X-Git-Url: https://git.whamcloud.com/?a=blobdiff_plain;f=lustre%2Ftarget%2Ftgt_lastrcvd.c;h=cbd9258be5847193203bdec3c5e82d20a7fe2fd3;hb=HEAD;hp=50147d313acc414babbd957b5c4364b009a32383;hpb=34bdd30ec6a2010dfd98c5110487df210d03815d;p=fs%2Flustre-release.git

diff --git a/lustre/target/tgt_lastrcvd.c b/lustre/target/tgt_lastrcvd.c
index 50147d3..8801607 100644
--- a/lustre/target/tgt_lastrcvd.c
+++ b/lustre/target/tgt_lastrcvd.c
@@ -15,11 +15,7 @@
  *
  * You should have received a copy of the GNU General Public License
  * version 2 along with this program; If not, see
- * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
- *
- * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
- * CA 95054 USA or visit www.sun.com if you need additional information or
- * have any questions.
+ * http://www.gnu.org/licenses/gpl-2.0.html
  *
  * GPL HEADER END
  */
@@ -27,11 +23,10 @@
  * Copyright (c) 2009, 2010, Oracle and/or its affiliates. All rights reserved.
  * Use is subject to license terms.
  *
- * Copyright (c) 2011, 2014, Intel Corporation.
+ * Copyright (c) 2011, 2017, Intel Corporation.
  */
 /*
  * This file is part of Lustre, http://www.lustre.org/
- * Lustre is a trademark of Sun Microsystems, Inc.
  *
  * Lustre Unified Target
  * These are common function to work with last_received file
@@ -44,6 +39,346 @@
 
 #include "tgt_internal.h"
 
