LU-6231 osp: prepare OUT RPC after remote transaction start

[fs/lustre-release.git] / lustre / osp / osp_object.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) 2014, Intel Corporation.
 */
/*
 * lustre/osp/osp_object.c
 *
 * Lustre OST Proxy Device (OSP) is the agent on the local MDT for the OST
 * or remote MDT.
 *
 * OSP object attributes cache
 * ---------------------------
 * OSP object is the stub of the remote OST-object or MDT-object. Both the
 * attribute and the extended attributes are stored on the peer side remotely.
 * It is inefficient to send RPC to peer to fetch those attributes when every
 * get_attr()/get_xattr() called. For a large system, the LFSCK synchronous
 * mode scanning is prohibitively inefficient.
 *
 * So the OSP maintains the OSP object attributes cache to cache some
 * attributes on the local MDT. The cache is organized against the OSP
 * object as follows:
 *
 * struct osp_xattr_entry {
 *      struct list_head         oxe_list;
 *      atomic_t                 oxe_ref;
 *      void                    *oxe_value;
 *      int                      oxe_buflen;
 *      int                      oxe_namelen;
 *      int                      oxe_vallen;
 *      unsigned int             oxe_exist:1,
 *                               oxe_ready:1;
 *      char                     oxe_buf[0];
 * };
 *
 * struct osp_object_attr {
 *      struct lu_attr          ooa_attr;
 *      struct list_head        ooa_xattr_list;
 * };
 *
 * struct osp_object {
 *      ...
 *      struct osp_object_attr *opo_ooa;
 *      spinlock_t              opo_lock;
 *      ...
 * };
 *
 * The basic attributes, such as owner/mode/flags, are stored in the
 * osp_object_attr::ooa_attr. The extended attributes will be stored
 * as osp_xattr_entry. Every extended attribute has an independent
 * osp_xattr_entry, and all the osp_xattr_entry are linked into the
 * osp_object_attr::ooa_xattr_list. The OSP object attributes cache
 * is protected by the osp_object::opo_lock.
 *
 * Not all OSP objects have an attributes cache because maintaining
 * the cache requires some resources. Currently, the OSP object
 * attributes cache will be initialized when the attributes or the
 * extended attributes are pre-fetched via osp_declare_attr_get()
 * or osp_declare_xattr_get(). That is usually for LFSCK purpose,
 * but it also can be shared by others.
 *
 *
 * XXX: NOT prepare out RPC for remote transaction. ((please refer to the
 *      comment of osp_trans_create() for remote transaction)
 *
 * According to our current transaction/dt_object_lock framework (to make
 * the cross-MDTs modification for DNE1 to be workable), the transaction
 * sponsor will start the transaction firstly, then try to acquire related
 * dt_object_lock if needed. Under such rules, if we want to prepare the
 * OUT RPC in the transaction declare phase, then related attr/xattr
 * should be known without dt_object_lock. But such condition maybe not
 * true for some remote transaction case. For example:
 *
 * For linkEA repairing (by LFSCK) case, before the LFSCK thread obtained
 * the dt_object_lock on the target MDT-object, it cannot know whether
 * the MDT-object has linkEA or not, neither invalid or not.
 *
 * Since the LFSCK thread cannot hold dt_object_lock before the remote
 * transaction start (otherwise there will be some potential deadlock),
 * it cannot prepare related OUT RPC for repairing during the declare
 * phase as other normal transactions do.
 *
 * To resolve the trouble, we will make OSP to prepare related OUT RPC
 * after remote transaction started, and trigger the remote updating
 * (send RPC) when trans_stop. Then the up layer users, such as LFSCK,
 * can follow the general rule to handle trans_start/dt_object_lock
 * for repairing linkEA inconsistency without distinguishing remote
 * MDT-object.
 *
 * In fact, above solution for remote transaction should be the normal
 * model without considering DNE1. The trouble brought by DNE1 will be
 * resolved in DNE2. At that time, this patch can be removed.
 *
 *
 * Author: Alex Zhuravlev <alexey.zhuravlev@intel.com>
 * Author: Mikhail Pershin <mike.tappro@intel.com>
 */

#define DEBUG_SUBSYSTEM S_MDS

#include "osp_internal.h"

static inline __u32 osp_dev2node(struct osp_device *osp)
{
        return osp->opd_storage->dd_lu_dev.ld_site->ld_seq_site->ss_node_id;
}

static inline bool is_ost_obj(struct lu_object *lo)
{
        return !lu2osp_dev(lo->lo_dev)->opd_connect_mdt;
}

/**
 * Assign FID to the OST object.
 *
 * This function will assign the FID to the OST object of a striped file.
 *
 * \param[in] env       pointer to the thread context
 * \param[in] d         pointer to the OSP device
 * \param[in] o         pointer to the OSP object that the FID will be
 *                      assigned to
 */
static void osp_object_assign_fid(const struct lu_env *env,
                                  struct osp_device *d, struct osp_object *o)
{
        struct osp_thread_info *osi = osp_env_info(env);

        LASSERT(fid_is_zero(lu_object_fid(&o->opo_obj.do_lu)));
        LASSERT(o->opo_reserved);
        o->opo_reserved = 0;

        osp_precreate_get_fid(env, d, &osi->osi_fid);

        lu_object_assign_fid(env, &o->opo_obj.do_lu, &osi->osi_fid);
}

/**
 * Initialize the OSP object attributes cache.
 *
 * \param[in] obj       pointer to the OSP object
 *
 * \retval              0 for success
 * \retval              negative error number on failure
 */
static int osp_oac_init(struct osp_object *obj)
{
        struct osp_object_attr *ooa;

        OBD_ALLOC_PTR(ooa);
        if (ooa == NULL)
                return -ENOMEM;

        INIT_LIST_HEAD(&ooa->ooa_xattr_list);
        spin_lock(&obj->opo_lock);
        if (likely(obj->opo_ooa == NULL)) {
                obj->opo_ooa = ooa;
                spin_unlock(&obj->opo_lock);
        } else {
                spin_unlock(&obj->opo_lock);
                OBD_FREE_PTR(ooa);
        }

        return 0;
}

/**
 * Find the named extended attribute in the OSP object attributes cache.
 *
 * The caller should take the osp_object::opo_lock before calling
 * this function.
 *
 * \param[in] ooa       pointer to the OSP object attributes cache
 * \param[in] name      the name of the extended attribute
 * \param[in] namelen   the name length of the extended attribute
 *
 * \retval              pointer to the found extended attribute entry
 * \retval              NULL if the specified extended attribute is not
 *                      in the cache
 */
static struct osp_xattr_entry *
osp_oac_xattr_find_locked(struct osp_object_attr *ooa,
                          const char *name, size_t namelen)
{
        struct osp_xattr_entry *oxe;

        list_for_each_entry(oxe, &ooa->ooa_xattr_list, oxe_list) {
                if (namelen == oxe->oxe_namelen &&
                    strncmp(name, oxe->oxe_buf, namelen) == 0)
                        return oxe;
        }

        return NULL;
}

/**
 * Find the named extended attribute in the OSP object attributes cache.
 *
 * Call osp_oac_xattr_find_locked() with the osp_object::opo_lock held.
 *
 * \param[in] obj       pointer to the OSP object
 * \param[in] name      the name of the extended attribute
 * \param[in] unlink    true if the extended attribute entry is to be removed
 *                      from the cache
 *
 * \retval              pointer to the found extended attribute entry
 * \retval              NULL if the specified extended attribute is not
 *                      in the cache
 */
static struct osp_xattr_entry *osp_oac_xattr_find(struct osp_object *obj,
                                                  const char *name, bool unlink)
{
        struct osp_xattr_entry *oxe = NULL;

        spin_lock(&obj->opo_lock);
        if (obj->opo_ooa != NULL) {
                oxe = osp_oac_xattr_find_locked(obj->opo_ooa, name,
                                                strlen(name));
                if (oxe != NULL) {
                        if (unlink)
                                list_del_init(&oxe->oxe_list);
                        else
                                atomic_inc(&oxe->oxe_ref);
                }
        }
        spin_unlock(&obj->opo_lock);

        return oxe;
}

/**
 * Find the named extended attribute in the OSP object attributes cache.
 *
 * If it is not in the cache, then add an empty entry (that will be
 * filled later) to cache with the given name.
 *
 * \param[in] obj       pointer to the OSP object
 * \param[in] name      the name of the extended attribute
 * \param[in] len       the length of the extended attribute value
 *
 * \retval              pointer to the found or new-created extended
 *                      attribute entry
 * \retval              NULL if the specified extended attribute is not in the
 *                      cache or fail to add new empty entry to the cache.
 */
static struct osp_xattr_entry *
osp_oac_xattr_find_or_add(struct osp_object *obj, const char *name, size_t len)
{
        struct osp_object_attr *ooa     = obj->opo_ooa;
        struct osp_xattr_entry *oxe;
        struct osp_xattr_entry *tmp     = NULL;
        size_t                  namelen = strlen(name);
        size_t                  size    = sizeof(*oxe) + namelen + 1 + len;

        LASSERT(ooa != NULL);

        oxe = osp_oac_xattr_find(obj, name, false);
        if (oxe != NULL)
                return oxe;

        OBD_ALLOC(oxe, size);
        if (unlikely(oxe == NULL))
                return NULL;

