LU-4008 mdt: update mdt_getattr comments about EA sizing

[fs/lustre-release.git] / lustre / mdt / mdt_handler.c

/*
 * GPL HEADER START
 *
 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
 *
 * This program is free software; you can redistribute it and/or modify
 * it under the terms of the GNU General Public License version 2 only,
 * as published by the Free Software Foundation.
 *
 * 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).
 *
 * 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
 *
 * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
 * CA 95054 USA or visit www.sun.com if you need additional information or
 * have any questions.
 *
 * GPL HEADER END
 */
/*
 * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
 * Use is subject to license terms.
 *
 * Copyright (c) 2010, 2013, Intel Corporation.
 */
/*
 * This file is part of Lustre, http://www.lustre.org/
 * Lustre is a trademark of Sun Microsystems, Inc.
 *
 * lustre/mdt/mdt_handler.c
 *
 * Lustre Metadata Target (mdt) request handler
 *
 * 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: Yury Umanets <umka@clusterfs.com>
 */

#define DEBUG_SUBSYSTEM S_MDS

#include <linux/module.h>
/*
 * struct OBD_{ALLOC,FREE}*()
 */
#include <obd_support.h>
/* struct ptlrpc_request */
#include <lustre_net.h>
/* struct obd_export */
#include <lustre_export.h>
/* struct obd_device */
#include <obd.h>
/* lu2dt_dev() */
#include <dt_object.h>
#include <lustre_mds.h>
#include <lustre_log.h>
#include "mdt_internal.h"
#include <lustre_acl.h>
#include <lustre_param.h>
#include <lustre_quota.h>
#include <lustre_lfsck.h>

mdl_mode_t mdt_mdl_lock_modes[] = {
        [LCK_MINMODE] = MDL_MINMODE,
        [LCK_EX]      = MDL_EX,
        [LCK_PW]      = MDL_PW,
        [LCK_PR]      = MDL_PR,
        [LCK_CW]      = MDL_CW,
        [LCK_CR]      = MDL_CR,
        [LCK_NL]      = MDL_NL,
        [LCK_GROUP]   = MDL_GROUP
};

ldlm_mode_t mdt_dlm_lock_modes[] = {
        [MDL_MINMODE] = LCK_MINMODE,
        [MDL_EX]      = LCK_EX,
        [MDL_PW]      = LCK_PW,
        [MDL_PR]      = LCK_PR,
        [MDL_CW]      = LCK_CW,
        [MDL_CR]      = LCK_CR,
        [MDL_NL]      = LCK_NL,
        [MDL_GROUP]   = LCK_GROUP
};

static struct mdt_device *mdt_dev(struct lu_device *d);
static int mdt_unpack_req_pack_rep(struct mdt_thread_info *info, __u32 flags);

static const struct lu_object_operations mdt_obj_ops;

/* Slab for MDT object allocation */
static struct kmem_cache *mdt_object_kmem;

/* For HSM restore handles */
struct kmem_cache *mdt_hsm_cdt_kmem;

/* For HSM request handles */
struct kmem_cache *mdt_hsm_car_kmem;

static struct lu_kmem_descr mdt_caches[] = {
        {
                .ckd_cache = &mdt_object_kmem,
                .ckd_name  = "mdt_obj",
                .ckd_size  = sizeof(struct mdt_object)
        },
        {
                .ckd_cache      = &mdt_hsm_cdt_kmem,
                .ckd_name       = "mdt_cdt_restore_handle",
                .ckd_size       = sizeof(struct cdt_restore_handle)
        },
        {
                .ckd_cache      = &mdt_hsm_car_kmem,
                .ckd_name       = "mdt_cdt_agent_req",
                .ckd_size       = sizeof(struct cdt_agent_req)
        },
        {
                .ckd_cache = NULL
        }
};

int mdt_get_disposition(struct ldlm_reply *rep, int flag)
{
        if (!rep)
                return 0;
        return (rep->lock_policy_res1 & flag);
}

void mdt_clear_disposition(struct mdt_thread_info *info,
                           struct ldlm_reply *rep, int flag)
{
        if (info) {
                info->mti_opdata &= ~flag;
                tgt_opdata_clear(info->mti_env, flag);
        }
        if (rep)
                rep->lock_policy_res1 &= ~flag;
}

void mdt_set_disposition(struct mdt_thread_info *info,
                         struct ldlm_reply *rep, int flag)
{
        if (info) {
                info->mti_opdata |= flag;
                tgt_opdata_set(info->mti_env, flag);
        }
        if (rep)
                rep->lock_policy_res1 |= flag;
}

void mdt_lock_reg_init(struct mdt_lock_handle *lh, ldlm_mode_t lm)
{
        lh->mlh_pdo_hash = 0;
        lh->mlh_reg_mode = lm;
        lh->mlh_rreg_mode = lm;
        lh->mlh_type = MDT_REG_LOCK;
}

void mdt_lock_pdo_init(struct mdt_lock_handle *lh, ldlm_mode_t lock_mode,
                       const struct lu_name *lname)
{
        lh->mlh_reg_mode = lock_mode;
        lh->mlh_rreg_mode = lock_mode;
        lh->mlh_type = MDT_PDO_LOCK;

        if (lu_name_is_valid(lname)) {
                lh->mlh_pdo_hash = full_name_hash(lname->ln_name,
                                                  lname->ln_namelen);
                /* XXX Workaround for LU-2856
                 *
                 * Zero is a valid return value of full_name_hash, but
                 * several users of mlh_pdo_hash assume a non-zero
                 * hash value. We therefore map zero onto an
                 * arbitrary, but consistent value (1) to avoid
                 * problems further down the road. */
                if (unlikely(lh->mlh_pdo_hash == 0))
                        lh->mlh_pdo_hash = 1;
        } else {
                lh->mlh_pdo_hash = 0;
        }
}

static void mdt_lock_pdo_mode(struct mdt_thread_info *info, struct mdt_object *o,
                              struct mdt_lock_handle *lh)
{
        mdl_mode_t mode;
        ENTRY;

        /*
         * Any dir access needs couple of locks:
         *
         * 1) on part of dir we gonna take lookup/modify;
         *
         * 2) on whole dir to protect it from concurrent splitting and/or to
         * flush client's cache for readdir().
         *
         * so, for a given mode and object this routine decides what lock mode
         * to use for lock #2:
         *
         * 1) if caller's gonna lookup in dir then we need to protect dir from
         * being splitted only - LCK_CR
         *
         * 2) if caller's gonna modify dir then we need to protect dir from
         * being splitted and to flush cache - LCK_CW
         *
         * 3) if caller's gonna modify dir and that dir seems ready for
         * splitting then we need to protect it from any type of access
         * (lookup/modify/split) - LCK_EX --bzzz
         */

        LASSERT(lh->mlh_reg_mode != LCK_MINMODE);
        LASSERT(lh->mlh_pdo_mode == LCK_MINMODE);

        /*
         * Ask underlaying level its opinion about preferable PDO lock mode
         * having access type passed as regular lock mode:
         *
         * - MDL_MINMODE means that lower layer does not want to specify lock
         * mode;
         *
         * - MDL_NL means that no PDO lock should be taken. This is used in some
         * cases. Say, for non-splittable directories no need to use PDO locks
         * at all.
         */
        mode = mdo_lock_mode(info->mti_env, mdt_object_child(o),
                             mdt_dlm_mode2mdl_mode(lh->mlh_reg_mode));

        if (mode != MDL_MINMODE) {
                lh->mlh_pdo_mode = mdt_mdl_mode2dlm_mode(mode);
        } else {
                /*
                 * Lower layer does not want to specify locking mode. We do it
                 * our selves. No special protection is needed, just flush
                 * client's cache on modification and allow concurrent
                 * mondification.
                 */
                switch (lh->mlh_reg_mode) {
                case LCK_EX:
                        lh->mlh_pdo_mode = LCK_EX;
                        break;
                case LCK_PR:
                        lh->mlh_pdo_mode = LCK_CR;
                        break;
                case LCK_PW:
                        lh->mlh_pdo_mode = LCK_CW;
                        break;
                default:
                        CERROR("Not expected lock type (0x%x)\n",
                               (int)lh->mlh_reg_mode);
                        LBUG();
                }
        }

        LASSERT(lh->mlh_pdo_mode != LCK_MINMODE);
        EXIT;
}

static int mdt_getstatus(struct tgt_session_info *tsi)
{
        struct mdt_thread_info  *info = tsi2mdt_info(tsi);
        struct mdt_device       *mdt = info->mti_mdt;
        struct mdt_body         *repbody;
        int                      rc;

        ENTRY;

        rc = mdt_check_ucred(info);
        if (rc)
                GOTO(out, rc = err_serious(rc));

        if (OBD_FAIL_CHECK(OBD_FAIL_MDS_GETSTATUS_PACK))
                GOTO(out, rc = err_serious(-ENOMEM));

        repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
        repbody->fid1 = mdt->mdt_md_root_fid;
        repbody->valid |= OBD_MD_FLID;

        if (tsi->tsi_tgt->lut_mds_capa &&
            exp_connect_flags(info->mti_exp) & OBD_CONNECT_MDS_CAPA) {
                struct mdt_object       *root;
                struct lustre_capa      *capa;

                root = mdt_object_find(info->mti_env, mdt, &repbody->fid1);
                if (IS_ERR(root))
                        GOTO(out, rc = PTR_ERR(root));

                capa = req_capsule_server_get(info->mti_pill, &RMF_CAPA1);
                LASSERT(capa);
                capa->lc_opc = CAPA_OPC_MDS_DEFAULT;
                rc = mo_capa_get(info->mti_env, mdt_object_child(root), capa,
                                 0);
                mdt_object_put(info->mti_env, root);
                if (rc == 0)
                        repbody->valid |= OBD_MD_FLMDSCAPA;
        }
        EXIT;
out:
        mdt_thread_info_fini(info);
        return rc;
}

static int mdt_statfs(struct tgt_session_info *tsi)
{
        struct ptlrpc_request           *req = tgt_ses_req(tsi);
        struct mdt_thread_info          *info = tsi2mdt_info(tsi);
        struct md_device                *next = info->mti_mdt->mdt_child;
        struct ptlrpc_service_part      *svcpt;
        struct obd_statfs               *osfs;
        int                             rc;

        ENTRY;

        svcpt = req->rq_rqbd->rqbd_svcpt;

        /* This will trigger a watchdog timeout */
        OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_STATFS_LCW_SLEEP,
                         (MDT_SERVICE_WATCHDOG_FACTOR *
                          at_get(&svcpt->scp_at_estimate)) + 1);

        rc = mdt_check_ucred(info);
        if (rc)
                GOTO(out, rc = err_serious(rc));

        if (OBD_FAIL_CHECK(OBD_FAIL_MDS_STATFS_PACK))
                GOTO(out, rc = err_serious(-ENOMEM));

        osfs = req_capsule_server_get(info->mti_pill, &RMF_OBD_STATFS);
        if (!osfs)
                GOTO(out, rc = -EPROTO);

        /** statfs information are cached in the mdt_device */
        if (cfs_time_before_64(info->mti_mdt->mdt_osfs_age,
                               cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS))) {
                /** statfs data is too old, get up-to-date one */
                rc = next->md_ops->mdo_statfs(info->mti_env, next, osfs);
                if (rc)
                        GOTO(out, rc);
                spin_lock(&info->mti_mdt->mdt_osfs_lock);
                info->mti_mdt->mdt_osfs = *osfs;
                info->mti_mdt->mdt_osfs_age = cfs_time_current_64();
                spin_unlock(&info->mti_mdt->mdt_osfs_lock);
        } else {
                /** use cached statfs data */
                spin_lock(&info->mti_mdt->mdt_osfs_lock);
                *osfs = info->mti_mdt->mdt_osfs;
                spin_unlock(&info->mti_mdt->mdt_osfs_lock);
        }

        if (rc == 0)
                mdt_counter_incr(req, LPROC_MDT_STATFS);
out:
        mdt_thread_info_fini(info);
        RETURN(rc);
}

/**
 * Pack SOM attributes into the reply.
 * Call under a DLM UPDATE lock.
 */
static void mdt_pack_size2body(struct mdt_thread_info *info,
                               struct mdt_object *mo)
{
        struct mdt_body *b;
        struct md_attr *ma = &info->mti_attr;

        LASSERT(ma->ma_attr.la_valid & LA_MODE);
        b = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);

        /* Check if Size-on-MDS is supported, if this is a regular file,
         * if SOM is enabled on the object and if SOM cache exists and valid.
         * Otherwise do not pack Size-on-MDS attributes to the reply. */
        if (!(mdt_conn_flags(info) & OBD_CONNECT_SOM) ||
            !S_ISREG(ma->ma_attr.la_mode) ||
            !mdt_object_is_som_enabled(mo) ||
            !(ma->ma_valid & MA_SOM))
                return;

        b->valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS;
        b->size = ma->ma_som->msd_size;
        b->blocks = ma->ma_som->msd_blocks;
}

void mdt_pack_attr2body(struct mdt_thread_info *info, struct mdt_body *b,
                        const struct lu_attr *attr, const struct lu_fid *fid)
{
        struct md_attr *ma = &info->mti_attr;

        LASSERT(ma->ma_valid & MA_INODE);

        b->atime      = attr->la_atime;
        b->mtime      = attr->la_mtime;
        b->ctime      = attr->la_ctime;
        b->mode       = attr->la_mode;
        b->size       = attr->la_size;
        b->blocks     = attr->la_blocks;
        b->uid        = attr->la_uid;
        b->gid        = attr->la_gid;
        b->flags      = attr->la_flags;
        b->nlink      = attr->la_nlink;
        b->rdev       = attr->la_rdev;