+/** version recovery epoch */
+#define LR_EPOCH_BITS	32
+
+/* Allocate a bitmap for a chunk of reply data slots */
+static int tgt_bitmap_chunk_alloc(struct lu_target *lut, int chunk)
+{
+	unsigned long *bm;
+
+	OBD_ALLOC_LARGE(bm, BITS_TO_LONGS(LUT_REPLY_SLOTS_PER_CHUNK) *
+			sizeof(long));
+	if (bm == NULL)
+		return -ENOMEM;
+
+	spin_lock(&lut->lut_client_bitmap_lock);
+
+	if (lut->lut_reply_bitmap[chunk] != NULL) {
+		/* someone else already allocated the bitmap for this chunk */
+		spin_unlock(&lut->lut_client_bitmap_lock);
+		OBD_FREE_LARGE(bm, BITS_TO_LONGS(LUT_REPLY_SLOTS_PER_CHUNK) *
+			 sizeof(long));
+		return 0;
+	}
+
+	lut->lut_reply_bitmap[chunk] = bm;
+
+	spin_unlock(&lut->lut_client_bitmap_lock);
+
+	return 0;
+}
+
+/* Look for an available reply data slot in the bitmap
+ * of the target @lut
+ * Allocate bitmap chunk when first used
+ * XXX algo could be improved if this routine limits performance
+ */
+static int tgt_find_free_reply_slot(struct lu_target *lut)
+{
+	unsigned long *bmp;
+	int chunk = 0;
+	int rc;
+	int b;
+
+	for (chunk = 0; chunk < LUT_REPLY_SLOTS_MAX_CHUNKS; chunk++) {
+		/* allocate the bitmap chunk if necessary */
+		if (unlikely(lut->lut_reply_bitmap[chunk] == NULL)) {
+			rc = tgt_bitmap_chunk_alloc(lut, chunk);
+			if (rc != 0)
+				return rc;
+		}
+		bmp = lut->lut_reply_bitmap[chunk];
+
+		/* look for an available slot in this chunk */
+		do {
+			b = find_first_zero_bit(bmp, LUT_REPLY_SLOTS_PER_CHUNK);
+			if (b >= LUT_REPLY_SLOTS_PER_CHUNK)
+				break;
+
+			/* found one */
+			if (test_and_set_bit(b, bmp) == 0)
+				return chunk * LUT_REPLY_SLOTS_PER_CHUNK + b;
+		} while (true);
+	}
+
+	return -ENOSPC;
+}
+
+/* Mark the reply data slot @idx 'used' in the corresponding bitmap chunk
+ * of the target @lut
+ * Allocate the bitmap chunk if necessary
+ */
+static int tgt_set_reply_slot(struct lu_target *lut, int idx)
+{
+	int chunk;
+	int b;
+	int rc;
+
+	chunk = idx / LUT_REPLY_SLOTS_PER_CHUNK;
+	b = idx % LUT_REPLY_SLOTS_PER_CHUNK;
+
+	LASSERT(chunk < LUT_REPLY_SLOTS_MAX_CHUNKS);
+	LASSERT(b < LUT_REPLY_SLOTS_PER_CHUNK);
+
+	/* allocate the bitmap chunk if necessary */
+	if (unlikely(lut->lut_reply_bitmap[chunk] == NULL)) {
+		rc = tgt_bitmap_chunk_alloc(lut, chunk);
+		if (rc != 0)
+			return rc;
+	}
+
+	/* mark the slot 'used' in this chunk */
+	if (test_and_set_bit(b, lut->lut_reply_bitmap[chunk]) != 0) {
+		CERROR("%s: slot %d already set in bitmap\n",
+		       tgt_name(lut), idx);
+		return -EALREADY;
+	}
+
+	return 0;
+}
+
+
+/* Mark the reply data slot @idx 'unused' in the corresponding bitmap chunk
+ * of the target @lut
+ */
+static int tgt_clear_reply_slot(struct lu_target *lut, int idx)
+{
+	int chunk;
+	int b;
+
+	if (lut->lut_obd->obd_stopping)
+		/*
+		 * in case of failover keep the bit set in order to
+		 * avoid overwriting slots in reply_data which might
+		 * be required by resent rpcs
+		 */
+		return 0;
+	chunk = idx / LUT_REPLY_SLOTS_PER_CHUNK;
+	b = idx % LUT_REPLY_SLOTS_PER_CHUNK;
+
+	LASSERT(chunk < LUT_REPLY_SLOTS_MAX_CHUNKS);
+	LASSERT(b < LUT_REPLY_SLOTS_PER_CHUNK);
+
+	if (lut->lut_reply_bitmap[chunk] == NULL) {
+		CERROR("%s: slot %d not allocated\n",
+		       tgt_name(lut), idx);
+		return -ENOENT;
+	}
+
+	if (test_and_clear_bit(b, lut->lut_reply_bitmap[chunk]) == 0) {
+		CERROR("%s: slot %d already clear in bitmap\n",
+		       tgt_name(lut), idx);
+		return -EALREADY;
+	}
+
+	return 0;
+}
+
+
+/* Read header of reply_data file of target @tgt into structure @lrh */
+static int tgt_reply_header_read(const struct lu_env *env,
+				 struct lu_target *tgt,
+				 struct lsd_reply_header *lrh)
+{
+	int			 rc;
+	struct lsd_reply_header	 buf;
+	struct tgt_thread_info	*tti = tgt_th_info(env);
+
+	tti->tti_off = 0;
+	tti->tti_buf.lb_buf = &buf;
+	tti->tti_buf.lb_len = sizeof(buf);
+
+	rc = dt_record_read(env, tgt->lut_reply_data, &tti->tti_buf,
+			    &tti->tti_off);
+	if (rc != 0)
+		return rc;
+
+	lrh->lrh_magic = le32_to_cpu(buf.lrh_magic);
+	lrh->lrh_header_size = le32_to_cpu(buf.lrh_header_size);
+	lrh->lrh_reply_size = le32_to_cpu(buf.lrh_reply_size);
+
+	CDEBUG(D_HA, "%s: read %s header. magic=0x%08x "
+	       "header_size=%d reply_size=%d\n",
+		tgt->lut_obd->obd_name, REPLY_DATA,
+		lrh->lrh_magic, lrh->lrh_header_size, lrh->lrh_reply_size);
+
+	return 0;
+}
+
+/* Write header into replay_data file of target @tgt from structure @lrh */
+static int tgt_reply_header_write(const struct lu_env *env,
+				  struct lu_target *tgt,
+				  struct lsd_reply_header *lrh)
+{
+	int			 rc;
+	struct lsd_reply_header	 buf;
+	struct tgt_thread_info	*tti = tgt_th_info(env);
+	struct thandle		*th;
+	struct dt_object	*dto;
+
+	CDEBUG(D_HA, "%s: write %s header. magic=0x%08x "
+	       "header_size=%d reply_size=%d\n",
+		tgt->lut_obd->obd_name, REPLY_DATA,
+		lrh->lrh_magic, lrh->lrh_header_size, lrh->lrh_reply_size);
+
+	if (tgt->lut_bottom->dd_rdonly)
+		RETURN(0);
+
+	buf.lrh_magic = cpu_to_le32(lrh->lrh_magic);
+	buf.lrh_header_size = cpu_to_le32(lrh->lrh_header_size);
+	buf.lrh_reply_size = cpu_to_le32(lrh->lrh_reply_size);
+
+	th = dt_trans_create(env, tgt->lut_bottom);
+	if (IS_ERR(th))
+		return PTR_ERR(th);
+	th->th_sync = 1;
+
+	tti->tti_off = 0;
+	tti->tti_buf.lb_buf = &buf;
+	tti->tti_buf.lb_len = sizeof(buf);
+
+	rc = dt_declare_record_write(env, tgt->lut_reply_data,
+				     &tti->tti_buf, tti->tti_off, th);
+	if (rc)
+		GOTO(out, rc);
+
+	rc = dt_trans_start(env, tgt->lut_bottom, th);
+	if (rc)
+		GOTO(out, rc);
+
+	dto = dt_object_locate(tgt->lut_reply_data, th->th_dev);
+	rc = dt_record_write(env, dto, &tti->tti_buf, &tti->tti_off, th);
+out:
+	dt_trans_stop(env, tgt->lut_bottom, th);
+	return rc;
+}
+
+/* Write the reply data @lrd into reply_data file of target @tgt
+ * at offset @off
+ */
+static int tgt_reply_data_write(const struct lu_env *env, struct lu_target *tgt,
+				struct lsd_reply_data *lrd, loff_t off,
+				struct thandle *th)
+{
+	struct tgt_thread_info *tti = tgt_th_info(env);
+	struct lsd_reply_data *buf = &tti->tti_lrd;
+	struct lsd_reply_header *lrh = &tgt->lut_reply_header;
+	struct dt_object *dto;
+
+	lrd->lrd_result = ptlrpc_status_hton(lrd->lrd_result);
+
+	buf->lrd_transno	 = cpu_to_le64(lrd->lrd_transno);
+	buf->lrd_xid		 = cpu_to_le64(lrd->lrd_xid);
+	buf->lrd_data		 = cpu_to_le64(lrd->lrd_data);
+	buf->lrd_result		 = cpu_to_le32(lrd->lrd_result);
+	buf->lrd_client_gen	 = cpu_to_le32(lrd->lrd_client_gen);
+
+	lrd->lrd_result = ptlrpc_status_ntoh(lrd->lrd_result);
+
+	if (lrh->lrh_magic > LRH_MAGIC_V1)
+		buf->lrd_batch_idx = cpu_to_le32(lrd->lrd_batch_idx);
+
+	tti->tti_off = off;
+	tti->tti_buf.lb_buf = buf;
+	tti->tti_buf.lb_len = lrh->lrh_reply_size;
+
+	dto = dt_object_locate(tgt->lut_reply_data, th->th_dev);
+	return dt_record_write(env, dto, &tti->tti_buf, &tti->tti_off, th);
+}
+
+/* Read the reply data from reply_data file of target @tgt at offset @off
+ * into structure @lrd
+ */
+static int tgt_reply_data_read(const struct lu_env *env, struct lu_target *tgt,
+			       struct lsd_reply_data *lrd, loff_t off,
+			       struct lsd_reply_header *lrh)
+{
+	struct tgt_thread_info *tti = tgt_th_info(env);
+	struct lsd_reply_data *buf = &tti->tti_lrd;
+	int rc;
+
+	tti->tti_off = off;
+	tti->tti_buf.lb_buf = buf;
+	tti->tti_buf.lb_len = lrh->lrh_reply_size;
+
+	rc = dt_record_read(env, tgt->lut_reply_data, &tti->tti_buf,
+			    &tti->tti_off);
+	if (rc != 0)
+		return rc;
+
+	lrd->lrd_transno = le64_to_cpu(buf->lrd_transno);
+	lrd->lrd_xid = le64_to_cpu(buf->lrd_xid);
+	lrd->lrd_data = le64_to_cpu(buf->lrd_data);
+	lrd->lrd_result = le32_to_cpu(buf->lrd_result);
+	lrd->lrd_client_gen = le32_to_cpu(buf->lrd_client_gen);
+
+	if (lrh->lrh_magic > LRH_MAGIC_V1)
+		lrd->lrd_batch_idx = le32_to_cpu(buf->lrd_batch_idx);
+	else
+		lrd->lrd_batch_idx = 0;
+
+	return 0;
+}
+
+/* Free the in-memory reply data structure @trd and release
+ * the corresponding slot in the reply_data file of target @lut
+ * Called with ted_lcd_lock held
+ */
+static void tgt_free_reply_data(struct lu_target *lut,
+				struct tg_export_data *ted,
+				struct tg_reply_data *trd)
+{
+	CDEBUG(D_TRACE, "%s: free reply data %p: xid %llu, transno %llu, "
+	       "client gen %u, slot idx %d\n",
+	       lut == NULL ? "" : tgt_name(lut), trd, trd->trd_reply.lrd_xid,
+	       trd->trd_reply.lrd_transno, trd->trd_reply.lrd_client_gen,
+	       trd->trd_index);
+
+	LASSERT(mutex_is_locked(&ted->ted_lcd_lock));
+
+	list_del(&trd->trd_list);
+	ted->ted_reply_cnt--;
+	if (lut != NULL && trd->trd_index != TRD_INDEX_MEMORY)
+		tgt_clear_reply_slot(lut, trd->trd_index);
+	OBD_FREE_PTR(trd);
+}
+
+/* Release the reply data @trd from target @lut
+ * The reply data with the highest transno for this export
+ * is retained to ensure correctness of target recovery
+ * Called with ted_lcd_lock held
+ */
+static void tgt_release_reply_data(struct lu_target *lut,
+				   struct tg_export_data *ted,
+				   struct tg_reply_data *trd)
+{
+	CDEBUG(D_TRACE, "%s: release reply data %p: xid %llu, transno %llu, "
+	       "client gen %u, slot idx %d\n",
+	       lut == NULL ? "" : tgt_name(lut), trd, trd->trd_reply.lrd_xid,
+	       trd->trd_reply.lrd_transno, trd->trd_reply.lrd_client_gen,
+	       trd->trd_index);
+
+	LASSERT(mutex_is_locked(&ted->ted_lcd_lock));
+
+	/* Do not free the reply data corresponding to the
+	 * highest transno of this export.
+	 * This ensures on-disk reply data is kept and
+	 * last committed transno can be restored from disk in case
+	 * of target recovery
+	 */
+	if (trd->trd_reply.lrd_transno == ted->ted_lcd->lcd_last_transno) {
+		/* free previous retained reply */
+		if (ted->ted_reply_last != NULL)
+			tgt_free_reply_data(lut, ted, ted->ted_reply_last);
+		/* retain the reply */
+		list_del_init(&trd->trd_list);
+		ted->ted_reply_last = trd;
+	} else {
+		tgt_free_reply_data(lut, ted, trd);
+	}
+}
+
 static inline struct lu_buf *tti_buf_lsd(struct tgt_thread_info *tti)
 {
 	tti->tti_buf.lb_buf = &tti->tti_lsd;
@@ -58,6 +393,13 @@ static inline struct lu_buf *tti_buf_lcd(struct tgt_thread_info *tti)
 	return &tti->tti_buf;
 }
 