        INIT_LIST_HEAD(&oxe->oxe_list);
        oxe->oxe_buflen = size;
        oxe->oxe_namelen = namelen;
        memcpy(oxe->oxe_buf, name, namelen);
        oxe->oxe_value = oxe->oxe_buf + namelen + 1;
        /* One ref is for the caller, the other is for the entry on the list. */
        atomic_set(&oxe->oxe_ref, 2);

        spin_lock(&obj->opo_lock);
        tmp = osp_oac_xattr_find_locked(ooa, name, namelen);
        if (tmp == NULL)
                list_add_tail(&oxe->oxe_list, &ooa->ooa_xattr_list);
        else
                atomic_inc(&tmp->oxe_ref);
        spin_unlock(&obj->opo_lock);

        if (tmp != NULL) {
                OBD_FREE(oxe, size);
                oxe = tmp;
        }

        return oxe;
}

/**
 * Add the given extended attribute to the OSP object attributes cache.
 *
 * If there is an old extended attributed entry with the same name,
 * remove it from the cache and return it via the parameter \a poxe.
 *
 * \param[in] obj       pointer to the OSP object
 * \param[in,out] poxe  double pointer to the OSP object extended attribute
 *                      entry: the new extended attribute entry is transfered
 *                      via such pointer target, and if old the extended
 *                      attribute entry exists, then it will be returned back
 *                      via such pointer target.
 * \param[in] len       the length of the (new) extended attribute value
 *
 * \retval              pointer to the new extended attribute entry
 * \retval              NULL for failure cases.
 */
static struct osp_xattr_entry *
osp_oac_xattr_replace(struct osp_object *obj,
                      struct osp_xattr_entry **poxe, size_t len)
{
        struct osp_object_attr *ooa     = obj->opo_ooa;
        struct osp_xattr_entry *oxe;
        size_t                  namelen = (*poxe)->oxe_namelen;
        size_t                  size    = sizeof(*oxe) + namelen + 1 + len;

        LASSERT(ooa != NULL);

        OBD_ALLOC(oxe, size);
        if (unlikely(oxe == NULL))
                return NULL;

        INIT_LIST_HEAD(&oxe->oxe_list);
        oxe->oxe_buflen = size;
        oxe->oxe_namelen = namelen;
        memcpy(oxe->oxe_buf, (*poxe)->oxe_buf, namelen);
        oxe->oxe_value = oxe->oxe_buf + namelen + 1;
        /* One ref is for the caller, the other is for the entry on the list. */
        atomic_set(&oxe->oxe_ref, 2);

        spin_lock(&obj->opo_lock);
        *poxe = osp_oac_xattr_find_locked(ooa, oxe->oxe_buf, namelen);
        LASSERT(*poxe != NULL);

        list_del_init(&(*poxe)->oxe_list);
        list_add_tail(&oxe->oxe_list, &ooa->ooa_xattr_list);
        spin_unlock(&obj->opo_lock);

        return oxe;
}

/**
 * Release reference from the OSP object extended attribute entry.
 *
 * If it is the last reference, then free the entry.
 *
 * \param[in] oxe       pointer to the OSP object extended attribute entry.
 */
static inline void osp_oac_xattr_put(struct osp_xattr_entry *oxe)
{
        if (atomic_dec_and_test(&oxe->oxe_ref)) {
                LASSERT(list_empty(&oxe->oxe_list));

                OBD_FREE(oxe, oxe->oxe_buflen);
        }
}

/**
 * Parse the OSP object attribute from the RPC reply.
 *
 * If the attribute is valid, then it will be added to the OSP object
 * attributes cache.
 *
 * \param[in] env       pointer to the thread context
 * \param[in] reply     pointer to the RPC reply
 * \param[in] req       pointer to the RPC request
 * \param[out] attr     pointer to buffer to hold the output attribute
 * \param[in] obj       pointer to the OSP object
 * \param[in] index     the index of the attribute buffer in the reply
 *
 * \retval              0 for success
 * \retval              negative error number on failure
 */
static int osp_get_attr_from_reply(const struct lu_env *env,
                                   struct object_update_reply *reply,
                                   struct ptlrpc_request *req,
                                   struct lu_attr *attr,
                                   struct osp_object *obj, int index)
{
        struct osp_thread_info  *osi    = osp_env_info(env);
        struct lu_buf           *rbuf   = &osi->osi_lb2;
        struct obdo             *lobdo  = &osi->osi_obdo;
        struct obdo             *wobdo;
        int                     rc;

        rc = object_update_result_data_get(reply, rbuf, index);
        if (rc < 0)
                return rc;

        wobdo = rbuf->lb_buf;
        if (rbuf->lb_len != sizeof(*wobdo))
                return -EPROTO;

        LASSERT(req != NULL);
        if (ptlrpc_req_need_swab(req))
                lustre_swab_obdo(wobdo);

        lustre_get_wire_obdo(NULL, lobdo, wobdo);
        spin_lock(&obj->opo_lock);
        if (obj->opo_ooa != NULL) {
                la_from_obdo(&obj->opo_ooa->ooa_attr, lobdo, lobdo->o_valid);
                if (attr != NULL)
                        *attr = obj->opo_ooa->ooa_attr;
        } else {
                LASSERT(attr != NULL);

                la_from_obdo(attr, lobdo, lobdo->o_valid);
        }
        spin_unlock(&obj->opo_lock);

        return 0;
}

/**
 * Interpreter function for getting OSP object attribute asynchronously.
 *
 * Called to interpret the result of an async mode RPC for getting the
 * OSP object attribute.
 *
 * \param[in] env       pointer to the thread context
 * \param[in] reply     pointer to the RPC reply
 * \param[in] req       pointer to the RPC request
 * \param[in] obj       pointer to the OSP object
 * \param[out] data     pointer to buffer to hold the output attribute
 * \param[in] index     the index of the attribute buffer in the reply
 * \param[in] rc        the result for handling the RPC
 *
 * \retval              0 for success
 * \retval              negative error number on failure
 */
static int osp_attr_get_interpterer(const struct lu_env *env,
                                    struct object_update_reply *reply,
                                    struct ptlrpc_request *req,
                                    struct osp_object *obj,
                                    void *data, int index, int rc)
{
        struct lu_attr *attr = data;

        LASSERT(obj->opo_ooa != NULL);

        if (rc == 0) {
                osp2lu_obj(obj)->lo_header->loh_attr |= LOHA_EXISTS;
                obj->opo_non_exist = 0;

                return osp_get_attr_from_reply(env, reply, req, NULL, obj,
                                               index);
        } else {
                if (rc == -ENOENT) {
                        osp2lu_obj(obj)->lo_header->loh_attr &= ~LOHA_EXISTS;
                        obj->opo_non_exist = 1;
                }

                spin_lock(&obj->opo_lock);
                attr->la_valid = 0;
                spin_unlock(&obj->opo_lock);
        }

        return 0;
}

/**
 * Implement OSP layer dt_object_operations::do_declare_attr_get() interface.
 *
 * Declare that the caller will get attribute from the specified OST object.
 *
 * This function adds an Object Unified Target (OUT) sub-request to the per-OSP
 * based shared asynchronous request queue. The osp_attr_get_interpterer()
 * is registered as the interpreter function to handle the result of this
 * sub-request.
 *
 * \param[in] env       pointer to the thread context
 * \param[in] dt        pointer to the OSP layer dt_object
 * \param[in] capa      the capability for this operation
 *
 * \retval              0 for success
 * \retval              negative error number on failure
 */
static int osp_declare_attr_get(const struct lu_env *env, struct dt_object *dt,
                                struct lustre_capa *capa)
{
        struct osp_object       *obj    = dt2osp_obj(dt);
        struct osp_device       *osp    = lu2osp_dev(dt->do_lu.lo_dev);
        int                      rc     = 0;

        if (obj->opo_ooa == NULL) {
                rc = osp_oac_init(obj);
                if (rc != 0)
                        return rc;
        }

        mutex_lock(&osp->opd_async_requests_mutex);
        rc = osp_insert_async_request(env, OUT_ATTR_GET, obj, 0, NULL, NULL,
                                      &obj->opo_ooa->ooa_attr,
                                      osp_attr_get_interpterer);
        mutex_unlock(&osp->opd_async_requests_mutex);

        return rc;
}

/**
 * Implement OSP layer dt_object_operations::do_attr_get() interface.
 *
 * Get attribute from the specified MDT/OST object.
 *
 * If the attribute is in the OSP object attributes cache, then return
 * the cached attribute directly. Otherwise it will trigger an OUT RPC
 * to the peer to get the attribute synchronously, if successful, add it
 * to the OSP attributes cache. (\see lustre/osp/osp_trans.c for OUT RPC.)
 *
 * \param[in] env       pointer to the thread context
 * \param[in] dt        pointer to the OSP layer dt_object
 * \param[out] attr     pointer to the buffer to hold the output attribute
 * \param[in] capa      the capability for this operation
 *
 * \retval              0 for success
 * \retval              negative error number on failure
 */
int osp_attr_get(const struct lu_env *env, struct dt_object *dt,
                 struct lu_attr *attr, struct lustre_capa *capa)
{
        struct osp_device               *osp = lu2osp_dev(dt->do_lu.lo_dev);
        struct osp_object               *obj = dt2osp_obj(dt);
        struct dt_device                *dev = &osp->opd_dt_dev;
        struct dt_update_request        *update;
        struct object_update_reply      *reply;
        struct ptlrpc_request           *req = NULL;
        int                             rc = 0;
        ENTRY;

        if (is_ost_obj(&dt->do_lu) && obj->opo_non_exist)
                RETURN(-ENOENT);

        if (obj->opo_ooa != NULL) {
                spin_lock(&obj->opo_lock);
                if (obj->opo_ooa->ooa_attr.la_valid != 0) {
                        *attr = obj->opo_ooa->ooa_attr;
                        spin_unlock(&obj->opo_lock);

                        RETURN(0);
                }
                spin_unlock(&obj->opo_lock);
        }

        update = dt_update_request_create(dev);
        if (IS_ERR(update))
                RETURN(PTR_ERR(update));

        rc = out_attr_get_pack(env, &update->dur_buf,
                               lu_object_fid(&dt->do_lu));
        if (rc != 0) {
                CERROR("%s: Insert update error "DFID": rc = %d\n",
                       dev->dd_lu_dev.ld_obd->obd_name,
                       PFID(lu_object_fid(&dt->do_lu)), rc);

                GOTO(out, rc);
        }

        rc = out_remote_sync(env, osp->opd_obd->u.cli.cl_import, update, &req);
        if (rc != 0) {
                if (rc == -ENOENT) {
                        osp2lu_obj(obj)->lo_header->loh_attr &= ~LOHA_EXISTS;
                        obj->opo_non_exist = 1;
                } else {
                        CERROR("%s:osp_attr_get update error "DFID": rc = %d\n",
                               dev->dd_lu_dev.ld_obd->obd_name,
                               PFID(lu_object_fid(&dt->do_lu)), rc);
                }

