[fs/lustre-release.git] / lustre / osp / osp_precreate.c

/*
 * GPL HEADER START
 *
 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
 *
 * This program is free software; you can redistribute it and/or modify
 * it under the terms of the GNU General Public License version 2 only,
 * as published by the Free Software Foundation.
 *
 * This program is distributed in the hope that it will be useful, but
 * WITHOUT ANY WARRANTY; without even the implied warranty of
 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
 * General Public License version 2 for more details (a copy is included
 * in the LICENSE file that accompanied this code).
 *
 * You should have received a copy of the GNU General Public License
 * version 2 along with this program; If not, see
 * http://www.gnu.org/licenses/gpl-2.0.html
 *
 * GPL HEADER END
 */
/*
 * Copyright (c) 2007, 2010, Oracle and/or its affiliates. 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 is a trademark of Sun Microsystems, Inc.
 *
 * lustre/osp/osp_precreate.c
 *
 * Lustre OST Proxy Device
 *
 * Author: Alex Zhuravlev <alexey.zhuravlev@intel.com>
 * Author: Mikhail Pershin <mike.pershin@intel.com>
 * Author: Di Wang <di.wang@intel.com>
 */

#define DEBUG_SUBSYSTEM S_MDS

#include <linux/kthread.h>

#include <lustre_obdo.h>

#include "osp_internal.h"

/*
 * there are two specific states to take care about:
 *
 * = import is disconnected =
 *
 * = import is inactive =
 *   in this case osp_declare_create() returns an error
 *
 */

/**
 * Check whether statfs data is expired
 *
 * OSP device caches statfs data for the target, the function checks
 * whether the data is expired or not.
 *
 * \param[in] d         OSP device
 *
 * \retval              0 - not expired, 1 - expired
 */
static inline int osp_statfs_need_update(struct osp_device *d)
{
        return !ktime_before(ktime_get(), d->opd_statfs_fresh_till);
}

/*
 * OSP tries to maintain pool of available objects so that calls to create
 * objects don't block most of time
 *
 * each time OSP gets connected to OST, we should start from precreation cleanup
 */
static inline bool osp_precreate_running(struct osp_device *d)
{
        return !!(d->opd_pre_thread.t_flags & SVC_RUNNING);
}

static inline bool osp_precreate_stopped(struct osp_device *d)
{
        return !!(d->opd_pre_thread.t_flags & SVC_STOPPED);
}

static void osp_statfs_timer_cb(cfs_timer_cb_arg_t data)
{
        struct osp_device *d = cfs_from_timer(d, data, opd_statfs_timer);

        LASSERT(d);
        if (osp_precreate_running(d))
                wake_up(&d->opd_pre_waitq);
}

static void osp_pre_update_msfs(struct osp_device *d, struct obd_statfs *msfs);

/*
 * The function updates current precreation status if broken, and
 * updates that cached statfs state if functional, then wakes up waiters.
 * We don't clear opd_pre_status directly here, but rather leave this
 * to osp_pre_update_msfs() to do if everything is OK so that we don't
 * have a race to clear opd_pre_status and then set it to -ENOSPC again.
 *
 * \param[in] d         OSP device
 * \param[in] msfs      statfs data
 * \param[in] rc        new precreate status for device \a d
 */
static void osp_pre_update_status_msfs(struct osp_device *d,
                                       struct obd_statfs *msfs, int rc)
{
        if (rc)
                d->opd_pre_status = rc;
        else
                osp_pre_update_msfs(d, msfs);

        wake_up(&d->opd_pre_user_waitq);
}

/* Pass in the old statfs data in case the limits have changed */
void osp_pre_update_status(struct osp_device *d, int rc)
{
        osp_pre_update_status_msfs(d, &d->opd_statfs, rc);
}


/**
 * RPC interpret callback for OST_STATFS RPC
 *
 * An interpretation callback called by ptlrpc for OST_STATFS RPC when it is
 * replied by the target. It's used to maintain statfs cache for the target.
 * The function fills data from the reply if successful and schedules another
 * update.
 *
 * \param[in] env       LU environment provided by the caller
 * \param[in] req       RPC replied
 * \param[in] aa        callback data
 * \param[in] rc        RPC result
 *
 * \retval 0            on success
 * \retval negative     negated errno on error
 */
static int osp_statfs_interpret(const struct lu_env *env,
                                struct ptlrpc_request *req, void *args, int rc)
{
        union ptlrpc_async_args *aa = args;
        struct obd_import *imp = req->rq_import;
        struct obd_statfs *msfs;
        struct osp_device *d;
        u64 maxage_ns;

        ENTRY;

        aa = ptlrpc_req_async_args(aa, req);
        d = aa->pointer_arg[0];
        LASSERT(d);

        if (rc != 0)
                GOTO(out, rc);

        msfs = req_capsule_server_get(&req->rq_pill, &RMF_OBD_STATFS);
        if (msfs == NULL)
                GOTO(out, rc = -EPROTO);

        if (d->opd_pre)
                osp_pre_update_status_msfs(d, msfs, 0);
        else
                d->opd_statfs = *msfs;

        /* schedule next update */
        maxage_ns = d->opd_statfs_maxage * NSEC_PER_SEC;
        d->opd_statfs_fresh_till = ktime_add_ns(ktime_get(), maxage_ns);
        mod_timer(&d->opd_statfs_timer,
                  jiffies + cfs_time_seconds(d->opd_statfs_maxage));
        d->opd_statfs_update_in_progress = 0;

        CDEBUG(D_CACHE, "updated statfs %p\n", d);

        RETURN(0);
out:
        /* couldn't update statfs, try again with a small delay */
        d->opd_statfs_fresh_till = ktime_add_ns(ktime_get(), 10 * NSEC_PER_SEC);
        d->opd_statfs_update_in_progress = 0;
        if (d->opd_pre != NULL && osp_precreate_running(d))
                wake_up(&d->opd_pre_waitq);

        if (req->rq_import_generation == imp->imp_generation)
                CDEBUG(D_CACHE, "%s: couldn't update statfs: rc = %d\n",
                       d->opd_obd->obd_name, rc);
        RETURN(rc);
}

/**
 * Send OST_STATFS RPC
 *
 * Sends OST_STATFS RPC to refresh cached statfs data for the target.
 * Also disables scheduled updates as times OSP may need to refresh
 * statfs data before expiration. The function doesn't block, instead
 * an interpretation callback osp_statfs_interpret() is used.
 *
 * \param[in] d         OSP device
 */
static int osp_statfs_update(const struct lu_env *env, struct osp_device *d)
{
        u64 expire = obd_timeout * 1000 * NSEC_PER_SEC;
        struct ptlrpc_request   *req;
        struct obd_import       *imp;
        union ptlrpc_async_args *aa;
        int rc;

        ENTRY;

        CDEBUG(D_CACHE, "going to update statfs\n");

        imp = d->opd_obd->u.cli.cl_import;
        LASSERT(imp);

        req = ptlrpc_request_alloc(imp,
                           d->opd_pre ? &RQF_OST_STATFS : &RQF_MDS_STATFS);
        if (req == NULL)
                RETURN(-ENOMEM);

        rc = ptlrpc_request_pack(req,
                         d->opd_pre ? LUSTRE_OST_VERSION : LUSTRE_MDS_VERSION,
                         d->opd_pre ? OST_STATFS : MDS_STATFS);
        if (rc) {
                ptlrpc_request_free(req);
                RETURN(rc);
        }
        ptlrpc_request_set_replen(req);
        if (d->opd_pre)
                req->rq_request_portal = OST_CREATE_PORTAL;
        ptlrpc_at_set_req_timeout(req);

        req->rq_interpret_reply = osp_statfs_interpret;
        aa = ptlrpc_req_async_args(aa, req);
        aa->pointer_arg[0] = d;

        /*
         * no updates till reply
         */
        del_timer(&d->opd_statfs_timer);
        d->opd_statfs_fresh_till = ktime_add_ns(ktime_get(), expire);
        d->opd_statfs_update_in_progress = 1;

        ptlrpcd_add_req(req);

