Whamcloud - gitweb
LU-2903 tests: calculation of available space
[fs/lustre-release.git] / lustre / ofd / ofd_obd.c
index 29592a9..99de9eb 100644 (file)
@@ -27,7 +27,7 @@
  * Copyright (c) 2009, 2010, Oracle and/or its affiliates. All rights reserved.
  * Use is subject to license terms.
  *
- * Copyright (c) 2011, 2012, Whamcloud, Inc.
+ * Copyright (c) 2012, Intel Corporation.
  */
 /*
  * This file is part of Lustre, http://www.lustre.org/
@@ -80,7 +80,7 @@ static int ofd_export_stats_init(struct ofd_device *ofd,
                GOTO(clean, rc = -ENOMEM);
 
        for (i = 0; i < BRW_LAST; i++)
-               cfs_spin_lock_init(&stats->nid_brw_stats->hist[i].oh_lock);
+               spin_lock_init(&stats->nid_brw_stats->hist[i].oh_lock);
 
        rc = lprocfs_seq_create(stats->nid_proc, "brw_stats", 0644,
                                &ofd_per_nid_stats_fops, stats);
@@ -128,10 +128,11 @@ static int ofd_parse_connect_data(const struct lu_env *env,
                RETURN(0);
 
        CDEBUG(D_RPCTRACE, "%s: cli %s/%p ocd_connect_flags: "LPX64
-              " ocd_version: %x ocd_grant: %d ocd_index: %u\n",
+              " ocd_version: %x ocd_grant: %d ocd_index: %u"
+              " ocd_group %u\n",
               exp->exp_obd->obd_name, exp->exp_client_uuid.uuid, exp,
               data->ocd_connect_flags, data->ocd_version,
-              data->ocd_grant, data->ocd_index);
+              data->ocd_grant, data->ocd_index, data->ocd_group);
 
        if (fed->fed_group != 0 && fed->fed_group != data->ocd_group) {
                CWARN("!!! This export (nid %s) used object group %d "
@@ -145,7 +146,7 @@ static int ofd_parse_connect_data(const struct lu_env *env,
        fed->fed_group = data->ocd_group;
 
        data->ocd_connect_flags &= OST_CONNECT_SUPPORTED;
-       exp->exp_connect_flags = data->ocd_connect_flags;
+       exp->exp_connect_data = *data;
        data->ocd_version = LUSTRE_VERSION_CODE;
 
        /* Kindly make sure the SKIP_ORPHAN flag is from MDS. */
@@ -164,21 +165,14 @@ static int ofd_parse_connect_data(const struct lu_env *env,
                data->ocd_grant_extent = ofd->ofd_dt_conf.ddp_grant_frag >> 10;
        }
 
-       if (exp->exp_connect_flags & OBD_CONNECT_GRANT)
+       if (exp_connect_flags(exp) & OBD_CONNECT_GRANT)
                data->ocd_grant = ofd_grant_connect(env, exp, data->ocd_grant);
 
        if (data->ocd_connect_flags & OBD_CONNECT_INDEX) {
                struct lr_server_data *lsd = &ofd->ofd_lut.lut_lsd;
-               int                    index = lsd->lsd_ost_index;
+               int                    index = lsd->lsd_osd_index;
 