                GOTO(out, rc);
        }

        osp2lu_obj(obj)->lo_header->loh_attr |= LOHA_EXISTS;
        obj->opo_non_exist = 0;
        reply = req_capsule_server_sized_get(&req->rq_pill,
                                             &RMF_OUT_UPDATE_REPLY,
                                             OUT_UPDATE_REPLY_SIZE);
        if (reply == NULL || reply->ourp_magic != UPDATE_REPLY_MAGIC)
                GOTO(out, rc = -EPROTO);

        rc = osp_get_attr_from_reply(env, reply, req, attr, obj, 0);
        if (rc != 0)
                GOTO(out, rc);

        GOTO(out, rc = 0);

out:
        if (req != NULL)
                ptlrpc_req_finished(req);

        dt_update_request_destroy(update);

        return rc;
}

static int __osp_attr_set(const struct lu_env *env, struct dt_object *dt,
                          const struct lu_attr *attr, struct thandle *th)
{
        struct osp_device       *d = lu2osp_dev(dt->do_lu.lo_dev);
        struct osp_object       *o = dt2osp_obj(dt);
        struct lu_attr          *la;
        int                      rc = 0;
        ENTRY;

        /*
         * Usually we don't allow server stack to manipulate size
         * but there is a special case when striping is created
         * late, after stripeless file got truncated to non-zero.
         *
         * In this case we do the following:
         *
         * 1) grab id in declare - this can lead to leaked OST objects
         *    but we don't currently have proper mechanism and the only
         *    options we have are to do truncate RPC holding transaction
         *    open (very bad) or to grab id in declare at cost of leaked
         *    OST object in same very rare unfortunate case (just bad)
         *    notice 1.6-2.0 do assignment outside of running transaction
         *    all the time, meaning many more chances for leaked objects.
         *
         * 2) send synchronous truncate RPC with just assigned id
         */

        /* there are few places in MDD code still passing NULL
         * XXX: to be fixed soon */
        if (attr == NULL)
                RETURN(0);

        if (attr->la_valid & LA_SIZE && attr->la_size > 0 &&
            fid_is_zero(lu_object_fid(&o->opo_obj.do_lu))) {
                LASSERT(!dt_object_exists(dt));
                osp_object_assign_fid(env, d, o);
                rc = osp_object_truncate(env, dt, attr->la_size);
                if (rc)
                        RETURN(rc);
        }

        if (!(attr->la_valid & (LA_UID | LA_GID)))
                RETURN(0);

        if (!is_only_remote_trans(th)) {
                /*
                 * track all UID/GID changes via llog
                 */
                rc = osp_sync_declare_add(env, o, MDS_SETATTR64_REC, th);
        } else {
                /* It is for OST-object attr_set directly without updating
                 * local MDT-object attribute. It is usually used by LFSCK. */
                rc = __osp_md_attr_set(env, dt, attr, th);
        }

        if (rc != 0 || o->opo_ooa == NULL)
                RETURN(rc);

        /* Update the OSP object attributes cache. */
        la = &o->opo_ooa->ooa_attr;
        spin_lock(&o->opo_lock);
        if (attr->la_valid & LA_UID) {
                la->la_uid = attr->la_uid;
                la->la_valid |= LA_UID;
        }

        if (attr->la_valid & LA_GID) {
                la->la_gid = attr->la_gid;
                la->la_valid |= LA_GID;
        }
        spin_unlock(&o->opo_lock);

        RETURN(0);
}

/**
 * Implement OSP layer dt_object_operations::do_declare_attr_set() interface.
 *
 * If the transaction is not remote one, then declare the credits that will
 * be used for the subsequent llog record for the object's attributes.
 *
 * \param[in] env       pointer to the thread context
 * \param[in] dt        pointer to the OSP layer dt_object
 * \param[in] attr      pointer to the attribute to be set
 * \param[in] th        pointer to the transaction handler
 *
 * \retval              0 for success
 * \retval              negative error number on failure
 */
static int osp_declare_attr_set(const struct lu_env *env, struct dt_object *dt,
                                const struct lu_attr *attr, struct thandle *th)
{
        int rc = 0;

        if (!is_only_remote_trans(th)) {
                rc = __osp_attr_set(env, dt, attr, th);

                CDEBUG(D_INFO, "declare set attr "DFID": rc = %d\n",
                       PFID(&dt->do_lu.lo_header->loh_fid), rc);
        }

        return rc;
}

/**
 * Implement OSP layer dt_object_operations::do_attr_set() interface.
 *
 * Set attribute to the specified OST object.
 *
 * If the transaction is a remote one, then add OUT_ATTR_SET sub-request
 * in the OUT RPC that will be flushed when the remote transaction stop.
 * Otherwise, it will generate a MDS_SETATTR64_REC record in the llog that
 * will be handled by a dedicated thread asynchronously.
 *
 * If the attribute entry exists in the OSP object attributes cache,
 * then update the cached attribute according to given attribute.
 *
 * \param[in] env       pointer to the thread context
 * \param[in] dt        pointer to the OSP layer dt_object
 * \param[in] attr      pointer to the attribute to be set
 * \param[in] th        pointer to the transaction handler
 * \param[in] capa      the capability for this operation
 *
 * \retval              0 for success
 * \retval              negative error number on failure
 */
static int osp_attr_set(const struct lu_env *env, struct dt_object *dt,
                        const struct lu_attr *attr, struct thandle *th,
                        struct lustre_capa *capa)
{
        struct osp_object       *o = dt2osp_obj(dt);
        int                      rc = 0;
        ENTRY;

        if (is_only_remote_trans(th)) {
                rc = __osp_attr_set(env, dt, attr, th);

                CDEBUG(D_INFO, "(1) set attr "DFID": rc = %d\n",
                       PFID(&dt->do_lu.lo_header->loh_fid), rc);

                RETURN(rc);
        }

        /* we're interested in uid/gid changes only */
        if (!(attr->la_valid & (LA_UID | LA_GID)))
                RETURN(0);

        rc = osp_sync_add(env, o, MDS_SETATTR64_REC, th, attr);
        /* XXX: send new uid/gid to OST ASAP? */

        CDEBUG(D_INFO, "(2) set attr "DFID": rc = %d\n",
               PFID(&dt->do_lu.lo_header->loh_fid), rc);

        RETURN(rc);
}

/**
 * Interpreter function for getting OSP object extended attribute asynchronously
 *
 * Called to interpret the result of an async mode RPC for getting the
 * OSP object extended attribute.
 *
 * \param[in] env       pointer to the thread context
 * \param[in] reply     pointer to the RPC reply
 * \param[in] req       pointer to the RPC request
 * \param[in] obj       pointer to the OSP object
 * \param[out] data     pointer to OSP object attributes cache
 * \param[in] index     the index of the attribute buffer in the reply
 * \param[in] rc        the result for handling the RPC
 *
 * \retval              0 for success
 * \retval              negative error number on failure
 */
static int osp_xattr_get_interpterer(const struct lu_env *env,
                                     struct object_update_reply *reply,
                                     struct ptlrpc_request *req,
                                     struct osp_object *obj,
                                     void *data, int index, int rc)
{
        struct osp_object_attr  *ooa  = obj->opo_ooa;
        struct osp_xattr_entry  *oxe  = data;
        struct lu_buf           *rbuf = &osp_env_info(env)->osi_lb2;

        LASSERT(ooa != NULL);

        if (rc == 0) {
                size_t len = sizeof(*oxe) + oxe->oxe_namelen + 1;

                rc = object_update_result_data_get(reply, rbuf, index);
                if (rc < 0 || rbuf->lb_len > (oxe->oxe_buflen - len)) {
                        spin_lock(&obj->opo_lock);
                        oxe->oxe_ready = 0;
                        spin_unlock(&obj->opo_lock);
                        osp_oac_xattr_put(oxe);

                        return rc < 0 ? rc : -ERANGE;
                }

                spin_lock(&obj->opo_lock);
                oxe->oxe_vallen = rbuf->lb_len;
                memcpy(oxe->oxe_value, rbuf->lb_buf, rbuf->lb_len);
                oxe->oxe_exist = 1;
                oxe->oxe_ready = 1;
                spin_unlock(&obj->opo_lock);
        } else if (rc == -ENOENT || rc == -ENODATA) {
                spin_lock(&obj->opo_lock);
                oxe->oxe_exist = 0;
                oxe->oxe_ready = 1;
                spin_unlock(&obj->opo_lock);
        } else {
                spin_lock(&obj->opo_lock);
                oxe->oxe_ready = 0;
                spin_unlock(&obj->opo_lock);
        }

        osp_oac_xattr_put(oxe);

        return 0;
}

/**
 * Implement OSP dt_object_operations::do_declare_xattr_get() interface.
 *
 * Declare that the caller will get extended attribute from the specified
 * OST object.
 *
 * This function will add an OUT_XATTR_GET sub-request to the per OSP
 * based shared asynchronous request queue with the interpreter function:
 * osp_xattr_get_interpterer().
 *
 * \param[in] env       pointer to the thread context
 * \param[in] dt        pointer to the OSP layer dt_object
 * \param[out] buf      pointer to the lu_buf to hold the extended attribute
 * \param[in] name      the name for the expected extended attribute
 * \param[in] capa      the capability for this operation
 *
 * \retval              0 for success
 * \retval              negative error number on failure
 */
static int osp_declare_xattr_get(const struct lu_env *env, struct dt_object *dt,
                                 struct lu_buf *buf, const char *name,
                                 struct lustre_capa *capa)
{
        struct osp_object       *obj     = dt2osp_obj(dt);
        struct osp_device       *osp     = lu2osp_dev(dt->do_lu.lo_dev);
        struct osp_xattr_entry  *oxe;
        __u16                    namelen = strlen(name);
        int                      rc      = 0;

        LASSERT(buf != NULL);
        LASSERT(name != NULL);

        /* If only for xattr size, return directly. */
        if (unlikely(buf->lb_len == 0))
                return 0;

        if (obj->opo_ooa == NULL) {
                rc = osp_oac_init(obj);
                if (rc != 0)
                        return rc;
        }

        oxe = osp_oac_xattr_find_or_add(obj, name, buf->lb_len);
        if (oxe == NULL)
                return -ENOMEM;

        mutex_lock(&osp->opd_async_requests_mutex);
        rc = osp_insert_async_request(env, OUT_XATTR_GET, obj, 1,
                                      &namelen, (const void **)&name, oxe,
                                      osp_xattr_get_interpterer);
        if (rc != 0) {
                mutex_unlock(&osp->opd_async_requests_mutex);
                osp_oac_xattr_put(oxe);
        } else {
                struct dt_update_request *update;

