LU-16634 obdclass: improve iocontrol error messages

[fs/lustre-release.git] / lustre / lod / lod_dev.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 COPYING file that accompanied this code.
 *
 * You should have received a copy of the GNU General Public License
 * along with this program; if not, write to the Free Software
 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
 *
 * GPL HEADER END
 */
/*
 * Copyright  2009 Sun Microsystems, Inc. All rights reserved
 * Use is subject to license terms.
 *
 * Copyright (c) 2012, 2017, Intel Corporation.
 */
/*
 * This file is part of Lustre, http://www.lustre.org/
 *
 * lustre/lod/lod_dev.c
 *
 * Lustre Logical Object Device
 *
 * Author: Alex Zhuravlev <alexey.zhuravlev@intel.com>
 * Author: Mikhail Pershin <mike.pershin@intel.com>
 */
/**
 * The Logical Object Device (LOD) layer manages access to striped
 * objects (both regular files and directories). It implements the DT
 * device and object APIs and is responsible for creating, storing,
 * and loading striping information as an extended attribute of the
 * underlying OSD object. LOD is the server side analog of the LOV and
 * LMV layers on the client side.
 *
 * Metadata LU object stack (layers of the same compound LU object,
 * all have the same FID):
 *
 *        MDT
 *         |      MD API
 *        MDD
 *         |      DT API
 *        LOD
 *       /   \    DT API
 *     OSD   OSP
 *
 * During LOD object initialization the localness or remoteness of the
 * object FID dictates the choice between OSD and OSP.
 *
 * An LOD object (file or directory) with N stripes (each has a
 * different FID):
 *
 *          LOD
 *           |
 *   +---+---+---+...+
 *   |   |   |   |   |
 *   S0  S1  S2  S3  S(N-1)  OS[DP] objects, seen as DT objects by LOD
 *
 * When upper layers must access an object's stripes (which are
 * themselves OST or MDT LU objects) LOD finds these objects by their
 * FIDs and stores them as an array of DT object pointers on the
 * object. Declarations and operations on LOD objects are received by
 * LOD (as DT object operations) and performed on the underlying
 * OS[DP] object and (as needed) on the stripes. From the perspective
 * of LOD, a stripe-less file (created by mknod() or open with
 * O_LOV_DELAY_CREATE) is an object which does not yet have stripes,
 * while a non-striped directory (created by mkdir()) is an object
 * which will never have stripes.
 *
 * The LOD layer also implements a small subset of the OBD device API
 * to support MDT stack initialization and finalization (an MDD device
 * connects and disconnects itself to and from the underlying LOD
 * device), and pool management. In turn LOD uses the OBD device API
 * to connect it self to the underlying OSD, and to connect itself to
 * OSP devices representing the MDTs and OSTs that bear the stripes of
 * its objects.
 */

#define DEBUG_SUBSYSTEM S_MDS

#include <linux/kthread.h>
#include <obd_class.h>
#include <md_object.h>
#include <lustre_fid.h>
#include <uapi/linux/lustre/lustre_param.h>
#include <uapi/linux/lustre/lustre_ioctl.h>
#include <lustre_update.h>
#include <lustre_log.h>
#include <lustre_lmv.h>
#include <llog_swab.h>

#include "lod_internal.h"

static const char lod_update_log_name[] = "update_log";
static const char lod_update_log_dir_name[] = "update_log_dir";

/*
 * Lookup target by FID.
 *
 * Lookup MDT/OST target index by FID. Type of the target can be
 * specific or any.
 *
 * \param[in] env               LU environment provided by the caller
 * \param[in] lod               lod device
 * \param[in] fid               FID
 * \param[out] tgt              result target index
 * \param[in] type              expected type of the target:
 *                              LU_SEQ_RANGE_{MDT,OST,ANY}
 *
 * \retval 0                    on success
 * \retval negative             negated errno on error
 **/
int lod_fld_lookup(const struct lu_env *env, struct lod_device *lod,
                   const struct lu_fid *fid, u32 *tgt, int *type)
{
        struct lu_seq_range range = { 0 };
        struct lu_server_fld *server_fld;
        int rc;

        ENTRY;

        if (!fid_is_sane(fid)) {
                CERROR("%s: invalid FID "DFID"\n", lod2obd(lod)->obd_name,
                       PFID(fid));
                RETURN(-EIO);
        }

        if (fid_is_idif(fid)) {
                *tgt = fid_idif_ost_idx(fid);
                *type = LU_SEQ_RANGE_OST;
                RETURN(0);
        }

        if (fid_is_update_log(fid) || fid_is_update_log_dir(fid)) {
                *tgt = fid_oid(fid);
                *type = LU_SEQ_RANGE_MDT;
                RETURN(0);
        }

        if (!lod->lod_initialized || (!fid_seq_in_fldb(fid_seq(fid)))) {
                LASSERT(lu_site2seq(lod2lu_dev(lod)->ld_site) != NULL);

                *tgt = lu_site2seq(lod2lu_dev(lod)->ld_site)->ss_node_id;
                *type = LU_SEQ_RANGE_MDT;
                RETURN(0);
        }

        server_fld = lu_site2seq(lod2lu_dev(lod)->ld_site)->ss_server_fld;
        if (!server_fld)
                RETURN(-EIO);

        fld_range_set_type(&range, *type);
        rc = fld_server_lookup(env, server_fld, fid_seq(fid), &range);
        if (rc != 0)
                RETURN(rc);

        *tgt = range.lsr_index;
        *type = range.lsr_flags;

        CDEBUG(D_INFO, "%s: got tgt %x for sequence: %#llx\n",
               lod2obd(lod)->obd_name, *tgt, fid_seq(fid));

        RETURN(0);
}

/* Slab for OSD object allocation */
struct kmem_cache *lod_object_kmem;

/* Slab for dt_txn_callback */
struct kmem_cache *lod_txn_callback_kmem;
static struct lu_kmem_descr lod_caches[] = {
        {
                .ckd_cache = &lod_object_kmem,
                .ckd_name  = "lod_obj",
                .ckd_size  = sizeof(struct lod_object)
        },
        {
                .ckd_cache = &lod_txn_callback_kmem,
                .ckd_name  = "lod_txn_callback",
                .ckd_size  = sizeof(struct dt_txn_callback)
        },
        {
                .ckd_cache = NULL
        }
};

static struct lu_device *lod_device_fini(const struct lu_env *env,
                                         struct lu_device *d);

/**
 * Implementation of lu_device_operations::ldo_object_alloc() for LOD
 *
 * Allocates and initializes LOD's slice in the given object.
 *
 * see include/lu_object.h for the details.
 */
static struct lu_object *lod_object_alloc(const struct lu_env *env,
                                          const struct lu_object_header *hdr,
                                          struct lu_device *dev)
{
        struct lod_object *lod_obj;
        struct lu_object *lu_obj;

        ENTRY;

        OBD_SLAB_ALLOC_PTR_GFP(lod_obj, lod_object_kmem, GFP_NOFS);
        if (!lod_obj)
                RETURN(ERR_PTR(-ENOMEM));

        mutex_init(&lod_obj->ldo_layout_mutex);
        lu_obj = lod2lu_obj(lod_obj);
        dt_object_init(&lod_obj->ldo_obj, NULL, dev);
        lod_obj->ldo_obj.do_ops = &lod_obj_ops;
        lu_obj->lo_ops = &lod_lu_obj_ops;

        RETURN(lu_obj);
}

/**
 * Process the config log for all sub device.
 *
 * The function goes through all the targets in the given table
 * and apply given configuration command on to the targets.
 * Used to cleanup the targets at unmount.
 *
 * \param[in] env               LU environment provided by the caller
 * \param[in] lod               lod device
 * \param[in] ltd               target's table to go through
 * \param[in] lcfg              configuration command to apply
 *
 * \retval 0                    on success
 * \retval negative             negated errno on error
 **/
static int lod_sub_process_config(const struct lu_env *env,
                                 struct lod_device *lod,
                                 struct lod_tgt_descs *ltd,
                                 struct lustre_cfg *lcfg)
{
        struct lu_device *next;
        struct lu_tgt_desc *tgt;
        int rc = 0;

        lod_getref(ltd);
        ltd_foreach_tgt(ltd, tgt) {
                int rc1;

                LASSERT(tgt && tgt->ltd_tgt);
                next = &tgt->ltd_tgt->dd_lu_dev;
                rc1 = next->ld_ops->ldo_process_config(env, next, lcfg);
                if (rc1) {
                        CERROR("%s: error cleaning up LOD index %u: cmd %#x : rc = %d\n",
                               lod2obd(lod)->obd_name, tgt->ltd_index,
                               lcfg->lcfg_command, rc1);
                        rc = rc1;
                }
        }
        lod_putref(lod, ltd);
        return rc;
}

struct lod_recovery_data {
        struct lod_device       *lrd_lod;
        struct lod_tgt_desc     *lrd_ltd;
        struct task_struct      **lrd_task;
        u32                     lrd_idx;
        struct lu_env           lrd_env;
        struct completion       *lrd_started;
};

/**
 * process update recovery record
 *
 * Add the update recovery recode to the update recovery list in
 * lod_recovery_data. Then the recovery thread (target_recovery_thread)
 * will redo these updates.
 *
 * \param[in]env        execution environment
 * \param[in]llh        log handle of update record
 * \param[in]rec        update record to be replayed
 * \param[in]data       update recovery data which holds the necessary
 *                      arguments for recovery (see struct lod_recovery_data)
 *
 * \retval              0 if the record is processed successfully.
 * \retval              negative errno if the record processing fails.
 */
static int lod_process_recovery_updates(const struct lu_env *env,
                                        struct llog_handle *llh,
                                        struct llog_rec_hdr *rec,
                                        void *data)
{
        struct lod_recovery_data *lrd = data;
        struct llog_cookie *cookie = &lod_env_info(env)->lti_cookie;
        struct lu_target *lut;
        u32 index = 0;

        ENTRY;

        if (!lrd->lrd_ltd) {
                int rc;

                rc = lodname2mdt_index(lod2obd(lrd->lrd_lod)->obd_name, &index);
                if (rc != 0)
                        return rc;
        } else {
                index = lrd->lrd_ltd->ltd_index;
        }

        if (rec->lrh_len !=
                llog_update_record_size((struct llog_update_record *)rec)) {
                CERROR("%s: broken update record! index %u "DFID".%u: rc = %d\n",
                       lod2obd(lrd->lrd_lod)->obd_name, index,
                       PLOGID(&llh->lgh_id), rec->lrh_index, -EIO);
                return -EINVAL;
        }

        cookie->lgc_lgl = llh->lgh_id;
        cookie->lgc_index = rec->lrh_index;
        cookie->lgc_subsys = LLOG_UPDATELOG_ORIG_CTXT;

        CDEBUG(D_HA, "%s: process recovery updates "DFID".%u\n",
               lod2obd(lrd->lrd_lod)->obd_name,
               PLOGID(&llh->lgh_id), rec->lrh_index);
        lut = lod2lu_dev(lrd->lrd_lod)->ld_site->ls_tgt;

        if (obd_mdt_recovery_abort(lut->lut_obd))
                return -ESHUTDOWN;

        return insert_update_records_to_replay_list(lut->lut_tdtd,
                                        (struct llog_update_record *)rec,
                                        cookie, index);
}

/* retain old catalog, create new catalog and update catlist */
static int lod_sub_recreate_llog(const struct lu_env *env,
                                 struct lod_device *lod, struct dt_device *dt,
                                 int index)
{
        struct lod_thread_info *lti = lod_env_info(env);
        struct llog_ctxt *ctxt;
        struct llog_handle *lgh;
        struct llog_catid *cid = &lti->lti_cid;
        struct lu_fid *fid = &lti->lti_fid;
        struct obd_device *obd;
        int rc;