+static inline bool tgt_is_multimodrpcs_record(struct lu_target *tgt,
+					      struct lsd_client_data *lcd)
+{
+	return tgt->lut_lsd.lsd_feature_incompat & OBD_INCOMPAT_MULTI_RPCS &&
+		lcd->lcd_generation != 0;
+}
+
 /**
  * Allocate in-memory data for client slot related to export.
  */
@@ -66,11 +408,18 @@ int tgt_client_alloc(struct obd_export *exp)
 	ENTRY;
 	LASSERT(exp != exp->exp_obd->obd_self_export);
 
+	spin_lock_init(&exp->exp_target_data.ted_nodemap_lock);
+	INIT_LIST_HEAD(&exp->exp_target_data.ted_nodemap_member);
+	spin_lock_init(&exp->exp_target_data.ted_fmd_lock);
+	INIT_LIST_HEAD(&exp->exp_target_data.ted_fmd_list);
+
 	OBD_ALLOC_PTR(exp->exp_target_data.ted_lcd);
 	if (exp->exp_target_data.ted_lcd == NULL)
 		RETURN(-ENOMEM);
 	/* Mark that slot is not yet valid, 0 doesn't work here */
 	exp->exp_target_data.ted_lr_idx = -1;
+	INIT_LIST_HEAD(&exp->exp_target_data.ted_reply_list);
+	mutex_init(&exp->exp_target_data.ted_lcd_lock);
 	RETURN(0);
 }
 EXPORT_SYMBOL(tgt_client_alloc);
@@ -82,17 +431,40 @@ void tgt_client_free(struct obd_export *exp)
 {
 	struct tg_export_data	*ted = &exp->exp_target_data;
 	struct lu_target	*lut = class_exp2tgt(exp);
+	struct tg_reply_data	*trd, *tmp;
 
 	LASSERT(exp != exp->exp_obd->obd_self_export);
 
+	tgt_fmd_cleanup(exp);
+
+	/* free reply data */
+	mutex_lock(&ted->ted_lcd_lock);
+	list_for_each_entry_safe(trd, tmp, &ted->ted_reply_list, trd_list) {
+		tgt_release_reply_data(lut, ted, trd);
+	}
+	if (ted->ted_reply_last != NULL) {
+		tgt_free_reply_data(lut, ted, ted->ted_reply_last);
+		ted->ted_reply_last = NULL;
+	}
+	mutex_unlock(&ted->ted_lcd_lock);
+
+	if (!hlist_unhashed(&exp->exp_gen_hash))
+		cfs_hash_del(exp->exp_obd->obd_gen_hash,
+			     &ted->ted_lcd->lcd_generation,
+			     &exp->exp_gen_hash);
+
 	OBD_FREE_PTR(ted->ted_lcd);
 	ted->ted_lcd = NULL;
 
