Whamcloud - gitweb
LU-4877 mdt: tolerate zero md_packed in mdt_fix_reply
[fs/lustre-release.git] / lustre / mdt / mdt_lib.c
index bc04b2b..fa1d9ff 100644 (file)
@@ -1,44 +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>
+ * 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, 2013, Intel Corporation.
+ */
+/*
+ * 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 {
@@ -47,276 +58,282 @@ typedef enum ucred_init_type {
         REC_INIT        = 2
 } ucred_init_type_t;
 
-int groups_from_list(struct group_info *ginfo, gid_t *glist)
+void mdt_exit_ucred(struct mdt_thread_info *info)
 {
-        int i;
-        int count = ginfo->ngroups;
-
-        /* fill group_info from gid array */
-        for (i = 0; i < ginfo->nblocks; i++) {
-                int cp_count = min(NGROUPS_PER_BLOCK, count);
-                int off = i * NGROUPS_PER_BLOCK;
-                int len = cp_count * sizeof(*glist);
-
-                if (memcpy(ginfo->blocks[i], glist + off, len))
-                        return -EFAULT;
-
-                count -= cp_count;
-        }
-        return 0;
+       struct lu_ucred   *uc  = mdt_ucred(info);
+       struct mdt_device *mdt = info->mti_mdt;
+
+       LASSERT(uc != NULL);
+       if (uc->uc_valid != UCRED_INIT) {
+               uc->uc_suppgids[0] = uc->uc_suppgids[1] = -1;
+               if (uc->uc_ginfo) {
+                       put_group_info(uc->uc_ginfo);
+                       uc->uc_ginfo = NULL;
+               }
+               if (uc->uc_identity) {
+                       mdt_identity_put(mdt->mdt_identity_cache,
+                                        uc->uc_identity);
+                       uc->uc_identity = NULL;
+               }
+               uc->uc_valid = UCRED_INIT;
+       }
 }
 
-/* groups_sort() is copied from linux kernel! */
-/* a simple shell-metzner sort */
-void groups_sort(struct group_info *group_info)
+static int match_nosquash_list(struct rw_semaphore *sem,
+                              cfs_list_t *nidlist,
+                              lnet_nid_t peernid)
 {
-        int base, max, stride;
-        int gidsetsize = group_info->ngroups;
-
-        for (stride = 1; stride < gidsetsize; stride = 3 * stride + 1)
-                ; /* nothing */
-        stride /= 3;
-
-        while (stride) {
-                max = gidsetsize - stride;
-                for (base = 0; base < max; base++) {
-                        int left = base;
-                        int right = left + stride;
-                        gid_t tmp = GROUP_AT(group_info, right);
-
-                        while (left >= 0 && GROUP_AT(group_info, left) > tmp) {
-                                GROUP_AT(group_info, right) =
-                                    GROUP_AT(group_info, left);
-                                right = left;
-                                left -= stride;
-                        }
-                        GROUP_AT(group_info, right) = tmp;
-                }
-                stride /= 3;
-        }
+       int rc;
+       ENTRY;
+       down_read(sem);
+       rc = cfs_match_nid(peernid, nidlist);
+       up_read(sem);
+       RETURN(rc);
 }
 
-void mdt_exit_ucred(struct mdt_thread_info *info)
+/* root_squash for inter-MDS operations */
+static int mdt_root_squash(struct mdt_thread_info *info, lnet_nid_t peernid)
 {
-        struct md_ucred   *uc  = mdt_ucred(info);
-        struct mdt_device *mdt = info->mti_mdt;
-
-        if (uc->mu_valid != UCRED_INIT) {
-                uc->mu_suppgids[0] = uc->mu_suppgids[1] = -1;
-                if (uc->mu_ginfo) {
-                        groups_free(uc->mu_ginfo);
-                        uc->mu_ginfo = NULL;
-                }
-                if (uc->mu_identity) {
-                        mdt_identity_put(mdt->mdt_identity_cache,
-                                         uc->mu_identity);
-                        uc->mu_identity = NULL;
-                }
-                uc->mu_valid = UCRED_INIT;
-        }
+       struct lu_ucred *ucred = mdt_ucred(info);
+       struct root_squash_info *squash = &info->mti_mdt->mdt_squash;
+       ENTRY;
+
+       LASSERT(ucred != NULL);
+       if (!squash->rsi_uid || ucred->uc_fsuid)
+               RETURN(0);
+
+       if (match_nosquash_list(&squash->rsi_sem,
+                               &squash->rsi_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->uc_fsuid, ucred->uc_fsgid, ucred->uc_cap,
+              squash->rsi_uid, squash->rsi_gid, 0);
+
+       ucred->uc_fsuid = squash->rsi_uid;
+       ucred->uc_fsgid = squash->rsi_gid;
+       ucred->uc_cap = 0;
+       ucred->uc_suppgids[0] = -1;
+       ucred->uc_suppgids[1] = -1;
+
+       RETURN(0);
 }
 
-static int old_init_ucred(struct mdt_thread_info *info,
-                          struct mdt_body *body)
+static int new_init_ucred(struct mdt_thread_info *info, ucred_init_type_t type,
+                          void *buf)
 {
-        struct md_ucred     *uc  = mdt_ucred(info);
-        struct mdt_device   *mdt = info->mti_mdt;
-        struct mdt_identity *identity = NULL;
+        struct ptlrpc_request   *req = mdt_info_req(info);
+        struct mdt_device       *mdt = info->mti_mdt;
+        struct ptlrpc_user_desc *pud = req->rq_user_desc;
+       struct lu_ucred         *ucred = mdt_ucred(info);
+        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;
 
         ENTRY;
 
-        uc->mu_valid = UCRED_INVALID;
+        LASSERT(req->rq_auth_gss);
+        LASSERT(!req->rq_auth_usr_mdt);
+        LASSERT(req->rq_user_desc);
+       LASSERT(ucred != NULL);
 
-        if (!is_identity_get_disabled(mdt->mdt_identity_cache)) {
-                /* get identity info of this user */
-                identity = mdt_identity_get(mdt->mdt_identity_cache,
-                                            body->fsuid);
-                if (!identity) {
-                        CERROR("Deny access without identity: uid %d\n",
-                               body->fsuid);
-                        RETURN(-EACCES);
-                }
-        }
+       ucred->uc_valid = UCRED_INVALID;
 
-        uc->mu_valid = UCRED_OLD;
-        uc->mu_squash = SQUASH_NONE;
-        uc->mu_o_uid = uc->mu_uid = body->uid;
-        uc->mu_o_gid = uc->mu_gid = body->gid;
-        uc->mu_o_fsuid = uc->mu_fsuid = body->fsuid;
-        uc->mu_o_fsgid = uc->mu_fsgid = body->fsgid;
-        uc->mu_suppgids[0] = body->suppgid;
-        uc->mu_suppgids[1] = -1;
-        if (uc->mu_fsuid)
-                uc->mu_cap = body->capability & ~CAP_FS_MASK;
-        else
-                uc->mu_cap = body->capability;
-        uc->mu_ginfo = NULL;
-        uc->mu_identity = identity;
+       ucred->uc_o_uid   = pud->pud_uid;
+       ucred->uc_o_gid   = pud->pud_gid;
+       ucred->uc_o_fsuid = pud->pud_fsuid;
+       ucred->uc_o_fsgid = pud->pud_fsgid;
 
-        RETURN(0);
-}
+       if (type == BODY_INIT) {
+               struct mdt_body *body = (struct mdt_body *)buf;
 
-static int old_init_ucred_reint(struct mdt_thread_info *info)
-{
-        struct md_ucred     *uc  = mdt_ucred(info);
-        struct mdt_device   *mdt = info->mti_mdt;
-        struct mdt_identity *identity = NULL;
+               ucred->uc_suppgids[0] = body->suppgid;
+               ucred->uc_suppgids[1] = -1;
+       }
 