        ENTRY;
        lu_update_log_fid(fid, index);
        rc = lodname2mdt_index(lod2obd(lod)->obd_name, (__u32 *)&index);
        if (rc < 0)
                RETURN(rc);

        rc = llog_osd_get_cat_list(env, dt, index, 1, NULL, fid);
        if (rc < 0) {
                CERROR("%s: can't access update_log: rc = %d\n",
                       lod2obd(lod)->obd_name, rc);
                RETURN(rc);
        }

        obd = dt->dd_lu_dev.ld_obd;
        ctxt = llog_get_context(obd, LLOG_UPDATELOG_ORIG_CTXT);
        LASSERT(ctxt != NULL);
        if (ctxt->loc_handle) {
                /* retain old catalog */
                llog_retain(env, ctxt->loc_handle);
                llog_cat_close(env, ctxt->loc_handle);
                LASSERT(!ctxt->loc_handle);
        }

        ctxt->loc_flags |= LLOG_CTXT_FLAG_NORMAL_FID;
        ctxt->loc_chunk_size = LLOG_MIN_CHUNK_SIZE * 4;
        rc = llog_open_create(env, ctxt, &lgh, NULL, NULL);
        if (rc < 0)
                GOTO(out_put, rc);

        LASSERT(lgh != NULL);
        rc = llog_init_handle(env, lgh, LLOG_F_IS_CAT, NULL);
        if (rc != 0)
                GOTO(out_close, rc);

        cid->lci_logid = lgh->lgh_id;
        rc = llog_osd_put_cat_list(env, dt, index, 1, cid, fid);
        if (rc != 0)
                GOTO(out_close, rc);

        ctxt->loc_handle = lgh;

        CDEBUG(D_INFO, "%s: recreate catalog "DFID"\n",
               obd->obd_name, PLOGID(&cid->lci_logid));
out_close:
        if (rc)
                llog_cat_close(env, lgh);
out_put:
        llog_ctxt_put(ctxt);
        RETURN(rc);
}

/* retain update catalog and llogs, and create a new catalog */
static int lod_sub_cancel_llog(const struct lu_env *env,
                               struct lod_device *lod, struct dt_device *dt,
                               int index)
{
        struct llog_ctxt *ctxt;
        int rc = 0;

        ctxt = llog_get_context(dt->dd_lu_dev.ld_obd,
                                LLOG_UPDATELOG_ORIG_CTXT);
        if (!ctxt)
                return 0;

        if (ctxt->loc_handle) {
                LCONSOLE(D_INFO, "%s: cancel update llog "DFID"\n",
                         dt->dd_lu_dev.ld_obd->obd_name,
                         PLOGID(&ctxt->loc_handle->lgh_id));
                /* set retention on logs to simplify reclamation */
                llog_process_or_fork(env, ctxt->loc_handle, llog_cat_retain_cb,
                                     NULL, NULL, false);
                /* retain old catalog and create a new one */
                lod_sub_recreate_llog(env, lod, dt, index);
        }
        llog_ctxt_put(ctxt);
        return rc;
}

/**
 * recovery thread for update log
 *
 * Start recovery thread and prepare the sub llog, then it will retrieve
 * the update records from the correpondent MDT and do recovery.
 *
 * \param[in] arg       pointer to the recovery data
 *
 * \retval              0 if recovery succeeds
 * \retval              negative errno if recovery failed.
 */
static int lod_sub_recovery_thread(void *arg)
{
        struct lod_recovery_data *lrd = arg;
        struct lod_device *lod = lrd->lrd_lod;
        struct dt_device *dt;
        struct llog_ctxt *ctxt = NULL;
        struct lu_env *env = &lrd->lrd_env;
        struct lu_target *lut;
        struct lu_tgt_desc *mdt = NULL;
        struct lu_device *top_device;
        time64_t start;
        int retries = 0;
        int rc;

        ENTRY;

        lut = lod2lu_dev(lod)->ld_site->ls_tgt;
        atomic_inc(&lut->lut_tdtd->tdtd_recovery_threads_count);
        if (!lrd->lrd_ltd)
                dt = lod->lod_child;
        else
                dt = lrd->lrd_ltd->ltd_tgt;

        start = ktime_get_real_seconds();
        complete(lrd->lrd_started);

again:

        if (unlikely(OBD_FAIL_PRECHECK(OBD_FAIL_TGT_RECOVERY_CONNECT)) &&
            lrd->lrd_ltd) {
                OBD_FAIL_TIMEOUT(OBD_FAIL_TGT_RECOVERY_CONNECT, cfs_fail_val);
                rc = -EIO;
        } else {
                rc = lod_sub_prep_llog(env, lod, dt, lrd->lrd_idx);
        }

        if (!rc && !lod->lod_child->dd_rdonly) {
                /* Process the recovery record */
                ctxt = llog_get_context(dt->dd_lu_dev.ld_obd,
                                        LLOG_UPDATELOG_ORIG_CTXT);
                LASSERT(ctxt != NULL);
                LASSERT(ctxt->loc_handle != NULL);

                rc = llog_cat_process(env, ctxt->loc_handle,
                                      lod_process_recovery_updates, lrd, 0, 0);
        }

        top_device = lod->lod_dt_dev.dd_lu_dev.ld_site->ls_top_dev;
        if (rc < 0 && dt != lod->lod_child &&
            !obd_mdt_recovery_abort(top_device->ld_obd)) {
                if (rc == -EBADR) {
                        /* remote update llog is shorter than expected from
                         * local header. Cached copy could be de-synced during
                         * recovery, trust remote llog data
                         */
                        CDEBUG(D_HA, "%s update log data de-sync\n",
                               dt->dd_lu_dev.ld_obd->obd_name);
                        rc = 0;
                } else if (rc == -ETIMEDOUT || rc == -EAGAIN || rc == -EIO) {
                        /*
                         * the remote target might failover at the same time,
                         * let's retry here
                         */
                        if (ctxt) {
                                if (ctxt->loc_handle)
                                        llog_cat_close(env, ctxt->loc_handle);
                                llog_ctxt_put(ctxt);
                                ctxt = NULL;
                        }
                        retries++;
                        CDEBUG(D_HA, "%s get update log failed %d, retry\n",
                               dt->dd_lu_dev.ld_obd->obd_name, rc);
                        goto again;
                }
        }

        llog_ctxt_put(ctxt);
        if (rc < 0) {
                CERROR("%s: get update log duration %lld, retries %d, failed: rc = %d\n",
                       dt->dd_lu_dev.ld_obd->obd_name,
                       ktime_get_real_seconds() - start, retries, rc);
                /* abort MDT recovery of this target, but not all targets,
                 * because recovery still has chance to succeed.
                 */
                if (!obd_mdt_recovery_abort(top_device->ld_obd))
                        lod_sub_cancel_llog(env, lod, dt, lrd->lrd_idx);
        } else {
                CDEBUG(D_HA,
                       "%s retrieved update log, duration %lld, retries %d\n",
                       dt->dd_lu_dev.ld_obd->obd_name,
                       ktime_get_real_seconds() - start, retries);
        }

        spin_lock(&lod->lod_lock);
        if (!lrd->lrd_ltd)
                lod->lod_child_got_update_log = 1;
        else
                lrd->lrd_ltd->ltd_got_update_log = 1;

        if (!lod->lod_child_got_update_log) {
                spin_unlock(&lod->lod_lock);
                GOTO(out, rc);
        }

        lod_foreach_mdt(lod, mdt) {
                if (!mdt->ltd_got_update_log) {
                        spin_unlock(&lod->lod_lock);
                        GOTO(out, rc);
                }
        }
        lut->lut_tdtd->tdtd_replay_ready = 1;
        spin_unlock(&lod->lod_lock);

        CDEBUG(D_HA, "%s got update logs from all MDTs.\n",
               lut->lut_obd->obd_name);
        wake_up(&lut->lut_obd->obd_next_transno_waitq);
        EXIT;

out:
        atomic_dec(&lut->lut_tdtd->tdtd_recovery_threads_count);
        wake_up(&lut->lut_tdtd->tdtd_recovery_threads_waitq);
        if (xchg(lrd->lrd_task, NULL) == NULL)
                /* Someone is waiting for us to finish, need
                 * to synchronize cleanly.
                 */
                wait_var_event(lrd, kthread_should_stop());
        lu_env_fini(env);
        OBD_FREE_PTR(lrd);
        return rc;
}

/**
 * finish sub llog context
 *
 * Stop update recovery thread for the sub device, then cleanup the
 * correspondent llog ctxt.
 *
 * \param[in] env      execution environment
 * \param[in] lod      lod device to do update recovery
 * \param[in] thread   recovery thread on this sub device
 */
void lod_sub_fini_llog(const struct lu_env *env,
                       struct dt_device *dt, struct task_struct **thread)
{
        struct obd_device *obd;
        struct llog_ctxt *ctxt;
        struct task_struct *task = NULL;

        ENTRY;

        obd = dt->dd_lu_dev.ld_obd;
        CDEBUG(D_INFO, "%s: finish sub llog\n", obd->obd_name);
        /* Wait for recovery thread to complete */
        if (thread)
                task = xchg(thread, NULL);
        if (task)
                kthread_stop(task);

        ctxt = llog_get_context(obd, LLOG_UPDATELOG_ORIG_CTXT);
        if (!ctxt)
                RETURN_EXIT;

        if (ctxt->loc_handle)
                llog_cat_close(env, ctxt->loc_handle);

        llog_cleanup(env, ctxt);

        RETURN_EXIT;
}

/**
 * Extract MDT target index from a device name.
 *
 * a helper function to extract index from the given device name
 * like "fsname-MDTxxxx-mdtlov"
 *
 * \param[in] lodname           device name
 * \param[out] mdt_index        extracted index
 *
 * \retval 0            on success
 * \retval -EINVAL      if the name is invalid
 */
int lodname2mdt_index(char *lodname, u32 *mdt_index)
{
        u32 index;
        const char *ptr, *tmp;
        int rc;

        /* 1.8 configs don't have "-MDT0000" at the end */
        ptr = strstr(lodname, "-MDT");
        if (!ptr) {
                *mdt_index = 0;
                return 0;
        }

        ptr = strrchr(lodname, '-');
        if (!ptr) {
                rc = -EINVAL;
                CERROR("invalid MDT index in '%s': rc = %d\n", lodname, rc);
                return rc;
        }

        if (strncmp(ptr, "-mdtlov", 7) != 0) {
                rc = -EINVAL;
                CERROR("invalid MDT index in '%s': rc = %d\n", lodname, rc);
                return rc;
        }

        if ((unsigned long)ptr - (unsigned long)lodname <= 8) {
                rc = -EINVAL;
                CERROR("invalid MDT index in '%s': rc = %d\n", lodname, rc);
                return rc;
        }

        if (strncmp(ptr - 8, "-MDT", 4) != 0) {
                rc = -EINVAL;
                CERROR("invalid MDT index in '%s': rc = %d\n", lodname, rc);
                return rc;
        }

        rc = target_name2index(ptr - 7, &index, &tmp);
        if (rc < 0 || rc & LDD_F_SV_ALL || *tmp != '-') {
                rc = -EINVAL;
                CERROR("invalid MDT index in '%s': rc = %d\n", lodname, rc);
                return rc;
        }
        *mdt_index = index;
        return 0;
}

/**
 * Init sub llog context
 *
 * Setup update llog ctxt for update recovery threads, then start the
 * recovery thread (lod_sub_recovery_thread) to read update llog from
 * the correspondent MDT to do update recovery.
 *
 * \param[in] env       execution environment
 * \param[in] lod       lod device to do update recovery
 * \param[in] dt        sub dt device for which the recovery thread is
 *
 * \retval              0 if initialization succeeds.
 * \retval              negative errno if initialization fails.
 */
int lod_sub_init_llog(const struct lu_env *env, struct lod_device *lod,
                      struct dt_device *dt)
{
        struct obd_device *obd;
        struct lod_recovery_data *lrd = NULL;
        DECLARE_COMPLETION_ONSTACK(started);
        struct task_struct **taskp;
        struct task_struct *task;
        struct lod_tgt_desc *subtgt = NULL;
        u32 index;
        u32 master_index;
        int rc;