        /*XXX should pack the reply body according to lu_valid*/
        b->valid |= OBD_MD_FLCTIME | OBD_MD_FLUID   |
                    OBD_MD_FLGID   | OBD_MD_FLTYPE  |
                    OBD_MD_FLMODE  | OBD_MD_FLNLINK | OBD_MD_FLFLAGS |
                    OBD_MD_FLATIME | OBD_MD_FLMTIME ;

        if (!S_ISREG(attr->la_mode)) {
                b->valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS | OBD_MD_FLRDEV;
        } else if (ma->ma_need & MA_LOV && !(ma->ma_valid & MA_LOV)) {
                /* means no objects are allocated on osts. */
                LASSERT(!(ma->ma_valid & MA_LOV));
                /* just ignore blocks occupied by extend attributes on MDS */
                b->blocks = 0;
                /* if no object is allocated on osts, the size on mds is valid. b=22272 */
                b->valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS;
        } else if ((ma->ma_valid & MA_LOV) && ma->ma_lmm != NULL &&
                   ma->ma_lmm->lmm_pattern & LOV_PATTERN_F_RELEASED) {
                /* A released file stores its size on MDS. */
                /* But return 1 block for released file, unless tools like tar
                 * will consider it fully sparse. (LU-3864)
                 */
                if (unlikely(b->size == 0))
                        b->blocks = 0;
                else
                        b->blocks = 1;
                b->valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS;
        }

        if (fid) {
                b->fid1 = *fid;
                b->valid |= OBD_MD_FLID;
                CDEBUG(D_INODE, DFID": nlink=%d, mode=%o, size="LPU64"\n",
                                PFID(fid), b->nlink, b->mode, b->size);
        }

        if (info)
                mdt_body_reverse_idmap(info, b);

        if (fid != NULL && (b->valid & OBD_MD_FLSIZE))
                CDEBUG(D_VFSTRACE, DFID": returning size %llu\n",
                       PFID(fid), (unsigned long long)b->size);
}

static inline int mdt_body_has_lov(const struct lu_attr *la,
                                   const struct mdt_body *body)
{
        return ((S_ISREG(la->la_mode) && (body->valid & OBD_MD_FLEASIZE)) ||
                (S_ISDIR(la->la_mode) && (body->valid & OBD_MD_FLDIREA )) );
}

void mdt_client_compatibility(struct mdt_thread_info *info)
{
        struct mdt_body       *body;
        struct ptlrpc_request *req = mdt_info_req(info);
        struct obd_export     *exp = req->rq_export;
        struct md_attr        *ma = &info->mti_attr;
        struct lu_attr        *la = &ma->ma_attr;
        ENTRY;

        if (exp_connect_layout(exp))
                /* the client can deal with 16-bit lmm_stripe_count */
                RETURN_EXIT;

        body = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);

        if (!mdt_body_has_lov(la, body))
                RETURN_EXIT;

        /* now we have a reply with a lov for a client not compatible with the
         * layout lock so we have to clean the layout generation number */
        if (S_ISREG(la->la_mode))
                ma->ma_lmm->lmm_layout_gen = 0;
        EXIT;
}

int mdt_attr_get_eabuf_size(struct mdt_thread_info *info, struct mdt_object *o)
{
        const struct lu_env *env = info->mti_env;
        int rc, rc2;

        rc = mo_xattr_get(env, mdt_object_child(o), &LU_BUF_NULL,
                          XATTR_NAME_LOV);

        if (rc == -ENODATA)
                rc = 0;

        if (rc < 0)
                goto out;

        /* Is it a directory? Let's check for the LMV as well */
        if (S_ISDIR(lu_object_attr(&mdt_object_child(o)->mo_lu))) {
                rc2 = mo_xattr_get(env, mdt_object_child(o), &LU_BUF_NULL,
                                   XATTR_NAME_LMV);
                if ((rc2 < 0 && rc2 != -ENODATA) || (rc2 > rc))
                        rc = rc2;
        }

out:
        return rc;
}

static int mdt_big_xattr_get(struct mdt_thread_info *info, struct mdt_object *o,
                             const char *name)
{
        const struct lu_env *env = info->mti_env;
        int rc;
        ENTRY;

        LASSERT(info->mti_big_lmm_used == 0);
        rc = mo_xattr_get(env, mdt_object_child(o), &LU_BUF_NULL, name);
        if (rc < 0)
                RETURN(rc);

        /* big_lmm may need to be grown */
        if (info->mti_big_lmmsize < rc) {
                int size = size_roundup_power2(rc);

                if (info->mti_big_lmmsize > 0) {
                        /* free old buffer */
                        LASSERT(info->mti_big_lmm);
                        OBD_FREE_LARGE(info->mti_big_lmm,
                                       info->mti_big_lmmsize);
                        info->mti_big_lmm = NULL;
                        info->mti_big_lmmsize = 0;
                }

                OBD_ALLOC_LARGE(info->mti_big_lmm, size);
                if (info->mti_big_lmm == NULL)
                        RETURN(-ENOMEM);
                info->mti_big_lmmsize = size;
        }
        LASSERT(info->mti_big_lmmsize >= rc);

        info->mti_buf.lb_buf = info->mti_big_lmm;
        info->mti_buf.lb_len = info->mti_big_lmmsize;
        rc = mo_xattr_get(env, mdt_object_child(o), &info->mti_buf, name);

        RETURN(rc);
}

int mdt_stripe_get(struct mdt_thread_info *info, struct mdt_object *o,
                   struct md_attr *ma, const char *name)
{
        struct md_object *next = mdt_object_child(o);
        struct lu_buf    *buf = &info->mti_buf;
        int rc;

        if (strcmp(name, XATTR_NAME_LOV) == 0) {
                buf->lb_buf = ma->ma_lmm;
                buf->lb_len = ma->ma_lmm_size;
                LASSERT(!(ma->ma_valid & MA_LOV));
        } else if (strcmp(name, XATTR_NAME_LMV) == 0) {
                buf->lb_buf = ma->ma_lmv;
                buf->lb_len = ma->ma_lmv_size;
                LASSERT(!(ma->ma_valid & MA_LMV));
        } else if (strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0) {
                buf->lb_buf = ma->ma_lmv;
                buf->lb_len = ma->ma_lmv_size;
                LASSERT(!(ma->ma_valid & MA_LMV_DEF));
        } else {
                return -EINVAL;
        }

        rc = mo_xattr_get(info->mti_env, next, buf, name);
        if (rc > 0) {
                if (strcmp(name, XATTR_NAME_LOV) == 0) {
                        ma->ma_lmm_size = rc;
                        ma->ma_valid |= MA_LOV;
                } else if (strcmp(name, XATTR_NAME_LMV) == 0) {
                        ma->ma_lmv_size = rc;
                        ma->ma_valid |= MA_LMV;
                } else if (strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0) {
                        ma->ma_lmv_size = rc;
                        ma->ma_valid |= MA_LMV_DEF;
                }

                rc = 0;
        } else if (rc == -ENODATA) {
                /* no LOV EA */
                rc = 0;
        } else if (rc == -ERANGE) {
                /* Default LMV has fixed size, so it must be able to fit
                 * in the original buffer */
                if (strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0)
                        return rc;
                rc = mdt_big_xattr_get(info, o, name);
                if (rc > 0) {
                        info->mti_big_lmm_used = 1;
                        if (!strcmp(name, XATTR_NAME_LOV)) {
                                ma->ma_valid |= MA_LOV;
                                ma->ma_lmm = info->mti_big_lmm;
                                ma->ma_lmm_size = rc;
                        } else if (!strcmp(name, XATTR_NAME_LMV)) {
                                ma->ma_valid |= MA_LMV;
                                ma->ma_lmv = info->mti_big_lmm;
                                ma->ma_lmv_size = rc;
                        } else {
                                return -EINVAL;
                        }

                        /* update mdt_max_mdsize so all clients
                         * will be aware about that */
                        if (info->mti_mdt->mdt_max_mdsize < rc)
                                info->mti_mdt->mdt_max_mdsize = rc;
                        rc = 0;
                }
        }

        return rc;
}

static int mdt_attr_get_pfid(struct mdt_thread_info *info,
                             struct mdt_object *o, struct lu_fid *pfid)
{
        struct lu_buf           *buf = &info->mti_buf;
        struct link_ea_header   *leh;
        struct link_ea_entry    *lee;
        int                      rc;
        ENTRY;

        buf->lb_buf = info->mti_big_lmm;
        buf->lb_len = info->mti_big_lmmsize;
        rc = mo_xattr_get(info->mti_env, mdt_object_child(o),
                          buf, XATTR_NAME_LINK);
        /* ignore errors, MA_PFID won't be set and it is
         * up to the caller to treat this as an error */
        if (rc == -ERANGE || buf->lb_len == 0) {
                rc = mdt_big_xattr_get(info, o, XATTR_NAME_LINK);
                buf->lb_buf = info->mti_big_lmm;
                buf->lb_len = info->mti_big_lmmsize;
        }

        if (rc < 0)
                RETURN(rc);
        if (rc < sizeof(*leh)) {
                CERROR("short LinkEA on "DFID": rc = %d\n",
                       PFID(mdt_object_fid(o)), rc);
                RETURN(-ENODATA);
        }

        leh = (struct link_ea_header *) buf->lb_buf;
        lee = (struct link_ea_entry *)(leh + 1);
        if (leh->leh_magic == __swab32(LINK_EA_MAGIC)) {
                leh->leh_magic = LINK_EA_MAGIC;
                leh->leh_reccount = __swab32(leh->leh_reccount);
                leh->leh_len = __swab64(leh->leh_len);
        }
        if (leh->leh_magic != LINK_EA_MAGIC)
                RETURN(-EINVAL);
        if (leh->leh_reccount == 0)
                RETURN(-ENODATA);

        memcpy(pfid, &lee->lee_parent_fid, sizeof(*pfid));
        fid_be_to_cpu(pfid, pfid);

        RETURN(0);
}

int mdt_attr_get_complex(struct mdt_thread_info *info,
                         struct mdt_object *o, struct md_attr *ma)
{
        const struct lu_env *env = info->mti_env;
        struct md_object    *next = mdt_object_child(o);
        struct lu_buf       *buf = &info->mti_buf;
        u32                  mode = lu_object_attr(&next->mo_lu);
        int                  need = ma->ma_need;
        int                  rc = 0, rc2;
        ENTRY;

        ma->ma_valid = 0;

        if (need & MA_INODE) {
                ma->ma_need = MA_INODE;
                rc = mo_attr_get(env, next, ma);
                if (rc)
                        GOTO(out, rc);
                ma->ma_valid |= MA_INODE;
        }

        if (need & MA_PFID) {
                rc = mdt_attr_get_pfid(info, o, &ma->ma_pfid);
                if (rc == 0)
                        ma->ma_valid |= MA_PFID;
                /* ignore this error, parent fid is not mandatory */
                rc = 0;
        }

        if (need & MA_LOV && (S_ISREG(mode) || S_ISDIR(mode))) {
                rc = mdt_stripe_get(info, o, ma, XATTR_NAME_LOV);
                if (rc)
                        GOTO(out, rc);
        }

        if (need & MA_LMV && S_ISDIR(mode)) {
                rc = mdt_stripe_get(info, o, ma, XATTR_NAME_LMV);
                if (rc != 0)
                        GOTO(out, rc);
        }

        if (need & MA_LMV_DEF && S_ISDIR(mode)) {
                rc = mdt_stripe_get(info, o, ma, XATTR_NAME_DEFAULT_LMV);
                if (rc != 0)
                        GOTO(out, rc);
        }

        if (need & MA_SOM && S_ISREG(mode)) {
                buf->lb_buf = info->mti_xattr_buf;
                buf->lb_len = sizeof(info->mti_xattr_buf);
                CLASSERT(sizeof(struct som_attrs) <=
                         sizeof(info->mti_xattr_buf));
                rc2 = mo_xattr_get(info->mti_env, next, buf, XATTR_NAME_SOM);
                rc2 = lustre_buf2som(info->mti_xattr_buf, rc2, ma->ma_som);
                if (rc2 == 0)
                        ma->ma_valid |= MA_SOM;
                else if (rc2 < 0 && rc2 != -ENODATA)
                        GOTO(out, rc = rc2);
        }

        if (need & MA_HSM && S_ISREG(mode)) {
                buf->lb_buf = info->mti_xattr_buf;
                buf->lb_len = sizeof(info->mti_xattr_buf);
                CLASSERT(sizeof(struct hsm_attrs) <=
                         sizeof(info->mti_xattr_buf));
                rc2 = mo_xattr_get(info->mti_env, next, buf, XATTR_NAME_HSM);
                rc2 = lustre_buf2hsm(info->mti_xattr_buf, rc2, &ma->ma_hsm);
                if (rc2 == 0)
                        ma->ma_valid |= MA_HSM;
                else if (rc2 < 0 && rc2 != -ENODATA)
                        GOTO(out, rc = rc2);
        }

#ifdef CONFIG_FS_POSIX_ACL
        if (need & MA_ACL_DEF && S_ISDIR(mode)) {
                buf->lb_buf = ma->ma_acl;
                buf->lb_len = ma->ma_acl_size;
                rc2 = mo_xattr_get(env, next, buf, XATTR_NAME_ACL_DEFAULT);
                if (rc2 > 0) {
                        ma->ma_acl_size = rc2;
                        ma->ma_valid |= MA_ACL_DEF;
                } else if (rc2 == -ENODATA) {
                        /* no ACLs */
                        ma->ma_acl_size = 0;
                } else
                        GOTO(out, rc = rc2);
        }
#endif
out:
        ma->ma_need = need;
        CDEBUG(D_INODE, "after getattr rc = %d, ma_valid = "LPX64" ma_lmm=%p\n",
               rc, ma->ma_valid, ma->ma_lmm);
        RETURN(rc);
}

static int mdt_getattr_internal(struct mdt_thread_info *info,
                                struct mdt_object *o, int ma_need)
{
        struct md_object        *next = mdt_object_child(o);
        const struct mdt_body   *reqbody = info->mti_body;
        struct ptlrpc_request   *req = mdt_info_req(info);
        struct md_attr          *ma = &info->mti_attr;
        struct lu_attr          *la = &ma->ma_attr;
        struct req_capsule      *pill = info->mti_pill;
        const struct lu_env     *env = info->mti_env;
        struct mdt_body         *repbody;
        struct lu_buf           *buffer = &info->mti_buf;
        int                     rc;
        int                     is_root;
        ENTRY;

        if (OBD_FAIL_CHECK(OBD_FAIL_MDS_GETATTR_PACK))
                RETURN(err_serious(-ENOMEM));

        repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);

        ma->ma_valid = 0;

        if (mdt_object_remote(o)) {
                /* This object is located on remote node.*/
                /* Return -EIO for old client */
                if (!mdt_is_dne_client(req->rq_export))
                        GOTO(out, rc = -EIO);

                repbody->fid1 = *mdt_object_fid(o);
                repbody->valid = OBD_MD_FLID | OBD_MD_MDS;
                GOTO(out, rc = 0);
        }

        if (reqbody->eadatasize > 0) {
                buffer->lb_buf = req_capsule_server_get(pill, &RMF_MDT_MD);
                if (buffer->lb_buf == NULL)
                        GOTO(out, rc = -EPROTO);
                buffer->lb_len = req_capsule_get_size(pill, &RMF_MDT_MD,
                                                      RCL_SERVER);
        } else {
                buffer->lb_buf = NULL;
                buffer->lb_len = 0;
                ma_need &= ~(MA_LOV | MA_LMV);
                CDEBUG(D_INFO, "%s: RPC from %s: does not need LOVEA.\n",
                       mdt_obd_name(info->mti_mdt),
                       req->rq_export->exp_client_uuid.uuid);
        }