                /* XXX: Currently, we trigger the batched async OUT
                 *      RPC via dt_declare_xattr_get(). It is not
                 *      perfect solution, but works well now.
                 *
                 *      We will improve it in the future. */
                update = osp->opd_async_requests;
                if (update != NULL && update->dur_buf.ub_req != NULL &&
                    update->dur_buf.ub_req->ourq_count > 0) {
                        osp->opd_async_requests = NULL;
                        mutex_unlock(&osp->opd_async_requests_mutex);
                        rc = osp_unplug_async_request(env, osp, update);
                } else {
                        mutex_unlock(&osp->opd_async_requests_mutex);
                }
        }

        return rc;
}

/**
 * Implement OSP layer dt_object_operations::do_xattr_get() interface.
 *
 * Get extended attribute from the specified MDT/OST object.
 *
 * If the extended attribute is in the OSP object attributes cache, then
 * return the cached extended attribute directly. Otherwise it will get
 * the extended attribute synchronously, if successful, add it to the OSP
 * attributes cache. (\see lustre/osp/osp_trans.c for OUT RPC.)
 *
 * There is a race condition: some other thread has added the named extended
 * attributed entry to the OSP object attributes cache during the current
 * OUT_XATTR_GET handling. If such case happens, the OSP will replace the
 * (just) existing extended attribute entry with the new replied one.
 *
 * \param[in] env       pointer to the thread context
 * \param[in] dt        pointer to the OSP layer dt_object
 * \param[out] buf      pointer to the lu_buf to hold the extended attribute
 * \param[in] name      the name for the expected extended attribute
 * \param[in] capa      the capability for this operation
 *
 * \retval              0 for success
 * \retval              negative error number on failure
 */
int osp_xattr_get(const struct lu_env *env, struct dt_object *dt,
                  struct lu_buf *buf, const char *name,
                  struct lustre_capa *capa)
{
        struct osp_device       *osp    = lu2osp_dev(dt->do_lu.lo_dev);
        struct osp_object       *obj    = dt2osp_obj(dt);
        struct dt_device        *dev    = &osp->opd_dt_dev;
        struct lu_buf           *rbuf   = &osp_env_info(env)->osi_lb2;
        struct dt_update_request *update = NULL;
        struct ptlrpc_request   *req    = NULL;
        struct object_update_reply *reply;
        struct osp_xattr_entry  *oxe    = NULL;
        const char              *dname  = dt->do_lu.lo_dev->ld_obd->obd_name;
        int                      rc     = 0;
        ENTRY;

        LASSERT(buf != NULL);
        LASSERT(name != NULL);

        if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_NETWORK) &&
            osp->opd_index == cfs_fail_val) {
                if (is_ost_obj(&dt->do_lu)) {
                        if (osp_dev2node(osp) == cfs_fail_val)
                                RETURN(-ENOTCONN);
                } else {
                        if (strcmp(name, XATTR_NAME_LINK) == 0)
                                RETURN(-ENOTCONN);
                }
        }

        if (unlikely(obj->opo_non_exist))
                RETURN(-ENOENT);

        oxe = osp_oac_xattr_find(obj, name, false);
        if (oxe != NULL) {
                spin_lock(&obj->opo_lock);
                if (oxe->oxe_ready) {
                        if (!oxe->oxe_exist)
                                GOTO(unlock, rc = -ENODATA);

                        if (buf->lb_buf == NULL)
                                GOTO(unlock, rc = oxe->oxe_vallen);

                        if (buf->lb_len < oxe->oxe_vallen)
                                GOTO(unlock, rc = -ERANGE);

                        memcpy(buf->lb_buf, oxe->oxe_value, oxe->oxe_vallen);

                        GOTO(unlock, rc = oxe->oxe_vallen);

unlock:
                        spin_unlock(&obj->opo_lock);
                        osp_oac_xattr_put(oxe);

                        return rc;
                }
                spin_unlock(&obj->opo_lock);
        }

        update = dt_update_request_create(dev);
        if (IS_ERR(update))
                GOTO(out, rc = PTR_ERR(update));

        rc = out_xattr_get_pack(env, &update->dur_buf,
                                lu_object_fid(&dt->do_lu), name);
        if (rc != 0) {
                CERROR("%s: Insert update error "DFID": rc = %d\n",
                       dname, PFID(lu_object_fid(&dt->do_lu)), rc);
                GOTO(out, rc);
        }

        rc = out_remote_sync(env, osp->opd_obd->u.cli.cl_import, update, &req);
        if (rc != 0) {
                if (rc == -ENOENT) {
                        dt->do_lu.lo_header->loh_attr &= ~LOHA_EXISTS;
                        obj->opo_non_exist = 1;
                }

                if (obj->opo_ooa == NULL)
                        GOTO(out, rc);

                if (oxe == NULL)
                        oxe = osp_oac_xattr_find_or_add(obj, name, buf->lb_len);

                if (oxe == NULL) {
                        CWARN("%s: Fail to add xattr (%s) to cache for "
                              DFID" (1): rc = %d\n", dname, name,
                              PFID(lu_object_fid(&dt->do_lu)), rc);

                        GOTO(out, rc);
                }

                spin_lock(&obj->opo_lock);
                if (rc == -ENOENT || rc == -ENODATA) {
                        oxe->oxe_exist = 0;
                        oxe->oxe_ready = 1;
                } else {
                        oxe->oxe_ready = 0;
                }
                spin_unlock(&obj->opo_lock);

                GOTO(out, rc);
        }

        reply = req_capsule_server_sized_get(&req->rq_pill,
                                             &RMF_OUT_UPDATE_REPLY,
                                             OUT_UPDATE_REPLY_SIZE);
        if (reply->ourp_magic != UPDATE_REPLY_MAGIC) {
                CERROR("%s: Wrong version %x expected %x "DFID": rc = %d\n",
                       dname, reply->ourp_magic, UPDATE_REPLY_MAGIC,
                       PFID(lu_object_fid(&dt->do_lu)), -EPROTO);

                GOTO(out, rc = -EPROTO);
        }

        rc = object_update_result_data_get(reply, rbuf, 0);
        if (rc < 0)
                GOTO(out, rc);

        if (buf->lb_buf == NULL)
                GOTO(out, rc = rbuf->lb_len);

        if (unlikely(buf->lb_len < rbuf->lb_len))
                GOTO(out, rc = -ERANGE);

        memcpy(buf->lb_buf, rbuf->lb_buf, rbuf->lb_len);
        rc = rbuf->lb_len;
        if (obj->opo_ooa == NULL)
                GOTO(out, rc);

        if (oxe == NULL) {
                oxe = osp_oac_xattr_find_or_add(obj, name, rbuf->lb_len);
                if (oxe == NULL) {
                        CWARN("%s: Fail to add xattr (%s) to "
                              "cache for "DFID" (2): rc = %d\n",
                              dname, name, PFID(lu_object_fid(&dt->do_lu)), rc);

                        GOTO(out, rc);
                }
        }

        if (oxe->oxe_buflen - oxe->oxe_namelen - 1 < rbuf->lb_len) {
                struct osp_xattr_entry *old = oxe;
                struct osp_xattr_entry *tmp;

                tmp = osp_oac_xattr_replace(obj, &old, rbuf->lb_len);
                osp_oac_xattr_put(oxe);
                oxe = tmp;
                if (tmp == NULL) {
                        CWARN("%s: Fail to update xattr (%s) to "
                              "cache for "DFID": rc = %d\n",
                              dname, name, PFID(lu_object_fid(&dt->do_lu)), rc);
                        spin_lock(&obj->opo_lock);
                        old->oxe_ready = 0;
                        spin_unlock(&obj->opo_lock);

                        GOTO(out, rc);
                }

                /* Drop the ref for entry on list. */
                osp_oac_xattr_put(old);
        }

        spin_lock(&obj->opo_lock);
        oxe->oxe_vallen = rbuf->lb_len;
        memcpy(oxe->oxe_value, rbuf->lb_buf, rbuf->lb_len);
        oxe->oxe_exist = 1;
        oxe->oxe_ready = 1;
        spin_unlock(&obj->opo_lock);

        GOTO(out, rc);

out:
        if (req != NULL)
                ptlrpc_req_finished(req);

        if (update != NULL && !IS_ERR(update))
                dt_update_request_destroy(update);

        if (oxe != NULL)
                osp_oac_xattr_put(oxe);

        return rc;
}

static int __osp_xattr_set(const struct lu_env *env, struct dt_object *dt,
                           const struct lu_buf *buf, const char *name,
                           int flag, struct thandle *th)
{
        struct osp_object       *o = dt2osp_obj(dt);
        struct dt_update_request *update;
        struct osp_xattr_entry  *oxe;
        int                     rc;
        ENTRY;

        LASSERT(buf->lb_len > 0 && buf->lb_buf != NULL);

        update = dt_update_request_find_or_create(th, dt);
        if (IS_ERR(update)) {
                CERROR("%s: Get OSP update buf failed "DFID": rc = %d\n",
                       dt->do_lu.lo_dev->ld_obd->obd_name,
                       PFID(lu_object_fid(&dt->do_lu)),
                       (int)PTR_ERR(update));

                RETURN(PTR_ERR(update));
        }

        rc = out_xattr_set_pack(env, &update->dur_buf,
                                lu_object_fid(&dt->do_lu),
                                buf, name, flag, update->dur_batchid);
        if (rc != 0 || o->opo_ooa == NULL)
                RETURN(rc);

        oxe = osp_oac_xattr_find_or_add(o, name, buf->lb_len);
        if (oxe == NULL) {
                CWARN("%s: cannot cache xattr '%s' of "DFID"\n",
                      dt->do_lu.lo_dev->ld_obd->obd_name,
                      name, PFID(lu_object_fid(&dt->do_lu)));