-               if (!(lsd->lsd_feature_compat & OBD_COMPAT_OST)) {
-                       /* this will only happen on the first connect */
-                       lsd->lsd_ost_index = data->ocd_index;
-                       lsd->lsd_feature_compat |= OBD_COMPAT_OST;
-                       /* sync is not needed here as lut_client_add will
-                        * set exp_need_sync flag */
-                       tgt_server_data_update(env, &ofd->ofd_lut, 0);
-               } else if (index != data->ocd_index) {
+               if (index != data->ocd_index) {
                        LCONSOLE_ERROR_MSG(0x136, "Connection from %s to index"
                                           " %u doesn't match actual OST index"
                                           " %u in last_rcvd file, bad "
@@ -187,13 +181,19 @@ static int ofd_parse_connect_data(const struct lu_env *env,
                                           data->ocd_index);
                        RETURN(-EBADF);
                }
+               if (!(lsd->lsd_feature_compat & OBD_COMPAT_OST)) {
+                       /* this will only happen on the first connect */
+                       lsd->lsd_feature_compat |= OBD_COMPAT_OST;
+                       /* sync is not needed here as lut_client_add will
+                        * set exp_need_sync flag */
+                       tgt_server_data_update(env, &ofd->ofd_lut, 0);
+               }
        }
-
        if (OBD_FAIL_CHECK(OBD_FAIL_OST_BRW_SIZE)) {
                data->ocd_brw_size = 65536;
        } else if (data->ocd_connect_flags & OBD_CONNECT_BRW_SIZE) {
                data->ocd_brw_size = min(data->ocd_brw_size,
-                             (__u32)(PTLRPC_MAX_BRW_PAGES << CFS_PAGE_SHIFT));
+                                        (__u32)DT_MAX_BRW_SIZE);
                if (data->ocd_brw_size == 0) {
                        CERROR("%s: cli %s/%p ocd_connect_flags: "LPX64
                               " ocd_version: %x ocd_grant: %d ocd_index: %u "
@@ -237,14 +237,24 @@ static int ofd_parse_connect_data(const struct lu_env *env,
        if (data->ocd_connect_flags & OBD_CONNECT_MAXBYTES)
                data->ocd_maxbytes = ofd->ofd_dt_conf.ddp_maxbytes;
 
-        RETURN(0);
+       if (data->ocd_connect_flags & OBD_CONNECT_PINGLESS) {
+               if (suppress_pings) {
+                       spin_lock(&exp->exp_obd->obd_dev_lock);
+                       list_del_init(&exp->exp_obd_chain_timed);
+                       spin_unlock(&exp->exp_obd->obd_dev_lock);
+               } else {
+                       data->ocd_connect_flags &= ~OBD_CONNECT_PINGLESS;
+               }
+       }
+
+       RETURN(0);
 }
 
 static int ofd_obd_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)
 {
-       struct ofd_device       *ofd = ofd_dev(obd->obd_lu_dev);
+       struct ofd_device       *ofd;
        int                      rc;
 
        ENTRY;
@@ -252,6 +262,8 @@ static int ofd_obd_reconnect(const struct lu_env *env, struct obd_export *exp,
        if (exp == NULL || obd == NULL || cluuid == NULL)
                RETURN(-EINVAL);
 
+       ofd = ofd_dev(obd->obd_lu_dev);
+
        rc = lu_env_refill((struct lu_env *)env);
        if (rc != 0) {
                CERROR("Failure to refill session: '%d'\n", rc);
@@ -273,8 +285,7 @@ static int ofd_obd_connect(const struct lu_env *env, struct obd_export **_exp,
        struct obd_export       *exp;
        struct ofd_device       *ofd;
        struct lustre_handle     conn = { 0 };
-       int                      rc, group;
-
+       int                      rc;
        ENTRY;
 
        if (_exp == NULL || obd == NULL || cluuid == NULL)
@@ -301,7 +312,6 @@ static int ofd_obd_connect(const struct lu_env *env, struct obd_export **_exp,
        if (rc)
                GOTO(out, rc);
 
-       group = data->ocd_group;
        if (obd->obd_replayable) {
                struct tg_export_data *ted = &exp->exp_target_data;
 
@@ -312,14 +322,10 @@ static int ofd_obd_connect(const struct lu_env *env, struct obd_export **_exp,
                        GOTO(out, rc);
                ofd_export_stats_init(ofd, exp, localdata);
        }
-       if (group == 0)
-               GOTO(out, rc = 0);
 
-       /* init new group */
-       if (group > ofd->ofd_max_group) {
-               ofd->ofd_max_group = group;
-               rc = ofd_group_load(env, ofd, group);
-       }
+       CDEBUG(D_HA, "%s: get connection from MDS %d\n", obd->obd_name,
+              data->ocd_group);
+
 out:
        if (rc != 0) {
                class_disconnect(exp);
@@ -366,11 +372,11 @@ static int ofd_init_export(struct obd_export *exp)
 {
        int rc;
 
