X-Git-Url: https://git.whamcloud.com/?a=blobdiff_plain;f=lustre%2Fofd%2Fofd_dev.c;h=a27e4b07e8d5dd289b0fe7d8a4e105f012cc45bd;hb=001b8dbfacb747f1649a2eb047a5f118ce32fdc7;hp=2bdbde6326670659ddab61851ad33a3578a4dba5;hpb=9761d5c52aeef31a8c3112c3fb3ec9e24b37c800;p=fs%2Flustre-release.git

diff --git a/lustre/ofd/ofd_dev.c b/lustre/ofd/ofd_dev.c
index 2bdbde6..a27e4b0 100644
--- a/lustre/ofd/ofd_dev.c
+++ b/lustre/ofd/ofd_dev.c
@@ -193,6 +193,40 @@ static struct cfg_interop_param ofd_interop_param[] = {
 	{ NULL }
 };
 
+/* Some parameters were moved from ofd to osd and only their
+ * symlinks were kept in ofd by LU-3106. They are:
+ * -writehthrough_cache_enable
+ * -readcache_max_filese
+ * -read_cache_enable
+ * -brw_stats
+ * Since they are not included by the static lprocfs var list,
+ * a pre-check is added for them to avoid "unknown param" error
+ * message confuses the customer. If they are matched in this
+ * check, they will be passed to the osd directly.
+ */
+static bool match_symlink_param(char *param)
+{
+	char *sval;
+	int paramlen;
+
+	if (class_match_param(param, PARAM_OST, &param) == 0) {
+		sval = strchr(param, '=');
+		if (sval != NULL) {
+			paramlen = sval - param;
+			if (strncmp(param, "writethrough_cache_enable",
+				    paramlen) == 0 ||
+			    strncmp(param, "readcache_max_filesize",
+				    paramlen) == 0 ||
+			    strncmp(param, "read_cache_enable",
+				    paramlen) == 0 ||
+			    strncmp(param, "brw_stats", paramlen) == 0)
+				return true;
+		}
+	}
+
+	return false;
+}
+
 /* used by MGS to process specific configurations */
 static int ofd_process_config(const struct lu_env *env, struct lu_device *d,
 			      struct lustre_cfg *cfg)
@@ -240,12 +274,20 @@ static int ofd_process_config(const struct lu_env *env, struct lu_device *d,
 			}
 		}
 
+		if (match_symlink_param(param)) {
+			rc = next->ld_ops->ldo_process_config(env, next, cfg);
+			break;
+		}
+
 		lprocfs_ofd_init_vars(&lvars);
 		rc = class_process_proc_param(PARAM_OST, lvars.obd_vars, cfg,
 					      d->ld_obd);
-		if (rc > 0 || rc == -ENOSYS)
+		if (rc > 0 || rc == -ENOSYS) {
+			CDEBUG(D_CONFIG, "pass param %s down the stack.\n",
+			       param);
 			/* we don't understand; pass it on */
 			rc = next->ld_ops->ldo_process_config(env, next, cfg);
+		}
 		break;
 	}
 	case LCFG_SPTLRPC_CONF: {
@@ -320,7 +362,7 @@ static struct lu_object *ofd_object_alloc(const struct lu_env *env,
 
 	ENTRY;
 
-	OBD_SLAB_ALLOC_PTR_GFP(of, ofd_object_kmem, __GFP_IO);
+	OBD_SLAB_ALLOC_PTR_GFP(of, ofd_object_kmem, GFP_NOFS);
 	if (of != NULL) {
 		struct lu_object	*o;
 		struct lu_object_header *h;
@@ -339,6 +381,38 @@ static struct lu_object *ofd_object_alloc(const struct lu_env *env,
 
 extern int ost_handle(struct ptlrpc_request *req);
 
+static int ofd_lfsck_out_notify(const struct lu_env *env, void *data,
+				enum lfsck_events event)
+{
+	struct ofd_device *ofd = data;
+	struct obd_device *obd = ofd_obd(ofd);
+
+	switch (event) {
+	case LE_LASTID_REBUILDING:
+		CWARN("%s: Found crashed LAST_ID, deny creating new OST-object "
+		      "on the device until the LAST_ID rebuilt successfully.\n",
+		      obd->obd_name);
+		down_write(&ofd->ofd_lastid_rwsem);
+		ofd->ofd_lastid_rebuilding = 1;
+		up_write(&ofd->ofd_lastid_rwsem);
+		break;
+	case LE_LASTID_REBUILT: {
+		down_write(&ofd->ofd_lastid_rwsem);
+		ofd_seqs_free(env, ofd);
+		ofd->ofd_lastid_rebuilding = 0;
+		ofd->ofd_lastid_gen++;
+		up_write(&ofd->ofd_lastid_rwsem);
+		break;
+	}
+	default:
+		CERROR("%s: unknown lfsck event: rc = %d\n",
+		       ofd_obd(ofd)->obd_name, event);
+		return -EINVAL;
+	}
+
+	return 0;
+}
+
 static int ofd_prepare(const struct lu_env *env, struct lu_device *pdev,
 		       struct lu_device *dev)
 {
@@ -351,12 +425,6 @@ static int ofd_prepare(const struct lu_env *env, struct lu_device *pdev,
 