        ENTRY;

        rc = lodname2mdt_index(lod2obd(lod)->obd_name, &master_index);
        if (rc != 0)
                RETURN(rc);

        OBD_ALLOC_PTR(lrd);
        if (!lrd)
                RETURN(-ENOMEM);

        if (lod->lod_child == dt) {
                taskp = &lod->lod_child_recovery_task;
                index = master_index;
        } else {
                struct lu_tgt_desc *mdt;

                lod_foreach_mdt(lod, mdt) {
                        if (mdt->ltd_tgt == dt) {
                                index = mdt->ltd_index;
                                subtgt = mdt;
                                break;
                        }
                }
                LASSERT(subtgt != NULL);
                taskp = &subtgt->ltd_recovery_task;
        }

        CDEBUG(D_INFO, "%s init sub log %s\n", lod2obd(lod)->obd_name,
               dt->dd_lu_dev.ld_obd->obd_name);
        lrd->lrd_lod = lod;
        lrd->lrd_ltd = subtgt;
        lrd->lrd_task = taskp;
        lrd->lrd_idx = index;
        lrd->lrd_started = &started;

        obd = dt->dd_lu_dev.ld_obd;
        obd->obd_lvfs_ctxt.dt = dt;
        rc = llog_setup(env, obd, &obd->obd_olg, LLOG_UPDATELOG_ORIG_CTXT,
                        NULL, &llog_common_cat_ops);
        if (rc < 0) {
                CERROR("%s: cannot setup updatelog llog: rc = %d\n",
                       obd->obd_name, rc);
                GOTO(free_lrd, rc);
        }

        rc = lu_env_init(&lrd->lrd_env, LCT_LOCAL | LCT_MD_THREAD);
        if (rc != 0) {
                CERROR("%s: can't initialize env: rc = %d\n",
                       lod2obd(lod)->obd_name, rc);
                GOTO(free_lrd, rc);
        }

        /* Start the recovery thread */
        task = kthread_create(lod_sub_recovery_thread, lrd, "lod%04x_rec%04x",
                              master_index, index);
        if (IS_ERR(task)) {
                rc = PTR_ERR(task);
                CERROR("%s: cannot start recovery thread: rc = %d\n",
                       obd->obd_name, rc);
                lu_env_fini(&lrd->lrd_env);
                GOTO(out_llog, rc);
        }
        *taskp = task;
        wake_up_process(task);
        wait_for_completion(&started);

        RETURN(0);
out_llog:
        lod_sub_fini_llog(env, dt, taskp);
free_lrd:
        OBD_FREE_PTR(lrd);
        RETURN(rc);
}

/**
 * Stop sub recovery thread
 *
 * Stop sub recovery thread on all subs.
 *
 * \param[in] env       execution environment
 * \param[in] lod       lod device to do update recovery
 */
static void lod_sub_stop_recovery_threads(const struct lu_env *env,
                                          struct lod_device *lod)
{
        struct task_struct *task;
        struct lu_tgt_desc *mdt;

        /*
         * Stop the update log commit cancel threads and finish master
         * llog ctxt
         */
        task = xchg(&lod->lod_child_recovery_task, NULL);
        if (task)
                kthread_stop(task);

        lod_getref(&lod->lod_mdt_descs);
        lod_foreach_mdt(lod, mdt) {
                task = xchg(&mdt->ltd_recovery_task, NULL);
                if (task)
                        kthread_stop(task);
        }
        lod_putref(lod, &lod->lod_mdt_descs);
}

/**
 * finish all sub llog
 *
 * cleanup all of sub llog ctxt on the LOD.
 *
 * \param[in] env       execution environment
 * \param[in] lod       lod device to do update recovery
 */
static void lod_sub_fini_all_llogs(const struct lu_env *env,
                                   struct lod_device *lod)
{
        struct lu_tgt_desc *mdt;

        /*
         * Stop the update log commit cancel threads and finish master
         * llog ctxt
         */
        lod_sub_fini_llog(env, lod->lod_child,
                          &lod->lod_child_recovery_task);
        lod_getref(&lod->lod_mdt_descs);
        lod_foreach_mdt(lod, mdt)
                lod_sub_fini_llog(env, mdt->ltd_tgt,
                                  &mdt->ltd_recovery_task);
        lod_putref(lod, &lod->lod_mdt_descs);
}

static char *lod_show_update_logs_retrievers(void *data, int *size, int *count)
{
        struct lod_device *lod = (struct lod_device *)data;
        struct lu_target *lut = lod2lu_dev(lod)->ld_site->ls_tgt;
        struct lu_tgt_desc *mdt = NULL;
        char *buf;
        int len = 0;
        int rc;
        int i;

        *count = atomic_read(&lut->lut_tdtd->tdtd_recovery_threads_count);
        if (*count == 0) {
                *size = 0;
                return NULL;
        }

        *size = 5 * *count + 1;
        OBD_ALLOC(buf, *size);
        if (!buf)
                return NULL;

        *count = 0;
        memset(buf, 0, *size);

        if (!lod->lod_child_got_update_log) {
                rc = lodname2mdt_index(lod2obd(lod)->obd_name, &i);
                LASSERTF(rc == 0, "Fail to parse target index: rc = %d\n", rc);

                rc = scnprintf(buf + len, *size - len, " %04x", i);
                LASSERT(rc > 0);

                len += rc;
                (*count)++;
        }

        lod_foreach_mdt(lod, mdt) {
                if (!mdt->ltd_got_update_log) {
                        rc = scnprintf(buf + len, *size - len, " %04x",
                                       mdt->ltd_index);
                        if (unlikely(rc <= 0))
                                break;

                        len += rc;
                        (*count)++;
                }
        }

        return buf;
}

/**
 * Prepare distribute txn
 *
 * Prepare distribute txn structure for LOD
 *
 * \param[in] env       execution environment
 * \param[in] lod_device  LOD device
 *
 * \retval              0 if preparation succeeds.
 * \retval              negative errno if preparation fails.
 */
static int lod_prepare_distribute_txn(const struct lu_env *env,
                                      struct lod_device *lod)
{
        struct target_distribute_txn_data *tdtd;
        struct lu_target *lut;
        int rc;

        ENTRY;

        /* Init update recovery data */
        OBD_ALLOC_PTR(tdtd);
        if (!tdtd)
                RETURN(-ENOMEM);

        lut = lod2lu_dev(lod)->ld_site->ls_tgt;
        tdtd->tdtd_dt = &lod->lod_dt_dev;
        rc = distribute_txn_init(env, lut, tdtd,
                lu_site2seq(lod2lu_dev(lod)->ld_site)->ss_node_id);

        if (rc < 0) {
                CERROR("%s: cannot init distribute txn: rc = %d\n",
                       lod2obd(lod)->obd_name, rc);
                OBD_FREE_PTR(tdtd);
                RETURN(rc);
        }

        tdtd->tdtd_show_update_logs_retrievers =
                lod_show_update_logs_retrievers;
        tdtd->tdtd_show_retrievers_cbdata = lod;

        lut->lut_tdtd = tdtd;

        RETURN(0);
}

/**
 * Finish distribute txn
 *
 * Release the resource holding by distribute txn, i.e. stop distribute
 * txn thread.
 *
 * \param[in] env       execution environment
 * \param[in] lod       lod device
 */
static void lod_fini_distribute_txn(const struct lu_env *env,
                                    struct lod_device *lod)
{
        struct lu_target *lut;

        lut = lod2lu_dev(lod)->ld_site->ls_tgt;
        target_recovery_fini(lut->lut_obd);
        if (!lut->lut_tdtd)
                return;

        distribute_txn_fini(env, lut->lut_tdtd);

        OBD_FREE_PTR(lut->lut_tdtd);
        lut->lut_tdtd = NULL;
}

/**
 * Implementation of lu_device_operations::ldo_process_config() for LOD
 *
 * The method is called by the configuration subsystem during setup,
 * cleanup and when the configuration changes. The method processes
 * few specific commands like adding/removing the targets, changing
 * the runtime parameters.

 * \param[in] env               LU environment provided by the caller
 * \param[in] dev               lod device
 * \param[in] lcfg              configuration command to apply
 *
 * \retval 0                    on success
 * \retval negative             negated errno on error
 *
 * The examples are below.
 *
 * Add osc config log:
 * marker  20 (flags=0x01, v2.2.49.56) lustre-OST0001  'add osc'
 * add_uuid  nid=192.168.122.162@tcp(0x20000c0a87aa2)  0:  1:nidxxx
 * attach    0:lustre-OST0001-osc-MDT0001  1:osc  2:lustre-MDT0001-mdtlov_UUID
 * setup     0:lustre-OST0001-osc-MDT0001  1:lustre-OST0001_UUID  2:nid
 * lov_modify_tgts add 0:lustre-MDT0001-mdtlov  1:lustre-OST0001_UUID  2:1  3:1
 * marker  20 (flags=0x02, v2.2.49.56) lustre-OST0001  'add osc'
 *
 * Add mdc config log:
 * marker  10 (flags=0x01, v2.2.49.56) lustre-MDT0000  'add osp'
 * add_uuid  nid=192.168.122.162@tcp(0x20000c0a87aa2)  0:  1:nid
 * attach 0:lustre-MDT0000-osp-MDT0001  1:osp  2:lustre-MDT0001-mdtlov_UUID
 * setup     0:lustre-MDT0000-osp-MDT0001  1:lustre-MDT0000_UUID  2:nid
 * modify_mdc_tgts add 0:lustre-MDT0001  1:lustre-MDT0000_UUID  2:0  3:1
 * marker  10 (flags=0x02, v2.2.49.56) lustre-MDT0000_UUID  'add osp'
 */
static int lod_process_config(const struct lu_env *env,
                              struct lu_device *dev,
                              struct lustre_cfg *lcfg)
{
        struct lod_device *lod = lu2lod_dev(dev);
        struct lu_device *next = &lod->lod_child->dd_lu_dev;
        char *arg1;
        int rc = 0;

        ENTRY;

        switch (lcfg->lcfg_command) {
        case LCFG_LOV_DEL_OBD:
        case LCFG_LOV_ADD_INA:
        case LCFG_LOV_ADD_OBD:
        case LCFG_ADD_MDC: {
                u32 index;
                u32 mdt_index;
                int gen;
                /*
                 * lov_modify_tgts add  0:lov_mdsA  1:osp  2:0  3:1
                 * modify_mdc_tgts add  0:lustre-MDT0001
                 *                    1:lustre-MDT0001-mdc0002
                 *                    2:2  3:1
                 */
                arg1 = lustre_cfg_string(lcfg, 1);

                if (sscanf(lustre_cfg_buf(lcfg, 2), "%d", &index) != 1)
                        GOTO(out, rc = -EINVAL);
                if (sscanf(lustre_cfg_buf(lcfg, 3), "%d", &gen) != 1)
                        GOTO(out, rc = -EINVAL);

                if (lcfg->lcfg_command == LCFG_LOV_ADD_OBD) {
                        u32 mdt_index;

                        rc = lodname2mdt_index(lustre_cfg_string(lcfg, 0),
                                               &mdt_index);
                        if (rc != 0)
                                GOTO(out, rc);

                        rc = lod_add_device(env, lod, arg1, index, gen,
                                            mdt_index, LUSTRE_OSC_NAME, 1);
                } else if (lcfg->lcfg_command == LCFG_ADD_MDC) {
                        mdt_index = index;
                        rc = lod_add_device(env, lod, arg1, index, gen,
                                            mdt_index, LUSTRE_MDC_NAME, 1);
                } else if (lcfg->lcfg_command == LCFG_LOV_ADD_INA) {
                        /*FIXME: Add mdt_index for LCFG_LOV_ADD_INA*/
                        mdt_index = 0;
                        rc = lod_add_device(env, lod, arg1, index, gen,
                                            mdt_index, LUSTRE_OSC_NAME, 0);
                } else {
                        rc = lod_del_device(env, lod, &lod->lod_ost_descs,
                                            arg1, index, gen);
                }

                break;
        }

        case LCFG_PARAM: {
                struct obd_device *obd;
                ssize_t count;
                char *param;