-       cfs_spin_lock_init(&exp->exp_filter_data.fed_lock);
+       spin_lock_init(&exp->exp_filter_data.fed_lock);
        CFS_INIT_LIST_HEAD(&exp->exp_filter_data.fed_mod_list);
-       cfs_spin_lock(&exp->exp_lock);
+       spin_lock(&exp->exp_lock);
        exp->exp_connecting = 1;
-       cfs_spin_unlock(&exp->exp_lock);
+       spin_unlock(&exp->exp_lock);
 
        /* self-export doesn't need client data and ldlm initialization */
        if (unlikely(obd_uuid_equals(&exp->exp_obd->obd_uuid,
@@ -413,7 +419,7 @@ static int ofd_destroy_export(struct obd_export *exp)
        ofd_grant_discard(exp);
        ofd_fmd_cleanup(exp);
 
-       if (exp->exp_connect_flags & OBD_CONNECT_GRANT_SHRINK) {
+       if (exp_connect_flags(exp) & OBD_CONNECT_GRANT_SHRINK) {
                if (ofd->ofd_tot_granted_clients > 0)
                        ofd->ofd_tot_granted_clients --;
        }
@@ -469,10 +475,10 @@ static int ofd_adapt_sptlrpc_conf(const struct lu_env *env,
 
        sptlrpc_target_update_exp_flavor(obd, &tmp_rset);
 
-       cfs_write_lock(&fo->fo_sptlrpc_lock);
+       write_lock(&fo->fo_sptlrpc_lock);
        sptlrpc_rule_set_free(&fo->fo_sptlrpc_rset);
        fo->fo_sptlrpc_rset = tmp_rset;
-       cfs_write_unlock(&fo->fo_sptlrpc_lock);
+       write_unlock(&fo->fo_sptlrpc_lock);
 
        return 0;
 }
@@ -492,7 +498,7 @@ static int ofd_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 ofd_device       *ofd = ofd_exp(exp);
+       struct ofd_device       *ofd;
        int                      rc = 0;
 
        ENTRY;
@@ -502,6 +508,8 @@ static int ofd_set_info_async(const struct lu_env *env, struct obd_export *exp,
                RETURN(-EINVAL);
        }
 
+       ofd = ofd_exp(exp);
+
        if (KEY_IS(KEY_CAPA_KEY)) {
                rc = ofd_update_capa_key(ofd, val);
                if (rc)
@@ -514,6 +522,7 @@ static int ofd_set_info_async(const struct lu_env *env, struct obd_export *exp,
        } else if (KEY_IS(KEY_GRANT_SHRINK)) {
                struct ost_body *body = val;
 
+               ofd_info_init(env, exp);
                /** handle grant shrink, similar to a read request */
                ofd_grant_prepare_read(env, exp, &body->oa);
        } else {
@@ -529,7 +538,7 @@ static int ofd_get_info(const struct lu_env *env, struct obd_export *exp,
                        __u32 keylen, void *key, __u32 *vallen, void *val,
                        struct lov_stripe_md *lsm)
 {
-       struct ofd_device       *ofd = ofd_exp(exp);
+       struct ofd_device       *ofd;
        int                      rc = 0;
 
        ENTRY;
@@ -539,6 +548,8 @@ static int ofd_get_info(const struct lu_env *env, struct obd_export *exp,
                RETURN(-EINVAL);
        }
 