 	ENTRY;
 
-	rc = lu_env_refill((struct lu_env *)env);
-	if (rc != 0) {
-		CERROR("Failure to refill session: '%d'\n", rc);
-		RETURN(rc);
-	}
-
 	info = ofd_info_init(env, NULL);
 	if (info == NULL)
 		RETURN(-EFAULT);
@@ -366,15 +434,21 @@ static int ofd_prepare(const struct lu_env *env, struct lu_device *pdev,
 	if (rc != 0)
 		RETURN(rc);
 
-	rc = lfsck_register(env, ofd->ofd_osd, &ofd->ofd_dt_dev, false);
+	rc = lfsck_register(env, ofd->ofd_osd, ofd->ofd_osd, obd,
+			    ofd_lfsck_out_notify, ofd, false);
 	if (rc != 0) {
 		CERROR("%s: failed to initialize lfsck: rc = %d\n",
 		       obd->obd_name, rc);
 		RETURN(rc);
 	}
 
+	rc = lfsck_register_namespace(env, ofd->ofd_osd, ofd->ofd_namespace);
+	/* The LFSCK instance is registered just now, so it must be there when
+	 * register the namespace to such instance. */
+	LASSERTF(rc == 0, "register namespace failed: rc = %d\n", rc);
+
 	lsp.lsp_start = NULL;
-	lsp.lsp_namespace = ofd->ofd_namespace;
+	lsp.lsp_index_valid = 0;
 	rc = lfsck_start(env, ofd->ofd_osd, &lsp);
 	if (rc != 0) {
 		CWARN("%s: auto trigger paused LFSCK failed: rc = %d\n",
@@ -679,8 +753,8 @@ int ofd_set_info_hdl(struct tgt_session_info *tsi)
 	RETURN(rc);
 }
 
-static int ofd_fiemap_get(const struct lu_env *env, struct ofd_device *ofd,
-			  struct lu_fid *fid, struct ll_user_fiemap *fiemap)
+int ofd_fiemap_get(const struct lu_env *env, struct ofd_device *ofd,
+		   struct lu_fid *fid, struct ll_user_fiemap *fiemap)
 {
 	struct ofd_object	*fo;
 	int			 rc;
@@ -821,7 +895,7 @@ int ofd_get_info_hdl(struct tgt_session_info *tsi)
 	} else if (KEY_IS(KEY_FIEMAP)) {
 		struct ll_fiemap_info_key	*fm_key;
 		struct ll_user_fiemap		*fiemap;
-		struct lu_fid			*fid = &fti->fti_fid;
+		struct lu_fid			*fid;
 
 		req_capsule_extend(tsi->tsi_pill, &RQF_OST_GET_INFO_FIEMAP);
 
@@ -830,6 +904,10 @@ int ofd_get_info_hdl(struct tgt_session_info *tsi)
 		if (rc)
 			RETURN(err_serious(rc));
 
+		fid = &fm_key->oa.o_oi.oi_fid;
+
+		CDEBUG(D_INODE, "get FIEMAP of object "DFID"\n", PFID(fid));
+
 		replylen = fiemap_count_to_size(fm_key->fiemap.fm_extent_count);
 		req_capsule_set_size(tsi->tsi_pill, &RMF_FIEMAP_VAL,
 				     RCL_SERVER, replylen);
@@ -842,12 +920,6 @@ int ofd_get_info_hdl(struct tgt_session_info *tsi)
 		if (fiemap == NULL)
 			RETURN(-ENOMEM);
 
-		rc = ostid_to_fid(fid, &fm_key->oa.o_oi, 0);
-		if (rc != 0)
-			RETURN(rc);
-
-		CDEBUG(D_INODE, "get FIEMAP of object "DFID"\n", PFID(fid));
-
 		*fiemap = fm_key->fiemap;
 		rc = ofd_fiemap_get(tsi->tsi_env, ofd, fid, fiemap);
 
