
 
         /* check permission of setgid */
         if (setgid && !(perm & CFS_SETGID_PERM)) {
-                CWARN("mdt blocked setgid attempt (%u:%u/%u:%u -> %u) "
-                      "from "LPX64"\n", pud->pud_uid, pud->pud_gid,
-                      pud->pud_fsuid, pud->pud_fsgid,
-                      ucred->mu_identity->mi_gid, peernid);
+                CDEBUG(D_SEC, "mdt blocked setgid attempt (%u:%u/%u:%u -> %u) "
+                       "from %s\n", pud->pud_uid, pud->pud_gid,
+                       pud->pud_fsuid, pud->pud_fsgid,
+                       ucred->mu_identity->mi_gid, libcfs_nid2str(peernid));
                 GOTO(out, rc = -EACCES);
         }
 
         /*
          * NB: remote client not allowed to setgroups anyway.
          */
-        if (!med->med_rmtclient && perm & CFS_SETGRP_PERM) {
+        if (!remote && perm & CFS_SETGRP_PERM) {
                 if (pud->pud_ngroups) {
-                /* setgroups for local client */
-                        ucred->mu_ginfo = groups_alloc(pud->pud_ngroups);
+                        /* setgroups for local client */
+                        ucred->mu_ginfo = cfs_groups_alloc(pud->pud_ngroups);
                         if (!ucred->mu_ginfo) {
-                        CERROR("failed to alloc %d groups\n",
-                               pud->pud_ngroups);
-                        GOTO(out, rc = -ENOMEM);
-                }
+                                CERROR("failed to alloc %d groups\n",
+                                       pud->pud_ngroups);
+                                GOTO(out, rc = -ENOMEM);
+                        }
 
                         lustre_groups_from_list(ucred->mu_ginfo,
                                                 pud->pud_groups);
                         lustre_groups_sort(ucred->mu_ginfo);
-        } else {
-                ucred->mu_ginfo = NULL;
-        }
+                } else {
+                        ucred->mu_ginfo = NULL;
+                }
         } else {
                 ucred->mu_suppgids[0] = -1;
                 ucred->mu_suppgids[1] = -1;
@@ -214,14 +271,17 @@ static int new_init_ucred(struct mdt_thread_info *info, ucred_init_type_t type,
         ucred->mu_fsuid = pud->pud_fsuid;
         ucred->mu_fsgid = pud->pud_fsgid;
 
-        /* XXX: need to process root_squash here. */
-        mdt_root_squash(info);
+        /* process root_squash here. */
+        mdt_root_squash(info, peernid);
 
-        /* remove fs privilege for non-root user */
+        /* remove fs privilege for non-root user. */
         if (ucred->mu_fsuid)
-                ucred->mu_cap = pud->pud_cap & ~CAP_FS_MASK;
+                ucred->mu_cap = pud->pud_cap & ~CFS_CAP_FS_MASK;
         else
                 ucred->mu_cap = pud->pud_cap;
+        if (remote && !(perm & CFS_RMTOWN_PERM))
+                ucred->mu_cap &= ~(CFS_CAP_SYS_RESOURCE_MASK |
+                                   CFS_CAP_CHOWN_MASK);
         ucred->mu_valid = UCRED_NEW;
 
         EXIT;
@@ -229,7 +289,7 @@ static int new_init_ucred(struct mdt_thread_info *info, ucred_init_type_t type,
 out:
         if (rc) {
                 if (ucred->mu_ginfo) {
-                        groups_free(ucred->mu_ginfo);
+                        cfs_put_group_info(ucred->mu_ginfo);
                         ucred->mu_ginfo = NULL;
                 }
                 if (ucred->mu_identity) {
@@ -245,13 +305,13 @@ out:
 int mdt_check_ucred(struct mdt_thread_info *info)
 {
         struct ptlrpc_request   *req = mdt_info_req(info);
-        struct mdt_export_data  *med = mdt_req2med(req);
         struct mdt_device       *mdt = info->mti_mdt;
         struct ptlrpc_user_desc *pud = req->rq_user_desc;
         struct md_ucred         *ucred = mdt_ucred(info);
         struct md_identity      *identity = NULL;
-        lnet_nid_t              peernid = req->rq_peer.nid;
+        lnet_nid_t               peernid = req->rq_peer.nid;
         __u32                    perm = 0;
+        __u32                    remote = exp_connect_rmtclient(info->mti_exp);
         int                      setuid;
         int                      setgid;
         int                      rc = 0;
@@ -266,38 +326,38 @@ int mdt_check_ucred(struct mdt_thread_info *info)
 
         /* sanity check: if we use strong authentication, we expect the
          * uid which client claimed is true */
-                if (med->med_rmtclient) {
-                        if (req->rq_auth_mapped_uid == INVALID_UID) {
-                                CWARN("remote user not mapped, deny access!\n");
-                                RETURN(-EACCES);
-                        }
+        if (remote) {
+                if (req->rq_auth_mapped_uid == INVALID_UID) {
+                        CDEBUG(D_SEC, "remote user not mapped, deny access!\n");
+                        RETURN(-EACCES);
+                }
 
-                        if (ptlrpc_user_desc_do_idmap(req, pud))
-                                RETURN(-EACCES);
+                if (ptlrpc_user_desc_do_idmap(req, pud))
+                        RETURN(-EACCES);
 
-                        if (req->rq_auth_mapped_uid != pud->pud_uid) {
-                        CERROR("remote client "LPU64": auth/mapped uid %u/%u "
-                                       "while client claim %u:%u/%u:%u\n",
-                               peernid, req->rq_auth_uid,
+                if (req->rq_auth_mapped_uid != pud->pud_uid) {
+                        CDEBUG(D_SEC, "remote client %s: auth/mapped uid %u/%u "
+                               "while client claims %u:%u/%u:%u\n",
+                               libcfs_nid2str(peernid), req->rq_auth_uid,
                                req->rq_auth_mapped_uid,
                                pud->pud_uid, pud->pud_gid,
                                pud->pud_fsuid, pud->pud_fsgid);
-                                RETURN(-EACCES);
-                        }
-                } else {
-                        if (req->rq_auth_uid != pud->pud_uid) {
-                                CERROR("local client "LPU64": auth uid %u "
-                                       "while client claim %u:%u/%u:%u\n",
-                                       peernid, req->rq_auth_uid, pud->pud_uid,
-                                       pud->pud_gid, pud->pud_fsuid,
-                                       pud->pud_fsgid);
-                                RETURN(-EACCES);
-                        }
+                        RETURN(-EACCES);
+                }
+        } else {
+                if (req->rq_auth_uid != pud->pud_uid) {
+                        CDEBUG(D_SEC, "local client %s: auth uid %u "
+                               "while client claims %u:%u/%u:%u\n",
+                               libcfs_nid2str(peernid), req->rq_auth_uid,
+                               pud->pud_uid, pud->pud_gid,
+                               pud->pud_fsuid, pud->pud_fsgid);
+                        RETURN(-EACCES);
                 }
+        }
 
