[fs/lustre-release.git] / lustre / osp / osp_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 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_dev.c
 *
 * Author: Alex Zhuravlev <alexey.zhuravlev@intel.com>
 * Author: Mikhail Pershin <mike.pershin@intel.com>
 * Author: Di Wang <di.wang@intel.com>
 */
/*
 * The Object Storage Proxy (OSP) module provides an implementation of
 * the DT API for remote MDTs and OSTs. Every local OSP device (or
 * object) is a proxy for a remote OSD device (or object). Thus OSP
 * converts DT operations into RPCs, which are sent to the OUT service
 * on a remote target, converted back to DT operations, and
 * executed. Of course there are many ways in which this description
 * is inaccurate but it's a good enough mental model. OSP is used by
 * the MDT stack in several ways:
 *
 * - OSP devices allocate FIDs for the stripe sub-objects of a striped
 *   file or directory.
 *
 * - OSP objects represent the remote MDT and OST objects that are
 *   the stripes of a striped object.
 *
 * - OSP devices log, send, and track synchronous operations (setattr
 *   and unlink) to remote targets.
 *
 * - OSP objects are the bottom slice of the compound LU object
 *   representing a remote MDT object: MDT/MDD/LOD/OSP.
 *
 * - OSP objects are used by LFSCK to represent remote OST objects
 *   during the verification of MDT-OST consistency.
 *
 * - OSP devices batch idempotent requests (declare_attr_get() and
 *   declare_xattr_get()) to the remote target and cache their results.
 *
 * In addition the OSP layer implements a subset of the OBD device API
 * to support being a client of a remote target, connecting to other
 * layers, and FID allocation.
 */

#define DEBUG_SUBSYSTEM S_MDS

#include <linux/kthread.h>

#include <uapi/linux/lustre/lustre_ioctl.h>
#include <lustre_log.h>
#include <lustre_obdo.h>
#include <uapi/linux/lustre/lustre_param.h>
#include <obd_class.h>

#include "osp_internal.h"

/* Slab for OSP object allocation */
struct kmem_cache *osp_object_kmem;

static struct lu_kmem_descr osp_caches[] = {
        {
                .ckd_cache = &osp_object_kmem,
                .ckd_name  = "osp_obj",
                .ckd_size  = sizeof(struct osp_object)
        },
        {
                .ckd_cache = NULL
        }
};

/**
 * Implementation of lu_device_operations::ldo_object_alloc
 *
 * Allocates an OSP object in memory, whose FID is on the remote target.
 *
 * \param[in] env       execution environment
 * \param[in] hdr       The header of the object stack. If it is NULL, it
 *                      means the object is not built from top device, i.e.
 *                      it is a sub-stripe object of striped directory or
 *                      an OST object.
 * \param[in] d         OSP device
 *
 * \retval object       object being created if the creation succeed.
 * \retval NULL         NULL if the creation failed.
 */
static struct lu_object *osp_object_alloc(const struct lu_env *env,
                                          const struct lu_object_header *hdr,
                                          struct lu_device *d)
{
        struct lu_object_header *h = NULL;
        struct osp_object       *o;
        struct lu_object        *l;

        OBD_SLAB_ALLOC_PTR_GFP(o, osp_object_kmem, GFP_NOFS);
        if (o != NULL) {
                l = &o->opo_obj.do_lu;

                /* If hdr is NULL, it means the object is not built
                 * from the top dev(MDT/OST), usually it happens when
                 * building striped object, like data object on MDT or
                 * striped object for directory */
                if (hdr == NULL) {
                        h = &o->opo_header;
                        lu_object_header_init(h);
                        dt_object_init(&o->opo_obj, h, d);
                        lu_object_add_top(h, l);
                } else {
                        dt_object_init(&o->opo_obj, h, d);
                }

                l->lo_ops = &osp_lu_obj_ops;

                return l;
        } else {
                return NULL;
        }
}

/**
 * Find or create the local object
 *
 * Finds or creates the local file referenced by \a reg_id and return the
 * attributes of the local file.
 *
 * \param[in] env       execution environment
 * \param[in] osp       OSP device
 * \param[out] attr     attributes of the object
 * \param[in] reg_id    the local object ID of the file. It will be used
 *                      to compose a local FID{FID_SEQ_LOCAL_FILE, reg_id, 0}
 *                      to identify the object.
 *
 * \retval object               object(dt_object) found or created
 * \retval ERR_PTR(errno)       ERR_PTR(errno) if not get the object.
 */
static struct dt_object
*osp_find_or_create_local_file(const struct lu_env *env, struct osp_device *osp,
                               struct lu_attr *attr, __u32 reg_id)
{
        struct osp_thread_info *osi = osp_env_info(env);
        struct dt_object_format dof = { 0 };
        struct dt_object       *dto;
        int                  rc;
        ENTRY;

        lu_local_obj_fid(&osi->osi_fid, reg_id);
        attr->la_valid = LA_MODE;
        attr->la_mode = S_IFREG | 0644;
        dof.dof_type = DFT_REGULAR;
        /* Find or create the local object by osi_fid. */
        dto = dt_find_or_create(env, osp->opd_storage, &osi->osi_fid,
                                &dof, attr);
        if (IS_ERR(dto))
                RETURN(dto);

        /* Get attributes of the local object. */
        rc = dt_attr_get(env, dto, attr);
        if (rc) {
                CERROR("%s: can't be initialized: rc = %d\n",
                       osp->opd_obd->obd_name, rc);
                dt_object_put(env, dto);
                RETURN(ERR_PTR(rc));
        }
        RETURN(dto);
}

/**
 * Write data buffer to a local file object.
 *
 * \param[in] env       execution environment
 * \param[in] osp       OSP device
 * \param[in] dt_obj    object written to
 * \param[in] buf       buffer containing byte array and length
 * \param[in] offset    write offset in the object in bytes
 *
 * \retval 0            0 if write succeed
 * \retval -EFAULT      -EFAULT if only part of buffer is written.
 * \retval negative             other negative errno if write failed.
 */
static int osp_write_local_file(const struct lu_env *env,
                                struct osp_device *osp,
                                struct dt_object *dt_obj,
                                struct lu_buf *buf,
                                loff_t offset)
{
        struct thandle *th;
        int rc;

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

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

        rc = dt_declare_record_write(env, dt_obj, buf, offset, th);
        if (rc)
                GOTO(out, rc);
        rc = dt_trans_start_local(env, osp->opd_storage, th);
        if (rc)
                GOTO(out, rc);

        rc = dt_record_write(env, dt_obj, buf, &offset, th);
out:
        dt_trans_stop(env, osp->opd_storage, th);
        RETURN(rc);
}

/**
 * Initialize last ID object.
 *
 * This function initializes the LAST_ID file, which stores the current last
 * used id of data objects. The MDT will use the last used id and the last_seq
 * (\see osp_init_last_seq()) to synchronize the precreate object cache with
 * OSTs.
 *
 * \param[in] env       execution environment
 * \param[in] osp       OSP device
 *
 * \retval 0            0 if initialization succeed
 * \retval negative     negative errno if initialization failed
 */
static int osp_init_last_objid(const struct lu_env *env, struct osp_device *osp)
{
        struct osp_thread_info  *osi = osp_env_info(env);
        struct lu_fid           *fid = &osp->opd_last_used_fid;
        struct dt_object        *dto;
        int                     rc = -EFAULT;
        ENTRY;

        dto = osp_find_or_create_local_file(env, osp, &osi->osi_attr,
                                            MDD_LOV_OBJ_OID);
        if (IS_ERR(dto))
                RETURN(PTR_ERR(dto));

        /* object will be released in device cleanup path */
        if (osi->osi_attr.la_size >=
            sizeof(osi->osi_id) * (osp->opd_index + 1)) {
                osp_objid_buf_prep(&osi->osi_lb, &osi->osi_off, &osi->osi_id,
                                   osp->opd_index);
                rc = dt_record_read(env, dto, &osi->osi_lb, &osi->osi_off);
                if (rc != 0 && rc != -EFAULT)
                        GOTO(out, rc);
                /* In case of idif bits 32-48 go to f_seq
                 * (see osp_init_last_seq). So don't care
                 * about u64->u32 convertion. */
                fid->f_oid = osi->osi_id;
        }

        if (rc == -EFAULT) { /* fresh LAST_ID */
                osi->osi_id = 0;
                fid->f_oid = 0;
                osp_objid_buf_prep(&osi->osi_lb, &osi->osi_off, &osi->osi_id,
                                   osp->opd_index);
                rc = osp_write_local_file(env, osp, dto, &osi->osi_lb,
                                          osi->osi_off);
                if (rc != 0)
                        GOTO(out, rc);
        }
        osp->opd_last_used_oid_file = dto;
        RETURN(0);
out:
        /* object will be released in device cleanup path */
        CERROR("%s: can't initialize lov_objid: rc = %d\n",
               osp->opd_obd->obd_name, rc);
        dt_object_put(env, dto);
        osp->opd_last_used_oid_file = NULL;
        RETURN(rc);
}