        /* If it is dir object and client require MEA, then we got MEA */
        if (S_ISDIR(lu_object_attr(&next->mo_lu)) &&
            (reqbody->valid & (OBD_MD_MEA | OBD_MD_DEFAULT_MEA))) {
                /* Assumption: MDT_MD size is enough for lmv size. */
                ma->ma_lmv = buffer->lb_buf;
                ma->ma_lmv_size = buffer->lb_len;
                ma->ma_need = MA_INODE;
                if (ma->ma_lmv_size > 0) {
                        if (reqbody->valid & OBD_MD_MEA)
                                ma->ma_need |= MA_LMV;
                        else if (reqbody->valid & OBD_MD_DEFAULT_MEA)
                                ma->ma_need |= MA_LMV_DEF;
                }
        } else {
                ma->ma_lmm = buffer->lb_buf;
                ma->ma_lmm_size = buffer->lb_len;
                ma->ma_need = MA_INODE | MA_HSM;
                if (ma->ma_lmm_size > 0)
                        ma->ma_need |= MA_LOV;
        }

        if (S_ISDIR(lu_object_attr(&next->mo_lu)) &&
            reqbody->valid & OBD_MD_FLDIREA  &&
            lustre_msg_get_opc(req->rq_reqmsg) == MDS_GETATTR) {
                /* get default stripe info for this dir. */
                ma->ma_need |= MA_LOV_DEF;
        }
        ma->ma_need |= ma_need;
        if (ma->ma_need & MA_SOM)
                ma->ma_som = &info->mti_u.som.data;

        rc = mdt_attr_get_complex(info, o, ma);
        if (unlikely(rc)) {
                CERROR("%s: getattr error for "DFID": rc = %d\n",
                       mdt_obd_name(info->mti_mdt),
                       PFID(mdt_object_fid(o)), rc);
                RETURN(rc);
        }

        /* if file is released, check if a restore is running */
        if ((ma->ma_valid & MA_HSM) && (ma->ma_hsm.mh_flags & HS_RELEASED) &&
            mdt_hsm_restore_is_running(info, mdt_object_fid(o))) {
                repbody->t_state = MS_RESTORE;
                repbody->valid |= OBD_MD_TSTATE;
        }

        is_root = lu_fid_eq(mdt_object_fid(o), &info->mti_mdt->mdt_md_root_fid);

        /* the Lustre protocol supposes to return default striping
         * on the user-visible root if explicitly requested */
        if ((ma->ma_valid & MA_LOV) == 0 && S_ISDIR(la->la_mode) &&
            (ma->ma_need & MA_LOV_DEF && is_root) && ma->ma_need & MA_LOV) {
                struct lu_fid      rootfid;
                struct mdt_object *root;
                struct mdt_device *mdt = info->mti_mdt;

                rc = dt_root_get(env, mdt->mdt_bottom, &rootfid);
                if (rc)
                        RETURN(rc);
                root = mdt_object_find(env, mdt, &rootfid);
                if (IS_ERR(root))
                        RETURN(PTR_ERR(root));
                rc = mdt_stripe_get(info, root, ma, XATTR_NAME_LOV);
                mdt_object_put(info->mti_env, root);
                if (unlikely(rc)) {
                        CERROR("%s: getattr error for "DFID": rc = %d\n",
                               mdt_obd_name(info->mti_mdt),
                               PFID(mdt_object_fid(o)), rc);
                        RETURN(rc);
                }
        }

        if (likely(ma->ma_valid & MA_INODE))
                mdt_pack_attr2body(info, repbody, la, mdt_object_fid(o));
        else
                RETURN(-EFAULT);

        if (mdt_body_has_lov(la, reqbody)) {
                if (ma->ma_valid & MA_LOV) {
                        LASSERT(ma->ma_lmm_size);
                        repbody->eadatasize = ma->ma_lmm_size;
                        if (S_ISDIR(la->la_mode))
                                repbody->valid |= OBD_MD_FLDIREA;
                        else
                                repbody->valid |= OBD_MD_FLEASIZE;
                        mdt_dump_lmm(D_INFO, ma->ma_lmm, repbody->valid);
                }
                if (ma->ma_valid & MA_LMV) {
                        LASSERT(S_ISDIR(la->la_mode));
                        mdt_dump_lmv(D_INFO, ma->ma_lmv);
                        repbody->eadatasize = ma->ma_lmv_size;
                        repbody->valid |= (OBD_MD_FLDIREA|OBD_MD_MEA);
                }
                if (ma->ma_valid & MA_LMV_DEF) {
                        LASSERT(S_ISDIR(la->la_mode));
                        repbody->eadatasize = ma->ma_lmv_size;
                        repbody->valid |= (OBD_MD_FLDIREA|OBD_MD_DEFAULT_MEA);
                }
        } else if (S_ISLNK(la->la_mode) &&
                   reqbody->valid & OBD_MD_LINKNAME) {
                buffer->lb_buf = ma->ma_lmm;
                /* eadatasize from client includes NULL-terminator, so
                 * there is no need to read it */
                buffer->lb_len = reqbody->eadatasize - 1;
                rc = mo_readlink(env, next, buffer);
                if (unlikely(rc <= 0)) {
                        CERROR("%s: readlink failed for "DFID": rc = %d\n",
                               mdt_obd_name(info->mti_mdt),
                               PFID(mdt_object_fid(o)), rc);
                        rc = -EFAULT;
                } else {
                        int print_limit = min_t(int, PAGE_CACHE_SIZE - 128, rc);

                        if (OBD_FAIL_CHECK(OBD_FAIL_MDS_READLINK_EPROTO))
                                rc -= 2;
                        repbody->valid |= OBD_MD_LINKNAME;
                        /* we need to report back size with NULL-terminator
                         * because client expects that */
                        repbody->eadatasize = rc + 1;
                        if (repbody->eadatasize != reqbody->eadatasize)
                                CDEBUG(D_INODE, "%s: Read shorter symlink %d "
                                       "on "DFID ", expected %d\n",
                                       mdt_obd_name(info->mti_mdt),
                                       rc, PFID(mdt_object_fid(o)),
                                       reqbody->eadatasize - 1);
                        /* NULL terminate */
                        ((char *)ma->ma_lmm)[rc] = 0;

                        /* If the total CDEBUG() size is larger than a page, it
                         * will print a warning to the console, avoid this by
                         * printing just the last part of the symlink. */
                        CDEBUG(D_INODE, "symlink dest %s%.*s, len = %d\n",
                               print_limit < rc ? "..." : "", print_limit,
                               (char *)ma->ma_lmm + rc - print_limit, rc);
                        rc = 0;
                }
        }

        if (reqbody->valid & OBD_MD_FLMODEASIZE) {
                repbody->max_cookiesize = 0;
                repbody->max_mdsize = info->mti_mdt->mdt_max_mdsize;
                repbody->valid |= OBD_MD_FLMODEASIZE;
                CDEBUG(D_INODE, "I am going to change the MAX_MD_SIZE & "
                       "MAX_COOKIE to : %d:%d\n", repbody->max_mdsize,
                       repbody->max_cookiesize);
        }

        if (exp_connect_rmtclient(info->mti_exp) &&
            reqbody->valid & OBD_MD_FLRMTPERM) {
                void *buf = req_capsule_server_get(pill, &RMF_ACL);

                /* mdt_getattr_lock only */
                rc = mdt_pack_remote_perm(info, o, buf);
                if (rc) {
                        repbody->valid &= ~OBD_MD_FLRMTPERM;
                        repbody->aclsize = 0;
                        RETURN(rc);
                } else {
                        repbody->valid |= OBD_MD_FLRMTPERM;
                        repbody->aclsize = sizeof(struct mdt_remote_perm);
                }
        }
#ifdef CONFIG_FS_POSIX_ACL
        else if ((exp_connect_flags(req->rq_export) & OBD_CONNECT_ACL) &&
                 (reqbody->valid & OBD_MD_FLACL)) {
                buffer->lb_buf = req_capsule_server_get(pill, &RMF_ACL);
                buffer->lb_len = req_capsule_get_size(pill,
                                                      &RMF_ACL, RCL_SERVER);
                if (buffer->lb_len > 0) {
                        rc = mo_xattr_get(env, next, buffer,
                                          XATTR_NAME_ACL_ACCESS);
                        if (rc < 0) {
                                if (rc == -ENODATA) {
                                        repbody->aclsize = 0;
                                        repbody->valid |= OBD_MD_FLACL;
                                        rc = 0;
                                } else if (rc == -EOPNOTSUPP) {
                                        rc = 0;
                                } else {
                                        CERROR("%s: unable to read "DFID
                                               " ACL: rc = %d\n",
                                               mdt_obd_name(info->mti_mdt),
                                               PFID(mdt_object_fid(o)), rc);
                                }
                        } else {
                                repbody->aclsize = rc;
                                repbody->valid |= OBD_MD_FLACL;
                                rc = 0;
                        }
                }
        }
#endif

        if (reqbody->valid & OBD_MD_FLMDSCAPA &&
            info->mti_mdt->mdt_lut.lut_mds_capa &&
            exp_connect_flags(info->mti_exp) & OBD_CONNECT_MDS_CAPA) {
                struct lustre_capa *capa;

                capa = req_capsule_server_get(pill, &RMF_CAPA1);
                LASSERT(capa);
                capa->lc_opc = CAPA_OPC_MDS_DEFAULT;
                rc = mo_capa_get(env, next, capa, 0);
                if (rc)
                        RETURN(rc);
                repbody->valid |= OBD_MD_FLMDSCAPA;
        }

out:
        if (rc == 0)
                mdt_counter_incr(req, LPROC_MDT_GETATTR);

        RETURN(rc);
}

static int mdt_renew_capa(struct mdt_thread_info *info)
{
        struct mdt_object  *obj = info->mti_object;
        struct mdt_body    *body;
        struct lustre_capa *capa, *c;
        int rc;
        ENTRY;

        /* if object doesn't exist, or server has disabled capability,
         * return directly, client will find body->valid OBD_MD_FLOSSCAPA
         * flag not set.
         */
        if (!obj || !info->mti_mdt->mdt_lut.lut_oss_capa ||
            !(exp_connect_flags(info->mti_exp) & OBD_CONNECT_OSS_CAPA))
                RETURN(0);

        body = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
        LASSERT(body != NULL);

        c = req_capsule_client_get(info->mti_pill, &RMF_CAPA1);
        LASSERT(c);

        capa = req_capsule_server_get(info->mti_pill, &RMF_CAPA2);
        LASSERT(capa);

        *capa = *c;
        rc = mo_capa_get(info->mti_env, mdt_object_child(obj), capa, 1);
        if (rc == 0)
                body->valid |= OBD_MD_FLOSSCAPA;
        RETURN(rc);
}

static int mdt_getattr(struct tgt_session_info *tsi)
{
        struct mdt_thread_info  *info = tsi2mdt_info(tsi);
        struct mdt_object       *obj = info->mti_object;
        struct req_capsule      *pill = info->mti_pill;
        struct mdt_body         *reqbody;
        struct mdt_body         *repbody;
        int rc, rc2;
        ENTRY;

        reqbody = req_capsule_client_get(pill, &RMF_MDT_BODY);
        LASSERT(reqbody);

        if (reqbody->valid & OBD_MD_FLOSSCAPA) {
                rc = req_capsule_server_pack(pill);
                if (unlikely(rc))
                        RETURN(err_serious(rc));
                rc = mdt_renew_capa(info);
                GOTO(out_shrink, rc);
        }

        LASSERT(obj != NULL);
        LASSERT(lu_object_assert_exists(&obj->mot_obj));

        /* Unlike intent case where we need to pre-fill out buffers early on
         * in intent policy for ldlm reasons, here we can have a much better
         * guess at EA size by just reading it from disk.
         * Exceptions are readdir and (missing) directory striping */
        /* Readlink */
        if (reqbody->valid & OBD_MD_LINKNAME) {
                /* No easy way to know how long is the symlink, but it cannot
                 * be more than PATH_MAX, so we allocate +1 */
                rc = PATH_MAX + 1;

        /* A special case for fs ROOT: getattr there might fetch
         * default EA for entire fs, not just for this dir!
         */
        } else if (lu_fid_eq(mdt_object_fid(obj),
                             &info->mti_mdt->mdt_md_root_fid) &&
                   (reqbody->valid & OBD_MD_FLDIREA) &&
                   (lustre_msg_get_opc(mdt_info_req(info)->rq_reqmsg) ==
                                                                 MDS_GETATTR)) {
                /* Should the default strping be bigger, mdt_fix_reply
                 * will reallocate */
                rc = DEF_REP_MD_SIZE;
        } else {
                /* Read the actual EA size from disk */
                rc = mdt_attr_get_eabuf_size(info, obj);
        }

        if (rc < 0)
                GOTO(out_shrink, rc);

        req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER, rc);

        rc = req_capsule_server_pack(pill);
        if (unlikely(rc != 0))
                GOTO(out, rc = err_serious(rc));

        repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
        LASSERT(repbody != NULL);
        repbody->eadatasize = 0;
        repbody->aclsize = 0;

        if (reqbody->valid & OBD_MD_FLRMTPERM)
                rc = mdt_init_ucred(info, reqbody);
        else
                rc = mdt_check_ucred(info);
        if (unlikely(rc))
                GOTO(out_shrink, rc);

        info->mti_cross_ref = !!(reqbody->valid & OBD_MD_FLCROSSREF);

        /*
         * Don't check capability at all, because rename might getattr for
         * remote obj, and at that time no capability is available.
         */
        mdt_set_capainfo(info, 1, &reqbody->fid1, BYPASS_CAPA);
        rc = mdt_getattr_internal(info, obj, 0);
        if (reqbody->valid & OBD_MD_FLRMTPERM)
                mdt_exit_ucred(info);
        EXIT;
out_shrink:
        mdt_client_compatibility(info);
        rc2 = mdt_fix_reply(info);
        if (rc == 0)
                rc = rc2;
out:
        mdt_thread_info_fini(info);
        return rc;
}

static int mdt_is_subdir(struct tgt_session_info *tsi)
{
        struct mdt_thread_info  *info = tsi2mdt_info(tsi);
        struct mdt_object     *o = info->mti_object;
        struct req_capsule    *pill = info->mti_pill;
        const struct mdt_body *body = info->mti_body;
        struct mdt_body       *repbody;
        int                    rc;
        ENTRY;

        LASSERT(o != NULL);

        repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);