         if (is_identity_get_disabled(mdt->mdt_identity_cache)) {
-                if (med->med_rmtclient) {
-                        CERROR("remote client must run with identity_get "
+                if (remote) {
+                        CDEBUG(D_SEC, "remote client must run with identity_get "
                                "enabled!\n");
                         RETURN(-EACCES);
                 }
@@ -305,12 +365,18 @@ int mdt_check_ucred(struct mdt_thread_info *info)
         }
 
         identity = mdt_identity_get(mdt->mdt_identity_cache, pud->pud_uid);
-        if (!identity) {
-                CERROR("Deny access without identity: uid %d\n", pud->pud_uid);
-                RETURN(-EACCES);
+        if (IS_ERR(identity)) {
+                if (unlikely(PTR_ERR(identity) == -EREMCHG &&
+                             !remote)) {
+                        RETURN(0);
+                } else {
+                        CDEBUG(D_SEC, "Deny access without identity: uid %u\n",
+                               pud->pud_uid);
+                        RETURN(-EACCES);
+               }
         }
 
-        perm = mdt_identity_get_perm(identity, med->med_rmtclient, peernid);
+        perm = mdt_identity_get_perm(identity, remote, peernid);
         /* find out the setuid/setgid attempt */
         setuid = (pud->pud_uid != pud->pud_fsuid);
         setgid = (pud->pud_gid != pud->pud_fsgid ||
@@ -318,17 +384,17 @@ int mdt_check_ucred(struct mdt_thread_info *info)
 
         /* check permission of setuid */
         if (setuid && !(perm & CFS_SETUID_PERM)) {
-                CWARN("mdt blocked setuid attempt (%u -> %u) from "
-                      LPX64"\n", pud->pud_uid, pud->pud_fsuid, peernid);
+                CDEBUG(D_SEC, "mdt blocked setuid attempt (%u -> %u) from %s\n",
+                       pud->pud_uid, pud->pud_fsuid, libcfs_nid2str(peernid));
                 GOTO(out, rc = -EACCES);
         }
 
         /* check permission of setgid */
         if (setgid && !(perm & CFS_SETGID_PERM)) {
-                CWARN("mdt blocked setgid attempt (%u:%u/%u:%u -> %u) "
-                      "from "LPX64"\n", pud->pud_uid, pud->pud_gid,
-                      pud->pud_fsuid, pud->pud_fsgid, identity->mi_gid,
-                      peernid);
+                CDEBUG(D_SEC, "mdt blocked setgid attempt (%u:%u/%u:%u -> %u) "
+                       "from %s\n", pud->pud_uid, pud->pud_gid,
+                       pud->pud_fsuid, pud->pud_fsgid, identity->mi_gid,
+                       libcfs_nid2str(peernid));
                 GOTO(out, rc = -EACCES);
         }
 
@@ -359,20 +425,24 @@ static int old_init_ucred(struct mdt_thread_info *info,
         if (!is_identity_get_disabled(mdt->mdt_identity_cache)) {
                 identity = mdt_identity_get(mdt->mdt_identity_cache,
                                             uc->mu_fsuid);
-                if (!identity) {
-                        CERROR("Deny access without identity: uid %d\n",
-                               uc->mu_fsuid);
-                        RETURN(-EACCES);
+                if (IS_ERR(identity)) {
+                        if (unlikely(PTR_ERR(identity) == -EREMCHG)) {
+                                identity = NULL;
+                        } else {
+                                CDEBUG(D_SEC, "Deny access without identity: "
+                                       "uid %u\n", uc->mu_fsuid);
+                                RETURN(-EACCES);
+                        }
                 }
         }
         uc->mu_identity = identity;
 
-        /* XXX: need to process root_squash here. */
-        mdt_root_squash(info);
+        /* process root_squash here. */
+        mdt_root_squash(info, mdt_info_req(info)->rq_peer.nid);
 
-        /* remove fs privilege for non-root user */
+        /* remove fs privilege for non-root user. */
         if (uc->mu_fsuid)
-                uc->mu_cap = body->capability & ~CAP_FS_MASK;
+                uc->mu_cap = body->capability & ~CFS_CAP_FS_MASK;
         else
                 uc->mu_cap = body->capability;
         uc->mu_valid = UCRED_OLD;
@@ -395,20 +465,24 @@ static int old_init_ucred_reint(struct mdt_thread_info *info)
         if (!is_identity_get_disabled(mdt->mdt_identity_cache)) {
                 identity = mdt_identity_get(mdt->mdt_identity_cache,
                                             uc->mu_fsuid);
-                if (!identity) {
-                        CERROR("Deny access without identity: uid %d\n",
-                               uc->mu_fsuid);
-                        RETURN(-EACCES);
-        }
+                if (IS_ERR(identity)) {
+                        if (unlikely(PTR_ERR(identity) == -EREMCHG)) {
+                                identity = NULL;
+                        } else {
+                                CDEBUG(D_SEC, "Deny access without identity: "
+                                       "uid %u\n", uc->mu_fsuid);
+                                RETURN(-EACCES);
+                        }
+                }
         }
         uc->mu_identity = identity;
 
-        /* XXX: need to process root_squash here. */
-        mdt_root_squash(info);
+        /* process root_squash here. */
+        mdt_root_squash(info, mdt_info_req(info)->rq_peer.nid);
 
-        /* remove fs privilege for non-root user */
+        /* remove fs privilege for non-root user. */
         if (uc->mu_fsuid)
-                uc->mu_cap &= ~CAP_FS_MASK;
+                uc->mu_cap &= ~CFS_CAP_FS_MASK;
         uc->mu_valid = UCRED_OLD;
 