-	/* Slot may be not yet assigned */
-	if (ted->ted_lr_idx < 0)
+	/* Target may have been freed (see LU-7430)
+	 * Slot may be not yet assigned */
+	if (((struct obd_device_target *)(&exp->exp_obd->u))->obt_magic !=
+	    OBT_MAGIC ||
+	    ted->ted_lr_idx < 0)
 		return;
+
 	/* Clear bit when lcd is freed */
-	LASSERT(lut->lut_client_bitmap);
+	LASSERT(lut && lut->lut_client_bitmap);
 	if (!test_and_clear_bit(ted->ted_lr_idx, lut->lut_client_bitmap)) {
 		CERROR("%s: client %u bit already clear in bitmap\n",
 		       exp->exp_obd->obd_name, ted->ted_lr_idx);
@@ -101,8 +473,22 @@ void tgt_client_free(struct obd_export *exp)
 }
 EXPORT_SYMBOL(tgt_client_free);
 
-int tgt_client_data_read(const struct lu_env *env, struct lu_target *tgt,
-			 struct lsd_client_data *lcd, loff_t *off, int index)
+static inline void tgt_check_lcd(const char *obd_name, int index,
+				 struct lsd_client_data *lcd)
+{
+	size_t uuid_size = sizeof(lcd->lcd_uuid);
+
+	if (strnlen((char*)lcd->lcd_uuid, uuid_size) == uuid_size) {
+		lcd->lcd_uuid[uuid_size - 1] = '\0';
+
+		LCONSOLE_ERROR("the client UUID (%s) on %s for exports stored in last_rcvd(index = %d) is bad!\n",
+			       lcd->lcd_uuid, obd_name, index);
+	}
+}
+
+static int tgt_client_data_read(const struct lu_env *env, struct lu_target *tgt,
+				struct lsd_client_data *lcd,
+				loff_t *off, int index)
 {
 	struct tgt_thread_info	*tti = tgt_th_info(env);
 	int			 rc;
@@ -110,16 +496,16 @@ int tgt_client_data_read(const struct lu_env *env, struct lu_target *tgt,
 	tti_buf_lcd(tti);
 	rc = dt_record_read(env, tgt->lut_last_rcvd, &tti->tti_buf, off);
 	if (rc == 0) {
-		check_lcd(tgt->lut_obd->obd_name, index, &tti->tti_lcd);
+		tgt_check_lcd(tgt->lut_obd->obd_name, index, &tti->tti_lcd);
 		lcd_le_to_cpu(&tti->tti_lcd, lcd);
 		lcd->lcd_last_result = ptlrpc_status_ntoh(lcd->lcd_last_result);
 		lcd->lcd_last_close_result =
 			ptlrpc_status_ntoh(lcd->lcd_last_close_result);
 	}
 
-	CDEBUG(D_INFO, "%s: read lcd @%lld uuid = %s, last_transno = "LPU64
-	       ", last_xid = "LPU64", last_result = %u, last_data = %u, "
-	       "last_close_transno = "LPU64", last_close_xid = "LPU64", "
+	CDEBUG(D_INFO, "%s: read lcd @%lld uuid = %s, last_transno = %llu"
+	       ", last_xid = %llu, last_result = %u, last_data = %u, "
+	       "last_close_transno = %llu, last_close_xid = %llu, "
 	       "last_close_result = %u, rc = %d\n", tgt->lut_obd->obd_name,
 	       *off, lcd->lcd_uuid, lcd->lcd_last_transno, lcd->lcd_last_xid,
 	       lcd->lcd_last_result, lcd->lcd_last_data,
@@ -128,11 +514,13 @@ int tgt_client_data_read(const struct lu_env *env, struct lu_target *tgt,
 	return rc;
 }
 
-int tgt_client_data_write(const struct lu_env *env, struct lu_target *tgt,
-			  struct lsd_client_data *lcd, loff_t *off,
-			  struct thandle *th)
+static int tgt_client_data_write(const struct lu_env *env,
+				 struct lu_target *tgt,
+				 struct lsd_client_data *lcd,
+				 loff_t *off, struct thandle *th)
 {
 	struct tgt_thread_info *tti = tgt_th_info(env);
+	struct dt_object	*dto;
 
 	lcd->lcd_last_result = ptlrpc_status_hton(lcd->lcd_last_result);
 	lcd->lcd_last_close_result =
@@ -140,7 +528,61 @@ int tgt_client_data_write(const struct lu_env *env, struct lu_target *tgt,
 	lcd_cpu_to_le(lcd, &tti->tti_lcd);
 	tti_buf_lcd(tti);
 
-	return dt_record_write(env, tgt->lut_last_rcvd, &tti->tti_buf, off, th);
+	dto = dt_object_locate(tgt->lut_last_rcvd, th->th_dev);
+	return dt_record_write(env, dto, &tti->tti_buf, off, th);
+}
+
+struct tgt_new_client_callback {
+	struct dt_txn_commit_cb	 lncc_cb;
+	struct obd_export	*lncc_exp;
+};
+
+static void tgt_cb_new_client(struct lu_env *env, struct thandle *th,
+			      struct dt_txn_commit_cb *cb, int err)
+{
+	struct tgt_new_client_callback *ccb;
+
+	ccb = container_of(cb, struct tgt_new_client_callback, lncc_cb);
+
+	LASSERT(ccb->lncc_exp->exp_obd);
+
+	CDEBUG(D_RPCTRACE, "%s: committing for initial connect of %s\n",
+	       ccb->lncc_exp->exp_obd->obd_name,
+	       ccb->lncc_exp->exp_client_uuid.uuid);
+
+	spin_lock(&ccb->lncc_exp->exp_lock);
+
+	ccb->lncc_exp->exp_need_sync = 0;
+
+	spin_unlock(&ccb->lncc_exp->exp_lock);
+	class_export_cb_put(ccb->lncc_exp);
+
+	OBD_FREE_PTR(ccb);
+}
+
+static int tgt_new_client_cb_add(struct thandle *th, struct obd_export *exp)
+{
+	struct tgt_new_client_callback *ccb;
+	struct dt_txn_commit_cb	*dcb;
+	int rc;
+
+	OBD_ALLOC_PTR(ccb);
+	if (ccb == NULL)
+		return -ENOMEM;
+
+	ccb->lncc_exp = class_export_cb_get(exp);
+
+	dcb = &ccb->lncc_cb;
+	dcb->dcb_func = tgt_cb_new_client;
+	INIT_LIST_HEAD(&dcb->dcb_linkage);
+	strscpy(dcb->dcb_name, "tgt_cb_new_client", sizeof(dcb->dcb_name));
+
+	rc = dt_trans_cb_add(th, dcb);
+	if (rc) {
+		class_export_cb_put(exp);
+		OBD_FREE_PTR(ccb);
+	}
+	return rc;
 }
 