-        ENTRY;
+       /* sanity check: we expect the uid which client claimed is true */
+       if (remote) {
+               if (!uid_valid(make_kuid(&init_user_ns, req->rq_auth_mapped_uid))) {
+                       CDEBUG(D_SEC, "remote user not mapped, deny access!\n");
+                       CDEBUG(D_SEC, "remote user not mapped, deny access!\n");
+                       RETURN(-EACCES);
+               }
 
-        uc->mu_valid = UCRED_INVALID;
+               if (ptlrpc_user_desc_do_idmap(req, pud))
+                       RETURN(-EACCES);
 
-        if (!is_identity_get_disabled(mdt->mdt_identity_cache)) {
-                /* get identity info of this user */
-                identity = mdt_identity_get(mdt->mdt_identity_cache,
-                                            uc->mu_fsuid);
-                if (!identity) {
-                        CERROR("Deny access without identity: uid %d\n",
-                               uc->mu_fsuid);
+                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);
                 }
-        }
-
-        uc->mu_valid = UCRED_OLD;
-        uc->mu_squash = SQUASH_NONE;
-        uc->mu_o_uid = uc->mu_o_fsuid = uc->mu_uid = uc->mu_fsuid;
-        uc->mu_o_gid = uc->mu_o_fsgid = uc->mu_gid = uc->mu_fsgid;
-        if (uc->mu_fsuid)
-                uc->mu_cap &= ~CAP_FS_MASK;
-        uc->mu_ginfo = NULL;
-        uc->mu_identity = identity;
-
-        RETURN(0);
-}
-
-static int nid_nosquash(struct mdt_device *mdt, lnet_nid_t nid)
-{
-        struct rootsquash_info *rsi = mdt->mdt_rootsquash_info;
-        int i;
-
-        for (i = 0; i < rsi->rsi_n_nosquash_nids; i++)
-                if ((rsi->rsi_nosquash_nids[i] == nid) ||
-                    (rsi->rsi_nosquash_nids[i] == LNET_NID_ANY))
-                        return 1;
-
-        return 0;
-}
-
-static int mdt_squash_root(struct mdt_device *mdt, struct md_ucred *ucred,
-                           struct ptlrpc_user_desc *pud, lnet_nid_t peernid)
-{
-        struct rootsquash_info *rsi = mdt->mdt_rootsquash_info;
-
-        if (!rsi || (!rsi->rsi_uid && !rsi->rsi_gid) ||
-            nid_nosquash(mdt, peernid))
-                return 0;
-
-        CDEBUG(D_SEC, "squash req from "LPX64":"
-               "(%u:%u-%u:%u/%x)=>(%u:%u-%u:%u/%x)\n", peernid,
-               pud->pud_uid, pud->pud_gid,
-               pud->pud_fsuid, pud->pud_fsgid, pud->pud_cap,
-               pud->pud_uid ? pud->pud_uid : rsi->rsi_uid,
-               pud->pud_uid ? pud->pud_gid : rsi->rsi_gid,
-               pud->pud_fsuid ? pud->pud_fsuid : rsi->rsi_uid,
-               pud->pud_fsuid ? pud->pud_fsgid : rsi->rsi_gid,
-               pud->pud_cap & ~CAP_FS_MASK);
-
-        if (rsi->rsi_uid) {
-                if (!pud->pud_uid) {
-                        ucred->mu_uid = rsi->rsi_uid;
-                        ucred->mu_squash |= SQUASH_UID;
-                } else {
-                        ucred->mu_uid = pud->pud_uid;
+        } 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 (!pud->pud_fsuid) {
-                        ucred->mu_fsuid = rsi->rsi_uid;
-                        ucred->mu_squash |= SQUASH_UID;
+        if (is_identity_get_disabled(mdt->mdt_identity_cache)) {
+                if (remote) {
+                        CDEBUG(D_SEC, "remote client must run with identity_get "
+                               "enabled!\n");
+                        RETURN(-EACCES);
                 } else {
-                        ucred->mu_fsuid = pud->pud_fsuid;
+                       ucred->uc_identity = NULL;
+                        perm = CFS_SETUID_PERM | CFS_SETGID_PERM |
+                               CFS_SETGRP_PERM;
                 }
         } else {
-                ucred->mu_uid   = pud->pud_uid;
-                ucred->mu_fsuid = pud->pud_fsuid;
-        }
+                struct md_identity *identity;
 
-        if (rsi->rsi_gid) {
-                int i;
-
-                if (!pud->pud_gid) {
-                        ucred->mu_gid = rsi->rsi_gid;
-                        ucred->mu_squash |= SQUASH_GID;
+                identity = mdt_identity_get(mdt->mdt_identity_cache,
+                                            pud->pud_uid);
+                if (IS_ERR(identity)) {
+                        if (unlikely(PTR_ERR(identity) == -EREMCHG &&
+                                     !remote)) {
+                               ucred->uc_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_gid = pud->pud_gid;
+                       ucred->uc_identity = identity;
+                       perm = mdt_identity_get_perm(ucred->uc_identity,
+                                                    remote, peernid);
                 }
+        }
 
-                if (!pud->pud_fsgid) {
-                        ucred->mu_fsgid = rsi->rsi_gid;
-                        ucred->mu_squash |= SQUASH_GID;
-                } else {
-                        ucred->mu_fsgid = pud->pud_fsgid;
-                }
+        /* find out the setuid/setgid attempt */
+        setuid = (pud->pud_uid != pud->pud_fsuid);
+       setgid = ((pud->pud_gid != pud->pud_fsgid) ||
+                 (ucred->uc_identity &&
+                  (pud->pud_gid != ucred->uc_identity->mi_gid)));
 
-                for (i = 0; i < 2; i++) {
-                        if (!ucred->mu_suppgids[i]) {
-                                ucred->mu_suppgids[i] = rsi->rsi_gid;
-                                ucred->mu_squash |= SQUASH_GID;
-                        }
-                }
+        /* check permission of setuid */
+        if (setuid && !(perm & CFS_SETUID_PERM)) {
+                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);
+        }
 
-                for (i = 0; i < pud->pud_ngroups; i++) {
-                        if (!pud->pud_groups[i]) {
-                                pud->pud_groups[i] = rsi->rsi_gid;
-                                ucred->mu_squash |= SQUASH_GID;
-                        }
-                }
-        } else {
-                ucred->mu_gid   = pud->pud_gid;
-                ucred->mu_fsgid = pud->pud_fsgid;
+        /* check permission of setgid */
+        if (setgid && !(perm & CFS_SETGID_PERM)) {
+               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->uc_identity->mi_gid, libcfs_nid2str(peernid));
+               GOTO(out, rc = -EACCES);
         }
 