         RETURN(0);
@@ -450,30 +524,34 @@ int mdt_init_ucred_reint(struct mdt_thread_info *info)
 void mdt_dump_lmm(int level, const struct lov_mds_md *lmm)
 {
         const struct lov_ost_data_v1 *lod;
-        int i;
-        __s16 stripe_count =
-                le16_to_cpu(((struct lov_user_md*)lmm)->lmm_stripe_count);
+        int                           i;
+        __u16                         count;
+
+        count = le16_to_cpu(((struct lov_user_md*)lmm)->lmm_stripe_count);
 
         CDEBUG(level, "objid "LPX64", magic 0x%08X, pattern %#X\n",
                le64_to_cpu(lmm->lmm_object_id), le32_to_cpu(lmm->lmm_magic),
                le32_to_cpu(lmm->lmm_pattern));
         CDEBUG(level,"stripe_size=0x%x, stripe_count=0x%x\n",
-               le32_to_cpu(lmm->lmm_stripe_size),
-               le32_to_cpu(lmm->lmm_stripe_count));
-        LASSERT(stripe_count <= (__s16)LOV_MAX_STRIPE_COUNT);
-        for (i = 0, lod = lmm->lmm_objects; i < stripe_count; i++, lod++) {
+               le32_to_cpu(lmm->lmm_stripe_size), count);
+        if (count == LOV_ALL_STRIPES)
+                return;
+        LASSERT(count <= LOV_MAX_STRIPE_COUNT);
+        for (i = 0, lod = lmm->lmm_objects; i < count; i++, lod++)
                 CDEBUG(level, "stripe %u idx %u subobj "LPX64"/"LPX64"\n",
                        i, le32_to_cpu(lod->l_ost_idx),
-                       le64_to_cpu(lod->l_object_gr),
+                       le64_to_cpu(lod->l_object_seq),
                        le64_to_cpu(lod->l_object_id));
-        }
 }
 
-void mdt_shrink_reply(struct mdt_thread_info *info)
+/* Shrink and/or grow reply buffers */
+int mdt_fix_reply(struct mdt_thread_info *info)
 {
-        struct req_capsule    *pill = &info->mti_pill;
-        struct mdt_body       *body;
-        int acl_size, md_size, adjust = 0;
+        struct req_capsule *pill = info->mti_pill;
+        struct mdt_body    *body;
+        int                md_size, md_packed = 0;
+        int                acl_size;
+        int                rc = 0;
         ENTRY;
 
         body = req_capsule_server_get(pill, &RMF_MDT_BODY);
@@ -486,11 +564,17 @@ void mdt_shrink_reply(struct mdt_thread_info *info)
 
         acl_size = body->aclsize;
 
-        CDEBUG(D_INFO, "Shrink to md_size = %d cookie/acl_size = %d" 
-                        " MDSCAPA = %d, OSSCAPA = %d\n",
+        /* this replay - not send info to client */
+        if (info->mti_spec.no_create == 1) {
+                md_size = 0;
+                acl_size = 0;
+        }
+
+        CDEBUG(D_INFO, "Shrink to md_size = %d cookie/acl_size = %d"
+                        " MDSCAPA = %llx, OSSCAPA = %llx\n",
                         md_size, acl_size,
-                        (int)(body->valid & OBD_MD_FLMDSCAPA),
-                        (int)(body->valid & OBD_MD_FLOSSCAPA));
+                        (unsigned long long)(body->valid & OBD_MD_FLMDSCAPA),
+                        (unsigned long long)(body->valid & OBD_MD_FLOSSCAPA));
 /*
             &RMF_MDT_BODY,
             &RMF_MDT_MD,
@@ -499,39 +583,75 @@ void mdt_shrink_reply(struct mdt_thread_info *info)
 (optional)  &RMF_CAPA2,
 (optional)  something else
 */
-        adjust += req_capsule_shrink(pill, &RMF_MDT_MD,
-                                    md_size, adjust, 1);
+
+        /* MDT_MD buffer may be bigger than packed value, let's shrink all
+         * buffers before growing it */
+       if (info->mti_big_lmm_used) {
+                LASSERT(req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER));
+                md_packed = req_capsule_get_size(pill, &RMF_MDT_MD,
+                                                 RCL_SERVER);
+                LASSERT(md_packed > 0);
+                /* buffer must be allocated separately */
+                LASSERT(info->mti_attr.ma_lmm !=
+                        req_capsule_server_get(pill, &RMF_MDT_MD));
+                req_capsule_shrink(pill, &RMF_MDT_MD, 0, RCL_SERVER);
+                /* free big lmm if md_size is not needed */
+                if (md_size == 0)
+                       info->mti_big_lmm_used = 0;
+        } else if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER)) {
+                req_capsule_shrink(pill, &RMF_MDT_MD, md_size, RCL_SERVER);
+        }
 
         if (req_capsule_has_field(pill, &RMF_ACL, RCL_SERVER))
-                adjust += req_capsule_shrink(pill, &RMF_ACL,
-                                            acl_size, adjust, 1);
+                req_capsule_shrink(pill, &RMF_ACL, acl_size, RCL_SERVER);
         else if (req_capsule_has_field(pill, &RMF_LOGCOOKIES, RCL_SERVER))
-                adjust += req_capsule_shrink(pill, &RMF_LOGCOOKIES,
-                                            acl_size, adjust, 1);
-
-        /* RMF_CAPA1 on server-side maybe for OBD_MD_FLMDSCAPA or
-         * OBD_MD_FLOSSCAPA. If RMF_CAPA2 exist also, RMF_CAPA1 is
-         * for OBD_MD_FLMDSCAPA only. */
-        if (req_capsule_has_field(pill, &RMF_CAPA1, RCL_SERVER)) {
-                if ((req_capsule_has_field(pill, &RMF_CAPA2, RCL_SERVER) &&
-                    !(body->valid & OBD_MD_FLMDSCAPA)) ||
-                    (!req_capsule_has_field(pill, &RMF_CAPA2, RCL_SERVER) &&
-                    !(body->valid & OBD_MD_FLMDSCAPA) &&
-                    !(body->valid & OBD_MD_FLOSSCAPA)))
-                        adjust += req_capsule_shrink(pill, &RMF_CAPA1,
-                                                     0, adjust, 1);
-        }
+                req_capsule_shrink(pill, &RMF_LOGCOOKIES,
+                                   acl_size, RCL_SERVER);
+
+        if (req_capsule_has_field(pill, &RMF_CAPA1, RCL_SERVER) &&
+            !(body->valid & OBD_MD_FLMDSCAPA))
+                req_capsule_shrink(pill, &RMF_CAPA1, 0, RCL_SERVER);
 