        /*
         * We save last checked parent fid to @repbody->fid1 for remote
         * directory case.
         */
        LASSERT(fid_is_sane(&body->fid2));
        LASSERT(mdt_object_exists(o) && !mdt_object_remote(o));
        rc = mdo_is_subdir(info->mti_env, mdt_object_child(o),
                           &body->fid2, &repbody->fid1);
        if (rc == 0 || rc == -EREMOTE)
                repbody->valid |= OBD_MD_FLID;

        mdt_thread_info_fini(info);
        RETURN(rc);
}

static int mdt_swap_layouts(struct tgt_session_info *tsi)
{
        struct mdt_thread_info  *info;
        struct ptlrpc_request   *req = tgt_ses_req(tsi);
        struct obd_export       *exp = req->rq_export;
        struct mdt_object       *o1, *o2, *o;
        struct mdt_lock_handle  *lh1, *lh2;
        struct mdc_swap_layouts *msl;
        int                      rc;
        ENTRY;

        /* client does not support layout lock, so layout swaping
         * is disabled.
         * FIXME: there is a problem for old clients which don't support
         * layout lock yet. If those clients have already opened the file
         * they won't be notified at all so that old layout may still be
         * used to do IO. This can be fixed after file release is landed by
         * doing exclusive open and taking full EX ibits lock. - Jinshan */
        if (!exp_connect_layout(exp))
                RETURN(-EOPNOTSUPP);

        info = tsi2mdt_info(tsi);

        if (info->mti_dlm_req != NULL)
                ldlm_request_cancel(req, info->mti_dlm_req, 0);

        if (req_capsule_get_size(info->mti_pill, &RMF_CAPA1, RCL_CLIENT))
                mdt_set_capainfo(info, 0, &info->mti_body->fid1,
                                 req_capsule_client_get(info->mti_pill,
                                                        &RMF_CAPA1));

        if (req_capsule_get_size(info->mti_pill, &RMF_CAPA2, RCL_CLIENT))
                mdt_set_capainfo(info, 1, &info->mti_body->fid2,
                                 req_capsule_client_get(info->mti_pill,
                                                        &RMF_CAPA2));

        o1 = info->mti_object;
        o = o2 = mdt_object_find(info->mti_env, info->mti_mdt,
                                &info->mti_body->fid2);
        if (IS_ERR(o))
                GOTO(out, rc = PTR_ERR(o));

        if (mdt_object_remote(o) || !mdt_object_exists(o)) /* remote object */
                GOTO(put, rc = -ENOENT);

        rc = lu_fid_cmp(&info->mti_body->fid1, &info->mti_body->fid2);
        if (unlikely(rc == 0)) /* same file, you kidding me? no-op. */
                GOTO(put, rc);

        if (rc < 0)
                swap(o1, o2);

        /* permission check. Make sure the calling process having permission
         * to write both files. */
        rc = mo_permission(info->mti_env, NULL, mdt_object_child(o1), NULL,
                                MAY_WRITE);
        if (rc < 0)
                GOTO(put, rc);

        rc = mo_permission(info->mti_env, NULL, mdt_object_child(o2), NULL,
                                MAY_WRITE);
        if (rc < 0)
                GOTO(put, rc);

        msl = req_capsule_client_get(info->mti_pill, &RMF_SWAP_LAYOUTS);
        if (msl == NULL)
                GOTO(put, rc = -EPROTO);

        lh1 = &info->mti_lh[MDT_LH_NEW];
        mdt_lock_reg_init(lh1, LCK_EX);
        lh2 = &info->mti_lh[MDT_LH_OLD];
        mdt_lock_reg_init(lh2, LCK_EX);

        rc = mdt_object_lock(info, o1, lh1, MDS_INODELOCK_LAYOUT |
                             MDS_INODELOCK_XATTR, MDT_LOCAL_LOCK);
        if (rc < 0)
                GOTO(put, rc);

        rc = mdt_object_lock(info, o2, lh2, MDS_INODELOCK_LAYOUT |
                             MDS_INODELOCK_XATTR, MDT_LOCAL_LOCK);
        if (rc < 0)
                GOTO(unlock1, rc);

        rc = mo_swap_layouts(info->mti_env, mdt_object_child(o1),
                             mdt_object_child(o2), msl->msl_flags);
        GOTO(unlock2, rc);
unlock2:
        mdt_object_unlock(info, o2, lh2, rc);
unlock1:
        mdt_object_unlock(info, o1, lh1, rc);
put:
        mdt_object_put(info->mti_env, o);
out:
        mdt_thread_info_fini(info);
        RETURN(rc);
}

static int mdt_raw_lookup(struct mdt_thread_info *info,
                          struct mdt_object *parent,
                          const struct lu_name *lname,
                          struct ldlm_reply *ldlm_rep)
{
        struct md_object *next = mdt_object_child(info->mti_object);
        const struct mdt_body *reqbody = info->mti_body;
        struct lu_fid *child_fid = &info->mti_tmp_fid1;
        struct mdt_body *repbody;
        int rc;
        ENTRY;

        if (reqbody->valid != OBD_MD_FLID)
                RETURN(0);

        LASSERT(!info->mti_cross_ref);

        /* Only got the fid of this obj by name */
        fid_zero(child_fid);
        rc = mdo_lookup(info->mti_env, next, lname, child_fid,
                        &info->mti_spec);
#if 0
        /* XXX is raw_lookup possible as intent operation? */
        if (rc != 0) {
                if (rc == -ENOENT)
                        mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_NEG);
                RETURN(rc);
        } else
                mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_POS);

        repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
#endif
        if (rc == 0) {
                repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
                repbody->fid1 = *child_fid;
                repbody->valid = OBD_MD_FLID;
        }
        RETURN(1);
}

/*
 * UPDATE lock should be taken against parent, and be release before exit;
 * child_bits lock should be taken against child, and be returned back:
 *            (1)normal request should release the child lock;
 *            (2)intent request will grant the lock to client.
 */
static int mdt_getattr_name_lock(struct mdt_thread_info *info,
                                 struct mdt_lock_handle *lhc,
                                 __u64 child_bits,
                                 struct ldlm_reply *ldlm_rep)
{
        struct ptlrpc_request  *req       = mdt_info_req(info);
        struct mdt_body        *reqbody   = NULL;
        struct mdt_object      *parent    = info->mti_object;
        struct mdt_object      *child;
        struct md_object       *next      = mdt_object_child(parent);
        struct lu_fid          *child_fid = &info->mti_tmp_fid1;
        struct lu_name         *lname     = NULL;
        struct mdt_lock_handle *lhp       = NULL;
        struct ldlm_lock       *lock;
        bool                    is_resent;
        bool                    try_layout;
        int                     ma_need = 0;
        int                     rc;
        ENTRY;

        is_resent = lustre_handle_is_used(&lhc->mlh_reg_lh);
        LASSERT(ergo(is_resent,
                     lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT));

        LASSERT(parent != NULL);

        if (info->mti_cross_ref) {
                /* Only getattr on the child. Parent is on another node. */
                mdt_set_disposition(info, ldlm_rep,
                                    DISP_LOOKUP_EXECD | DISP_LOOKUP_POS);
                child = parent;
                CDEBUG(D_INODE, "partial getattr_name child_fid = "DFID", "
                       "ldlm_rep = %p\n",
                       PFID(mdt_object_fid(child)), ldlm_rep);

                if (is_resent) {
                        /* Do not take lock for resent case. */
                        lock = ldlm_handle2lock(&lhc->mlh_reg_lh);
                        LASSERTF(lock != NULL, "Invalid lock handle "LPX64"\n",
                                 lhc->mlh_reg_lh.cookie);
                        LASSERT(fid_res_name_eq(mdt_object_fid(child),
                                                &lock->l_resource->lr_name));
                        LDLM_LOCK_PUT(lock);
                } else {
                        mdt_lock_handle_init(lhc);
                        mdt_lock_reg_init(lhc, LCK_PR);

                        /*
                         * Object's name is on another MDS, no lookup or layout
                         * lock is needed here but update lock is.
                         */
                        child_bits &= ~(MDS_INODELOCK_LOOKUP |
                                        MDS_INODELOCK_LAYOUT);
                        child_bits |= MDS_INODELOCK_PERM | MDS_INODELOCK_UPDATE;

                        rc = mdt_object_lock(info, child, lhc, child_bits,
                                             MDT_LOCAL_LOCK);
                        if (rc < 0)
                                RETURN(rc);
                }

                /* Finally, we can get attr for child. */
                if (!mdt_object_exists(child)) {
                        LU_OBJECT_DEBUG(D_INFO, info->mti_env,
                                        &child->mot_obj,
                                        "remote object doesn't exist.\n");
                        mdt_object_unlock(info, child, lhc, 1);
                        RETURN(-ENOENT);
                }

                mdt_set_capainfo(info, 0, mdt_object_fid(child), BYPASS_CAPA);
                rc = mdt_getattr_internal(info, child, 0);
                if (unlikely(rc != 0))
                        mdt_object_unlock(info, child, lhc, 1);

                RETURN(rc);
        }

        lname = &info->mti_name;
        mdt_name_unpack(info->mti_pill, &RMF_NAME, lname, MNF_FIX_ANON);

        if (lu_name_is_valid(lname)) {
                CDEBUG(D_INODE, "getattr with lock for "DFID"/"DNAME", "
                       "ldlm_rep = %p\n", PFID(mdt_object_fid(parent)),
                       PNAME(lname), ldlm_rep);
        } else {
                reqbody = req_capsule_client_get(info->mti_pill, &RMF_MDT_BODY);
                if (unlikely(reqbody == NULL))
                        RETURN(err_serious(-EPROTO));

                *child_fid = reqbody->fid2;

                if (unlikely(!fid_is_sane(child_fid)))
                        RETURN(err_serious(-EINVAL));

                CDEBUG(D_INODE, "getattr with lock for "DFID"/"DFID", "
                       "ldlm_rep = %p\n",
                       PFID(mdt_object_fid(parent)),
                       PFID(&reqbody->fid2), ldlm_rep);
        }

        mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_EXECD);

        if (unlikely(!mdt_object_exists(parent)) && lu_name_is_valid(lname)) {
                LU_OBJECT_DEBUG(D_INODE, info->mti_env,
                                &parent->mot_obj,
                                "Parent doesn't exist!\n");
                RETURN(-ESTALE);
        }

        if (mdt_object_remote(parent)) {
                CERROR("%s: parent "DFID" is on remote target\n",
                       mdt_obd_name(info->mti_mdt),
                       PFID(mdt_object_fid(parent)));
                RETURN(-EIO);
        }

        if (lu_name_is_valid(lname)) {
                rc = mdt_raw_lookup(info, parent, lname, ldlm_rep);
                if (rc != 0) {
                        if (rc > 0)
                                rc = 0;
                        RETURN(rc);
                }

                /* step 1: lock parent only if parent is a directory */
                if (S_ISDIR(lu_object_attr(&parent->mot_obj))) {
                        lhp = &info->mti_lh[MDT_LH_PARENT];
                        mdt_lock_pdo_init(lhp, LCK_PR, lname);
                        rc = mdt_object_lock(info, parent, lhp,
                                             MDS_INODELOCK_UPDATE,
                                             MDT_LOCAL_LOCK);
                        if (unlikely(rc != 0))
                                RETURN(rc);
                }

                /* step 2: lookup child's fid by name */
                fid_zero(child_fid);
                rc = mdo_lookup(info->mti_env, next, lname, child_fid,
                                &info->mti_spec);
                if (rc == -ENOENT)
                        mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_NEG);

                if (rc != 0)
                        GOTO(out_parent, rc);
        }

        mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_POS);

        /*
         *step 3: find the child object by fid & lock it.
         *        regardless if it is local or remote.
         *
         *Note: LU-3240 (commit 762f2114d282a98ebfa4dbbeea9298a8088ad24e)
         *      set parent dir fid the same as child fid in getattr by fid case
         *      we should not lu_object_find() the object again, could lead
         *      to hung if there is a concurrent unlink destroyed the object.
         */
        if (lu_fid_eq(mdt_object_fid(parent), child_fid)) {
                mdt_object_get(info->mti_env, parent);
                child = parent;
        } else {
                child = mdt_object_find(info->mti_env, info->mti_mdt,
                                        child_fid);
        }

        if (unlikely(IS_ERR(child)))
                GOTO(out_parent, rc = PTR_ERR(child));
        if (is_resent) {
                /* Do not take lock for resent case. */
                lock = ldlm_handle2lock(&lhc->mlh_reg_lh);
                LASSERTF(lock != NULL, "Invalid lock handle "LPX64"\n",
                         lhc->mlh_reg_lh.cookie);

                if (!fid_res_name_eq(mdt_object_fid(child),
                                     &lock->l_resource->lr_name)) {
                        LASSERTF(fid_res_name_eq(mdt_object_fid(parent),
                                                 &lock->l_resource->lr_name),
                                 "Lock res_id: "DLDLMRES", fid: "DFID"\n",
                                 PLDLMRES(lock->l_resource),
                                 PFID(mdt_object_fid(parent)));
                        CWARN("Although resent, but still not get child lock"
                              "parent:"DFID" child:"DFID"\n",
                              PFID(mdt_object_fid(parent)),
                              PFID(mdt_object_fid(child)));
                        lustre_msg_clear_flags(req->rq_reqmsg, MSG_RESENT);
                        LDLM_LOCK_PUT(lock);
                        GOTO(relock, 0);
                }
                LDLM_LOCK_PUT(lock);
        } else {
relock:
                OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_RESEND, obd_timeout*2);
                mdt_lock_handle_init(lhc);
                mdt_lock_reg_init(lhc, LCK_PR);
                try_layout = false;

                if (!mdt_object_exists(child)) {
                        LU_OBJECT_DEBUG(D_INODE, info->mti_env,
                                        &child->mot_obj,
                                        "Object doesn't exist!\n");
                        GOTO(out_child, rc = -ENOENT);
                }

                if (!(child_bits & MDS_INODELOCK_UPDATE) &&
                      mdt_object_exists(child) && !mdt_object_remote(child)) {
                        struct md_attr *ma = &info->mti_attr;

                        ma->ma_valid = 0;
                        ma->ma_need = MA_INODE;
                        rc = mdt_attr_get_complex(info, child, ma);
                        if (unlikely(rc != 0))
                                GOTO(out_child, rc);

                        /* If the file has not been changed for some time, we
                         * return not only a LOOKUP lock, but also an UPDATE
                         * lock and this might save us RPC on later STAT. For
                         * directories, it also let negative dentry cache start
                         * working for this dir. */
                        if (ma->ma_valid & MA_INODE &&
                            ma->ma_attr.la_valid & LA_CTIME &&
                            info->mti_mdt->mdt_namespace->ns_ctime_age_limit +
                                ma->ma_attr.la_ctime < cfs_time_current_sec())
                                child_bits |= MDS_INODELOCK_UPDATE;
                }

