X-Git-Url: https://git.whamcloud.com/?p=fs%2Flustre-release.git;a=blobdiff_plain;f=lustre%2Fmdd%2Fmdd_device.c;h=1d049eac833ba95f2f9ef3c881eca26fc794e3d6;hp=75abe60694cf5f70320e0112bf42db8900cf590c;hb=df1d59429cbfd1ea2464e863458b6a4a268e516b;hpb=845d4b760ccdc685b6bc9746985532429e717224

diff --git a/lustre/mdd/mdd_device.c b/lustre/mdd/mdd_device.c
index 75abe60..1d049ea 100644
--- a/lustre/mdd/mdd_device.c
+++ b/lustre/mdd/mdd_device.c
@@ -27,7 +27,7 @@
  * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
  * Use is subject to license terms.
  *
- * Copyright (c) 2011, 2012, Whamcloud, Inc.
+ * Copyright (c) 2011, 2013, Intel Corporation.
  */
 /*
  * This file is part of Lustre, http://www.lustre.org/
@@ -43,27 +43,10 @@
 #define DEBUG_SUBSYSTEM S_MDS
 
 #include <linux/module.h>
-#ifdef HAVE_EXT4_LDISKFS
-#include <ldiskfs/ldiskfs_jbd2.h>
-#else
-#include <linux/jbd.h>
-#endif
-#include <obd.h>
 #include <obd_class.h>
-#include <lustre_ver.h>
-#include <obd_support.h>
-#include <lprocfs_status.h>
-
-#include <lustre_disk.h>
-#include <lustre_fid.h>
-#ifdef HAVE_EXT4_LDISKFS
-#include <ldiskfs/ldiskfs.h>
-#else
-#include <linux/ldiskfs_fs.h>
-#endif
 #include <lustre_mds.h>
-#include <lustre/lustre_idl.h>
-#include <lustre_disk.h>      /* for changelogs */
+#include <obd_support.h>
+#include <lu_object.h>
 #include <lustre_param.h>
 #include <lustre_fid.h>
 
@@ -75,27 +58,88 @@ static struct lu_device_type mdd_device_type;
 static const char mdd_root_dir_name[] = "ROOT";
 static const char mdd_obf_dir_name[] = "fid";
 
-static int mdd_device_init(const struct lu_env *env, struct lu_device *d,
-                           const char *name, struct lu_device *next)
+/* Slab for MDD object allocation */
+struct kmem_cache *mdd_object_kmem;
+
+static struct lu_kmem_descr mdd_caches[] = {
+	{
+		.ckd_cache = &mdd_object_kmem,
+		.ckd_name  = "mdd_obj",
+		.ckd_size  = sizeof(struct mdd_object)
+	},
+	{
+		.ckd_cache = NULL
+	}
+};
+
+static int mdd_connect_to_next(const struct lu_env *env, struct mdd_device *m,
+			       const char *nextdev)
 {
-        struct mdd_device *mdd = lu2mdd_dev(d);
-        int rc;
-        ENTRY;
+	struct obd_connect_data *data = NULL;
+	struct lu_device	*lud = mdd2lu_dev(m);
+	struct obd_device       *obd;
+	int			 rc;
+	ENTRY;
+
+	LASSERT(m->mdd_child_exp == NULL);
+
+	OBD_ALLOC(data, sizeof(*data));
+	if (data == NULL)
+		GOTO(out, rc = -ENOMEM);
+
+	obd = class_name2obd(nextdev);
+	if (obd == NULL) {
+		CERROR("can't locate next device: %s\n", nextdev);
+		GOTO(out, rc = -ENOTCONN);
+	}
 
-        mdd->mdd_child = lu2dt_dev(next);
+	data->ocd_connect_flags = OBD_CONNECT_VERSION;
+	data->ocd_version = LUSTRE_VERSION_CODE;
 
-        /* Prepare transactions callbacks. */
-        mdd->mdd_txn_cb.dtc_txn_start = NULL;
-        mdd->mdd_txn_cb.dtc_txn_stop = mdd_txn_stop_cb;
-        mdd->mdd_txn_cb.dtc_txn_commit = NULL;
-        mdd->mdd_txn_cb.dtc_cookie = mdd;
-        mdd->mdd_txn_cb.dtc_tag = LCT_MD_THREAD;
-        CFS_INIT_LIST_HEAD(&mdd->mdd_txn_cb.dtc_linkage);
-        mdd->mdd_atime_diff = MAX_ATIME_DIFF;
+	rc = obd_connect(NULL, &m->mdd_child_exp, obd, &obd->obd_uuid, data, NULL);
+	if (rc) {
+		CERROR("cannot connect to next dev %s (%d)\n", nextdev, rc);
+		GOTO(out, rc);
+	}
+
+	lud->ld_site = m->mdd_child_exp->exp_obd->obd_lu_dev->ld_site;
+	LASSERT(lud->ld_site);
+	m->mdd_child = lu2dt_dev(m->mdd_child_exp->exp_obd->obd_lu_dev);
+	m->mdd_bottom = lu2dt_dev(lud->ld_site->ls_bottom_dev);
+	lu_dev_add_linkage(lud->ld_site, lud);
+
+out:
+	if (data)
+		OBD_FREE(data, sizeof(*data));
+	RETURN(rc);
+}
+
+static int mdd_init0(const struct lu_env *env, struct mdd_device *mdd,
+		struct lu_device_type *t, struct lustre_cfg *lcfg)
+{
+	int rc;
+	ENTRY;
+
+	mdd->mdd_md_dev.md_lu_dev.ld_ops = &mdd_lu_ops;
+	mdd->mdd_md_dev.md_ops = &mdd_ops;
+
+	rc = mdd_connect_to_next(env, mdd, lustre_cfg_string(lcfg, 3));
+	if (rc)
+		RETURN(rc);
+
+	mdd->mdd_atime_diff = MAX_ATIME_DIFF;
         /* sync permission changes */
         mdd->mdd_sync_permission = 1;
 
-        rc = mdd_procfs_init(mdd, name);
+	dt_conf_get(env, mdd->mdd_child, &mdd->mdd_dt_conf);
+
+	/* we are using service name but not mdd obd name
+	 * for compatibility reasons.
+	 * It is passed from MDT in lustre_cfg[2] buffer */
+	rc = mdd_procfs_init(mdd, lustre_cfg_string(lcfg, 2));
+	if (rc < 0)
+		obd_disconnect(mdd->mdd_child_exp);
+
         RETURN(rc);
 }
 
@@ -103,167 +147,268 @@ static struct lu_device *mdd_device_fini(const struct lu_env *env,
                                          struct lu_device *d)
 {
         struct mdd_device *mdd = lu2mdd_dev(d);
-        struct lu_device *next = &mdd->mdd_child->dd_lu_dev;
         int rc;
 
+	if (d->ld_site)
+		lu_dev_del_linkage(d->ld_site, d);
+
         rc = mdd_procfs_fini(mdd);
         if (rc) {
                 CERROR("proc fini error %d \n", rc);
                 return ERR_PTR(rc);
         }
-        return next;
-}
-
-static void mdd_changelog_fini(const struct lu_env *env,
-                               struct mdd_device *mdd);
-
-static void mdd_device_shutdown(const struct lu_env *env,
-                                struct mdd_device *m, struct lustre_cfg *cfg)
-{
-        ENTRY;
-        mdd_changelog_fini(env, m);
-        dt_txn_callback_del(m->mdd_child, &m->mdd_txn_cb);
-        if (m->mdd_dot_lustre_objs.mdd_obf)
-                mdd_object_put(env, m->mdd_dot_lustre_objs.mdd_obf);
-        if (m->mdd_dot_lustre)
-                mdd_object_put(env, m->mdd_dot_lustre);
-        if (m->mdd_obd_dev)
-                mdd_fini_obd(env, m, cfg);
-        orph_index_fini(env, m);
-        if (m->mdd_capa != NULL) {
-                lu_object_put(env, &m->mdd_capa->do_lu);
-                m->mdd_capa = NULL;
-        }
-        /* remove upcall device*/
-        md_upcall_fini(&m->mdd_md_dev);
-        EXIT;
+	return NULL;
 }
 
-static int changelog_init_cb(struct llog_handle *llh, struct llog_rec_hdr *hdr,
-                             void *data)
+static int changelog_init_cb(const struct lu_env *env, struct llog_handle *llh,
+			     struct llog_rec_hdr *hdr, void *data)
 {
-        struct mdd_device *mdd = (struct mdd_device *)data;
-        struct llog_changelog_rec *rec = (struct llog_changelog_rec *)hdr;
-        ENTRY;
+	struct mdd_device *mdd = (struct mdd_device *)data;
+	struct llog_changelog_rec *rec = (struct llog_changelog_rec *)hdr;
 
-        LASSERT(llh->lgh_hdr->llh_flags & LLOG_F_IS_PLAIN);
-        LASSERT(rec->cr_hdr.lrh_type == CHANGELOG_REC);
+	LASSERT(llh->lgh_hdr->llh_flags & LLOG_F_IS_PLAIN);
+	LASSERT(rec->cr_hdr.lrh_type == CHANGELOG_REC);
 
-        CDEBUG(D_INFO,
-               "seeing record at index %d/%d/"LPU64" t=%x %.*s in log "LPX64"\n",
-               hdr->lrh_index, rec->cr_hdr.lrh_index, rec->cr.cr_index,
-               rec->cr.cr_type, rec->cr.cr_namelen, rec->cr.cr_name,
-               llh->lgh_id.lgl_oid);
+	CDEBUG(D_INFO,
+	       "seeing record at index %d/%d/"LPU64" t=%x %.*s in log"
+	       DOSTID"\n", hdr->lrh_index, rec->cr_hdr.lrh_index,
+	       rec->cr.cr_index, rec->cr.cr_type, rec->cr.cr_namelen,
+	       rec->cr.cr_name, POSTID(&llh->lgh_id.lgl_oi));
 
-        mdd->mdd_cl.mc_index = rec->cr.cr_index;
-        RETURN(LLOG_PROC_BREAK);
+	mdd->mdd_cl.mc_index = rec->cr.cr_index;
+	return LLOG_PROC_BREAK;
 }
 
-static int changelog_user_init_cb(struct llog_handle *llh,
-                                  struct llog_rec_hdr *hdr, void *data)
+static int changelog_user_init_cb(const struct lu_env *env,
+				  struct llog_handle *llh,
+				  struct llog_rec_hdr *hdr, void *data)
 {
         struct mdd_device *mdd = (struct mdd_device *)data;
         struct llog_changelog_user_rec *rec =
                 (struct llog_changelog_user_rec *)hdr;
-        ENTRY;
 
         LASSERT(llh->lgh_hdr->llh_flags & LLOG_F_IS_PLAIN);
         LASSERT(rec->cur_hdr.lrh_type == CHANGELOG_USER_REC);
 
         CDEBUG(D_INFO, "seeing user at index %d/%d id=%d endrec="LPU64
-               " in log "LPX64"\n", hdr->lrh_index, rec->cur_hdr.lrh_index,
-               rec->cur_id, rec->cur_endrec, llh->lgh_id.lgl_oid);
+               " in log "DOSTID"\n", hdr->lrh_index, rec->cur_hdr.lrh_index,
+               rec->cur_id, rec->cur_endrec, POSTID(&llh->lgh_id.lgl_oi));
 
-        cfs_spin_lock(&mdd->mdd_cl.mc_user_lock);
-        mdd->mdd_cl.mc_lastuser = rec->cur_id;
-        cfs_spin_unlock(&mdd->mdd_cl.mc_user_lock);
+	spin_lock(&mdd->mdd_cl.mc_user_lock);
+	mdd->mdd_cl.mc_lastuser = rec->cur_id;
+	if (rec->cur_endrec > mdd->mdd_cl.mc_index)
+		mdd->mdd_cl.mc_index = rec->cur_endrec;
+	spin_unlock(&mdd->mdd_cl.mc_user_lock);
 