/**
 * Initialize last sequence object.
 *
 * This function initializes the LAST_SEQ file in the local OSD, which stores
 * the current last used sequence of data objects. The MDT will use the last
 * sequence and last id (\see osp_init_last_objid()) to synchronize the
 * precreate object cache with OSTs.
 *
 * \param[in] env       execution environment
 * \param[in] osp       OSP device
 *
 * \retval 0            0 if initialization succeed
 * \retval negative     negative errno if initialization failed
 */
static int osp_init_last_seq(const struct lu_env *env, struct osp_device *osp)
{
        struct osp_thread_info  *osi = osp_env_info(env);
        struct lu_fid           *fid = &osp->opd_last_used_fid;
        struct dt_object        *dto;
        int                     rc = -EFAULT;
        ENTRY;

        dto = osp_find_or_create_local_file(env, osp, &osi->osi_attr,
                                            MDD_LOV_OBJ_OSEQ);
        if (IS_ERR(dto))
                RETURN(PTR_ERR(dto));

        /* object will be released in device cleanup path */
        if (osi->osi_attr.la_size >=
            sizeof(osi->osi_id) * (osp->opd_index + 1)) {
                osp_objseq_buf_prep(&osi->osi_lb, &osi->osi_off, &fid->f_seq,
                                   osp->opd_index);
                rc = dt_record_read(env, dto, &osi->osi_lb, &osi->osi_off);
                if (rc != 0 && rc != -EFAULT)
                        GOTO(out, rc);
                if (fid_is_idif(fid))
                        fid->f_seq = fid_idif_seq(osi->osi_id, osp->opd_index);
        }

        if (rc == -EFAULT) { /* fresh OSP */
                fid->f_seq = 0;
                osp_objseq_buf_prep(&osi->osi_lb, &osi->osi_off, &fid->f_seq,
                                    osp->opd_index);
                rc = osp_write_local_file(env, osp, dto, &osi->osi_lb,
                                          osi->osi_off);
                if (rc != 0)
                        GOTO(out, rc);
        }
        osp->opd_last_used_seq_file = dto;
        RETURN(0);
out:
        /* object will be released in device cleanup path */
        CERROR("%s: can't initialize lov_seq: rc = %d\n",
               osp->opd_obd->obd_name, rc);
        dt_object_put(env, dto);
        osp->opd_last_used_seq_file = NULL;
        RETURN(rc);
}

/**
 * Initialize last OID and sequence object.
 *
 * If the MDT is just upgraded to 2.4 from the lower version, where the
 * LAST_SEQ file does not exist, the file will be created and IDIF sequence
 * will be written into the file.
 *
 * \param[in] env       execution environment
 * \param[in] osp       OSP device
 *
 * \retval 0            0 if initialization succeed
 * \retval negative     negative error if initialization failed
 */
static int osp_last_used_init(const struct lu_env *env, struct osp_device *osp)
{
        struct osp_thread_info *osi = osp_env_info(env);
        int                  rc;
        ENTRY;

        fid_zero(&osp->opd_last_used_fid);
        rc = osp_init_last_objid(env, osp);
        if (rc < 0) {
                CERROR("%s: Can not get ids %d from old objid!\n",
                       osp->opd_obd->obd_name, rc);
                RETURN(rc);
        }

        rc = osp_init_last_seq(env, osp);
        if (rc < 0) {
                CERROR("%s: Can not get sequence %d from old objseq!\n",
                       osp->opd_obd->obd_name, rc);
                GOTO(out, rc);
        }

        if (fid_oid(&osp->opd_last_used_fid) != 0 &&
            fid_seq(&osp->opd_last_used_fid) == 0) {
                /* Just upgrade from the old version,
                 * set the seq to be IDIF */
                osp->opd_last_used_fid.f_seq =
                   fid_idif_seq(fid_oid(&osp->opd_last_used_fid),
                                osp->opd_index);
                osp_objseq_buf_prep(&osi->osi_lb, &osi->osi_off,
                                    &osp->opd_last_used_fid.f_seq,
                                    osp->opd_index);
                rc = osp_write_local_file(env, osp, osp->opd_last_used_seq_file,
                                          &osi->osi_lb, osi->osi_off);
                if (rc) {
                        CERROR("%s : Can not write seq file: rc = %d\n",
                               osp->opd_obd->obd_name, rc);
                        GOTO(out, rc);
                }
        }

        if (!fid_is_zero(&osp->opd_last_used_fid) &&
                 !fid_is_sane(&osp->opd_last_used_fid)) {
                CERROR("%s: Got invalid FID "DFID"\n", osp->opd_obd->obd_name,
                        PFID(&osp->opd_last_used_fid));
                GOTO(out, rc = -EINVAL);
        }

        CDEBUG(D_INFO, "%s: Init last used fid "DFID"\n",
               osp->opd_obd->obd_name, PFID(&osp->opd_last_used_fid));
out:
        if (rc != 0) {
                if (osp->opd_last_used_oid_file != NULL) {
                        dt_object_put(env, osp->opd_last_used_oid_file);
                        osp->opd_last_used_oid_file = NULL;
                }
                if (osp->opd_last_used_seq_file != NULL) {
                        dt_object_put(env, osp->opd_last_used_seq_file);
                        osp->opd_last_used_seq_file = NULL;
                }
        }

        RETURN(rc);
}

/**
 * Release the last sequence and OID file objects in OSP device.
 *
 * \param[in] env       execution environment
 * \param[in] osp       OSP device
 */
static void osp_last_used_fini(const struct lu_env *env, struct osp_device *osp)
{
        /* release last_used file */
        if (osp->opd_last_used_oid_file != NULL) {
                dt_object_put(env, osp->opd_last_used_oid_file);
                osp->opd_last_used_oid_file = NULL;
        }

        if (osp->opd_last_used_seq_file != NULL) {
                dt_object_put(env, osp->opd_last_used_seq_file);
                osp->opd_last_used_seq_file = NULL;
        }
}

/**
 * Disconnects the connection between OSP and its correspondent MDT or OST, and
 * the import will be marked as inactive. It will only be called during OSP
 * cleanup process.
 *
 * \param[in] d         OSP device being disconnected
 *
 * \retval 0            0 if disconnection succeed
 * \retval negative     negative errno if disconnection failed
 */
static int osp_disconnect(struct osp_device *d)
{
        struct obd_device *obd = d->opd_obd;
        struct obd_import *imp;
        int rc = 0;

        imp = obd->u.cli.cl_import;

        /* Mark import deactivated now, so we don't try to reconnect if any
         * of the cleanup RPCs fails (e.g. ldlm cancel, etc).  We don't
         * fully deactivate the import, or that would drop all requests. */
        LASSERT(imp != NULL);
        spin_lock(&imp->imp_lock);
        imp->imp_deactive = 1;
        spin_unlock(&imp->imp_lock);

        ptlrpc_deactivate_import(imp);

        /* Some non-replayable imports (MDS's OSCs) are pinged, so just
         * delete it regardless.  (It's safe to delete an import that was
         * never added.) */
        (void)ptlrpc_pinger_del_import(imp);

        rc = ptlrpc_disconnect_import(imp, 0);
        if (rc != 0)
                CERROR("%s: can't disconnect: rc = %d\n", obd->obd_name, rc);

        ptlrpc_invalidate_import(imp);

        RETURN(rc);
}

/**
 * Initialize the osp_update structure in OSP device
 *
 * Allocate osp update structure and start update thread.
 *
 * \param[in] osp       OSP device
 *
 * \retval              0 if initialization succeeds.
 * \retval              negative errno if initialization fails.
 */
static int osp_update_init(struct osp_device *osp)
{
        struct l_wait_info      lwi = { 0 };
        struct task_struct      *task;

        ENTRY;

        LASSERT(osp->opd_connect_mdt);