        /* we still want to sync changes if no new changes are coming */
        if (ktime_before(ktime_get(), d->opd_sync_next_commit_cb))
                GOTO(out, rc);

        if (atomic_read(&d->opd_sync_changes)) {
                struct thandle *th;

                th = dt_trans_create(env, d->opd_storage);
                if (IS_ERR(th)) {
                        CERROR("%s: can't sync\n", d->opd_obd->obd_name);
                        GOTO(out, rc);
                }
                rc = dt_trans_start_local(env, d->opd_storage, th);
                if (rc == 0) {
                        CDEBUG(D_OTHER, "%s: sync forced, %d changes\n",
                               d->opd_obd->obd_name,
                               atomic_read(&d->opd_sync_changes));
                        osp_sync_add_commit_cb_1s(env, d, th);
                        dt_trans_stop(env, d->opd_storage, th);
                }
        }

out:
        RETURN(0);
}

/**
 * Schedule an immediate update for statfs data
 *
 * If cached statfs data claim no free space, but OSP has got a request to
 * destroy an object (so release some space probably), then we may need to
 * refresh cached statfs data sooner than planned. The function checks there
 * is no statfs update going and schedules immediate update if so.
 * XXX: there might be a case where removed object(s) do not add free space (empty
 * object). If the number of such deletions is high, then we can start to update
 * statfs too often causing a RPC storm. some throttling is needed...
 *
 * \param[in] d         OSP device where statfs data needs to be refreshed
 */
void osp_statfs_need_now(struct osp_device *d)
{
        if (!d->opd_statfs_update_in_progress) {
                /*
                 * if current status is -ENOSPC (lack of free space on OST)
                 * then we should poll OST immediately once object destroy
                 * is replied
                 */
                d->opd_statfs_fresh_till = ktime_sub_ns(ktime_get(), NSEC_PER_SEC);
                del_timer(&d->opd_statfs_timer);
                wake_up(&d->opd_pre_waitq);
        }
}

/**
 * Return number of precreated objects
 *
 * A simple helper to calculate the number of precreated objects on the device.
 *
 * \param[in] env       LU environment provided by the caller
 * \param[in] osp       OSP device
 *
 * \retval              the number of the precreated objects
 */
static inline int osp_objs_precreated(const struct lu_env *env,
                                      struct osp_device *osp)
{
        return osp_fid_diff(&osp->opd_pre_last_created_fid,
                            &osp->opd_pre_used_fid);
}

/**
 * Check pool of precreated objects is nearly empty
 *
 * We should not wait till the pool of the precreated objects is exhausted,
 * because then there will be a long period of OSP being unavailable for the
 * new creations due to lenghty precreate RPC. Instead we ask for another
 * precreation ahead and hopefully have it ready before the current pool is
 * empty. Notice this function relies on an external locking.
 *
 * \param[in] env       LU environment provided by the caller
 * \param[in] d         OSP device
 *
 * \retval              0 - current pool is good enough, 1 - time to precreate
 */
static inline int osp_precreate_near_empty_nolock(const struct lu_env *env,
                                                  struct osp_device *d)
{
        int window = osp_objs_precreated(env, d);

        /* don't consider new precreation till OST is healty and
         * has free space */
        return ((window - d->opd_pre_reserved < d->opd_pre_create_count / 2) &&
                (d->opd_pre_status == 0));
}

/**
 * Check pool of precreated objects
 *
 * This is protected version of osp_precreate_near_empty_nolock(), check that
 * for the details.
 *
 * \param[in] env       LU environment provided by the caller
 * \param[in] d         OSP device
 *
 * \retval              0 - current pool is good enough, 1 - time to precreate
 */
static inline int osp_precreate_near_empty(const struct lu_env *env,
                                           struct osp_device *d)
{
        int rc;

        if (d->opd_pre == NULL)
                return 0;

        /* XXX: do we really need locking here? */
        spin_lock(&d->opd_pre_lock);
        rc = osp_precreate_near_empty_nolock(env, d);
        spin_unlock(&d->opd_pre_lock);
        return rc;
}

/**
 * Check given sequence is empty
 *
 * Returns a binary result whether the given sequence has some IDs left
 * or not. Find the details in osp_fid_end_seq(). This is a lock protected
 * version of that function.
 *
 * \param[in] env       LU environment provided by the caller
 * \param[in] osp       OSP device
 *
 * \retval              0 - current sequence has no IDs, 1 - otherwise
 */
static inline int osp_create_end_seq(const struct lu_env *env,
                                     struct osp_device *osp)
{
        struct lu_fid *fid = &osp->opd_pre_used_fid;
        int rc;

        spin_lock(&osp->opd_pre_lock);
        rc = osp_fid_end_seq(env, fid);
        spin_unlock(&osp->opd_pre_lock);
        return rc;
}

/**
 * Write FID into into last_oid/last_seq file
 *
 * The function stores the sequence and the in-sequence id into two dedicated
 * files. The sync argument can be used to request synchronous commit, so the
 * function won't return until the updates are committed.
 *
 * \param[in] env       LU environment provided by the caller
 * \param[in] osp       OSP device
 * \param[in] fid       fid where sequence/id is taken
 * \param[in] sync      update mode: 0 - asynchronously, 1 - synchronously
 *
 * \retval 0            on success
 * \retval negative     negated errno on error
 **/
int osp_write_last_oid_seq_files(struct lu_env *env, struct osp_device *osp,
                                 struct lu_fid *fid, int sync)
{
        struct osp_thread_info  *oti = osp_env_info(env);
        struct lu_buf      *lb_oid = &oti->osi_lb;
        struct lu_buf      *lb_oseq = &oti->osi_lb2;
        loff_t             oid_off;
        u64                oid;
        loff_t             oseq_off;
        struct thandle    *th;
        int                   rc;
        ENTRY;

        if (osp->opd_storage->dd_rdonly)
                RETURN(0);

        /* Note: through f_oid is only 32 bits, it will also write 64 bits
         * for oid to keep compatibility with the previous version. */
        oid = fid->f_oid;
        osp_objid_buf_prep(lb_oid, &oid_off,
                           &oid, osp->opd_index);

        osp_objseq_buf_prep(lb_oseq, &oseq_off,
                            &fid->f_seq, osp->opd_index);

        th = dt_trans_create(env, osp->opd_storage);
        if (IS_ERR(th))
                RETURN(PTR_ERR(th));

        th->th_sync |= sync;
        rc = dt_declare_record_write(env, osp->opd_last_used_oid_file,
                                     lb_oid, oid_off, th);
        if (rc != 0)
                GOTO(out, rc);

        rc = dt_declare_record_write(env, osp->opd_last_used_seq_file,
                                     lb_oseq, oseq_off, th);
        if (rc != 0)
                GOTO(out, rc);

        rc = dt_trans_start_local(env, osp->opd_storage, th);
        if (rc != 0)
                GOTO(out, rc);

        rc = dt_record_write(env, osp->opd_last_used_oid_file, lb_oid,
                             &oid_off, th);
        if (rc != 0) {
                CERROR("%s: can not write to last seq file: rc = %d\n",
                        osp->opd_obd->obd_name, rc);
                GOTO(out, rc);
        }
        rc = dt_record_write(env, osp->opd_last_used_seq_file, lb_oseq,
                             &oseq_off, th);
        if (rc) {
                CERROR("%s: can not write to last seq file: rc = %d\n",
                        osp->opd_obd->obd_name, rc);
                GOTO(out, rc);
        }
out:
        dt_trans_stop(env, osp->opd_storage, th);
        RETURN(rc);
}

/**
 * Switch to another sequence
 *
 * When a current sequence has no available IDs left, OSP has to switch to
 * another new sequence. OSP requests it using the regular FLDB protocol
 * and stores synchronously before that is used in precreated. This is needed
 * to basically have the sequences referenced (not orphaned), otherwise it's
 * possible that OST has some objects precreated and the clients have data
 * written to it, but after MDT failover nobody refers those objects and OSP
 * has no idea that the sequence need cleanup to be done.
 * While this is very expensive operation, it's supposed to happen very very
 * infrequently because sequence has 2^32 or 2^48 objects (depending on type)
 *
 * \param[in] env       LU environment provided by the caller
 * \param[in] osp       OSP device
 *
 * \retval 0            on success
 * \retval negative     negated errno on error
 */
static int osp_precreate_rollover_new_seq(struct lu_env *env,
                                          struct osp_device *osp)
{
        struct lu_fid   *fid = &osp_env_info(env)->osi_fid;
        struct lu_fid   *last_fid = &osp->opd_last_used_fid;
        int             rc;
        ENTRY;

        rc = seq_client_get_seq(env, osp->opd_obd->u.cli.cl_seq, &fid->f_seq);
        if (rc != 0) {
                CERROR("%s: alloc fid error: rc = %d\n",
                       osp->opd_obd->obd_name, rc);
                RETURN(rc);
        }

        fid->f_oid = 1;
        fid->f_ver = 0;
        LASSERTF(fid_seq(fid) != fid_seq(last_fid),
                 "fid "DFID", last_fid "DFID"\n", PFID(fid),
                 PFID(last_fid));

        rc = osp_write_last_oid_seq_files(env, osp, fid, 1);
        if (rc != 0) {
                CERROR("%s: Can not update oid/seq file: rc = %d\n",
                       osp->opd_obd->obd_name, rc);
                RETURN(rc);
        }