-        RETURN(LLOG_PROC_BREAK);
+	return LLOG_PROC_BREAK;
 }
 
-
-static int mdd_changelog_llog_init(struct mdd_device *mdd)
+static int llog_changelog_cancel_cb(const struct lu_env *env,
+				    struct llog_handle *llh,
+				    struct llog_rec_hdr *hdr, void *data)
 {
-        struct obd_device *obd = mdd2obd_dev(mdd);
-        struct llog_ctxt *ctxt;
-        int rc;
+	struct llog_changelog_rec *rec = (struct llog_changelog_rec *)hdr;
+	struct llog_cookie	 cookie;
+	long long		 endrec = *(long long *)data;
+	int			 rc;
 
-        /* Find last changelog entry number */
-        ctxt = llog_get_context(obd, LLOG_CHANGELOG_ORIG_CTXT);
-        if (ctxt == NULL) {
-                CERROR("no changelog context\n");
-                return -EINVAL;
-        }
-        if (!ctxt->loc_handle) {
-                llog_ctxt_put(ctxt);
-                return -EINVAL;
-        }
+	ENTRY;
 
-        rc = llog_cat_reverse_process(ctxt->loc_handle, changelog_init_cb, mdd);
-        llog_ctxt_put(ctxt);
+	/* This is always a (sub)log, not the catalog */
+	LASSERT(llh->lgh_hdr->llh_flags & LLOG_F_IS_PLAIN);
 
-        if (rc < 0) {
-                CERROR("changelog init failed: %d\n", rc);
-                return rc;
-        }
-        CDEBUG(D_IOCTL, "changelog starting index="LPU64"\n",
-               mdd->mdd_cl.mc_index);
-
-        /* Find last changelog user id */
-        ctxt = llog_get_context(obd, LLOG_CHANGELOG_USER_ORIG_CTXT);
-        if (ctxt == NULL) {
-                CERROR("no changelog user context\n");
-                return -EINVAL;
-        }
-        if (!ctxt->loc_handle) {
-                llog_ctxt_put(ctxt);
-                return -EINVAL;
-        }
+	if (rec->cr.cr_index > endrec)
+		/* records are in order, so we're done */
+		RETURN(LLOG_PROC_BREAK);
 
-        rc = llog_cat_reverse_process(ctxt->loc_handle, changelog_user_init_cb,
-                                      mdd);
-        llog_ctxt_put(ctxt);
+	cookie.lgc_lgl = llh->lgh_id;
+	cookie.lgc_index = hdr->lrh_index;
 
-        if (rc < 0) {
-                CERROR("changelog user init failed: %d\n", rc);
-                return rc;
-        }
+	/* cancel them one at a time.  I suppose we could store up the cookies
+	 * and cancel them all at once; probably more efficient, but this is
+	 * done as a user call, so who cares... */
+	rc = llog_cat_cancel_records(env, llh->u.phd.phd_cat_handle, 1,
+				     &cookie);
+	RETURN(rc < 0 ? rc : 0);
+}
 
-        /* If we have registered users, assume we want changelogs on */
-        if (mdd->mdd_cl.mc_lastuser > 0)
-                rc = mdd_changelog_on(mdd, 1);
+static int llog_changelog_cancel(const struct lu_env *env,
+				 struct llog_ctxt *ctxt,
+				 struct lov_stripe_md *lsm, int count,
+				 struct llog_cookie *cookies, int flags)
+{
+	struct llog_handle	*cathandle = ctxt->loc_handle;
+	int			 rc;
 
-        return rc;
+	ENTRY;
+
+	/* This should only be called with the catalog handle */
+	LASSERT(cathandle->lgh_hdr->llh_flags & LLOG_F_IS_CAT);
+
+	rc = llog_cat_process(env, cathandle, llog_changelog_cancel_cb,
+			      (void *)cookies, 0, 0);
+	if (rc >= 0)
+		/* 0 or 1 means we're done */
+		rc = 0;
+	else
+		CERROR("%s: cancel idx %u of catalog "DOSTID" rc=%d\n",
+		       ctxt->loc_obd->obd_name, cathandle->lgh_last_idx,
+		       POSTID(&cathandle->lgh_id.lgl_oi), rc);
+
+	RETURN(rc);
+}
+
+static struct llog_operations changelog_orig_logops;
+
+int mdd_changelog_on(const struct lu_env *env, struct mdd_device *mdd, int on);
+
+static int mdd_changelog_llog_init(const struct lu_env *env,
+				   struct mdd_device *mdd)
+{
+	struct obd_device	*obd = mdd2obd_dev(mdd);
+	struct llog_ctxt	*ctxt = NULL, *uctxt = NULL;
+	int			 rc;
+
+	ENTRY;
+
+	/* LU-2844 mdd setup failure should not cause umount oops */
+	if (OBD_FAIL_CHECK(OBD_FAIL_MDS_CHANGELOG_INIT))
+		RETURN(-EIO);
+
+	OBD_SET_CTXT_MAGIC(&obd->obd_lvfs_ctxt);
+	obd->obd_lvfs_ctxt.dt = mdd->mdd_bottom;
+	rc = llog_setup(env, obd, &obd->obd_olg, LLOG_CHANGELOG_ORIG_CTXT,
+			obd, &changelog_orig_logops);
+	if (rc) {
+		CERROR("%s: changelog llog setup failed: rc = %d\n",
+		       obd->obd_name, rc);
+		RETURN(rc);
+	}
+
+	ctxt = llog_get_context(obd, LLOG_CHANGELOG_ORIG_CTXT);
+	LASSERT(ctxt);
+
+	rc = llog_open_create(env, ctxt, &ctxt->loc_handle, NULL,
+			      CHANGELOG_CATALOG);
+	if (rc)
+		GOTO(out_cleanup, rc);
+
+	rc = llog_cat_init_and_process(env, ctxt->loc_handle);
+	if (rc)
+		GOTO(out_close, rc);
+
+	rc = llog_cat_reverse_process(env, ctxt->loc_handle,
+				      changelog_init_cb, mdd);
+
+	if (rc < 0) {
+		CERROR("%s: changelog init failed: rc = %d\n", obd->obd_name,
+		       rc);
+		GOTO(out_close, rc);
+	}
+
+	CDEBUG(D_IOCTL, "changelog starting index="LPU64"\n",
+	       mdd->mdd_cl.mc_index);
+
+	/* setup user changelog */
+	rc = llog_setup(env, obd, &obd->obd_olg, LLOG_CHANGELOG_USER_ORIG_CTXT,
+			obd, &changelog_orig_logops);
+	if (rc) {
+		CERROR("%s: changelog users llog setup failed: rc = %d\n",
+		       obd->obd_name, rc);
+		GOTO(out_close, rc);
+	}
+
+	uctxt = llog_get_context(obd, LLOG_CHANGELOG_USER_ORIG_CTXT);
+	LASSERT(ctxt);
+
+	rc = llog_open_create(env, uctxt, &uctxt->loc_handle, NULL,
+			      CHANGELOG_USERS);
+	if (rc)
+		GOTO(out_ucleanup, rc);
+
+	uctxt->loc_handle->lgh_logops->lop_add = llog_cat_add_rec;
+	uctxt->loc_handle->lgh_logops->lop_declare_add = llog_cat_declare_add_rec;
+
+	rc = llog_cat_init_and_process(env, uctxt->loc_handle);
+	if (rc)
+		GOTO(out_uclose, rc);
+
+	rc = llog_cat_reverse_process(env, uctxt->loc_handle,
+				      changelog_user_init_cb, mdd);
+	if (rc < 0) {
+		CERROR("%s: changelog user init failed: rc = %d\n",
+		       obd->obd_name, rc);
+		GOTO(out_uclose, rc);
+	}
+
+	/* If we have registered users, assume we want changelogs on */
+	if (mdd->mdd_cl.mc_lastuser > 0) {
+		rc = mdd_changelog_on(env, mdd, 1);
+		if (rc < 0)
+			GOTO(out_uclose, rc);
+	}
+	llog_ctxt_put(ctxt);
+	llog_ctxt_put(uctxt);
+	RETURN(0);
+out_uclose:
+	llog_cat_close(env, uctxt->loc_handle);
+out_ucleanup:
+	llog_cleanup(env, uctxt);
+out_close:
+	llog_cat_close(env, ctxt->loc_handle);
+out_cleanup:
+	llog_cleanup(env, ctxt);
+	return rc;
 }
 
 static int mdd_changelog_init(const struct lu_env *env, struct mdd_device *mdd)
 {
-        int rc;
+	struct obd_device	*obd = mdd2obd_dev(mdd);
+	int			 rc;
 
-        mdd->mdd_cl.mc_index = 0;
-        cfs_spin_lock_init(&mdd->mdd_cl.mc_lock);
-        mdd->mdd_cl.mc_starttime = cfs_time_current_64();
-        mdd->mdd_cl.mc_flags = 0; /* off by default */
-        mdd->mdd_cl.mc_mask = CHANGELOG_DEFMASK;
-        cfs_spin_lock_init(&mdd->mdd_cl.mc_user_lock);
-        mdd->mdd_cl.mc_lastuser = 0;
+	mdd->mdd_cl.mc_index = 0;
+	spin_lock_init(&mdd->mdd_cl.mc_lock);
+	mdd->mdd_cl.mc_starttime = cfs_time_current_64();
+	mdd->mdd_cl.mc_flags = 0; /* off by default */
+	mdd->mdd_cl.mc_mask = CHANGELOG_DEFMASK;
+	spin_lock_init(&mdd->mdd_cl.mc_user_lock);
+	mdd->mdd_cl.mc_lastuser = 0;
 
-        rc = mdd_changelog_llog_init(mdd);
-        if (rc) {
-                CERROR("Changelog setup during init failed %d\n", rc);
-                mdd->mdd_cl.mc_flags |= CLM_ERR;
-        }
+	rc = mdd_changelog_llog_init(env, mdd);
+	if (rc) {
+		CERROR("%s: changelog setup during init failed: rc = %d\n",
+		       obd->obd_name, rc);
+		mdd->mdd_cl.mc_flags |= CLM_ERR;
+	}
 
-        return rc;
+	return rc;
 }
 
-static void mdd_changelog_fini(const struct lu_env *env, struct mdd_device *mdd)
+static void mdd_changelog_fini(const struct lu_env *env,
+			       struct mdd_device *mdd)
 {
-        mdd->mdd_cl.mc_flags = 0;
+	struct obd_device	*obd = mdd2obd_dev(mdd);
+	struct llog_ctxt	*ctxt;
+
+	mdd->mdd_cl.mc_flags = 0;
+
+	ctxt = llog_get_context(obd, LLOG_CHANGELOG_ORIG_CTXT);
+	if (ctxt) {
+		llog_cat_close(env, ctxt->loc_handle);
+		llog_cleanup(env, ctxt);
+	}
+	ctxt = llog_get_context(obd, LLOG_CHANGELOG_USER_ORIG_CTXT);
+	if (ctxt) {
+		llog_cat_close(env, ctxt->loc_handle);
+		llog_cleanup(env, ctxt);
+	}
 }
 
+int mdd_changelog_write_header(const struct lu_env *env,
+			       struct mdd_device *mdd, int markerflags);
+
 /* Start / stop recording */
-int mdd_changelog_on(struct mdd_device *mdd, int on)
+int mdd_changelog_on(const struct lu_env *env, struct mdd_device *mdd, int on)
 {
         int rc = 0;
 