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

        OBD_ALLOC_PTR(osp->opd_update);
        if (osp->opd_update == NULL)
                RETURN(-ENOMEM);

        init_waitqueue_head(&osp->opd_update_thread.t_ctl_waitq);
        init_waitqueue_head(&osp->opd_update->ou_waitq);
        spin_lock_init(&osp->opd_update->ou_lock);
        INIT_LIST_HEAD(&osp->opd_update->ou_list);
        osp->opd_update->ou_rpc_version = 1;
        osp->opd_update->ou_version = 1;
        osp->opd_update->ou_generation = 0;

        /* start thread handling sending updates to the remote MDT */
        task = kthread_run(osp_send_update_thread, osp,
                           "osp_up%u-%u", osp->opd_index, osp->opd_group);
        if (IS_ERR(task)) {
                int rc = PTR_ERR(task);

                OBD_FREE_PTR(osp->opd_update);
                osp->opd_update = NULL;
                CERROR("%s: can't start precreate thread: rc = %d\n",
                       osp->opd_obd->obd_name, rc);
                RETURN(rc);
        }

        l_wait_event(osp->opd_update_thread.t_ctl_waitq,
                     osp_send_update_thread_running(osp) ||
                     osp_send_update_thread_stopped(osp), &lwi);

        RETURN(0);
}

/**
 * Finialize osp_update structure in OSP device
 *
 * Stop the OSP update sending thread, then delete the left
 * osp thandle in the sending list.
 *
 * \param [in] osp      OSP device.
 */
static void osp_update_fini(const struct lu_env *env, struct osp_device *osp)
{
        struct osp_update_request *our;
        struct osp_update_request *tmp;
        struct osp_updates *ou = osp->opd_update;

        if (ou == NULL)
                return;

        osp->opd_update_thread.t_flags = SVC_STOPPING;
        wake_up(&ou->ou_waitq);

        wait_event(osp->opd_update_thread.t_ctl_waitq,
                   osp->opd_update_thread.t_flags & SVC_STOPPED);

        /* Remove the left osp thandle from the list */
        spin_lock(&ou->ou_lock);
        list_for_each_entry_safe(our, tmp, &ou->ou_list,
                                 our_list) {
                list_del_init(&our->our_list);
                LASSERT(our->our_th != NULL);
                osp_trans_callback(env, our->our_th, -EIO);
                /* our will be destroyed in osp_thandle_put() */
                osp_thandle_put(env, our->our_th);
        }
        spin_unlock(&ou->ou_lock);

        OBD_FREE_PTR(ou);
        osp->opd_update = NULL;
}

/**
 * Cleanup OSP, which includes disconnect import, cleanup unlink log, stop
 * precreate threads etc.
 *
 * \param[in] env       execution environment.
 * \param[in] d         OSP device being disconnected.
 *
 * \retval 0            0 if cleanup succeed
 * \retval negative     negative errno if cleanup failed
 */
static int osp_shutdown(const struct lu_env *env, struct osp_device *d)
{
        int                      rc = 0;
        ENTRY;

        LASSERT(env);

        rc = osp_disconnect(d);

        if (!d->opd_connect_mdt) {
                /* stop sync thread */
                osp_sync_fini(d);

                /* stop precreate thread */
                osp_precreate_fini(d);

                /* release last_used file */
                osp_last_used_fini(env, d);
        }

        obd_fid_fini(d->opd_obd);

        RETURN(rc);
}

/**
 * Implementation of osp_lu_ops::ldo_process_config
 *
 * This function processes config log records in OSP layer. It is usually
 * called from the top layer of MDT stack, and goes through the stack by calling
 * ldo_process_config of next layer.
 *
 * \param[in] env       execution environment
 * \param[in] dev       lu_device of OSP
 * \param[in] lcfg      config log
 *
 * \retval 0            0 if the config log record is executed correctly.
 * \retval negative     negative errno if the record execution fails.
 */
static int osp_process_config(const struct lu_env *env,
                              struct lu_device *dev, struct lustre_cfg *lcfg)
{
        struct osp_device *d = lu2osp_dev(dev);
        struct dt_device *dt = lu2dt_dev(dev);
        struct obd_device *obd = d->opd_obd;
        ssize_t count;
        int rc;

        ENTRY;

        switch (lcfg->lcfg_command) {
        case LCFG_PRE_CLEANUP:
                rc = osp_disconnect(d);
                osp_update_fini(env, d);
                if (obd->obd_namespace != NULL)
                        ldlm_namespace_free_prior(obd->obd_namespace, NULL, 1);
                break;
        case LCFG_CLEANUP:
                lu_dev_del_linkage(dev->ld_site, dev);
                rc = osp_shutdown(env, d);
                break;
        case LCFG_PARAM:
                count = class_modify_config(lcfg, d->opd_connect_mdt ?
                                                  PARAM_OSP : PARAM_OSC,
                                            &dt->dd_kobj);
                if (count < 0) {
                        /* class_modify_config() haven't found matching
                         * parameter and returned an error so that layer(s)
                         * below could use that. But OSP is the bottom, so
                         * just ignore it
                         */
                        CERROR("%s: unknown param %s\n",
                               (char *)lustre_cfg_string(lcfg, 0),
                               (char *)lustre_cfg_string(lcfg, 1));
                }
                rc = 0;
                break;
        default:
                CERROR("%s: unknown command %u\n",
                       (char *)lustre_cfg_string(lcfg, 0), lcfg->lcfg_command);
                rc = 0;
                break;
        }

        RETURN(rc);
}

/**
 * Implementation of osp_lu_ops::ldo_recovery_complete
 *
 * This function is called after recovery is finished, and OSP layer
 * will wake up precreate thread here.
 *
 * \param[in] env       execution environment
 * \param[in] dev       lu_device of OSP
 *
 * \retval 0            0 unconditionally
 */
static int osp_recovery_complete(const struct lu_env *env,
                                 struct lu_device *dev)
{
        struct osp_device       *osp = lu2osp_dev(dev);

        ENTRY;
        osp->opd_recovery_completed = 1;

        if (!osp->opd_connect_mdt && osp->opd_pre != NULL)
                wake_up(&osp->opd_pre_waitq);

        RETURN(0);
}

const struct lu_device_operations osp_lu_ops = {
        .ldo_object_alloc       = osp_object_alloc,
        .ldo_process_config     = osp_process_config,
        .ldo_recovery_complete  = osp_recovery_complete,
};

/**
 * Implementation of dt_device_operations::dt_statfs
 *
 * This function provides statfs status (for precreation) from
 * corresponding OST. Note: this function only retrieves the status
 * from the OSP device, and the real statfs RPC happens inside
 * precreate thread (\see osp_statfs_update). Note: OSP for MDT does
 * not need to retrieve statfs data for now.
 *
 * \param[in] env       execution environment.
 * \param[in] dev       dt_device of OSP.
 * \param[out] sfs      holds the retrieved statfs data.
 *
 * \retval 0            0 statfs data was retrieved successfully or
 *                      retrieval was not needed
 * \retval negative     negative errno if get statfs failed.
 */
static int osp_statfs(const struct lu_env *env, struct dt_device *dev,
                      struct obd_statfs *sfs)
{
        struct osp_device *d = dt2osp_dev(dev);
        struct obd_import *imp = d->opd_obd->u.cli.cl_import;

        ENTRY;

        if (imp->imp_state == LUSTRE_IMP_CLOSED)
                RETURN(-ESHUTDOWN);

        if (unlikely(d->opd_imp_active == 0))
                RETURN(-ENOTCONN);

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

        /* return recently updated data */
        *sfs = d->opd_statfs;

        /*
         * layer above osp (usually lod) can use ffree to estimate
         * how many objects are available for immediate creation
         */
        spin_lock(&d->opd_pre_lock);
        LASSERTF(fid_seq(&d->opd_pre_last_created_fid) ==
                 fid_seq(&d->opd_pre_used_fid),
                 "last_created "DFID", next_fid "DFID"\n",
                 PFID(&d->opd_pre_last_created_fid),
                 PFID(&d->opd_pre_used_fid));
        sfs->os_fprecreated = fid_oid(&d->opd_pre_last_created_fid) -
                              fid_oid(&d->opd_pre_used_fid);
        sfs->os_fprecreated -= d->opd_pre_reserved;
        LASSERTF(sfs->os_fprecreated <= OST_MAX_PRECREATE * 2,
                 "last_created "DFID", next_fid "DFID", reserved %llu\n",
                 PFID(&d->opd_pre_last_created_fid), PFID(&d->opd_pre_used_fid),
                 d->opd_pre_reserved);
        spin_unlock(&d->opd_pre_lock);