@@ -1069,12 +1141,86 @@ out:
 	return rc;
 }
 
+static int ofd_orphans_destroy(const struct lu_env *env,
+			       struct obd_export *exp,
+			       struct ofd_device *ofd, struct obdo *oa)
+{
+	struct ofd_thread_info	*info	= ofd_info(env);
+	struct lu_fid		*fid	= &info->fti_fid;
+	struct ost_id		*oi	= &oa->o_oi;
+	struct ofd_seq		*oseq;
+	obd_seq 		 seq	= ostid_seq(oi);
+	obd_id			 end_id = ostid_id(oi);
+	obd_id			 last;
+	obd_id			 oid;
+	int			 skip_orphan;
+	int			 rc	= 0;
+
+	ENTRY;
+
+	oseq = ofd_seq_get(ofd, seq);
+	if (oseq == NULL) {
+		CERROR("%s: Can not find seq for "DOSTID"\n",
+		       ofd_name(ofd), POSTID(oi));
+		RETURN(-EINVAL);
+	}
+
+	*fid = oi->oi_fid;
+	last = ofd_seq_last_oid(oseq);
+	oid = last;
+
+	LASSERT(exp != NULL);
+	skip_orphan = !!(exp_connect_flags(exp) & OBD_CONNECT_SKIP_ORPHAN);
+
+	LCONSOLE(D_INFO, "%s: deleting orphan objects from "DOSTID
+		 " to "DOSTID"\n", ofd_name(ofd), seq, end_id + 1, seq, last);
+
+	while (oid > end_id) {
+		rc = fid_set_id(fid, oid);
+		if (unlikely(rc != 0))
+			GOTO(out_put, rc);
+
+		rc = ofd_destroy_by_fid(env, ofd, fid, 1);
+		if (rc != 0 && rc != -ENOENT) /* this is pretty fatal... */
+			CEMERG("%s: error destroying precreated id "DFID
+			       ": rc = %d\n", ofd_name(ofd), PFID(fid), rc);
+
+		oid--;
+		if (!skip_orphan) {
+			ofd_seq_last_oid_set(oseq, oid);
+			/* 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 ((oid & 511) == 0)
+				ofd_seq_last_oid_write(env, ofd, oseq);
+		}
+	}
+
+	CDEBUG(D_HA, "%s: after destroy: set last_id to "DOSTID"\n",
+	       ofd_name(ofd), seq, oid);
+
+	if (!skip_orphan) {
+		rc = ofd_seq_last_oid_write(env, ofd, oseq);
+	} else {
+		/* don't reuse orphan object, return last used objid */
+		ostid_set_id(oi, last);
+		rc = 0;
+	}
+
+	GOTO(out_put, rc);
+
+out_put:
+	ofd_seq_put(env, oseq);
+	return rc;
+}
+
 static int ofd_create_hdl(struct tgt_session_info *tsi)
 {
 	struct ost_body		*repbody;
 	const struct obdo	*oa = &tsi->tsi_ost_body->oa;
 	struct obdo		*rep_oa;
-	struct ofd_device	*ofd = ofd_exp(tsi->tsi_exp);
+	struct obd_export	*exp = tsi->tsi_exp;
+	struct ofd_device	*ofd = ofd_exp(exp);
 	obd_seq			 seq = ostid_seq(&oa->o_oi);
 	obd_id			 oid = ostid_id(&oa->o_oi);
 	struct ofd_seq		*oseq;
@@ -1090,6 +1236,13 @@ static int ofd_create_hdl(struct tgt_session_info *tsi)
 	if (repbody == NULL)
 		RETURN(-ENOMEM);
 
+	down_read(&ofd->ofd_lastid_rwsem);
+	/* Currently, for safe, we do not distinguish which LAST_ID is broken,
+	 * we may do that in the future.
+	 * Return -ENOSPC until the LAST_ID rebuilt. */
+	if (unlikely(ofd->ofd_lastid_rebuilding))
+		GOTO(out_sem, rc = -ENOSPC);
+
 	rep_oa = &repbody->oa;
 	rep_oa->o_oi = oa->o_oi;
 
@@ -1102,7 +1255,7 @@ static int ofd_create_hdl(struct tgt_session_info *tsi)
 	if (IS_ERR(oseq)) {
 		CERROR("%s: Can't find FID Sequence "LPX64": rc = %ld\n",
 		       ofd_name(ofd), seq, PTR_ERR(oseq));
-		RETURN(-EINVAL);
+		GOTO(out_sem, rc = -EINVAL);
 	}
 