                /*
                 * Check if it is activate/deactivate mdc
                 * lustre-MDTXXXX-osp-MDTXXXX.active=1
                 */
                param = lustre_cfg_buf(lcfg, 1);
                if (strstr(param, "osp") && strstr(param, ".active=")) {
                        struct lod_tgt_desc *sub_tgt = NULL;
                        struct lu_tgt_desc *mdt;
                        char *ptr;
                        char *tmp;

                        ptr = strstr(param, ".");
                        *ptr = '\0';
                        obd = class_name2obd(param);
                        if (!obd) {
                                CERROR("%s: can not find %s: rc = %d\n",
                                       lod2obd(lod)->obd_name, param, -EINVAL);
                                *ptr = '.';
                                GOTO(out, rc);
                        }

                        lod_foreach_mdt(lod, mdt) {
                                if (mdt->ltd_tgt->dd_lu_dev.ld_obd == obd) {
                                        sub_tgt = mdt;
                                        break;
                                }
                        }

                        if (!sub_tgt) {
                                CERROR("%s: can not find %s: rc = %d\n",
                                       lod2obd(lod)->obd_name, param, -EINVAL);
                                *ptr = '.';
                                GOTO(out, rc);
                        }

                        *ptr = '.';
                        tmp = strstr(param, "=");
                        tmp++;
                        if (*tmp == '1') {
                                struct llog_ctxt *ctxt;

                                obd = sub_tgt->ltd_tgt->dd_lu_dev.ld_obd;
                                ctxt = llog_get_context(obd,
                                                LLOG_UPDATELOG_ORIG_CTXT);
                                if (!ctxt) {
                                        rc = llog_setup(env, obd, &obd->obd_olg,
                                                       LLOG_UPDATELOG_ORIG_CTXT,
                                                    NULL, &llog_common_cat_ops);
                                        if (rc < 0)
                                                GOTO(out, rc);
                                } else {
                                        llog_ctxt_put(ctxt);
                                }
                                rc = lod_sub_prep_llog(env, lod,
                                                       sub_tgt->ltd_tgt,
                                                       sub_tgt->ltd_index);
                                if (rc == 0)
                                        sub_tgt->ltd_active = 1;
                        } else {
                                lod_sub_fini_llog(env, sub_tgt->ltd_tgt,
                                                  NULL);
                                sub_tgt->ltd_active = 0;
                        }
                        GOTO(out, rc);
                }


                if (strstr(param, PARAM_LOD) != NULL)
                        count = class_modify_config(lcfg, PARAM_LOD,
                                                    &lod->lod_dt_dev.dd_kobj);
                else
                        count = class_modify_config(lcfg, PARAM_LOV,
                                                    &lod->lod_dt_dev.dd_kobj);
                rc = count > 0 ? 0 : count;
                GOTO(out, rc);
        }
        case LCFG_PRE_CLEANUP: {
                lod_sub_process_config(env, lod, &lod->lod_mdt_descs, lcfg);
                lod_sub_process_config(env, lod, &lod->lod_ost_descs, lcfg);
                OBD_FAIL_TIMEOUT(OBD_FAIL_TGT_RECOVERY_CONNECT, cfs_fail_val * 2);
                next = &lod->lod_child->dd_lu_dev;
                rc = next->ld_ops->ldo_process_config(env, next, lcfg);
                if (rc != 0)
                        CDEBUG(D_HA, "%s: can't process %u: %d\n",
                               lod2obd(lod)->obd_name, lcfg->lcfg_command, rc);

                lod_sub_stop_recovery_threads(env, lod);
                lod_fini_distribute_txn(env, lod);
                lod_sub_fini_all_llogs(env, lod);
                break;
        }
        case LCFG_CLEANUP: {
                if (lod->lod_md_root) {
                        dt_object_put(env, &lod->lod_md_root->ldo_obj);
                        lod->lod_md_root = NULL;
                }

                /*
                 * do cleanup on underlying storage only when
                 * all OSPs are cleaned up, as they use that OSD as well
                 */
                lu_dev_del_linkage(dev->ld_site, dev);
                lod_sub_process_config(env, lod, &lod->lod_mdt_descs, lcfg);
                lod_sub_process_config(env, lod, &lod->lod_ost_descs, lcfg);
                next = &lod->lod_child->dd_lu_dev;
                rc = next->ld_ops->ldo_process_config(env, next, lcfg);
                if (rc)
                        CERROR("%s: can't process %u: rc = %d\n",
                               lod2obd(lod)->obd_name, lcfg->lcfg_command, rc);

                rc = obd_disconnect(lod->lod_child_exp);
                if (rc)
                        CERROR("error in disconnect from storage: rc = %d\n",
                               rc);
                break;
        }
        default:
                CERROR("%s: unknown command %u\n", lod2obd(lod)->obd_name,
                       lcfg->lcfg_command);
                rc = -EINVAL;
                break;
        }

out:
        RETURN(rc);
}

/**
 * Implementation of lu_device_operations::ldo_recovery_complete() for LOD
 *
 * The method is called once the recovery is complete. This implementation
 * distributes the notification to all the known targets.
 *
 * see include/lu_object.h for the details
 */
static int lod_recovery_complete(const struct lu_env *env,
                                 struct lu_device *dev)
{
        struct lod_device *lod = lu2lod_dev(dev);
        struct lu_device *next = &lod->lod_child->dd_lu_dev;
        struct lod_tgt_desc *tgt;
        int rc;

        ENTRY;

        LASSERT(lod->lod_recovery_completed == 0);
        lod->lod_recovery_completed = 1;

        rc = next->ld_ops->ldo_recovery_complete(env, next);

        lod_getref(&lod->lod_ost_descs);
        if (lod->lod_ost_descs.ltd_tgts_size > 0) {
                lod_foreach_ost(lod, tgt) {
                        LASSERT(tgt && tgt->ltd_tgt);
                        next = &tgt->ltd_tgt->dd_lu_dev;
                        rc = next->ld_ops->ldo_recovery_complete(env, next);
                        if (rc)
                                CERROR("%s: can't complete recovery on #%d: rc = %d\n",
                                       lod2obd(lod)->obd_name, tgt->ltd_index,
                                       rc);
                }
        }
        lod_putref(lod, &lod->lod_ost_descs);
        RETURN(rc);
}

/**
 * Init update logs on all sub device
 *
 * LOD initialize update logs on all of sub devices. Because the initialization
 * process might need FLD lookup, see llog_osd_open()->dt_locate()->...->
 * lod_object_init(), this API has to be called after LOD is initialized.
 * \param[in] env       execution environment
 * \param[in] lod       lod device
 *
 * \retval              0 if update log is initialized successfully.
 * \retval              negative errno if initialization fails.
 */
static int lod_sub_init_llogs(const struct lu_env *env, struct lod_device *lod)
{
        struct lu_tgt_desc *mdt;
        int rc;

        ENTRY;

        /*
         * llog must be setup after LOD is initialized, because llog
         * initialization include FLD lookup
         */
        LASSERT(lod->lod_initialized);

        /* Init the llog in its own stack */
        rc = lod_sub_init_llog(env, lod, lod->lod_child);
        if (rc < 0)
                RETURN(rc);

        lod_foreach_mdt(lod, mdt) {
                rc = lod_sub_init_llog(env, lod, mdt->ltd_tgt);
                if (rc != 0)
                        break;
        }

        RETURN(rc);
}

#define UPDATE_LOG_MAX_AGE      (30 * 24 * 60 * 60)     /* 30 days, in sec */

static int lod_update_log_stale(const struct lu_env *env, struct dt_object *dto,
                                struct lu_buf *buf)
{
        struct lu_attr *attr = &lod_env_info(env)->lti_attr;
        struct llog_log_hdr *hdr;
        loff_t off = 0;
        int rc;

        ENTRY;
        rc = dt_attr_get(env, dto, attr);
        if (rc)
                RETURN(rc);

        if (!(attr->la_valid & (LA_CTIME | LA_SIZE)))
                RETURN(-EFAULT);

        /* by default update log ctime is not set */
        if (attr->la_ctime == 0)
                RETURN(0);

        /* update log not expired yet */
        if (attr->la_ctime + UPDATE_LOG_MAX_AGE > ktime_get_real_seconds())
                RETURN(0);

        if (attr->la_size == 0)
                RETURN(-EFAULT);

        rc = dt_read(env, dto, buf, &off);
        if (rc < 0)
                RETURN(rc);

        hdr = (struct llog_log_hdr *)buf->lb_buf;
        if (LLOG_REC_HDR_NEEDS_SWABBING(&hdr->llh_hdr))
                lustre_swab_llog_hdr(hdr);
        /* log header is sane and flag LLOG_F_MAX_AGE|LLOG_F_RM_ON_ERR is set */
        if (rc >= sizeof(*hdr) &&
            hdr->llh_hdr.lrh_type == LLOG_HDR_MAGIC &&
            (hdr->llh_flags & (LLOG_F_MAX_AGE | LLOG_F_RM_ON_ERR)) ==
            (LLOG_F_MAX_AGE | LLOG_F_RM_ON_ERR))
                RETURN(1);

        RETURN(0);
}

/**
 * Reclaim stale update log.
 *
 * When update log is canceld (upon recovery abort), it's not destroy, but
 * canceled from catlist, and set ctime and LLOG_F_MAX_AGE|LLOG_F_RM_ON_ERR,
 * which is kept for debug. If it expired (more than UPDATE_LOG_MAX_AGE seconds
 * passed), destroy it to save space.
 */
static int lod_update_log_gc(const struct lu_env *env, struct lod_device *lod,
                             struct dt_object *dir, struct dt_object *dto,
                             const char *name)
{
        struct dt_device *dt = lod->lod_child;
        struct thandle *th;
        int rc;

        ENTRY;
        th = dt_trans_create(env, dt);
        if (IS_ERR(th))
                RETURN(PTR_ERR(th));

        rc = dt_declare_delete(env, dir, (const struct dt_key *)name, th);
        if (rc)
                GOTO(out_trans, rc);

        rc = dt_declare_destroy(env, dto, th);
        if (rc)
                GOTO(out_trans, rc);

        rc = dt_trans_start_local(env, dt, th);
        if (rc)
                GOTO(out_trans, rc);

        rc = dt_delete(env, dir, (const struct dt_key *)name, th);
        if (rc)
                GOTO(out_trans, rc);

        rc = dt_destroy(env, dto, th);
        GOTO(out_trans, rc);
out_trans:
        dt_trans_stop(env, dt, th);

        return rc;
}

/* reclaim stale update llogs under "update_log_dir" */
static int lod_update_log_dir_gc(const struct lu_env *env,
                                 struct lod_device *lod,
                                 struct dt_object *dir)
{
        struct lod_thread_info *info = lod_env_info(env);
        struct lu_buf *buf = &info->lti_linkea_buf;
        struct lu_dirent *ent = (struct lu_dirent *)info->lti_key;
        struct lu_fid *fid = &info->lti_fid;
        struct dt_it *it;
        const struct dt_it_ops *iops;
        struct dt_object *dto;
        int rc;

        ENTRY;

        if (unlikely(!dt_try_as_dir(env, dir, true)))
                RETURN(-ENOTDIR);

        lu_buf_alloc(buf, sizeof(struct llog_log_hdr));
        if (!buf->lb_buf)
                RETURN(-ENOMEM);

        iops = &dir->do_index_ops->dio_it;
        it = iops->init(env, dir, LUDA_64BITHASH);
        if (IS_ERR(it))
                GOTO(out, rc = PTR_ERR(it));

        rc = iops->load(env, it, 0);
        if (rc == 0)
                rc = iops->next(env, it);
        else if (rc > 0)
                rc = 0;

        while (rc == 0) {
                rc = iops->rec(env, it, (struct dt_rec *)ent, LUDA_64BITHASH);
                if (rc != 0)
                        break;

                ent->lde_namelen = le16_to_cpu(ent->lde_namelen);
                if (ent->lde_name[0] == '.') {
                        if (ent->lde_namelen == 1)
                                goto next;