        LCONSOLE_INFO("%s: update sequence from %#llx to %#llx\n",
                      osp->opd_obd->obd_name, fid_seq(last_fid),
                      fid_seq(fid));
        /* Update last_xxx to the new seq */
        spin_lock(&osp->opd_pre_lock);
        osp->opd_last_used_fid = *fid;
        osp_fid_to_obdid(fid, &osp->opd_last_id);
        osp->opd_gap_start_fid = *fid;
        osp->opd_pre_used_fid = *fid;
        osp->opd_pre_last_created_fid = *fid;
        spin_unlock(&osp->opd_pre_lock);

        RETURN(rc);
}

/**
 * Find IDs available in current sequence
 *
 * The function calculates the highest possible ID and the number of IDs
 * available in the current sequence OSP is using. The number is limited
 * artifically by the caller (grow param) and the number of IDs available
 * in the sequence by nature. The function doesn't require an external
 * locking.
 *
 * \param[in] env       LU environment provided by the caller
 * \param[in] osp       OSP device
 * \param[in] fid       FID the caller wants to start with
 * \param[in] grow      how many the caller wants
 * \param[out] fid      the highest calculated FID
 * \param[out] grow     the number of available IDs calculated
 *
 * \retval              0 on success, 1 - the sequence is empty
 */
static int osp_precreate_fids(const struct lu_env *env, struct osp_device *osp,
                              struct lu_fid *fid, int *grow)
{
        struct osp_thread_info  *osi = osp_env_info(env);
        __u64                   end;
        int                     i = 0;

        if (fid_is_idif(fid)) {
                struct lu_fid   *last_fid;
                struct ost_id   *oi = &osi->osi_oi;
                int rc;

                spin_lock(&osp->opd_pre_lock);
                last_fid = &osp->opd_pre_last_created_fid;
                fid_to_ostid(last_fid, oi);
                end = min(ostid_id(oi) + *grow, IDIF_MAX_OID);
                *grow = end - ostid_id(oi);
                rc = ostid_set_id(oi, ostid_id(oi) + *grow);
                spin_unlock(&osp->opd_pre_lock);

                if (*grow == 0 || rc)
                        return 1;

                ostid_to_fid(fid, oi, osp->opd_index);
                return 0;
        }

        spin_lock(&osp->opd_pre_lock);
        *fid = osp->opd_pre_last_created_fid;
        end = fid->f_oid;
        end = min((end + *grow), (__u64)LUSTRE_DATA_SEQ_MAX_WIDTH);
        *grow = end - fid->f_oid;
        fid->f_oid += end - fid->f_oid;
        spin_unlock(&osp->opd_pre_lock);

        CDEBUG(D_INFO, "Expect %d, actual %d ["DFID" -- "DFID"]\n",
               *grow, i, PFID(fid), PFID(&osp->opd_pre_last_created_fid));

        return *grow > 0 ? 0 : 1;
}

/**
 * Prepare and send precreate RPC
 *
 * The function finds how many objects should be precreated.  Then allocates,
 * prepares and schedules precreate RPC synchronously. Upon reply the function
 * wakes up the threads waiting for the new objects on this target. If the
 * target wasn't able to create all the objects requested, then the next
 * precreate will be asking for fewer objects (i.e. slow precreate down).
 *
 * \param[in] env       LU environment provided by the caller
 * \param[in] d         OSP device
 *
 * \retval 0            on success
 * \retval negative     negated errno on error
 **/
static int osp_precreate_send(const struct lu_env *env, struct osp_device *d)
{
        struct osp_thread_info  *oti = osp_env_info(env);
        struct ptlrpc_request   *req;
        struct obd_import       *imp;
        struct ost_body         *body;
        int                      rc, grow, diff;
        struct lu_fid           *fid = &oti->osi_fid;
        ENTRY;

        /* don't precreate new objects till OST healthy and has free space */
        if (unlikely(d->opd_pre_status)) {
                CDEBUG(D_INFO, "%s: don't send new precreate: rc = %d\n",
                       d->opd_obd->obd_name, d->opd_pre_status);
                RETURN(0);
        }

        /*
         * if not connection/initialization is compeleted, ignore
         */
        imp = d->opd_obd->u.cli.cl_import;
        LASSERT(imp);

        req = ptlrpc_request_alloc(imp, &RQF_OST_CREATE);
        if (req == NULL)
                RETURN(-ENOMEM);
        req->rq_request_portal = OST_CREATE_PORTAL;
        /* we should not resend create request - anyway we will have delorphan
         * and kill these objects */
        req->rq_no_delay = req->rq_no_resend = 1;

        rc = ptlrpc_request_pack(req, LUSTRE_OST_VERSION, OST_CREATE);
        if (rc) {
                ptlrpc_request_free(req);
                RETURN(rc);
        }

        spin_lock(&d->opd_pre_lock);
        if (d->opd_pre_create_count > d->opd_pre_max_create_count / 2)
                d->opd_pre_create_count = d->opd_pre_max_create_count / 2;
        grow = d->opd_pre_create_count;
        spin_unlock(&d->opd_pre_lock);

        body = req_capsule_client_get(&req->rq_pill, &RMF_OST_BODY);
        LASSERT(body);

        *fid = d->opd_pre_last_created_fid;
        rc = osp_precreate_fids(env, d, fid, &grow);
        if (rc == 1)
                /* Current seq has been used up*/
                GOTO(out_req, rc = -ENOSPC);

        if (!osp_is_fid_client(d)) {
                /* Non-FID client will always send seq 0 because of
                 * compatiblity */
                LASSERTF(fid_is_idif(fid), "Invalid fid "DFID"\n", PFID(fid));
                fid->f_seq = 0;
        }

        fid_to_ostid(fid, &body->oa.o_oi);
        body->oa.o_valid = OBD_MD_FLGROUP;

        ptlrpc_request_set_replen(req);

        if (OBD_FAIL_CHECK(OBD_FAIL_OSP_FAKE_PRECREATE))
                GOTO(ready, rc = 0);

        rc = ptlrpc_queue_wait(req);
        if (rc) {
                CERROR("%s: can't precreate: rc = %d\n", d->opd_obd->obd_name,
                       rc);
                GOTO(out_req, rc);
        }
        LASSERT(req->rq_transno == 0);

        body = req_capsule_server_get(&req->rq_pill, &RMF_OST_BODY);
        if (body == NULL)
                GOTO(out_req, rc = -EPROTO);

        ostid_to_fid(fid, &body->oa.o_oi, d->opd_index);

ready:
        if (osp_fid_diff(fid, &d->opd_pre_used_fid) <= 0) {
                CERROR("%s: precreate fid "DFID" <= local used fid "DFID
                       ": rc = %d\n", d->opd_obd->obd_name,
                       PFID(fid), PFID(&d->opd_pre_used_fid), -ESTALE);
                GOTO(out_req, rc = -ESTALE);
        }

        diff = osp_fid_diff(fid, &d->opd_pre_last_created_fid);

        spin_lock(&d->opd_pre_lock);
        if (diff < grow) {
                /* the OST has not managed to create all the
                 * objects we asked for */
                d->opd_pre_create_count = max(diff, OST_MIN_PRECREATE);
                d->opd_pre_create_slow = 1;
        } else {
                /* the OST is able to keep up with the work,
                 * we could consider increasing create_count
                 * next time if needed */
                d->opd_pre_create_slow = 0;
        }

        body = req_capsule_client_get(&req->rq_pill, &RMF_OST_BODY);
        fid_to_ostid(fid, &body->oa.o_oi);

        d->opd_pre_last_created_fid = *fid;
        spin_unlock(&d->opd_pre_lock);