-        /* RMF_CAPA2 on server-side is for OBD_MD_FLOSSCAPA only. */
-        if ((req_capsule_has_field(pill, &RMF_CAPA2, RCL_SERVER) &&
-                !(body->valid & OBD_MD_FLOSSCAPA)))
-                adjust += req_capsule_shrink(pill, &RMF_CAPA2, 0, adjust, 0);
+        if (req_capsule_has_field(pill, &RMF_CAPA2, RCL_SERVER) &&
+            !(body->valid & OBD_MD_FLOSSCAPA))
+                req_capsule_shrink(pill, &RMF_CAPA2, 0, RCL_SERVER);
 
         /*
          * Some more field should be shrinked if needed.
          * This should be done by those who added fields to reply message.
          */
-        EXIT;
+
+        /* Grow MD buffer if needed finally */
+       if (info->mti_big_lmm_used) {
+                void *lmm;
+
+                LASSERT(md_size > md_packed);
+                CDEBUG(D_INFO, "Enlarge reply buffer, need extra %d bytes\n",
+                       md_size - md_packed);
+                rc = req_capsule_server_grow(pill, &RMF_MDT_MD, md_size);
+                if (rc) {
+                        /* we can't answer with proper LOV EA, drop flags,
+                         * the rc is also returned so this request is
+                         * considered as failed */
+                        body->valid &= ~(OBD_MD_FLDIREA | OBD_MD_FLEASIZE);
+                        /* don't return transno along with error */
+                        lustre_msg_set_transno(pill->rc_req->rq_repmsg, 0);
+                } else {
+                        /* now we need to pack right LOV EA */
+                        lmm = req_capsule_server_get(pill, &RMF_MDT_MD);
+                        LASSERT(req_capsule_get_size(pill, &RMF_MDT_MD,
+                                                     RCL_SERVER) ==
+                                info->mti_attr.ma_lmm_size);
+                        memcpy(lmm, info->mti_attr.ma_lmm,
+                               info->mti_attr.ma_lmm_size);
+                }
+                /* update mdt_max_mdsize so clients will be aware about that */
+                if (info->mti_mdt->mdt_max_mdsize < info->mti_attr.ma_lmm_size)
+                        info->mti_mdt->mdt_max_mdsize =
+                                                    info->mti_attr.ma_lmm_size;
+               info->mti_big_lmm_used = 0;
+        }
+        RETURN(rc);
 }
 
 
@@ -542,9 +662,10 @@ int mdt_handle_last_unlink(struct mdt_thread_info *info, struct mdt_object *mo,
 {
         struct mdt_body       *repbody;
         const struct lu_attr *la = &ma->ma_attr;
+        int rc;
         ENTRY;
 
-        repbody = req_capsule_server_get(&info->mti_pill, &RMF_MDT_BODY);
+        repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
         LASSERT(repbody != NULL);
 
         if (ma->ma_valid & MA_INODE)
@@ -576,6 +697,21 @@ int mdt_handle_last_unlink(struct mdt_thread_info *info, struct mdt_object *mo,
                 repbody->valid |= OBD_MD_FLCOOKIE;
         }
 
+        if (info->mti_mdt->mdt_opts.mo_oss_capa &&
+            info->mti_exp->exp_connect_flags & OBD_CONNECT_OSS_CAPA &&
+            repbody->valid & OBD_MD_FLEASIZE) {
+                struct lustre_capa *capa;
+
+                capa = req_capsule_server_get(info->mti_pill, &RMF_CAPA2);
+                LASSERT(capa);
+                capa->lc_opc = CAPA_OPC_OSS_DESTROY;
+                rc = mo_capa_get(info->mti_env, mdt_object_child(mo), capa, 0);
+                if (rc)
+                        RETURN(rc);
+
+                repbody->valid |= OBD_MD_FLOSSCAPA;
+        }
+
         RETURN(0);
 }
 
@@ -637,7 +773,7 @@ static __u64 mdt_attr_valid_xlate(__u64 in, struct mdt_reint_record *rr,
                 out |= LA_BLOCKS;
 
         if (in & ATTR_FROM_OPEN)
-                rr->rr_flags |= MRF_SETATTR_LOCKED;
+                rr->rr_flags |= MRF_OPEN_TRUNC;
 
         if (in & ATTR_ATIME_SET)
                 out |= LA_ATIME;
@@ -651,16 +787,26 @@ static __u64 mdt_attr_valid_xlate(__u64 in, struct mdt_reint_record *rr,
         if (in & ATTR_ATTR_FLAG)
                 out |= LA_FLAGS;
 
+        if (in & ATTR_KILL_SUID)
+                out |= LA_KILL_SUID;
+
+        if (in & ATTR_KILL_SGID)
+                out |= LA_KILL_SGID;
+
         if (in & MDS_OPEN_OWNEROVERRIDE)
-                out |= MDS_OPEN_OWNEROVERRIDE;
+                ma->ma_attr_flags |= MDS_OPEN_OWNEROVERRIDE;
+
+        if (in & ATTR_FORCE)
+                ma->ma_attr_flags |= MDS_PERM_BYPASS;
 
         /*XXX need ATTR_RAW?*/
         in &= ~(ATTR_MODE|ATTR_UID|ATTR_GID|ATTR_SIZE|ATTR_BLOCKS|
                 ATTR_ATIME|ATTR_MTIME|ATTR_CTIME|ATTR_FROM_OPEN|
                 ATTR_ATIME_SET|ATTR_CTIME_SET|ATTR_MTIME_SET|
-                ATTR_ATTR_FLAG|ATTR_RAW|MDS_OPEN_OWNEROVERRIDE);
+                ATTR_ATTR_FLAG|ATTR_RAW|MDS_OPEN_OWNEROVERRIDE|
+                ATTR_FORCE|ATTR_KILL_SUID|ATTR_KILL_SGID);
         if (in != 0)
-                CERROR("Unknown attr bits: %#llx\n", in);
+                CERROR("Unknown attr bits: "LPX64"\n", in);
         return out;
 }
 /* unpacking */
@@ -670,12 +816,13 @@ static int mdt_setattr_unpack_rec(struct mdt_thread_info *info)
         struct md_ucred         *uc  = mdt_ucred(info);
         struct md_attr          *ma = &info->mti_attr;
         struct lu_attr          *la = &ma->ma_attr;
-        struct req_capsule      *pill = &info->mti_pill;
+        struct req_capsule      *pill = info->mti_pill;
         struct mdt_reint_record *rr = &info->mti_rr;
         struct mdt_rec_setattr  *rec;
         ENTRY;
 