        CDEBUG(D_OTHER, "%s: %llu blocks, %llu free, %llu avail, "
               "%llu files, %llu free files\n", d->opd_obd->obd_name,
               sfs->os_blocks, sfs->os_bfree, sfs->os_bavail,
               sfs->os_files, sfs->os_ffree);
        RETURN(0);
}

static int osp_sync_timeout(void *data)
{
        return 1;
}

/**
 * Implementation of dt_device_operations::dt_sync
 *
 * This function synchronizes the OSP cache to the remote target. It wakes
 * up unlink log threads and sends out unlink records to the remote OST.
 *
 * \param[in] env       execution environment
 * \param[in] dev       dt_device of OSP
 *
 * \retval 0            0 if synchronization succeeds
 * \retval negative     negative errno if synchronization fails
 */
static int osp_sync(const struct lu_env *env, struct dt_device *dev)
{
        struct osp_device *d = dt2osp_dev(dev);
        struct l_wait_info lwi = { 0 };
        time64_t start = ktime_get_seconds();
        int recs, rc = 0;
        u64 old;

        ENTRY;

        /* No Sync between MDTs yet. */
        if (d->opd_connect_mdt)
                RETURN(0);

        recs = atomic_read(&d->opd_sync_changes);
        old = atomic64_read(&d->opd_sync_processed_recs);

        osp_sync_force(env, dt2osp_dev(dev));

        if (unlikely(d->opd_imp_active == 0))
                RETURN(-ENOTCONN);

        down_write(&d->opd_async_updates_rwsem);

        CDEBUG(D_OTHER, "%s: async updates %d\n", d->opd_obd->obd_name,
               atomic_read(&d->opd_async_updates_count));

        /* make sure the connection is fine */
        lwi = LWI_TIMEOUT(cfs_time_seconds(obd_timeout), osp_sync_timeout, d);
        rc = l_wait_event(d->opd_sync_barrier_waitq,
                          atomic_read(&d->opd_async_updates_count) == 0,
                          &lwi);
        up_write(&d->opd_async_updates_rwsem);
        if (rc != 0)
                GOTO(out, rc);

        CDEBUG(D_CACHE, "%s: processed %lu\n", d->opd_obd->obd_name,
               atomic64_read(&d->opd_sync_processed_recs));

        while (atomic64_read(&d->opd_sync_processed_recs) < old + recs) {
                __u64 last = atomic64_read(&d->opd_sync_processed_recs);
                /* make sure the connection is fine */
                lwi = LWI_TIMEOUT(cfs_time_seconds(obd_timeout),
                                  osp_sync_timeout, d);
                l_wait_event(d->opd_sync_barrier_waitq,
                             atomic64_read(&d->opd_sync_processed_recs)
                             >= old + recs, &lwi);

                if (atomic64_read(&d->opd_sync_processed_recs) >= old + recs)
                        break;

                if (atomic64_read(&d->opd_sync_processed_recs) != last) {
                        /* some progress have been made,
                         * keep trying... */
                        continue;
                }

                /* no changes and expired, something is wrong */
                GOTO(out, rc = -ETIMEDOUT);
        }

        /* block new processing (barrier>0 - few callers are possible */
        atomic_inc(&d->opd_sync_barrier);

        CDEBUG(D_CACHE, "%s: %u in flight\n", d->opd_obd->obd_name,
               atomic_read(&d->opd_sync_rpcs_in_flight));

        /* wait till all-in-flight are replied, so executed by the target */
        /* XXX: this is used by LFSCK at the moment, which doesn't require
         *      all the changes to be committed, but in general it'd be
         *      better to wait till commit */
        while (atomic_read(&d->opd_sync_rpcs_in_flight) > 0) {
                old = atomic_read(&d->opd_sync_rpcs_in_flight);

                lwi = LWI_TIMEOUT(cfs_time_seconds(obd_timeout),
                                  osp_sync_timeout, d);
                l_wait_event(d->opd_sync_barrier_waitq,
                             atomic_read(&d->opd_sync_rpcs_in_flight) == 0,
                             &lwi);

                if (atomic_read(&d->opd_sync_rpcs_in_flight) == 0)
                        break;

                if (atomic_read(&d->opd_sync_rpcs_in_flight) != old) {
                        /* some progress have been made */
                        continue;
                }

                /* no changes and expired, something is wrong */
                GOTO(out, rc = -ETIMEDOUT);
        }

out:
        /* resume normal processing (barrier=0) */
        atomic_dec(&d->opd_sync_barrier);
        osp_sync_check_for_work(d);

        CDEBUG(D_CACHE, "%s: done in %lld: rc = %d\n", d->opd_obd->obd_name,
               ktime_get_seconds() - start, rc);

        RETURN(rc);
}

const struct dt_device_operations osp_dt_ops = {
        .dt_statfs       = osp_statfs,
        .dt_sync         = osp_sync,
        .dt_trans_create = osp_trans_create,
        .dt_trans_start  = osp_trans_start,
        .dt_trans_stop   = osp_trans_stop,
        .dt_trans_cb_add   = osp_trans_cb_add,
};

/**
 * Connect OSP to local OSD.
 *
 * Locate the local OSD referenced by \a nextdev and connect to it. Sometimes,
 * OSP needs to access the local OSD to store some information. For example,
 * during precreate, it needs to update last used OID and sequence file
 * (LAST_SEQ) in local OSD.
 *
 * \param[in] env       execution environment
 * \param[in] osp       OSP device
 * \param[in] nextdev   the name of local OSD
 *
 * \retval 0            0 connection succeeded
 * \retval negative     negative errno connection failed
 */
static int osp_connect_to_osd(const struct lu_env *env, struct osp_device *osp,
                              const char *nextdev)
{
        struct obd_connect_data *data = NULL;
        struct obd_device       *obd;
        int                      rc;

        ENTRY;

        LASSERT(osp->opd_storage_exp == NULL);

        OBD_ALLOC_PTR(data);
        if (data == NULL)
                RETURN(-ENOMEM);

        obd = class_name2obd(nextdev);
        if (obd == NULL) {
                CERROR("%s: can't locate next device: %s\n",
                       osp->opd_obd->obd_name, nextdev);
                GOTO(out, rc = -ENOTCONN);
        }

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

        osp->opd_dt_dev.dd_lu_dev.ld_site =
                osp->opd_storage_exp->exp_obd->obd_lu_dev->ld_site;
        LASSERT(osp->opd_dt_dev.dd_lu_dev.ld_site);
        osp->opd_storage = lu2dt_dev(osp->opd_storage_exp->exp_obd->obd_lu_dev);

out:
        OBD_FREE_PTR(data);
        RETURN(rc);
}

/**
 * Determine if the lock needs to be cancelled
 *
 * Determine if the unused lock should be cancelled before replay, see
 * (ldlm_cancel_no_wait_policy()). Currently, only inode bits lock exists
 * between MDTs.
 *
 * \param[in] lock      lock to be checked.
 *
 * \retval              1 if the lock needs to be cancelled before replay.
 * \retval              0 if the lock does not need to be cancelled before
 *                      replay.
 */
static int osp_cancel_weight(struct ldlm_lock *lock)
{
        if (lock->l_resource->lr_type != LDLM_IBITS)
                RETURN(0);

        RETURN(1);
}

/**
 * Initialize OSP device according to the parameters in the configuration
 * log \a cfg.
 *
 * Reconstruct the local device name from the configuration profile, and
 * initialize necessary threads and structures according to the OSP type
 * (MDT or OST).
 *
 * Since there is no record in the MDT configuration for the local disk
 * device, we have to extract this from elsewhere in the profile.
 * The only information we get at setup is from the OSC records:
 * setup 0:{fsname}-OSTxxxx-osc[-MDTxxxx] 1:lustre-OST0000_UUID 2:NID
 *
 * Note: configs generated by Lustre 1.8 are missing the -MDTxxxx part,
 * so, we need to reconstruct the name of the underlying OSD from this:
 * {fsname}-{svname}-osd, for example "lustre-MDT0000-osd".
 *
 * \param[in] env       execution environment
 * \param[in] osp       OSP device
 * \param[in] ldt       lu device type of OSP
 * \param[in] cfg       configuration log
 *
 * \retval 0            0 if OSP initialization succeeded.
 * \retval negative     negative errno if OSP initialization failed.
 */
static int osp_init0(const struct lu_env *env, struct osp_device *osp,
                     struct lu_device_type *ldt, struct lustre_cfg *cfg)
{
        struct obd_device       *obd;
        struct obd_import       *imp;
        class_uuid_t            uuid;
        char                    *src, *tgt, *mdt, *osdname = NULL;
        int                     rc;
        long                    idx;