                        if (ent->lde_namelen == 2 && ent->lde_name[1] == '.')
                                goto next;
                }

                fid_le_to_cpu(fid, &ent->lde_fid);
                dto = dt_locate(env, lod->lod_child, fid);
                if (IS_ERR(dto))
                        goto next;

                buf->lb_len = sizeof(struct llog_log_hdr);
                if (lod_update_log_stale(env, dto, buf) == 1)
                        lod_update_log_gc(env, lod, dir, dto, ent->lde_name);
                dt_object_put(env, dto);
next:
                rc = iops->next(env, it);
        }

        iops->put(env, it);
        iops->fini(env, it);
out:
        buf->lb_len = sizeof(struct llog_log_hdr);
        lu_buf_free(buf);

        RETURN(rc > 0 ? 0 : rc);
}

/**
 * Implementation of lu_device_operations::ldo_prepare() for LOD
 *
 * see include/lu_object.h for the details.
 */
static int lod_prepare(const struct lu_env *env, struct lu_device *pdev,
                       struct lu_device *cdev)
{
        struct lod_device *lod = lu2lod_dev(cdev);
        struct lu_device *next = &lod->lod_child->dd_lu_dev;
        struct lu_fid *fid = &lod_env_info(env)->lti_fid;
        int rc;
        struct dt_object *root;
        struct dt_object *dto;
        u32 index;

        ENTRY;

        rc = next->ld_ops->ldo_prepare(env, pdev, next);
        if (rc != 0) {
                CERROR("%s: prepare bottom error: rc = %d\n",
                       lod2obd(lod)->obd_name, rc);
                RETURN(rc);
        }

        lod->lod_initialized = 1;

        rc = dt_root_get(env, lod->lod_child, fid);
        if (rc < 0)
                RETURN(rc);

        root = dt_locate(env, lod->lod_child, fid);
        if (IS_ERR(root))
                RETURN(PTR_ERR(root));

        /* Create update log object */
        index = lu_site2seq(lod2lu_dev(lod)->ld_site)->ss_node_id;
        lu_update_log_fid(fid, index);

        dto = local_file_find_or_create_with_fid(env, lod->lod_child,
                                                 fid, root,
                                                 lod_update_log_name,
                                                 S_IFREG | 0644);
        if (IS_ERR(dto))
                GOTO(out_put, rc = PTR_ERR(dto));

        dt_object_put(env, dto);

        /* Create update log dir */
        lu_update_log_dir_fid(fid, index);
        dto = local_file_find_or_create_with_fid(env, lod->lod_child,
                                                 fid, root,
                                                 lod_update_log_dir_name,
                                                 S_IFDIR | 0644);
        if (IS_ERR(dto))
                GOTO(out_put, rc = PTR_ERR(dto));

        lod_update_log_dir_gc(env, lod, dto);
        dt_object_put(env, dto);

        rc = lod_prepare_distribute_txn(env, lod);
        if (rc != 0)
                GOTO(out_put, rc);

        rc = lod_sub_init_llogs(env, lod);
        if (rc != 0)
                GOTO(out_put, rc);

out_put:
        dt_object_put(env, root);

        RETURN(rc);
}

/**
 * Implementation of lu_device_operations::ldo_fid_alloc() for LOD
 *
 * Find corresponding device by passed parent and name, and allocate FID from
 * there.
 *
 * see include/lu_object.h for the details.
 */
static int lod_fid_alloc(const struct lu_env *env, struct lu_device *d,
                         struct lu_fid *fid, struct lu_object *parent,
                         const struct lu_name *name)
{
        struct lod_device *lod = lu2lod_dev(d);
        struct lod_object *lo = lu2lod_obj(parent);
        struct dt_device *next;
        int rc;

        ENTRY;

        /* if @parent is remote, we don't know whether its layout was changed,
         * always reload layout.
         */
        if (lu_object_remote(parent))
                lod_striping_free(env, lo);

        rc = lod_striping_load(env, lo);
        if (rc)
                RETURN(rc);

        if (lo->ldo_dir_stripe_count > 0 && name) {
                struct dt_object *stripe;
                int idx;

                idx = __lmv_name_to_stripe_index(lo->ldo_dir_hash_type,
                                                 lo->ldo_dir_stripe_count,
                                                 lo->ldo_dir_migrate_hash,
                                                 lo->ldo_dir_migrate_offset,
                                                 name->ln_name,
                                                 name->ln_namelen, true);
                if (idx < 0)
                        RETURN(idx);

                stripe = lo->ldo_stripe[idx];
                if (!stripe || !dt_object_exists(stripe))
                        RETURN(-ENODEV);

                next = lu2dt_dev(stripe->do_lu.lo_dev);
        } else {
                next = lod->lod_child;
        }

        rc = dt_fid_alloc(env, next, fid, parent, name);

        RETURN(rc);
}

const struct lu_device_operations lod_lu_ops = {
        .ldo_object_alloc       = lod_object_alloc,
        .ldo_process_config     = lod_process_config,
        .ldo_recovery_complete  = lod_recovery_complete,
        .ldo_prepare            = lod_prepare,
        .ldo_fid_alloc          = lod_fid_alloc,
};

/**
 * Implementation of dt_device_operations::dt_root_get() for LOD
 *
 * see include/dt_object.h for the details.
 */
static int lod_root_get(const struct lu_env *env,
                        struct dt_device *dev, struct lu_fid *f)
{
        return dt_root_get(env, dt2lod_dev(dev)->lod_child, f);
}

static void lod_statfs_sum(struct obd_statfs *sfs,
                             struct obd_statfs *ost_sfs, int *bs)
{
        while (ost_sfs->os_bsize < *bs) {
                *bs >>= 1;
                sfs->os_bsize >>= 1;
                sfs->os_bavail <<= 1;
                sfs->os_blocks <<= 1;
                sfs->os_bfree <<= 1;
                sfs->os_granted <<= 1;
        }
        while (ost_sfs->os_bsize > *bs) {
                ost_sfs->os_bsize >>= 1;
                ost_sfs->os_bavail <<= 1;
                ost_sfs->os_blocks <<= 1;
                ost_sfs->os_bfree <<= 1;
                ost_sfs->os_granted <<= 1;
        }
        sfs->os_bavail += ost_sfs->os_bavail;
        sfs->os_blocks += ost_sfs->os_blocks;
        sfs->os_bfree += ost_sfs->os_bfree;
        sfs->os_granted += ost_sfs->os_granted;
}

/**
 * Implementation of dt_device_operations::dt_statfs() for LOD
 *
 * see include/dt_object.h for the details.
 */
static int lod_statfs(const struct lu_env *env, struct dt_device *dev,
                      struct obd_statfs *sfs, struct obd_statfs_info *info)
{
        struct lod_device *lod = dt2lod_dev(dev);
        struct lu_tgt_desc *tgt;
        struct obd_statfs ost_sfs;
        u64 ost_files = 0;
        u64 ost_ffree = 0;
        int rc, bs;

        rc = dt_statfs(env, dt2lod_dev(dev)->lod_child, sfs);
        if (rc)
                GOTO(out, rc);

        bs = sfs->os_bsize;

        sfs->os_bavail = 0;
        sfs->os_blocks = 0;
        sfs->os_bfree = 0;
        sfs->os_granted = 0;

        lod_getref(&lod->lod_mdt_descs);
        lod_foreach_mdt(lod, tgt) {
                rc = dt_statfs(env, tgt->ltd_tgt, &ost_sfs);
                /* ignore errors */
                if (rc)
                        continue;
                sfs->os_files += ost_sfs.os_files;
                sfs->os_ffree += ost_sfs.os_ffree;
                lod_statfs_sum(sfs, &ost_sfs, &bs);
        }
        lod_putref(lod, &lod->lod_mdt_descs);

        /*
         * at some point we can check whether DoM is enabled and
         * decide how to account MDT space. for simplicity let's
         * just fallback to pre-DoM policy if any OST is alive
         */
        lod_getref(&lod->lod_ost_descs);
        lod_foreach_ost(lod, tgt) {
                rc = dt_statfs(env, tgt->ltd_tgt, &ost_sfs);
                /* ignore errors */
                if (rc || ost_sfs.os_bsize == 0)
                        continue;
                if (!ost_files) {
                        /*
                         * if only MDTs with DoM then report only MDT blocks,
                         * otherwise show only OST blocks, and DoM is "free"
                         */
                        sfs->os_bavail = 0;
                        sfs->os_blocks = 0;
                        sfs->os_bfree = 0;
                        sfs->os_granted = 0;
                }
                ost_files += ost_sfs.os_files;
                ost_ffree += ost_sfs.os_ffree;
                ost_sfs.os_bavail += ost_sfs.os_granted;
                lod_statfs_sum(sfs, &ost_sfs, &bs);
                LASSERTF(bs == ost_sfs.os_bsize, "%d != %d\n",
                        (int)sfs->os_bsize, (int)ost_sfs.os_bsize);
        }
        lod_putref(lod, &lod->lod_ost_descs);
        sfs->os_state |= OS_STATFS_SUM;

        /* If we have _some_ OSTs, but don't have as many free objects on the
         * OSTs as inodes on the MDTs, reduce the reported number of inodes
         * to compensate, so that the "inodes in use" number is correct.
         * This should be kept in sync with ll_statfs_internal().
         */
        if (ost_files && ost_ffree < sfs->os_ffree) {
                sfs->os_files = (sfs->os_files - sfs->os_ffree) + ost_ffree;
                sfs->os_ffree = ost_ffree;
        }

        /* a single successful statfs should be enough */
        rc = 0;

out:
        RETURN(rc);
}

/**
 * Implementation of dt_device_operations::dt_trans_create() for LOD
 *
 * Creates a transaction using local (to this node) OSD.
 *
 * see include/dt_object.h for the details.
 */
static struct thandle *lod_trans_create(const struct lu_env *env,
                                        struct dt_device *dt)
{
        struct thandle *th;

        th = top_trans_create(env, dt2lod_dev(dt)->lod_child);
        if (IS_ERR(th))
                return th;

        th->th_dev = dt;

        return th;
}

/* distributed transaction failure may cause object missing or disconnected
 * directories, check space before transaction start.
 */
static int lod_trans_space_check(const struct lu_env *env,
                                 struct lod_device *lod,
                                 struct thandle *th)
{
        struct lod_thread_info *info = lod_env_info(env);
        struct obd_statfs *sfs = &info->lti_osfs;
        struct top_thandle *top_th = container_of(th, struct top_thandle,
                                                  tt_super);
        struct top_multiple_thandle *tmt = top_th->tt_multiple_thandle;
        struct sub_thandle *st;
        int rc;

        if (likely(!tmt))
                return 0;

        list_for_each_entry(st, &tmt->tmt_sub_thandle_list, st_sub_list) {
                struct dt_device *sub_dt;

                if (st->st_sub_th == NULL)
                        continue;

                if (st->st_sub_th == top_th->tt_master_sub_thandle)
                        continue;

                sub_dt = st->st_sub_th->th_dev;
                rc = dt_statfs(env, sub_dt, sfs);
                if (rc) {
                        CDEBUG(D_INFO, "%s: fail - statfs error: rc = %d\n",
                               sub_dt->dd_lu_dev.ld_obd->obd_name, rc);
                        /* statfs may fail during recovery, skip check */
                        if (!lod->lod_recovery_completed)
                                rc = 0;
                        return rc;
                }

                if (unlikely(sfs->os_state &
                             (OS_STATFS_ENOINO | OS_STATFS_ENOSPC))) {
                        CDEBUG(D_INFO, "%s: fail - target state %x: rc = %d\n",
                               sub_dt->dd_lu_dev.ld_obd->obd_name,
                               sfs->os_state, -ENOSPC);
                        return -ENOSPC;
                }
        }

        return 0;
}

/**
 * Implementation of dt_device_operations::dt_trans_start() for LOD
 *
 * Starts the set of local transactions using the targets involved
 * in declare phase. Initial support for the distributed transactions.
 *
 * see include/dt_object.h for the details.
 */
static int lod_trans_start(const struct lu_env *env, struct dt_device *dt,
                           struct thandle *th)
{
        struct lod_device *lod = dt2lod_dev(dt);

        if (lod->lod_dist_txn_check_space) {
                int rc;

                rc = lod_trans_space_check(env, lod, th);
                if (rc)
                        return rc;
        }

        return top_trans_start(env, lod->lod_child, th);
}

static int lod_trans_cb_add(struct thandle *th,
                            struct dt_txn_commit_cb *dcb)
{
        struct top_thandle      *top_th = container_of(th, struct top_thandle,
                                                       tt_super);
        return dt_trans_cb_add(top_th->tt_master_sub_thandle, dcb);
}