@@ -275,62 +420,19 @@ int mdd_changelog_on(struct mdd_device *mdd, int on)
                                mdd2obd_dev(mdd)->obd_name);
                         rc = -ESRCH;
                 } else {
-                        cfs_spin_lock(&mdd->mdd_cl.mc_lock);
-                        mdd->mdd_cl.mc_flags |= CLM_ON;
-                        cfs_spin_unlock(&mdd->mdd_cl.mc_lock);
-                        rc = mdd_changelog_write_header(mdd, CLM_START);
-                }
-        } else if ((on == 0) && ((mdd->mdd_cl.mc_flags & CLM_ON) == CLM_ON)) {
-                LCONSOLE_INFO("%s: changelog off\n",mdd2obd_dev(mdd)->obd_name);
-                rc = mdd_changelog_write_header(mdd, CLM_FINI);
-                cfs_spin_lock(&mdd->mdd_cl.mc_lock);
-                mdd->mdd_cl.mc_flags &= ~CLM_ON;
-                cfs_spin_unlock(&mdd->mdd_cl.mc_lock);
-        }
-        return rc;
-}
-
-static __u64 cl_time(void) {
-        cfs_fs_time_t time;
-
-        cfs_fs_time_current(&time);
-        return (((__u64)time.tv_sec) << 30) + time.tv_nsec;
-}
-
-/** Add a changelog entry \a rec to the changelog llog
- * \param mdd
- * \param rec
- * \param handle - currently ignored since llogs start their own transaction;
- *                 this will hopefully be fixed in llog rewrite
- * \retval 0 ok
- */
-int mdd_changelog_llog_write(struct mdd_device         *mdd,
-                             struct llog_changelog_rec *rec,
-                             struct thandle            *handle)
-{
-        struct obd_device *obd = mdd2obd_dev(mdd);
-        struct llog_ctxt *ctxt;
-        int rc;
-
-        rec->cr_hdr.lrh_len = llog_data_len(sizeof(*rec) + rec->cr.cr_namelen);
-        /* llog_lvfs_write_rec sets the llog tail len */
-        rec->cr_hdr.lrh_type = CHANGELOG_REC;
-        rec->cr.cr_time = cl_time();
-        cfs_spin_lock(&mdd->mdd_cl.mc_lock);
-        /* NB: I suppose it's possible llog_add adds out of order wrt cr_index,
-           but as long as the MDD transactions are ordered correctly for e.g.
-           rename conflicts, I don't think this should matter. */
-        rec->cr.cr_index = ++mdd->mdd_cl.mc_index;
-        cfs_spin_unlock(&mdd->mdd_cl.mc_lock);
-        ctxt = llog_get_context(obd, LLOG_CHANGELOG_ORIG_CTXT);
-        if (ctxt == NULL)
-                return -ENXIO;
-
-        /* nested journal transaction */
-        rc = llog_add(ctxt, &rec->cr_hdr, NULL, NULL, 0);
-        llog_ctxt_put(ctxt);
-
-        return rc;
+			spin_lock(&mdd->mdd_cl.mc_lock);
+			mdd->mdd_cl.mc_flags |= CLM_ON;
+			spin_unlock(&mdd->mdd_cl.mc_lock);
+			rc = mdd_changelog_write_header(env, mdd, CLM_START);
+		}
+	} else if ((on == 0) && ((mdd->mdd_cl.mc_flags & CLM_ON) == CLM_ON)) {
+		LCONSOLE_INFO("%s: changelog off\n",mdd2obd_dev(mdd)->obd_name);
+		rc = mdd_changelog_write_header(env, mdd, CLM_FINI);
+		spin_lock(&mdd->mdd_cl.mc_lock);
+		mdd->mdd_cl.mc_flags &= ~CLM_ON;
+		spin_unlock(&mdd->mdd_cl.mc_lock);
+	}
+	return rc;
 }
 
 /** Remove entries with indicies up to and including \a endrec from the
@@ -339,7 +441,8 @@ int mdd_changelog_llog_write(struct mdd_device         *mdd,
  * \param endrec
  * \retval 0 ok
  */
-int mdd_changelog_llog_cancel(struct mdd_device *mdd, long long endrec)
+int mdd_changelog_llog_cancel(const struct lu_env *env,
+			      struct mdd_device *mdd, long long endrec)
 {
         struct obd_device *obd = mdd2obd_dev(mdd);
         struct llog_ctxt *ctxt;
@@ -350,9 +453,9 @@ int mdd_changelog_llog_cancel(struct mdd_device *mdd, long long endrec)
         if (ctxt == NULL)
                 return -ENXIO;
 
-        cfs_spin_lock(&mdd->mdd_cl.mc_lock);
-        cur = (long long)mdd->mdd_cl.mc_index;
-        cfs_spin_unlock(&mdd->mdd_cl.mc_lock);
+	spin_lock(&mdd->mdd_cl.mc_lock);
+	cur = (long long)mdd->mdd_cl.mc_index;
+	spin_unlock(&mdd->mdd_cl.mc_lock);
         if (endrec > cur)
                 endrec = cur;
 
@@ -366,7 +469,7 @@ int mdd_changelog_llog_cancel(struct mdd_device *mdd, long long endrec)
            allright. */
         if (endrec == cur) {
                 /* XXX: transaction is started by llog itself */
-                rc = mdd_changelog_write_header(mdd, CLM_PURGE);
+                rc = mdd_changelog_write_header(env, mdd, CLM_PURGE);
                 if (rc)
                       goto out;
         }
@@ -376,8 +479,7 @@ int mdd_changelog_llog_cancel(struct mdd_device *mdd, long long endrec)
            changed since the last purge) */
         mdd->mdd_cl.mc_starttime = cfs_time_current_64();
 
-        /* XXX: transaction is started by llog itself */
-        rc = llog_cancel(ctxt, NULL, 1, (struct llog_cookie *)&endrec, 0);
+	rc = llog_cancel(env, ctxt, NULL, 1, (struct llog_cookie *)&endrec, 0);
 out:
         llog_ctxt_put(ctxt);
         return rc;
@@ -388,432 +490,56 @@ out:
  * \param markerflags - CLM_*
  * \retval 0 ok
  */
-int mdd_changelog_write_header(struct mdd_device *mdd, int markerflags)
-{
-        struct obd_device *obd = mdd2obd_dev(mdd);
-        struct llog_changelog_rec *rec;
-        int reclen;
-        int len = strlen(obd->obd_name);
-        int rc;
-        ENTRY;
-
-        reclen = llog_data_len(sizeof(*rec) + len);
-        OBD_ALLOC(rec, reclen);
-        if (rec == NULL)
-                RETURN(-ENOMEM);
+int mdd_changelog_write_header(const struct lu_env *env,
+			       struct mdd_device *mdd, int markerflags)
+{
+	struct obd_device		*obd = mdd2obd_dev(mdd);
+	struct llog_changelog_rec	*rec;
+	struct lu_buf			*buf;
+	struct llog_ctxt		*ctxt;
+	int				 reclen;
+	int				 len = strlen(obd->obd_name);
+	int				 rc;
+
+	ENTRY;
+
+	if (mdd->mdd_cl.mc_mask & (1 << CL_MARK)) {
+		mdd->mdd_cl.mc_starttime = cfs_time_current_64();
+		RETURN(0);
+	}
+
+	reclen = llog_data_len(sizeof(*rec) + len);
+	buf = lu_buf_check_and_alloc(&mdd_env_info(env)->mti_big_buf, reclen);
+	if (buf->lb_buf == NULL)
+		RETURN(-ENOMEM);
+	rec = buf->lb_buf;
 