 	if ((oa->o_valid & OBD_MD_FLFLAGS) &&
@@ -1121,9 +1274,11 @@ static int ofd_create_hdl(struct tgt_session_info *tsi)
 	/* former ofd_handle_precreate */
 	if ((oa->o_valid & OBD_MD_FLFLAGS) &&
 	    (oa->o_flags & OBD_FL_DELORPHAN)) {
+		exp->exp_filter_data.fed_lastid_gen = ofd->ofd_lastid_gen;
+
 		/* destroy orphans */
 		if (lustre_msg_get_conn_cnt(tgt_ses_req(tsi)->rq_reqmsg) <
-		    tsi->tsi_exp->exp_conn_cnt) {
+		    exp->exp_conn_cnt) {
 			CERROR("%s: dropping old orphan cleanup request\n",
 			       ofd_name(ofd));
 			GOTO(out_nolock, rc = 0);
@@ -1144,7 +1299,7 @@ static int ofd_create_hdl(struct tgt_session_info *tsi)
 			/* FIXME: should reset precreate_next_id on MDS */
 			rc = 0;
 		} else if (diff < 0) {
-			rc = ofd_orphans_destroy(tsi->tsi_env, tsi->tsi_exp,
+			rc = ofd_orphans_destroy(tsi->tsi_env, exp,
 						 ofd, rep_oa);
 			oseq->os_destroys_in_progress = 0;
 		} else {
@@ -1152,9 +1307,15 @@ static int ofd_create_hdl(struct tgt_session_info *tsi)
 			oseq->os_destroys_in_progress = 0;
 		}
 	} else {
+		if (unlikely(exp->exp_filter_data.fed_lastid_gen !=
+			     ofd->ofd_lastid_gen)) {
+			ofd_obd_disconnect(exp);
+			GOTO(out_nolock, rc = -ENOTCONN);
+		}
+
 		mutex_lock(&oseq->os_create_lock);
 		if (lustre_msg_get_conn_cnt(tgt_ses_req(tsi)->rq_reqmsg) <
-		    tsi->tsi_exp->exp_conn_cnt) {
+		    exp->exp_conn_cnt) {
 			CERROR("%s: dropping old precreate request\n",
 			       ofd_name(ofd));
 			GOTO(out, rc = 0);
@@ -1259,7 +1420,7 @@ static int ofd_create_hdl(struct tgt_session_info *tsi)
 		ostid_set_id(&rep_oa->o_oi, ofd_seq_last_oid(oseq));
 	}
 	EXIT;
-	ofd_counter_incr(tsi->tsi_exp, LPROC_OFD_STATS_CREATE,
+	ofd_counter_incr(exp, LPROC_OFD_STATS_CREATE,
 			 tsi->tsi_jobid, 1);
 out:
 	mutex_unlock(&oseq->os_create_lock);
@@ -1268,6 +1429,9 @@ out_nolock:
 		rep_oa->o_valid |= OBD_MD_FLID | OBD_MD_FLGROUP;
 
 	ofd_seq_put(tsi->tsi_env, oseq);
+
+out_sem:
+	up_read(&ofd->ofd_lastid_rwsem);
 	return rc;
 }
 
@@ -1277,6 +1441,8 @@ static int ofd_destroy_hdl(struct tgt_session_info *tsi)
 	struct ost_body		*repbody;
 	struct ofd_device	*ofd = ofd_exp(tsi->tsi_exp);
 	struct ofd_thread_info	*fti = tsi2ofd_info(tsi);
+	struct lu_fid		*fid = &fti->fti_fid;
+	obd_id			 oid;
 	obd_count		 count;
 	int			 rc = 0;
 
@@ -1297,8 +1463,11 @@ static int ofd_destroy_hdl(struct tgt_session_info *tsi)
 		ldlm_request_cancel(tgt_ses_req(tsi), dlm, 0);
 	}
 
+	*fid = body->oa.o_oi.oi_fid;
+	oid = ostid_id(&body->oa.o_oi);
+	LASSERT(oid != 0);
+
 	repbody = req_capsule_server_get(tsi->tsi_pill, &RMF_OST_BODY);
-	repbody->oa.o_oi = body->oa.o_oi;
 
 	/* check that o_misc makes sense */
 	if (body->oa.o_valid & OBD_MD_FLOBJCOUNT)
@@ -1308,37 +1477,39 @@ static int ofd_destroy_hdl(struct tgt_session_info *tsi)
 
 	CDEBUG(D_HA, "%s: Destroy object "DOSTID" count %d\n", ofd_name(ofd),
 	       POSTID(&body->oa.o_oi), count);
+
 	while (count > 0) {
 		int lrc;
 
-		lrc = ostid_to_fid(&fti->fti_fid, &repbody->oa.o_oi, 0);
-		if (lrc != 0) {
-			if (rc == 0)
-				rc = lrc;
-			GOTO(out, rc);
-		}
-		lrc = ofd_destroy_by_fid(tsi->tsi_env, ofd, &fti->fti_fid, 0);
+		lrc = ofd_destroy_by_fid(tsi->tsi_env, ofd, fid, 0);
 		if (lrc == -ENOENT) {
 			CDEBUG(D_INODE,
 			       "%s: destroying non-existent object "DFID"\n",
-			       ofd_name(ofd), PFID(&fti->fti_fid));
+			       ofd_name(ofd), PFID(fid));
 			/* rewrite rc with -ENOENT only if it is 0 */
 			if (rc == 0)
 				rc = lrc;
 		} else if (lrc != 0) {
 			CERROR("%s: error destroying object "DFID": %d\n",
-			       ofd_name(ofd), PFID(&fti->fti_fid),
-			       rc);
+			       ofd_name(ofd), PFID(fid), lrc);
 			rc = lrc;
 		}
+
 		count--;
-		ostid_inc_id(&repbody->oa.o_oi);
+		oid++;
+		lrc = fid_set_id(fid, oid);
+		if (unlikely(lrc != 0 && count > 0))
+			GOTO(out, rc = lrc);
 	}
 
 	ofd_counter_incr(tsi->tsi_exp, LPROC_OFD_STATS_DESTROY,
 			 tsi->tsi_jobid, 1);
+
+	GOTO(out, rc);
+
 out:
-	RETURN(rc);
+	fid_to_ostid(fid, &repbody->oa.o_oi);
+	return rc;
 }
 
 static int ofd_statfs_hdl(struct tgt_session_info *tsi)
@@ -1485,7 +1656,7 @@ static int ofd_punch_hdl(struct tgt_session_info *tsi)
 	}
 
 	rc = ofd_object_punch(tsi->tsi_env, fo, start, end, &info->fti_attr,
-			      ff);
+			      ff, (struct obdo *)oa);
 	if (rc)
 		GOTO(out_put, rc);
 