+       ofd = ofd_exp(exp);
+
        if (KEY_IS(KEY_BLOCKSIZE)) {
                __u32 *blocksize = val;
                if (blocksize) {
@@ -557,12 +568,24 @@ static int ofd_get_info(const struct lu_env *env, struct obd_export *exp,
                *vallen = sizeof(*blocksize_bits);
        } else if (KEY_IS(KEY_LAST_ID)) {
                obd_id *last_id = val;
+               struct ofd_seq *oseq;
+
+               if (val == NULL) {
+                       *vallen = sizeof(obd_id);
+                       RETURN(0);
+               }
+               ofd_info_init(env, exp);
+               oseq = ofd_seq_load(env, ofd,
+                                   (obd_seq)exp->exp_filter_data.fed_group);
+               LASSERT(!IS_ERR(oseq));
                if (last_id) {
-                       if (*vallen < sizeof(*last_id))
+                       if (*vallen < sizeof(*last_id)) {
+                               ofd_seq_put(env, oseq);
                                RETURN(-EOVERFLOW);
-                       *last_id = ofd_last_id(ofd,
-                                              exp->exp_filter_data.fed_group);
+                       }
+                       *last_id = ofd_seq_last_oid(oseq);
                }
+               ofd_seq_put(env, oseq);
                *vallen = sizeof(*last_id);
        } else if (KEY_IS(KEY_FIEMAP)) {
                struct ofd_thread_info          *info;
@@ -606,6 +629,41 @@ static int ofd_get_info(const struct lu_env *env, struct obd_export *exp,
        } else if (KEY_IS(KEY_SYNC_LOCK_CANCEL)) {
                *((__u32 *) val) = ofd->ofd_sync_lock_cancel;
                *vallen = sizeof(__u32);
+       } else if (KEY_IS(KEY_LAST_FID)) {
+               struct lu_env           env;
+               struct ofd_device       *ofd = ofd_exp(exp);
+               struct ofd_seq          *oseq;
+               struct ost_id           *oid = val;
+               int                     rc;
+
+               if (oid == NULL) {
+                       *vallen = sizeof(struct ost_id);
+                       RETURN(0);
+               }
+
+               if (*vallen < sizeof(*oid))
+                       RETURN(-EOVERFLOW);
+
+               rc = lu_env_init(&env, LCT_DT_THREAD);
+               if (rc != 0)
+                       RETURN(rc);
+               ofd_info_init(&env, exp);
+
+               ostid_le_to_cpu(oid, oid);
+               CDEBUG(D_HA, "Get LAST FID for seq "LPX64"\n", oid->oi_seq);
+
+               oseq = ofd_seq_load(&env, ofd, oid->oi_seq);
+               if (IS_ERR(oseq))
+                       GOTO(out_fini, rc = PTR_ERR(oseq));
+
+               CDEBUG(D_HA, "LAST FID is "POSTID"\n", oseq->os_last_oid,
+                      oseq->os_seq);
+
+               *oid = oseq->os_oi;
+               *vallen = sizeof(*oid);
+               ofd_seq_put(&env, oseq);
+out_fini:
+               lu_env_fini(&env);
        } else {
                CERROR("Not supported key %s\n", (char*)key);
                rc = -EOPNOTSUPP;
@@ -620,7 +678,7 @@ int ofd_statfs_internal(const struct lu_env *env, struct ofd_device *ofd,
 {
        int rc;
 
-       cfs_spin_lock(&ofd->ofd_osfs_lock);
+       spin_lock(&ofd->ofd_osfs_lock);
        if (cfs_time_before_64(ofd->ofd_osfs_age, max_age) || max_age == 0) {
                obd_size unstable;
 
@@ -639,7 +697,7 @@ int ofd_statfs_internal(const struct lu_env *env, struct ofd_device *ofd,
                /* record value of inflight counter before running statfs to
                 * compute the diff once statfs is completed */
                unstable = ofd->ofd_osfs_inflight;
-               cfs_spin_unlock(&ofd->ofd_osfs_lock);
+               spin_unlock(&ofd->ofd_osfs_lock);
 
                /* statfs can sleep ... hopefully not for too long since we can
                 * call it fairly often as space fills up */
@@ -647,8 +705,8 @@ int ofd_statfs_internal(const struct lu_env *env, struct ofd_device *ofd,
                if (unlikely(rc))
                        return rc;
 
-               cfs_spin_lock(&ofd->ofd_grant_lock);
-               cfs_spin_lock(&ofd->ofd_osfs_lock);
+               spin_lock(&ofd->ofd_grant_lock);
+               spin_lock(&ofd->ofd_osfs_lock);
                /* calculate how much space was written while we released the
                 * ofd_osfs_lock */
                unstable = ofd->ofd_osfs_inflight - unstable;
@@ -672,7 +730,7 @@ int ofd_statfs_internal(const struct lu_env *env, struct ofd_device *ofd,
                /* similarly, there is some uncertainty on write requests
                 * between prepare & commit */
                ofd->ofd_osfs_unstable += ofd->ofd_tot_pending;
-               cfs_spin_unlock(&ofd->ofd_grant_lock);
+               spin_unlock(&ofd->ofd_grant_lock);
 