         rec->cr.cr_flags = CLF_VERSION;
         rec->cr.cr_type = CL_MARK;
         rec->cr.cr_namelen = len;
         memcpy(rec->cr.cr_name, obd->obd_name, rec->cr.cr_namelen);
         /* Status and action flags */
-        rec->cr.cr_markerflags = mdd->mdd_cl.mc_flags | markerflags;
-
-        /* XXX: transaction is started by llog itself */
-        rc = (mdd->mdd_cl.mc_mask & (1 << CL_MARK)) ?
-                mdd_changelog_llog_write(mdd, rec, NULL) : 0;
-
-        /* assume on or off event; reset repeat-access time */
-        mdd->mdd_cl.mc_starttime = cfs_time_current_64();
-
-        OBD_FREE(rec, reclen);
-        RETURN(rc);
-}
-
-/**
- * Create ".lustre" directory.
- */
-static int create_dot_lustre_dir(const struct lu_env *env, struct mdd_device *m)
-{
-        struct lu_fid *fid = &mdd_env_info(env)->mti_fid;
-        struct md_object *mdo;
-        int rc;
-
-        memcpy(fid, &LU_DOT_LUSTRE_FID, sizeof(struct lu_fid));
-        mdo = llo_store_create_index(env, &m->mdd_md_dev, m->mdd_child,
-                                     mdd_root_dir_name, dot_lustre_name,
-                                     fid, &dt_directory_features);
-        /* .lustre dir may be already present */
-        if (IS_ERR(mdo) && PTR_ERR(mdo) != -EEXIST) {
-                rc = PTR_ERR(mdo);
-                CERROR("creating obj [%s] fid = "DFID" rc = %d\n",
-                        dot_lustre_name, PFID(fid), rc);
-                RETURN(rc);
-        }
-
-        if (!IS_ERR(mdo))
-                lu_object_put(env, &mdo->mo_lu);
-
-        return 0;
-}
-
-
-static int dot_lustre_mdd_permission(const struct lu_env *env,
-                                     struct md_object *pobj,
-                                     struct md_object *cobj,
-                                     struct md_attr *attr, int mask)
-{
-        if (mask & ~(MAY_READ | MAY_EXEC))
-                return -EPERM;
-        else
-                return 0;
-}
-
-static int dot_lustre_mdd_attr_get(const struct lu_env *env,
-                                   struct md_object *obj, struct md_attr *ma)
-{
-        struct mdd_object *mdd_obj = md2mdd_obj(obj);
-
-        return mdd_attr_get_internal_locked(env, mdd_obj, ma);
-}
-
-static int dot_lustre_mdd_attr_set(const struct lu_env *env,
-                                   struct md_object *obj,
-                                   const struct md_attr *ma)
-{
-        return -EPERM;
-}
-
-static int dot_lustre_mdd_xattr_get(const struct lu_env *env,
-                                    struct md_object *obj, struct lu_buf *buf,
-                                    const char *name)
-{
-        return 0;
-}
-
-static int dot_lustre_mdd_xattr_list(const struct lu_env *env,
-                                     struct md_object *obj, struct lu_buf *buf)
-{
-        return 0;
-}
-
-static int dot_lustre_mdd_xattr_set(const struct lu_env *env,
-                                    struct md_object *obj,
-                                    const struct lu_buf *buf, const char *name,
-                                    int fl)
-{
-        return -EPERM;
-}
-
-static int dot_lustre_mdd_xattr_del(const struct lu_env *env,
-                                    struct md_object *obj,
-                                    const char *name)
-{
-        return -EPERM;
-}
-
-static int dot_lustre_mdd_readlink(const struct lu_env *env,
-                                   struct md_object *obj, struct lu_buf *buf)
-{
-        return 0;
-}
-
-static int dot_lustre_mdd_object_create(const struct lu_env *env,
-                                        struct md_object *obj,
-                                        const struct md_op_spec *spec,
-                                        struct md_attr *ma)
-{
-        return -EPERM;
-}
-
-static int dot_lustre_mdd_ref_add(const struct lu_env *env,
-                                  struct md_object *obj,
-                                  const struct md_attr *ma)
-{
-        return -EPERM;
-}
-
-static int dot_lustre_mdd_ref_del(const struct lu_env *env,
-                                  struct md_object *obj,
-                                  struct md_attr *ma)
-{
-        return -EPERM;
-}
-
-static int dot_lustre_mdd_open(const struct lu_env *env, struct md_object *obj,
-                               int flags)
-{
-        struct mdd_object *mdd_obj = md2mdd_obj(obj);
-
-        mdd_write_lock(env, mdd_obj, MOR_TGT_CHILD);
-        mdd_obj->mod_count++;
-        mdd_write_unlock(env, mdd_obj);
-
-        return 0;
-}
-
-static int dot_lustre_mdd_close(const struct lu_env *env, struct md_object *obj,
-                                struct md_attr *ma, int mode)
-{
-        struct mdd_object *mdd_obj = md2mdd_obj(obj);
-
-        mdd_write_lock(env, mdd_obj, MOR_TGT_CHILD);
-        mdd_obj->mod_count--;
-        mdd_write_unlock(env, mdd_obj);
-
-        return 0;
-}
-
-static int dot_lustre_mdd_object_sync(const struct lu_env *env,
-                                      struct md_object *obj)
-{
-        return -ENOSYS;
-}
-
-static int dot_lustre_mdd_path(const struct lu_env *env, struct md_object *obj,
-                           char *path, int pathlen, __u64 *recno, int *linkno)
-{
-        return -ENOSYS;
-}
-
-static int dot_file_lock(const struct lu_env *env, struct md_object *obj,
-                         struct lov_mds_md *lmm, struct ldlm_extent *extent,
-                         struct lustre_handle *lockh)
-{
-        return -ENOSYS;
-}
-
-static int dot_file_unlock(const struct lu_env *env, struct md_object *obj,
-                           struct lov_mds_md *lmm, struct lustre_handle *lockh)
-{
-        return -ENOSYS;
-}
-
-static struct md_object_operations mdd_dot_lustre_obj_ops = {
-        .moo_permission    = dot_lustre_mdd_permission,
-        .moo_attr_get      = dot_lustre_mdd_attr_get,
-        .moo_attr_set      = dot_lustre_mdd_attr_set,
-        .moo_xattr_get     = dot_lustre_mdd_xattr_get,
-        .moo_xattr_list    = dot_lustre_mdd_xattr_list,
-        .moo_xattr_set     = dot_lustre_mdd_xattr_set,
-        .moo_xattr_del     = dot_lustre_mdd_xattr_del,
-        .moo_readpage      = mdd_readpage,
-        .moo_readlink      = dot_lustre_mdd_readlink,
-        .moo_object_create = dot_lustre_mdd_object_create,
-        .moo_ref_add       = dot_lustre_mdd_ref_add,
-        .moo_ref_del       = dot_lustre_mdd_ref_del,
-        .moo_open          = dot_lustre_mdd_open,
-        .moo_close         = dot_lustre_mdd_close,
-        .moo_capa_get      = mdd_capa_get,
-        .moo_object_sync   = dot_lustre_mdd_object_sync,
-        .moo_path          = dot_lustre_mdd_path,
-        .moo_file_lock     = dot_file_lock,
-        .moo_file_unlock   = dot_file_unlock,
-};
-
-
-static int dot_lustre_mdd_lookup(const struct lu_env *env, struct md_object *p,
-                                 const struct lu_name *lname, struct lu_fid *f,
-                                 struct md_op_spec *spec)
-{
-        if (strcmp(lname->ln_name, mdd_obf_dir_name) == 0)
-                *f = LU_OBF_FID;
-        else
-                return -ENOENT;
-
-        return 0;
-}
-
-static mdl_mode_t dot_lustre_mdd_lock_mode(const struct lu_env *env,
-                                           struct md_object *obj,
-                                           mdl_mode_t mode)
-{
-        return MDL_MINMODE;
-}
-
-static int dot_lustre_mdd_create(const struct lu_env *env,
-                                 struct md_object *pobj,
-                                 const struct lu_name *lname,
-                                 struct md_object *child,
-                                 struct md_op_spec *spec,
-                                 struct md_attr* ma)
-{
-        return -EPERM;
-}
-
-static int dot_lustre_mdd_create_data(const struct lu_env *env,
-                                      struct md_object *p,
-                                      struct md_object *o,
-                                      const struct md_op_spec *spec,
-                                      struct md_attr *ma)
-{
-        return -EPERM;
-}
-
-static int dot_lustre_mdd_rename(const struct lu_env *env,
-                                 struct md_object *src_pobj,
-                                 struct md_object *tgt_pobj,
-                                 const struct lu_fid *lf,
-                                 const struct lu_name *lsname,
-                                 struct md_object *tobj,
-                                 const struct lu_name *ltname,
-                                 struct md_attr *ma)
-{
-        return -EPERM;
-}
-
-static int dot_lustre_mdd_link(const struct lu_env *env,
-                               struct md_object *tgt_obj,
-                               struct md_object *src_obj,
-                               const struct lu_name *lname,
-                               struct md_attr *ma)
-{
-        return -EPERM;
-}
-
-static int dot_lustre_mdd_unlink(const struct lu_env *env,
-                                 struct md_object *pobj,
-                                 struct md_object *cobj,
-                                 const struct lu_name *lname,
-                                 struct md_attr *ma)
-{
-        return -EPERM;
-}
-
-static int dot_lustre_mdd_name_insert(const struct lu_env *env,
-                                      struct md_object *obj,
-                                      const struct lu_name *lname,
-                                      const struct lu_fid *fid,
-                                      const struct md_attr *ma)
-{
-        return -EPERM;
-}
-
-static int dot_lustre_mdd_name_remove(const struct lu_env *env,
-                                      struct md_object *obj,
-                                      const struct lu_name *lname,
-                                      const struct md_attr *ma)
-{
-        return -EPERM;
-}
-
-static int dot_lustre_mdd_rename_tgt(const struct lu_env *env,
-                                     struct md_object *pobj,
-                                     struct md_object *tobj,
-                                     const struct lu_fid *fid,
-                                     const struct lu_name *lname,
-                                     struct md_attr *ma)
-{
-        return -EPERM;
-}
-
-
-static struct md_dir_operations mdd_dot_lustre_dir_ops = {
-        .mdo_is_subdir   = mdd_is_subdir,
-        .mdo_lookup      = dot_lustre_mdd_lookup,
-        .mdo_lock_mode   = dot_lustre_mdd_lock_mode,
-        .mdo_create      = dot_lustre_mdd_create,
-        .mdo_create_data = dot_lustre_mdd_create_data,
-        .mdo_rename      = dot_lustre_mdd_rename,
-        .mdo_link        = dot_lustre_mdd_link,
-        .mdo_unlink      = dot_lustre_mdd_unlink,
-        .mdo_name_insert = dot_lustre_mdd_name_insert,
-        .mdo_name_remove = dot_lustre_mdd_name_remove,
-        .mdo_rename_tgt  = dot_lustre_mdd_rename_tgt,
-};
-
-static int obf_attr_get(const struct lu_env *env, struct md_object *obj,
-                        struct md_attr *ma)
-{
-        int rc = 0;
-
-        if (ma->ma_need & MA_INODE) {
-                struct mdd_device *mdd = mdo2mdd(obj);
-
-                /* "fid" is a virtual object and hence does not have any "real"
-                 * attributes. So we reuse attributes of .lustre for "fid" dir */
-                ma->ma_need |= MA_INODE;
-                rc = dot_lustre_mdd_attr_get(env, &mdd->mdd_dot_lustre->mod_obj,
-                                             ma);
-                if (rc)
-                        return rc;
-                ma->ma_valid |= MA_INODE;
-        }
-
-        /* "fid" directory does not have any striping information. */
-        if (ma->ma_need & MA_LOV) {
-                struct mdd_object *mdd_obj = md2mdd_obj(obj);
-
-                if (ma->ma_valid & MA_LOV)
-                        return 0;
-
-                if (!(S_ISREG(mdd_object_type(mdd_obj)) ||
-                      S_ISDIR(mdd_object_type(mdd_obj))))
-                        return 0;
-
-                if (ma->ma_need & MA_LOV_DEF) {
-                        rc = mdd_get_default_md(mdd_obj, ma->ma_lmm);
-                        if (rc > 0) {
-                                ma->ma_lmm_size = rc;
-                                ma->ma_valid |= MA_LOV;
-                                rc = 0;
-                        }
-                }
-        }
-
-        return rc;
-}
-
-static int obf_attr_set(const struct lu_env *env, struct md_object *obj,
-                        const struct md_attr *ma)
-{
-        return -EPERM;
-}
-
-static int obf_xattr_get(const struct lu_env *env,
-                         struct md_object *obj, struct lu_buf *buf,
-                         const char *name)
-{
-        return 0;
-}
-
-static int obf_mdd_open(const struct lu_env *env, struct md_object *obj,
-                        int flags)
-{
-        struct mdd_object *mdd_obj = md2mdd_obj(obj);
-
-        mdd_write_lock(env, mdd_obj, MOR_TGT_CHILD);
-        mdd_obj->mod_count++;
-        mdd_write_unlock(env, mdd_obj);
+	rec->cr.cr_markerflags = mdd->mdd_cl.mc_flags | markerflags;
+	rec->cr_hdr.lrh_len = llog_data_len(sizeof(*rec) + rec->cr.cr_namelen);
+	rec->cr_hdr.lrh_type = CHANGELOG_REC;
+	rec->cr.cr_time = cl_time();
+	spin_lock(&mdd->mdd_cl.mc_lock);
+	rec->cr.cr_index = ++mdd->mdd_cl.mc_index;
+	spin_unlock(&mdd->mdd_cl.mc_lock);
 
-        return 0;
-}
-
-static int obf_mdd_close(const struct lu_env *env, struct md_object *obj,
-                         struct md_attr *ma, int mode)
-{
-        struct mdd_object *mdd_obj = md2mdd_obj(obj);
-
-        mdd_write_lock(env, mdd_obj, MOR_TGT_CHILD);
-        mdd_obj->mod_count--;
-        mdd_write_unlock(env, mdd_obj);
+	ctxt = llog_get_context(obd, LLOG_CHANGELOG_ORIG_CTXT);
+	LASSERT(ctxt);
 
-        return 0;
-}
-
-/** Nothing to list in "fid" directory */
-static int obf_mdd_readpage(const struct lu_env *env, struct md_object *obj,
-                            const struct lu_rdpg *rdpg)
-{
-        return -EPERM;
-}
+	rc = llog_cat_add(env, ctxt->loc_handle, &rec->cr_hdr, NULL, NULL);
+	if (rc > 0)
+		rc = 0;
+	llog_ctxt_put(ctxt);
 
-static int obf_path(const struct lu_env *env, struct md_object *obj,
-                    char *path, int pathlen, __u64 *recno, int *linkno)
-{
-        return -ENOSYS;
+	/* assume on or off event; reset repeat-access time */
+	mdd->mdd_cl.mc_starttime = cfs_time_current_64();
+	RETURN(rc);
 }
 
-static struct md_object_operations mdd_obf_obj_ops = {
-        .moo_attr_get   = obf_attr_get,
-        .moo_attr_set   = obf_attr_set,
-        .moo_xattr_get  = obf_xattr_get,
-        .moo_open       = obf_mdd_open,
-        .moo_close      = obf_mdd_close,
-        .moo_readpage   = obf_mdd_readpage,
-        .moo_path       = obf_path
-};
-
 /**
  * Lookup method for "fid" object. Only filenames with correct SEQ:OID format
  * are valid. We also check if object with passed fid exists or not.
@@ -832,11 +558,22 @@ static int obf_lookup(const struct lu_env *env, struct md_object *p,
 
         sscanf(name, SFID, RFID(f));
         if (!fid_is_sane(f)) {
-                CWARN("bad FID format [%s], should be "DFID"\n", lname->ln_name,
-                      (__u64)1, 2, 0);
+		CWARN("%s: bad FID format [%s], should be "DFID"\n",
+		      mdd2obd_dev(mdd)->obd_name, lname->ln_name,
+		      (__u64)FID_SEQ_NORMAL, 1, 0);
                 GOTO(out, rc = -EINVAL);
         }
 
+	if (!fid_is_norm(f) && !fid_is_igif(f) && !fid_is_root(f) &&
+	    !fid_seq_is_dot(f->f_seq)) {
+		CWARN("%s: "DFID" is invalid, sequence should be "
+		      ">= "LPX64" or within ["LPX64","LPX64"].\n",
+		      mdd2obd_dev(mdd)->obd_name, PFID(f),
+		      (__u64)FID_SEQ_NORMAL, (__u64)FID_SEQ_IGIF,
+		      (__u64)FID_SEQ_IGIF_MAX);
+		GOTO(out, rc = -EINVAL);
+	}
+
         /* Check if object with this fid exists */
         child = mdd_object_find(env, mdd, f);
         if (child == NULL)
@@ -877,10 +614,10 @@ static int obf_link(const struct lu_env *env, struct md_object *tgt_obj,
 }
 
 static int obf_unlink(const struct lu_env *env, struct md_object *pobj,
-                      struct md_object *cobj, const struct lu_name *lname,
-                      struct md_attr *ma)
+		      struct md_object *cobj, const struct lu_name *lname,
+		      struct md_attr *ma, int no_name)
 {
-        return -EPERM;
+	return -EPERM;
 }
 
 static struct md_dir_operations mdd_obf_dir_ops = {
@@ -891,65 +628,178 @@ static struct md_dir_operations mdd_obf_dir_ops = {
         .mdo_unlink = obf_unlink
 };
 