 /**
@@ -157,6 +599,15 @@ static int tgt_client_data_update(const struct lu_env *env,
 
 	ENTRY;
 
+	if (unlikely(tgt == NULL)) {
+		CDEBUG(D_ERROR, "%s: No target for connected export\n",
+			  class_exp2obd(exp)->obd_name);
+		RETURN(-EINVAL);
+	}
+
+	if (tgt->lut_bottom->dd_rdonly)
+		RETURN(0);
+
 	th = dt_trans_create(env, tgt->lut_bottom);
 	if (IS_ERR(th))
 		RETURN(PTR_ERR(th));
@@ -171,6 +622,9 @@ static int tgt_client_data_update(const struct lu_env *env,
 	rc = dt_trans_start_local(env, tgt->lut_bottom, th);
 	if (rc)
 		GOTO(out, rc);
+
+	mutex_lock(&ted->ted_lcd_lock);
+
 	/*
 	 * Until this operations will be committed the sync is needed
 	 * for this export. This should be done _after_ starting the
@@ -189,17 +643,20 @@ static int tgt_client_data_update(const struct lu_env *env,
 
 	tti->tti_off = ted->ted_lr_off;
 	rc = tgt_client_data_write(env, tgt, ted->ted_lcd, &tti->tti_off, th);
+
+	mutex_unlock(&ted->ted_lcd_lock);
+
 	EXIT;
 out:
 	dt_trans_stop(env, tgt->lut_bottom, th);
 	CDEBUG(D_INFO, "%s: update last_rcvd client data for UUID = %s, "
-	       "last_transno = "LPU64": rc = %d\n", tgt->lut_obd->obd_name,
+	       "last_transno = %llu: rc = %d\n", tgt->lut_obd->obd_name,
 	       tgt->lut_lsd.lsd_uuid, tgt->lut_lsd.lsd_last_transno, rc);
 
 	return rc;
 }
 
-int tgt_server_data_read(const struct lu_env *env, struct lu_target *tgt)
+static int tgt_server_data_read(const struct lu_env *env, struct lu_target *tgt)
 {
 	struct tgt_thread_info	*tti = tgt_th_info(env);
 	int			 rc;
@@ -212,15 +669,16 @@ int tgt_server_data_read(const struct lu_env *env, struct lu_target *tgt)
 		lsd_le_to_cpu(&tti->tti_lsd, &tgt->lut_lsd);
 
 	CDEBUG(D_INFO, "%s: read last_rcvd server data for UUID = %s, "
-	       "last_transno = "LPU64": rc = %d\n", tgt->lut_obd->obd_name,
+	       "last_transno = %llu: rc = %d\n", tgt->lut_obd->obd_name,
 	       tgt->lut_lsd.lsd_uuid, tgt->lut_lsd.lsd_last_transno, rc);
         return rc;
 }
 
-int tgt_server_data_write(const struct lu_env *env, struct lu_target *tgt,
-			  struct thandle *th)
+static int tgt_server_data_write(const struct lu_env *env,
+				 struct lu_target *tgt, struct thandle *th)
 {
 	struct tgt_thread_info	*tti = tgt_th_info(env);
+	struct dt_object	*dto;
 	int			 rc;
 
 	ENTRY;
@@ -229,11 +687,11 @@ int tgt_server_data_write(const struct lu_env *env, struct lu_target *tgt,
 	tti_buf_lsd(tti);
 	lsd_cpu_to_le(&tgt->lut_lsd, &tti->tti_lsd);
 
-	rc = dt_record_write(env, tgt->lut_last_rcvd, &tti->tti_buf,
-			     &tti->tti_off, th);
+	dto = dt_object_locate(tgt->lut_last_rcvd, th->th_dev);
+	rc = dt_record_write(env, dto, &tti->tti_buf, &tti->tti_off, th);
 
 	CDEBUG(D_INFO, "%s: write last_rcvd server data for UUID = %s, "
-	       "last_transno = "LPU64": rc = %d\n", tgt->lut_obd->obd_name,
+	       "last_transno = %llu: rc = %d\n", tgt->lut_obd->obd_name,
 	       tgt->lut_lsd.lsd_uuid, tgt->lut_lsd.lsd_last_transno, rc);
 
 	RETURN(rc);
@@ -252,8 +710,8 @@ int tgt_server_data_update(const struct lu_env *env, struct lu_target *tgt,
 	ENTRY;
 
 	CDEBUG(D_SUPER,
-	       "%s: mount_count is "LPU64", last_transno is "LPU64"\n",
-	       tgt->lut_lsd.lsd_uuid, tgt->lut_obd->u.obt.obt_mount_count,
+	       "%s: mount_count is %llu, last_transno is %llu\n",
+	       tgt->lut_lsd.lsd_uuid, obd2obt(tgt->lut_obd)->obt_mount_count,
 	       tgt->lut_last_transno);
 
 	/* Always save latest transno to keep it fresh */
@@ -261,6 +719,9 @@ int tgt_server_data_update(const struct lu_env *env, struct lu_target *tgt,
 	tgt->lut_lsd.lsd_last_transno = tgt->lut_last_transno;
 	spin_unlock(&tgt->lut_translock);
 
+	if (tgt->lut_bottom->dd_rdonly)
+		RETURN(0);
+
 	th = dt_trans_create(env, tgt->lut_bottom);
 	if (IS_ERR(th))
 		RETURN(PTR_ERR(th));
@@ -282,22 +743,24 @@ out:
 	dt_trans_stop(env, tgt->lut_bottom, th);
 