-        rec = req_capsule_client_get(pill, &RMF_REC_SETATTR);
+        CLASSERT(sizeof(struct mdt_rec_setattr)== sizeof(struct mdt_rec_reint));
+        rec = req_capsule_client_get(pill, &RMF_REC_REINT);
         if (rec == NULL)
                 RETURN(-EFAULT);
 
@@ -705,34 +852,36 @@ static int mdt_setattr_unpack_rec(struct mdt_thread_info *info)
         RETURN(0);
 }
 
-static int mdt_epoch_unpack(struct mdt_thread_info *info)
+static int mdt_ioepoch_unpack(struct mdt_thread_info *info)
 {
-        struct req_capsule *pill = &info->mti_pill;
+        struct req_capsule *pill = info->mti_pill;
         ENTRY;
 
         if (req_capsule_get_size(pill, &RMF_MDT_EPOCH, RCL_CLIENT))
-                info->mti_epoch = req_capsule_client_get(pill, &RMF_MDT_EPOCH);
+                info->mti_ioepoch =
+                        req_capsule_client_get(pill, &RMF_MDT_EPOCH);
         else
-                info->mti_epoch = NULL;
-        RETURN(info->mti_epoch == NULL ? -EFAULT : 0);
+                info->mti_ioepoch = NULL;
+        RETURN(info->mti_ioepoch == NULL ? -EFAULT : 0);
 }
 
 static inline int mdt_dlmreq_unpack(struct mdt_thread_info *info) {
-        struct req_capsule      *pill = &info->mti_pill;
+        struct req_capsule      *pill = info->mti_pill;
 
         if (req_capsule_get_size(pill, &RMF_DLM_REQ, RCL_CLIENT)) {
                 info->mti_dlm_req = req_capsule_client_get(pill, &RMF_DLM_REQ);
                 if (info->mti_dlm_req == NULL)
                         RETURN(-EFAULT);
         }
-        
+
         RETURN(0);
 }
 
 static int mdt_setattr_unpack(struct mdt_thread_info *info)
 {
+        struct mdt_reint_record *rr = &info->mti_rr;
         struct md_attr          *ma = &info->mti_attr;
-        struct req_capsule      *pill = &info->mti_pill;
+        struct req_capsule      *pill = info->mti_pill;
         int rc;
         ENTRY;
 
@@ -741,12 +890,17 @@ static int mdt_setattr_unpack(struct mdt_thread_info *info)
                 RETURN(rc);
 
         /* Epoch may be absent */
-        mdt_epoch_unpack(info);
-
-        ma->ma_lmm_size = req_capsule_get_size(pill, &RMF_EADATA, RCL_CLIENT);
-        if (ma->ma_lmm_size) {
-                ma->ma_lmm = req_capsule_client_get(pill, &RMF_EADATA);
-                ma->ma_valid |= MA_LOV;
+        mdt_ioepoch_unpack(info);
+
+        if (req_capsule_field_present(pill, &RMF_EADATA, RCL_CLIENT)) {
+                rr->rr_eadata = req_capsule_client_get(pill, &RMF_EADATA);
+                rr->rr_eadatalen = req_capsule_get_size(pill, &RMF_EADATA,
+                                                        RCL_CLIENT);
+                ma->ma_lmm_size = rr->rr_eadatalen;
+                if (ma->ma_lmm_size > 0) {
+                        ma->ma_lmm = (void *)rr->rr_eadata;
+                        ma->ma_valid |= MA_LOV;
+                }
         }
 
         ma->ma_cookie_size = req_capsule_get_size(pill, &RMF_LOGCOOKIES,
@@ -765,7 +919,7 @@ int mdt_close_unpack(struct mdt_thread_info *info)
         int rc;
         ENTRY;
 
-        rc = mdt_epoch_unpack(info);
+        rc = mdt_ioepoch_unpack(info);
         if (rc)
                 RETURN(rc);
 
@@ -778,12 +932,13 @@ static int mdt_create_unpack(struct mdt_thread_info *info)
         struct mdt_rec_create   *rec;
         struct lu_attr          *attr = &info->mti_attr.ma_attr;
         struct mdt_reint_record *rr = &info->mti_rr;
-        struct req_capsule      *pill = &info->mti_pill;
+        struct req_capsule      *pill = info->mti_pill;
         struct md_op_spec       *sp = &info->mti_spec;
         int rc;
         ENTRY;
 
-        rec = req_capsule_client_get(pill, &RMF_REC_CREATE);
+        CLASSERT(sizeof(struct mdt_rec_create) == sizeof(struct mdt_rec_reint));
+        rec = req_capsule_client_get(pill, &RMF_REC_REINT);
         if (rec == NULL)
                 RETURN(-EFAULT);
 
@@ -805,7 +960,7 @@ static int mdt_create_unpack(struct mdt_thread_info *info)
         attr->la_valid = LA_MODE | LA_RDEV | LA_UID | LA_GID |
                          LA_CTIME | LA_MTIME | LA_ATIME;
         memset(&sp->u, 0, sizeof(sp->u));
-        sp->sp_cr_flags = rec->cr_flags;
+        sp->sp_cr_flags = get_mrc_cr_flags(rec);
         sp->sp_ck_split = !!(rec->cr_bias & MDS_CHECK_SPLIT);
         info->mti_cross_ref = !!(rec->cr_bias & MDS_CROSS_REF);
 
@@ -814,10 +969,16 @@ static int mdt_create_unpack(struct mdt_thread_info *info)
                                  req_capsule_client_get(pill, &RMF_CAPA1));
         mdt_set_capainfo(info, 1, rr->rr_fid2, BYPASS_CAPA);
 