                /* finally udpate cached statfs data */
                ofd->ofd_osfs = *osfs;
@@ -681,14 +739,14 @@ int ofd_statfs_internal(const struct lu_env *env, struct ofd_device *ofd,
                ofd->ofd_statfs_inflight--; /* stop tracking */
                if (ofd->ofd_statfs_inflight == 0)
                        ofd->ofd_osfs_inflight = 0;
-               cfs_spin_unlock(&ofd->ofd_osfs_lock);
+               spin_unlock(&ofd->ofd_osfs_lock);
 
                if (from_cache)
                        *from_cache = 0;
        } else {
                /* use cached statfs data */
                *osfs = ofd->ofd_osfs;
-               cfs_spin_unlock(&ofd->ofd_osfs_lock);
+               spin_unlock(&ofd->ofd_osfs_lock);
                if (from_cache)
                        *from_cache = 1;
        }
@@ -724,7 +782,7 @@ static int ofd_statfs(const struct lu_env *env,  struct obd_export *exp,
 
        /* The QoS code on the MDS does not care about space reserved for
         * precreate, so take it out. */
-       if (exp->exp_connect_flags & OBD_CONNECT_MDS) {
+       if (exp_connect_flags(exp) & OBD_CONNECT_MDS) {
                struct filter_export_data *fed;
 
                fed = &obd->obd_self_export->exp_filter_data;
@@ -739,7 +797,7 @@ static int ofd_statfs(const struct lu_env *env,  struct obd_export *exp,
               osfs->os_files, osfs->os_ffree, osfs->os_state);
 
        if (OBD_FAIL_CHECK_VALUE(OBD_FAIL_OST_ENOINO,
-                                ofd->ofd_lut.lut_lsd.lsd_ost_index))
+                                ofd->ofd_lut.lut_lsd.lsd_osd_index))
                osfs->os_ffree = 0;
 
        /* OS_STATE_READONLY can be set by OSD already */
@@ -759,7 +817,7 @@ static int ofd_statfs(const struct lu_env *env,  struct obd_export *exp,
        }
 
        if (OBD_FAIL_CHECK_VALUE(OBD_FAIL_OST_ENOSPC,
-                                ofd->ofd_lut.lut_lsd.lsd_ost_index))
+                                ofd->ofd_lut.lut_lsd.lsd_osd_index))
                osfs->os_bfree = osfs->os_bavail = 2;
 
        EXIT;
@@ -961,7 +1019,7 @@ static int ofd_destroy_by_fid(const struct lu_env *env,
        rc = ldlm_cli_enqueue_local(ofd->ofd_namespace, &info->fti_resid,
                                    LDLM_EXTENT, &policy, LCK_PW, &flags,
                                    ldlm_blocking_ast, ldlm_completion_ast,
-                                   NULL, NULL, 0, NULL, &lockh);
+                                   NULL, NULL, 0, LVB_T_NONE, NULL, &lockh);
 
        /* We only care about the side-effects, just drop the lock. */
        if (rc == ELDLM_OK)
@@ -1058,15 +1116,24 @@ static int ofd_orphans_destroy(const struct lu_env *env,
        int                      skip_orphan;
        int                      rc = 0;
        struct ost_id            oi = oa->o_oi;
+       struct ofd_seq          *oseq;
 
        ENTRY;
 
+       oseq = ofd_seq_get(ofd, oa->o_seq);
+       if (oseq == NULL) {
+               CERROR("%s: Can not find seq for "LPU64":"LPU64"\n",
+                      ofd_name(ofd), oa->o_seq, oa->o_id);
+               RETURN(-EINVAL);
+       }
+
        LASSERT(exp != NULL);
-       skip_orphan = !!(exp->exp_connect_flags & OBD_CONNECT_SKIP_ORPHAN);
+       skip_orphan = !!(exp_connect_flags(exp) & OBD_CONNECT_SKIP_ORPHAN);
 