+static struct md_object *mdo_locate(const struct lu_env *env,
+				    struct md_device *md,
+				    const struct lu_fid *fid)
+{
+	struct lu_object *obj;
+	struct md_object *mdo;
+
+	obj = lu_object_find(env, &md->md_lu_dev, fid, NULL);
+	if (!IS_ERR(obj)) {
+		obj = lu_object_locate(obj->lo_header, md->md_lu_dev.ld_type);
+		LASSERT(obj != NULL);
+		mdo = lu2md(obj);
+	} else {
+		mdo = ERR_PTR(PTR_ERR(obj));
+	}
+	return mdo;
+}
+
 /**
  * Create special in-memory "fid" object for open-by-fid.
  */
 static int mdd_obf_setup(const struct lu_env *env, struct mdd_device *m)
 {
-        struct mdd_object *mdd_obf;
-        struct lu_object *obf_lu_obj;
-        int rc = 0;
+	struct md_object	*mdo;
+	struct mdd_object	*mdd_obf;
+	struct lu_fid		 fid = LU_OBF_FID;
+	int			 rc;
 
-        m->mdd_dot_lustre_objs.mdd_obf = mdd_object_find(env, m,
-                                                         &LU_OBF_FID);
-        if (m->mdd_dot_lustre_objs.mdd_obf == NULL ||
-            IS_ERR(m->mdd_dot_lustre_objs.mdd_obf))
-                GOTO(out, rc = -ENOENT);
+	rc = mdd_local_file_create(env, m, mdd_object_fid(m->mdd_dot_lustre),
+				   mdd_obf_dir_name, S_IFDIR | S_IXUSR, &fid);
+	if (rc < 0)
+		RETURN(rc);
 
-        mdd_obf = m->mdd_dot_lustre_objs.mdd_obf;
-        mdd_obf->mod_obj.mo_dir_ops = &mdd_obf_dir_ops;
-        mdd_obf->mod_obj.mo_ops = &mdd_obf_obj_ops;
-        /* Don't allow objects to be created in "fid" dir */
-        mdd_obf->mod_flags |= IMMUTE_OBJ;
+	mdo = mdo_locate(env, &m->mdd_md_dev, &fid);
+	if (IS_ERR(mdo))
+		RETURN(PTR_ERR(mdo));
 
-        obf_lu_obj = mdd2lu_obj(mdd_obf);
-        obf_lu_obj->lo_header->loh_attr |= (LOHA_EXISTS | S_IFDIR);
+	LASSERT(lu_object_exists(&mdo->mo_lu));
 
-out:
-        return rc;
+	mdd_obf = md2mdd_obj(mdo);
+	mdd_obf->mod_obj.mo_dir_ops = &mdd_obf_dir_ops;
+	m->mdd_dot_lustre_objs.mdd_obf = mdd_obf;
+
+	return 0;
 }
 
 /** Setup ".lustre" directory object */
 static int mdd_dot_lustre_setup(const struct lu_env *env, struct mdd_device *m)
 {
-        struct dt_object *dt_dot_lustre;
-        struct lu_fid *fid = &mdd_env_info(env)->mti_fid;
-        int rc;
+	struct md_object	*mdo;
+	struct lu_fid		 fid;
+	int			 rc;
+
+	ENTRY;
+	/* Create ".lustre" directory in ROOT. */
+	fid = LU_DOT_LUSTRE_FID;
+	rc = mdd_local_file_create(env, m, &m->mdd_root_fid,
+				   dot_lustre_name,
+				   S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO,
+				   &fid);
+	if (rc < 0)
+		RETURN(rc);
+	mdo = mdo_locate(env, &m->mdd_md_dev, &fid);
+	if (IS_ERR(mdo))
+		RETURN(PTR_ERR(mdo));
+	LASSERT(lu_object_exists(&mdo->mo_lu));
+
+	m->mdd_dot_lustre = md2mdd_obj(mdo);
+
+	rc = mdd_obf_setup(env, m);
+	if (rc) {
+		CERROR("%s: error initializing \"fid\" object: rc = %d.\n",
+		       mdd2obd_dev(m)->obd_name, rc);
+		GOTO(out, rc);
+	}
+	RETURN(0);
+out:
+	mdd_object_put(env, m->mdd_dot_lustre);
+	m->mdd_dot_lustre = NULL;
+	return rc;
+}
 
-        rc = create_dot_lustre_dir(env, m);
-        if (rc)
-                return rc;
 
-        dt_dot_lustre = dt_store_open(env, m->mdd_child, mdd_root_dir_name,
-                                      dot_lustre_name, fid);
-        if (IS_ERR(dt_dot_lustre)) {
-                rc = PTR_ERR(dt_dot_lustre);
-                GOTO(out, rc);
-        }
+static struct llog_operations hsm_actions_logops;
 
-        /* references are released in mdd_device_shutdown() */
-        m->mdd_dot_lustre = lu2mdd_obj(lu_object_locate(dt_dot_lustre->do_lu.lo_header,
-                                                        &mdd_device_type));
+/**
+ * set llog methods and create LLOG_AGENT_ORIG_CTXT llog
+ * object in obd_device
+ */
+static int mdd_hsm_actions_llog_init(const struct lu_env *env,
+				     struct mdd_device *m)
+{
+	struct obd_device	*obd = mdd2obd_dev(m);
+	struct llog_ctxt	*ctxt = NULL;
+	int			 rc;
+	ENTRY;
 
-        m->mdd_dot_lustre->mod_obj.mo_dir_ops = &mdd_dot_lustre_dir_ops;
-        m->mdd_dot_lustre->mod_obj.mo_ops = &mdd_dot_lustre_obj_ops;
+	OBD_SET_CTXT_MAGIC(&obd->obd_lvfs_ctxt);
+	obd->obd_lvfs_ctxt.dt = m->mdd_bottom;
 
-        rc = mdd_obf_setup(env, m);
-        if (rc)
-                CERROR("Error initializing \"fid\" object - %d.\n", rc);
+	rc = llog_setup(env, obd, &obd->obd_olg, LLOG_AGENT_ORIG_CTXT,
+			obd, &hsm_actions_logops);
+	if (rc) {
+		CERROR("%s: hsm actions llog setup failed: rc = %d\n",
+			obd->obd_name, rc);
+		RETURN(rc);
+	}
 
-out:
-        RETURN(rc);
+	ctxt = llog_get_context(obd, LLOG_AGENT_ORIG_CTXT);
+	LASSERT(ctxt);
+
+	rc = llog_open_create(env, ctxt, &ctxt->loc_handle, NULL,
+			      HSM_ACTIONS);
+	if (rc) {
+		CERROR("%s: hsm actions llog open_create failed: rc = %d\n",
+			obd->obd_name, rc);
+		GOTO(out_cleanup, rc);
+	}
+
+	rc = llog_cat_init_and_process(env, ctxt->loc_handle);
+	if (rc)
+		GOTO(out_close, rc);
+
+	llog_ctxt_put(ctxt);
+	RETURN(0);
+
+out_close:
+	llog_cat_close(env, ctxt->loc_handle);
+	ctxt->loc_handle = NULL;
+out_cleanup:
+	llog_cleanup(env, ctxt);
+
+	return rc;
+}
+
+/**
+ * cleanup the context created by llog_setup_named()
+ */
+static int mdd_hsm_actions_llog_fini(const struct lu_env *env,
+				     struct mdd_device *m)
+{
+	struct obd_device	*obd = mdd2obd_dev(m);
+	struct llog_ctxt	*lctxt;
+	ENTRY;
+
+	lctxt = llog_get_context(obd, LLOG_AGENT_ORIG_CTXT);
+	if (lctxt) {
+		llog_cat_close(env, lctxt->loc_handle);
+		lctxt->loc_handle = NULL;
+		llog_cleanup(env, lctxt);
+	}
+
+	RETURN(0);
+}
+
+static void mdd_device_shutdown(const struct lu_env *env, struct mdd_device *m,
+				struct lustre_cfg *cfg)
+{
+	lfsck_degister(env, m->mdd_bottom);
+	mdd_hsm_actions_llog_fini(env, m);
+	mdd_changelog_fini(env, m);
+	orph_index_fini(env, m);
+	if (m->mdd_dot_lustre_objs.mdd_obf)
+		mdd_object_put(env, m->mdd_dot_lustre_objs.mdd_obf);
+	if (m->mdd_dot_lustre)
+		mdd_object_put(env, m->mdd_dot_lustre);
+	if (m->mdd_los != NULL)
+		local_oid_storage_fini(env, m->mdd_los);
+	lu_site_purge(env, mdd2lu_dev(m)->ld_site, ~0);
+
+	if (m->mdd_child_exp)
+		obd_disconnect(m->mdd_child_exp);
 }
 
 static int mdd_process_config(const struct lu_env *env,
@@ -976,18 +826,13 @@ static int mdd_process_config(const struct lu_env *env,
                 rc = next->ld_ops->ldo_process_config(env, next, cfg);
                 if (rc)
                         GOTO(out, rc);
-                dt->dd_ops->dt_conf_get(env, dt, &m->mdd_dt_conf);
-
-                rc = mdd_init_obd(env, m, cfg);
-                if (rc) {
-                        CERROR("lov init error %d\n", rc);
-                        GOTO(out, rc);
-                }
-
-                mdd_changelog_init(env, m);
+		dt_conf_get(env, dt, &m->mdd_dt_conf);
                 break;
         case LCFG_CLEANUP:
-                mdd_device_shutdown(env, m, cfg);
+		rc = next->ld_ops->ldo_process_config(env, next, cfg);
+		lu_dev_del_linkage(d->ld_site, d);
+		mdd_device_shutdown(env, m, cfg);
+		break;
         default:
                 rc = next->ld_ops->ldo_process_config(env, next, cfg);
                 break;
@@ -996,67 +841,16 @@ out:
         RETURN(rc);
 }
 
-#if 0
-static int mdd_lov_set_nextid(const struct lu_env *env,
-                              struct mdd_device *mdd)
-{
-        struct mds_obd *mds = &mdd->mdd_obd_dev->u.mds;
-        int rc;
-        ENTRY;
-
-        LASSERT(mds->mds_lov_objids != NULL);
-        rc = obd_set_info_async(mds->mds_lov_exp, strlen(KEY_NEXT_ID),
-                                KEY_NEXT_ID, mds->mds_lov_desc.ld_tgt_count,
-                                mds->mds_lov_objids, NULL);
-
-        RETURN(rc);
-}
-
-static int mdd_cleanup_unlink_llog(const struct lu_env *env,
-                                   struct mdd_device *mdd)
-{
-        /* XXX: to be implemented! */
-        return 0;
-}
-#endif
-
 static int mdd_recovery_complete(const struct lu_env *env,
                                  struct lu_device *d)
 {
         struct mdd_device *mdd = lu2mdd_dev(d);
-        struct lu_device *next = &mdd->mdd_child->dd_lu_dev;
-        struct obd_device *obd = mdd2obd_dev(mdd);
+	struct lu_device *next;
         int rc;
         ENTRY;
 