-        rr->rr_name = req_capsule_client_get(pill, &RMF_NAME);
-        rr->rr_namelen = req_capsule_get_size(pill, &RMF_NAME, RCL_CLIENT) - 1;
-        LASSERT(rr->rr_namelen > 0);
-        
+        if (!info->mti_cross_ref) {
+                rr->rr_name = req_capsule_client_get(pill, &RMF_NAME);
+                rr->rr_namelen = req_capsule_get_size(pill, &RMF_NAME,
+                                                      RCL_CLIENT) - 1;
+                LASSERT(rr->rr_name && rr->rr_namelen > 0);
+        } else {
+                rr->rr_name = NULL;
+                rr->rr_namelen = 0;
+        }
+
 #ifdef CONFIG_FS_POSIX_ACL
         if (sp->sp_cr_flags & MDS_CREATE_RMT_ACL) {
                 if (S_ISDIR(attr->la_mode))
@@ -825,9 +986,11 @@ static int mdt_create_unpack(struct mdt_thread_info *info)
                 req_capsule_extend(pill, &RQF_MDS_REINT_CREATE_RMT_ACL);
                 LASSERT(req_capsule_field_present(pill, &RMF_EADATA,
                                                   RCL_CLIENT));
-                sp->u.sp_ea.eadata = req_capsule_client_get(pill, &RMF_EADATA);
-                sp->u.sp_ea.eadatalen = req_capsule_get_size(pill, &RMF_EADATA,
-                                                                RCL_CLIENT);
+                rr->rr_eadata = req_capsule_client_get(pill, &RMF_EADATA);
+                rr->rr_eadatalen = req_capsule_get_size(pill, &RMF_EADATA,
+                                                        RCL_CLIENT);
+                sp->u.sp_ea.eadata = rr->rr_eadata;
+                sp->u.sp_ea.eadatalen = rr->rr_eadatalen;
                 sp->u.sp_ea.fid = rr->rr_fid1;
                 RETURN(0);
         }
@@ -842,11 +1005,13 @@ static int mdt_create_unpack(struct mdt_thread_info *info)
                        req_capsule_extend(pill, &RQF_MDS_REINT_CREATE_SLAVE);
                        LASSERT(req_capsule_field_present(pill, &RMF_EADATA,
                                                          RCL_CLIENT));
-                       sp->u.sp_ea.eadata = req_capsule_client_get(pill,
-                                                                   &RMF_EADATA);
-                       sp->u.sp_ea.eadatalen = req_capsule_get_size(pill,
-                                                                    &RMF_EADATA,
-                                                                    RCL_CLIENT);
+                       rr->rr_eadata = req_capsule_client_get(pill,
+                                                              &RMF_EADATA);
+                       rr->rr_eadatalen = req_capsule_get_size(pill,
+                                                               &RMF_EADATA,
+                                                               RCL_CLIENT);
+                       sp->u.sp_ea.eadata = rr->rr_eadata;
+                       sp->u.sp_ea.eadatalen = rr->rr_eadatalen;
                        sp->u.sp_ea.fid = rr->rr_fid1;
                        RETURN(0);
                 }
@@ -874,11 +1039,12 @@ static int mdt_link_unpack(struct mdt_thread_info *info)
         struct mdt_rec_link     *rec;
         struct lu_attr          *attr = &info->mti_attr.ma_attr;
         struct mdt_reint_record *rr = &info->mti_rr;
-        struct req_capsule      *pill = &info->mti_pill;
+        struct req_capsule      *pill = info->mti_pill;
         int rc;
         ENTRY;
 
-        rec = req_capsule_client_get(pill, &RMF_REC_LINK);
+        CLASSERT(sizeof(struct mdt_rec_link) == sizeof(struct mdt_rec_reint));
+        rec = req_capsule_client_get(pill, &RMF_REC_REINT);
         if (rec == NULL)
                 RETURN(-EFAULT);
 
@@ -903,13 +1069,14 @@ static int mdt_link_unpack(struct mdt_thread_info *info)
                 mdt_set_capainfo(info, 1, rr->rr_fid2,
                                  req_capsule_client_get(pill, &RMF_CAPA2));
 
+        info->mti_spec.sp_ck_split = !!(rec->lk_bias & MDS_CHECK_SPLIT);
+        info->mti_cross_ref = !!(rec->lk_bias & MDS_CROSS_REF);
         rr->rr_name = req_capsule_client_get(pill, &RMF_NAME);
         if (rr->rr_name == NULL)
                 RETURN(-EFAULT);
         rr->rr_namelen = req_capsule_get_size(pill, &RMF_NAME, RCL_CLIENT) - 1;
-        LASSERT(rr->rr_namelen > 0);
-        info->mti_spec.sp_ck_split = !!(rec->lk_bias & MDS_CHECK_SPLIT);
-        info->mti_cross_ref = !!(rec->lk_bias & MDS_CROSS_REF);
+        if (!info->mti_cross_ref)
+                LASSERT(rr->rr_namelen > 0);
 
         rc = mdt_dlmreq_unpack(info);
         RETURN(rc);
@@ -922,18 +1089,19 @@ static int mdt_unlink_unpack(struct mdt_thread_info *info)
         struct md_attr          *ma = &info->mti_attr;
         struct lu_attr          *attr = &info->mti_attr.ma_attr;
         struct mdt_reint_record *rr = &info->mti_rr;
-        struct req_capsule      *pill = &info->mti_pill;
+        struct req_capsule      *pill = info->mti_pill;
         int rc;
         ENTRY;
 
-        rec = req_capsule_client_get(pill, &RMF_REC_UNLINK);
+        CLASSERT(sizeof(struct mdt_rec_unlink) == sizeof(struct mdt_rec_reint));
+        rec = req_capsule_client_get(pill, &RMF_REC_REINT);
         if (rec == NULL)
                 RETURN(-EFAULT);
 
         uc->mu_fsuid = rec->ul_fsuid;
         uc->mu_fsgid = rec->ul_fsgid;
         uc->mu_cap   = rec->ul_cap;
-        uc->mu_suppgids[0] = rec->ul_suppgid;
+        uc->mu_suppgids[0] = rec->ul_suppgid1;
         uc->mu_suppgids[1] = -1;
 
         attr->la_uid = rec->ul_fsuid;
@@ -949,18 +1117,26 @@ static int mdt_unlink_unpack(struct mdt_thread_info *info)
                 mdt_set_capainfo(info, 0, rr->rr_fid1,
                                  req_capsule_client_get(pill, &RMF_CAPA1));
 