                RETURN(0);
        }

        if (oxe->oxe_buflen - oxe->oxe_namelen - 1 < buf->lb_len) {
                struct osp_xattr_entry *old = oxe;
                struct osp_xattr_entry *tmp;

                tmp = osp_oac_xattr_replace(o, &old, buf->lb_len);
                osp_oac_xattr_put(oxe);
                oxe = tmp;
                if (tmp == NULL) {
                        CWARN("%s: cannot update cached xattr '%s' of "DFID"\n",
                              dt->do_lu.lo_dev->ld_obd->obd_name,
                              name, PFID(lu_object_fid(&dt->do_lu)));
                        spin_lock(&o->opo_lock);
                        old->oxe_ready = 0;
                        spin_unlock(&o->opo_lock);

                        RETURN(0);
                }

                /* Drop the ref for entry on list. */
                osp_oac_xattr_put(old);
        }

        spin_lock(&o->opo_lock);
        oxe->oxe_vallen = buf->lb_len;
        memcpy(oxe->oxe_value, buf->lb_buf, buf->lb_len);
        oxe->oxe_exist = 1;
        oxe->oxe_ready = 1;
        spin_unlock(&o->opo_lock);
        osp_oac_xattr_put(oxe);

        RETURN(0);
}

/**
 * Implement OSP layer dt_object_operations::do_declare_xattr_set() interface.
 *
 * Declare that the caller will set extended attribute to the specified
 * MDT/OST object.
 *
 * If it is non-remote transaction, it will add an OUT_XATTR_SET sub-request
 * to the OUT RPC that will be flushed when the transaction start. And if the
 * OSP attributes cache is initialized, then check whether the name extended
 * attribute entry exists in the cache or not. If yes, replace it; otherwise,
 * add the extended attribute to the cache.
 *
 * \param[in] env       pointer to the thread context
 * \param[in] dt        pointer to the OSP layer dt_object
 * \param[in] buf       pointer to the lu_buf to hold the extended attribute
 * \param[in] name      the name of the extended attribute to be set
 * \param[in] flag      to indicate the detailed set operation: LU_XATTR_CREATE
 *                      or LU_XATTR_REPLACE or others
 * \param[in] th        pointer to the transaction handler
 *
 * \retval              0 for success
 * \retval              negative error number on failure
 */
int osp_declare_xattr_set(const struct lu_env *env, struct dt_object *dt,
                          const struct lu_buf *buf, const char *name,
                          int flag, struct thandle *th)
{
        int rc = 0;

        if (!is_only_remote_trans(th)) {
                rc = __osp_xattr_set(env, dt, buf, name, flag, th);

                CDEBUG(D_INFO, "declare xattr %s set object "DFID": rc = %d\n",
                       name, PFID(&dt->do_lu.lo_header->loh_fid), rc);
        }

        return rc;
}

/**
 * Implement OSP layer dt_object_operations::do_xattr_set() interface.
 *
 * Set extended attribute to the specified MDT/OST object.
 *
 * If it is remote transaction, it will add an OUT_XATTR_SET sub-request into
 * the OUT RPC that will be flushed when the transaction stop. And if the OSP
 * attributes cache is initialized, then check whether the name extended
 * attribute entry exists in the cache or not. If yes, replace it; otherwise,
 * add the extended attribute to the cache.
 *
 * \param[in] env       pointer to the thread context
 * \param[in] dt        pointer to the OSP layer dt_object
 * \param[in] buf       pointer to the lu_buf to hold the extended attribute
 * \param[in] name      the name of the extended attribute to be set
 * \param[in] fl        to indicate the detailed set operation: LU_XATTR_CREATE
 *                      or LU_XATTR_REPLACE or others
 * \param[in] th        pointer to the transaction handler
 * \param[in] capa      the capability for this operation
 *
 * \retval              0 for success
 * \retval              negative error number on failure
 */
int osp_xattr_set(const struct lu_env *env, struct dt_object *dt,
                  const struct lu_buf *buf, const char *name, int fl,
                  struct thandle *th, struct lustre_capa *capa)
{
        int rc = 0;

        if (is_only_remote_trans(th)) {
                rc = __osp_xattr_set(env, dt, buf, name, fl, th);

                CDEBUG(D_INFO, "xattr %s set object "DFID": rc = %d\n",
                       name, PFID(&dt->do_lu.lo_header->loh_fid), rc);
        }

        return rc;
}

static int __osp_xattr_del(const struct lu_env *env, struct dt_object *dt,
                           const char *name, struct thandle *th)
{
        struct dt_update_request *update;
        const struct lu_fid      *fid;
        struct osp_object        *o     = dt2osp_obj(dt);
        struct osp_xattr_entry   *oxe;
        int                       rc;

        update = dt_update_request_find_or_create(th, dt);
        if (IS_ERR(update))
                return PTR_ERR(update);

        fid = lu_object_fid(&dt->do_lu);

        rc = out_xattr_del_pack(env, &update->dur_buf, fid, name,
                                update->dur_batchid);

        if (rc != 0 || o->opo_ooa == NULL)
                return rc;

        oxe = osp_oac_xattr_find(o, name, true);
        if (oxe != NULL)
                /* Drop the ref for entry on list. */
                osp_oac_xattr_put(oxe);

        return 0;
}

/**
 * Implement OSP layer dt_object_operations::do_declare_xattr_del() interface.
 *
 * Declare that the caller will delete extended attribute on the specified
 * MDT/OST object.
 *
 * If it is non-remote transaction, it will add an OUT_XATTR_DEL sub-request
 * to the OUT RPC that will be flushed when the transaction start. And if the
 * name extended attribute entry exists in the OSP attributes cache, then remove
 * it from the cache.
 *
 * \param[in] env       pointer to the thread context
 * \param[in] dt        pointer to the OSP layer dt_object
 * \param[in] name      the name of the extended attribute to be set
 * \param[in] th        pointer to the transaction handler
 *
 * \retval              0 for success
 * \retval              negative error number on failure
 */
int osp_declare_xattr_del(const struct lu_env *env, struct dt_object *dt,
                          const char *name, struct thandle *th)
{
        int rc = 0;

        if (!is_only_remote_trans(th)) {
                rc = __osp_xattr_del(env, dt, name, th);

                CDEBUG(D_INFO, "declare xattr %s del object "DFID": rc = %d\n",
                       name, PFID(&dt->do_lu.lo_header->loh_fid), rc);
        }

        return rc;
}

/**
 * Implement OSP layer dt_object_operations::do_xattr_del() interface.
 *
 * Delete extended attribute on the specified MDT/OST object.
 *
 * If it is remote transaction, it will add an OUT_XATTR_DEL sub-request into
 * the OUT RPC that will be flushed when the transaction stop. And if the name
 * extended attribute entry exists in the OSP attributes cache, then remove it
 * from the cache.
 *
 * \param[in] env       pointer to the thread context
 * \param[in] dt        pointer to the OSP layer dt_object
 * \param[in] name      the name of the extended attribute to be set
 * \param[in] th        pointer to the transaction handler
 * \param[in] capa      the capability for this operation
 *
 * \retval              0 for success
 * \retval              negative error number on failure
 */
int osp_xattr_del(const struct lu_env *env, struct dt_object *dt,
                  const char *name, struct thandle *th,
                  struct lustre_capa *capa)
{
        int rc = 0;

        if (is_only_remote_trans(th)) {
                rc = __osp_xattr_del(env, dt, name, th);

                CDEBUG(D_INFO, "xattr %s del object "DFID": rc = %d\n",
                       name, PFID(&dt->do_lu.lo_header->loh_fid), rc);
        }

        return rc;
}

/**
 * Implement OSP layer dt_object_operations::do_declare_create() interface.
 *
 * Declare that the caller will create the OST object.
 *
 * If the transaction is a remote transaction and the FID for the OST-object
 * has been assigned already, then handle it as creating (remote) MDT object
 * via osp_md_declare_object_create(). This function is usually used for LFSCK
 * to re-create the lost OST object. Otherwise, if it is not replay case, the
 * OSP will reserve pre-created object for the subsequent create operation;
 * if the MDT side cached pre-created objects are less than some threshold,
 * then it will wakeup the pre-create thread.
 *
 * \param[in] env       pointer to the thread context
 * \param[in] dt        pointer to the OSP layer dt_object
 * \param[in] attr      the attribute for the object to be created
 * \param[in] hint      pointer to the hint for creating the object, such as
 *                      the parent object
 * \param[in] dof       pointer to the dt_object_format for help the creation
 * \param[in] th        pointer to the transaction handler
 *
 * \retval              0 for success
 * \retval              negative error number on failure
 */
static int osp_declare_object_create(const struct lu_env *env,
                                     struct dt_object *dt,
                                     struct lu_attr *attr,
                                     struct dt_allocation_hint *hint,
                                     struct dt_object_format *dof,
                                     struct thandle *th)
{
        struct osp_thread_info  *osi = osp_env_info(env);
        struct osp_device       *d = lu2osp_dev(dt->do_lu.lo_dev);
        struct osp_object       *o = dt2osp_obj(dt);
        const struct lu_fid     *fid = lu_object_fid(&dt->do_lu);
        int                      rc = 0;

        ENTRY;

        if (is_only_remote_trans(th) && !fid_is_zero(fid)) {
                LASSERT(fid_is_sane(fid));

                rc = osp_md_declare_object_create(env, dt, attr, hint, dof, th);

                RETURN(rc);
        }

        /* should happen to non-0 OSP only so that at least one object
         * has been already declared in the scenario and LOD should
         * cleanup that */
        if (OBD_FAIL_CHECK(OBD_FAIL_MDS_OSC_CREATE_FAIL) && d->opd_index == 1)
                RETURN(-ENOSPC);

        LASSERT(d->opd_last_used_oid_file);

        /*
         * There can be gaps in precreated ids and record to unlink llog
         * XXX: we do not handle gaps yet, implemented before solution
         *      was found to be racy, so we disabled that. there is no
         *      point in making useless but expensive llog declaration.
         */
        /* rc = osp_sync_declare_add(env, o, MDS_UNLINK64_REC, th); */

        if (unlikely(!fid_is_zero(fid))) {
                /* replay case: caller knows fid */
                osi->osi_off = sizeof(osi->osi_id) * d->opd_index;
                osi->osi_lb.lb_len = sizeof(osi->osi_id);
                osi->osi_lb.lb_buf = NULL;
                rc = dt_declare_record_write(env, d->opd_last_used_oid_file,
                                             &osi->osi_lb, osi->osi_off, th);
                RETURN(rc);
        }