@@ -1513,7 +1684,6 @@ out:
 	return rc;
 }
 
-
 static int ofd_quotactl(struct tgt_session_info *tsi)
 {
 	struct obd_quotactl	*oqctl, *repoqc;
@@ -1543,6 +1713,302 @@ static int ofd_quotactl(struct tgt_session_info *tsi)
 	RETURN(rc);
 }
 
+/* High priority request handlers for OFD */
+
+/* prolong locks for the current service time of the corresponding
+ * portal (= OST_IO_PORTAL)
+ */
+static inline int prolong_timeout(struct ptlrpc_request *req)
+{
+	struct ptlrpc_service_part *svcpt = req->rq_rqbd->rqbd_svcpt;
+
+	if (AT_OFF)
+		return obd_timeout / 2;
+
+	return max(at_est2timeout(at_get(&svcpt->scp_at_estimate)),
+		   ldlm_timeout);
+}
+
+static int ofd_prolong_one_lock(struct tgt_session_info *tsi,
+				struct ldlm_lock *lock,
+				struct ldlm_extent *extent, int timeout)
+{
+
+	if (lock->l_flags & LDLM_FL_DESTROYED) /* lock already cancelled */
+		return 0;
+
+	/* XXX: never try to grab resource lock here because we're inside
+	 * exp_bl_list_lock; in ldlm_lockd.c to handle waiting list we take
+	 * res lock and then exp_bl_list_lock. */
+
+	if (!(lock->l_flags & LDLM_FL_AST_SENT))
+		/* ignore locks not being cancelled */
+		return 0;
+
+	LDLM_DEBUG(lock, "refreshed for req x"LPU64" ext("LPU64"->"LPU64") "
+			 "to %ds.\n", tgt_ses_req(tsi)->rq_xid, extent->start,
+			 extent->end, timeout);
+
+	/* OK. this is a possible lock the user holds doing I/O
+	 * let's refresh eviction timer for it */
+	ldlm_refresh_waiting_lock(lock, timeout);
+	return 1;
+}
+
+static int ofd_prolong_extent_locks(struct tgt_session_info *tsi,
+				    __u64 start, __u64 end)
+{
+	struct obd_export	*exp = tsi->tsi_exp;
+	struct obdo		*oa  = &tsi->tsi_ost_body->oa;
+	struct ldlm_extent	 extent = {
+		.start = start,
+		.end = end
+	};
+	struct ldlm_lock	*lock;
+	int			 timeout = prolong_timeout(tgt_ses_req(tsi));
+	int			 lock_count = 0;
+
+	ENTRY;
+
+	if (oa->o_valid & OBD_MD_FLHANDLE) {
+		/* mostly a request should be covered by only one lock, try
+		 * fast path. */
+		lock = ldlm_handle2lock(&oa->o_handle);
+		if (lock != NULL) {
+			/* Fast path to check if the lock covers the whole IO
+			 * region exclusively. */
+			if (lock->l_granted_mode == LCK_PW &&
+			    ldlm_extent_contain(&lock->l_policy_data.l_extent,
+						&extent)) {
+				/* bingo */
+				LASSERT(lock->l_export == exp);
+				lock_count = ofd_prolong_one_lock(tsi, lock,
+							     &extent, timeout);
+				LDLM_LOCK_PUT(lock);
+				RETURN(lock_count);
+			}
+			LDLM_LOCK_PUT(lock);
+		}
+	}
+
+	spin_lock_bh(&exp->exp_bl_list_lock);
+	list_for_each_entry(lock, &exp->exp_bl_list, l_exp_list) {
+		LASSERT(lock->l_flags & LDLM_FL_AST_SENT);
+		LASSERT(lock->l_resource->lr_type == LDLM_EXTENT);
+
+		if (!ldlm_res_eq(&tsi->tsi_resid, &lock->l_resource->lr_name))
+			continue;
+
+		if (!ldlm_extent_overlap(&lock->l_policy_data.l_extent,
+					 &extent))
+			continue;
+
+		lock_count += ofd_prolong_one_lock(tsi, lock, &extent, timeout);
+	}
+	spin_unlock_bh(&exp->exp_bl_list_lock);
+
+	RETURN(lock_count);
+}
+
+/**
+ * Returns 1 if the given PTLRPC matches the given LDLM lock, or 0 if it does
+ * not.
+ */
+static int ofd_rw_hpreq_lock_match(struct ptlrpc_request *req,
+				   struct ldlm_lock *lock)
+{
+	struct niobuf_remote	*rnb;
+	struct obd_ioobj	*ioo;
+	ldlm_mode_t		 mode;
+	struct ldlm_extent	 ext;
+	__u32			 opc = lustre_msg_get_opc(req->rq_reqmsg);
+
+	ENTRY;
+
+	ioo = req_capsule_client_get(&req->rq_pill, &RMF_OBD_IOOBJ);
+	LASSERT(ioo != NULL);
+
+	rnb = req_capsule_client_get(&req->rq_pill, &RMF_NIOBUF_REMOTE);
+	LASSERT(rnb != NULL);
+
+	ext.start = rnb->offset;