-        rr->rr_name = req_capsule_client_get(pill, &RMF_NAME);
-        if (rr->rr_name == NULL)
-                RETURN(-EFAULT);
-        rr->rr_namelen = req_capsule_get_size(pill, &RMF_NAME, RCL_CLIENT) - 1;
-        LASSERT(rr->rr_namelen > 0);
-        info->mti_spec.sp_ck_split = !!(rec->ul_bias & MDS_CHECK_SPLIT);
         info->mti_cross_ref = !!(rec->ul_bias & MDS_CROSS_REF);
+        if (!info->mti_cross_ref) {
+                rr->rr_name = req_capsule_client_get(pill, &RMF_NAME);
+                rr->rr_namelen = req_capsule_get_size(pill, &RMF_NAME, RCL_CLIENT) - 1;
+                if (rr->rr_name == NULL || rr->rr_namelen == 0)
+                        RETURN(-EFAULT);
+        } else {
+                rr->rr_name = NULL;
+                rr->rr_namelen = 0;
+        }
+        info->mti_spec.sp_ck_split = !!(rec->ul_bias & MDS_CHECK_SPLIT);
         if (rec->ul_bias & MDS_VTX_BYPASS)
                 ma->ma_attr_flags |= MDS_VTX_BYPASS;
         else
                 ma->ma_attr_flags &= ~MDS_VTX_BYPASS;
 
+        info->mti_spec.no_create = !!req_is_replay(mdt_info_req(info));
+        /* last unlink need LOV EA sent back */
+        rr->rr_eadatalen = info->mti_mdt->mdt_max_mdsize;
+
         rc = mdt_dlmreq_unpack(info);
         RETURN(rc);
 }
@@ -972,11 +1148,12 @@ static int mdt_rename_unpack(struct mdt_thread_info *info)
         struct md_attr          *ma = &info->mti_attr;
         struct lu_attr          *attr = &info->mti_attr.ma_attr;
         struct mdt_reint_record *rr = &info->mti_rr;
-        struct req_capsule      *pill = &info->mti_pill;
+        struct req_capsule      *pill = info->mti_pill;
         int rc;
         ENTRY;
 
-        rec = req_capsule_client_get(pill, &RMF_REC_RENAME);
+        CLASSERT(sizeof(struct mdt_rec_rename) == sizeof(struct mdt_rec_reint));
+        rec = req_capsule_client_get(pill, &RMF_REC_REINT);
         if (rec == NULL)
                 RETURN(-EFAULT);
 
@@ -1003,21 +1180,25 @@ static int mdt_rename_unpack(struct mdt_thread_info *info)
                 mdt_set_capainfo(info, 1, rr->rr_fid2,
                                  req_capsule_client_get(pill, &RMF_CAPA2));
 
+        info->mti_spec.sp_ck_split = !!(rec->rn_bias & MDS_CHECK_SPLIT);
+        info->mti_cross_ref = !!(rec->rn_bias & MDS_CROSS_REF);
         rr->rr_name = req_capsule_client_get(pill, &RMF_NAME);
         rr->rr_tgt = req_capsule_client_get(pill, &RMF_SYMTGT);
         if (rr->rr_name == NULL || rr->rr_tgt == NULL)
                 RETURN(-EFAULT);
         rr->rr_namelen = req_capsule_get_size(pill, &RMF_NAME, RCL_CLIENT) - 1;
-        LASSERT(rr->rr_namelen > 0);
         rr->rr_tgtlen = req_capsule_get_size(pill, &RMF_SYMTGT, RCL_CLIENT) - 1;
-        LASSERT(rr->rr_tgtlen > 0);
-        info->mti_spec.sp_ck_split = !!(rec->rn_bias & MDS_CHECK_SPLIT);
-        info->mti_cross_ref = !!(rec->rn_bias & MDS_CROSS_REF);
+        if (!info->mti_cross_ref)
+                LASSERT(rr->rr_namelen > 0 && rr->rr_tgtlen > 0);
         if (rec->rn_bias & MDS_VTX_BYPASS)
                 ma->ma_attr_flags |= MDS_VTX_BYPASS;
         else
                 ma->ma_attr_flags &= ~MDS_VTX_BYPASS;
 
+        info->mti_spec.no_create = !!req_is_replay(mdt_info_req(info));
+        /* rename may contain unlink so we might need LOV EA sent back */
+        rr->rr_eadatalen = info->mti_mdt->mdt_max_mdsize;
+
         rc = mdt_dlmreq_unpack(info);
         RETURN(rc);
 }
@@ -1027,13 +1208,14 @@ static int mdt_open_unpack(struct mdt_thread_info *info)
         struct md_ucred         *uc = mdt_ucred(info);
         struct mdt_rec_create   *rec;
         struct lu_attr          *attr = &info->mti_attr.ma_attr;
-        struct req_capsule      *pill = &info->mti_pill;
+        struct req_capsule      *pill = info->mti_pill;
         struct mdt_reint_record *rr   = &info->mti_rr;
         struct ptlrpc_request   *req  = mdt_info_req(info);
         struct md_op_spec       *sp   = &info->mti_spec;
         ENTRY;
 
-        rec = req_capsule_client_get(pill, &RMF_REC_CREATE);
+        CLASSERT(sizeof(struct mdt_rec_create) == sizeof(struct mdt_rec_reint));
+        rec = req_capsule_client_get(pill, &RMF_REC_REINT);
         if (rec == NULL)
                 RETURN(-EFAULT);
 