-        return 1;
+        /*
+         * NB: remote client not allowed to setgroups anyway.
+         */
+       if (!remote && perm & CFS_SETGRP_PERM) {
+               if (pud->pud_ngroups) {
+                       /* setgroups for local client */
+                       ucred->uc_ginfo = groups_alloc(pud->pud_ngroups);
+                       if (!ucred->uc_ginfo) {
+                               CERROR("failed to alloc %d groups\n",
+                                      pud->pud_ngroups);
+                               GOTO(out, rc = -ENOMEM);
+                       }
+
+                       lustre_groups_from_list(ucred->uc_ginfo,
+                                               pud->pud_groups);
+                       lustre_groups_sort(ucred->uc_ginfo);
+               } else {
+                       ucred->uc_ginfo = NULL;
+               }
+       } else {
+               ucred->uc_suppgids[0] = -1;
+               ucred->uc_suppgids[1] = -1;
+               ucred->uc_ginfo = NULL;
+       }
+
+       ucred->uc_uid   = pud->pud_uid;
+       ucred->uc_gid   = pud->pud_gid;
+       ucred->uc_fsuid = pud->pud_fsuid;
+       ucred->uc_fsgid = pud->pud_fsgid;
+
+       /* process root_squash here. */
+       mdt_root_squash(info, peernid);
+
+       /* remove fs privilege for non-root user. */
+       if (ucred->uc_fsuid)
+               ucred->uc_cap = pud->pud_cap & ~CFS_CAP_FS_MASK;
+       else
+               ucred->uc_cap = pud->pud_cap;
+       if (remote && !(perm & CFS_RMTOWN_PERM))
+               ucred->uc_cap &= ~(CFS_CAP_SYS_RESOURCE_MASK |
+                                  CFS_CAP_CHOWN_MASK);
+       ucred->uc_valid = UCRED_NEW;
+
+       EXIT;
+
+out:
+       if (rc) {
+               if (ucred->uc_ginfo) {
+                       put_group_info(ucred->uc_ginfo);
+                       ucred->uc_ginfo = NULL;
+               }
+               if (ucred->uc_identity) {
+                       mdt_identity_put(mdt->mdt_identity_cache,
+                                        ucred->uc_identity);
+                       ucred->uc_identity = NULL;
+               }
+       }
+
+       return rc;
 }
 
-static int new_init_ucred(struct mdt_thread_info *info, ucred_init_type_t type,
-                          void *buf)
+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 mdt_identity     *identity = NULL;
-        lnet_nid_t              peernid = req->rq_peer.nid;
-        __u32                   setxid_perm = 0;
-        int                     setuid;
-        int                     setgid;
-        int                     rc = 0;
+       struct lu_ucred         *ucred = mdt_ucred(info);
+        struct md_identity      *identity = NULL;
+        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;
 
         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;
-        ucred->mu_o_gid   = pud->pud_gid;
-        ucred->mu_o_fsuid = pud->pud_fsuid;
-        ucred->mu_o_fsgid = pud->pud_fsgid;
+       LASSERT(ucred != NULL);
+       if ((ucred->uc_valid == UCRED_OLD) || (ucred->uc_valid == UCRED_NEW))
+               RETURN(0);
 
-        if (type == BODY_INIT) {
-                struct mdt_body *body = (struct mdt_body *)buf;
-
-                ucred->mu_suppgids[0] = body->suppgid;
-                ucred->mu_suppgids[1] = -1;
-        }
+        if (!req->rq_auth_gss || req->rq_auth_usr_mdt || !req->rq_user_desc)
+                RETURN(0);
 
-        /* sanity check: 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");
+        /* sanity check: if we use strong authentication, we expect the
+         * uid which client claimed is true */
+        if (remote) {
+               if (!uid_valid(make_kuid(&init_user_ns, req->rq_auth_mapped_uid))) {
+                        CDEBUG(D_SEC, "remote user not mapped, deny access!\n");
                         RETURN(-EACCES);
                 }
 
@@ -324,191 +341,168 @@ 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 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, "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);
+                        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 {
-                        setxid_perm |= LUSTRE_SETGRP_PERM;
-                        goto check_squash;
                 }
+                RETURN(0);
         }
 
         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);
+               }
         }
 
-        setxid_perm = mdt_identity_get_setxid_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 ||
                   pud->pud_gid != identity->mi_gid);
 
         /* check permission of setuid */
-        if (setuid && !(setxid_perm & LUSTRE_SETUID_PERM)) {
-                CWARN("mdt blocked setuid attempt (%u -> %u) from "
-                      LPX64"\n", pud->pud_uid, pud->pud_fsuid, peernid);
+        if (setuid && !(perm & CFS_SETUID_PERM)) {
+                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 && !(setxid_perm & LUSTRE_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);
+        if (setgid && !(perm & CFS_SETGID_PERM)) {
+                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);
         }
 
-check_squash:
-        /* FIXME: The exact behavior of root_squash is not defined. */
-        ucred->mu_squash = SQUASH_NONE;
-        if (mdt_squash_root(mdt, ucred, pud, peernid) == 0) {
-                ucred->mu_uid   = pud->pud_uid;
-                ucred->mu_gid   = pud->pud_gid;
-                ucred->mu_fsuid = pud->pud_fsuid;
-                ucred->mu_fsgid = pud->pud_fsgid;
-        }
-
-        /* remove fs privilege for non-root user */
-        if (ucred->mu_fsuid)
-                ucred->mu_cap = pud->pud_cap & ~CAP_FS_MASK;
-        else
-                ucred->mu_cap = pud->pud_cap;
-
-        /*
-         * NB: remote client not allowed to setgroups anyway.
-         */
-        if (!med->med_rmtclient && pud->pud_ngroups &&
-            (setxid_perm & LUSTRE_SETGRP_PERM)) {
-                struct group_info *ginfo;
-
-                /* setgroups for local client */
-                ginfo = groups_alloc(pud->pud_ngroups);
-                if (!ginfo) {
-                        CERROR("failed to alloc %d groups\n",
-                               pud->pud_ngroups);
-                        GOTO(out, rc = -ENOMEM);
-                }
-                groups_from_list(ginfo, pud->pud_groups);
-                groups_sort(ginfo);
-                ucred->mu_ginfo = ginfo;
-        } else {
-                ucred->mu_ginfo = NULL;
-        }
-
-        ucred->mu_identity = identity;
-        ucred->mu_valid = UCRED_NEW;
-
         EXIT;
 
 out:
-        if (rc && identity)
-                mdt_identity_put(mdt->mdt_identity_cache, identity);
-
+        mdt_identity_put(mdt->mdt_identity_cache, identity);
         return rc;
 }
 
-int mdt_check_ucred(struct mdt_thread_info *info)
+static int old_init_ucred(struct mdt_thread_info *info,
+                         struct mdt_body *body)
 {
-        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 mdt_identity     *identity;
-        lnet_nid_t              peernid = req->rq_peer.nid;
-
-        ENTRY;
-
-        if ((ucred->mu_valid == UCRED_OLD) || (ucred->mu_valid == UCRED_NEW))
-                RETURN(0);
-
-        if (!req->rq_user_desc)
-                RETURN(0);
-
-        /* sanity check: if we use strong authentication, we expect the
-         * uid which client claimed is true */
-        if (req->rq_auth_gss) {
-                if (med->med_rmtclient) {
-                        if (req->rq_auth_mapped_uid == INVALID_UID) {
-                                CWARN("remote user not mapped, deny access!\n");
-                                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 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);
-                        }
-                } 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);
-                        }
-                }
-        }
-
-        if (is_identity_get_disabled(mdt->mdt_identity_cache)) {
-                if (med->med_rmtclient) {
-                        CERROR("remote client must run with "
-                               "identity_get enabled!\n");
-                        RETURN(-EACCES);
-                }
-        } else {
-                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);
-                }
-
-                mdt_identity_put(mdt->mdt_identity_cache, identity);
-        }
+       struct lu_ucred *uc = mdt_ucred(info);
+       struct mdt_device  *mdt = info->mti_mdt;
+       struct md_identity *identity = NULL;
+
+       ENTRY;
+
+       LASSERT(uc != NULL);
+       uc->uc_valid = UCRED_INVALID;
+       uc->uc_o_uid = uc->uc_uid = body->uid;
+       uc->uc_o_gid = uc->uc_gid = body->gid;
+       uc->uc_o_fsuid = uc->uc_fsuid = body->fsuid;
+       uc->uc_o_fsgid = uc->uc_fsgid = body->fsgid;
+       uc->uc_suppgids[0] = body->suppgid;
+       uc->uc_suppgids[1] = -1;
+       uc->uc_ginfo = NULL;
+       if (!is_identity_get_disabled(mdt->mdt_identity_cache)) {
+               identity = mdt_identity_get(mdt->mdt_identity_cache,
+                                           uc->uc_fsuid);
+               if (IS_ERR(identity)) {
+                       if (unlikely(PTR_ERR(identity) == -EREMCHG)) {
+                               identity = NULL;
+                       } else {
+                               CDEBUG(D_SEC, "Deny access without identity: "
+                                      "uid %u\n", uc->uc_fsuid);
+                               RETURN(-EACCES);
+                       }
+               }
+       }
+       uc->uc_identity = identity;
+
+       /* process root_squash here. */
+       mdt_root_squash(info, mdt_info_req(info)->rq_peer.nid);
+
+       /* remove fs privilege for non-root user. */
+       if (uc->uc_fsuid)
+               uc->uc_cap = body->capability & ~CFS_CAP_FS_MASK;
+       else
+               uc->uc_cap = body->capability;
+       uc->uc_valid = UCRED_OLD;
+
+       RETURN(0);
+}
 