-       last = ofd_last_id(ofd, oa->o_seq);
-       CWARN("%s: deleting orphan objects from "LPU64" to "LPU64"\n",
-             ofd_obd(ofd)->obd_name, oa->o_id + 1, last);
+       last = ofd_seq_last_oid(oseq);
+       LCONSOLE_INFO("%s: deleting orphan objects from "LPX64":"LPU64
+                     " to "LPU64"\n", ofd_name(ofd), oa->o_seq,
+                     oa->o_id + 1, last);
 
        for (oi.oi_id = last; oi.oi_id > oa->o_id; oi.oi_id--) {
                fid_ostid_unpack(&info->fti_fid, &oi, 0);
@@ -1075,23 +1142,24 @@ static int ofd_orphans_destroy(const struct lu_env *env,
                        CEMERG("error destroying precreated id "LPU64": %d\n",
                               oi.oi_id, rc);
                if (!skip_orphan) {
-                       ofd_last_id_set(ofd, oi.oi_id - 1, oa->o_seq);
+                       ofd_seq_last_oid_set(oseq, oi.oi_id - 1);
                        /* update last_id on disk periodically so that if we
                         * restart * we don't need to re-scan all of the just
                         * deleted objects. */
                        if ((oi.oi_id & 511) == 0)
-                               ofd_last_id_write(env, ofd, oa->o_seq);
+                               ofd_seq_last_oid_write(env, ofd, oseq);
                }
        }
        CDEBUG(D_HA, "%s: after destroy: set last_objids["LPU64"] = "LPU64"\n",
               ofd_obd(ofd)->obd_name, oa->o_seq, oa->o_id);
        if (!skip_orphan) {
-               rc = ofd_last_id_write(env, ofd, oa->o_seq);
+               rc = ofd_seq_last_oid_write(env, ofd, oseq);
        } else {
                /* don't reuse orphan object, return last used objid */
                oa->o_id = last;
                rc = 0;
        }
+       ofd_seq_put(env, oseq);
        RETURN(rc);
 }
 
@@ -1101,7 +1169,10 @@ int ofd_create(const struct lu_env *env, struct obd_export *exp,
 {
        struct ofd_device       *ofd = ofd_exp(exp);
        struct ofd_thread_info  *info;
-       int                      rc = 0, diff;
+       obd_seq                 seq = oa->o_seq;
+       struct ofd_seq          *oseq;
+       int                     rc = 0, diff;
+       int                     sync_trans = 0;
 
        ENTRY;
 
@@ -1112,14 +1183,21 @@ int ofd_create(const struct lu_env *env, struct obd_export *exp,
        LASSERT(oa->o_valid & OBD_MD_FLGROUP);
 
        CDEBUG(D_INFO, "ofd_create(oa->o_seq="LPU64",oa->o_id="LPU64")\n",
-              oa->o_seq, oa->o_id);
+              seq, oa->o_id);
+
+       oseq = ofd_seq_load(env, ofd, seq);
+       if (IS_ERR(oseq)) {
+               CERROR("%s: Can't find FID Sequence "LPX64": rc = %ld\n",
+                      ofd_name(ofd), seq, PTR_ERR(oseq));
+               RETURN(-EINVAL);
+       }
 