        ENTRY;

        mutex_init(&osp->opd_async_requests_mutex);
        INIT_LIST_HEAD(&osp->opd_async_updates);
        init_rwsem(&osp->opd_async_updates_rwsem);
        atomic_set(&osp->opd_async_updates_count, 0);

        obd = class_name2obd(lustre_cfg_string(cfg, 0));
        if (obd == NULL) {
                CERROR("Cannot find obd with name %s\n",
                       lustre_cfg_string(cfg, 0));
                RETURN(-ENODEV);
        }
        osp->opd_obd = obd;

        src = lustre_cfg_string(cfg, 0);
        if (src == NULL)
                RETURN(-EINVAL);

        tgt = strrchr(src, '-');
        if (tgt == NULL) {
                CERROR("%s: invalid target name %s: rc = %d\n",
                       osp->opd_obd->obd_name, lustre_cfg_string(cfg, 0),
                       -EINVAL);
                RETURN(-EINVAL);
        }

        if (strncmp(tgt, "-osc", 4) == 0) {
                /* Old OSC name fsname-OSTXXXX-osc */
                for (tgt--; tgt > src && *tgt != '-'; tgt--)
                        ;
                if (tgt == src) {
                        CERROR("%s: invalid target name %s: rc = %d\n",
                               osp->opd_obd->obd_name,
                               lustre_cfg_string(cfg, 0), -EINVAL);
                        RETURN(-EINVAL);
                }

                if (strncmp(tgt, "-OST", 4) != 0) {
                        CERROR("%s: invalid target name %s: rc = %d\n",
                               osp->opd_obd->obd_name,
                               lustre_cfg_string(cfg, 0), -EINVAL);
                        RETURN(-EINVAL);
                }

                idx = simple_strtol(tgt + 4, &mdt, 16);
                if (mdt[0] != '-' || idx > INT_MAX || idx < 0) {
                        CERROR("%s: invalid OST index in '%s': rc = %d\n",
                               osp->opd_obd->obd_name, src, -EINVAL);
                        RETURN(-EINVAL);
                }
                osp->opd_index = idx;
                osp->opd_group = 0;
                idx = tgt - src;
        } else {
                /* New OSC name fsname-OSTXXXX-osc-MDTXXXX */
                if (strncmp(tgt, "-MDT", 4) != 0 &&
                    strncmp(tgt, "-OST", 4) != 0) {
                        CERROR("%s: invalid target name %s: rc = %d\n",
                               osp->opd_obd->obd_name,
                               lustre_cfg_string(cfg, 0), -EINVAL);
                        RETURN(-EINVAL);
                }

                idx = simple_strtol(tgt + 4, &mdt, 16);
                if (*mdt != '\0' || idx > INT_MAX || idx < 0) {
                        CERROR("%s: invalid OST index in '%s': rc = %d\n",
                               osp->opd_obd->obd_name, src, -EINVAL);
                        RETURN(-EINVAL);
                }

                /* Get MDT index from the name and set it to opd_group,
                 * which will be used by OSP to connect with OST */
                osp->opd_group = idx;
                if (tgt - src <= 12) {
                        CERROR("%s: invalid mdt index from %s: rc =%d\n",
                               osp->opd_obd->obd_name,
                               lustre_cfg_string(cfg, 0), -EINVAL);
                        RETURN(-EINVAL);
                }

                if (strncmp(tgt - 12, "-MDT", 4) == 0)
                        osp->opd_connect_mdt = 1;

                idx = simple_strtol(tgt - 8, &mdt, 16);
                if (mdt[0] != '-' || idx > INT_MAX || idx < 0) {
                        CERROR("%s: invalid OST index in '%s': rc =%d\n",
                               osp->opd_obd->obd_name, src, -EINVAL);
                        RETURN(-EINVAL);
                }

                osp->opd_index = idx;
                idx = tgt - src - 12;
        }
        /* check the fsname length, and after this everything else will fit */
        if (idx > MTI_NAME_MAXLEN) {
                CERROR("%s: fsname too long in '%s': rc = %d\n",
                       osp->opd_obd->obd_name, src, -EINVAL);
                RETURN(-EINVAL);
        }

        OBD_ALLOC(osdname, MAX_OBD_NAME);
        if (osdname == NULL)
                RETURN(-ENOMEM);

        memcpy(osdname, src, idx); /* copy just the fsname part */
        osdname[idx] = '\0';

        mdt = strstr(mdt, "-MDT");
        if (mdt == NULL) /* 1.8 configs don't have "-MDT0000" at the end */
                strcat(osdname, "-MDT0000");
        else
                strcat(osdname, mdt);
        strcat(osdname, "-osd");
        CDEBUG(D_HA, "%s: connect to %s (%s)\n", obd->obd_name, osdname, src);

        if (osp->opd_connect_mdt) {
                struct client_obd *cli = &osp->opd_obd->u.cli;

                OBD_ALLOC(cli->cl_rpc_lock, sizeof(*cli->cl_rpc_lock));
                if (!cli->cl_rpc_lock)
                        GOTO(out_fini, rc = -ENOMEM);
                osp_init_rpc_lock(cli->cl_rpc_lock);
        }

        osp->opd_dt_dev.dd_lu_dev.ld_ops = &osp_lu_ops;
        osp->opd_dt_dev.dd_ops = &osp_dt_ops;

        obd->obd_lu_dev = &osp->opd_dt_dev.dd_lu_dev;

        rc = osp_connect_to_osd(env, osp, osdname);
        if (rc)
                GOTO(out_fini, rc);

        rc = ptlrpcd_addref();
        if (rc)
                GOTO(out_disconnect, rc);

        rc = client_obd_setup(obd, cfg);
        if (rc) {
                CERROR("%s: can't setup obd: rc = %d\n", osp->opd_obd->obd_name,
                       rc);
                GOTO(out_ref, rc);
        }

        osp_tunables_init(osp);

        rc = obd_fid_init(osp->opd_obd, NULL, osp->opd_connect_mdt ?
                          LUSTRE_SEQ_METADATA : LUSTRE_SEQ_DATA);
        if (rc) {
                CERROR("%s: fid init error: rc = %d\n",
                       osp->opd_obd->obd_name, rc);
                GOTO(out_proc, rc);
        }

        if (!osp->opd_connect_mdt) {
                /* Initialize last id from the storage - will be
                 * used in orphan cleanup. */
                if (!osp->opd_storage->dd_rdonly) {
                        rc = osp_last_used_init(env, osp);
                        if (rc)
                                GOTO(out_fid, rc);
                }

                /* Initialize precreation thread, it handles new
                 * connections as well. */
                rc = osp_init_precreate(osp);
                if (rc)
                        GOTO(out_last_used, rc);

                /*
                 * Initialize synhronization mechanism taking
                 * care of propogating changes to OST in near
                 * transactional manner.
                 */
                rc = osp_sync_init(env, osp);
                if (rc < 0)
                        GOTO(out_precreat, rc);
        } else {
                rc = osp_update_init(osp);
                if (rc != 0)
                        GOTO(out_fid, rc);
        }

        ns_register_cancel(obd->obd_namespace, osp_cancel_weight);

        /*
         * Initiate connect to OST
         */
        ll_generate_random_uuid(uuid);
        class_uuid_unparse(uuid, &osp->opd_cluuid);

        imp = obd->u.cli.cl_import;

        rc = ptlrpc_init_import(imp);
        if (rc)
                GOTO(out, rc);
        if (osdname)
                OBD_FREE(osdname, MAX_OBD_NAME);
        RETURN(0);

out:
        if (!osp->opd_connect_mdt)
                /* stop sync thread */
                osp_sync_fini(osp);
out_precreat:
        /* stop precreate thread */
        if (!osp->opd_connect_mdt)
                osp_precreate_fini(osp);
        else
                osp_update_fini(env, osp);
out_last_used:
        if (!osp->opd_connect_mdt)
                osp_last_used_fini(env, osp);
out_fid:
        obd_fid_fini(osp->opd_obd);
out_proc:
        osp_tunables_fini(osp);
        client_obd_cleanup(obd);
out_ref:
        ptlrpcd_decref();
out_disconnect:
        if (osp->opd_connect_mdt) {
                struct client_obd *cli = &osp->opd_obd->u.cli;
                if (cli->cl_rpc_lock != NULL) {
                        OBD_FREE_PTR(cli->cl_rpc_lock);
                        cli->cl_rpc_lock = NULL;
                }
        }
        obd_disconnect(osp->opd_storage_exp);
out_fini:
        if (osdname)
                OBD_FREE(osdname, MAX_OBD_NAME);
        RETURN(rc);
}