-        RETURN(0);
+static int old_init_ucred_reint(struct mdt_thread_info *info)
+{
+       struct lu_ucred *uc = mdt_ucred(info);
+       struct mdt_device  *mdt = info->mti_mdt;
+       struct md_identity *identity = NULL;
+
+       ENTRY;
+
+       LASSERT(uc != NULL);
+       uc->uc_valid = UCRED_INVALID;
+       uc->uc_o_uid = uc->uc_o_fsuid = uc->uc_uid = uc->uc_fsuid;
+       uc->uc_o_gid = uc->uc_o_fsgid = uc->uc_gid = uc->uc_fsgid;
+       uc->uc_ginfo = NULL;
+       if (!is_identity_get_disabled(mdt->mdt_identity_cache)) {
+               identity = mdt_identity_get(mdt->mdt_identity_cache,
+                                           uc->uc_fsuid);
+               if (IS_ERR(identity)) {
+                       if (unlikely(PTR_ERR(identity) == -EREMCHG)) {
+                               identity = NULL;
+                       } else {
+                               CDEBUG(D_SEC, "Deny access without identity: "
+                                      "uid %u\n", uc->uc_fsuid);
+                               RETURN(-EACCES);
+                       }
+               }
+       }
+       uc->uc_identity = identity;
+
+       /* process root_squash here. */
+       mdt_root_squash(info, mdt_info_req(info)->rq_peer.nid);
+
+       /* remove fs privilege for non-root user. */
+       if (uc->uc_fsuid)
+               uc->uc_cap &= ~CFS_CAP_FS_MASK;
+       uc->uc_valid = UCRED_OLD;
+
+       RETURN(0);
 }
 
 int mdt_init_ucred(struct mdt_thread_info *info, struct mdt_body *body)
 {
         struct ptlrpc_request *req = mdt_info_req(info);
-        struct md_ucred       *uc  = mdt_ucred(info);
+       struct lu_ucred       *uc  = mdt_ucred(info);
 
-        if ((uc->mu_valid == UCRED_OLD) || (uc->mu_valid == UCRED_NEW))
-                return 0;
+       LASSERT(uc != NULL);
+       if ((uc->uc_valid == UCRED_OLD) || (uc->uc_valid == UCRED_NEW))
+               return 0;
 
         mdt_exit_ucred(info);
 
@@ -521,10 +515,11 @@ int mdt_init_ucred(struct mdt_thread_info *info, struct mdt_body *body)
 int mdt_init_ucred_reint(struct mdt_thread_info *info)
 {
         struct ptlrpc_request *req = mdt_info_req(info);
-        struct md_ucred       *uc  = mdt_ucred(info);
+       struct lu_ucred       *uc  = mdt_ucred(info);
 
-        if ((uc->mu_valid == UCRED_OLD) || (uc->mu_valid == UCRED_NEW))
-                return 0;
+       LASSERT(uc != NULL);
+       if ((uc->uc_valid == UCRED_OLD) || (uc->uc_valid == UCRED_NEW))
+               return 0;
 
         mdt_exit_ucred(info);
 
@@ -535,33 +530,72 @@ int mdt_init_ucred_reint(struct mdt_thread_info *info)
 }
 
 /* copied from lov/lov_ea.c, just for debugging, will be removed later */
-void mdt_dump_lmm(int level, const struct lov_mds_md *lmm)
+void mdt_dump_lmm(int level, const struct lov_mds_md *lmm, __u64 valid)
 {
-        const struct lov_ost_data_v1 *lod;
-        int i;
-        __s16 stripe_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++) {
-                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_id));
-        }
+       const struct lov_ost_data_v1    *lod;
+       int                              i;
+       __u16                            count;
+
+       if (likely(!cfs_cdebug_show(level, DEBUG_SUBSYSTEM)))
+               return;
+
+       count = le16_to_cpu(((struct lov_user_md *)lmm)->lmm_stripe_count);
+
+       CDEBUG(level, "objid "DOSTID", magic 0x%08X, pattern %#X\n",
+              POSTID(&lmm->lmm_oi), 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), count);
+
+       /* If it's a directory or a released file, then there are
+        * no actual objects to print, so bail out. */
+       if (valid & OBD_MD_FLDIREA ||
+           le32_to_cpu(lmm->lmm_pattern) & LOV_PATTERN_F_RELEASED)
+               return;
+
+       LASSERT(count <= LOV_MAX_STRIPE_COUNT);
+       for (i = 0, lod = lmm->lmm_objects; i < count; i++, lod++) {
+               struct ost_id oi;
+
+               ostid_le_to_cpu(&lod->l_ost_oi, &oi);
+               CDEBUG(level, "stripe %u idx %u subobj "DOSTID"\n",
+                      i, le32_to_cpu(lod->l_ost_idx), POSTID(&oi));
+       }
 }
 
-void mdt_shrink_reply(struct mdt_thread_info *info)
+void mdt_dump_lmv(unsigned int level, const union lmv_mds_md *lmv)
 {
-        struct req_capsule    *pill = &info->mti_pill;
-        struct mdt_body       *body;
-        int acl_size, md_size, adjust = 0;
+       const struct lmv_mds_md_v1 *lmm1;
+       int                        i;
+
+       if (likely(!cfs_cdebug_show(level, DEBUG_SUBSYSTEM)))
+               return;
+
+       lmm1 = &lmv->lmv_md_v1;
+       CDEBUG(level, "magic 0x%08X, master %#X stripe_count %#x\n",
+              le32_to_cpu(lmm1->lmv_magic),
+              le32_to_cpu(lmm1->lmv_master_mdt_index),
+              le32_to_cpu(lmm1->lmv_stripe_count));
+
+       if (le32_to_cpu(lmm1->lmv_magic) == LMV_MAGIC_STRIPE)
+               return;
+
+       for (i = 0; i < le32_to_cpu(lmm1->lmv_stripe_count); i++) {
+               struct lu_fid fid;
+
+               fid_le_to_cpu(&fid, &lmm1->lmv_stripe_fids[i]);
+               CDEBUG(level, "idx %u subobj "DFID"\n", i, PFID(&fid));
+       }
+}
+
+/* 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                md_size, md_packed = 0;
+        int                acl_size;
+        int                rc = 0;
         ENTRY;
 
         body = req_capsule_server_get(pill, &RMF_MDT_BODY);
@@ -574,11 +608,17 @@ 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) {
+               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,
@@ -587,39 +627,87 @@ 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) {
+               /* big_lmm buffer may be used even without packing the result
+                * into reply, just for internal server needs */
+               if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER))
+                       md_packed = req_capsule_get_size(pill, &RMF_MDT_MD,
+                                                        RCL_SERVER);
+
+               /* free big lmm if md_size is not needed */
+               if (md_size == 0 || md_packed == 0) {
+                       info->mti_big_lmm_used = 0;
+               } else {
+                       /* 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);
+               }
+        } 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);
 