        CDEBUG(D_HA, "%s: current precreated pool: "DFID"-"DFID"\n",
               d->opd_obd->obd_name, PFID(&d->opd_pre_used_fid),
               PFID(&d->opd_pre_last_created_fid));
out_req:
        /* now we can wakeup all users awaiting for objects */
        osp_pre_update_status(d, rc);
        wake_up(&d->opd_pre_user_waitq);

        ptlrpc_req_finished(req);
        RETURN(rc);
}

/**
 * Get last precreated object from target (OST)
 *
 * Sends synchronous RPC to the target (OST) to learn the last precreated
 * object. This later is used to remove all unused objects (cleanup orphan
 * procedure). Also, the next object after one we got will be used as a
 * starting point for the new precreates.
 *
 * \param[in] env       LU environment provided by the caller
 * \param[in] d         OSP device
 *
 * \retval 0            on success
 * \retval negative     negated errno on error
 **/
static int osp_get_lastfid_from_ost(const struct lu_env *env,
                                    struct osp_device *d)
{
        struct ptlrpc_request   *req = NULL;
        struct obd_import       *imp;
        struct lu_fid           *last_fid;
        char                    *tmp;
        int                     rc;
        ENTRY;

        imp = d->opd_obd->u.cli.cl_import;
        LASSERT(imp);

        req = ptlrpc_request_alloc(imp, &RQF_OST_GET_INFO_LAST_FID);
        if (req == NULL)
                RETURN(-ENOMEM);

        req_capsule_set_size(&req->rq_pill, &RMF_GETINFO_KEY, RCL_CLIENT,
                             sizeof(KEY_LAST_FID));

        rc = ptlrpc_request_pack(req, LUSTRE_OST_VERSION, OST_GET_INFO);
        if (rc) {
                ptlrpc_request_free(req);
                RETURN(rc);
        }

        tmp = req_capsule_client_get(&req->rq_pill, &RMF_GETINFO_KEY);
        memcpy(tmp, KEY_LAST_FID, sizeof(KEY_LAST_FID));

        req->rq_no_delay = req->rq_no_resend = 1;
        last_fid = req_capsule_client_get(&req->rq_pill, &RMF_FID);
        fid_cpu_to_le(last_fid, &d->opd_last_used_fid);

        ptlrpc_request_set_replen(req);

        rc = ptlrpc_queue_wait(req);
        if (rc) {
                /* bad-bad OST.. let sysadm sort this out */
                if (rc == -ENOTSUPP) {
                        CERROR("%s: server does not support FID: rc = %d\n",
                               d->opd_obd->obd_name, -ENOTSUPP);
                }
                ptlrpc_set_import_active(imp, 0);
                GOTO(out, rc);
        }

        last_fid = req_capsule_server_get(&req->rq_pill, &RMF_FID);
        if (last_fid == NULL) {
                CERROR("%s: Got last_fid failed.\n", d->opd_obd->obd_name);
                GOTO(out, rc = -EPROTO);
        }

        if (!fid_is_sane(last_fid)) {
                CERROR("%s: Got insane last_fid "DFID"\n",
                       d->opd_obd->obd_name, PFID(last_fid));
                GOTO(out, rc = -EPROTO);
        }

        /* Only update the last used fid, if the OST has objects for
         * this sequence, i.e. fid_oid > 0 */
        if (fid_oid(last_fid) > 0)
                d->opd_last_used_fid = *last_fid;

        CDEBUG(D_HA, "%s: Got last_fid "DFID"\n", d->opd_obd->obd_name,
               PFID(last_fid));

out:
        ptlrpc_req_finished(req);
        RETURN(rc);
}

/**
 * Cleanup orphans on OST
 *
 * This function is called in a contex of a dedicated thread handling
 * all the precreation suff. The function waits till local recovery
 * is complete, then identify all the unreferenced objects (orphans)
 * using the highest ID referenced by a local and the highest object
 * precreated by the target. The found range is a subject to removal
 * using specially flagged RPC. During this process OSP is marked
 * unavailable for new objects.
 *
 * \param[in] env       LU environment provided by the caller
 * \param[in] d         OSP device
 *
 * \retval 0            on success
 * \retval negative     negated errno on error
 */
static int osp_precreate_cleanup_orphans(struct lu_env *env,
                                         struct osp_device *d)
{
        struct osp_thread_info  *osi = osp_env_info(env);
        struct lu_fid           *last_fid = &osi->osi_fid;
        struct ptlrpc_request   *req = NULL;
        struct obd_import       *imp;
        struct ost_body         *body;
        int                      update_status = 0;
        int                      rc;
        int                      diff;

        ENTRY;

        /*
         * wait for local recovery to finish, so we can cleanup orphans
         * orphans are all objects since "last used" (assigned), but
         * there might be objects reserved and in some cases they won't
         * be used. we can't cleanup them till we're sure they won't be
         * used. also can't we allow new reservations because they may
         * end up getting orphans being cleaned up below. so we block
         * new reservations and wait till all reserved objects either
         * user or released.
         */
        spin_lock(&d->opd_pre_lock);
        d->opd_pre_recovering = 1;
        spin_unlock(&d->opd_pre_lock);
        /*
         * The locking above makes sure the opd_pre_reserved check below will
         * catch all osp_precreate_reserve() calls who find
         * "!opd_pre_recovering".
         */
        wait_event_idle(d->opd_pre_waitq,
                        (!d->opd_pre_reserved && d->opd_recovery_completed) ||
                        !osp_precreate_running(d) || d->opd_got_disconnected);
        if (!osp_precreate_running(d) || d->opd_got_disconnected)
                GOTO(out, rc = -EAGAIN);

        CDEBUG(D_HA, "%s: going to cleanup orphans since "DFID"\n",
               d->opd_obd->obd_name, PFID(&d->opd_last_used_fid));

        *last_fid = d->opd_last_used_fid;
        /* The OSP should already get the valid seq now */
        LASSERT(!fid_is_zero(last_fid));
        if (fid_oid(&d->opd_last_used_fid) < 2) {
                /* lastfid looks strange... ask OST */
                rc = osp_get_lastfid_from_ost(env, d);
                if (rc)
                        GOTO(out, rc);
        }

        imp = d->opd_obd->u.cli.cl_import;
        LASSERT(imp);

        req = ptlrpc_request_alloc(imp, &RQF_OST_CREATE);
        if (req == NULL)
                GOTO(out, rc = -ENOMEM);

        rc = ptlrpc_request_pack(req, LUSTRE_OST_VERSION, OST_CREATE);
        if (rc) {
                ptlrpc_request_free(req);
                req = NULL;
                GOTO(out, rc);
        }

        body = req_capsule_client_get(&req->rq_pill, &RMF_OST_BODY);
        if (body == NULL)
                GOTO(out, rc = -EPROTO);

        body->oa.o_flags = OBD_FL_DELORPHAN;
        body->oa.o_valid = OBD_MD_FLFLAGS | OBD_MD_FLGROUP;

        fid_to_ostid(&d->opd_last_used_fid, &body->oa.o_oi);

        ptlrpc_request_set_replen(req);

        /* Don't resend the delorphan req */
        req->rq_no_resend = req->rq_no_delay = 1;

        rc = ptlrpc_queue_wait(req);
        if (rc) {
                update_status = 1;
                GOTO(out, rc);
        }

        body = req_capsule_server_get(&req->rq_pill, &RMF_OST_BODY);
        if (body == NULL)
                GOTO(out, rc = -EPROTO);

        /*
         * OST provides us with id new pool starts from in body->oa.o_id
         */
        ostid_to_fid(last_fid, &body->oa.o_oi, d->opd_index);

        spin_lock(&d->opd_pre_lock);
        diff = osp_fid_diff(&d->opd_last_used_fid, last_fid);
        if (diff > 0) {
                d->opd_pre_create_count = OST_MIN_PRECREATE + diff;
                d->opd_pre_last_created_fid = d->opd_last_used_fid;
        } else {
                d->opd_pre_create_count = OST_MIN_PRECREATE;
                d->opd_pre_last_created_fid = *last_fid;
        }
        /*
         * This empties the pre-creation pool and effectively blocks any new
         * reservations.
         */
        LASSERT(fid_oid(&d->opd_pre_last_created_fid) <=
                LUSTRE_DATA_SEQ_MAX_WIDTH);
        d->opd_pre_used_fid = d->opd_pre_last_created_fid;
        d->opd_pre_create_slow = 0;
        spin_unlock(&d->opd_pre_lock);