                /* layout lock must be granted in a best-effort way
                 * for IT operations */
                LASSERT(!(child_bits & MDS_INODELOCK_LAYOUT));
                if (!OBD_FAIL_CHECK(OBD_FAIL_MDS_NO_LL_GETATTR) &&
                    exp_connect_layout(info->mti_exp) &&
                    S_ISREG(lu_object_attr(&child->mot_obj)) &&
                    !mdt_object_remote(child) && ldlm_rep != NULL) {
                        /* try to grant layout lock for regular file. */
                        try_layout = true;
                }

                rc = 0;
                if (try_layout) {
                        child_bits |= MDS_INODELOCK_LAYOUT;
                        /* try layout lock, it may fail to be granted due to
                         * contention at LOOKUP or UPDATE */
                        if (!mdt_object_lock_try(info, child, lhc, child_bits,
                                                 MDT_CROSS_LOCK)) {
                                child_bits &= ~MDS_INODELOCK_LAYOUT;
                                LASSERT(child_bits != 0);
                                rc = mdt_object_lock(info, child, lhc,
                                                child_bits, MDT_CROSS_LOCK);
                        } else {
                                ma_need |= MA_LOV;
                        }
                } else {
                        /* Do not enqueue the UPDATE lock from MDT(cross-MDT),
                         * client will enqueue the lock to the remote MDT */
                        if (mdt_object_remote(child))
                                child_bits &= ~MDS_INODELOCK_UPDATE;
                        rc = mdt_object_lock(info, child, lhc, child_bits,
                                                MDT_CROSS_LOCK);
                }
                if (unlikely(rc != 0))
                        GOTO(out_child, rc);
        }

        lock = ldlm_handle2lock(&lhc->mlh_reg_lh);
        /* Get MA_SOM attributes if update lock is given. */
        if (lock &&
            lock->l_policy_data.l_inodebits.bits & MDS_INODELOCK_UPDATE &&
            S_ISREG(lu_object_attr(&mdt_object_child(child)->mo_lu)))
                ma_need |= MA_SOM;

        /* finally, we can get attr for child. */
        mdt_set_capainfo(info, 1, child_fid, BYPASS_CAPA);
        rc = mdt_getattr_internal(info, child, ma_need);
        if (unlikely(rc != 0)) {
                mdt_object_unlock(info, child, lhc, 1);
        } else if (lock) {
                /* Debugging code. */
                LDLM_DEBUG(lock, "Returning lock to client");
                LASSERTF(fid_res_name_eq(mdt_object_fid(child),
                                         &lock->l_resource->lr_name),
                         "Lock res_id: "DLDLMRES", fid: "DFID"\n",
                         PLDLMRES(lock->l_resource),
                         PFID(mdt_object_fid(child)));
                if (mdt_object_exists(child) && !mdt_object_remote(child))
                        mdt_pack_size2body(info, child);
        }
        if (lock)
                LDLM_LOCK_PUT(lock);

        EXIT;
out_child:
        mdt_object_put(info->mti_env, child);
out_parent:
        if (lhp)
                mdt_object_unlock(info, parent, lhp, 1);
        return rc;
}

/* normal handler: should release the child lock */
static int mdt_getattr_name(struct tgt_session_info *tsi)
{
        struct mdt_thread_info  *info = tsi2mdt_info(tsi);
        struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_CHILD];
        struct mdt_body        *reqbody;
        struct mdt_body        *repbody;
        int rc, rc2;
        ENTRY;

        reqbody = req_capsule_client_get(info->mti_pill, &RMF_MDT_BODY);
        LASSERT(reqbody != NULL);
        repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
        LASSERT(repbody != NULL);

        info->mti_cross_ref = !!(reqbody->valid & OBD_MD_FLCROSSREF);
        repbody->eadatasize = 0;
        repbody->aclsize = 0;

        rc = mdt_init_ucred(info, reqbody);
        if (unlikely(rc))
                GOTO(out_shrink, rc);

        rc = mdt_getattr_name_lock(info, lhc, MDS_INODELOCK_UPDATE, NULL);
        if (lustre_handle_is_used(&lhc->mlh_reg_lh)) {
                ldlm_lock_decref(&lhc->mlh_reg_lh, lhc->mlh_reg_mode);
                lhc->mlh_reg_lh.cookie = 0;
        }
        mdt_exit_ucred(info);
        EXIT;
out_shrink:
        mdt_client_compatibility(info);
        rc2 = mdt_fix_reply(info);
        if (rc == 0)
                rc = rc2;
        mdt_thread_info_fini(info);
        return rc;
}

static int mdt_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
                         void *karg, void *uarg);

static int mdt_set_info(struct tgt_session_info *tsi)
{
        struct ptlrpc_request   *req = tgt_ses_req(tsi);
        char                    *key;
        void                    *val;
        int                      keylen, vallen, rc = 0;

        ENTRY;

        key = req_capsule_client_get(tsi->tsi_pill, &RMF_SETINFO_KEY);
        if (key == NULL) {
                DEBUG_REQ(D_HA, req, "no set_info key");
                RETURN(err_serious(-EFAULT));
        }

        keylen = req_capsule_get_size(tsi->tsi_pill, &RMF_SETINFO_KEY,
                                      RCL_CLIENT);

        val = req_capsule_client_get(tsi->tsi_pill, &RMF_SETINFO_VAL);
        if (val == NULL) {
                DEBUG_REQ(D_HA, req, "no set_info val");
                RETURN(err_serious(-EFAULT));
        }

        vallen = req_capsule_get_size(tsi->tsi_pill, &RMF_SETINFO_VAL,
                                      RCL_CLIENT);

        /* Swab any part of val you need to here */
        if (KEY_IS(KEY_READ_ONLY)) {
                spin_lock(&req->rq_export->exp_lock);
                if (*(__u32 *)val)
                        *exp_connect_flags_ptr(req->rq_export) |=
                                OBD_CONNECT_RDONLY;
                else
                        *exp_connect_flags_ptr(req->rq_export) &=
                                ~OBD_CONNECT_RDONLY;
                spin_unlock(&req->rq_export->exp_lock);
        } else if (KEY_IS(KEY_CHANGELOG_CLEAR)) {
                struct changelog_setinfo *cs = val;

                if (vallen != sizeof(*cs)) {
                        CERROR("%s: bad changelog_clear setinfo size %d\n",
                               tgt_name(tsi->tsi_tgt), vallen);
                        RETURN(-EINVAL);
                }
                if (ptlrpc_req_need_swab(req)) {
                        __swab64s(&cs->cs_recno);
                        __swab32s(&cs->cs_id);
                }

                rc = mdt_iocontrol(OBD_IOC_CHANGELOG_CLEAR, req->rq_export,
                                   vallen, val, NULL);
        } else {
                RETURN(-EINVAL);
        }
        RETURN(rc);
}

static int mdt_readpage(struct tgt_session_info *tsi)
{
        struct mdt_thread_info  *info = mdt_th_info(tsi->tsi_env);
        struct mdt_object       *object = mdt_obj(tsi->tsi_corpus);
        struct lu_rdpg          *rdpg = &info->mti_u.rdpg.mti_rdpg;
        const struct mdt_body   *reqbody = tsi->tsi_mdt_body;
        struct mdt_body         *repbody;
        int                      rc;
        int                      i;

        ENTRY;

        if (OBD_FAIL_CHECK(OBD_FAIL_MDS_READPAGE_PACK))
                RETURN(err_serious(-ENOMEM));

        repbody = req_capsule_server_get(tsi->tsi_pill, &RMF_MDT_BODY);
        if (repbody == NULL || reqbody == NULL)
                RETURN(err_serious(-EFAULT));

        /*
         * prepare @rdpg before calling lower layers and transfer itself. Here
         * reqbody->size contains offset of where to start to read and
         * reqbody->nlink contains number bytes to read.
         */
        rdpg->rp_hash = reqbody->size;
        if (rdpg->rp_hash != reqbody->size) {
                CERROR("Invalid hash: "LPX64" != "LPX64"\n",
                       rdpg->rp_hash, reqbody->size);
                RETURN(-EFAULT);
        }

        rdpg->rp_attrs = reqbody->mode;
        if (exp_connect_flags(tsi->tsi_exp) & OBD_CONNECT_64BITHASH)
                rdpg->rp_attrs |= LUDA_64BITHASH;
        rdpg->rp_count  = min_t(unsigned int, reqbody->nlink,
                                exp_max_brw_size(tsi->tsi_exp));
        rdpg->rp_npages = (rdpg->rp_count + PAGE_CACHE_SIZE - 1) >>
                          PAGE_CACHE_SHIFT;
        OBD_ALLOC(rdpg->rp_pages, rdpg->rp_npages * sizeof rdpg->rp_pages[0]);
        if (rdpg->rp_pages == NULL)
                RETURN(-ENOMEM);

        for (i = 0; i < rdpg->rp_npages; ++i) {
                rdpg->rp_pages[i] = alloc_page(GFP_IOFS);
                if (rdpg->rp_pages[i] == NULL)
                        GOTO(free_rdpg, rc = -ENOMEM);
        }

        /* call lower layers to fill allocated pages with directory data */
        rc = mo_readpage(tsi->tsi_env, mdt_object_child(object), rdpg);
        if (rc < 0)
                GOTO(free_rdpg, rc);

        /* send pages to client */
        rc = tgt_sendpage(tsi, rdpg, rc);

        EXIT;
free_rdpg:

        for (i = 0; i < rdpg->rp_npages; i++)
                if (rdpg->rp_pages[i] != NULL)
                        __free_page(rdpg->rp_pages[i]);
        OBD_FREE(rdpg->rp_pages, rdpg->rp_npages * sizeof rdpg->rp_pages[0]);

        if (OBD_FAIL_CHECK(OBD_FAIL_MDS_SENDPAGE))
                RETURN(0);

        return rc;
}

static int mdt_reint_internal(struct mdt_thread_info *info,
                              struct mdt_lock_handle *lhc,
                              __u32 op)
{
        struct req_capsule      *pill = info->mti_pill;
        struct mdt_body         *repbody;
        int                      rc = 0, rc2;

        ENTRY;

        rc = mdt_reint_unpack(info, op);
        if (rc != 0) {
                CERROR("Can't unpack reint, rc %d\n", rc);
                RETURN(err_serious(rc));
        }

        /* for replay (no_create) lmm is not needed, client has it already */
        if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER))
                req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER,
                                     DEF_REP_MD_SIZE);

        /* llog cookies are always 0, the field is kept for compatibility */
        if (req_capsule_has_field(pill, &RMF_LOGCOOKIES, RCL_SERVER))
                req_capsule_set_size(pill, &RMF_LOGCOOKIES, RCL_SERVER, 0);

        rc = req_capsule_server_pack(pill);
        if (rc != 0) {
                CERROR("Can't pack response, rc %d\n", rc);
                RETURN(err_serious(rc));
        }

        if (req_capsule_has_field(pill, &RMF_MDT_BODY, RCL_SERVER)) {
                repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
                LASSERT(repbody);
                repbody->eadatasize = 0;
                repbody->aclsize = 0;
        }

        OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_REINT_DELAY, 10);

        /* for replay no cookkie / lmm need, because client have this already */
        if (info->mti_spec.no_create)
                if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER))
                        req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER, 0);

        rc = mdt_init_ucred_reint(info);
        if (rc)
                GOTO(out_shrink, rc);

        rc = mdt_fix_attr_ucred(info, op);
        if (rc != 0)
                GOTO(out_ucred, rc = err_serious(rc));

        if (mdt_check_resent(info, mdt_reconstruct, lhc)) {
                DEBUG_REQ(D_INODE, mdt_info_req(info), "resent opt.");
                rc = lustre_msg_get_status(mdt_info_req(info)->rq_repmsg);
                GOTO(out_ucred, rc);
        }
        rc = mdt_reint_rec(info, lhc);
        EXIT;
out_ucred:
        mdt_exit_ucred(info);
out_shrink:
        mdt_client_compatibility(info);
        rc2 = mdt_fix_reply(info);
        if (rc == 0)
                rc = rc2;
        return rc;
}

static long mdt_reint_opcode(struct ptlrpc_request *req,
                             const struct req_format **fmt)
{
        struct mdt_device       *mdt;
        struct mdt_rec_reint    *rec;
        long                     opc;

        rec = req_capsule_client_get(&req->rq_pill, &RMF_REC_REINT);
        if (rec != NULL) {
                opc = rec->rr_opcode;
                DEBUG_REQ(D_INODE, req, "reint opt = %ld", opc);
                if (opc < REINT_MAX && fmt[opc] != NULL)
                        req_capsule_extend(&req->rq_pill, fmt[opc]);
                else {
                        mdt = mdt_exp2dev(req->rq_export);
                        CERROR("%s: Unsupported opcode '%ld' from client '%s':"
                               " rc = %d\n", req->rq_export->exp_obd->obd_name,
                               opc, mdt->mdt_ldlm_client->cli_name, -EFAULT);
                        opc = err_serious(-EFAULT);
                }
        } else {
                opc = err_serious(-EFAULT);
        }
        return opc;
}

static int mdt_reint(struct tgt_session_info *tsi)
{
        long opc;
        int  rc;
        static const struct req_format *reint_fmts[REINT_MAX] = {
                [REINT_SETATTR]  = &RQF_MDS_REINT_SETATTR,
                [REINT_CREATE]   = &RQF_MDS_REINT_CREATE,
                [REINT_LINK]     = &RQF_MDS_REINT_LINK,
                [REINT_UNLINK]   = &RQF_MDS_REINT_UNLINK,
                [REINT_RENAME]   = &RQF_MDS_REINT_RENAME,
                [REINT_OPEN]     = &RQF_MDS_REINT_OPEN,
                [REINT_SETXATTR] = &RQF_MDS_REINT_SETXATTR,
                [REINT_RMENTRY]  = &RQF_MDS_REINT_UNLINK,
                [REINT_MIGRATE]  = &RQF_MDS_REINT_RENAME
        };

        ENTRY;

        opc = mdt_reint_opcode(tgt_ses_req(tsi), reint_fmts);
        if (opc >= 0) {
                struct mdt_thread_info *info = tsi2mdt_info(tsi);
                /*
                 * No lock possible here from client to pass it to reint code
                 * path.
                 */
                rc = mdt_reint_internal(info, NULL, opc);
                mdt_thread_info_fini(info);
        } else {
                rc = opc;
        }

        tsi->tsi_reply_fail_id = OBD_FAIL_MDS_REINT_NET_REP;
        RETURN(rc);
}

/* this should sync the whole device */
static int mdt_device_sync(const struct lu_env *env, struct mdt_device *mdt)
{
        struct dt_device *dt = mdt->mdt_bottom;
        int rc;
        ENTRY;

        rc = dt->dd_ops->dt_sync(env, dt);
        RETURN(rc);
}

/* this should sync this object */
static int mdt_object_sync(struct mdt_thread_info *info)
{
        struct md_object *next;
        int rc;
        ENTRY;

        if (!mdt_object_exists(info->mti_object)) {
                CWARN("Non existing object  "DFID"!\n",
                      PFID(mdt_object_fid(info->mti_object)));
                RETURN(-ESTALE);
        }
        next = mdt_object_child(info->mti_object);
        rc = mo_object_sync(info->mti_env, next);