-        /* 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_CAPA1, RCL_SERVER) &&
+            !(body->valid & OBD_MD_FLMDSCAPA))
+                req_capsule_shrink(pill, &RMF_CAPA1, 0, RCL_SERVER);
+
+        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/LMV EA */
+                       lmm = req_capsule_server_get(pill, &RMF_MDT_MD);
+                       if (info->mti_attr.ma_valid & MA_LOV) {
+                               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);
+                       } else if (info->mti_attr.ma_valid & MA_LMV) {
+                               LASSERT(req_capsule_get_size(pill, &RMF_MDT_MD,
+                                                            RCL_SERVER) ==
+                                               info->mti_attr.ma_lmv_size);
+                               memcpy(lmm, info->mti_attr.ma_lmv,
+                                      info->mti_attr.ma_lmv_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);
 }
 
 
@@ -630,151 +718,194 @@ 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)
                 mdt_pack_attr2body(info, repbody, la, mdt_object_fid(mo));
 
         if (ma->ma_valid & MA_LOV) {
-                __u32 mode;
-
-                if (mdt_object_exists(mo) < 0)
-                        /* If it is a remote object, and we do not retrieve
-                         * EA back unlink reg file*/
-                        mode = S_IFREG;
-                else
-                        mode = lu_object_attr(&mo->mot_obj.mo_lu);
-
-                LASSERT(ma->ma_lmm_size);
-                mdt_dump_lmm(D_INFO, ma->ma_lmm);
-                repbody->eadatasize = ma->ma_lmm_size;
-                if (S_ISREG(mode))
-                        repbody->valid |= OBD_MD_FLEASIZE;
-                else if (S_ISDIR(mode))
-                        repbody->valid |= OBD_MD_FLDIREA;
-                else
-                        LBUG();
+               CERROR("No need in LOV EA upon unlink\n");
+               dump_stack();
         }
+       repbody->eadatasize = 0;
+
+       if (info->mti_mdt->mdt_lut.lut_oss_capa &&
+           exp_connect_flags(info->mti_exp) & OBD_CONNECT_OSS_CAPA &&
+           repbody->valid & OBD_MD_FLEASIZE) {
+                struct lustre_capa *capa;
 
-        if (ma->ma_cookie_size && (ma->ma_valid & MA_COOKIE)) {
-                repbody->aclsize = ma->ma_cookie_size;
-                repbody->valid |= OBD_MD_FLCOOKIE;
+                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);
 }
 
-static inline unsigned int attr_unpack(__u64 sa_valid) {
-        unsigned int ia_valid = 0;
-
-        if (sa_valid & MDS_ATTR_MODE)
-                ia_valid |= ATTR_MODE;
-        if (sa_valid & MDS_ATTR_UID)
-                ia_valid |= ATTR_UID;
-        if (sa_valid & MDS_ATTR_GID)
-                ia_valid |= ATTR_GID;
-        if (sa_valid & MDS_ATTR_SIZE)
-                ia_valid |= ATTR_SIZE;
-        if (sa_valid & MDS_ATTR_ATIME)
-                ia_valid |= ATTR_ATIME;
-        if (sa_valid & MDS_ATTR_MTIME)
-                ia_valid |= ATTR_MTIME;
-        if (sa_valid & MDS_ATTR_CTIME)
-                ia_valid |= ATTR_CTIME;
-        if (sa_valid & MDS_ATTR_ATIME_SET)
-                ia_valid |= ATTR_ATIME_SET;
-        if (sa_valid & MDS_ATTR_MTIME_SET)
-                ia_valid |= ATTR_MTIME_SET;
-        if (sa_valid & MDS_ATTR_FORCE)
-                ia_valid |= ATTR_FORCE;
-        if (sa_valid & MDS_ATTR_ATTR_FLAG)
-                ia_valid |= ATTR_ATTR_FLAG;
-        if (sa_valid & MDS_ATTR_KILL_SUID)
-                ia_valid |=  ATTR_KILL_SUID;
-        if (sa_valid & MDS_ATTR_KILL_SGID)
-                ia_valid |= ATTR_KILL_SGID;
-        if (sa_valid & MDS_ATTR_CTIME_SET)
-                ia_valid |= ATTR_CTIME_SET;
-        if (sa_valid & MDS_ATTR_FROM_OPEN)
-                ia_valid |= ATTR_FROM_OPEN;
-        if (sa_valid & MDS_ATTR_BLOCKS)
-                ia_valid |= ATTR_BLOCKS;
-        if (sa_valid & MDS_OPEN_OWNEROVERRIDE)
-                ia_valid |= MDS_OPEN_OWNEROVERRIDE;
-        return ia_valid;
-}
-
 static __u64 mdt_attr_valid_xlate(__u64 in, struct mdt_reint_record *rr,
                                   struct md_attr *ma)
 {
-        __u64 out;
-
-        out = 0;
-        if (in & ATTR_MODE)
-                out |= LA_MODE;
-        if (in & ATTR_UID)
-                out |= LA_UID;
-        if (in & ATTR_GID)
-                out |= LA_GID;
-        if (in & ATTR_SIZE)
-                out |= LA_SIZE;
-        if (in & ATTR_BLOCKS)
-                out |= LA_BLOCKS;
-
-        if (in & ATTR_FROM_OPEN)
-                rr->rr_flags |= MRF_SETATTR_LOCKED;
-
-        if (in & ATTR_ATIME_SET)
-                out |= LA_ATIME;
-
-        if (in & ATTR_CTIME_SET)
-                out |= LA_CTIME;
-
-        if (in & ATTR_MTIME_SET)
-                out |= LA_MTIME;
-
-        if (in & ATTR_ATTR_FLAG)
-                out |= LA_FLAGS;
-
-        if (in & MDS_OPEN_OWNEROVERRIDE)
-                out |= MDS_OPEN_OWNEROVERRIDE;
-
-        /*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);
-        if (in != 0)
-                CERROR("Unknown attr bits: %#llx\n", in);
-        return out;
+       __u64 out;
+
+       out = 0;
+       if (in & MDS_ATTR_MODE)
+               out |= LA_MODE;
+       if (in & MDS_ATTR_UID)
+               out |= LA_UID;
+       if (in & MDS_ATTR_GID)
+               out |= LA_GID;
+       if (in & MDS_ATTR_SIZE)
+               out |= LA_SIZE;
+       if (in & MDS_ATTR_BLOCKS)
+               out |= LA_BLOCKS;
+       if (in & MDS_ATTR_ATIME_SET)
+               out |= LA_ATIME;
+       if (in & MDS_ATTR_CTIME_SET)
+               out |= LA_CTIME;
+       if (in & MDS_ATTR_MTIME_SET)
+               out |= LA_MTIME;
+       if (in & MDS_ATTR_ATTR_FLAG)
+               out |= LA_FLAGS;
+       if (in & MDS_ATTR_KILL_SUID)
+               out |= LA_KILL_SUID;
+       if (in & MDS_ATTR_KILL_SGID)
+               out |= LA_KILL_SGID;
+
+       if (in & MDS_ATTR_FROM_OPEN)
+               rr->rr_flags |= MRF_OPEN_TRUNC;
+       if (in & MDS_OPEN_OWNEROVERRIDE)
+               ma->ma_attr_flags |= MDS_OWNEROVERRIDE;
+       if (in & MDS_ATTR_FORCE)
+               ma->ma_attr_flags |= MDS_PERM_BYPASS;
+
+       in &= ~(MDS_ATTR_MODE | MDS_ATTR_UID | MDS_ATTR_GID |
+               MDS_ATTR_ATIME | MDS_ATTR_MTIME | MDS_ATTR_CTIME |
+               MDS_ATTR_ATIME_SET | MDS_ATTR_CTIME_SET | MDS_ATTR_MTIME_SET |
+               MDS_ATTR_SIZE | MDS_ATTR_BLOCKS | MDS_ATTR_ATTR_FLAG |
+               MDS_ATTR_FORCE | MDS_ATTR_KILL_SUID | MDS_ATTR_KILL_SGID |
+               MDS_ATTR_FROM_OPEN | MDS_OPEN_OWNEROVERRIDE);
+       if (in != 0)
+               CERROR("Unknown attr bits: "LPX64"\n", in);
+       return out;
 }
+
+void mdt_set_capainfo(struct mdt_thread_info *info, int offset,
+                     const struct lu_fid *fid, struct lustre_capa *capa)
+{
+       struct lu_capainfo *lci;
+
+       LASSERT(offset >= 0 && offset < LU_CAPAINFO_MAX);
+       if (!info->mti_mdt->mdt_lut.lut_mds_capa ||
+           !(exp_connect_flags(info->mti_exp) & OBD_CONNECT_MDS_CAPA))
+               return;
+
+       lci = lu_capainfo_get(info->mti_env);
+       LASSERT(lci);
+       lci->lci_fid[offset]  = *fid;
+       lci->lci_capa[offset] = capa;
+}
+
+#ifdef DEBUG_CAPA
+void mdt_dump_capainfo(struct mdt_thread_info *info)
+{
+       struct lu_capainfo *lci = lu_capainfo_get(info->mti_env);
+       int i;
+
+       if (lci == NULL)
+               return;
+
+       for (i = 0; i < LU_CAPAINFO_MAX; i++) {
+               if (lci->lci_capa[i] == NULL) {
+                       CERROR("no capa for index %d "DFID"\n",
+                              i, PFID(&lci->lci_fid[i]));
+                       continue;
+               }
+               if (lci->lci_capa[i] == BYPASS_CAPA) {
+                       CERROR("bypass for index %d "DFID"\n",
+                              i, PFID(&lci->lci_fid[i]));
+                       continue;
+               }
+               DEBUG_CAPA(D_ERROR, lci->lci_capa[i], "index %d", i);
+       }
+}
+#endif /* DEBUG_CAPA */
+
 /* unpacking */
 