 	CDEBUG(D_INFO, "%s: update last_rcvd server data for UUID = %s, "
-	       "last_transno = "LPU64": rc = %d\n", tgt->lut_obd->obd_name,
+	       "last_transno = %llu: rc = %d\n", tgt->lut_obd->obd_name,
 	       tgt->lut_lsd.lsd_uuid, tgt->lut_lsd.lsd_last_transno, rc);
 	RETURN(rc);
 }
 EXPORT_SYMBOL(tgt_server_data_update);
 
-int tgt_truncate_last_rcvd(const struct lu_env *env, struct lu_target *tgt,
-			   loff_t size)
+static int tgt_truncate_object(const struct lu_env *env, struct lu_target *tgt,
+			       struct dt_object *dt, loff_t size)
 {
-	struct dt_object *dt = tgt->lut_last_rcvd;
 	struct thandle	 *th;
 	struct lu_attr	  attr;
 	int		  rc;
 
 	ENTRY;
 
+	if (tgt->lut_bottom->dd_rdonly)
+		RETURN(0);
+
 	attr.la_size = size;
 	attr.la_valid = LA_SIZE;
 
@@ -314,9 +777,9 @@ int tgt_truncate_last_rcvd(const struct lu_env *env, struct lu_target *tgt,
 	if (rc)
 		GOTO(cleanup, rc);
 
-	rc = dt_punch(env, dt, size, OBD_OBJECT_EOF, th, BYPASS_CAPA);
+	rc = dt_punch(env, dt, size, OBD_OBJECT_EOF, th);
 	if (rc == 0)
-		rc = dt_attr_set(env, dt, &attr, th, BYPASS_CAPA);
+		rc = dt_attr_set(env, dt, &attr, th);
 
 cleanup:
 	dt_trans_stop(env, tgt->lut_bottom, th);
@@ -330,7 +793,7 @@ static void tgt_client_epoch_update(const struct lu_env *env,
 	struct lsd_client_data	*lcd = exp->exp_target_data.ted_lcd;
 	struct lu_target	*tgt = class_exp2tgt(exp);
 
-	LASSERT(tgt->lut_bottom);
+	LASSERT(tgt && tgt->lut_bottom);
 	/** VBR: set client last_epoch to current epoch */
 	if (lcd->lcd_last_epoch >= tgt->lut_lsd.lsd_start_epoch)
 		return;
@@ -338,6 +801,52 @@ static void tgt_client_epoch_update(const struct lu_env *env,
 	tgt_client_data_update(env, exp);
 }
 
+static int tgt_reply_data_upgrade_check(const struct lu_env *env,
+					struct lu_target *tgt)
+{
+	struct lsd_reply_header *lrh = &tgt->lut_reply_header;
+	int rc;
+
+	/*
+	 * Reply data is supported by MDT targets only for now.
+	 * When reply data object @lut_reply_data is NULL, it indicates the
+	 * target type is OST and it should skip the upgrade check.
+	 */
+	if (tgt->lut_reply_data == NULL)
+		RETURN(0);
+
+	rc = tgt_reply_header_read(env, tgt, lrh);
+	if (rc) {
+		CERROR("%s: failed to read %s: rc = %d\n",
+		       tgt_name(tgt), REPLY_DATA, rc);
+		RETURN(rc);
+	}
+
+	if (lrh->lrh_magic == LRH_MAGIC)
+		RETURN(0);
+
+	rc = tgt_truncate_object(env, tgt, tgt->lut_reply_data, 0);
+	if (rc) {
+		CERROR("%s: failed to truncate %s: rc = %d\n",
+		       tgt_name(tgt), REPLY_DATA, rc);
+		RETURN(rc);
+	}
+
+	lrh->lrh_magic = LRH_MAGIC;
+	lrh->lrh_header_size = sizeof(struct lsd_reply_header);
+	if (lrh->lrh_magic == LRH_MAGIC_V1)
+		lrh->lrh_reply_size = sizeof(struct lsd_reply_data_v1);
+	else
+		lrh->lrh_reply_size = sizeof(struct lsd_reply_data_v2);
+
+	rc = tgt_reply_header_write(env, tgt, lrh);
+	if (rc)
+		CERROR("%s: failed to write header for %s: rc = %d\n",
+		       tgt_name(tgt), REPLY_DATA, rc);
+
+	RETURN(rc);
+}
+
 /**
  * Update boot epoch when recovery ends
  */
@@ -346,7 +855,7 @@ void tgt_boot_epoch_update(struct lu_target *tgt)
 	struct lu_env		 env;
 	struct ptlrpc_request	*req;
 	__u32			 start_epoch;
-	struct list_head	 client_list;
+	LIST_HEAD(client_list);
 	int			 rc;
 
 	if (tgt->lut_obd->obd_stopping)
@@ -360,12 +869,11 @@ void tgt_boot_epoch_update(struct lu_target *tgt)
 	}
 
 	spin_lock(&tgt->lut_translock);
-	start_epoch = lr_epoch(tgt->lut_last_transno) + 1;
+	start_epoch = (tgt->lut_last_transno >> LR_EPOCH_BITS) + 1;
 	tgt->lut_last_transno = (__u64)start_epoch << LR_EPOCH_BITS;
 	tgt->lut_lsd.lsd_start_epoch = start_epoch;
 	spin_unlock(&tgt->lut_translock);
 
-	INIT_LIST_HEAD(&client_list);
 	/**
 	 * The recovery is not yet finished and final queue can still be updated
 	 * with resend requests. Move final list to separate one for processing
@@ -387,13 +895,22 @@ void tgt_boot_epoch_update(struct lu_target *tgt)
 	spin_lock(&tgt->lut_obd->obd_recovery_task_lock);
 	list_splice_init(&client_list, &tgt->lut_obd->obd_final_req_queue);
 	spin_unlock(&tgt->lut_obd->obd_recovery_task_lock);
+
+	/**
+	 * Clear MULTI RPCS incompatibility flag if there is no multi-rpcs
+	 * client in last_rcvd file
+	 */
+	if (atomic_read(&tgt->lut_num_clients) == 0)
+		tgt->lut_lsd.lsd_feature_incompat &= ~OBD_INCOMPAT_MULTI_RPCS;
+
 	/** update server epoch */
 	tgt_server_data_update(&env, tgt, 1);
+	tgt_reply_data_upgrade_check(&env, tgt);
 	lu_env_fini(&env);
 }
 