        CDEBUG(D_HA, "%s: Got last_id "DFID" from OST, last_created "DFID
               "last_used is "DFID"\n", d->opd_obd->obd_name, PFID(last_fid),
               PFID(&d->opd_pre_last_created_fid), PFID(&d->opd_last_used_fid));
out:
        if (req)
                ptlrpc_req_finished(req);

        /*
         * If rc is zero, the pre-creation window should have been emptied.
         * Since waking up the herd would be useless without pre-created
         * objects, we defer the signal to osp_precreate_send() in that case.
         */
        if (rc != 0) {
                if (update_status) {
                        CERROR("%s: cannot cleanup orphans: rc = %d\n",
                               d->opd_obd->obd_name, rc);
                        /* we can't proceed from here, OST seem to
                         * be in a bad shape, better to wait for
                         * a new instance of the server and repeat
                         * from the beginning. notify possible waiters
                         * this OSP isn't quite functional yet */
                        osp_pre_update_status(d, rc);
                } else {
                        wake_up(&d->opd_pre_user_waitq);
                }
        } else {
                spin_lock(&d->opd_pre_lock);
                d->opd_pre_recovering = 0;
                spin_unlock(&d->opd_pre_lock);
        }

        RETURN(rc);
}

/**
 * Update precreate status using statfs data
 *
 * The function decides whether this OSP should be used for new objects.
 * IOW, whether this OST is used up or has some free space. Cached statfs
 * data is used to make this decision. If the latest result of statfs
 * request (rc argument) is not success, then just mark OSP unavailable
 * right away.
 *
 * The new statfs data is passed in \a msfs and needs to be stored into
 * opd_statfs, but only after the various flags in os_state are set, so
 * that the new statfs data is not visible without appropriate flags set.
 * As such, there is no need to clear the flags here, since this is called
 * with new statfs data, and they should not be cleared if sent from OST.
 *
 * Add a bit of hysteresis so this flag isn't continually flapping, and
 * ensure that new files don't get extremely fragmented due to only a
 * small amount of available space in the filesystem.  We want to set
 * the ENOSPC/ENOINO flags unconditionally when there is less than the
 * reserved size free, and still copy them from the old state when there
 * is less than 2*reserved size free space or inodes.
 *
 * \param[in] d         OSP device
 * \param[in] msfs      statfs data
 */
static void osp_pre_update_msfs(struct osp_device *d, struct obd_statfs *msfs)
{
        u32 old_state = d->opd_statfs.os_state;
        u32 reserved_ino_low = 32;      /* could be tunable in the future */
        u32 reserved_ino_high = reserved_ino_low * 2;
        u64 available_mb;

        /* statfs structure not initialized yet */
        if (unlikely(!msfs->os_type))
                return;

        /* if the low and high watermarks have not been initialized yet */
        if (unlikely(d->opd_reserved_mb_high == 0 &&
                     d->opd_reserved_mb_low == 0)) {
                /* Use ~0.1% by default to disable object allocation,
                 * and ~0.2% to enable, size in MB, set both watermark
                 */
                spin_lock(&d->opd_pre_lock);
                if (d->opd_reserved_mb_high == 0 &&
                    d->opd_reserved_mb_low == 0) {
                        d->opd_reserved_mb_low = ((msfs->os_bsize >> 10) *
                                                  msfs->os_blocks) >> 20;
                        if (d->opd_reserved_mb_low == 0)
                                d->opd_reserved_mb_low = 1;
                        d->opd_reserved_mb_high =
                                (d->opd_reserved_mb_low << 1) + 1;
                }
                spin_unlock(&d->opd_pre_lock);
        }

        available_mb = (msfs->os_bavail * (msfs->os_bsize >> 10)) >> 10;
        if (msfs->os_ffree < reserved_ino_low)
                msfs->os_state |= OS_STATE_ENOINO;
        else if (msfs->os_ffree <= reserved_ino_high)
                msfs->os_state |= old_state & OS_STATE_ENOINO;
        /* else don't clear flags in new msfs->os_state sent from OST */

        CDEBUG(D_INFO,
               "%s: blocks=%llu free=%llu avail=%llu avail_mb=%llu hwm_mb=%u files=%llu ffree=%llu state=%x: rc = %d\n",
               d->opd_obd->obd_name, msfs->os_blocks, msfs->os_bfree,
               msfs->os_bavail, available_mb, d->opd_reserved_mb_high,
               msfs->os_files, msfs->os_ffree, msfs->os_state,
               d->opd_pre_status);
        if (available_mb < d->opd_reserved_mb_low)
                msfs->os_state |= OS_STATE_ENOSPC;
        else if (available_mb <= d->opd_reserved_mb_high)
                msfs->os_state |= old_state & OS_STATE_ENOSPC;
        /* else don't clear flags in new msfs->os_state sent from OST */

        if (msfs->os_state & (OS_STATE_ENOINO | OS_STATE_ENOSPC)) {
                d->opd_pre_status = -ENOSPC;
                if (!(old_state & (OS_STATE_ENOINO | OS_STATE_ENOSPC)))
                        CDEBUG(D_INFO, "%s: full: state=%x: rc = %x\n",
                               d->opd_obd->obd_name, msfs->os_state,
                               d->opd_pre_status);
                CDEBUG(D_INFO, "uncommitted changes=%u in_progress=%u\n",
                       atomic_read(&d->opd_sync_changes),
                       atomic_read(&d->opd_sync_rpcs_in_progress));
        } else if (old_state & (OS_STATE_ENOINO | OS_STATE_ENOSPC)) {
                d->opd_pre_status = 0;
                spin_lock(&d->opd_pre_lock);
                d->opd_pre_create_slow = 0;
                d->opd_pre_create_count = OST_MIN_PRECREATE;
                spin_unlock(&d->opd_pre_lock);
                wake_up(&d->opd_pre_waitq);

                CDEBUG(D_INFO,
                       "%s: available: state=%x: rc = %d\n",
                       d->opd_obd->obd_name, msfs->os_state,
                       d->opd_pre_status);
        } else {
                /* we only get here if rc == 0 in the caller */
                d->opd_pre_status = 0;
        }

        /* Object precreation skipped on OST if manually disabled */
        if (d->opd_pre_max_create_count == 0)
                msfs->os_state |= OS_STATE_NOPRECREATE;
        /* else don't clear flags in new msfs->os_state sent from OST */

        /* copy only new statfs state to make it visible to MDS threads */
        if (&d->opd_statfs != msfs)
                d->opd_statfs = *msfs;
}

/**
 * Initialize FID for precreation
 *
 * For a just created new target, a new sequence should be taken.
 * The function checks there is no IDIF in use (if the target was
 * added with the older version of Lustre), then requests a new
 * sequence from FLDB using the regular protocol. Then this new
 * sequence is stored on a persisten storage synchronously to prevent
 * possible object leakage (for the detail see the description for
 * osp_precreate_rollover_new_seq()).
 *
 * \param[in] osp       OSP device
 *
 * \retval 0            on success
 * \retval negative     negated errno on error
 */
int osp_init_pre_fid(struct osp_device *osp)
{
        struct lu_env           env;
        struct osp_thread_info  *osi;
        struct lu_client_seq    *cli_seq;
        struct lu_fid           *last_fid;
        int                     rc;
        ENTRY;

        LASSERT(osp->opd_pre != NULL);