+	rnb += ioo->ioo_bufcnt - 1;
+	ext.end = rnb->offset + rnb->len - 1;
+
+	LASSERT(lock->l_resource != NULL);
+	if (!ostid_res_name_eq(&ioo->ioo_oid, &lock->l_resource->lr_name))
+		RETURN(0);
+
+	mode = LCK_PW;
+	if (opc == OST_READ)
+		mode |= LCK_PR;
+
+	if (!(lock->l_granted_mode & mode))
+		RETURN(0);
+
+	RETURN(ldlm_extent_overlap(&lock->l_policy_data.l_extent, &ext));
+}
+
+/**
+ * High-priority queue request check for whether the given PTLRPC request
+ * (\a req) is blocking an LDLM lock cancel.
+ *
+ * Returns 1 if the given given PTLRPC request (\a req) is blocking an LDLM lock
+ * cancel, 0 if it is not, and -EFAULT if the request is malformed.
+ *
+ * Only OST_READs, OST_WRITEs and OST_PUNCHes go on the h-p RPC queue.  This
+ * function looks only at OST_READs and OST_WRITEs.
+ */
+static int ofd_rw_hpreq_check(struct ptlrpc_request *req)
+{
+	struct tgt_session_info	*tsi;
+	struct obd_ioobj	*ioo;
+	struct niobuf_remote	*rnb;
+	__u64			 start, end;
+	int			 lock_count;
+
+	ENTRY;
+
+	/* Don't use tgt_ses_info() to get session info, because lock_match()
+	 * can be called while request has no processing thread yet. */
+	tsi = lu_context_key_get(&req->rq_session, &tgt_session_key);
+	LASSERT(tsi != NULL);
+
+	/*
+	 * Use LASSERT below because malformed RPCs should have
+	 * been filtered out in tgt_hpreq_handler().
+	 */
+	ioo = req_capsule_client_get(&req->rq_pill, &RMF_OBD_IOOBJ);
+	LASSERT(ioo != NULL);
+
+	rnb = req_capsule_client_get(&req->rq_pill, &RMF_NIOBUF_REMOTE);
+	LASSERT(rnb != NULL);
+	LASSERT(!(rnb->flags & OBD_BRW_SRVLOCK));
+
+	start = rnb->offset;
+	rnb += ioo->ioo_bufcnt - 1;
+	end = rnb->offset + rnb->len - 1;
+
+	DEBUG_REQ(D_RPCTRACE, req, "%s %s: refresh rw locks: "DFID
+				   " ("LPU64"->"LPU64")\n",
+		  tgt_name(tsi->tsi_tgt), current->comm,
+		  PFID(&tsi->tsi_fid), start, end);
+
+	lock_count = ofd_prolong_extent_locks(tsi, start, end);
+
+	CDEBUG(D_DLMTRACE, "%s: refreshed %u locks timeout for req %p.\n",
+	       tgt_name(tsi->tsi_tgt), lock_count, req);
+
+	RETURN(lock_count > 0);
+}
+
+static void ofd_rw_hpreq_fini(struct ptlrpc_request *req)
+{
+	ofd_rw_hpreq_check(req);
+}
+
+/**
+ * Like tgt_rw_hpreq_lock_match(), but for OST_PUNCH RPCs.
+ */
+static int ofd_punch_hpreq_lock_match(struct ptlrpc_request *req,
+				      struct ldlm_lock *lock)
+{
+	struct tgt_session_info	*tsi;
+
+	/* Don't use tgt_ses_info() to get session info, because lock_match()
+	 * can be called while request has no processing thread yet. */
+	tsi = lu_context_key_get(&req->rq_session, &tgt_session_key);
+	LASSERT(tsi != NULL);
+
+	LASSERT(tsi->tsi_ost_body != NULL);
+	if (tsi->tsi_ost_body->oa.o_valid & OBD_MD_FLHANDLE &&
+	    tsi->tsi_ost_body->oa.o_handle.cookie == lock->l_handle.h_cookie)
+		return 1;
+
+	return 0;
+}
+
+/**
+ * Like ost_rw_hpreq_check(), but for OST_PUNCH RPCs.
+ */
+static int ofd_punch_hpreq_check(struct ptlrpc_request *req)
+{
+	struct tgt_session_info	*tsi;
+	struct obdo		*oa;
+	int			 lock_count;
+
+	ENTRY;
+
+	/* Don't use tgt_ses_info() to get session info, because lock_match()
+	 * can be called while request has no processing thread yet. */
+	tsi = lu_context_key_get(&req->rq_session, &tgt_session_key);
+	LASSERT(tsi != NULL);
+	oa = &tsi->tsi_ost_body->oa;
+
+	LASSERT(!(oa->o_valid & OBD_MD_FLFLAGS &&
+		  oa->o_flags & OBD_FL_SRVLOCK));
+
+	CDEBUG(D_DLMTRACE,
+	       "%s: refresh locks: "LPU64"/"LPU64" ("LPU64"->"LPU64")\n",
+	       tgt_name(tsi->tsi_tgt), tsi->tsi_resid.name[0],
+	       tsi->tsi_resid.name[1], oa->o_size, oa->o_blocks);
+