 /**
- * commit callback, need to update last_commited value
+ * commit callback, need to update last_committed value
  */
 struct tgt_last_committed_callback {
 	struct dt_txn_commit_cb	 llcc_cb;
@@ -407,32 +924,61 @@ static void tgt_cb_last_committed(struct lu_env *env, struct thandle *th,
 {
 	struct tgt_last_committed_callback *ccb;
 
-	ccb = container_of0(cb, struct tgt_last_committed_callback, llcc_cb);
+	ccb = container_of(cb, struct tgt_last_committed_callback, llcc_cb);
 
+	LASSERT(ccb->llcc_exp);
 	LASSERT(ccb->llcc_tgt != NULL);
 	LASSERT(ccb->llcc_exp->exp_obd == ccb->llcc_tgt->lut_obd);
 
+	if (th->th_reserved_quota.lqi_space > 0) {
+		CDEBUG(D_QUOTA, "free quota %llu %llu\n",
+		       th->th_reserved_quota.lqi_id.qid_gid,
+		       th->th_reserved_quota.lqi_space);
+
+		/* env can be NULL for freeing reserved quota */
+		th->th_reserved_quota.lqi_space *= -1;
+		dt_reserve_or_free_quota(NULL, th->th_dev,
+					 &th->th_reserved_quota);
+	}
+
+	/* error hit, don't update last committed to provide chance to
+	 * replay data after fail */
+	if (err != 0)
+		goto out;
+
+	/* Fast path w/o spinlock, if exp_last_committed was updated
+	 * with higher transno, no need to take spinlock and check,
+	 * also no need to update obd_last_committed. */
+	if (ccb->llcc_transno <= ccb->llcc_exp->exp_last_committed)
+		goto out;
 	spin_lock(&ccb->llcc_tgt->lut_translock);
 	if (ccb->llcc_transno > ccb->llcc_tgt->lut_obd->obd_last_committed)
 		ccb->llcc_tgt->lut_obd->obd_last_committed = ccb->llcc_transno;
 
-	LASSERT(ccb->llcc_exp);
 	if (ccb->llcc_transno > ccb->llcc_exp->exp_last_committed) {
 		ccb->llcc_exp->exp_last_committed = ccb->llcc_transno;
 		spin_unlock(&ccb->llcc_tgt->lut_translock);
+
 		ptlrpc_commit_replies(ccb->llcc_exp);
+		tgt_cancel_slc_locks(ccb->llcc_tgt, ccb->llcc_transno);
 	} else {
 		spin_unlock(&ccb->llcc_tgt->lut_translock);
 	}
+
+	CDEBUG(D_HA, "%s: transno %lld is committed\n",
+	       ccb->llcc_tgt->lut_obd->obd_name, ccb->llcc_transno);
+
+out:
 	class_export_cb_put(ccb->llcc_exp);
-	if (ccb->llcc_transno)
-		CDEBUG(D_HA, "%s: transno "LPD64" is committed\n",
-		       ccb->llcc_tgt->lut_obd->obd_name, ccb->llcc_transno);
 	OBD_FREE_PTR(ccb);
 }
 
-int tgt_last_commit_cb_add(struct thandle *th, struct lu_target *tgt,
-			   struct obd_export *exp, __u64 transno)
+/**
+ * Add commit callback function, it returns a non-zero value to inform
+ * caller to use sync transaction if necessary.
+ */
+static int tgt_last_commit_cb_add(struct thandle *th, struct lu_target *tgt,
+				  struct obd_export *exp, __u64 transno)
 {
 	struct tgt_last_committed_callback	*ccb;
 	struct dt_txn_commit_cb			*dcb;
@@ -442,82 +988,48 @@ int tgt_last_commit_cb_add(struct thandle *th, struct lu_target *tgt,
 	if (ccb == NULL)
 		return -ENOMEM;
 
-	ccb->llcc_tgt = tgt;
-	ccb->llcc_exp = class_export_cb_get(exp);
-	ccb->llcc_transno = transno;
-
-	dcb = &ccb->llcc_cb;
-	dcb->dcb_func = tgt_cb_last_committed;
-	INIT_LIST_HEAD(&dcb->dcb_linkage);
-	strlcpy(dcb->dcb_name, "tgt_cb_last_committed", sizeof(dcb->dcb_name));
-
-	rc = dt_trans_cb_add(th, dcb);
-	if (rc) {
-		class_export_cb_put(exp);
-		OBD_FREE_PTR(ccb);
-	}
-
-	if (exp_connect_flags(exp) & OBD_CONNECT_LIGHTWEIGHT)
-		/* report failure to force synchronous operation */
-		return -EPERM;
-
-	return rc;
-}
-
-struct tgt_new_client_callback {
-	struct dt_txn_commit_cb	 lncc_cb;
-	struct obd_export	*lncc_exp;
-};
-
-static void tgt_cb_new_client(struct lu_env *env, struct thandle *th,
-			      struct dt_txn_commit_cb *cb, int err)
-{
-	struct tgt_new_client_callback *ccb;
-
-	ccb = container_of0(cb, struct tgt_new_client_callback, lncc_cb);
-
-	LASSERT(ccb->lncc_exp->exp_obd);
-
-	CDEBUG(D_RPCTRACE, "%s: committing for initial connect of %s\n",
-	       ccb->lncc_exp->exp_obd->obd_name,
-	       ccb->lncc_exp->exp_client_uuid.uuid);
-
-	spin_lock(&ccb->lncc_exp->exp_lock);
-	/* XXX: Currently, we use per-export based sync/async policy for
-	 *	the update via OUT RPC, it is coarse-grained policy, and
-	 *	will be changed as per-request based by DNE II patches. */
-	if (!ccb->lncc_exp->exp_keep_sync)
-		ccb->lncc_exp->exp_need_sync = 0;
-
-	spin_unlock(&ccb->lncc_exp->exp_lock);
-	class_export_cb_put(ccb->lncc_exp);
-
-	OBD_FREE_PTR(ccb);
-}
-
-int tgt_new_client_cb_add(struct thandle *th, struct obd_export *exp)
-{
-	struct tgt_new_client_callback	*ccb;
-	struct dt_txn_commit_cb		*dcb;
-	int				 rc;
-
-	OBD_ALLOC_PTR(ccb);
-	if (ccb == NULL)
-		return -ENOMEM;
-
-	ccb->lncc_exp = class_export_cb_get(exp);
+	ccb->llcc_tgt = tgt;
+	ccb->llcc_exp = class_export_cb_get(exp);
+	ccb->llcc_transno = transno;
 