+int mdt_name_unpack(struct req_capsule *pill,
+                   const struct req_msg_field *field,
+                   struct lu_name *ln,
+                   enum mdt_name_flags flags)
+{
+       ln->ln_name = req_capsule_client_get(pill, field);
+       ln->ln_namelen = req_capsule_get_size(pill, field, RCL_CLIENT) - 1;
+
+       if (!lu_name_is_valid(ln)) {
+               ln->ln_name = NULL;
+               ln->ln_namelen = 0;
+
+               return -EPROTO;
+       }
+
+       if ((flags & MNF_FIX_ANON) &&
+           ln->ln_namelen == 1 && ln->ln_name[0] == '/') {
+               /* Newer (3.x) kernels use a name of "/" for the
+                * "anonymous" disconnected dentries from NFS
+                * filehandle conversion. See d_obtain_alias(). */
+               ln->ln_name = NULL;
+               ln->ln_namelen = 0;
+       }
+
+       return 0;
+}
+
 static int mdt_setattr_unpack_rec(struct mdt_thread_info *info)
 {
-        struct md_ucred         *uc  = mdt_ucred(info);
+       struct lu_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);
 
-        uc->mu_fsuid = rec->sa_fsuid;
-        uc->mu_fsgid = rec->sa_fsgid;
-        uc->mu_cap   = rec->sa_cap;
-        uc->mu_suppgids[0] = rec->sa_suppgid;
-        uc->mu_suppgids[1] = -1;
+       /* This prior initialization is needed for old_init_ucred_reint() */
+       uc->uc_fsuid = rec->sa_fsuid;
+       uc->uc_fsgid = rec->sa_fsgid;
+       uc->uc_cap   = rec->sa_cap;
+       uc->uc_suppgids[0] = rec->sa_suppgid;
+       uc->uc_suppgids[1] = -1;
 
         rr->rr_fid1 = &rec->sa_fid;
-        la->la_valid = mdt_attr_valid_xlate(attr_unpack(rec->sa_valid), rr, ma);
+       la->la_valid = mdt_attr_valid_xlate(rec->sa_valid, rr, ma);
+       /*  If MDS_ATTR_xTIME is set without MDS_ATTR_xTIME_SET and
+        *  the client does not have OBD_CONNECT_FULL20, convert it
+        *  to LA_xTIME. LU-3036 */
+       if (!(exp_connect_flags(info->mti_exp) & OBD_CONNECT_FULL20)) {
+               if (!(rec->sa_valid & MDS_ATTR_ATIME_SET) &&
+                    (rec->sa_valid & MDS_ATTR_ATIME))
+                       la->la_valid |= LA_ATIME;
+               if (!(rec->sa_valid & MDS_ATTR_MTIME_SET) &&
+                    (rec->sa_valid & MDS_ATTR_MTIME))
+                       la->la_valid |= LA_MTIME;
+               if (!(rec->sa_valid & MDS_ATTR_CTIME_SET) &&
+                    (rec->sa_valid & MDS_ATTR_CTIME))
+                       la->la_valid |= LA_CTIME;
+       }
         la->la_mode  = rec->sa_mode;
         la->la_flags = rec->sa_attr_flags;
         la->la_uid   = rec->sa_uid;
@@ -786,41 +917,53 @@ static int mdt_setattr_unpack_rec(struct mdt_thread_info *info)
         la->la_mtime = rec->sa_mtime;
         ma->ma_valid = MA_INODE;
 
-        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 (rec->sa_bias & MDS_DATA_MODIFIED)
+               ma->ma_attr_flags |= MDS_DATA_MODIFIED;
+       else
+               ma->ma_attr_flags &= ~MDS_DATA_MODIFIED;
 
-        RETURN(0);
+       if (rec->sa_bias & MDS_HSM_RELEASE)
+               ma->ma_attr_flags |= MDS_HSM_RELEASE;
+       else
+               ma->ma_attr_flags &= ~MDS_HSM_RELEASE;
+
+       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));
+
+       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;
 
@@ -829,23 +972,50 @@ static int mdt_setattr_unpack(struct mdt_thread_info *info)
                 RETURN(rc);
 
         /* Epoch may be absent */
-        mdt_epoch_unpack(info);
+        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);
+               if (rr->rr_eadatalen > 0) {
+                       const struct lmv_user_md        *lum;
+
+                       lum = rr->rr_eadata;
+                       /* Sigh ma_valid(from req) does not indicate whether
+                        * it will set LOV/LMV EA, so we have to check magic */
+                       if (le32_to_cpu(lum->lum_magic) == LMV_USER_MAGIC) {
+                               ma->ma_valid |= MA_LMV;
+                               ma->ma_lmv = (void *)rr->rr_eadata;
+                               ma->ma_lmv_size = rr->rr_eadatalen;
+                       } else {
+                               ma->ma_valid |= MA_LOV;
+                               ma->ma_lmm = (void *)rr->rr_eadata;
+                               ma->ma_lmm_size = rr->rr_eadatalen;
+                       }
+               }
+       }
+
+       rc = mdt_dlmreq_unpack(info);
+       RETURN(rc);
+}
 
-        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;
-        }
+static int mdt_hsm_release_unpack(struct mdt_thread_info *info)
+{
+       struct md_attr          *ma = &info->mti_attr;
+       struct req_capsule      *pill = info->mti_pill;
+       ENTRY;
 