        /* Let's check if the current last_seq/fid is valid,
         * otherwise request new sequence from the controller */
        if (osp_is_fid_client(osp) && osp->opd_group != 0) {
                /* Non-MDT0 can only use normal sequence for
                 * OST objects */
                if (fid_is_norm(&osp->opd_last_used_fid))
                        RETURN(0);
        } else {
                /* Initially MDT0 will start with IDIF, after
                 * that it will request new sequence from the
                 * controller */
                if (fid_is_idif(&osp->opd_last_used_fid) ||
                    fid_is_norm(&osp->opd_last_used_fid))
                        RETURN(0);
        }

        if (!fid_is_zero(&osp->opd_last_used_fid))
                CWARN("%s: invalid last used fid "DFID
                      ", try to get new sequence.\n",
                      osp->opd_obd->obd_name,
                      PFID(&osp->opd_last_used_fid));

        rc = lu_env_init(&env, osp->opd_dt_dev.dd_lu_dev.ld_type->ldt_ctx_tags);
        if (rc) {
                CERROR("%s: init env error: rc = %d\n",
                       osp->opd_obd->obd_name, rc);
                RETURN(rc);
        }

        osi = osp_env_info(&env);
        last_fid = &osi->osi_fid;
        fid_zero(last_fid);
        /* For a freshed fs, it will allocate a new sequence first */
        if (osp_is_fid_client(osp) && osp->opd_group != 0) {
                cli_seq = osp->opd_obd->u.cli.cl_seq;
                rc = seq_client_get_seq(&env, cli_seq, &last_fid->f_seq);
                if (rc != 0) {
                        CERROR("%s: alloc fid error: rc = %d\n",
                               osp->opd_obd->obd_name, rc);
                        GOTO(out, rc);
                }
        } else {
                last_fid->f_seq = fid_idif_seq(0, osp->opd_index);
        }
        last_fid->f_oid = 1;
        last_fid->f_ver = 0;

        spin_lock(&osp->opd_pre_lock);
        osp->opd_last_used_fid = *last_fid;
        osp->opd_pre_used_fid = *last_fid;
        osp->opd_pre_last_created_fid = *last_fid;
        spin_unlock(&osp->opd_pre_lock);
        rc = osp_write_last_oid_seq_files(&env, osp, last_fid, 1);
        if (rc != 0) {
                CERROR("%s: write fid error: rc = %d\n",
                       osp->opd_obd->obd_name, rc);
                GOTO(out, rc);
        }
out:
        lu_env_fini(&env);
        RETURN(rc);
}

/**
 * The core of precreate functionality
 *
 * The function implements the main precreation loop. Basically it
 * involves connecting to the target, precerate FID initialization,
 * identifying and removing orphans, then serving precreation. As
 * part of the latter, the thread is responsible for statfs data
 * updates. The precreation is mostly driven by another threads
 * asking for new OST objects - those askers wake the thread when
 * the number of precreated objects reach low watermark.
 * After a disconnect, the sequence above repeats. This is keep going
 * until the thread is requested to stop.
 *
 * \param[in] _arg      private data the thread (OSP device to handle)
 *
 * \retval 0            on success
 * \retval negative     negated errno on error
 */
static int osp_precreate_thread(void *_arg)
{
        struct osp_device       *d = _arg;
        struct ptlrpc_thread    *thread = &d->opd_pre_thread;
        struct l_wait_info       lwi2 = LWI_TIMEOUT(cfs_time_seconds(5),
                                                    back_to_sleep, NULL);
        struct lu_env            env;
        int                      rc;

        ENTRY;

        rc = lu_env_init(&env, d->opd_dt_dev.dd_lu_dev.ld_type->ldt_ctx_tags);
        if (rc) {
                CERROR("%s: init env error: rc = %d\n", d->opd_obd->obd_name,
                       rc);

                spin_lock(&d->opd_pre_lock);
                thread->t_flags = SVC_STOPPED;
                spin_unlock(&d->opd_pre_lock);
                wake_up(&thread->t_ctl_waitq);

                RETURN(rc);
        }

        spin_lock(&d->opd_pre_lock);
        thread->t_flags = SVC_RUNNING;
        spin_unlock(&d->opd_pre_lock);
        wake_up(&thread->t_ctl_waitq);

        while (osp_precreate_running(d)) {
                /*
                 * need to be connected to OST
                 */
                while (osp_precreate_running(d)) {
                        if ((d->opd_pre == NULL || d->opd_pre_recovering) &&
                            d->opd_imp_connected &&
                            !d->opd_got_disconnected)
                                break;
                        wait_event_idle(d->opd_pre_waitq,
                                        !osp_precreate_running(d) ||
                                        d->opd_new_connection);

                        if (!d->opd_new_connection)
                                continue;

                        d->opd_new_connection = 0;
                        d->opd_got_disconnected = 0;
                        break;
                }

                if (!osp_precreate_running(d))
                        break;

                if (d->opd_pre) {
                        LASSERT(d->opd_obd->u.cli.cl_seq != NULL);
                        /* Sigh, fid client is not ready yet */
                        if (d->opd_obd->u.cli.cl_seq->lcs_exp == NULL)
                                continue;

                        /* Init fid for osp_precreate if necessary */
                        rc = osp_init_pre_fid(d);
                        if (rc != 0) {
                                class_export_put(d->opd_exp);
                                d->opd_obd->u.cli.cl_seq->lcs_exp = NULL;
                                CERROR("%s: init pre fid error: rc = %d\n",
                                                d->opd_obd->obd_name, rc);
                                continue;
                        }
                }

                if (osp_statfs_update(&env, d)) {
                        l_wait_event(d->opd_pre_waitq,
                                     !osp_precreate_running(d), &lwi2);
                        continue;
                }

                if (d->opd_pre) {
                        /*
                         * Clean up orphans or recreate missing objects.
                         */
                        rc = osp_precreate_cleanup_orphans(&env, d);
                        if (rc != 0) {
                                schedule_timeout_interruptible(cfs_time_seconds(1));
                                continue;
                        }
                }

                /*
                 * connected, can handle precreates now
                 */
                while (osp_precreate_running(d)) {
                        wait_event_idle(d->opd_pre_waitq,
                                        !osp_precreate_running(d) ||
                                        osp_precreate_near_empty(&env, d) ||
                                        osp_statfs_need_update(d) ||
                                        d->opd_got_disconnected);

                        if (!osp_precreate_running(d))
                                break;

                        /* something happened to the connection
                         * have to start from the beginning */
                        if (d->opd_got_disconnected)
                                break;

                        if (osp_statfs_need_update(d))
                                if (osp_statfs_update(&env, d))
                                        break;

                        if (d->opd_pre == NULL)
                                continue;

                        /* To avoid handling different seq in precreate/orphan
                         * cleanup, it will hold precreate until current seq is
                         * used up. */
                        if (unlikely(osp_precreate_end_seq(&env, d) &&
                            !osp_create_end_seq(&env, d)))
                                continue;

                        if (unlikely(osp_precreate_end_seq(&env, d) &&
                                     osp_create_end_seq(&env, d))) {
                                LCONSOLE_INFO("%s:%#llx is used up."
                                              " Update to new seq\n",
                                              d->opd_obd->obd_name,
                                         fid_seq(&d->opd_pre_last_created_fid));
                                rc = osp_precreate_rollover_new_seq(&env, d);
                                if (rc)
                                        continue;
                        }

                        if (osp_precreate_near_empty(&env, d)) {
                                rc = osp_precreate_send(&env, d);
                                /* osp_precreate_send() sets opd_pre_status
                                 * in case of error, that prevent the using of
                                 * failed device. */
                                if (rc < 0 && rc != -ENOSPC &&
                                    rc != -ETIMEDOUT && rc != -ENOTCONN)
                                        CERROR("%s: cannot precreate objects:"
                                               " rc = %d\n",
                                               d->opd_obd->obd_name, rc);
                        }
                }
        }

        thread->t_flags = SVC_STOPPED;
        lu_env_fini(&env);
        wake_up(&thread->t_ctl_waitq);

        RETURN(0);
}

/**
 * Check when to stop to wait for precreate objects.
 *
 * The caller wanting a new OST object can't wait undefinitely. The
 * function checks for few conditions including available new OST
 * objects, disconnected OST, lack of space with no pending destroys,
 * etc. IOW, it checks whether the current OSP state is good to keep
 * waiting or it's better to give up.
 *
 * \param[in] env       LU environment provided by the caller
 * \param[in] d         OSP device
 *
 * \retval              0 - keep waiting, 1 - no luck
 */
static int osp_precreate_ready_condition(const struct lu_env *env,
                                         struct osp_device *d)
{
        if (d->opd_pre_recovering)
                return 0;

        /* ready if got enough precreated objects */
        /* we need to wait for others (opd_pre_reserved) and our object (+1) */
        if (d->opd_pre_reserved + 1 < osp_objs_precreated(env, d))
                return 1;