/**
 * Implementation of lu_device_type_operations::ldto_device_free
 *
 * Free the OSP device in memory.  No return value is needed for now,
 * so always return NULL to comply with the interface.
 *
 * \param[in] env       execution environment
 * \param[in] lu        lu_device of OSP
 *
 * \retval NULL         NULL unconditionally
 */
static struct lu_device *osp_device_free(const struct lu_env *env,
                                         struct lu_device *lu)
{
        struct osp_device *osp = lu2osp_dev(lu);

        if (atomic_read(&lu->ld_ref) && lu->ld_site) {
                LIBCFS_DEBUG_MSG_DATA_DECL(msgdata, D_ERROR, NULL);
                lu_site_print(env, lu->ld_site, &msgdata, lu_cdebug_printer);
        }
        dt_device_fini(&osp->opd_dt_dev);
        OBD_FREE_PTR(osp);

        return NULL;
}

/**
 * Implementation of lu_device_type_operations::ldto_device_alloc
 *
 * This function allocates and initializes OSP device in memory according to
 * the config log.
 *
 * \param[in] env       execution environment
 * \param[in] type      device type of OSP
 * \param[in] lcfg      config log
 *
 * \retval pointer              the pointer of allocated OSP if succeed.
 * \retval ERR_PTR(errno)       ERR_PTR(errno) if failed.
 */
static struct lu_device *osp_device_alloc(const struct lu_env *env,
                                          struct lu_device_type *type,
                                          struct lustre_cfg *lcfg)
{
        struct osp_device *osp;
        struct lu_device  *ld;

        OBD_ALLOC_PTR(osp);
        if (osp == NULL) {
                ld = ERR_PTR(-ENOMEM);
        } else {
                int rc;

                ld = osp2lu_dev(osp);
                dt_device_init(&osp->opd_dt_dev, type);
                rc = osp_init0(env, osp, type, lcfg);
                if (rc != 0) {
                        osp_device_free(env, ld);
                        ld = ERR_PTR(rc);
                }
        }
        return ld;
}

/**
 * Implementation of lu_device_type_operations::ldto_device_fini
 *
 * This function cleans up the OSP device, i.e. release and free those
 * attached items in osp_device.
 *
 * \param[in] env       execution environment
 * \param[in] ld        lu_device of OSP
 *
 * \retval NULL                 NULL if cleanup succeeded.
 * \retval ERR_PTR(errno)       ERR_PTR(errno) if cleanup failed.
 */
static struct lu_device *osp_device_fini(const struct lu_env *env,
                                         struct lu_device *ld)
{
        struct osp_device *osp = lu2osp_dev(ld);
        int                rc;

        ENTRY;

        if (osp->opd_async_requests != NULL) {
                osp_update_request_destroy(env, osp->opd_async_requests);
                osp->opd_async_requests = NULL;
        }

        if (osp->opd_storage_exp) {
                /* wait for the commit callbacks to complete */
                wait_event(osp->opd_sync_waitq,
                          atomic_read(&osp->opd_commits_registered) == 0);
                obd_disconnect(osp->opd_storage_exp);
        }

        LASSERT(osp->opd_obd);
        osp_tunables_fini(osp);

        if (osp->opd_connect_mdt) {
                struct client_obd *cli = &osp->opd_obd->u.cli;
                if (cli->cl_rpc_lock != NULL) {
                        OBD_FREE_PTR(cli->cl_rpc_lock);
                        cli->cl_rpc_lock = NULL;
                }
        }

        rc = client_obd_cleanup(osp->opd_obd);
        if (rc != 0) {
                ptlrpcd_decref();
                RETURN(ERR_PTR(rc));
        }

        ptlrpcd_decref();

        RETURN(NULL);
}

/**
 * Implementation of obd_ops::o_reconnect
 *
 * This function is empty and does not need to do anything for now.
 */
static int osp_reconnect(const struct lu_env *env,
                         struct obd_export *exp, struct obd_device *obd,
                         struct obd_uuid *cluuid,
                         struct obd_connect_data *data,
                         void *localdata)
{
        return 0;
}

/*
 * Implementation of obd_ops::o_connect
 *
 * Connect OSP to the remote target (MDT or OST). Allocate the
 * export and return it to the LOD, which calls this function
 * for each OSP to connect it to the remote target. This function
 * is currently only called once per OSP.
 *
 * \param[in] env       execution environment
 * \param[out] exp      export connected to OSP
 * \param[in] obd       OSP device
 * \param[in] cluuid    OSP device client uuid
 * \param[in] data      connect_data to be used to connect to the remote
 *                      target
 * \param[in] localdata necessary for the API interface, but not used in
 *                      this function
 *
 * \retval 0            0 if the connection succeeded.
 * \retval negative     negative errno if the connection failed.
 */
static int osp_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 osp_device       *osp = lu2osp_dev(obd->obd_lu_dev);
        struct obd_connect_data *ocd;
        struct obd_import       *imp;
        struct lustre_handle     conn;
        int                      rc;

        ENTRY;

        CDEBUG(D_CONFIG, "connect #%d\n", osp->opd_connects);

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

        *exp = class_conn2export(&conn);
        /* Why should there ever be more than 1 connect? */
        osp->opd_connects++;
        LASSERT(osp->opd_connects == 1);

        osp->opd_exp = *exp;

        imp = osp->opd_obd->u.cli.cl_import;
        imp->imp_dlm_handle = conn;

        LASSERT(data != NULL);
        LASSERT(data->ocd_connect_flags & OBD_CONNECT_INDEX);
        ocd = &imp->imp_connect_data;
        *ocd = *data;

        imp->imp_connect_flags_orig = ocd->ocd_connect_flags;
        imp->imp_connect_flags2_orig = ocd->ocd_connect_flags2;

        ocd->ocd_version = LUSTRE_VERSION_CODE;
        ocd->ocd_index = data->ocd_index;

        rc = ptlrpc_connect_import(imp);
        if (rc) {
                CERROR("%s: can't connect obd: rc = %d\n", obd->obd_name, rc);
                GOTO(out, rc);
        } else {
                osp->opd_obd->u.cli.cl_seq->lcs_exp =
                                class_export_get(osp->opd_exp);
        }

        ptlrpc_pinger_add_import(imp);
out:
        RETURN(rc);
}

/**
 * Implementation of obd_ops::o_disconnect
 *
 * Disconnect the export for the OSP.  This is called by LOD to release the
 * OSP during cleanup (\see lod_del_device()). The OSP will be released after
 * the export is released.
 *
 * \param[in] exp       export to be disconnected.
 *
 * \retval 0            0 if disconnection succeed
 * \retval negative     negative errno if disconnection failed
 */
static int osp_obd_disconnect(struct obd_export *exp)
{
        struct obd_device *obd = exp->exp_obd;
        struct osp_device *osp = lu2osp_dev(obd->obd_lu_dev);
        int                rc;
        ENTRY;

        /* Only disconnect the underlying layers on the final disconnect. */
        LASSERT(osp->opd_connects == 1);
        osp->opd_connects--;

        rc = class_disconnect(exp);
        if (rc) {
                CERROR("%s: class disconnect error: rc = %d\n",
                       obd->obd_name, rc);
                RETURN(rc);
        }

        /* destroy the device */
        class_manual_cleanup(obd);

        RETURN(rc);
}

/**
 * Implementation of obd_ops::o_statfs
 *
 * Send a RPC to the remote target to get statfs status. This is only used
 * in lprocfs helpers by obd_statfs.
 *
 * \param[in] env       execution environment
 * \param[in] exp       connection state from this OSP to the parent (LOD)
 *                      device
 * \param[out] osfs     hold the statfs result
 * \param[in] unused    Not used in this function for now
 * \param[in] flags     flags to indicate how OSP will issue the RPC
 *
 * \retval 0            0 if statfs succeeded.
 * \retval negative     negative errno if statfs failed.
 */
static int osp_obd_statfs(const struct lu_env *env, struct obd_export *exp,
                          struct obd_statfs *osfs, time64_t unused, __u32 flags)
{
        struct obd_statfs       *msfs;
        struct ptlrpc_request   *req;
        struct obd_import       *imp = NULL;
        int                      rc;