-        ma->ma_cookie_size = req_capsule_get_size(pill, &RMF_LOGCOOKIES,
-                                                  RCL_CLIENT);
-        if (ma->ma_cookie_size) {
-                ma->ma_cookie = req_capsule_client_get(pill, &RMF_LOGCOOKIES);
-                ma->ma_valid |= MA_COOKIE;
-        }
+       if (!(ma->ma_attr_flags & MDS_HSM_RELEASE))
+               RETURN(0);
 
-        rc = mdt_dlmreq_unpack(info);
-        RETURN(rc);
+       req_capsule_extend(pill, &RQF_MDS_RELEASE_CLOSE);
+
+       if (!(req_capsule_has_field(pill, &RMF_CLOSE_DATA, RCL_CLIENT) &&
+           req_capsule_field_present(pill, &RMF_CLOSE_DATA, RCL_CLIENT)))
+               RETURN(-EFAULT);
+
+       RETURN(0);
 }
 
 int mdt_close_unpack(struct mdt_thread_info *info)
@@ -853,33 +1023,44 @@ 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);
 
-        RETURN(mdt_setattr_unpack_rec(info));
+       rc = mdt_setattr_unpack_rec(info);
+       if (rc)
+               RETURN(rc);
+
+       rc = mdt_hsm_release_unpack(info);
+       if (rc)
+               RETURN(rc);
+
+       RETURN(mdt_init_ucred_reint(info));
 }
 
 static int mdt_create_unpack(struct mdt_thread_info *info)
 {
-        struct md_ucred         *uc  = mdt_ucred(info);
+       struct lu_ucred         *uc  = mdt_ucred(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);
 
-        uc->mu_fsuid = rec->cr_fsuid;
-        uc->mu_fsgid = rec->cr_fsgid;
-        uc->mu_cap   = rec->cr_cap;
-        uc->mu_suppgids[0] = rec->cr_suppgid1;
-        uc->mu_suppgids[1] = -1;
+       /* This prior initialization is needed for old_init_ucred_reint() */
+       uc->uc_fsuid = rec->cr_fsuid;
+       uc->uc_fsgid = rec->cr_fsgid;
+       uc->uc_cap   = rec->cr_cap;
+       uc->uc_suppgids[0] = rec->cr_suppgid1;
+       uc->uc_suppgids[1] = -1;
+       uc->uc_umask = rec->cr_umask;
 
         rr->rr_fid1 = &rec->cr_fid1;
         rr->rr_fid2 = &rec->cr_fid2;
@@ -890,56 +1071,21 @@ static int mdt_create_unpack(struct mdt_thread_info *info)
         attr->la_ctime = rec->cr_time;
         attr->la_mtime = rec->cr_time;
         attr->la_atime = rec->cr_time;
-        attr->la_valid = LA_MODE | LA_RDEV | LA_UID | LA_GID |
-                         LA_CTIME | LA_MTIME | LA_ATIME;
+       attr->la_valid = LA_MODE | LA_RDEV | LA_UID | LA_GID | LA_TYPE |
+                        LA_CTIME | LA_MTIME | LA_ATIME;
         memset(&sp->u, 0, sizeof(sp->u));
-        sp->sp_cr_flags = rec->cr_flags;
-        sp->sp_ck_split = !!(rec->cr_bias & MDS_CHECK_SPLIT);
-        info->mti_cross_ref = !!(rec->cr_bias & MDS_CROSS_REF);
+        sp->sp_cr_flags = get_mrc_cr_flags(rec);
 
         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));
         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);
-        
-#ifdef CONFIG_FS_POSIX_ACL
-        if (sp->sp_cr_flags & MDS_CREATE_RMT_ACL) {
-                if (S_ISDIR(attr->la_mode))
-                        sp->u.sp_pfid = rr->rr_fid1;
-                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);
-                sp->u.sp_ea.fid = rr->rr_fid1;
-                RETURN(0);
-        }
-#endif
-        if (S_ISDIR(attr->la_mode)) {
-                /* pass parent fid for cross-ref cases */
-                sp->u.sp_pfid = rr->rr_fid1;
-                if (sp->sp_cr_flags & MDS_CREATE_SLAVE_OBJ) {
-                        /* create salve object req, need
-                         * unpack split ea here
-                         */
-                       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);
-                       sp->u.sp_ea.fid = rr->rr_fid1;
-                       RETURN(0);
-                }
-                req_capsule_extend(pill, &RQF_MDS_REINT_CREATE_RMT_ACL);
-        } else if (S_ISLNK(attr->la_mode)) {
+       rc = mdt_name_unpack(pill, &RMF_NAME, &rr->rr_name, 0);
+       if (rc < 0)
+               RETURN(rc);
+
+       if (S_ISLNK(attr->la_mode)) {
                 const char *tgt = NULL;
 
                 req_capsule_extend(pill, &RQF_MDS_REINT_CREATE_SYM);
@@ -951,30 +1097,42 @@ static int mdt_create_unpack(struct mdt_thread_info *info)
                         RETURN(-EFAULT);
         } else {
                 req_capsule_extend(pill, &RQF_MDS_REINT_CREATE_RMT_ACL);
-        }
-        rc = mdt_dlmreq_unpack(info);
-        RETURN(rc);
+               if (S_ISDIR(attr->la_mode) &&
+                   req_capsule_get_size(pill, &RMF_EADATA, RCL_CLIENT) > 0) {
+                       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);
+                       sp->sp_cr_flags |= MDS_OPEN_HAS_EA;
+               }
+       }
+
+       rc = mdt_dlmreq_unpack(info);
+       RETURN(rc);
 }
 
 static int mdt_link_unpack(struct mdt_thread_info *info)
 {
-        struct md_ucred         *uc  = mdt_ucred(info);
+       struct lu_ucred         *uc  = mdt_ucred(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);
 
-        uc->mu_fsuid = rec->lk_fsuid;
-        uc->mu_fsgid = rec->lk_fsgid;
-        uc->mu_cap   = rec->lk_cap;
-        uc->mu_suppgids[0] = rec->lk_suppgid1;
-        uc->mu_suppgids[1] = rec->lk_suppgid2;
+       /* This prior initialization is needed for old_init_ucred_reint() */
+       uc->uc_fsuid = rec->lk_fsuid;
+       uc->uc_fsgid = rec->lk_fsgid;
+       uc->uc_cap   = rec->lk_cap;
+       uc->uc_suppgids[0] = rec->lk_suppgid1;
+       uc->uc_suppgids[1] = rec->lk_suppgid2;
 
         attr->la_uid = rec->lk_fsuid;
         attr->la_gid = rec->lk_fsgid;
@@ -991,38 +1149,37 @@ 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));
 
-        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);
+       rc = mdt_name_unpack(pill, &RMF_NAME, &rr->rr_name, 0);
+       if (rc < 0)
+               RETURN(rc);
 
-        rc = mdt_dlmreq_unpack(info);
-        RETURN(rc);
+       rc = mdt_dlmreq_unpack(info);
+
+       RETURN(rc);
 }
 
 static int mdt_unlink_unpack(struct mdt_thread_info *info)
 {
-        struct md_ucred         *uc  = mdt_ucred(info);
+       struct lu_ucred         *uc  = mdt_ucred(info);
         struct mdt_rec_unlink   *rec;
         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[1] = -1;
+       /* This prior initialization is needed for old_init_ucred_reint() */
+       uc->uc_fsuid = rec->ul_fsuid;
+       uc->uc_fsgid = rec->ul_fsgid;
+       uc->uc_cap   = rec->ul_cap;
+       uc->uc_suppgids[0] = rec->ul_suppgid1;
+       uc->uc_suppgids[1] = -1;
 
         attr->la_uid = rec->ul_fsuid;
         attr->la_gid = rec->ul_fsgid;
@@ -1037,42 +1194,49 @@ 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);
+       rc = mdt_name_unpack(pill, &RMF_NAME, &rr->rr_name, 0);
+       if (rc < 0)
+               RETURN(rc);
+
         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));
+
         rc = mdt_dlmreq_unpack(info);
         RETURN(rc);
 }
 