        /* ready if OST reported no space and no destroys in progress */
        if (atomic_read(&d->opd_sync_changes) +
            atomic_read(&d->opd_sync_rpcs_in_progress) == 0 &&
            d->opd_pre_status == -ENOSPC)
                return 1;

        /* Bail out I/O fails to OST */
        if (d->opd_pre_status != 0 &&
            d->opd_pre_status != -EAGAIN &&
            d->opd_pre_status != -ENODEV &&
            d->opd_pre_status != -ENOTCONN &&
            d->opd_pre_status != -ENOSPC) {
                /* DEBUG LU-3230 */
                if (d->opd_pre_status != -EIO)
                        CERROR("%s: precreate failed opd_pre_status %d\n",
                               d->opd_obd->obd_name, d->opd_pre_status);
                return 1;
        }

        return 0;
}

static int osp_precreate_timeout_condition(void *data)
{
        struct osp_device *d = data;

        CDEBUG(D_HA, "%s: slow creates, last="DFID", next="DFID", "
              "reserved=%llu, sync_changes=%u, "
              "sync_rpcs_in_progress=%d, status=%d\n",
              d->opd_obd->obd_name, PFID(&d->opd_pre_last_created_fid),
              PFID(&d->opd_pre_used_fid), d->opd_pre_reserved,
              atomic_read(&d->opd_sync_changes),
              atomic_read(&d->opd_sync_rpcs_in_progress),
              d->opd_pre_status);

        return 1;
}

/**
 * Reserve object in precreate pool
 *
 * When the caller wants to create a new object on this target (target
 * represented by the given OSP), it should declare this intention using
 * a regular ->dt_declare_create() OSD API method. Then OSP will be trying
 * to reserve an object in the existing precreated pool or wait up to
 * obd_timeout for the available object to appear in the pool (a dedicated
 * thread will be doing real precreation in background). The object can be
 * consumed later with osp_precreate_get_fid() or be released with call to
 * lu_object_put(). Notice the function doesn't reserve a specific ID, just
 * some ID. The actual ID assignment happen in osp_precreate_get_fid().
 * If the space on the target is short and there is a pending object destroy,
 * then the function forces local commit to speedup space release (see
 * osp_sync.c for the details).
 *
 * \param[in] env       LU environment provided by the caller
 * \param[in] d         OSP device
 *
 * \retval              0 on success
 * \retval              -ENOSPC when no space on OST
 * \retval              -EAGAIN try later, slow precreation in progress
 * \retval              -EIO when no access to OST
 */
int osp_precreate_reserve(const struct lu_env *env, struct osp_device *d)
{
        time64_t expire = ktime_get_seconds() + obd_timeout;
        struct l_wait_info lwi;
        int precreated, rc, synced = 0;

        ENTRY;

        LASSERTF(osp_objs_precreated(env, d) >= 0, "Last created FID "DFID
                 "Next FID "DFID"\n", PFID(&d->opd_pre_last_created_fid),
                 PFID(&d->opd_pre_used_fid));

        /* opd_pre_max_create_count 0 to not use specified OST. */
        if (d->opd_pre_max_create_count == 0)
                RETURN(-ENOBUFS);

        /*
         * wait till:
         *  - preallocation is done
         *  - no free space expected soon
         *  - can't connect to OST for too long (obd_timeout)
         *  - OST can allocate fid sequence.
         */
        while ((rc = d->opd_pre_status) == 0 || rc == -ENOSPC ||
                rc == -ENODEV || rc == -EAGAIN || rc == -ENOTCONN) {

                /*
                 * increase number of precreations
                 */
                precreated = osp_objs_precreated(env, d);
                if (d->opd_pre_create_count < d->opd_pre_max_create_count &&
                    d->opd_pre_create_slow == 0 &&
                    precreated <= (d->opd_pre_create_count / 4 + 1)) {
                        spin_lock(&d->opd_pre_lock);
                        d->opd_pre_create_slow = 1;
                        d->opd_pre_create_count *= 2;
                        spin_unlock(&d->opd_pre_lock);
                }

                spin_lock(&d->opd_pre_lock);
                precreated = osp_objs_precreated(env, d);
                if (precreated > d->opd_pre_reserved &&
                    !d->opd_pre_recovering) {
                        d->opd_pre_reserved++;
                        spin_unlock(&d->opd_pre_lock);
                        rc = 0;

                        /* XXX: don't wake up if precreation is in progress */
                        if (osp_precreate_near_empty_nolock(env, d) &&
                           !osp_precreate_end_seq_nolock(env, d))
                                wake_up(&d->opd_pre_waitq);

                        break;
                }
                spin_unlock(&d->opd_pre_lock);

                /*
                 * all precreated objects have been used and no-space
                 * status leave us no chance to succeed very soon
                 * but if there is destroy in progress, then we should
                 * wait till that is done - some space might be released
                 */
                if (unlikely(rc == -ENOSPC)) {
                        if (atomic_read(&d->opd_sync_changes) && synced == 0) {
                                /* force local commit to release space */
                                dt_commit_async(env, d->opd_storage);
                                osp_sync_check_for_work(d);
                                synced = 1;
                        }
                        if (atomic_read(&d->opd_sync_rpcs_in_progress)) {
                                /* just wait till destroys are done */
                                /* see l_wait_even() few lines below */
                        }
                        if (atomic_read(&d->opd_sync_changes) +
                            atomic_read(&d->opd_sync_rpcs_in_progress) == 0) {
                                /* no hope for free space */
                                break;
                        }
                }

                /* XXX: don't wake up if precreation is in progress */
                wake_up(&d->opd_pre_waitq);

                lwi = LWI_TIMEOUT(cfs_time_seconds(obd_timeout),
                                  osp_precreate_timeout_condition, d);
                if (ktime_get_seconds() >= expire) {
                        rc = -ETIMEDOUT;
                        break;
                }

                l_wait_event(d->opd_pre_user_waitq,
                             osp_precreate_ready_condition(env, d), &lwi);
        }

        RETURN(rc);
}

/**
 * Get a FID from precreation pool
 *
 * The function is a companion for osp_precreate_reserve() - it assigns
 * a specific FID from the precreate. The function should be called only
 * if the call to osp_precreate_reserve() was successful. The function
 * updates a local storage to remember the highest object ID referenced
 * by the node in the given sequence.
 *
 * A very importan details: this is supposed to be called once the
 * transaction is started, so on-disk update will be atomic with the
 * data (like LOVEA) refering this object. Then the object won't be leaked:
 * either it's referenced by the committed transaction or it's a subject
 * to the orphan cleanup procedure.
 *
 * \param[in] env       LU environment provided by the caller
 * \param[in] d         OSP device
 * \param[out] fid      generated FID
 *
 * \retval 0            on success
 * \retval negative     negated errno on error
 */
int osp_precreate_get_fid(const struct lu_env *env, struct osp_device *d,
                          struct lu_fid *fid)
{
        struct lu_fid *pre_used_fid = &d->opd_pre_used_fid;
        /* grab next id from the pool */
        spin_lock(&d->opd_pre_lock);

        LASSERTF(osp_fid_diff(&d->opd_pre_used_fid,
                             &d->opd_pre_last_created_fid) < 0,
                 "next fid "DFID" last created fid "DFID"\n",
                 PFID(&d->opd_pre_used_fid),
                 PFID(&d->opd_pre_last_created_fid));

        /*
         * When sequence is used up, new one should be allocated in
         * osp_precreate_rollover_new_seq. So ASSERT here to avoid
         * objid overflow.
         */
        LASSERTF(osp_fid_end_seq(env, pre_used_fid) == 0,
                 "next fid "DFID" last created fid "DFID"\n",
                 PFID(&d->opd_pre_used_fid),
                 PFID(&d->opd_pre_last_created_fid));
        /* Non IDIF fids shoulnd't get here with oid == 0xFFFFFFFF. */
        if (fid_is_idif(pre_used_fid) &&
            unlikely(fid_oid(pre_used_fid) == LUSTRE_DATA_SEQ_MAX_WIDTH))
                pre_used_fid->f_seq++;

        d->opd_pre_used_fid.f_oid++;
        memcpy(fid, &d->opd_pre_used_fid, sizeof(*fid));
        d->opd_pre_reserved--;
        /*
         * last_used_id must be changed along with getting new id otherwise
         * we might miscalculate gap causing object loss or leak
         */
        osp_update_last_fid(d, fid);
        spin_unlock(&d->opd_pre_lock);