+	lock_count = ofd_prolong_extent_locks(tsi, oa->o_size, oa->o_blocks);
+
+	CDEBUG(D_DLMTRACE, "%s: refreshed %u locks timeout for req %p.\n",
+	       tgt_name(tsi->tsi_tgt), lock_count, req);
+
+	RETURN(lock_count > 0);
+}
+
+static void ofd_punch_hpreq_fini(struct ptlrpc_request *req)
+{
+	ofd_punch_hpreq_check(req);
+}
+
+struct ptlrpc_hpreq_ops ofd_hpreq_rw = {
+	.hpreq_lock_match	= ofd_rw_hpreq_lock_match,
+	.hpreq_check		= ofd_rw_hpreq_check,
+	.hpreq_fini		= ofd_rw_hpreq_fini
+};
+
+struct ptlrpc_hpreq_ops ofd_hpreq_punch = {
+	.hpreq_lock_match	= ofd_punch_hpreq_lock_match,
+	.hpreq_check		= ofd_punch_hpreq_check,
+	.hpreq_fini		= ofd_punch_hpreq_fini
+};
+
+/** Assign high priority operations to the IO requests */
+static void ofd_hp_brw(struct tgt_session_info *tsi)
+{
+	struct niobuf_remote	*rnb;
+	struct obd_ioobj	*ioo;
+
+	ENTRY;
+
+	ioo = req_capsule_client_get(tsi->tsi_pill, &RMF_OBD_IOOBJ);
+	LASSERT(ioo != NULL); /* must exist after request preprocessing */
+	if (ioo->ioo_bufcnt > 0) {
+		rnb = req_capsule_client_get(tsi->tsi_pill, &RMF_NIOBUF_REMOTE);
+		LASSERT(rnb != NULL); /* must exist after request preprocessing */
+
+		/* no high priority if server lock is needed */
+		if (rnb->flags & OBD_BRW_SRVLOCK)
+			return;
+	}
+	tgt_ses_req(tsi)->rq_ops = &ofd_hpreq_rw;
+}
+
+static void ofd_hp_punch(struct tgt_session_info *tsi)
+{
+	LASSERT(tsi->tsi_ost_body != NULL); /* must exists if we are here */
+	/* no high-priority if server lock is needed */
+	if (tsi->tsi_ost_body->oa.o_valid & OBD_MD_FLFLAGS &&
+	    tsi->tsi_ost_body->oa.o_flags & OBD_FL_SRVLOCK)
+		return;
+	tgt_ses_req(tsi)->rq_ops = &ofd_hpreq_punch;
+}
+
 #define OBD_FAIL_OST_READ_NET	OBD_FAIL_OST_BRW_NET
 #define OBD_FAIL_OST_WRITE_NET	OBD_FAIL_OST_BRW_NET
 #define OST_BRW_READ	OST_READ
@@ -1567,11 +2033,15 @@ TGT_OST_HDL(0		| HABEO_REFERO | MUTABOR,
 TGT_OST_HDL(0		| HABEO_REFERO | MUTABOR,
 					OST_DESTROY,	ofd_destroy_hdl),
 TGT_OST_HDL(0		| HABEO_REFERO,	OST_STATFS,	ofd_statfs_hdl),
-TGT_OST_HDL(HABEO_CORPUS| HABEO_REFERO,	OST_BRW_READ,	tgt_brw_read),
+TGT_OST_HDL_HP(HABEO_CORPUS| HABEO_REFERO,
+					OST_BRW_READ,	tgt_brw_read,
+							ofd_hp_brw),
 /* don't set CORPUS flag for brw_write because -ENOENT may be valid case */
-TGT_OST_HDL(MUTABOR,			OST_BRW_WRITE,	tgt_brw_write),
-TGT_OST_HDL(HABEO_CORPUS| HABEO_REFERO | MUTABOR,
-					OST_PUNCH,	ofd_punch_hdl),
+TGT_OST_HDL_HP(HABEO_CORPUS| MUTABOR,	OST_BRW_WRITE,	tgt_brw_write,
+							ofd_hp_brw),
+TGT_OST_HDL_HP(HABEO_CORPUS| HABEO_REFERO | MUTABOR,
+					OST_PUNCH,	ofd_punch_hdl,
+							ofd_hp_punch),
 TGT_OST_HDL(HABEO_CORPUS| HABEO_REFERO,	OST_SYNC,	ofd_sync_hdl),
 TGT_OST_HDL(0		| HABEO_REFERO,	OST_QUOTACTL,	ofd_quotactl),
 };
@@ -1593,8 +2063,8 @@ static struct tgt_opc_slice ofd_common_slice[] = {
 		.tos_hs		= tgt_dlm_handlers
 	},
 	{
-		.tos_opc_start	= UPDATE_OBJ,
-		.tos_opc_end	= UPDATE_LAST_OPC,
+		.tos_opc_start	= OUT_UPDATE_FIRST_OPC,
+		.tos_opc_end	= OUT_UPDATE_LAST_OPC,
 		.tos_hs		= tgt_out_handlers
 	},
 	{
@@ -1603,6 +2073,11 @@ static struct tgt_opc_slice ofd_common_slice[] = {
 		.tos_hs		= seq_handlers
 	},
 	{
+		.tos_opc_start	= LFSCK_FIRST_OPC,
+		.tos_opc_end	= LFSCK_LAST_OPC,
+		.tos_hs		= tgt_lfsck_handlers
+	},
+	{
 		.tos_hs		= NULL
 	}
 };
@@ -1653,10 +2128,14 @@ static int ofd_init0(const struct lu_env *env, struct ofd_device *m,
 	m->ofd_tot_granted = 0;
 	m->ofd_tot_pending = 0;
 	m->ofd_seq_count = 0;
+	init_waitqueue_head(&m->ofd_inconsistency_thread.t_ctl_waitq);
+	INIT_LIST_HEAD(&m->ofd_inconsistency_list);
+	spin_lock_init(&m->ofd_inconsistency_lock);
 