+static int mdt_rmentry_unpack(struct mdt_thread_info *info)
+{
+       info->mti_spec.sp_rm_entry = 1;
+       return mdt_unlink_unpack(info);
+}
+
 static int mdt_rename_unpack(struct mdt_thread_info *info)
 {
-        struct md_ucred         *uc = mdt_ucred(info);
+       struct lu_ucred         *uc = mdt_ucred(info);
         struct mdt_rec_rename   *rec;
         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);
 
-        uc->mu_fsuid = rec->rn_fsuid;
-        uc->mu_fsgid = rec->rn_fsgid;
-        uc->mu_cap   = rec->rn_cap;
-        uc->mu_suppgids[0] = rec->rn_suppgid1;
-        uc->mu_suppgids[1] = rec->rn_suppgid2;
+       /* This prior initialization is needed for old_init_ucred_reint() */
+       uc->uc_fsuid = rec->rn_fsuid;
+       uc->uc_fsgid = rec->rn_fsgid;
+       uc->uc_cap   = rec->rn_cap;
+       uc->uc_suppgids[0] = rec->rn_suppgid1;
+       uc->uc_suppgids[1] = rec->rn_suppgid2;
 
         attr->la_uid = rec->rn_fsuid;
         attr->la_gid = rec->rn_fsgid;
@@ -1091,45 +1255,74 @@ 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));
 
-        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);
+       rc = mdt_name_unpack(pill, &RMF_NAME, &rr->rr_name, 0);
+       if (rc < 0)
+               RETURN(rc);
+
+       rc = mdt_name_unpack(pill, &RMF_SYMTGT, &rr->rr_tgt_name, 0);
+       if (rc < 0)
+               RETURN(rc);
+
         if (rec->rn_bias & MDS_VTX_BYPASS)
                 ma->ma_attr_flags |= MDS_VTX_BYPASS;
         else
                 ma->ma_attr_flags &= ~MDS_VTX_BYPASS;
 
-        rc = mdt_dlmreq_unpack(info);
-        RETURN(rc);
+        info->mti_spec.no_create = !!req_is_replay(mdt_info_req(info));
+
+
+       rc = mdt_dlmreq_unpack(info);
+
+       RETURN(rc);
+}
+
+/*
+ * please see comment above LOV_MAGIC_V1_DEF
+ */
+static void mdt_fix_lov_magic(struct mdt_thread_info *info)
+{
+       struct mdt_reint_record *rr = &info->mti_rr;
+       struct lov_user_md_v1   *v1;
+
+       v1 = (void *)rr->rr_eadata;
+       LASSERT(v1);
+
+       if (unlikely(req_is_replay(mdt_info_req(info)))) {
+               if (v1->lmm_magic == LOV_USER_MAGIC_V1) {
+                       v1->lmm_magic = LOV_MAGIC_V1_DEF;
+               } else if (v1->lmm_magic == __swab32(LOV_USER_MAGIC_V1)) {
+                       v1->lmm_magic = __swab32(LOV_MAGIC_V1_DEF);
+               } else if (v1->lmm_magic == LOV_USER_MAGIC_V3) {
+                       v1->lmm_magic = LOV_MAGIC_V3_DEF;
+               } else if (v1->lmm_magic == __swab32(LOV_USER_MAGIC_V3)) {
+                       v1->lmm_magic = __swab32(LOV_MAGIC_V3_DEF);
+               }
+       }
 }
 
 static int mdt_open_unpack(struct mdt_thread_info *info)
 {
-        struct md_ucred         *uc = mdt_ucred(info);
+       struct lu_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);
 
-        uc->mu_fsuid = rec->cr_fsuid;
-        uc->mu_fsgid = rec->cr_fsgid;
-        uc->mu_cap   = rec->cr_cap;
-        uc->mu_suppgids[0] = rec->cr_suppgid1;
-        uc->mu_suppgids[1] = rec->cr_suppgid2;
+       /* This prior initialization is needed for old_init_ucred_reint() */
+       uc->uc_fsuid = rec->cr_fsuid;
+       uc->uc_fsgid = rec->cr_fsgid;
+       uc->uc_cap   = rec->cr_cap;
+       uc->uc_suppgids[0] = rec->cr_suppgid1;
+       uc->uc_suppgids[1] = rec->cr_suppgid2;
+       uc->uc_umask = rec->cr_umask;
 
         rr->rr_fid1   = &rec->cr_fid1;
         rr->rr_fid2   = &rec->cr_fid2;
@@ -1144,17 +1337,19 @@ 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);
         info->mti_cross_ref = !!(rec->cr_bias & MDS_CROSS_REF);
 
         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));
@@ -1168,32 +1363,110 @@ static int mdt_open_unpack(struct mdt_thread_info *info)
 #endif
         }
 
-        rr->rr_name = req_capsule_client_get(pill, &RMF_NAME);
-        if (rr->rr_name == NULL)
+       mdt_name_unpack(pill, &RMF_NAME, &rr->rr_name, MNF_FIX_ANON);
+
+        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);
+                       mdt_fix_lov_magic(info);
+                }
+
+                /*
+                 * 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 lu_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;
+       int                      rc;
+       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);
+
+       /* This prior initialization is needed for old_init_ucred_reint() */
+       uc->uc_fsuid  = rec->sx_fsuid;
+       uc->uc_fsgid  = rec->sx_fsgid;
+       uc->uc_cap    = rec->sx_cap;
+       uc->uc_suppgids[0] = rec->sx_suppgid1;
+       uc->uc_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);
+
+       rc = mdt_name_unpack(pill, &RMF_NAME, &rr->rr_name, 0);
+       if (rc < 0)
+               RETURN(rc);
+
+        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);
-        rr->rr_namelen = req_capsule_get_size(pill, &RMF_NAME, RCL_CLIENT) - 1;
-        LASSERT(rr->rr_namelen > 0);
-
-        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 (mdt_dlmreq_unpack(info) < 0)
+               RETURN(-EPROTO);
+
         RETURN(0);
 }
 
+
 typedef int (*reint_unpacker)(struct mdt_thread_info *info);
 
 static reint_unpacker mdt_reint_unpackers[REINT_MAX] = {
-        [REINT_SETATTR]  = mdt_setattr_unpack,
-        [REINT_CREATE]   = mdt_create_unpack,
-        [REINT_LINK]     = mdt_link_unpack,
-        [REINT_UNLINK]   = mdt_unlink_unpack,
-        [REINT_RENAME]   = mdt_rename_unpack,
-        [REINT_OPEN]     = mdt_open_unpack
+       [REINT_SETATTR]  = mdt_setattr_unpack,
+       [REINT_CREATE]   = mdt_create_unpack,
+       [REINT_LINK]     = mdt_link_unpack,
+       [REINT_UNLINK]   = mdt_unlink_unpack,
+       [REINT_RENAME]   = mdt_rename_unpack,
+       [REINT_OPEN]     = mdt_open_unpack,
+       [REINT_SETXATTR] = mdt_setxattr_unpack,
+       [REINT_RMENTRY]  = mdt_rmentry_unpack,
+       [REINT_MIGRATE]  = mdt_rename_unpack,
 };
 
 int mdt_reint_unpack(struct mdt_thread_info *info, __u32 op)