/**
 * add noop update to the update records
 *
 * Add noop updates to the update records, which is only used in
 * test right now.
 *
 * \param[in] env       execution environment
 * \param[in] dt        dt device of lod
 * \param[in] th        thandle
 * \param[in] count     the count of update records to be added.
 *
 * \retval              0 if adding succeeds.
 * \retval              negative errno if adding fails.
 */
static int lod_add_noop_records(const struct lu_env *env,
                                struct dt_device *dt, struct thandle *th,
                                int count)
{
        struct top_thandle *top_th;
        struct lu_fid *fid = &lod_env_info(env)->lti_fid;
        int i;
        int rc = 0;

        top_th = container_of(th, struct top_thandle, tt_super);
        if (!top_th->tt_multiple_thandle)
                return 0;

        fid_zero(fid);
        for (i = 0; i < count; i++) {
                rc = update_record_pack(noop, th, fid);
                if (rc < 0)
                        return rc;
        }
        return rc;
}

/**
 * Implementation of dt_device_operations::dt_trans_stop() for LOD
 *
 * Stops the set of local transactions using the targets involved
 * in declare phase. Initial support for the distributed transactions.
 *
 * see include/dt_object.h for the details.
 */
static int lod_trans_stop(const struct lu_env *env, struct dt_device *dt,
                          struct thandle *th)
{
        if (OBD_FAIL_CHECK(OBD_FAIL_SPLIT_UPDATE_REC)) {
                int rc;

                rc = lod_add_noop_records(env, dt, th, 5000);
                if (rc < 0)
                        RETURN(rc);
        }
        return top_trans_stop(env, dt2lod_dev(dt)->lod_child, th);
}

/**
 * Implementation of dt_device_operations::dt_conf_get() for LOD
 *
 * Currently returns the configuration provided by the local OSD.
 *
 * see include/dt_object.h for the details.
 */
static void lod_conf_get(const struct lu_env *env,
                         const struct dt_device *dev,
                         struct dt_device_param *param)
{
        dt_conf_get(env, dt2lod_dev((struct dt_device *)dev)->lod_child, param);
}

/**
 * Implementation of dt_device_operations::dt_sync() for LOD
 *
 * Syncs all known OST targets. Very very expensive and used
 * rarely by LFSCK now. Should not be used in general.
 *
 * see include/dt_object.h for the details.
 */
static int lod_sync(const struct lu_env *env, struct dt_device *dev)
{
        struct lod_device *lod = dt2lod_dev(dev);
        struct lu_tgt_desc *tgt;
        int rc = 0;

        ENTRY;

        lod_getref(&lod->lod_ost_descs);
        lod_foreach_ost(lod, tgt) {
                if (!tgt->ltd_active)
                        continue;
                rc = dt_sync(env, tgt->ltd_tgt);
                if (rc) {
                        if (rc != -ENOTCONN) {
                                CERROR("%s: can't sync ost %u: rc = %d\n",
                                       lod2obd(lod)->obd_name, tgt->ltd_index,
                                       rc);
                                break;
                        }
                        rc = 0;
                }
        }
        lod_putref(lod, &lod->lod_ost_descs);

        if (rc)
                RETURN(rc);

        lod_getref(&lod->lod_mdt_descs);
        lod_foreach_mdt(lod, tgt) {
                if (!tgt->ltd_active)
                        continue;
                rc = dt_sync(env, tgt->ltd_tgt);
                if (rc) {
                        if (rc != -ENOTCONN) {
                                CERROR("%s: can't sync mdt %u: rc = %d\n",
                                       lod2obd(lod)->obd_name, tgt->ltd_index,
                                       rc);
                                break;
                        }
                        rc = 0;
                }
        }
        lod_putref(lod, &lod->lod_mdt_descs);

        if (rc == 0)
                rc = dt_sync(env, lod->lod_child);

        RETURN(rc);
}

/**
 * Implementation of dt_device_operations::dt_ro() for LOD
 *
 * Turns local OSD read-only, used for the testing only.
 *
 * see include/dt_object.h for the details.
 */
static int lod_ro(const struct lu_env *env, struct dt_device *dev)
{
        return dt_ro(env, dt2lod_dev(dev)->lod_child);
}

/**
 * Implementation of dt_device_operations::dt_commit_async() for LOD
 *
 * Asks local OSD to commit sooner.
 *
 * see include/dt_object.h for the details.
 */
static int lod_commit_async(const struct lu_env *env, struct dt_device *dev)
{
        return dt_commit_async(env, dt2lod_dev(dev)->lod_child);
}

static const struct dt_device_operations lod_dt_ops = {
        .dt_root_get         = lod_root_get,
        .dt_statfs           = lod_statfs,
        .dt_trans_create     = lod_trans_create,
        .dt_trans_start      = lod_trans_start,
        .dt_trans_stop       = lod_trans_stop,
        .dt_conf_get         = lod_conf_get,
        .dt_sync             = lod_sync,
        .dt_ro               = lod_ro,
        .dt_commit_async     = lod_commit_async,
        .dt_trans_cb_add     = lod_trans_cb_add,
};

/**
 * Connect to a local OSD.
 *
 * Used to connect to the local OSD at mount. OSD name is taken from the
 * configuration command passed. This connection is used to identify LU
 * site and pin the OSD from early removal.
 *
 * \param[in] env               LU environment provided by the caller
 * \param[in] lod               lod device
 * \param[in] cfg               configuration command to apply
 *
 * \retval 0                    on success
 * \retval negative             negated errno on error
 **/
static int lod_connect_to_osd(const struct lu_env *env, struct lod_device *lod,
                              struct lustre_cfg *cfg)
{
        struct obd_connect_data *data = NULL;
        struct obd_device *obd;
        char *nextdev = NULL, *p, *s;
        int rc, len = 0;

        ENTRY;

        LASSERT(cfg);
        LASSERT(lod->lod_child_exp == NULL);

        /*
         * compatibility hack: we still use old config logs
         * which specify LOV, but we need to learn underlying
         * OSD device, which is supposed to be:
         *  <fsname>-MDTxxxx-osd
         *
         * 2.x MGS generates lines like the following:
         *   #03 (176)lov_setup 0:lustre-MDT0000-mdtlov  1:(struct lov_desc)
         * 1.8 MGS generates lines like the following:
         *   #03 (168)lov_setup 0:lustre-mdtlov  1:(struct lov_desc)
         *
         * we use "-MDT" to differentiate 2.x from 1.8
         */
        p = lustre_cfg_string(cfg, 0);
        if (p && strstr(p, "-mdtlov")) {
                len = strlen(p) + 6;
                OBD_ALLOC(nextdev, len);
                if (!nextdev)
                        GOTO(out, rc = -ENOMEM);

                strcpy(nextdev, p);
                s = strstr(nextdev, "-mdtlov");
                if (unlikely(!s)) {
                        CERROR("%s: unable to parse device name: rc = %d\n",
                               lustre_cfg_string(cfg, 0), -EINVAL);
                        GOTO(out, rc = -EINVAL);
                }

                if (strstr(nextdev, "-MDT")) {
                        /* 2.x config */
                        strcpy(s, "-osd");
                } else {
                        /* 1.8 config */
                        strcpy(s, "-MDT0000-osd");
                }
        } else {
                CERROR("%s: unable to parse device name: rc = %d\n",
                       lustre_cfg_string(cfg, 0), -EINVAL);
                GOTO(out, rc = -EINVAL);
        }

        OBD_ALLOC_PTR(data);
        if (!data)
                GOTO(out, rc = -ENOMEM);

        obd = class_name2obd(nextdev);
        if (!obd) {
                CERROR("%s: can not locate next device: rc = %d\n",
                       nextdev, -ENOTCONN);
                GOTO(out, rc = -ENOTCONN);
        }

        data->ocd_connect_flags = OBD_CONNECT_VERSION;
        data->ocd_version = LUSTRE_VERSION_CODE;

        rc = obd_connect(env, &lod->lod_child_exp, obd, &obd->obd_uuid,
                         data, NULL);
        if (rc) {
                CERROR("%s: cannot connect to next dev: rc = %d\n",
                       nextdev, rc);
                GOTO(out, rc);
        }

        lod->lod_dt_dev.dd_lu_dev.ld_site =
                lod->lod_child_exp->exp_obd->obd_lu_dev->ld_site;
        LASSERT(lod->lod_dt_dev.dd_lu_dev.ld_site);
        lod->lod_child = lu2dt_dev(lod->lod_child_exp->exp_obd->obd_lu_dev);

out:
        if (data)
                OBD_FREE_PTR(data);
        if (nextdev)
                OBD_FREE(nextdev, len);
        RETURN(rc);
}

static int lod_lsfs_init(const struct lu_env *env, struct lod_device *d)
{
        struct obd_statfs sfs;
        int rc;

        rc = dt_statfs(env, d->lod_child, &sfs);
        if (rc) {
                CDEBUG(D_LAYOUT, "%s: failed to get OSD statfs, rc = %d\n",
                       lod2obd(d)->obd_name, rc);
                return rc;
        }

        /* udpate local OSD cached statfs data */
        spin_lock_init(&d->lod_lsfs_lock);
        d->lod_lsfs_age = ktime_get_seconds();
        d->lod_lsfs_total_mb = (sfs.os_blocks * sfs.os_bsize) >> 20;
        d->lod_lsfs_free_mb = (sfs.os_bfree * sfs.os_bsize) >> 20;
        return 0;
}

/**
 * Initialize LOD device at setup.
 *
 * Initializes the given LOD device using the original configuration command.
 * The function initiates a connection to the local OSD and initializes few
 * internal structures like pools, target tables, etc.
 *
 * \param[in] env               LU environment provided by the caller
 * \param[in] lod               lod device
 * \param[in] ldt               not used
 * \param[in] cfg               configuration command
 *
 * \retval 0                    on success
 * \retval negative             negated errno on error
 **/
static int lod_init0(const struct lu_env *env, struct lod_device *lod,
                     struct lu_device_type *ldt, struct lustre_cfg *cfg)
{
        struct dt_device_param ddp;
        struct obd_device *obd;
        int rc;

        ENTRY;

        obd = class_name2obd(lustre_cfg_string(cfg, 0));
        if (!obd) {
                rc = -ENODEV;
                CERROR("Cannot find obd with name '%s': rc = %d\n",
                       lustre_cfg_string(cfg, 0), rc);
                RETURN(rc);
        }

        obd->obd_lu_dev = &lod->lod_dt_dev.dd_lu_dev;
        lod->lod_dt_dev.dd_lu_dev.ld_obd = obd;
        lod->lod_dt_dev.dd_lu_dev.ld_ops = &lod_lu_ops;
        lod->lod_dt_dev.dd_ops = &lod_dt_ops;

        rc = lod_connect_to_osd(env, lod, cfg);
        if (rc)
                RETURN(rc);

        dt_conf_get(env, &lod->lod_dt_dev, &ddp);
        lod->lod_osd_max_easize = ddp.ddp_max_ea_size;
        lod->lod_dom_stripesize_max_kb = (1ULL << 10); /* 1Mb is default */
        lod->lod_max_stripecount = 0;

        /* initialize local statfs cached values */
        rc = lod_lsfs_init(env, lod);
        if (rc)
                GOTO(out_disconnect, rc);