         LASSERT(mdd != NULL);
-        LASSERT(obd != NULL);
-#if 0
-        /* XXX: Do we need this in new stack? */
-        rc = mdd_lov_set_nextid(env, mdd);
-        if (rc) {
-                CERROR("mdd_lov_set_nextid() failed %d\n",
-                       rc);
-                RETURN(rc);
-        }
-
-        /* XXX: cleanup unlink. */
-        rc = mdd_cleanup_unlink_llog(env, mdd);
-        if (rc) {
-                CERROR("mdd_cleanup_unlink_llog() failed %d\n",
-                       rc);
-                RETURN(rc);
-        }
-#endif
-        /* Call that with obd_recovering = 1 just to update objids */
-        obd_notify(obd->u.mds.mds_lov_obd, NULL, (obd->obd_async_recov ?
-                    OBD_NOTIFY_SYNC_NONBLOCK : OBD_NOTIFY_SYNC), NULL);
-
-        /* Drop obd_recovering to 0 and call o_postrecov to recover mds_lov */
-        cfs_spin_lock(&obd->obd_dev_lock);
-        obd->obd_recovering = 0;
-        cfs_spin_unlock(&obd->obd_dev_lock);
-        obd->obd_type->typ_dt_ops->o_postrecov(obd);
+	next = &mdd->mdd_child->dd_lu_dev;
 
         /* XXX: orphans handling. */
         __mdd_orphan_cleanup(env, mdd);
@@ -1065,50 +859,137 @@ static int mdd_recovery_complete(const struct lu_env *env,
         RETURN(rc);
 }
 
+int mdd_local_file_create(const struct lu_env *env, struct mdd_device *mdd,
+			  const struct lu_fid *pfid, const char *name, __u32 mode,
+			  struct lu_fid *fid)
+{
+	struct dt_object	*parent, *dto;
+	int			 rc;
+
+	ENTRY;
+
+	LASSERT(!fid_is_zero(pfid));
+	parent = dt_locate(env, mdd->mdd_bottom, pfid);
+	if (unlikely(IS_ERR(parent)))
+		RETURN(PTR_ERR(parent));
+
+	/* create local file/dir, if @fid is passed then try to use it */
+	if (fid_is_zero(fid))
+		dto = local_file_find_or_create(env, mdd->mdd_los, parent,
+						name, mode);
+	else
+		dto = local_file_find_or_create_with_fid(env, mdd->mdd_bottom,
+							 fid, parent, name,
+							 mode);
+	if (IS_ERR(dto))
+		GOTO(out_put, rc = PTR_ERR(dto));
+	*fid = *lu_object_fid(&dto->do_lu);
+	/* since stack is not fully set up the local_storage uses own stack
+	 * and we should drop its object from cache */
+	lu_object_put_nocache(env, &dto->do_lu);
+	EXIT;
+out_put:
+	lu_object_put(env, &parent->do_lu);
+	return 0;
+}
+
 static int mdd_prepare(const struct lu_env *env,
                        struct lu_device *pdev,
                        struct lu_device *cdev)
 {
-        struct mdd_device *mdd = lu2mdd_dev(cdev);
-        struct lu_device *next = &mdd->mdd_child->dd_lu_dev;
-        struct dt_object *root;
-        struct lu_fid     fid;
-        int rc;
-
-        ENTRY;
-        rc = next->ld_ops->ldo_prepare(env, cdev, next);
-        if (rc)
-                GOTO(out, rc);
-
-        dt_txn_callback_add(mdd->mdd_child, &mdd->mdd_txn_cb);
-        root = dt_store_open(env, mdd->mdd_child, "", mdd_root_dir_name,
-                             &mdd->mdd_root_fid);
-        if (!IS_ERR(root)) {
-                LASSERT(root != NULL);
-                lu_object_put(env, &root->do_lu);
-                rc = orph_index_init(env, mdd);
-        } else {
-                rc = PTR_ERR(root);
-        }
-        if (rc)
-                GOTO(out, rc);
-
-        rc = mdd_dot_lustre_setup(env, mdd);
-        if (rc) {
-                CERROR("Error(%d) initializing .lustre objects\n", rc);
-                GOTO(out, rc);
-        }
-
-        /* we use capa file to declare llog changes,
-         * will be fixed with new llog in 2.3 */
-        root = dt_store_open(env, mdd->mdd_child, "", CAPA_KEYS, &fid);
-        if (!IS_ERR(root))
-                mdd->mdd_capa = root;
-        else
-                rc = PTR_ERR(root);
-
-out:
-        RETURN(rc);
+	struct mdd_device	*mdd = lu2mdd_dev(cdev);
+	struct lu_device	*next = &mdd->mdd_child->dd_lu_dev;
+	struct lu_fid		 fid;
+	int			 rc;
+
+	ENTRY;
+
+	rc = next->ld_ops->ldo_prepare(env, cdev, next);
+	if (rc)
+		RETURN(rc);
+
+	/* Setup local dirs */
+	fid.f_seq = FID_SEQ_LOCAL_NAME;
+	fid.f_oid = 1;
+	fid.f_ver = 0;
+	rc = local_oid_storage_init(env, mdd->mdd_bottom, &fid,
+				    &mdd->mdd_los);
+	if (rc)
+		RETURN(rc);
+
+	rc = dt_root_get(env, mdd->mdd_child, &mdd->mdd_local_root_fid);
+	if (rc < 0)
+		GOTO(out_los, rc);
+
+	lu_root_fid(&fid);
+	if (mdd_seq_site(mdd)->ss_node_id == 0) {
+		rc = mdd_local_file_create(env, mdd, &mdd->mdd_local_root_fid,
+					   mdd_root_dir_name, S_IFDIR |
+					   S_IRUGO | S_IWUSR | S_IXUGO, &fid);
+		if (rc != 0) {
+			CERROR("%s: create root fid failed: rc = %d\n",
+			       mdd2obd_dev(mdd)->obd_name, rc);
+			GOTO(out_los, rc);
+		}
+
+		mdd->mdd_root_fid = fid;
+		rc = mdd_dot_lustre_setup(env, mdd);
+		if (rc != 0) {
+			CERROR("%s: initializing .lustre failed: rc = %d\n",
+			       mdd2obd_dev(mdd)->obd_name, rc);
+			GOTO(out_los, rc);
+		}
+
+		rc = mdd_compat_fixes(env, mdd);
+		if (rc)
+			GOTO(out_los, rc);
+
+	} else {
+		/* Normal client usually send root access to MDT0 directly,
+		 * the root FID on non-MDT0 will only be used by echo client. */
+		mdd->mdd_root_fid = fid;
+	}
+
+	rc = orph_index_init(env, mdd);
+	if (rc < 0)
+		GOTO(out_dot, rc);
+
+	rc = mdd_changelog_init(env, mdd);
+	if (rc != 0) {
+		CERROR("%s: failed to initialize changelog: rc = %d\n",
+		       mdd2obd_dev(mdd)->obd_name, rc);
+		GOTO(out_orph, rc);
+	}
+
+	rc = mdd_hsm_actions_llog_init(env, mdd);
+	if (rc != 0)
+		GOTO(out_changelog, rc);
+
+	rc = lfsck_register(env, mdd->mdd_bottom, mdd->mdd_child, true);
+	if (rc != 0) {
+		CERROR("%s: failed to initialize lfsck: rc = %d\n",
+		       mdd2obd_dev(mdd)->obd_name, rc);
+		GOTO(out_hsm, rc);
+	}
+	RETURN(0);
+out_hsm:
+	mdd_hsm_actions_llog_fini(env, mdd);
+out_changelog:
+	mdd_changelog_fini(env, mdd);
+out_orph:
+	orph_index_fini(env, mdd);
+out_dot:
+	if (mdd_seq_site(mdd)->ss_node_id == 0) {
+		mdd_object_put(env, mdd->mdd_dot_lustre);
+		mdd->mdd_dot_lustre = NULL;
+		mdd_object_put(env, mdd->mdd_dot_lustre_objs.mdd_obf);
+		mdd->mdd_dot_lustre_objs.mdd_obf = NULL;
+	}
+out_los:
+	local_oid_storage_fini(env, mdd->mdd_los);
+	mdd->mdd_los = NULL;
+
+	return rc;
 }
 
 const struct lu_device_operations mdd_lu_ops = {
@@ -1118,82 +999,69 @@ const struct lu_device_operations mdd_lu_ops = {
         .ldo_prepare           = mdd_prepare,
 };
 
-/*
- * No permission check is needed.
- */
 static int mdd_root_get(const struct lu_env *env,
-                        struct md_device *m, struct lu_fid *f)
+			struct md_device *m, struct lu_fid *f)
 {
-        struct mdd_device *mdd = lu2mdd_dev(&m->md_lu_dev);
+	struct mdd_device *mdd = lu2mdd_dev(&m->md_lu_dev);
 
-        ENTRY;
-        *f = mdd->mdd_root_fid;
-        RETURN(0);
+	ENTRY;
+	*f = mdd->mdd_root_fid;
+	RETURN(0);
 }
 
 /*
  * No permission check is needed.
  */
 static int mdd_statfs(const struct lu_env *env, struct md_device *m,
-                      struct obd_statfs *sfs)
+		      struct obd_statfs *sfs)
 {
-        struct mdd_device *mdd = lu2mdd_dev(&m->md_lu_dev);
-        int rc;
+	struct mdd_device *mdd = lu2mdd_dev(&m->md_lu_dev);
+	int rc;
 
-        ENTRY;
+	ENTRY;
 
-        rc = mdd_child_ops(mdd)->dt_statfs(env, mdd->mdd_child, sfs);
+	rc = mdd_child_ops(mdd)->dt_statfs(env, mdd->mdd_child, sfs);
 
-        RETURN(rc);
+	sfs->os_namelen = min_t(__u32, sfs->os_namelen, NAME_MAX);
+
+	RETURN(rc);
 }
 
 /*
  * No permission check is needed.
  */
-static int mdd_maxsize_get(const struct lu_env *env, struct md_device *m,
-                           int *md_size, int *cookie_size)
-{
-        struct mdd_device *mdd = lu2mdd_dev(&m->md_lu_dev);
-        ENTRY;
-
-        *md_size = mdd_lov_mdsize(env, mdd);
-        *cookie_size = mdd_lov_cookiesize(env, mdd);
-
-        RETURN(0);
-}
-
 static int mdd_init_capa_ctxt(const struct lu_env *env, struct md_device *m,
                               int mode, unsigned long timeout, __u32 alg,
                               struct lustre_capa_key *keys)
 {
         struct mdd_device *mdd = lu2mdd_dev(&m->md_lu_dev);
-        struct mds_obd    *mds = &mdd2obd_dev(mdd)->u.mds;
         int rc;
         ENTRY;
 
         /* need barrier for mds_capa_keys access. */
-        cfs_down_write(&mds->mds_notify_lock);
-        mds->mds_capa_keys = keys;
-        cfs_up_write(&mds->mds_notify_lock);
 
         rc = mdd_child_ops(mdd)->dt_init_capa_ctxt(env, mdd->mdd_child, mode,
                                                    timeout, alg, keys);
         RETURN(rc);
 }
 
+static int mdd_maxeasize_get(const struct lu_env *env, struct md_device *m,
+				int *easize)
+{
+	struct mdd_device *mdd = lu2mdd_dev(&m->md_lu_dev);
+	ENTRY;
+
+	*easize = mdd->mdd_dt_conf.ddp_max_ea_size;
+
+	RETURN(0);
+}
+
 static int mdd_update_capa_key(const struct lu_env *env,
                                struct md_device *m,
                                struct lustre_capa_key *key)
 {
-        struct mds_capa_info info = { .uuid = NULL, .capa = key };
-        struct mdd_device *mdd = lu2mdd_dev(&m->md_lu_dev);
-        struct obd_export *lov_exp = mdd2obd_dev(mdd)->u.mds.mds_lov_exp;
-        int rc;
-        ENTRY;
-
-        rc = obd_set_info_async(env, lov_exp, sizeof(KEY_CAPA_KEY),
-                                KEY_CAPA_KEY, sizeof(info), &info, NULL);
-        RETURN(rc);
+	/* we do not support capabilities ... */
+	return -EINVAL;
 }
 
 static int mdd_llog_ctxt_get(const struct lu_env *env, struct md_device *m,
@@ -1205,6 +1073,18 @@ static int mdd_llog_ctxt_get(const struct lu_env *env, struct md_device *m,
         return (*h == NULL ? -ENOENT : 0);
 }
 