        /*
         * in declaration we need to reserve object so that we don't block
         * awaiting precreation RPC to complete
         */
        rc = osp_precreate_reserve(env, d);
        /*
         * we also need to declare update to local "last used id" file for
         * recovery if object isn't used for a reason, we need to release
         * reservation, this can be made in osd_object_release()
         */
        if (rc == 0) {
                /* mark id is reserved: in create we don't want to talk
                 * to OST */
                LASSERT(o->opo_reserved == 0);
                o->opo_reserved = 1;

                /* common for all OSPs file hystorically */
                osi->osi_off = sizeof(osi->osi_id) * d->opd_index;
                osi->osi_lb.lb_len = sizeof(osi->osi_id);
                osi->osi_lb.lb_buf = NULL;
                rc = dt_declare_record_write(env, d->opd_last_used_oid_file,
                                             &osi->osi_lb, osi->osi_off, th);
        } else {
                /* not needed in the cache anymore */
                set_bit(LU_OBJECT_HEARD_BANSHEE,
                            &dt->do_lu.lo_header->loh_flags);
        }
        RETURN(rc);
}

/**
 * Implement OSP layer dt_object_operations::do_create() interface.
 *
 * Create the OST object.
 *
 * If the transaction is a remote transaction and the FID for the OST-object
 * has been assigned already, then handle it as handling MDT object via the
 * osp_md_object_create(). For other cases, the OSP will assign FID to the
 * object to be created, and update last_used Object ID (OID) file.
 *
 * \param[in] env       pointer to the thread context
 * \param[in] dt        pointer to the OSP layer dt_object
 * \param[in] attr      the attribute for the object to be created
 * \param[in] hint      pointer to the hint for creating the object, such as
 *                      the parent object
 * \param[in] dof       pointer to the dt_object_format for help the creation
 * \param[in] th        pointer to the transaction handler
 *
 * \retval              0 for success
 * \retval              negative error number on failure
 */
static int osp_object_create(const struct lu_env *env, struct dt_object *dt,
                             struct lu_attr *attr,
                             struct dt_allocation_hint *hint,
                             struct dt_object_format *dof, struct thandle *th)
{
        struct osp_thread_info  *osi = osp_env_info(env);
        struct osp_device       *d = lu2osp_dev(dt->do_lu.lo_dev);
        struct osp_object       *o = dt2osp_obj(dt);
        int                     rc = 0;
        struct lu_fid           *fid = &osi->osi_fid;
        ENTRY;

        if (is_only_remote_trans(th) &&
            !fid_is_zero(lu_object_fid(&dt->do_lu))) {
                LASSERT(fid_is_sane(lu_object_fid(&dt->do_lu)));

                rc = osp_md_object_create(env, dt, attr, hint, dof, th);
                if (rc == 0)
                        o->opo_non_exist = 0;

                RETURN(rc);
        }

        o->opo_non_exist = 0;
        if (o->opo_reserved) {
                /* regular case, fid is assigned holding transaction open */
                 osp_object_assign_fid(env, d, o);
        }

        memcpy(fid, lu_object_fid(&dt->do_lu), sizeof(*fid));

        LASSERTF(fid_is_sane(fid), "fid for osp_object %p is insane"DFID"!\n",
                 o, PFID(fid));

        if (!o->opo_reserved) {
                /* special case, id was assigned outside of transaction
                 * see comments in osp_declare_attr_set */
                LASSERT(d->opd_pre != NULL);
                spin_lock(&d->opd_pre_lock);
                osp_update_last_fid(d, fid);
                spin_unlock(&d->opd_pre_lock);
        }

        CDEBUG(D_INODE, "fid for osp_object %p is "DFID"\n", o, PFID(fid));

        /* If the precreate ends, it means it will be ready to rollover to
         * the new sequence soon, all the creation should be synchronized,
         * otherwise during replay, the replay fid will be inconsistent with
         * last_used/create fid */
        if (osp_precreate_end_seq(env, d) && osp_is_fid_client(d))
                th->th_sync = 1;

        /*
         * it's OK if the import is inactive by this moment - id was created
         * by OST earlier, we just need to maintain it consistently on the disk
         * once import is reconnected, OSP will claim this and other objects
         * used and OST either keep them, if they exist or recreate
         */

        /* we might have lost precreated objects */
        if (unlikely(d->opd_gap_count) > 0) {
                LASSERT(d->opd_pre != NULL);
                spin_lock(&d->opd_pre_lock);
                if (d->opd_gap_count > 0) {
                        int count = d->opd_gap_count;

                        ostid_set_id(&osi->osi_oi,
                                     fid_oid(&d->opd_gap_start_fid));
                        d->opd_gap_count = 0;
                        spin_unlock(&d->opd_pre_lock);

                        CDEBUG(D_HA, "Writing gap "DFID"+%d in llog\n",
                               PFID(&d->opd_gap_start_fid), count);
                        /* real gap handling is disabled intil ORI-692 will be
                         * fixed, now we only report gaps */
                } else {
                        spin_unlock(&d->opd_pre_lock);
                }
        }

        /* Only need update last_used oid file, seq file will only be update
         * during seq rollover */
        osp_objid_buf_prep(&osi->osi_lb, &osi->osi_off,
                           &d->opd_last_used_fid.f_oid, d->opd_index);

        rc = dt_record_write(env, d->opd_last_used_oid_file, &osi->osi_lb,
                             &osi->osi_off, th);

        CDEBUG(D_HA, "%s: Wrote last used FID: "DFID", index %d: %d\n",
               d->opd_obd->obd_name, PFID(fid), d->opd_index, rc);

        RETURN(rc);
}

/**
 * Implement OSP layer dt_object_operations::do_declare_destroy() interface.
 *
 * Declare that the caller will destroy the specified OST object.
 *
 * The OST object destroy will be handled via llog asynchronously. This
 * function will declare the credits for generating MDS_UNLINK64_REC llog.
 *
 * \param[in] env       pointer to the thread context
 * \param[in] dt        pointer to the OSP layer dt_object to be destroyed
 * \param[in] th        pointer to the transaction handler
 *
 * \retval              0 for success
 * \retval              negative error number on failure
 */
int osp_declare_object_destroy(const struct lu_env *env,
                               struct dt_object *dt, struct thandle *th)
{
        struct osp_object       *o = dt2osp_obj(dt);
        int                      rc = 0;

        ENTRY;

        /*
         * track objects to be destroyed via llog
         */
        rc = osp_sync_declare_add(env, o, MDS_UNLINK64_REC, th);

        RETURN(rc);
}

/**
 * Implement OSP layer dt_object_operations::do_destroy() interface.
 *
 * Destroy the specified OST object.
 *
 * The OSP generates a MDS_UNLINK64_REC record in the llog. There
 * will be some dedicated thread to handle the llog asynchronously.
 *
 * It also marks the object as non-cached.
 *
 * \param[in] env       pointer to the thread context
 * \param[in] dt        pointer to the OSP layer dt_object to be destroyed
 * \param[in] th        pointer to the transaction handler
 *
 * \retval              0 for success
 * \retval              negative error number on failure
 */
int osp_object_destroy(const struct lu_env *env, struct dt_object *dt,
                       struct thandle *th)
{
        struct osp_object       *o = dt2osp_obj(dt);
        int                      rc = 0;

        ENTRY;

        o->opo_non_exist = 1;
        /*
         * once transaction is committed put proper command on
         * the queue going to our OST
         */
        rc = osp_sync_add(env, o, MDS_UNLINK64_REC, th, NULL);

        /* not needed in cache any more */
        set_bit(LU_OBJECT_HEARD_BANSHEE, &dt->do_lu.lo_header->loh_flags);

        RETURN(rc);
}

static int osp_orphan_index_lookup(const struct lu_env *env,
                                   struct dt_object *dt,
                                   struct dt_rec *rec,
                                   const struct dt_key *key,
                                   struct lustre_capa *capa)
{
        return -EOPNOTSUPP;
}

static int osp_orphan_index_declare_insert(const struct lu_env *env,
                                           struct dt_object *dt,
                                           const struct dt_rec *rec,
                                           const struct dt_key *key,
                                           struct thandle *handle)
{
        return -EOPNOTSUPP;
}

static int osp_orphan_index_insert(const struct lu_env *env,
                                   struct dt_object *dt,
                                   const struct dt_rec *rec,
                                   const struct dt_key *key,
                                   struct thandle *handle,
                                   struct lustre_capa *capa,
                                   int ignore_quota)
{
        return -EOPNOTSUPP;
}

static int osp_orphan_index_declare_delete(const struct lu_env *env,
                                           struct dt_object *dt,
                                           const struct dt_key *key,
                                           struct thandle *handle)
{
        return -EOPNOTSUPP;
}

static int osp_orphan_index_delete(const struct lu_env *env,
                                   struct dt_object *dt,
                                   const struct dt_key *key,
                                   struct thandle *handle,
                                   struct lustre_capa *capa)
{
        return -EOPNOTSUPP;
}

/**
 * Initialize the OSP layer index iteration.
 *
 * \param[in] env       pointer to the thread context
 * \param[in] dt        pointer to the index object to be iterated
 * \param[in] attr      unused
 * \param[in] capa      the capability for this operation
 *
 * \retval              pointer to the iteration structure
 * \retval              negative error number on failure
 */
struct dt_it *osp_it_init(const struct lu_env *env, struct dt_object *dt,
                          __u32 attr, struct lustre_capa *capa)
{
        struct osp_it *it;