        RETURN(rc);
}

static int mdt_sync(struct tgt_session_info *tsi)
{
        struct ptlrpc_request   *req = tgt_ses_req(tsi);
        struct req_capsule      *pill = tsi->tsi_pill;
        struct mdt_body         *body;
        int                      rc;

        ENTRY;

        if (OBD_FAIL_CHECK(OBD_FAIL_MDS_SYNC_PACK))
                RETURN(err_serious(-ENOMEM));

        if (fid_seq(&tsi->tsi_mdt_body->fid1) == 0) {
                rc = mdt_device_sync(tsi->tsi_env, mdt_exp2dev(tsi->tsi_exp));
        } else {
                struct mdt_thread_info *info = tsi2mdt_info(tsi);

                /* sync an object */
                rc = mdt_object_sync(info);
                if (rc == 0) {
                        const struct lu_fid *fid;
                        struct lu_attr *la = &info->mti_attr.ma_attr;

                        info->mti_attr.ma_need = MA_INODE;
                        info->mti_attr.ma_valid = 0;
                        rc = mdt_attr_get_complex(info, info->mti_object,
                                                  &info->mti_attr);
                        if (rc == 0) {
                                body = req_capsule_server_get(pill,
                                                              &RMF_MDT_BODY);
                                fid = mdt_object_fid(info->mti_object);
                                mdt_pack_attr2body(info, body, la, fid);
                        }
                }
                mdt_thread_info_fini(info);
        }
        if (rc == 0)
                mdt_counter_incr(req, LPROC_MDT_SYNC);

        RETURN(rc);
}

/*
 * Handle quota control requests to consult current usage/limit, but also
 * to configure quota enforcement
 */
static int mdt_quotactl(struct tgt_session_info *tsi)
{
        struct obd_export       *exp  = tsi->tsi_exp;
        struct req_capsule      *pill = tsi->tsi_pill;
        struct obd_quotactl     *oqctl, *repoqc;
        int                      id, rc;
        struct mdt_device       *mdt = mdt_exp2dev(exp);
        struct lu_device        *qmt = mdt->mdt_qmt_dev;
        ENTRY;

        oqctl = req_capsule_client_get(pill, &RMF_OBD_QUOTACTL);
        if (oqctl == NULL)
                RETURN(err_serious(-EPROTO));

        rc = req_capsule_server_pack(pill);
        if (rc)
                RETURN(err_serious(rc));

        switch (oqctl->qc_cmd) {
        case Q_QUOTACHECK:
        case LUSTRE_Q_INVALIDATE:
        case LUSTRE_Q_FINVALIDATE:
        case Q_QUOTAON:
        case Q_QUOTAOFF:
        case Q_INITQUOTA:
                /* deprecated, not used any more */
                RETURN(-EOPNOTSUPP);
                /* master quotactl */
        case Q_GETINFO:
        case Q_SETINFO:
        case Q_SETQUOTA:
        case Q_GETQUOTA:
                if (qmt == NULL)
                        RETURN(-EOPNOTSUPP);
                /* slave quotactl */
        case Q_GETOINFO:
        case Q_GETOQUOTA:
                break;
        default:
                CERROR("Unsupported quotactl command: %d\n", oqctl->qc_cmd);
                RETURN(-EFAULT);
        }

        /* map uid/gid for remote client */
        id = oqctl->qc_id;
        if (exp_connect_rmtclient(exp)) {
                struct lustre_idmap_table *idmap;

                idmap = exp->exp_mdt_data.med_idmap;

                if (unlikely(oqctl->qc_cmd != Q_GETQUOTA &&
                             oqctl->qc_cmd != Q_GETINFO))
                        RETURN(-EPERM);

                if (oqctl->qc_type == USRQUOTA)
                        id = lustre_idmap_lookup_uid(NULL, idmap, 0,
                                                     oqctl->qc_id);
                else if (oqctl->qc_type == GRPQUOTA)
                        id = lustre_idmap_lookup_gid(NULL, idmap, 0,
                                                     oqctl->qc_id);
                else
                        RETURN(-EINVAL);

                if (id == CFS_IDMAP_NOTFOUND) {
                        CDEBUG(D_QUOTA, "no mapping for id %u\n", oqctl->qc_id);
                        RETURN(-EACCES);
                }
        }

        repoqc = req_capsule_server_get(pill, &RMF_OBD_QUOTACTL);
        if (repoqc == NULL)
                RETURN(err_serious(-EFAULT));

        if (oqctl->qc_id != id)
                swap(oqctl->qc_id, id);

        switch (oqctl->qc_cmd) {

        case Q_GETINFO:
        case Q_SETINFO:
        case Q_SETQUOTA:
        case Q_GETQUOTA:
                /* forward quotactl request to QMT */
                rc = qmt_hdls.qmth_quotactl(tsi->tsi_env, qmt, oqctl);
                break;

        case Q_GETOINFO:
        case Q_GETOQUOTA:
                /* slave quotactl */
                rc = lquotactl_slv(tsi->tsi_env, tsi->tsi_tgt->lut_bottom,
                                   oqctl);
                break;

        default:
                CERROR("Unsupported quotactl command: %d\n", oqctl->qc_cmd);
                RETURN(-EFAULT);
        }

        if (oqctl->qc_id != id)
                swap(oqctl->qc_id, id);

        *repoqc = *oqctl;
        RETURN(rc);
}

/** clone llog ctxt from child (mdd)
 * This allows remote llog (replicator) access.
 * We can either pass all llog RPCs (eg mdt_llog_create) on to child where the
 * context was originally set up, or we can handle them directly.
 * I choose the latter, but that means I need any llog
 * contexts set up by child to be accessable by the mdt.  So we clone the
 * context into our context list here.
 */
static int mdt_llog_ctxt_clone(const struct lu_env *env, struct mdt_device *mdt,
                               int idx)
{
        struct md_device  *next = mdt->mdt_child;
        struct llog_ctxt *ctxt;
        int rc;

        if (!llog_ctxt_null(mdt2obd_dev(mdt), idx))
                return 0;

        rc = next->md_ops->mdo_llog_ctxt_get(env, next, idx, (void **)&ctxt);
        if (rc || ctxt == NULL) {
                return 0;
        }

        rc = llog_group_set_ctxt(&mdt2obd_dev(mdt)->obd_olg, ctxt, idx);
        if (rc)
                CERROR("Can't set mdt ctxt %d\n", rc);

        return rc;
}

static int mdt_llog_ctxt_unclone(const struct lu_env *env,
                                 struct mdt_device *mdt, int idx)
{
        struct llog_ctxt *ctxt;

        ctxt = llog_get_context(mdt2obd_dev(mdt), idx);
        if (ctxt == NULL)
                return 0;
        /* Put once for the get we just did, and once for the clone */
        llog_ctxt_put(ctxt);
        llog_ctxt_put(ctxt);
        return 0;
}

/*
 * sec context handlers
 */
static int mdt_sec_ctx_handle(struct tgt_session_info *tsi)
{
        int rc;

        rc = mdt_handle_idmap(tsi);
        if (unlikely(rc)) {
                struct ptlrpc_request   *req = tgt_ses_req(tsi);
                __u32                    opc;

                opc = lustre_msg_get_opc(req->rq_reqmsg);
                if (opc == SEC_CTX_INIT || opc == SEC_CTX_INIT_CONT)
                        sptlrpc_svc_ctx_invalidate(req);
        }

        CFS_FAIL_TIMEOUT(OBD_FAIL_SEC_CTX_HDL_PAUSE, cfs_fail_val);

        return rc;
}

/*
 * quota request handlers
 */
static int mdt_quota_dqacq(struct tgt_session_info *tsi)
{
        struct mdt_device       *mdt = mdt_exp2dev(tsi->tsi_exp);
        struct lu_device        *qmt = mdt->mdt_qmt_dev;
        int                      rc;
        ENTRY;

        if (qmt == NULL)
                RETURN(err_serious(-EOPNOTSUPP));

        rc = qmt_hdls.qmth_dqacq(tsi->tsi_env, qmt, tgt_ses_req(tsi));
        RETURN(rc);
}

struct mdt_object *mdt_object_new(const struct lu_env *env,
                                  struct mdt_device *d,
                                  const struct lu_fid *f)
{
        struct lu_object_conf conf = { .loc_flags = LOC_F_NEW };
        struct lu_object *o;
        struct mdt_object *m;
        ENTRY;

        CDEBUG(D_INFO, "Allocate object for "DFID"\n", PFID(f));
        o = lu_object_find(env, &d->mdt_lu_dev, f, &conf);
        if (unlikely(IS_ERR(o)))
                m = (struct mdt_object *)o;
        else
                m = mdt_obj(o);
        RETURN(m);
}

struct mdt_object *mdt_object_find(const struct lu_env *env,
                                   struct mdt_device *d,
                                   const struct lu_fid *f)
{
        struct lu_object *o;
        struct mdt_object *m;
        ENTRY;

        CDEBUG(D_INFO, "Find object for "DFID"\n", PFID(f));
        o = lu_object_find(env, &d->mdt_lu_dev, f, NULL);
        if (unlikely(IS_ERR(o)))
                m = (struct mdt_object *)o;
        else
                m = mdt_obj(o);

        RETURN(m);
}

/**
 * Asyncronous commit for mdt device.
 *
 * Pass asynchonous commit call down the MDS stack.
 *
 * \param env environment
 * \param mdt the mdt device
 */
static void mdt_device_commit_async(const struct lu_env *env,
                                    struct mdt_device *mdt)
{
        struct dt_device *dt = mdt->mdt_bottom;
        int rc;

        rc = dt->dd_ops->dt_commit_async(env, dt);
        if (unlikely(rc != 0))
                CWARN("async commit start failed with rc = %d", rc);
}

/**
 * Mark the lock as "synchonous".
 *
 * Mark the lock to deffer transaction commit to the unlock time.
 *
 * \param lock the lock to mark as "synchonous"
 *
 * \see mdt_is_lock_sync
 * \see mdt_save_lock
 */
static inline void mdt_set_lock_sync(struct ldlm_lock *lock)
{
        lock->l_ast_data = (void*)1;
}

/**
 * Check whehter the lock "synchonous" or not.
 *
 * \param lock the lock to check
 * \retval 1 the lock is "synchonous"
 * \retval 0 the lock isn't "synchronous"
 *
 * \see mdt_set_lock_sync
 * \see mdt_save_lock
 */
static inline int mdt_is_lock_sync(struct ldlm_lock *lock)
{
        return lock->l_ast_data != NULL;
}

/**
 * Blocking AST for mdt locks.
 *
 * Starts transaction commit if in case of COS lock conflict or
 * deffers such a commit to the mdt_save_lock.
 *
 * \param lock the lock which blocks a request or cancelling lock
 * \param desc unused
 * \param data unused
 * \param flag indicates whether this cancelling or blocking callback
 * \retval 0
 * \see ldlm_blocking_ast_nocheck
 */
int mdt_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
                     void *data, int flag)
{
        struct obd_device *obd = ldlm_lock_to_ns(lock)->ns_obd;
        struct mdt_device *mdt = mdt_dev(obd->obd_lu_dev);
        int rc;
        ENTRY;

        if (flag == LDLM_CB_CANCELING)
                RETURN(0);
        lock_res_and_lock(lock);
        if (lock->l_blocking_ast != mdt_blocking_ast) {
                unlock_res_and_lock(lock);
                RETURN(0);
        }
        if (mdt_cos_is_enabled(mdt) &&
            lock->l_req_mode & (LCK_PW | LCK_EX) &&
            lock->l_blocking_lock != NULL &&
            lock->l_client_cookie != lock->l_blocking_lock->l_client_cookie) {
                mdt_set_lock_sync(lock);
        }
        rc = ldlm_blocking_ast_nocheck(lock);

        /* There is no lock conflict if l_blocking_lock == NULL,
         * it indicates a blocking ast sent from ldlm_lock_decref_internal
         * when the last reference to a local lock was released */
        if (lock->l_req_mode == LCK_COS && lock->l_blocking_lock != NULL) {
                struct lu_env env;

                rc = lu_env_init(&env, LCT_LOCAL);
                if (unlikely(rc != 0))
                        CWARN("lu_env initialization failed with rc = %d,"
                              "cannot start asynchronous commit\n", rc);
                else
                        mdt_device_commit_async(&env, mdt);
                lu_env_fini(&env);
        }
        RETURN(rc);
}

/* Used for cross-MDT lock */
int mdt_remote_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
                            void *data, int flag)
{
        struct lustre_handle lockh;
        int               rc;

        switch (flag) {
        case LDLM_CB_BLOCKING:
                ldlm_lock2handle(lock, &lockh);
                rc = ldlm_cli_cancel(&lockh, LCF_ASYNC);
                if (rc < 0) {
                        CDEBUG(D_INODE, "ldlm_cli_cancel: %d\n", rc);
                        RETURN(rc);
                }
                break;
        case LDLM_CB_CANCELING:
                LDLM_DEBUG(lock, "Revoke remote lock\n");
                break;
        default:
                LBUG();
        }
        RETURN(0);
}

int mdt_remote_object_lock(struct mdt_thread_info *mti,
                           struct mdt_object *o, const struct lu_fid *fid,
                           struct lustre_handle *lh, ldlm_mode_t mode,
                           __u64 ibits)
{
        struct ldlm_enqueue_info *einfo = &mti->mti_einfo;
        ldlm_policy_data_t *policy = &mti->mti_policy;
        struct ldlm_res_id *res_id = &mti->mti_res_id;
        int rc = 0;
        ENTRY;

        LASSERT(mdt_object_remote(o));