+static struct lu_device *mdd_device_free(const struct lu_env *env,
+					 struct lu_device *lu)
+{
+	struct mdd_device *m = lu2mdd_dev(lu);
+	ENTRY;
+
+	LASSERT(atomic_read(&lu->ld_ref) == 0);
+	md_device_fini(&m->mdd_md_dev);
+	OBD_FREE_PTR(m);
+	RETURN(NULL);
+}
+
 static struct lu_device *mdd_device_alloc(const struct lu_env *env,
                                           struct lu_device_type *t,
                                           struct lustre_cfg *lcfg)
@@ -1216,97 +1096,109 @@ static struct lu_device *mdd_device_alloc(const struct lu_env *env,
         if (m == NULL) {
                 l = ERR_PTR(-ENOMEM);
         } else {
-                md_device_init(&m->mdd_md_dev, t);
+		int rc;
+
                 l = mdd2lu_dev(m);
-                l->ld_ops = &mdd_lu_ops;
-                m->mdd_md_dev.md_ops = &mdd_ops;
-                md_upcall_init(&m->mdd_md_dev, NULL);
+		md_device_init(&m->mdd_md_dev, t);
+		rc = mdd_init0(env, m, t, lcfg);
+		if (rc != 0) {
+			mdd_device_free(env, l);
+			l = ERR_PTR(rc);
+		}
         }
 
         return l;
 }
 
-static struct lu_device *mdd_device_free(const struct lu_env *env,
-                                         struct lu_device *lu)
+/*
+ * we use exports to track all mdd users
+ */
+static int mdd_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 mdd_device *m = lu2mdd_dev(lu);
-        struct lu_device  *next = &m->mdd_child->dd_lu_dev;
-        ENTRY;
+	struct mdd_device    *mdd = lu2mdd_dev(obd->obd_lu_dev);
+	struct lustre_handle  conn;
+	int                   rc;
+	ENTRY;
 
-        LASSERT(cfs_atomic_read(&lu->ld_ref) == 0);
-        md_device_fini(&m->mdd_md_dev);
-        OBD_FREE_PTR(m);
-        RETURN(next);
-}
+	CDEBUG(D_CONFIG, "connect #%d\n", mdd->mdd_connects);
 
-static struct obd_ops mdd_obd_device_ops = {
-        .o_owner = THIS_MODULE
-};
+	rc = class_connect(&conn, obd, cluuid);
+	if (rc)
+		RETURN(rc);
 
-/* context key constructor/destructor: mdd_ucred_key_init, mdd_ucred_key_fini */
-LU_KEY_INIT_FINI(mdd_ucred, struct md_ucred);
+	*exp = class_conn2export(&conn);
 
-static struct lu_context_key mdd_ucred_key = {
-        .lct_tags = LCT_SESSION,
-        .lct_init = mdd_ucred_key_init,
-        .lct_fini = mdd_ucred_key_fini
-};
+	/* Why should there ever be more than 1 connect? */
+	LASSERT(mdd->mdd_connects == 0);
+	mdd->mdd_connects++;
 
-struct md_ucred *md_ucred(const struct lu_env *env)
-{
-        LASSERT(env->le_ses != NULL);
-        return lu_context_key_get(env->le_ses, &mdd_ucred_key);
+	RETURN(0);
 }
-EXPORT_SYMBOL(md_ucred);
 
 /*
- * context key constructor/destructor:
- * mdd_capainfo_key_init, mdd_capainfo_key_fini
+ * once last export (we don't count self-export) disappeared
+ * mdd can be released
  */
-LU_KEY_INIT_FINI(mdd_capainfo, struct md_capainfo);
+static int mdd_obd_disconnect(struct obd_export *exp)
+{
+	struct obd_device *obd = exp->exp_obd;
+	struct mdd_device *mdd = lu2mdd_dev(obd->obd_lu_dev);
+	int                rc, release = 0;
+	ENTRY;
 
-struct lu_context_key mdd_capainfo_key = {
-        .lct_tags = LCT_SESSION,
-        .lct_init = mdd_capainfo_key_init,
-        .lct_fini = mdd_capainfo_key_fini
-};
+	mdd->mdd_connects--;
+	if (mdd->mdd_connects == 0)
+		release = 1;
 
-struct md_capainfo *md_capainfo(const struct lu_env *env)
-{
-        /* NB, in mdt_init0 */
-        if (env->le_ses == NULL)
-                return NULL;
-        return lu_context_key_get(env->le_ses, &mdd_capainfo_key);
+	rc = class_disconnect(exp);
+
+	if (rc == 0 && release)
+		class_manual_cleanup(obd);
+	RETURN(rc);
 }
-EXPORT_SYMBOL(md_capainfo);
 
-/*
- * context key constructor/destructor:
- * mdd_quota_key_init, mdd_quota_key_fini
- */
-LU_KEY_INIT_FINI(mdd_quota, struct md_quota);
+static int mdd_obd_get_info(const struct lu_env *env, struct obd_export *exp,
+			    __u32 keylen, void *key, __u32 *vallen, void *val,
+			    struct lov_stripe_md *lsm)
+{
+	int rc = -EINVAL;
 
-struct lu_context_key mdd_quota_key = {
-        .lct_tags = LCT_SESSION,
-        .lct_init = mdd_quota_key_init,
-        .lct_fini = mdd_quota_key_fini
-};
+	if (KEY_IS(KEY_OSP_CONNECTED)) {
+		struct obd_device	*obd = exp->exp_obd;
+		struct mdd_device	*mdd;
 
-struct md_quota *md_quota(const struct lu_env *env)
-{
-        LASSERT(env->le_ses != NULL);
-        return lu_context_key_get(env->le_ses, &mdd_quota_key);
+		if (!obd->obd_set_up || obd->obd_stopping)
+			RETURN(-EAGAIN);
+
+		mdd = lu2mdd_dev(obd->obd_lu_dev);
+		LASSERT(mdd);
+		rc = obd_get_info(env, mdd->mdd_child_exp, keylen, key, vallen,
+				  val, lsm);
+		RETURN(rc);
+	}
+
+	RETURN(rc);
 }
-EXPORT_SYMBOL(md_quota);
 
-static int mdd_changelog_user_register(struct mdd_device *mdd, int *id)
+static struct obd_ops mdd_obd_device_ops = {
+	.o_owner	= THIS_MODULE,
+	.o_connect	= mdd_obd_connect,
+	.o_disconnect	= mdd_obd_disconnect,
+	.o_get_info     = mdd_obd_get_info,
+};
+
+static int mdd_changelog_user_register(const struct lu_env *env,
+				       struct mdd_device *mdd, int *id)
 {
         struct llog_ctxt *ctxt;
         struct llog_changelog_user_rec *rec;
         int rc;
         ENTRY;
 
-        ctxt = llog_get_context(mdd2obd_dev(mdd),LLOG_CHANGELOG_USER_ORIG_CTXT);
+        ctxt = llog_get_context(mdd2obd_dev(mdd),
+				LLOG_CHANGELOG_USER_ORIG_CTXT);
         if (ctxt == NULL)
                 RETURN(-ENXIO);
 
@@ -1317,23 +1209,23 @@ static int mdd_changelog_user_register(struct mdd_device *mdd, int *id)
         }
 
         /* Assume we want it on since somebody registered */
-        rc = mdd_changelog_on(mdd, 1);
+        rc = mdd_changelog_on(env, mdd, 1);
         if (rc)
                 GOTO(out, rc);
 
         rec->cur_hdr.lrh_len = sizeof(*rec);
         rec->cur_hdr.lrh_type = CHANGELOG_USER_REC;
-        cfs_spin_lock(&mdd->mdd_cl.mc_user_lock);
-        if (mdd->mdd_cl.mc_lastuser == (unsigned int)(-1)) {
-                cfs_spin_unlock(&mdd->mdd_cl.mc_user_lock);
-                CERROR("Maximum number of changelog users exceeded!\n");
-                GOTO(out, rc = -EOVERFLOW);
-        }
-        *id = rec->cur_id = ++mdd->mdd_cl.mc_lastuser;
-        rec->cur_endrec = mdd->mdd_cl.mc_index;
-        cfs_spin_unlock(&mdd->mdd_cl.mc_user_lock);
-
-        rc = llog_add(ctxt, &rec->cur_hdr, NULL, NULL, 0);
+	spin_lock(&mdd->mdd_cl.mc_user_lock);
+	if (mdd->mdd_cl.mc_lastuser == (unsigned int)(-1)) {
+		spin_unlock(&mdd->mdd_cl.mc_user_lock);
+		CERROR("Maximum number of changelog users exceeded!\n");
+		GOTO(out, rc = -EOVERFLOW);
+	}
+	*id = rec->cur_id = ++mdd->mdd_cl.mc_lastuser;
+	rec->cur_endrec = mdd->mdd_cl.mc_index;
+	spin_unlock(&mdd->mdd_cl.mc_user_lock);
+
+	rc = llog_cat_add(env, ctxt->loc_handle, &rec->cur_hdr, NULL, NULL);
 
         CDEBUG(D_IOCTL, "Registered changelog user %d\n", *id);
 out:
@@ -1342,44 +1234,13 @@ out:
         RETURN(rc);
 }
 
-int mdd_declare_llog_cancel(const struct lu_env *env, struct mdd_device *mdd,
-                            struct thandle *handle)
-{
-        int rc;
-
-
-        /* XXX: this is a temporary solution to declare llog changes
-         *      will be fixed in 2.3 with new llog implementation */
-
-        LASSERT(mdd->mdd_capa);
-
-        /* the llog record could be canceled either by modifying
-         * the plain llog's header or by destroying the llog itself
-         * when this record is the last one in it, it can't be known
-         * here, but the catlog's header will also be modified for
-         * the second case, then the first case can be covered and
-         * is no need to declare it */
-
-        /* destroy empty plain log */
-        rc = dt_declare_destroy(env, mdd->mdd_capa, handle);
-        if (rc)
-                return rc;
-
-        /* record the catlog's header if an empty plain log was destroyed */
-        rc = dt_declare_record_write(env, mdd->mdd_capa,
-                                     sizeof(struct llog_logid_rec), 0, handle);
-        return rc;
-}
-
 struct mdd_changelog_user_data {
         __u64 mcud_endrec; /**< purge record for this user */
         __u64 mcud_minrec; /**< lowest changelog recno still referenced */
         __u32 mcud_id;
         __u32 mcud_minid;  /**< user id with lowest rec reference */
         __u32 mcud_usercount;
-        int   mcud_found:1;
-        struct mdd_device   *mcud_mdd;
-        const struct lu_env *mcud_env;
+	unsigned int mcud_found:1;
 };
 #define MCUD_UNREGISTER -1LL
 
@@ -1387,13 +1248,14 @@ struct mdd_changelog_user_data {
  * 1. Find the smallest record everyone is willing to purge
  * 2. Update the last purgeable record for this user
  */
-static int mdd_changelog_user_purge_cb(struct llog_handle *llh,
-                                       struct llog_rec_hdr *hdr, void *data)
+static int mdd_changelog_user_purge_cb(const struct lu_env *env,
+				       struct llog_handle *llh,
+				       struct llog_rec_hdr *hdr, void *data)
 {
-        struct llog_changelog_user_rec *rec;
-        struct mdd_changelog_user_data *mcud =
-                (struct mdd_changelog_user_data *)data;
-        int rc;
+	struct llog_changelog_user_rec	*rec;
+	struct mdd_changelog_user_data	*mcud = data;
+	int				 rc;
+
         ENTRY;
 
         LASSERT(llh->lgh_hdr->llh_flags & LLOG_F_IS_PLAIN);
@@ -1422,35 +1284,15 @@ static int mdd_changelog_user_purge_cb(struct llog_handle *llh,
         /* Special case: unregister this user */
         if (mcud->mcud_endrec == MCUD_UNREGISTER) {
                 struct llog_cookie cookie;
-                void *th;
-                struct mdd_device *mdd = mcud->mcud_mdd;
 
                 cookie.lgc_lgl = llh->lgh_id;
                 cookie.lgc_index = hdr->lrh_index;
 