        /* default threshold as half of total space, in MiB */
        lod->lod_dom_threshold_free_mb = lod->lod_lsfs_total_mb / 2;
        /* set default DoM stripe size based on free space amount */
        lod_dom_stripesize_recalc(lod);

        /* setup obd to be used with old lov code */
        rc = lod_pools_init(lod, cfg);
        if (rc)
                GOTO(out_disconnect, rc);

        rc = lod_procfs_init(lod);
        if (rc)
                GOTO(out_pools, rc);

        spin_lock_init(&lod->lod_lock);
        spin_lock_init(&lod->lod_connects_lock);
        lu_tgt_descs_init(&lod->lod_mdt_descs, true);
        lu_tgt_descs_init(&lod->lod_ost_descs, false);
        lu_qos_rr_init(&lod->lod_mdt_descs.ltd_qos.lq_rr);
        lu_qos_rr_init(&lod->lod_ost_descs.ltd_qos.lq_rr);
        lod->lod_dist_txn_check_space = 1;

        RETURN(0);

out_pools:
        lod_pools_fini(lod);
out_disconnect:
        obd_disconnect(lod->lod_child_exp);
        RETURN(rc);
}

/**
 * Implementation of lu_device_type_operations::ldto_device_free() for LOD
 *
 * Releases the memory allocated for LOD device.
 *
 * see include/lu_object.h for the details.
 */
static struct lu_device *lod_device_free(const struct lu_env *env,
                                         struct lu_device *lu)
{
        struct lod_device *lod = lu2lod_dev(lu);
        struct lu_device  *next = &lod->lod_child->dd_lu_dev;

        ENTRY;

        if (atomic_read(&lu->ld_site->ls_obj_hash.nelems)) {
                lu_site_print(env, lu->ld_site, &lu->ld_ref, D_ERROR,
                              lu_cdebug_printer);
        }
        LASSERTF(atomic_read(&lu->ld_ref) == 0, "lu is %p\n", lu);
        dt_device_fini(&lod->lod_dt_dev);
        OBD_FREE_PTR(lod);
        RETURN(next);
}

/**
 * Implementation of lu_device_type_operations::ldto_device_alloc() for LOD
 *
 * Allocates LOD device and calls the helpers to initialize it.
 *
 * see include/lu_object.h for the details.
 */
static struct lu_device *lod_device_alloc(const struct lu_env *env,
                                          struct lu_device_type *type,
                                          struct lustre_cfg *lcfg)
{
        struct lod_device *lod;
        struct lu_device *lu_dev;

        OBD_ALLOC_PTR(lod);
        if (!lod) {
                lu_dev = ERR_PTR(-ENOMEM);
        } else {
                int rc;

                lu_dev = lod2lu_dev(lod);
                dt_device_init(&lod->lod_dt_dev, type);
                rc = lod_init0(env, lod, type, lcfg);
                if (rc != 0) {
                        lod_device_free(env, lu_dev);
                        lu_dev = ERR_PTR(rc);
                }
        }

        return lu_dev;
}

static void lod_avoid_guide_fini(struct lod_avoid_guide *lag)
{
        if (lag->lag_oss_avoid_array)
                OBD_FREE_PTR_ARRAY(lag->lag_oss_avoid_array,
                                   lag->lag_oaa_size);
        bitmap_free(lag->lag_ost_avoid_bitmap);
}

/**
 * Implementation of lu_device_type_operations::ldto_device_fini() for LOD
 *
 * Releases the internal resources used by LOD device.
 *
 * see include/lu_object.h for the details.
 */
static struct lu_device *lod_device_fini(const struct lu_env *env,
                                         struct lu_device *d)
{
        struct lod_device *lod = lu2lod_dev(d);
        int rc;

        ENTRY;

        lod_pools_fini(lod);

        lod_procfs_fini(lod);

        rc = lod_fini_tgt(env, lod, &lod->lod_ost_descs);
        if (rc)
                CERROR("%s: can not fini ost descriptors: rc =  %d\n",
                        lod2obd(lod)->obd_name, rc);

        rc = lod_fini_tgt(env, lod, &lod->lod_mdt_descs);
        if (rc)
                CERROR("%s: can not fini mdt descriptors: rc =  %d\n",
                        lod2obd(lod)->obd_name, rc);

        RETURN(NULL);
}

/**
 * Implementation of obd_ops::o_connect() for LOD
 *
 * Used to track all the users of this specific LOD device,
 * so the device stays up until the last user disconnected.
 *
 * \param[in] env               LU environment provided by the caller
 * \param[out] exp              export the caller will be using to access LOD
 * \param[in] obd               OBD device representing LOD device
 * \param[in] cluuid            unique identifier of the caller
 * \param[in] data              not used
 * \param[in] localdata         not used
 *
 * \retval 0                    on success
 * \retval negative             negated errno on error
 **/
static int lod_obd_connect(const struct lu_env *env, struct obd_export **exp,
                           struct obd_device *obd, struct obd_uuid *cluuid,
                           struct obd_connect_data *data, void *localdata)
{
        struct lod_device *lod = lu2lod_dev(obd->obd_lu_dev);
        struct lustre_handle conn;
        int rc;

        ENTRY;

        CDEBUG(D_CONFIG, "connect #%d\n", lod->lod_connects);

        rc = class_connect(&conn, obd, cluuid);
        if (rc)
                RETURN(rc);

        *exp = class_conn2export(&conn);

        spin_lock(&lod->lod_connects_lock);
        lod->lod_connects++;
        /* at the moment we expect the only user */
        LASSERT(lod->lod_connects == 1);
        spin_unlock(&lod->lod_connects_lock);

        RETURN(0);
}

/**
 *
 * Implementation of obd_ops::o_disconnect() for LOD
 *
 * When the caller doesn't need to use this LOD instance, it calls
 * obd_disconnect() and LOD releases corresponding export/reference count.
 * Once all the users gone, LOD device is released.
 *
 * \param[in] exp               export provided to the caller in obd_connect()
 *
 * \retval 0                    on success
 * \retval negative             negated errno on error
 **/
static int lod_obd_disconnect(struct obd_export *exp)
{
        struct obd_device *obd = exp->exp_obd;
        struct lod_device *lod = lu2lod_dev(obd->obd_lu_dev);
        int rc, release = 0;

        ENTRY;

        /* Only disconnect the underlying layers on the final disconnect. */
        spin_lock(&lod->lod_connects_lock);
        lod->lod_connects--;
        if (lod->lod_connects != 0) {
                /* why should there be more than 1 connect? */
                spin_unlock(&lod->lod_connects_lock);
                CERROR("%s: disconnect #%d\n", exp->exp_obd->obd_name,
                       lod->lod_connects);
                goto out;
        }
        spin_unlock(&lod->lod_connects_lock);

        /* the last user of lod has gone, let's release the device */
        release = 1;

out:
        rc = class_disconnect(exp); /* bz 9811 */

        if (rc == 0 && release)
                class_manual_cleanup(obd);
        RETURN(rc);
}

LU_KEY_INIT(lod, struct lod_thread_info);

static void lod_key_fini(const struct lu_context *ctx,
                struct lu_context_key *key, void *data)
{
        struct lod_thread_info *info = data;
        struct lod_layout_component *lds =
                                info->lti_def_striping.lds_def_comp_entries;

        /*
         * allocated in lod_get_lov_ea
         * XXX: this is overload, a tread may have such store but used only
         * once. Probably better would be pool of such stores per LOD.
         */
        if (info->lti_ea_store) {
                OBD_FREE_LARGE(info->lti_ea_store, info->lti_ea_store_size);
                info->lti_ea_store = NULL;
                info->lti_ea_store_size = 0;
        }
        lu_buf_free(&info->lti_linkea_buf);

        if (lds)
                lod_free_def_comp_entries(&info->lti_def_striping);

        if (info->lti_comp_size > 0)
                OBD_FREE_PTR_ARRAY(info->lti_comp_idx,
                                   info->lti_comp_size);

        lod_avoid_guide_fini(&info->lti_avoid);

        OBD_FREE_PTR(info);
}

/* context key: lod_thread_key */
LU_CONTEXT_KEY_DEFINE(lod, LCT_MD_THREAD);

LU_TYPE_INIT_FINI(lod, &lod_thread_key);

static const struct lu_device_type_operations lod_device_type_ops = {
        .ldto_init              = lod_type_init,
        .ldto_fini              = lod_type_fini,

        .ldto_start             = lod_type_start,
        .ldto_stop              = lod_type_stop,

        .ldto_device_alloc      = lod_device_alloc,
        .ldto_device_free       = lod_device_free,

        .ldto_device_fini       = lod_device_fini
};

static struct lu_device_type lod_device_type = {
        .ldt_tags     = LU_DEVICE_DT,
        .ldt_name     = LUSTRE_LOD_NAME,
        .ldt_ops      = &lod_device_type_ops,
        .ldt_ctx_tags = LCT_MD_THREAD,
};

/**
 * Implementation of obd_ops::o_get_info() for LOD
 *
 * Currently, there is only one supported key: KEY_OSP_CONNECTED , to provide
 * the caller binary status whether LOD has seen connection to any OST target.
 * It will also check if the MDT update log context being initialized (if
 * needed).
 *
 * \param[in] env               LU environment provided by the caller
 * \param[in] exp               export of the caller
 * \param[in] keylen            len of the key
 * \param[in] key               the key
 * \param[in] vallen            not used
 * \param[in] val               not used
 *
 * \retval                      0 if a connection was seen
 * \retval                      -EAGAIN if LOD isn't running yet or no
 *                              connection has been seen yet
 * \retval                      -EINVAL if not supported key is requested
 **/
static int lod_obd_get_info(const struct lu_env *env, struct obd_export *exp,
                            u32 keylen, void *key, u32 *vallen, void *val)
{
        int rc = -EINVAL;

        if (KEY_IS(KEY_OSP_CONNECTED)) {
                struct obd_device *obd = exp->exp_obd;
                struct lod_device *d;
                struct lod_tgt_desc *tgt;
                int rc = 1;

                if (!obd->obd_set_up || obd->obd_stopping)
                        RETURN(-EAGAIN);

                d = lu2lod_dev(obd->obd_lu_dev);
                lod_getref(&d->lod_ost_descs);
                lod_foreach_ost(d, tgt) {
                        rc = obd_get_info(env, tgt->ltd_exp, keylen, key,
                                          vallen, val);
                        /* one healthy device is enough */
                        if (rc == 0)
                                break;
                }
                lod_putref(d, &d->lod_ost_descs);

                lod_getref(&d->lod_mdt_descs);
                lod_foreach_mdt(d, tgt) {
                        struct llog_ctxt *ctxt;

                        if (!tgt->ltd_active)
                                continue;

                        ctxt = llog_get_context(tgt->ltd_tgt->dd_lu_dev.ld_obd,
                                                LLOG_UPDATELOG_ORIG_CTXT);
                        if (!ctxt) {
                                CDEBUG(D_INFO, "%s: %s is not ready.\n",
                                       obd->obd_name,
                                      tgt->ltd_tgt->dd_lu_dev.ld_obd->obd_name);
                                rc = -EAGAIN;
                                break;
                        }
                        if (!ctxt->loc_handle) {
                                CDEBUG(D_INFO, "%s: %s is not ready.\n",
                                       obd->obd_name,
                                      tgt->ltd_tgt->dd_lu_dev.ld_obd->obd_name);
                                rc = -EAGAIN;
                                llog_ctxt_put(ctxt);
                                break;
                        }
                        llog_ctxt_put(ctxt);
                }
                lod_putref(d, &d->lod_mdt_descs);

                RETURN(rc);
        }

        RETURN(rc);
}

static int lod_obd_set_info_async(const struct lu_env *env,
                                  struct obd_export *exp,
                                  u32 keylen, void *key,
                                  u32 vallen, void *val,
                                  struct ptlrpc_request_set *set)
{
        struct obd_device *obd = class_exp2obd(exp);
        struct lod_device *d;
        struct lod_tgt_desc *tgt;
        int no_set = 0;
        int rc = 0, rc2;