        LASSERT(ibits == MDS_INODELOCK_UPDATE);

        fid_build_reg_res_name(fid, res_id);

        memset(einfo, 0, sizeof(*einfo));
        einfo->ei_type = LDLM_IBITS;
        einfo->ei_mode = mode;
        einfo->ei_cb_bl = mdt_remote_blocking_ast;
        einfo->ei_cb_cp = ldlm_completion_ast;
        einfo->ei_enq_slave = 0;
        einfo->ei_res_id = res_id;

        memset(policy, 0, sizeof(*policy));
        policy->l_inodebits.bits = ibits;

        rc = mo_object_lock(mti->mti_env, mdt_object_child(o), lh, einfo,
                            policy);
        RETURN(rc);
}

static int mdt_object_local_lock(struct mdt_thread_info *info,
                                 struct mdt_object *o,
                                 struct mdt_lock_handle *lh, __u64 ibits,
                                 bool nonblock, int locality)
{
        struct ldlm_namespace *ns = info->mti_mdt->mdt_namespace;
        ldlm_policy_data_t *policy = &info->mti_policy;
        struct ldlm_res_id *res_id = &info->mti_res_id;
        __u64 dlmflags;
        int rc;
        ENTRY;

        LASSERT(!lustre_handle_is_used(&lh->mlh_reg_lh));
        LASSERT(!lustre_handle_is_used(&lh->mlh_pdo_lh));
        LASSERT(lh->mlh_reg_mode != LCK_MINMODE);
        LASSERT(lh->mlh_type != MDT_NUL_LOCK);

        /* Only enqueue LOOKUP lock for remote object */
        if (mdt_object_remote(o))
                LASSERT(ibits == MDS_INODELOCK_LOOKUP);

        if (lh->mlh_type == MDT_PDO_LOCK) {
                /* check for exists after object is locked */
                if (mdt_object_exists(o) == 0) {
                        /* Non-existent object shouldn't have PDO lock */
                        RETURN(-ESTALE);
                } else {
                        /* Non-dir object shouldn't have PDO lock */
                        if (!S_ISDIR(lu_object_attr(&o->mot_obj)))
                                RETURN(-ENOTDIR);
                }
        }

        memset(policy, 0, sizeof(*policy));
        fid_build_reg_res_name(mdt_object_fid(o), res_id);

        dlmflags = LDLM_FL_ATOMIC_CB;
        if (nonblock)
                dlmflags |= LDLM_FL_BLOCK_NOWAIT;

        /*
         * Take PDO lock on whole directory and build correct @res_id for lock
         * on part of directory.
         */
        if (lh->mlh_pdo_hash != 0) {
                LASSERT(lh->mlh_type == MDT_PDO_LOCK);
                mdt_lock_pdo_mode(info, o, lh);
                if (lh->mlh_pdo_mode != LCK_NL) {
                        /*
                         * Do not use LDLM_FL_LOCAL_ONLY for parallel lock, it
                         * is never going to be sent to client and we do not
                         * want it slowed down due to possible cancels.
                         */
                        policy->l_inodebits.bits = MDS_INODELOCK_UPDATE;
                        rc = mdt_fid_lock(ns, &lh->mlh_pdo_lh, lh->mlh_pdo_mode,
                                          policy, res_id, dlmflags,
                                          info->mti_exp == NULL ? NULL :
                                          &info->mti_exp->exp_handle.h_cookie);
                        if (unlikely(rc))
                                RETURN(rc);
                }

                /*
                 * Finish res_id initializing by name hash marking part of
                 * directory which is taking modification.
                 */
                res_id->name[LUSTRE_RES_ID_HSH_OFF] = lh->mlh_pdo_hash;
        }

        policy->l_inodebits.bits = ibits;

        /*
         * Use LDLM_FL_LOCAL_ONLY for this lock. We do not know yet if it is
         * going to be sent to client. If it is - mdt_intent_policy() path will
         * fix it up and turn FL_LOCAL flag off.
         */
        rc = mdt_fid_lock(ns, &lh->mlh_reg_lh, lh->mlh_reg_mode, policy,
                          res_id, LDLM_FL_LOCAL_ONLY | dlmflags,
                          info->mti_exp == NULL ? NULL :
                          &info->mti_exp->exp_handle.h_cookie);
        if (rc)
                mdt_object_unlock(info, o, lh, 1);
        else if (unlikely(OBD_FAIL_PRECHECK(OBD_FAIL_MDS_PDO_LOCK)) &&
                 lh->mlh_pdo_hash != 0 &&
                 (lh->mlh_reg_mode == LCK_PW || lh->mlh_reg_mode == LCK_EX)) {
                OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_PDO_LOCK, 15);
        }

        RETURN(rc);
}

static int
mdt_object_lock_internal(struct mdt_thread_info *info, struct mdt_object *o,
                         struct mdt_lock_handle *lh, __u64 ibits,
                         bool nonblock, int locality)
{
        int rc;
        ENTRY;

        if (!mdt_object_remote(o))
                return mdt_object_local_lock(info, o, lh, ibits, nonblock,
                                             locality);

        if (locality == MDT_LOCAL_LOCK) {
                CERROR("%s: try to get local lock for remote object"
                       DFID".\n", mdt_obd_name(info->mti_mdt),
                       PFID(mdt_object_fid(o)));
                RETURN(-EPROTO);
        }

        /* XXX do not support PERM/LAYOUT/XATTR lock for remote object yet */
        ibits &= ~(MDS_INODELOCK_PERM | MDS_INODELOCK_LAYOUT |
                   MDS_INODELOCK_XATTR);
        if (ibits & MDS_INODELOCK_UPDATE) {
                /* Sigh, PDO needs to enqueue 2 locks right now, but
                 * enqueue RPC can only request 1 lock, to avoid extra
                 * RPC, so it will instead enqueue EX lock for remote
                 * object anyway XXX*/
                if (lh->mlh_type == MDT_PDO_LOCK &&
                    lh->mlh_pdo_hash != 0) {
                        CDEBUG(D_INFO, "%s: "DFID" convert PDO lock to"
                               "EX lock.\n", mdt_obd_name(info->mti_mdt),
                               PFID(mdt_object_fid(o)));
                        lh->mlh_pdo_hash = 0;
                        lh->mlh_rreg_mode = LCK_EX;
                        lh->mlh_type = MDT_REG_LOCK;
                }
                rc = mdt_remote_object_lock(info, o, mdt_object_fid(o),
                                            &lh->mlh_rreg_lh,
                                            lh->mlh_rreg_mode,
                                            MDS_INODELOCK_UPDATE);
                if (rc != ELDLM_OK)
                        RETURN(rc);
        }

        /* Only enqueue LOOKUP lock for remote object */
        if (ibits & MDS_INODELOCK_LOOKUP) {
                rc = mdt_object_local_lock(info, o, lh,
                                           MDS_INODELOCK_LOOKUP,
                                           nonblock, locality);
                if (rc != ELDLM_OK)
                        RETURN(rc);
        }

        RETURN(0);
}

int mdt_object_lock(struct mdt_thread_info *info, struct mdt_object *o,
                    struct mdt_lock_handle *lh, __u64 ibits, int locality)
{
        return mdt_object_lock_internal(info, o, lh, ibits, false, locality);
}

int mdt_object_lock_try(struct mdt_thread_info *info, struct mdt_object *o,
                        struct mdt_lock_handle *lh, __u64 ibits, int locality)
{
        struct mdt_lock_handle tmp = *lh;
        int rc;

        rc = mdt_object_lock_internal(info, o, &tmp, ibits, true, locality);
        if (rc == 0)
                *lh = tmp;

        return rc == 0;
}

/**
 * Save a lock within request object.
 *
 * Keep the lock referenced until whether client ACK or transaction
 * commit happens or release the lock immediately depending on input
 * parameters. If COS is ON, a write lock is converted to COS lock
 * before saving.
 *
 * \param info thead info object
 * \param h lock handle
 * \param mode lock mode
 * \param decref force immediate lock releasing
 */
static
void mdt_save_lock(struct mdt_thread_info *info, struct lustre_handle *h,
                   ldlm_mode_t mode, int decref)
{
        ENTRY;

        if (lustre_handle_is_used(h)) {
                if (decref || !info->mti_has_trans ||
                    !(mode & (LCK_PW | LCK_EX))){
                        mdt_fid_unlock(h, mode);
                } else {
                        struct mdt_device *mdt = info->mti_mdt;
                        struct ldlm_lock *lock = ldlm_handle2lock(h);
                        struct ptlrpc_request *req = mdt_info_req(info);
                        int no_ack = 0;

                        LASSERTF(lock != NULL, "no lock for cookie "LPX64"\n",
                                 h->cookie);
                        /* there is no request if mdt_object_unlock() is called
                         * from mdt_export_cleanup()->mdt_add_dirty_flag() */
                        if (likely(req != NULL)) {
                                CDEBUG(D_HA, "request = %p reply state = %p"
                                       " transno = "LPD64"\n", req,
                                       req->rq_reply_state, req->rq_transno);
                                if (mdt_cos_is_enabled(mdt)) {
                                        no_ack = 1;
                                        ldlm_lock_downgrade(lock, LCK_COS);
                                        mode = LCK_COS;
                                }
                                ptlrpc_save_lock(req, h, mode, no_ack);
                        } else {
                                ldlm_lock_decref(h, mode);
                        }
                        if (mdt_is_lock_sync(lock)) {
                                CDEBUG(D_HA, "found sync-lock,"
                                       " async commit started\n");
                                mdt_device_commit_async(info->mti_env,
                                                        mdt);
                        }
                        LDLM_LOCK_PUT(lock);
                }
                h->cookie = 0ull;
        }

        EXIT;
}

/**
 * Unlock mdt object.
 *
 * Immeditely release the regular lock and the PDO lock or save the
 * lock in reqeuest and keep them referenced until client ACK or
 * transaction commit.
 *
 * \param info thread info object
 * \param o mdt object
 * \param lh mdt lock handle referencing regular and PDO locks
 * \param decref force immediate lock releasing
 */
void mdt_object_unlock(struct mdt_thread_info *info, struct mdt_object *o,
                       struct mdt_lock_handle *lh, int decref)
{
        ENTRY;

        mdt_save_lock(info, &lh->mlh_pdo_lh, lh->mlh_pdo_mode, decref);
        mdt_save_lock(info, &lh->mlh_reg_lh, lh->mlh_reg_mode, decref);

        if (lustre_handle_is_used(&lh->mlh_rreg_lh))
                ldlm_lock_decref(&lh->mlh_rreg_lh, lh->mlh_rreg_mode);

        EXIT;
}

struct mdt_object *mdt_object_find_lock(struct mdt_thread_info *info,
                                        const struct lu_fid *f,
                                        struct mdt_lock_handle *lh,
                                        __u64 ibits)
{
        struct mdt_object *o;

        o = mdt_object_find(info->mti_env, info->mti_mdt, f);
        if (!IS_ERR(o)) {
                int rc;

                rc = mdt_object_lock(info, o, lh, ibits,
                                     MDT_LOCAL_LOCK);
                if (rc != 0) {
                        mdt_object_put(info->mti_env, o);
                        o = ERR_PTR(rc);
                }
        }
        return o;
}

void mdt_object_unlock_put(struct mdt_thread_info * info,
                           struct mdt_object * o,
                           struct mdt_lock_handle *lh,
                           int decref)
{
        mdt_object_unlock(info, o, lh, decref);
        mdt_object_put(info->mti_env, o);
}

/*
 * Generic code handling requests that have struct mdt_body passed in:
 *
 *  - extract mdt_body from request and save it in @info, if present;
 *
 *  - create lu_object, corresponding to the fid in mdt_body, and save it in
 *  @info;
 *
 *  - if HABEO_CORPUS flag is set for this request type check whether object
 *  actually exists on storage (lu_object_exists()).
 *
 */
static int mdt_body_unpack(struct mdt_thread_info *info, __u32 flags)
{
        const struct mdt_body    *body;
        struct mdt_object        *obj;
        const struct lu_env      *env;
        struct req_capsule       *pill;
        int                       rc;
        ENTRY;

        env = info->mti_env;
        pill = info->mti_pill;

        body = info->mti_body = req_capsule_client_get(pill, &RMF_MDT_BODY);
        if (body == NULL)
                RETURN(-EFAULT);

        if (!(body->valid & OBD_MD_FLID))
                RETURN(0);

        if (!fid_is_sane(&body->fid1)) {
                CERROR("Invalid fid: "DFID"\n", PFID(&body->fid1));
                RETURN(-EINVAL);
        }

        /*
         * Do not get size or any capa fields before we check that request
         * contains capa actually. There are some requests which do not, for
         * instance MDS_IS_SUBDIR.
         */
        if (req_capsule_has_field(pill, &RMF_CAPA1, RCL_CLIENT) &&
            req_capsule_get_size(pill, &RMF_CAPA1, RCL_CLIENT))
                mdt_set_capainfo(info, 0, &body->fid1,
                                 req_capsule_client_get(pill, &RMF_CAPA1));

        obj = mdt_object_find(env, info->mti_mdt, &body->fid1);
        if (!IS_ERR(obj)) {
                if ((flags & HABEO_CORPUS) &&
                    !mdt_object_exists(obj)) {
                        mdt_object_put(env, obj);
                        /* for capability renew ENOENT will be handled in
                         * mdt_renew_capa */
                        if (body->valid & OBD_MD_FLOSSCAPA)
                                rc = 0;
                        else
                                rc = -ENOENT;
                } else {
                        info->mti_object = obj;
                        rc = 0;
                }
        } else
                rc = PTR_ERR(obj);

        RETURN(rc);
}

static int mdt_unpack_req_pack_rep(struct mdt_thread_info *info, __u32 flags)
{
        struct req_capsule *pill = info->mti_pill;
        int rc;
        ENTRY;

        if (req_capsule_has_field(pill, &RMF_MDT_BODY, RCL_CLIENT))
                rc = mdt_body_unpack(info, flags);
        else
                rc = 0;

        if (rc == 0 && (flags & HABEO_REFERO)) {
                /* Pack reply. */
                if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER))
                        req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER,
                                             DEF_REP_MD_SIZE);
                if (req_capsule_has_field(pill, &RMF_LOGCOOKIES, RCL_SERVER))
                        req_capsule_set_size(pill, &RMF_LOGCOOKIES,
                                             RCL_SERVER, 0);

                rc = req_capsule_server_pack(pill);
        }
        RETURN(rc);
}

static int mdt_init_capa_ctxt(const struct lu_env *env, struct mdt_device *m)
{
        struct md_device *next = m->mdt_child;

        return next->md_ops->mdo_init_capa_ctxt(env, next,
                                                m->mdt_lut.lut_mds_capa,
                                                m->mdt_capa_timeout,
                                                m->mdt_capa_alg,
                                                m->mdt_capa_keys);
}

void mdt_lock_handle_init(struct mdt_lock_handle *lh)
{
        lh->mlh_type = MDT_NUL_LOCK;
        lh->mlh_reg_lh.cookie = 0ull;
        lh->mlh_reg_mode = LCK_MINMODE;
        lh->mlh_pdo_lh.cookie = 0ull;
        lh->mlh_pdo_mode = LCK_MINMODE;
        lh->mlh_rreg_lh.cookie = 0ull;
        lh->mlh_rreg_mode = LCK_MINMODE;
}

void mdt_lock_handle_fini(struct mdt_lock_handle *lh)
{
        LASSERT(!lustre_handle_is_used(&lh->mlh_reg_lh));
        LASSERT(!lustre_handle_is_used(&lh->mlh_pdo_lh));
}