        if ((oa->o_valid & OBD_MD_FLFLAGS) &&
            (oa->o_flags & OBD_FL_RECREATE_OBJS)) {
                if (!ofd_obd(ofd)->obd_recovering ||
-                   oa->o_id > ofd_last_id(ofd, oa->o_seq)) {
+                   oa->o_id > ofd_seq_last_oid(oseq)) {
                        CERROR("recreate objid "LPU64" > last id "LPU64"\n",
-                                       oa->o_id, ofd_last_id(ofd, oa->o_seq));
+                                       oa->o_id, ofd_seq_last_oid(oseq));
                        GOTO(out_nolock, rc = -EINVAL);
                }
                /* do nothing because we create objects during first write */
@@ -1131,42 +1209,59 @@ int ofd_create(const struct lu_env *env, struct obd_export *exp,
                /* destroy orphans */
                if (oti->oti_conn_cnt < exp->exp_conn_cnt) {
                        CERROR("%s: dropping old orphan cleanup request\n",
-                              ofd_obd(ofd)->obd_name);
+                              ofd_name(ofd));
                        GOTO(out_nolock, rc = 0);
                }
                /* This causes inflight precreates to abort and drop lock */
-               cfs_set_bit(oa->o_seq, &ofd->ofd_destroys_in_progress);
-               cfs_mutex_lock(&ofd->ofd_create_locks[oa->o_seq]);
-               if (!cfs_test_bit(oa->o_seq, &ofd->ofd_destroys_in_progress)) {
+               oseq->os_destroys_in_progress = 1;
+               mutex_lock(&oseq->os_create_lock);
+               if (!oseq->os_destroys_in_progress) {
                        CERROR("%s:["LPU64"] destroys_in_progress already cleared\n",
                               exp->exp_obd->obd_name, oa->o_seq);
                        GOTO(out, rc = 0);
                }
-               diff = oa->o_id - ofd_last_id(ofd, oa->o_seq);
+               diff = oa->o_id - ofd_seq_last_oid(oseq);
                CDEBUG(D_HA, "ofd_last_id() = "LPU64" -> diff = %d\n",
-                      ofd_last_id(ofd, oa->o_seq), diff);
+                       ofd_seq_last_oid(oseq), diff);
                if (-diff > OST_MAX_PRECREATE) {
                        /* FIXME: should reset precreate_next_id on MDS */
                        rc = 0;
                } else if (diff < 0) {
                        rc = ofd_orphans_destroy(env, exp, ofd, oa);
-                       cfs_clear_bit(oa->o_seq, &ofd->ofd_destroys_in_progress);
+                       oseq->os_destroys_in_progress = 0;
                } else {
                        /* XXX: Used by MDS for the first time! */
-                       cfs_clear_bit(oa->o_seq, &ofd->ofd_destroys_in_progress);
+                       oseq->os_destroys_in_progress = 0;
                }
        } else {
-               cfs_mutex_lock(&ofd->ofd_create_locks[oa->o_seq]);
+               mutex_lock(&oseq->os_create_lock);
                if (oti->oti_conn_cnt < exp->exp_conn_cnt) {
                        CERROR("%s: dropping old precreate request\n",
-                              ofd_obd(ofd)->obd_name);
+                               ofd_obd(ofd)->obd_name);
                        GOTO(out, rc = 0);
                }
-               /* only precreate if group == 0 and o_id is specfied */
-               if (!fid_seq_is_mdt(oa->o_seq) || oa->o_id == 0) {
+               /* only precreate if seq is 0, IDIF or normal and also o_id
+                * must be specfied */
+               if ((!fid_seq_is_mdt(oa->o_seq) &&
+                    !fid_seq_is_norm(oa->o_seq) &&
+                    !fid_seq_is_idif(oa->o_seq)) || oa->o_id == 0) {
                        diff = 1; /* shouldn't we create this right now? */
                } else {
-                       diff = oa->o_id - ofd_last_id(ofd, oa->o_seq);
+                       diff = oa->o_id - ofd_seq_last_oid(oseq);
+                       /* Do sync create if the seq is about to used up */
+                       if (fid_seq_is_idif(oa->o_seq) ||
+                           fid_seq_is_mdt0(oa->o_seq)) {
+                               if (unlikely(oa->o_id >= IDIF_MAX_OID - 1))
+                                       sync_trans = 1;
+                       } else if (fid_seq_is_norm(oa->o_seq)) {
+                               if (unlikely(oa->o_id >=
+                                            LUSTRE_DATA_SEQ_MAX_WIDTH - 1))
+                                       sync_trans = 1;
+                       } else {
+                               CERROR("%s : invalid o_seq "LPX64": rc = %d\n",
+                                      ofd_name(ofd), oa->o_seq, -EINVAL);
+                               GOTO(out, rc = -EINVAL);
+                       }
                }
        }
        if (diff > 0) {
@@ -1182,15 +1277,15 @@ int ofd_create(const struct lu_env *env, struct obd_export *exp,
                                              ofd_obd(ofd)->obd_self_export,
                                              &diff);
                        if (rc) {
-                               CDEBUG(D_HA, "%s: failed to acquire grant space"
-                                      "for precreate (%d)\n",
-                                      ofd_obd(ofd)->obd_name, diff);
+                               CDEBUG(D_HA, "%s: failed to acquire grant "
+                                      "space for precreate (%d): rc = %d\n",
+                                      ofd_name(ofd), diff, rc);
                                diff = 0;
                        }
                }
 