        ENTRY;

        /* Since the request might also come from lprocfs, so we need
         * sync this with client_disconnect_export Bug15684 */
        down_read(&exp->exp_obd->u.cli.cl_sem);
        if (exp->exp_obd->u.cli.cl_import)
                imp = class_import_get(exp->exp_obd->u.cli.cl_import);
        up_read(&exp->exp_obd->u.cli.cl_sem);
        if (!imp)
                RETURN(-ENODEV);

        req = ptlrpc_request_alloc(imp, &RQF_OST_STATFS);

        class_import_put(imp);

        if (req == NULL)
                RETURN(-ENOMEM);

        rc = ptlrpc_request_pack(req, LUSTRE_OST_VERSION, OST_STATFS);
        if (rc) {
                ptlrpc_request_free(req);
                RETURN(rc);
        }
        ptlrpc_request_set_replen(req);
        req->rq_request_portal = OST_CREATE_PORTAL;
        ptlrpc_at_set_req_timeout(req);

        if (flags & OBD_STATFS_NODELAY) {
                /* procfs requests not want stat in wait for avoid deadlock */
                req->rq_no_resend = 1;
                req->rq_no_delay = 1;
        }

        rc = ptlrpc_queue_wait(req);
        if (rc)
                GOTO(out, rc);

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

        *osfs = *msfs;

        EXIT;
out:
        ptlrpc_req_finished(req);
        return rc;
}

/**
 * Implementation of obd_ops::o_import_event
 *
 * This function is called when some related import event happens. It will
 * mark the necessary flags according to the event and notify the necessary
 * threads (mainly precreate thread).
 *
 * \param[in] obd       OSP OBD device
 * \param[in] imp       import attached from OSP to remote (OST/MDT) service
 * \param[in] event     event related to remote service (IMP_EVENT_*)
 *
 * \retval 0            0 if the event handling succeeded.
 * \retval negative     negative errno if the event handling failed.
 */
static int osp_import_event(struct obd_device *obd, struct obd_import *imp,
                            enum obd_import_event event)
{
        struct osp_device *d = lu2osp_dev(obd->obd_lu_dev);
        int rc;

        switch (event) {
        case IMP_EVENT_DISCON:
                d->opd_got_disconnected = 1;
                d->opd_imp_connected = 0;
                if (d->opd_connect_mdt)
                        break;

                if (d->opd_pre != NULL) {
                        osp_pre_update_status(d, -ENODEV);
                        wake_up(&d->opd_pre_waitq);
                }

                CDEBUG(D_HA, "got disconnected\n");
                break;
        case IMP_EVENT_INACTIVE:
                d->opd_imp_active = 0;
                d->opd_imp_connected = 0;
                d->opd_obd->obd_inactive = 1;
                if (d->opd_connect_mdt)
                        break;
                if (d->opd_pre != NULL) {
                        /* Import is invalid, we can`t get stripes so
                         * wakeup waiters */
                        rc = imp->imp_deactive ? -ESHUTDOWN : -ENODEV;
                        osp_pre_update_status(d, rc);
                        wake_up(&d->opd_pre_waitq);
                }

                CDEBUG(D_HA, "got inactive\n");
                break;
        case IMP_EVENT_ACTIVE:
                d->opd_imp_active = 1;

                if (d->opd_got_disconnected)
                        d->opd_new_connection = 1;
                d->opd_imp_connected = 1;
                d->opd_imp_seen_connected = 1;
                d->opd_obd->obd_inactive = 0;
                if (d->opd_connect_mdt)
                        break;

                if (d->opd_pre != NULL)
                        wake_up(&d->opd_pre_waitq);

                osp_sync_check_for_work(d);
                CDEBUG(D_HA, "got connected\n");
                break;
        case IMP_EVENT_INVALIDATE:
                if (d->opd_connect_mdt)
                        osp_invalidate_request(d);

                if (obd->obd_namespace == NULL)
                        break;
                ldlm_namespace_cleanup(obd->obd_namespace, LDLM_FL_LOCAL_ONLY);
                break;
        case IMP_EVENT_OCD:
        case IMP_EVENT_DEACTIVATE:
        case IMP_EVENT_ACTIVATE:
                break;
        default:
                CERROR("%s: unsupported import event: %#x\n",
                       obd->obd_name, event);
        }
        return 0;
}

/**
 * Implementation of obd_ops: o_iocontrol
 *
 * This function is the ioctl handler for OSP. Note: lctl will access the OSP
 * directly by ioctl, instead of through the MDS stack.
 *
 * param[in] cmd        ioctl command.
 * param[in] exp        export of this OSP.
 * param[in] len        data length of \a karg.
 * param[in] karg       input argument which is packed as
 *                      obd_ioctl_data
 * param[out] uarg      pointer to userspace buffer (must access by
 *                      copy_to_user()).
 *
 * \retval 0            0 if the ioctl handling succeeded.
 * \retval negative     negative errno if the ioctl handling failed.
 */
static int osp_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
                         void *karg, void __user *uarg)
{
        struct obd_device       *obd = exp->exp_obd;
        struct osp_device       *d;
        struct obd_ioctl_data   *data = karg;
        int                      rc = 0;

        ENTRY;

        LASSERT(obd->obd_lu_dev);
        d = lu2osp_dev(obd->obd_lu_dev);
        LASSERT(d->opd_dt_dev.dd_ops == &osp_dt_ops);

        if (!try_module_get(THIS_MODULE)) {
                CERROR("%s: cannot get module '%s'\n", obd->obd_name,
                       module_name(THIS_MODULE));
                return -EINVAL;
        }

        switch (cmd) {
        case OBD_IOC_CLIENT_RECOVER:
                rc = ptlrpc_recover_import(obd->u.cli.cl_import,
                                           data->ioc_inlbuf1, 0);
                if (rc > 0)
                        rc = 0;
                break;
        case IOC_OSC_SET_ACTIVE:
                rc = ptlrpc_set_import_active(obd->u.cli.cl_import,
                                              data->ioc_offset);
                break;
        case OBD_IOC_PING_TARGET:
                rc = ptlrpc_obd_ping(obd);
                break;
        default:
                CERROR("%s: unrecognized ioctl %#x by %s\n", obd->obd_name,
                       cmd, current_comm());
                rc = -ENOTTY;
        }
        module_put(THIS_MODULE);
        return rc;
}


/**
 * Implementation of obd_ops::o_get_info
 *
 * Retrieve information by key. Retrieval starts from the top layer
 * (MDT) of the MDS stack and traverses the stack by calling the
 * obd_get_info() method of the next sub-layer.
 *
 * \param[in] env       execution environment
 * \param[in] exp       export of this OSP
 * \param[in] keylen    length of \a key
 * \param[in] key       the key
 * \param[out] vallen   length of \a val
 * \param[out] val      holds the value returned by the key
 *
 * \retval 0            0 if getting information succeeded.
 * \retval negative     negative errno if getting information failed.
 */
static int osp_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 osp_device       *osp;

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

                osp = lu2osp_dev(obd->obd_lu_dev);
                LASSERT(osp);
                /*
                 * 1.8/2.0 behaviour is that OST being connected once at least
                 * is considered "healthy". and one "healthy" OST is enough to
                 * allow lustre clients to connect to MDS
                 */
                RETURN(!osp->opd_imp_seen_connected);
        }

        RETURN(rc);
}

static int osp_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 = exp->exp_obd;
        struct obd_import       *imp = obd->u.cli.cl_import;
        struct osp_device       *osp;
        struct ptlrpc_request   *req;
        char                    *tmp;
        int                      rc;

        if (KEY_IS(KEY_SPTLRPC_CONF)) {
                sptlrpc_conf_client_adapt(exp->exp_obd);
                RETURN(0);
        }

        LASSERT(set != NULL);
        if (!obd->obd_set_up || obd->obd_stopping)
                RETURN(-EAGAIN);
        osp = lu2osp_dev(obd->obd_lu_dev);

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

        req_capsule_set_size(&req->rq_pill, &RMF_SETINFO_KEY,
                             RCL_CLIENT, keylen);
        req_capsule_set_size(&req->rq_pill, &RMF_SETINFO_VAL,
                             RCL_CLIENT, vallen);
        if (osp->opd_connect_mdt)
                rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_SET_INFO);
        else
                rc = ptlrpc_request_pack(req, LUSTRE_OST_VERSION, OST_SET_INFO);
        if (rc) {
                ptlrpc_request_free(req);
                RETURN(rc);
        }

        tmp = req_capsule_client_get(&req->rq_pill, &RMF_SETINFO_KEY);
        memcpy(tmp, key, keylen);
        tmp = req_capsule_client_get(&req->rq_pill, &RMF_SETINFO_VAL);
        memcpy(tmp, val, vallen);

        ptlrpc_request_set_replen(req);
        ptlrpc_set_add_req(set, req);
        ptlrpc_check_set(NULL, set);