        /*
         * probably main thread suspended orphan cleanup till
         * all reservations are released, see comment in
         * osp_precreate_thread() just before orphan cleanup
         */
        if (unlikely(d->opd_pre_reserved == 0 &&
                     (d->opd_pre_recovering || d->opd_pre_status)))
                wake_up(&d->opd_pre_waitq);

        return 0;
}

/*
 * Set size regular attribute on an object
 *
 * When a striping is created late, it's possible that size is already
 * initialized on the file. Then the new striping should inherit size
 * from the file. The function sets size on the object using the regular
 * protocol (OST_PUNCH).
 * XXX: should be re-implemented using OUT ?
 *
 * \param[in] env       LU environment provided by the caller
 * \param[in] dt        object
 * \param[in] size      size to set.
 *
 * \retval 0            on success
 * \retval negative     negated errno on error
 */
int osp_object_truncate(const struct lu_env *env, struct dt_object *dt,
                        __u64 size)
{
        struct osp_device       *d = lu2osp_dev(dt->do_lu.lo_dev);
        struct ptlrpc_request   *req = NULL;
        struct obd_import       *imp;
        struct ost_body         *body;
        struct obdo             *oa = NULL;
        int                      rc;

        ENTRY;

        imp = d->opd_obd->u.cli.cl_import;
        LASSERT(imp);

        req = ptlrpc_request_alloc(imp, &RQF_OST_PUNCH);
        if (req == NULL)
                RETURN(-ENOMEM);

        rc = ptlrpc_request_pack(req, LUSTRE_OST_VERSION, OST_PUNCH);
        if (rc) {
                ptlrpc_request_free(req);
                RETURN(rc);
        }

        /*
         * XXX: decide how do we do here with resend
         * if we don't resend, then client may see wrong file size
         * if we do resend, then MDS thread can get stuck for quite long
         * and if we don't resend, then client will also get -EWOULDBLOCK !!
         * (see LU-7975 and sanity/test_27F use cases)
         * but let's decide not to resend/delay this truncate request to OST
         * and allow Client to decide to resend, in a less agressive way from
         * after_reply(), by returning -EINPROGRESS instead of
         * -EAGAIN/-EWOULDBLOCK upon return from ptlrpc_queue_wait() at the
         * end of this routine
         */
        req->rq_no_resend = req->rq_no_delay = 1;

        req->rq_request_portal = OST_IO_PORTAL; /* bug 7198 */
        ptlrpc_at_set_req_timeout(req);

        OBD_ALLOC_PTR(oa);
        if (oa == NULL)
                GOTO(out, rc = -ENOMEM);

        rc = fid_to_ostid(lu_object_fid(&dt->do_lu), &oa->o_oi);
        LASSERT(rc == 0);
        oa->o_size = size;
        oa->o_blocks = OBD_OBJECT_EOF;
        oa->o_valid = OBD_MD_FLSIZE | OBD_MD_FLBLOCKS |
                      OBD_MD_FLID | OBD_MD_FLGROUP;

        body = req_capsule_client_get(&req->rq_pill, &RMF_OST_BODY);
        LASSERT(body);
        lustre_set_wire_obdo(&req->rq_import->imp_connect_data, &body->oa, oa);

        /* XXX: capa support? */
        /* osc_pack_capa(req, body, capa); */

        ptlrpc_request_set_replen(req);

        rc = ptlrpc_queue_wait(req);
        if (rc) {
                /* -EWOULDBLOCK/-EAGAIN means OST is unreachable at the moment
                 * since we have decided not to resend/delay, but this could
                 * lead to wrong size to be seen at Client side and even process
                 * trying to open to exit/fail if not itself handling -EAGAIN.
                 * So it should be better to return -EINPROGRESS instead and
                 * leave the decision to resend at Client side in after_reply()
                 */
                if (rc == -EWOULDBLOCK) {
                        rc = -EINPROGRESS;
                        CDEBUG(D_HA, "returning -EINPROGRESS instead of "
                               "-EWOULDBLOCK/-EAGAIN to allow Client to "
                               "resend\n");
                } else {
                        CERROR("can't punch object: %d\n", rc);
                }
        }
out:
        ptlrpc_req_finished(req);
        if (oa)
                OBD_FREE_PTR(oa);
        RETURN(rc);
}

/**
 * Initialize precreation functionality of OSP
 *
 * Prepares all the internal structures and starts the precreate thread
 *
 * \param[in] d         OSP device
 *
 * \retval 0            on success
 * \retval negative     negated errno on error
 */
int osp_init_precreate(struct osp_device *d)
{
        ENTRY;

        OBD_ALLOC_PTR(d->opd_pre);
        if (d->opd_pre == NULL)
                RETURN(-ENOMEM);

        /* initially precreation isn't ready */
        init_waitqueue_head(&d->opd_pre_user_waitq);
        d->opd_pre_status = -EAGAIN;
        fid_zero(&d->opd_pre_used_fid);
        d->opd_pre_used_fid.f_oid = 1;
        fid_zero(&d->opd_pre_last_created_fid);
        d->opd_pre_last_created_fid.f_oid = 1;
        d->opd_last_id = 0;
        d->opd_pre_reserved = 0;
        d->opd_got_disconnected = 1;
        d->opd_pre_create_slow = 0;
        d->opd_pre_create_count = OST_MIN_PRECREATE;
        d->opd_pre_min_create_count = OST_MIN_PRECREATE;
        d->opd_pre_max_create_count = OST_MAX_PRECREATE;
        d->opd_reserved_mb_high = 0;
        d->opd_reserved_mb_low = 0;

        RETURN(0);
}

/**
 * Finish precreate functionality of OSP
 *
 *
 * Asks all the activity (the thread, update timer) to stop, then
 * wait till that is done.
 *
 * \param[in] d         OSP device
 */
void osp_precreate_fini(struct osp_device *d)
{
        ENTRY;

        if (d->opd_pre == NULL)
                RETURN_EXIT;

        OBD_FREE_PTR(d->opd_pre);
        d->opd_pre = NULL;

        EXIT;
}

int osp_init_statfs(struct osp_device *d)
{
        struct task_struct              *task;

        ENTRY;

        spin_lock_init(&d->opd_pre_lock);
        init_waitqueue_head(&d->opd_pre_waitq);
        thread_set_flags(&d->opd_pre_thread, SVC_INIT);
        init_waitqueue_head(&d->opd_pre_thread.t_ctl_waitq);

        /*
         * Initialize statfs-related things
         */
        d->opd_statfs_maxage = 5; /* defaultupdate interval */
        d->opd_statfs_fresh_till = ktime_sub_ns(ktime_get(),
                                                1000 * NSEC_PER_SEC);
        CDEBUG(D_OTHER, "current %lldns, fresh till %lldns\n",
               ktime_get_ns(),
               ktime_to_ns(d->opd_statfs_fresh_till));
        cfs_timer_setup(&d->opd_statfs_timer, osp_statfs_timer_cb,
                        (unsigned long)d, 0);

        if (d->opd_storage->dd_rdonly)
                RETURN(0);

        /*
         * start thread handling precreation and statfs updates
         */
        task = kthread_run(osp_precreate_thread, d,
                           "osp-pre-%u-%u", d->opd_index, d->opd_group);
        if (IS_ERR(task)) {
                CERROR("can't start precreate thread %ld\n", PTR_ERR(task));
                RETURN(PTR_ERR(task));
        }

        wait_event_idle(d->opd_pre_thread.t_ctl_waitq,
                        osp_precreate_running(d) || osp_precreate_stopped(d));


        RETURN(0);
}

void osp_statfs_fini(struct osp_device *d)
{
        struct ptlrpc_thread *thread = &d->opd_pre_thread;
        ENTRY;

        del_timer(&d->opd_statfs_timer);

        if (!thread_is_init(thread) && !thread_is_stopped(thread)) {
                thread->t_flags = SVC_STOPPING;
                wake_up(&d->opd_pre_waitq);
                wait_event(thread->t_ctl_waitq, thread_is_stopped(thread));
        }

        EXIT;
}