                while (diff > 0) {
-                       next_id = ofd_last_id(ofd, oa->o_seq) + 1;
+                       next_id = ofd_seq_last_oid(oseq) + 1;
                        count = ofd_precreate_batch(ofd, diff);
 
                        CDEBUG(D_HA, "%s: reserve %d objects in group "LPU64
@@ -1204,10 +1299,10 @@ int ofd_create(const struct lu_env *env, struct obd_export *exp,
                                              created, diff + created,
                                              created / DISK_TIMEOUT);
                                break;
-               }
+                       }
 
                        rc = ofd_precreate_objects(env, ofd, next_id,
-                                                  oa->o_seq, count);
+                                                  oseq, count, sync_trans);
                        if (rc > 0) {
                                created += rc;
                                diff -= rc;
@@ -1215,16 +1310,15 @@ int ofd_create(const struct lu_env *env, struct obd_export *exp,
                                break;
                        }
                }
-               if (created > 0) {
+               if (created > 0)
                        /* some objects got created, we can return
                         * them, even if last creation failed */
-                       oa->o_id = ofd_last_id(ofd, oa->o_seq);
                        rc = 0;
-               } else {
-                       CERROR("unable to precreate: %d\n", rc);
-                       oa->o_id = ofd_last_id(ofd, oa->o_seq);
-               }
+               else
+                       CERROR("%s: unable to precreate: rc = %d\n",
+                              ofd_name(ofd), rc);
 
+               oa->o_id = ofd_seq_last_oid(oseq);
                oa->o_valid |= OBD_MD_FLID | OBD_MD_FLGROUP;
 
                if (!(oa->o_valid & OBD_MD_FLFLAGS) ||
@@ -1235,14 +1329,15 @@ int ofd_create(const struct lu_env *env, struct obd_export *exp,
 
        ofd_info2oti(info, oti);
 out:
-       cfs_mutex_unlock(&ofd->ofd_create_locks[oa->o_seq]);
+       mutex_unlock(&oseq->os_create_lock);
 out_nolock:
        if (rc == 0 && ea != NULL) {
                struct lov_stripe_md *lsm = *ea;
 
                lsm->lsm_object_id = oa->o_id;
        }
-       return rc;
+       ofd_seq_put(env, oseq);
+       RETURN(rc);
 }
 
 int ofd_getattr(const struct lu_env *env, struct obd_export *exp,
@@ -1456,23 +1551,6 @@ out:
        return !!rc;
 }
 
-static int ofd_obd_notify(struct obd_device *obd, struct obd_device *unused,
-                         enum obd_notify_event ev, void *data)
-{
-       switch (ev) {
-       case OBD_NOTIFY_CONFIG:
-               LASSERT(obd->obd_no_conn);
-               cfs_spin_lock(&obd->obd_dev_lock);
-               obd->obd_no_conn = 0;
-               cfs_spin_unlock(&obd->obd_dev_lock);
-               break;
-       default:
-               CDEBUG(D_INFO, "%s: Unhandled notification %#x\n",
-                      obd->obd_name, ev);
-       }
-       return 0;
-}
-
 /*
  * Handle quota control requests to consult current usage/limit.
  *
@@ -1526,6 +1604,5 @@ struct obd_ops ofd_obd_ops = {
        .o_precleanup           = ofd_precleanup,
        .o_ping                 = ofd_ping,
        .o_health_check         = ofd_health_check,
-       .o_notify               = ofd_obd_notify,
        .o_quotactl             = ofd_quotactl,
 };