        RETURN(0);
}

/**
 * Implementation of obd_ops: o_fid_alloc
 *
 * Allocate a FID. There are two cases in which OSP performs
 * FID allocation.
 *
 * 1. FID precreation for data objects, which is done in
 *    osp_precreate_fids() instead of this function.
 * 2. FID allocation for each sub-stripe of a striped directory.
 *    Similar to other FID clients, OSP requests the sequence
 *    from its corresponding remote MDT, which in turn requests
 *    sequences from the sequence controller (MDT0).
 *
 * \param[in] env       execution environment
 * \param[in] exp       export of the OSP
 * \param[out] fid      FID being allocated
 * \param[in] unused    necessary for the interface but unused.
 *
 * \retval 0            0 FID allocated successfully.
 * \retval 1            1 FID allocated successfully and new sequence
 *                      requested from seq meta server
 * \retval negative     negative errno if FID allocation failed.
 */
static int osp_fid_alloc(const struct lu_env *env, struct obd_export *exp,
                         struct lu_fid *fid, struct md_op_data *unused)
{
        struct client_obd       *cli = &exp->exp_obd->u.cli;
        struct osp_device       *osp = lu2osp_dev(exp->exp_obd->obd_lu_dev);
        struct lu_client_seq    *seq = cli->cl_seq;
        ENTRY;

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

        RETURN(seq_client_alloc_fid(env, seq, fid));
}

/* context key constructor/destructor: mdt_key_init, mdt_key_fini */
LU_KEY_INIT_FINI(osp, struct osp_thread_info);
static void osp_key_exit(const struct lu_context *ctx,
                         struct lu_context_key *key, void *data)
{
        struct osp_thread_info *info = data;

        info->osi_attr.la_valid = 0;
}

struct lu_context_key osp_thread_key = {
        .lct_tags = LCT_MD_THREAD,
        .lct_init = osp_key_init,
        .lct_fini = osp_key_fini,
        .lct_exit = osp_key_exit
};

/* context key constructor/destructor: mdt_txn_key_init, mdt_txn_key_fini */
LU_KEY_INIT_FINI(osp_txn, struct osp_txn_info);

struct lu_context_key osp_txn_key = {
        .lct_tags = LCT_OSP_THREAD,
        .lct_init = osp_txn_key_init,
        .lct_fini = osp_txn_key_fini
};
LU_TYPE_INIT_FINI(osp, &osp_thread_key, &osp_txn_key);

static struct lu_device_type_operations osp_device_type_ops = {
        .ldto_init           = osp_type_init,
        .ldto_fini           = osp_type_fini,

        .ldto_start          = osp_type_start,
        .ldto_stop           = osp_type_stop,

        .ldto_device_alloc   = osp_device_alloc,
        .ldto_device_free    = osp_device_free,

        .ldto_device_fini    = osp_device_fini
};

static struct lu_device_type osp_device_type = {
        .ldt_tags     = LU_DEVICE_DT,
        .ldt_name     = LUSTRE_OSP_NAME,
        .ldt_ops      = &osp_device_type_ops,
        .ldt_ctx_tags = LCT_MD_THREAD | LCT_DT_THREAD,
};

static struct obd_ops osp_obd_device_ops = {
        .o_owner        = THIS_MODULE,
        .o_add_conn     = client_import_add_conn,
        .o_del_conn     = client_import_del_conn,
        .o_reconnect    = osp_reconnect,
        .o_connect      = osp_obd_connect,
        .o_disconnect   = osp_obd_disconnect,
        .o_get_info     = osp_obd_get_info,
        .o_set_info_async = osp_obd_set_info_async,
        .o_import_event = osp_import_event,
        .o_iocontrol    = osp_iocontrol,
        .o_statfs       = osp_obd_statfs,
        .o_fid_init     = client_fid_init,
        .o_fid_fini     = client_fid_fini,
        .o_fid_alloc    = osp_fid_alloc,
};

struct llog_operations osp_mds_ost_orig_logops;

static struct obd_type sym;

/**
 * Initialize OSP module.
 *
 * Register device types OSP and Light Weight Proxy (LWP) (\see lwp_dev.c)
 * in obd_types (\see class_obd.c).  Initialize procfs for the
 * the OSP device.  Note: OSP was called OSC before Lustre 2.4,
 * so for compatibility it still uses the name "osc" in procfs.
 * This is called at module load time.
 *
 * \retval 0            0 if initialization succeeds.
 * \retval negative     negative errno if initialization failed.
 */
static int __init osp_init(void)
{
        struct dentry *symlink;
        struct obd_type *type;
        struct kobject *kobj;
        struct qstr dname;
        int rc;

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

        rc = class_register_type(&osp_obd_device_ops, NULL, true, NULL,
                                 LUSTRE_OSP_NAME, &osp_device_type);
        if (rc != 0) {
                lu_kmem_fini(osp_caches);
                return rc;
        }

        rc = class_register_type(&lwp_obd_device_ops, NULL, false, NULL,
                                 LUSTRE_LWP_NAME, &lwp_device_type);
        if (rc != 0) {
                class_unregister_type(LUSTRE_OSP_NAME);
                lu_kmem_fini(osp_caches);
                return rc;
        }

        /* Note: add_rec/delcare_add_rec will be only used by catalogs */
        osp_mds_ost_orig_logops = llog_osd_ops;
        osp_mds_ost_orig_logops.lop_add = llog_cat_add_rec;
        osp_mds_ost_orig_logops.lop_declare_add = llog_cat_declare_add_rec;

        /* create "osc" entry for compatibility purposes */
        dname.name = "osc";
        dname.len = strlen(dname.name);
        dname.hash = ll_full_name_hash(debugfs_lustre_root, dname.name,
                                       dname.len);
        symlink = d_lookup(debugfs_lustre_root, &dname);
        if (!symlink) {
                symlink = debugfs_create_dir(dname.name, debugfs_lustre_root);
                if (IS_ERR_OR_NULL(symlink)) {
                        rc = symlink ? PTR_ERR(symlink) : -ENOMEM;
                        GOTO(no_osc, rc);
                }
                sym.typ_debugfs_entry = symlink;
        } else {
                dput(symlink);
        }

        kobj = kset_find_obj(lustre_kset, dname.name);
        if (kobj) {
                kobject_put(kobj);
                goto try_proc;
        }

        kobj = class_setup_tunables(dname.name);
        if (IS_ERR(kobj)) {
                rc = PTR_ERR(kobj);
                if (sym.typ_debugfs_entry)
                        ldebugfs_remove(&sym.typ_debugfs_entry);
                GOTO(no_osc, rc);
        }
        sym.typ_kobj = kobj;

try_proc:
        type = class_search_type(LUSTRE_OSC_NAME);
        if (type != NULL && type->typ_procroot != NULL)
                GOTO(no_osc, rc);

        type = class_search_type(LUSTRE_OSP_NAME);
        type->typ_procsym = lprocfs_register("osc", proc_lustre_root,
                                             NULL, NULL);
        if (IS_ERR(type->typ_procsym)) {
                CERROR("osp: can't create compat entry \"osc\": %d\n",
                       (int) PTR_ERR(type->typ_procsym));
                type->typ_procsym = NULL;
        }
no_osc:
        return rc;
}

/**
 * Finalize OSP module.
 *
 * This callback is called when kernel unloads OSP module from memory, and
 * it will deregister OSP and LWP device type from obd_types (\see class_obd.c).
 */
static void __exit osp_exit(void)
{
        ldebugfs_remove(&sym.typ_debugfs_entry);
        kobject_put(sym.typ_kobj);
        class_unregister_type(LUSTRE_LWP_NAME);
        class_unregister_type(LUSTRE_OSP_NAME);
        lu_kmem_fini(osp_caches);
}

MODULE_AUTHOR("OpenSFS, Inc. <http://www.lustre.org/>");
MODULE_DESCRIPTION("Lustre OSD Storage Proxy ("LUSTRE_OSP_NAME")");
MODULE_VERSION(LUSTRE_VERSION_STRING);
MODULE_LICENSE("GPL");

module_init(osp_init);
module_exit(osp_exit);