@@ -1056,7 +1238,10 @@ static int mdt_open_unpack(struct mdt_thread_info *info)
         attr->la_valid = LA_MODE  | LA_RDEV  | LA_UID   | LA_GID |
                          LA_CTIME | LA_MTIME | LA_ATIME;
         memset(&info->mti_spec.u, 0, sizeof(info->mti_spec.u));
-        info->mti_spec.sp_cr_flags = rec->cr_flags;
+        info->mti_spec.sp_cr_flags = get_mrc_cr_flags(rec);
+        /* Do not trigger ASSERTION if client miss to set such flags. */
+        if (unlikely(info->mti_spec.sp_cr_flags == 0))
+                RETURN(-EPROTO);
         info->mti_replayepoch = rec->cr_ioepoch;
 
         info->mti_spec.sp_ck_split = !!(rec->cr_bias & MDS_CHECK_SPLIT);
@@ -1065,8 +1250,8 @@ static int mdt_open_unpack(struct mdt_thread_info *info)
         if (req_capsule_get_size(pill, &RMF_CAPA1, RCL_CLIENT))
                 mdt_set_capainfo(info, 0, rr->rr_fid1,
                                  req_capsule_client_get(pill, &RMF_CAPA1));
-        if ((lustre_msg_get_flags(req->rq_reqmsg) & MSG_REPLAY) &&
-            (req_capsule_get_size(pill, &RMF_CAPA2, RCL_CLIENT))) {
+        if (req_is_replay(req) &&
+            req_capsule_get_size(pill, &RMF_CAPA2, RCL_CLIENT)) {
 #if 0
                 mdt_set_capainfo(info, 1, rr->rr_fid2,
                                  req_capsule_client_get(pill, &RMF_CAPA2));
@@ -1085,17 +1270,92 @@ static int mdt_open_unpack(struct mdt_thread_info *info)
                 RETURN(-EFAULT);
         rr->rr_namelen = req_capsule_get_size(pill, &RMF_NAME, RCL_CLIENT) - 1;
 
-        sp->u.sp_ea.eadatalen = req_capsule_get_size(pill, &RMF_EADATA,
-                                                     RCL_CLIENT);
-        if (sp->u.sp_ea.eadatalen) {
-                sp->u.sp_ea.eadata = req_capsule_client_get(pill, &RMF_EADATA);
-                if (lustre_msg_get_flags(req->rq_reqmsg) & MSG_REPLAY)
-                        sp->u.sp_ea.no_lov_create = 1;
+        if (req_capsule_field_present(pill, &RMF_EADATA, RCL_CLIENT)) {
+                rr->rr_eadatalen = req_capsule_get_size(pill, &RMF_EADATA,
+                                                        RCL_CLIENT);
+                if (rr->rr_eadatalen > 0) {
+                        rr->rr_eadata = req_capsule_client_get(pill,
+                                                               &RMF_EADATA);
+                        sp->u.sp_ea.eadatalen = rr->rr_eadatalen;
+                        sp->u.sp_ea.eadata = rr->rr_eadata;
+                        sp->no_create = !!req_is_replay(req);
+                }
+
+                /*
+                 * Client default md_size may be 0 right after client start,
+                 * until all osc are connected, set here just some reasonable
+                 * value to prevent misbehavior.
+                 */
+                if (rr->rr_eadatalen == 0 &&
+                    !(info->mti_spec.sp_cr_flags & MDS_OPEN_DELAY_CREATE))
+                       rr->rr_eadatalen = MIN_MD_SIZE;
+       }
+
+        RETURN(0);
+}
+
+static int mdt_setxattr_unpack(struct mdt_thread_info *info)
+{
+        struct mdt_reint_record   *rr   = &info->mti_rr;
+        struct md_ucred           *uc   = mdt_ucred(info);
+        struct lu_attr            *attr = &info->mti_attr.ma_attr;
+        struct req_capsule        *pill = info->mti_pill;
+        struct mdt_rec_setxattr   *rec;
+        ENTRY;
+
+
+        CLASSERT(sizeof(struct mdt_rec_setxattr) ==
+                         sizeof(struct mdt_rec_reint));
+
+        rec = req_capsule_client_get(pill, &RMF_REC_REINT);
+        if (rec == NULL)
+                RETURN(-EFAULT);
+
+        uc->mu_fsuid  = rec->sx_fsuid;
+        uc->mu_fsgid  = rec->sx_fsgid;
+        uc->mu_cap    = rec->sx_cap;
+        uc->mu_suppgids[0] = rec->sx_suppgid1;
+        uc->mu_suppgids[1] = -1;
+
+        rr->rr_opcode = rec->sx_opcode;
+        rr->rr_fid1   = &rec->sx_fid;
+        attr->la_valid = rec->sx_valid;
+        attr->la_ctime = rec->sx_time;
+        attr->la_size = rec->sx_size;
+        attr->la_flags = rec->sx_flags;
+
+        if (req_capsule_get_size(pill, &RMF_CAPA1, RCL_CLIENT))
+                mdt_set_capainfo(info, 0, rr->rr_fid1,
+                                 req_capsule_client_get(pill, &RMF_CAPA1));
+        else
+                mdt_set_capainfo(info, 0, rr->rr_fid1, BYPASS_CAPA);
+
+        rr->rr_name = req_capsule_client_get(pill, &RMF_NAME);
+        if (rr->rr_name == NULL)
+                RETURN(-EFAULT);
+        rr->rr_namelen = req_capsule_get_size(pill, &RMF_NAME, RCL_CLIENT) - 1;
+        LASSERT(rr->rr_namelen > 0);
+
+        if (req_capsule_field_present(pill, &RMF_EADATA, RCL_CLIENT)) {
+                rr->rr_eadatalen = req_capsule_get_size(pill, &RMF_EADATA,
+                                                        RCL_CLIENT);
+                if (rr->rr_eadatalen > 0) {
+                        rr->rr_eadata = req_capsule_client_get(pill,
+                                                               &RMF_EADATA);
+                        if (rr->rr_eadata == NULL)
+                                RETURN(-EFAULT);
+                } else {
+                        rr->rr_eadata = NULL;
+                }
+        } else if (!(attr->la_valid & OBD_MD_FLXATTRRM)) {
+                CDEBUG(D_INFO, "no xattr data supplied\n");
+                RETURN(-EFAULT);
         }
 
         RETURN(0);
 }
 
+
 typedef int (*reint_unpacker)(struct mdt_thread_info *info);
 
 static reint_unpacker mdt_reint_unpackers[REINT_MAX] = {
@@ -1104,7 +1364,8 @@ static reint_unpacker mdt_reint_unpackers[REINT_MAX] = {
         [REINT_LINK]     = mdt_link_unpack,
         [REINT_UNLINK]   = mdt_unlink_unpack,
         [REINT_RENAME]   = mdt_rename_unpack,
-        [REINT_OPEN]     = mdt_open_unpack
+        [REINT_OPEN]     = mdt_open_unpack,
+        [REINT_SETXATTR] = mdt_setxattr_unpack
 };
 
 int mdt_reint_unpack(struct mdt_thread_info *info, __u32 op)