-                /* XXX This is a workaround for the deadlock of changelog
-                 * adding vs. changelog cancelling. LU-81. */
-                th = mdd_trans_create(mcud->mcud_env, mdd);
-                if (IS_ERR(th)) {
-                        CERROR("Cannot get thandle\n");
-                        RETURN(-ENOMEM);
-                }
-
-                rc = mdd_declare_llog_cancel(mcud->mcud_env, mdd, th); 
-                if (rc)
-                        GOTO(stop, rc);
-
-                rc = mdd_trans_start(mcud->mcud_env, mdd, th);
-                if (rc)
-                        GOTO(stop, rc);
-
-                rc = llog_cat_cancel_records(llh->u.phd.phd_cat_handle,
-                                             1, &cookie);
+		rc = llog_cat_cancel_records(env, llh->u.phd.phd_cat_handle,
+					     1, &cookie);
                 if (rc == 0)
                         mcud->mcud_usercount--;
 
-stop:
-                mdd_trans_stop(mcud->mcud_env, mdd, 0, th);
                 RETURN(rc);
         }
 
@@ -1460,8 +1302,8 @@ stop:
 
         /* hdr+1 is loc of data */
         hdr->lrh_len -= sizeof(*hdr) + sizeof(struct llog_rec_tail);
-        rc = llog_write_rec(llh, hdr, NULL, 0, (void *)(hdr + 1),
-                            hdr->lrh_index);
+	rc = llog_write(env, llh, hdr, NULL, 0, (void *)(hdr + 1),
+			hdr->lrh_index);
 
         RETURN(rc);
 }
@@ -1482,28 +1324,29 @@ static int mdd_changelog_user_purge(const struct lu_env *env,
         data.mcud_minrec = 0;
         data.mcud_usercount = 0;
         data.mcud_endrec = endrec;
-        data.mcud_mdd = mdd;
-        data.mcud_env = env;
-        cfs_spin_lock(&mdd->mdd_cl.mc_lock);
-        endrec = mdd->mdd_cl.mc_index;
-        cfs_spin_unlock(&mdd->mdd_cl.mc_lock);
+	spin_lock(&mdd->mdd_cl.mc_lock);
+	endrec = mdd->mdd_cl.mc_index;
+	spin_unlock(&mdd->mdd_cl.mc_lock);
         if ((data.mcud_endrec == 0) ||
             ((data.mcud_endrec > endrec) &&
              (data.mcud_endrec != MCUD_UNREGISTER)))
                 data.mcud_endrec = endrec;
 
-        ctxt = llog_get_context(mdd2obd_dev(mdd),LLOG_CHANGELOG_USER_ORIG_CTXT);
+        ctxt = llog_get_context(mdd2obd_dev(mdd),
+				LLOG_CHANGELOG_USER_ORIG_CTXT);
         if (ctxt == NULL)
                 return -ENXIO;
+
         LASSERT(ctxt->loc_handle->lgh_hdr->llh_flags & LLOG_F_IS_CAT);
 
-        rc = llog_cat_process(ctxt->loc_handle, mdd_changelog_user_purge_cb,
-                              (void *)&data, 0, 0);
+	rc = llog_cat_process(env, ctxt->loc_handle,
+			      mdd_changelog_user_purge_cb, (void *)&data,
+			      0, 0);
         if ((rc >= 0) && (data.mcud_minrec > 0)) {
                 CDEBUG(D_IOCTL, "Purging changelog entries up to "LPD64
                        ", referenced by "CHANGELOG_USER_PREFIX"%d\n",
                        data.mcud_minrec, data.mcud_minid);
-                rc = mdd_changelog_llog_cancel(mdd, data.mcud_minrec);
+		rc = mdd_changelog_llog_cancel(env, mdd, data.mcud_minrec);
         } else {
                 CWARN("Could not determine changelog records to purge; rc=%d\n",
                       rc);
@@ -1520,7 +1363,7 @@ static int mdd_changelog_user_purge(const struct lu_env *env,
 
         if (!rc && data.mcud_usercount == 0)
                 /* No more users; turn changelogs off */
-                rc = mdd_changelog_on(mdd, 0);
+                rc = mdd_changelog_on(env, mdd, 0);
 
         RETURN (rc);
 }
@@ -1556,6 +1399,19 @@ static int mdd_iocontrol(const struct lu_env *env, struct md_device *m,
                 *mntopts = mdd->mdd_dt_conf.ddp_mntopts;
                 RETURN(0);
         }
+	case OBD_IOC_START_LFSCK: {
+		rc = lfsck_start(env, mdd->mdd_bottom,
+				 (struct lfsck_start_param *)karg);
+		RETURN(rc);
+	}
+	case OBD_IOC_STOP_LFSCK: {
+		rc = lfsck_stop(env, mdd->mdd_bottom, false);
+		RETURN(rc);
+	}
+	case OBD_IOC_PAUSE_LFSCK: {
+		rc = lfsck_stop(env, mdd->mdd_bottom, true);
+		RETURN(rc);
+	}
         }
 
         /* Below ioctls use obd_ioctl_data */
@@ -1571,7 +1427,7 @@ static int mdd_iocontrol(const struct lu_env *env, struct md_device *m,
 
         switch (cmd) {
         case OBD_IOC_CHANGELOG_REG:
-                rc = mdd_changelog_user_register(mdd, &data->ioc_u32_1);
+                rc = mdd_changelog_user_register(env, mdd, &data->ioc_u32_1);
                 break;
         case OBD_IOC_CHANGELOG_DEREG:
                 rc = mdd_changelog_user_purge(env, mdd, data->ioc_u32_1,
@@ -1585,36 +1441,16 @@ static int mdd_iocontrol(const struct lu_env *env, struct md_device *m,
 }
 
 /* type constructor/destructor: mdd_type_init, mdd_type_fini */
-LU_TYPE_INIT_FINI(mdd, &mdd_thread_key, &mdd_ucred_key, &mdd_capainfo_key,
-                  &mdd_quota_key);
+LU_TYPE_INIT_FINI(mdd, &mdd_thread_key);
 
 const struct md_device_operations mdd_ops = {
-        .mdo_statfs         = mdd_statfs,
-        .mdo_root_get       = mdd_root_get,
-        .mdo_maxsize_get    = mdd_maxsize_get,
-        .mdo_init_capa_ctxt = mdd_init_capa_ctxt,
-        .mdo_update_capa_key= mdd_update_capa_key,
-        .mdo_llog_ctxt_get  = mdd_llog_ctxt_get,
-        .mdo_iocontrol      = mdd_iocontrol,
-#ifdef HAVE_QUOTA_SUPPORT
-        .mdo_quota          = {
-                .mqo_notify      = mdd_quota_notify,
-                .mqo_setup       = mdd_quota_setup,
-                .mqo_cleanup     = mdd_quota_cleanup,
-                .mqo_recovery    = mdd_quota_recovery,
-                .mqo_check       = mdd_quota_check,
-                .mqo_on          = mdd_quota_on,
-                .mqo_off         = mdd_quota_off,
-                .mqo_setinfo     = mdd_quota_setinfo,
-                .mqo_getinfo     = mdd_quota_getinfo,
-                .mqo_setquota    = mdd_quota_setquota,
-                .mqo_getquota    = mdd_quota_getquota,
-                .mqo_getoinfo    = mdd_quota_getoinfo,
-                .mqo_getoquota   = mdd_quota_getoquota,
-                .mqo_invalidate  = mdd_quota_invalidate,
-                .mqo_finvalidate = mdd_quota_finvalidate
-        }
-#endif
+	.mdo_statfs         = mdd_statfs,
+	.mdo_root_get	    = mdd_root_get,
+	.mdo_init_capa_ctxt = mdd_init_capa_ctxt,
+	.mdo_update_capa_key= mdd_update_capa_key,
+	.mdo_llog_ctxt_get  = mdd_llog_ctxt_get,
+	.mdo_iocontrol      = mdd_iocontrol,
+	.mdo_maxeasize_get  = mdd_maxeasize_get,
 };
 
 static struct lu_device_type_operations mdd_device_type_ops = {
@@ -1627,7 +1463,6 @@ static struct lu_device_type_operations mdd_device_type_ops = {
         .ldto_device_alloc = mdd_device_alloc,
         .ldto_device_free  = mdd_device_free,
 
-        .ldto_device_init    = mdd_device_init,
         .ldto_device_fini    = mdd_device_fini
 };
 
@@ -1645,11 +1480,9 @@ static void mdd_key_fini(const struct lu_context *ctx,
                          struct lu_context_key *key, void *data)
 {
         struct mdd_thread_info *info = data;
-        if (info->mti_max_lmm != NULL)
-                OBD_FREE(info->mti_max_lmm, info->mti_max_lmm_size);
-        if (info->mti_max_cookie != NULL)
-                OBD_FREE(info->mti_max_cookie, info->mti_max_cookie_size);
-        mdd_buf_put(&info->mti_big_buf);
+
+	lu_buf_free(&info->mti_big_buf);
+	lu_buf_free(&info->mti_link_buf);
 
         OBD_FREE_PTR(info);
 }
@@ -1657,46 +1490,40 @@ static void mdd_key_fini(const struct lu_context *ctx,
 /* context key: mdd_thread_key */
 LU_CONTEXT_KEY_DEFINE(mdd, LCT_MD_THREAD);
 
-static struct lu_local_obj_desc llod_capa_key = {
-        .llod_name      = CAPA_KEYS,
-        .llod_oid       = MDD_CAPA_KEYS_OID,
-        .llod_is_index  = 0,
-};
+static int __init mdd_mod_init(void)
+{
+	struct lprocfs_static_vars lvars;
+	int rc;
 
-static struct lu_local_obj_desc llod_mdd_orphan = {
-        .llod_name      = orph_index_name,
-        .llod_oid       = MDD_ORPHAN_OID,
-        .llod_is_index  = 1,
-        .llod_feat      = &dt_directory_features,
-};
+	lprocfs_mdd_init_vars(&lvars);
 
-static struct lu_local_obj_desc llod_mdd_root = {
-        .llod_name      = mdd_root_dir_name,
-        .llod_oid       = MDD_ROOT_INDEX_OID,
-        .llod_is_index  = 1,
-        .llod_feat      = &dt_directory_features,
-};
+	rc = lu_kmem_init(mdd_caches);
+	if (rc)
+		return rc;
 
-static int __init mdd_mod_init(void)
-{
-        struct lprocfs_static_vars lvars;
-        lprocfs_mdd_init_vars(&lvars);
+	changelog_orig_logops = llog_osd_ops;
+	changelog_orig_logops.lop_cancel = llog_changelog_cancel;
+	changelog_orig_logops.lop_add = llog_cat_add_rec;
+	changelog_orig_logops.lop_declare_add = llog_cat_declare_add_rec;
 
-        llo_local_obj_register(&llod_capa_key);
-        llo_local_obj_register(&llod_mdd_orphan);
-        llo_local_obj_register(&llod_mdd_root);
+	hsm_actions_logops = llog_osd_ops;
+	hsm_actions_logops.lop_add = llog_cat_add_rec;
+	hsm_actions_logops.lop_declare_add = llog_cat_declare_add_rec;
 
-        return class_register_type(&mdd_obd_device_ops, NULL, lvars.module_vars,
-                                   LUSTRE_MDD_NAME, &mdd_device_type);
+	rc = class_register_type(&mdd_obd_device_ops, NULL, NULL,
+#ifndef HAVE_ONLY_PROCFS_SEQ
+				lvars.module_vars,
+#endif
+				LUSTRE_MDD_NAME, &mdd_device_type);
+	if (rc)
+		lu_kmem_fini(mdd_caches);
+	return rc;
 }
 
 static void __exit mdd_mod_exit(void)
 {
-        llo_local_obj_unregister(&llod_capa_key);
-        llo_local_obj_unregister(&llod_mdd_orphan);
-        llo_local_obj_unregister(&llod_mdd_root);
-
-        class_unregister_type(LUSTRE_MDD_NAME);
+	class_unregister_type(LUSTRE_MDD_NAME);
+	lu_kmem_fini(mdd_caches);
 }
 
 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");