        OBD_ALLOC_PTR(it);
        if (it == NULL)
                return ERR_PTR(-ENOMEM);

        it->ooi_pos_ent = -1;
        it->ooi_obj = dt;
        it->ooi_attr = attr;

        return (struct dt_it *)it;
}

/**
 * Finalize the OSP layer index iteration.
 *
 * \param[in] env       pointer to the thread context
 * \param[in] di        pointer to the iteration structure
 */
void osp_it_fini(const struct lu_env *env, struct dt_it *di)
{
        struct osp_it   *it = (struct osp_it *)di;
        struct page     **pages = it->ooi_pages;
        int             npages = it->ooi_total_npages;
        int             i;

        if (pages != NULL) {
                for (i = 0; i < npages; i++) {
                        if (pages[i] != NULL) {
                                if (pages[i] == it->ooi_cur_page) {
                                        kunmap(pages[i]);
                                        it->ooi_cur_page = NULL;
                                }
                                __free_page(pages[i]);
                        }
                }
                OBD_FREE(pages, npages * sizeof(*pages));
        }
        OBD_FREE_PTR(it);
}

/**
 * Get more records for the iteration from peer.
 *
 * The new records will be filled in an array of pages. The OSP side
 * allows 1MB bulk data to be transfered.
 *
 * \param[in] env       pointer to the thread context
 * \param[in] it        pointer to the iteration structure
 *
 * \retval              0 for success
 * \retval              negative error number on failure
 */
static int osp_it_fetch(const struct lu_env *env, struct osp_it *it)
{
        struct lu_device         *dev   = it->ooi_obj->do_lu.lo_dev;
        struct osp_device        *osp   = lu2osp_dev(dev);
        struct page             **pages;
        struct ptlrpc_request    *req   = NULL;
        struct ptlrpc_bulk_desc  *desc;
        struct idx_info          *ii;
        int                       npages;
        int                       rc;
        int                       i;
        ENTRY;

        /* 1MB bulk */
        npages = min_t(unsigned int, OFD_MAX_BRW_SIZE, 1 << 20);
        npages /= PAGE_CACHE_SIZE;

        OBD_ALLOC(pages, npages * sizeof(*pages));
        if (pages == NULL)
                RETURN(-ENOMEM);

        it->ooi_pages = pages;
        it->ooi_total_npages = npages;
        for (i = 0; i < npages; i++) {
                pages[i] = alloc_page(GFP_IOFS);
                if (pages[i] == NULL)
                        RETURN(-ENOMEM);
        }

        req = ptlrpc_request_alloc(osp->opd_obd->u.cli.cl_import,
                                   &RQF_OBD_IDX_READ);
        if (req == NULL)
                RETURN(-ENOMEM);

        rc = ptlrpc_request_pack(req, LUSTRE_OBD_VERSION, OBD_IDX_READ);
        if (rc != 0) {
                ptlrpc_request_free(req);
                RETURN(rc);
        }

        req->rq_request_portal = OUT_PORTAL;
        ii = req_capsule_client_get(&req->rq_pill, &RMF_IDX_INFO);
        memset(ii, 0, sizeof(*ii));
        if (fid_is_last_id(lu_object_fid(&it->ooi_obj->do_lu))) {
                /* LFSCK will iterate orphan object[FID_SEQ_LAYOUT_BTREE,
                 * ost_index, 0] with LAST_ID FID, so it needs to replace
                 * the FID with orphan FID here */
                ii->ii_fid.f_seq = FID_SEQ_LAYOUT_RBTREE;
                ii->ii_fid.f_oid = osp->opd_index;
                ii->ii_fid.f_ver = 0;
                ii->ii_flags = II_FL_NOHASH;
                ii->ii_attrs = osp_dev2node(osp);
        } else {
                ii->ii_fid = *lu_object_fid(&it->ooi_obj->do_lu);
                ii->ii_flags = II_FL_NOHASH | II_FL_NOKEY | II_FL_VARKEY |
                               II_FL_VARREC;
                ii->ii_attrs = it->ooi_attr;
        }
        ii->ii_magic = IDX_INFO_MAGIC;
        ii->ii_count = npages * LU_PAGE_COUNT;
        ii->ii_hash_start = it->ooi_next;

        ptlrpc_at_set_req_timeout(req);

        desc = ptlrpc_prep_bulk_imp(req, npages, 1, BULK_PUT_SINK,
                                    MDS_BULK_PORTAL);
        if (desc == NULL) {
                ptlrpc_request_free(req);
                RETURN(-ENOMEM);
        }

        for (i = 0; i < npages; i++)
                ptlrpc_prep_bulk_page_pin(desc, pages[i], 0, PAGE_CACHE_SIZE);

        ptlrpc_request_set_replen(req);
        rc = ptlrpc_queue_wait(req);
        if (rc != 0)
                GOTO(out, rc);

        rc = sptlrpc_cli_unwrap_bulk_read(req, req->rq_bulk,
                                          req->rq_bulk->bd_nob_transferred);
        if (rc < 0)
                GOTO(out, rc);
        rc = 0;

        ii = req_capsule_server_get(&req->rq_pill, &RMF_IDX_INFO);
        if (ii->ii_magic != IDX_INFO_MAGIC)
                 GOTO(out, rc = -EPROTO);

        npages = (ii->ii_count + LU_PAGE_COUNT - 1) >>
                 (PAGE_CACHE_SHIFT - LU_PAGE_SHIFT);
        if (npages > it->ooi_total_npages) {
                CERROR("%s: returned more pages than expected, %u > %u\n",
                       osp->opd_obd->obd_name, npages, it->ooi_total_npages);
                GOTO(out, rc = -EINVAL);
        }

        it->ooi_valid_npages = npages;
        if (ptlrpc_rep_need_swab(req))
                it->ooi_swab = 1;

        it->ooi_next = ii->ii_hash_end;

out:
        ptlrpc_req_finished(req);

        return rc;
}

/**
 * Move the iteration cursor to the next lu_page.
 *
 * One system page (PAGE_SIZE) may contain multiple lu_page (4KB),
 * that depends on the LU_PAGE_COUNT. If it is not the last lu_page
 * in current system page, then move the iteration cursor to the next
 * lu_page in current system page. Otherwise, if there are more system
 * pages in the cache, then move the iteration cursor to the next system
 * page. If all the cached records (pages) have been iterated, then fetch
 * more records via osp_it_fetch().
 *
 * \param[in] env       pointer to the thread context
 * \param[in] di        pointer to the iteration structure
 *
 * \retval              positive for end of the directory
 * \retval              0 for success
 * \retval              negative error number on failure
 */
int osp_it_next_page(const struct lu_env *env, struct dt_it *di)
{
        struct osp_it           *it = (struct osp_it *)di;
        struct lu_idxpage       *idxpage;
        struct page             **pages;
        int                     rc;
        int                     i;
        ENTRY;

again2:
        idxpage = it->ooi_cur_idxpage;
        if (idxpage != NULL) {
                if (idxpage->lip_nr == 0)
                        RETURN(1);

                if (it->ooi_pos_ent < idxpage->lip_nr) {
                        CDEBUG(D_INFO, "ooi_pos %d nr %d\n",
                               (int)it->ooi_pos_ent, (int)idxpage->lip_nr);
                        RETURN(0);
                }
                it->ooi_cur_idxpage = NULL;
                it->ooi_pos_lu_page++;

again1:
                if (it->ooi_pos_lu_page < LU_PAGE_COUNT) {
                        it->ooi_cur_idxpage = (void *)it->ooi_cur_page +
                                         LU_PAGE_SIZE * it->ooi_pos_lu_page;
                        if (it->ooi_swab)
                                lustre_swab_lip_header(it->ooi_cur_idxpage);
                        if (it->ooi_cur_idxpage->lip_magic != LIP_MAGIC) {
                                struct osp_device *osp =
                                        lu2osp_dev(it->ooi_obj->do_lu.lo_dev);

                                CERROR("%s: invalid magic (%x != %x) for page "
                                       "%d/%d while read layout orphan index\n",
                                       osp->opd_obd->obd_name,
                                       it->ooi_cur_idxpage->lip_magic,
                                       LIP_MAGIC, it->ooi_pos_page,
                                       it->ooi_pos_lu_page);
                                /* Skip this lu_page next time. */
                                it->ooi_pos_ent = idxpage->lip_nr - 1;
                                RETURN(-EINVAL);
                        }
                        it->ooi_pos_ent = -1;
                        goto again2;
                }

                kunmap(it->ooi_cur_page);
                it->ooi_cur_page = NULL;
                it->ooi_pos_page++;

again0:
                pages = it->ooi_pages;
                if (it->ooi_pos_page < it->ooi_valid_npages) {
                        it->ooi_cur_page = kmap(pages[it->ooi_pos_page]);
                        it->ooi_pos_lu_page = 0;
                        goto again1;
                }

                for (i = 0; i < it->ooi_total_npages; i++) {
                        if (pages[i] != NULL)
                                __free_page(pages[i]);
                }
                OBD_FREE(pages, it->ooi_total_npages * sizeof(*pages));

                it->ooi_pos_page = 0;
                it->ooi_total_npages = 0;
                it->ooi_valid_npages = 0;
                it->ooi_swab = 0;
                it->ooi_ent = NULL;
                it->ooi_cur_page = NULL;
                it->ooi_cur_idxpage = NULL;
                it->ooi_pages = NULL;
        }

        if (it->ooi_next == II_END_OFF)
                RETURN(1);

        rc = osp_it_fetch(env, it);
        if (rc == 0)
                goto again0;

        RETURN(rc);
}

/**
 * Move the iteration cursor to the next record.
 *
 * If there are more records in the lu_page, then move the iteration
 * cursor to the next record directly. Otherwise, move the iteration
 * cursor to the record in the next lu_page via osp_it_next_page()
 *
 * \param[in] env       pointer to the thread context
 * \param[in] di        pointer to the iteration structure
 *
 * \retval              positive for end of the directory
 * \retval              0 for success
 * \retval              negative error number on failure
 */
static int osp_orphan_it_next(const struct lu_env *env, struct dt_it *di)
{
        struct osp_it           *it = (struct osp_it *)di;
        struct lu_idxpage       *idxpage;
        int                     rc;
        ENTRY;

again:
        idxpage = it->ooi_cur_idxpage;
        if (idxpage != NULL) {
                if (idxpage->lip_nr == 0)
                        RETURN(1);

                it->ooi_pos_ent++;
                if (it->ooi_pos_ent < idxpage->lip_nr) {
                        it->ooi_ent =
                                (struct lu_orphan_ent *)idxpage->lip_entries +
                                                        it->ooi_pos_ent;
                        if (it->ooi_swab)
                                lustre_swab_orphan_ent(it->ooi_ent);
                        RETURN(0);
                }
        }

        rc = osp_it_next_page(env, di);
        if (rc == 0)
                goto again;