        ENTRY;

        if (!set) {
                no_set = 1;
                set = ptlrpc_prep_set();
                if (!set)
                        RETURN(-ENOMEM);
        }

        d = lu2lod_dev(obd->obd_lu_dev);
        lod_getref(&d->lod_ost_descs);
        lod_foreach_ost(d, tgt) {
                if (!tgt->ltd_active)
                        continue;

                rc2 = obd_set_info_async(env, tgt->ltd_exp, keylen, key,
                                         vallen, val, set);
                if (rc2 != 0 && rc == 0)
                        rc = rc2;
        }
        lod_putref(d, &d->lod_ost_descs);

        lod_getref(&d->lod_mdt_descs);
        lod_foreach_mdt(d, tgt) {
                if (!tgt->ltd_active)
                        continue;
                rc2 = obd_set_info_async(env, tgt->ltd_exp, keylen, key,
                                         vallen, val, set);
                if (rc2 != 0 && rc == 0)
                        rc = rc2;
        }
        lod_putref(d, &d->lod_mdt_descs);


        if (no_set) {
                rc2 = ptlrpc_set_wait(env, set);
                if (rc2 == 0 && rc == 0)
                        rc = rc2;
                ptlrpc_set_destroy(set);
        }
        RETURN(rc);
}


#define QMT0_DEV_NAME_LEN (LUSTRE_MAXFSNAME + sizeof("-QMT0000"))
static struct obd_device *obd_find_qmt0(char *obd_name)
{
        char qmt_name[QMT0_DEV_NAME_LEN];
        struct obd_device *qmt = NULL;

        if (!server_name2fsname(obd_name, qmt_name, NULL)) {
                strlcat(qmt_name, "-QMT0000", QMT0_DEV_NAME_LEN);
                qmt = class_name2obd(qmt_name);
        }

        return qmt;
}

/* Run QMT0000 pool operations only for MDT0000 */
static inline bool lod_pool_need_qmt0(const char *obd_name)
{
        __u32 idx;
        int type;

        type = server_name2index(obd_name, &idx, NULL);

        return type == LDD_F_SV_TYPE_MDT && idx == 0;
}

static int lod_pool_new_q(struct obd_device *obd, char *poolname)
{
        int err = lod_pool_new(obd, poolname);

        if (!err && lod_pool_need_qmt0(obd->obd_name)) {
                obd = obd_find_qmt0(obd->obd_name);
                if (obd)
                        obd_pool_new(obd, poolname);
        }

        return err;
}

static int lod_pool_remove_q(struct obd_device *obd, char *poolname,
                             char *ostname)
{
        int err = lod_pool_remove(obd, poolname, ostname);

        if (!err && lod_pool_need_qmt0(obd->obd_name)) {
                obd = obd_find_qmt0(obd->obd_name);
                if (obd)
                        obd_pool_rem(obd, poolname, ostname);
        }

        return err;
}

static int lod_pool_add_q(struct obd_device *obd, char *poolname, char *ostname)
{
        int err = lod_pool_add(obd, poolname, ostname);

        if (!err && lod_pool_need_qmt0(obd->obd_name)) {
                obd = obd_find_qmt0(obd->obd_name);
                if (obd)
                        obd_pool_add(obd, poolname, ostname);
        }

        return err;
}

static int lod_pool_del_q(struct obd_device *obd, char *poolname)
{
        int err = lod_pool_del(obd, poolname);

        if (!err && lod_pool_need_qmt0(obd->obd_name)) {
                obd = obd_find_qmt0(obd->obd_name);
                if (obd)
                        obd_pool_del(obd, poolname);
        }

        return err;
}

static inline int lod_sub_print_llog(const struct lu_env *env,
                                     struct dt_device *dt, void *data)
{
        struct llog_ctxt *ctxt;
        int rc = 0;

        ENTRY;
        ctxt = llog_get_context(dt->dd_lu_dev.ld_obd,
                                LLOG_UPDATELOG_ORIG_CTXT);
        if (!ctxt)
                RETURN(0);

        if (ctxt->loc_handle) {
                struct llog_print_data *lprd = data;
                struct obd_ioctl_data *ioc_data = lprd->lprd_data;
                int l, remains;
                long from;
                char *out;

                LASSERT(ioc_data);
                if (ioc_data->ioc_inllen1 > 0) {
                        remains = ioc_data->ioc_inllen4 +
                                  round_up(ioc_data->ioc_inllen1, 8) +
                                  round_up(ioc_data->ioc_inllen2, 8) +
                                  round_up(ioc_data->ioc_inllen3, 8);

                        rc = kstrtol(ioc_data->ioc_inlbuf2, 0, &from);
                        if (rc)
                                GOTO(ctxt_put, rc);

                        /* second iteration from jt_llog_print_iter() */
                        if (from > 1)
                                GOTO(ctxt_put, rc = 0);

                        out = ioc_data->ioc_bulk;
                        ioc_data->ioc_inllen1 = 0;
                } else {
                        out = ioc_data->ioc_bulk + ioc_data->ioc_offset;
                        remains = ioc_data->ioc_count;
                }

                l = snprintf(out, remains, "%s [catalog]: "DFID"\n",
                             ctxt->loc_obd->obd_name,
                             PLOGID(&ctxt->loc_handle->lgh_id));
                out += l;
                remains -= l;
                if (remains <= 0) {
                        CERROR("%s: not enough space for print log records: rc = %d\n",
                               ctxt->loc_obd->obd_name, -LLOG_EEMPTY);
                        GOTO(ctxt_put, rc = -LLOG_EEMPTY);
                }

                ioc_data->ioc_offset += l;
                ioc_data->ioc_count = remains;

                rc = llog_process_or_fork(env, ctxt->loc_handle, llog_print_cb,
                                          data, NULL, false);
        }
        GOTO(ctxt_put, rc);
ctxt_put:
        llog_ctxt_put(ctxt);

        return rc;
}

/* print update catalog and update logs FID of all sub devices */
static int lod_llog_print(const struct lu_env *env, struct lod_device *lod,
                          void *data)
{
        struct lod_tgt_desc *mdt;
        bool empty = true;
        int rc = 0;

        ENTRY;
        rc = lod_sub_print_llog(env, lod->lod_child, data);
        if (!rc) {
                empty = false;
        } else if (rc == -LLOG_EEMPTY) {
                rc = 0;
        } else {
                CERROR("%s: llog_print failed: rc = %d\n",
                       lod2obd(lod)->obd_name, rc);
                RETURN(rc);
        }

        lod_getref(&lod->lod_mdt_descs);
        lod_foreach_mdt(lod, mdt) {
                rc = lod_sub_print_llog(env, mdt->ltd_tgt, data);
                if (!rc) {
                        empty = false;
                } else if (rc == -LLOG_EEMPTY) {
                        rc = 0;
                } else {
                        CERROR("%s: llog_print of MDT %u failed: rc = %d\n",
                               lod2obd(lod)->obd_name, mdt->ltd_index, rc);
                        break;
                }
        }
        lod_putref(lod, &lod->lod_mdt_descs);

        RETURN(rc ? rc : empty ? -LLOG_EEMPTY : 0);
}

/* cancel update catalog from update catlist */
static int lod_llog_cancel(const struct lu_env *env, struct lod_device *lod)
{
        struct lod_tgt_desc *tgt;
        int index;
        int rc;
        int rc2;

        rc = lodname2mdt_index(lod2obd(lod)->obd_name, (__u32 *)&index);
        if (rc < 0)
                return rc;

        rc = lod_sub_cancel_llog(env, lod, lod->lod_child, index);

        lod_getref(&lod->lod_mdt_descs);
        lod_foreach_mdt(lod, tgt) {
                LASSERT(tgt && tgt->ltd_tgt);
                rc2 = lod_sub_cancel_llog(env, lod, tgt->ltd_tgt,
                                          tgt->ltd_index);
                if (rc2 && !rc)
                        rc = rc2;
        }
        lod_putref(lod, &lod->lod_mdt_descs);

        return rc;
}

static int lod_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
                         void *karg, void __user *uarg)
{
        struct obd_device *obd = exp->exp_obd;
        struct lod_device *lod = lu2lod_dev(obd->obd_lu_dev);
        struct obd_ioctl_data *data = karg;
        struct lu_env env;
        int rc;

        ENTRY;
        CDEBUG(D_IOCTL, "%s: cmd=%x len=%u karg=%pK uarg=%pK\n",
               obd->obd_name, cmd, len, karg, uarg);

        rc = lu_env_init(&env, LCT_LOCAL | LCT_MD_THREAD);
        if (rc) {
                CERROR("%s: can't initialize env: rc = %d\n",
                       lod2obd(lod)->obd_name, rc);
                RETURN(rc);
        }

        switch (cmd) {
        case OBD_IOC_LLOG_PRINT: {
                struct llog_print_data lprd = {
                        .lprd_data = data,
                        .lprd_raw = data->ioc_u32_1,
                };
                char *logname;

                logname = data->ioc_inlbuf1;
                if (strcmp(logname, lod_update_log_name) != 0) {
                        rc = -EINVAL;
                        CERROR("%s: llog iocontrol support %s only: rc = %d\n",
                               lod2obd(lod)->obd_name, lod_update_log_name, rc);
                        RETURN(rc);
                }

                LASSERT(data->ioc_inllen1 > 0);
                rc = lod_llog_print(&env, lod, &lprd);
                break;
        }
        case OBD_IOC_LLOG_CANCEL:
                rc = lod_llog_cancel(&env, lod);
                break;
        default:
                rc = OBD_IOC_ERROR(obd->obd_name, cmd, "unrecognized", -ENOTTY);
                break;
        }
        lu_env_fini(&env);

        RETURN(rc);
}

static const struct obd_ops lod_obd_device_ops = {
        .o_owner        = THIS_MODULE,
        .o_connect      = lod_obd_connect,
        .o_disconnect   = lod_obd_disconnect,
        .o_get_info     = lod_obd_get_info,
        .o_set_info_async = lod_obd_set_info_async,
        .o_pool_new     = lod_pool_new_q,
        .o_pool_rem     = lod_pool_remove_q,
        .o_pool_add     = lod_pool_add_q,
        .o_pool_del     = lod_pool_del_q,
        .o_iocontrol    = lod_iocontrol,
};

static int __init lod_init(void)
{
        struct obd_type *sym;
        int rc;

        rc = lu_kmem_init(lod_caches);
        if (rc)
                return rc;

        rc = class_register_type(&lod_obd_device_ops, NULL, true,
                                 LUSTRE_LOD_NAME, &lod_device_type);
        if (rc) {
                lu_kmem_fini(lod_caches);
                return rc;
        }

        /* create "lov" entry for compatibility purposes */
        sym = class_add_symlinks(LUSTRE_LOV_NAME, true);
        if (IS_ERR(sym)) {
                rc = PTR_ERR(sym);
                /* does real "lov" already exist ? */
                if (rc == -EEXIST)
                        rc = 0;
        }

        return rc;
}

static void __exit lod_exit(void)
{
        struct obd_type *sym = class_search_type(LUSTRE_LOV_NAME);

        /* if this was never fully initialized by the lov layer
         * then we are responsible for freeing this obd_type
         */
        if (sym) {
                /* final put if we manage this obd type */
                if (sym->typ_sym_filter)
                        kobject_put(&sym->typ_kobj);
                /* put reference taken by class_search_type */
                kobject_put(&sym->typ_kobj);
        }

        class_unregister_type(LUSTRE_LOD_NAME);
        lu_kmem_fini(lod_caches);
}

MODULE_AUTHOR("OpenSFS, Inc. <http://www.lustre.org/>");
MODULE_DESCRIPTION("Lustre Logical Object Device ("LUSTRE_LOD_NAME")");
MODULE_VERSION(LUSTRE_VERSION_STRING);
MODULE_LICENSE("GPL");

module_init(lod_init);
module_exit(lod_exit);