/*
 * Initialize fields of struct mdt_thread_info. Other fields are left in
 * uninitialized state, because it's too expensive to zero out whole
 * mdt_thread_info (> 1K) on each request arrival.
 */
void mdt_thread_info_init(struct ptlrpc_request *req,
                          struct mdt_thread_info *info)
{
        int i;

        info->mti_pill = &req->rq_pill;

        /* lock handle */
        for (i = 0; i < ARRAY_SIZE(info->mti_lh); i++)
                mdt_lock_handle_init(&info->mti_lh[i]);

        /* mdt device: it can be NULL while CONNECT */
        if (req->rq_export) {
                info->mti_mdt = mdt_dev(req->rq_export->exp_obd->obd_lu_dev);
                info->mti_exp = req->rq_export;
        } else
                info->mti_mdt = NULL;
        info->mti_env = req->rq_svc_thread->t_env;
        info->mti_transno = lustre_msg_get_transno(req->rq_reqmsg);

        memset(&info->mti_attr, 0, sizeof(info->mti_attr));
        info->mti_big_buf = LU_BUF_NULL;
        info->mti_body = NULL;
        info->mti_object = NULL;
        info->mti_dlm_req = NULL;
        info->mti_has_trans = 0;
        info->mti_cross_ref = 0;
        info->mti_opdata = 0;
        info->mti_big_lmm_used = 0;

        info->mti_spec.no_create = 0;
        info->mti_spec.sp_rm_entry = 0;

        info->mti_spec.u.sp_ea.eadata = NULL;
        info->mti_spec.u.sp_ea.eadatalen = 0;
}

void mdt_thread_info_fini(struct mdt_thread_info *info)
{
        int i;

        if (info->mti_object != NULL) {
                mdt_object_put(info->mti_env, info->mti_object);
                info->mti_object = NULL;
        }

        for (i = 0; i < ARRAY_SIZE(info->mti_lh); i++)
                mdt_lock_handle_fini(&info->mti_lh[i]);
        info->mti_env = NULL;
        info->mti_pill = NULL;
        info->mti_exp = NULL;

        if (unlikely(info->mti_big_buf.lb_buf != NULL))
                lu_buf_free(&info->mti_big_buf);
}

struct mdt_thread_info *tsi2mdt_info(struct tgt_session_info *tsi)
{
        struct mdt_thread_info  *mti;
        struct lustre_capa      *lc;

        mti = mdt_th_info(tsi->tsi_env);
        LASSERT(mti != NULL);

        mdt_thread_info_init(tgt_ses_req(tsi), mti);
        if (tsi->tsi_corpus != NULL) {
                struct req_capsule *pill = tsi->tsi_pill;

                mti->mti_object = mdt_obj(tsi->tsi_corpus);
                lu_object_get(tsi->tsi_corpus);

                /*
                 * XXX: must be part of tgt_mdt_body_unpack but moved here
                 * due to mdt_set_capainfo().
                 */
                if (req_capsule_has_field(pill, &RMF_CAPA1, RCL_CLIENT) &&
                    req_capsule_get_size(pill, &RMF_CAPA1, RCL_CLIENT) > 0) {
                        lc = req_capsule_client_get(pill, &RMF_CAPA1);
                        mdt_set_capainfo(mti, 0, &tsi->tsi_mdt_body->fid1, lc);
                }
        }
        mti->mti_body = tsi->tsi_mdt_body;
        mti->mti_dlm_req = tsi->tsi_dlm_req;

        return mti;
}

static int mdt_tgt_connect(struct tgt_session_info *tsi)
{
        struct ptlrpc_request   *req = tgt_ses_req(tsi);
        int                      rc;

        ENTRY;

        rc = tgt_connect(tsi);
        if (rc != 0)
                RETURN(rc);

        rc = mdt_init_idmap(tsi);
        if (rc != 0)
                GOTO(err, rc);
        RETURN(0);
err:
        obd_disconnect(class_export_get(req->rq_export));
        return rc;
}

enum mdt_it_code {
        MDT_IT_OPEN,
        MDT_IT_OCREAT,
        MDT_IT_CREATE,
        MDT_IT_GETATTR,
        MDT_IT_READDIR,
        MDT_IT_LOOKUP,
        MDT_IT_UNLINK,
        MDT_IT_TRUNC,
        MDT_IT_GETXATTR,
        MDT_IT_LAYOUT,
        MDT_IT_QUOTA,
        MDT_IT_NR
};

static int mdt_intent_getattr(enum mdt_it_code opcode,
                              struct mdt_thread_info *info,
                              struct ldlm_lock **,
                              __u64);

static int mdt_intent_getxattr(enum mdt_it_code opcode,
                                struct mdt_thread_info *info,
                                struct ldlm_lock **lockp,
                                __u64 flags);

static int mdt_intent_layout(enum mdt_it_code opcode,
                             struct mdt_thread_info *info,
                             struct ldlm_lock **,
                             __u64);
static int mdt_intent_reint(enum mdt_it_code opcode,
                            struct mdt_thread_info *info,
                            struct ldlm_lock **,
                            __u64);

static struct mdt_it_flavor {
        const struct req_format *it_fmt;
        __u32                    it_flags;
        int                    (*it_act)(enum mdt_it_code ,
                                         struct mdt_thread_info *,
                                         struct ldlm_lock **,
                                         __u64);
        long                     it_reint;
} mdt_it_flavor[] = {
        [MDT_IT_OPEN]     = {
                .it_fmt   = &RQF_LDLM_INTENT,
                /*.it_flags = HABEO_REFERO,*/
                .it_flags = 0,
                .it_act   = mdt_intent_reint,
                .it_reint = REINT_OPEN
        },
        [MDT_IT_OCREAT]   = {
                .it_fmt   = &RQF_LDLM_INTENT,
                /*
                 * OCREAT is not a MUTABOR request as if the file
                 * already exists.
                 * We do the extra check of OBD_CONNECT_RDONLY in
                 * mdt_reint_open() when we really need to create
                 * the object.
                 */
                .it_flags = 0,
                .it_act   = mdt_intent_reint,
                .it_reint = REINT_OPEN
        },
        [MDT_IT_CREATE]   = {
                .it_fmt   = &RQF_LDLM_INTENT,
                .it_flags = MUTABOR,
                .it_act   = mdt_intent_reint,
                .it_reint = REINT_CREATE
        },
        [MDT_IT_GETATTR]  = {
                .it_fmt   = &RQF_LDLM_INTENT_GETATTR,
                .it_flags = HABEO_REFERO,
                .it_act   = mdt_intent_getattr
        },
        [MDT_IT_READDIR]  = {
                .it_fmt   = NULL,
                .it_flags = 0,
                .it_act   = NULL
        },
        [MDT_IT_LOOKUP]   = {
                .it_fmt   = &RQF_LDLM_INTENT_GETATTR,
                .it_flags = HABEO_REFERO,
                .it_act   = mdt_intent_getattr
        },
        [MDT_IT_UNLINK]   = {
                .it_fmt   = &RQF_LDLM_INTENT_UNLINK,
                .it_flags = MUTABOR,
                .it_act   = NULL,
                .it_reint = REINT_UNLINK
        },
        [MDT_IT_TRUNC]    = {
                .it_fmt   = NULL,
                .it_flags = MUTABOR,
                .it_act   = NULL
        },
        [MDT_IT_GETXATTR] = {
                .it_fmt   = &RQF_LDLM_INTENT_GETXATTR,
                .it_flags = HABEO_CORPUS,
                .it_act   = mdt_intent_getxattr
        },
        [MDT_IT_LAYOUT] = {
                .it_fmt   = &RQF_LDLM_INTENT_LAYOUT,
                .it_flags = 0,
                .it_act   = mdt_intent_layout
        }
};

static int
mdt_intent_lock_replace(struct mdt_thread_info *info, struct ldlm_lock **lockp,
                        struct ldlm_lock *new_lock, struct mdt_lock_handle *lh,
                        __u64 flags)
{
        struct ptlrpc_request  *req = mdt_info_req(info);
        struct ldlm_lock       *lock = *lockp;

        /*
         * Get new lock only for cases when possible resent did not find any
         * lock.
         */
        if (new_lock == NULL)
                new_lock = ldlm_handle2lock_long(&lh->mlh_reg_lh, 0);

        if (new_lock == NULL && (flags & LDLM_FL_INTENT_ONLY)) {
                lh->mlh_reg_lh.cookie = 0;
                RETURN(0);
        }

        LASSERTF(new_lock != NULL,
                 "lockh "LPX64"\n", lh->mlh_reg_lh.cookie);

        /*
         * If we've already given this lock to a client once, then we should
         * have no readers or writers.  Otherwise, we should have one reader
         * _or_ writer ref (which will be zeroed below) before returning the
         * lock to a client.
         */
        if (new_lock->l_export == req->rq_export) {
                LASSERT(new_lock->l_readers + new_lock->l_writers == 0);
        } else {
                LASSERT(new_lock->l_export == NULL);
                LASSERT(new_lock->l_readers + new_lock->l_writers == 1);
        }

        *lockp = new_lock;

        if (new_lock->l_export == req->rq_export) {
                /*
                 * Already gave this to the client, which means that we
                 * reconstructed a reply.
                 */
                LASSERT(lustre_msg_get_flags(req->rq_reqmsg) &
                        MSG_RESENT);
                lh->mlh_reg_lh.cookie = 0;
                RETURN(ELDLM_LOCK_REPLACED);
        }

        /*
         * Fixup the lock to be given to the client.
         */
        lock_res_and_lock(new_lock);
        /* Zero new_lock->l_readers and new_lock->l_writers without triggering
         * possible blocking AST. */
        while (new_lock->l_readers > 0) {
                lu_ref_del(&new_lock->l_reference, "reader", new_lock);
                lu_ref_del(&new_lock->l_reference, "user", new_lock);
                new_lock->l_readers--;
        }
        while (new_lock->l_writers > 0) {
                lu_ref_del(&new_lock->l_reference, "writer", new_lock);
                lu_ref_del(&new_lock->l_reference, "user", new_lock);
                new_lock->l_writers--;
        }

        new_lock->l_export = class_export_lock_get(req->rq_export, new_lock);
        new_lock->l_blocking_ast = lock->l_blocking_ast;
        new_lock->l_completion_ast = lock->l_completion_ast;
        new_lock->l_remote_handle = lock->l_remote_handle;
        new_lock->l_flags &= ~LDLM_FL_LOCAL;

        unlock_res_and_lock(new_lock);

        cfs_hash_add(new_lock->l_export->exp_lock_hash,
                     &new_lock->l_remote_handle,
                     &new_lock->l_exp_hash);

        LDLM_LOCK_RELEASE(new_lock);
        lh->mlh_reg_lh.cookie = 0;

        RETURN(ELDLM_LOCK_REPLACED);
}

static void mdt_intent_fixup_resent(struct mdt_thread_info *info,
                                    struct ldlm_lock *new_lock,
                                    struct ldlm_lock **old_lock,
                                    struct mdt_lock_handle *lh,
                                    enum mdt_it_code opcode)
{
        struct ptlrpc_request  *req = mdt_info_req(info);
        struct obd_export      *exp = req->rq_export;
        struct lustre_handle    remote_hdl;
        struct ldlm_request    *dlmreq;
        struct ldlm_lock       *lock;

        if (!(lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT))
                return;

        dlmreq = req_capsule_client_get(info->mti_pill, &RMF_DLM_REQ);
        remote_hdl = dlmreq->lock_handle[0];
        /* If the client does not require open lock, it does not need to
         * search lock in exp_lock_hash, since the server thread will
         * make sure the lock will be released, and the resend request
         * can always re-enqueue the lock */
        if ((opcode != MDT_IT_OPEN) || (opcode == MDT_IT_OPEN &&
            info->mti_spec.sp_cr_flags & MDS_OPEN_LOCK)) {
                /* In the function below, .hs_keycmp resolves to
                 * ldlm_export_lock_keycmp() */
                /* coverity[overrun-buffer-val] */
                lock = cfs_hash_lookup(exp->exp_lock_hash, &remote_hdl);
                if (lock) {
                        lock_res_and_lock(lock);
                        if (lock != new_lock) {
                                lh->mlh_reg_lh.cookie = lock->l_handle.h_cookie;
                                lh->mlh_reg_mode = lock->l_granted_mode;

                                LDLM_DEBUG(lock, "Restoring lock cookie");
                                DEBUG_REQ(D_DLMTRACE, req,
                                          "restoring lock cookie "LPX64,
                                          lh->mlh_reg_lh.cookie);
                                if (old_lock)
                                        *old_lock = LDLM_LOCK_GET(lock);
                                cfs_hash_put(exp->exp_lock_hash,
                                             &lock->l_exp_hash);
                                unlock_res_and_lock(lock);
                                return;
                        }
                        cfs_hash_put(exp->exp_lock_hash, &lock->l_exp_hash);
                        unlock_res_and_lock(lock);
                }
        }
        /*
         * If the xid matches, then we know this is a resent request, and allow
         * it. (It's probably an OPEN, for which we don't send a lock.
         */
        if (req_xid_is_last(req))
                return;

        /*
         * This remote handle isn't enqueued, so we never received or processed
         * this request.  Clear MSG_RESENT, because it can be handled like any
         * normal request now.
         */
        lustre_msg_clear_flags(req->rq_reqmsg, MSG_RESENT);

        DEBUG_REQ(D_DLMTRACE, req, "no existing lock with rhandle "LPX64,
                  remote_hdl.cookie);
}

static int mdt_intent_getxattr(enum mdt_it_code opcode,
                                struct mdt_thread_info *info,
                                struct ldlm_lock **lockp,
                                __u64 flags)
{
        struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_RMT];
        struct ldlm_reply      *ldlm_rep = NULL;
        int rc, grc;

        /*
         * Initialize lhc->mlh_reg_lh either from a previously granted lock
         * (for the resend case) or a new lock. Below we will use it to
         * replace the original lock.
         */
        mdt_intent_fixup_resent(info, *lockp, NULL, lhc, opcode);
        if (!lustre_handle_is_used(&lhc->mlh_reg_lh)) {
                mdt_lock_reg_init(lhc, (*lockp)->l_req_mode);
                rc = mdt_object_lock(info, info->mti_object, lhc,
                                        MDS_INODELOCK_XATTR,
                                        MDT_LOCAL_LOCK);
                if (rc)
                        return rc;
        }