        RETURN(rc);
}

int osp_it_get(const struct lu_env *env, struct dt_it *di,
               const struct dt_key *key)
{
        return 1;
}

void osp_it_put(const struct lu_env *env, struct dt_it *di)
{
}

static struct dt_key *osp_orphan_it_key(const struct lu_env *env,
                                        const struct dt_it *di)
{
        struct osp_it   *it  = (struct osp_it *)di;
        struct lu_orphan_ent    *ent = (struct lu_orphan_ent *)it->ooi_ent;

        if (likely(ent != NULL))
                return (struct dt_key *)(&ent->loe_key);

        return NULL;
}

static int osp_orphan_it_key_size(const struct lu_env *env,
                                  const struct dt_it *di)
{
        return sizeof(struct lu_fid);
}

static int osp_orphan_it_rec(const struct lu_env *env, const struct dt_it *di,
                             struct dt_rec *rec, __u32 attr)
{
        struct osp_it   *it  = (struct osp_it *)di;
        struct lu_orphan_ent    *ent = (struct lu_orphan_ent *)it->ooi_ent;

        if (likely(ent != NULL)) {
                *(struct lu_orphan_rec *)rec = ent->loe_rec;
                return 0;
        }

        return -EINVAL;
}

__u64 osp_it_store(const struct lu_env *env, const struct dt_it *di)
{
        struct osp_it   *it = (struct osp_it *)di;

        return it->ooi_next;
}

/**
 * Locate the iteration cursor to the specified position (cookie).
 *
 * \param[in] env       pointer to the thread context
 * \param[in] di        pointer to the iteration structure
 * \param[in] hash      the specified position
 *
 * \retval              positive number for locating to the exactly position
 *                      or the next
 * \retval              0 for arriving at the end of the iteration
 * \retval              negative error number on failure
 */
int osp_orphan_it_load(const struct lu_env *env, const struct dt_it *di,
                       __u64 hash)
{
        struct osp_it   *it     = (struct osp_it *)di;
        int              rc;

        it->ooi_next = hash;
        rc = osp_orphan_it_next(env, (struct dt_it *)di);
        if (rc == 1)
                return 0;

        if (rc == 0)
                return 1;

        return rc;
}

int osp_it_key_rec(const struct lu_env *env, const struct dt_it *di,
                   void *key_rec)
{
        return 0;
}

static const struct dt_index_operations osp_orphan_index_ops = {
        .dio_lookup             = osp_orphan_index_lookup,
        .dio_declare_insert     = osp_orphan_index_declare_insert,
        .dio_insert             = osp_orphan_index_insert,
        .dio_declare_delete     = osp_orphan_index_declare_delete,
        .dio_delete             = osp_orphan_index_delete,
        .dio_it = {
                .init           = osp_it_init,
                .fini           = osp_it_fini,
                .next           = osp_orphan_it_next,
                .get            = osp_it_get,
                .put            = osp_it_put,
                .key            = osp_orphan_it_key,
                .key_size       = osp_orphan_it_key_size,
                .rec            = osp_orphan_it_rec,
                .store          = osp_it_store,
                .load           = osp_orphan_it_load,
                .key_rec        = osp_it_key_rec,
        }
};

/**
 * Implement OSP layer dt_object_operations::do_index_try() interface.
 *
 * Negotiate the index type.
 *
 * If the target index is an IDIF object, then use osp_orphan_index_ops.
 * Otherwise, assign osp_md_index_ops to the dt_object::do_index_ops.
 * (\see lustre/include/lustre_fid.h for IDIF.)
 *
 * \param[in] env       pointer to the thread context
 * \param[in] dt        pointer to the OSP layer dt_object
 * \param[in] feat      unused
 *
 * \retval              0 for success
 */
static int osp_index_try(const struct lu_env *env,
                         struct dt_object *dt,
                         const struct dt_index_features *feat)
{
        const struct lu_fid *fid = lu_object_fid(&dt->do_lu);

        if (fid_is_last_id(fid) && fid_is_idif(fid))
                dt->do_index_ops = &osp_orphan_index_ops;
        else
                dt->do_index_ops = &osp_md_index_ops;
        return 0;
}

static struct dt_object_operations osp_obj_ops = {
        .do_declare_attr_get    = osp_declare_attr_get,
        .do_attr_get            = osp_attr_get,
        .do_declare_attr_set    = osp_declare_attr_set,
        .do_attr_set            = osp_attr_set,
        .do_declare_xattr_get   = osp_declare_xattr_get,
        .do_xattr_get           = osp_xattr_get,
        .do_declare_xattr_set   = osp_declare_xattr_set,
        .do_xattr_set           = osp_xattr_set,
        .do_declare_create      = osp_declare_object_create,
        .do_create              = osp_object_create,
        .do_declare_destroy     = osp_declare_object_destroy,
        .do_destroy             = osp_object_destroy,
        .do_index_try           = osp_index_try,
};

/**
 * Implement OSP layer lu_object_operations::loo_object_init() interface.
 *
 * Initialize the object.
 *
 * If it is a remote MDT object, then call do_attr_get() to fetch
 * the attribute from the peer.
 *
 * \param[in] env       pointer to the thread context
 * \param[in] o         pointer to the OSP layer lu_object
 * \param[in] conf      unused
 *
 * \retval              0 for success
 * \retval              negative error number on failure
 */
static int osp_object_init(const struct lu_env *env, struct lu_object *o,
                           const struct lu_object_conf *conf)
{
        struct osp_object       *po = lu2osp_obj(o);
        int                     rc = 0;
        ENTRY;

        spin_lock_init(&po->opo_lock);
        o->lo_header->loh_attr |= LOHA_REMOTE;

        if (is_ost_obj(o)) {
                po->opo_obj.do_ops = &osp_obj_ops;
        } else {
                struct lu_attr *la = &osp_env_info(env)->osi_attr;

                po->opo_obj.do_ops = &osp_md_obj_ops;
                po->opo_obj.do_body_ops = &osp_md_body_ops;
                rc = po->opo_obj.do_ops->do_attr_get(env, lu2dt_obj(o),
                                                     la, NULL);
                if (rc == 0)
                        o->lo_header->loh_attr |=
                                LOHA_EXISTS | (la->la_mode & S_IFMT);
                if (rc == -ENOENT) {
                        po->opo_non_exist = 1;
                        rc = 0;
                }
                init_rwsem(&po->opo_sem);
        }
        RETURN(rc);
}

/**
 * Implement OSP layer lu_object_operations::loo_object_free() interface.
 *
 * Finalize the object.
 *
 * If the OSP object has attributes cache, then destroy the cache.
 * Free the object finally.
 *
 * \param[in] env       pointer to the thread context
 * \param[in] o         pointer to the OSP layer lu_object
 */
static void osp_object_free(const struct lu_env *env, struct lu_object *o)
{
        struct osp_object       *obj = lu2osp_obj(o);
        struct lu_object_header *h = o->lo_header;

        dt_object_fini(&obj->opo_obj);
        lu_object_header_fini(h);
        if (obj->opo_ooa != NULL) {
                struct osp_xattr_entry *oxe;
                struct osp_xattr_entry *tmp;
                int                     count;

                list_for_each_entry_safe(oxe, tmp,
                                         &obj->opo_ooa->ooa_xattr_list,
                                         oxe_list) {
                        list_del(&oxe->oxe_list);
                        count = atomic_read(&oxe->oxe_ref);
                        LASSERTF(count == 1,
                                 "Still has %d users on the xattr entry %.*s\n",
                                 count-1, (int)oxe->oxe_namelen, oxe->oxe_buf);

                        OBD_FREE(oxe, oxe->oxe_buflen);
                }
                OBD_FREE_PTR(obj->opo_ooa);
        }
        OBD_SLAB_FREE_PTR(obj, osp_object_kmem);
}

/**
 * Implement OSP layer lu_object_operations::loo_object_release() interface.
 *
 * Cleanup (not free) the object.
 *
 * If it is a reserved object but failed to be created, or it is an OST
 * object, then mark the object as non-cached.
 *
 * \param[in] env       pointer to the thread context
 * \param[in] o         pointer to the OSP layer lu_object
 */
static void osp_object_release(const struct lu_env *env, struct lu_object *o)
{
        struct osp_object       *po = lu2osp_obj(o);
        struct osp_device       *d  = lu2osp_dev(o->lo_dev);

        ENTRY;

        /*
         * release reservation if object was declared but not created
         * this may require lu_object_put() in LOD
         */
        if (unlikely(po->opo_reserved)) {
                LASSERT(d->opd_pre != NULL);
                LASSERT(d->opd_pre_reserved > 0);
                spin_lock(&d->opd_pre_lock);
                d->opd_pre_reserved--;
                spin_unlock(&d->opd_pre_lock);

                /* not needed in cache any more */
                set_bit(LU_OBJECT_HEARD_BANSHEE, &o->lo_header->loh_flags);
        }

        if (is_ost_obj(o))
                /* XXX: Currently, NOT cache OST-object on MDT because:
                 *      1. it is not often accessed on MDT.
                 *      2. avoid up layer (such as LFSCK) to load too many
                 *         once-used OST-objects. */
                set_bit(LU_OBJECT_HEARD_BANSHEE, &o->lo_header->loh_flags);

        EXIT;
}

static int osp_object_print(const struct lu_env *env, void *cookie,
                            lu_printer_t p, const struct lu_object *l)
{
        const struct osp_object *o = lu2osp_obj((struct lu_object *)l);

        return (*p)(env, cookie, LUSTRE_OSP_NAME"-object@%p", o);
}

static int osp_object_invariant(const struct lu_object *o)
{
        LBUG();
}

struct lu_object_operations osp_lu_obj_ops = {
        .loo_object_init        = osp_object_init,
        .loo_object_free        = osp_object_free,
        .loo_object_release     = osp_object_release,
        .loo_object_print       = osp_object_print,
        .loo_object_invariant   = osp_object_invariant
};