 	spin_lock_init(&m->ofd_batch_lock);
 	rwlock_init(&obd->u.filter.fo_sptlrpc_lock);
 	sptlrpc_rule_set_init(&obd->u.filter.fo_sptlrpc_rset);
+	init_rwsem(&m->ofd_lastid_rwsem);
 
 	obd->u.filter.fo_fl_oss_capa = 0;
 	CFS_INIT_LIST_HEAD(&obd->u.filter.fo_capa_keys);
@@ -1713,7 +2192,7 @@ static int ofd_init0(const struct lu_env *env, struct ofd_device *m,
 				obd->obd_name, osfs->os_bsize);
 		GOTO(err_fini_stack, rc = -EPROTO);
 	}
-	m->ofd_blockbits = fls(osfs->os_bsize) - 1;
+	m->ofd_blockbits = __fls(osfs->os_bsize) - 1;
 
 	m->ofd_precreate_batch = OFD_PRECREATE_BATCH_DEFAULT;
 	if (osfs->os_bsize * osfs->os_blocks < OFD_PRECREATE_SMALL_FS)
@@ -1756,7 +2235,14 @@ static int ofd_init0(const struct lu_env *env, struct ofd_device *m,
 	if (rc)
 		GOTO(err_fini_lut, rc);
 
+	rc = ofd_start_inconsistency_verification_thread(m);
+	if (rc != 0)
+		GOTO(err_fini_fs, rc);
+
 	RETURN(0);
+
+err_fini_fs:
+	ofd_fs_cleanup(env, m);
 err_fini_lut:
 	tgt_fini(env, &m->ofd_lut);
 err_free_ns:
@@ -1771,16 +2257,20 @@ err_fini_proc:
 
 static void ofd_fini(const struct lu_env *env, struct ofd_device *m)
 {
-	struct obd_device *obd = ofd_obd(m);
-	struct lu_device  *d = &m->ofd_dt_dev.dd_lu_dev;
+	struct obd_device	*obd = ofd_obd(m);
+	struct lu_device	*d   = &m->ofd_dt_dev.dd_lu_dev;
+	struct lfsck_stop	 stop;
 
-	lfsck_stop(env, m->ofd_osd, true);
-	lfsck_degister(env, m->ofd_osd);
+	stop.ls_status = LS_PAUSED;
+	stop.ls_flags = 0;
+	lfsck_stop(env, m->ofd_osd, &stop);
 	target_recovery_fini(obd);
 	obd_exports_barrier(obd);
 	obd_zombie_barrier();
 
 	tgt_fini(env, &m->ofd_lut);
+	ofd_stop_inconsistency_verification_thread(m);
+	lfsck_degister(env, m->ofd_osd);
 	ofd_fs_cleanup(env, m);
 
 	ofd_free_capa_keys(m);
@@ -1794,7 +2284,7 @@ static void ofd_fini(const struct lu_env *env, struct ofd_device *m)
 
 	ofd_stack_fini(env, m, &m->ofd_dt_dev.dd_lu_dev);
 	ofd_procfs_fini(m);
-	LASSERT(cfs_atomic_read(&d->ld_ref) == 0);
+	LASSERT(atomic_read(&d->ld_ref) == 0);
 	server_put_mount(obd->obd_name);
 	EXIT;
 }