-	dcb = &ccb->lncc_cb;
-	dcb->dcb_func = tgt_cb_new_client;
+	dcb = &ccb->llcc_cb;
+	dcb->dcb_func = tgt_cb_last_committed;
 	INIT_LIST_HEAD(&dcb->dcb_linkage);
-	strlcpy(dcb->dcb_name, "tgt_cb_new_client", sizeof(dcb->dcb_name));
+	strscpy(dcb->dcb_name, "tgt_cb_last_committed", sizeof(dcb->dcb_name));
 
 	rc = dt_trans_cb_add(th, dcb);
 	if (rc) {
 		class_export_cb_put(exp);
 		OBD_FREE_PTR(ccb);
 	}
-	return rc;
+
+	if (exp_connect_flags(exp) & OBD_CONNECT_LIGHTWEIGHT)
+		/* report failure to force synchronous operation */
+		return -EPERM;
+
+	/* if exp_need_sync is set, return non-zero value to force
+	 * a sync transaction. */
+	return rc ? rc : exp->exp_need_sync;
+}
+
+static int tgt_is_local_client(const struct lu_env *env,
+				      struct obd_export *exp)
+{
+	struct lu_target	*tgt = class_exp2tgt(exp);
+	struct tgt_session_info *tsi = tgt_ses_info(env);
+	struct ptlrpc_request	*req = tgt_ses_req(tsi);
+
+	if (exp_connect_flags(exp) & OBD_CONNECT_MDS ||
+	    exp_connect_flags(exp) & OBD_CONNECT_MDS_MDS)
+		return 0;
+	if (tgt->lut_local_recovery)
+		return 0;
+	if (!req)
+		return 0;
+	if (!LNetIsPeerLocal(&req->rq_peer.nid))
+		return 0;
+
+	return 1;
 }
 
 /**
@@ -534,22 +1046,27 @@ int tgt_client_new(const struct lu_env *env, struct obd_export *exp)
 
 	ENTRY;
 
-	LASSERT(tgt->lut_client_bitmap != NULL);
+	LASSERT(tgt && tgt->lut_client_bitmap != NULL);
 	if (!strcmp(ted->ted_lcd->lcd_uuid, tgt->lut_obd->obd_uuid.uuid))
 		RETURN(0);
 
-	mutex_init(&ted->ted_lcd_lock);
-
 	if (exp_connect_flags(exp) & OBD_CONNECT_LIGHTWEIGHT)
 		RETURN(0);
 
+	if (tgt_is_local_client(env, exp)) {
+		LCONSOLE_WARN("%s: local client %s w/o recovery\n",
+			      exp->exp_obd->obd_name, ted->ted_lcd->lcd_uuid);
+		exp->exp_no_recovery = 1;
+		RETURN(0);
+	}
+
 	/* the bitmap operations can handle cl_idx > sizeof(long) * 8, so
 	 * there's no need for extra complication here
 	 */
 	idx = find_first_zero_bit(tgt->lut_client_bitmap, LR_MAX_CLIENTS);
 repeat:
 	if (idx >= LR_MAX_CLIENTS ||
-	    OBD_FAIL_CHECK(OBD_FAIL_MDS_CLIENT_ADD)) {
+	    CFS_FAIL_CHECK(OBD_FAIL_MDS_CLIENT_ADD)) {
 		CERROR("%s: no room for %u clients - fix LR_MAX_CLIENTS\n",
 		       tgt->lut_obd->obd_name,  idx);
 		RETURN(-EOVERFLOW);
@@ -560,35 +1077,60 @@ repeat:
 		goto repeat;
 	}
 
-	CDEBUG(D_INFO, "%s: client at idx %d with UUID '%s' added\n",
-	       tgt->lut_obd->obd_name, idx, ted->ted_lcd->lcd_uuid);
-
 	ted->ted_lr_idx = idx;
 	ted->ted_lr_off = tgt->lut_lsd.lsd_client_start +
 			  idx * tgt->lut_lsd.lsd_client_size;
 
 	LASSERTF(ted->ted_lr_off > 0, "ted_lr_off = %llu\n", ted->ted_lr_off);
 
-	CDEBUG(D_INFO, "%s: new client at index %d (%llu) with UUID '%s'\n",
+	if (tgt_is_multimodrpcs_client(exp)) {
+		/* Set MULTI RPCS incompatibility flag to prevent previous
+		 * Lustre versions to mount a target with reply_data file */
+		if (!(tgt->lut_lsd.lsd_feature_incompat &
+		      OBD_INCOMPAT_MULTI_RPCS)) {
+			tgt->lut_lsd.lsd_feature_incompat |=
+							OBD_INCOMPAT_MULTI_RPCS;
+			rc = tgt_server_data_update(env, tgt, 1);
+			if (rc < 0) {
+				CERROR("%s: unable to set MULTI RPCS "
+				       "incompatibility flag\n",
+				       exp->exp_obd->obd_name);
+				RETURN(rc);
+			}
+		}
+
+		/* assign client slot generation */
+		ted->ted_lcd->lcd_generation =
+				atomic_inc_return(&tgt->lut_client_generation);
+	} else {
+		ted->ted_lcd->lcd_generation = 0;
+	}
+
+	CDEBUG(D_INFO, "%s: new client at index %d (%llu) with UUID '%s' "
+	       "generation %d\n",
 	       tgt->lut_obd->obd_name, ted->ted_lr_idx, ted->ted_lr_off,
-	       ted->ted_lcd->lcd_uuid);
+	       ted->ted_lcd->lcd_uuid, ted->ted_lcd->lcd_generation);
 
-	if (OBD_FAIL_CHECK(OBD_FAIL_TGT_CLIENT_ADD))
+	if (CFS_FAIL_CHECK(OBD_FAIL_TGT_CLIENT_ADD))
 		RETURN(-ENOSPC);
 
 	rc = tgt_client_data_update(env, exp);
-	if (rc)
+	if (rc) {
 		CERROR("%s: Failed to write client lcd at idx %d, rc %d\n",
 		       tgt->lut_obd->obd_name, idx, rc);
+		RETURN(rc);
+	}
 
-	RETURN(rc);
+	if (tgt_is_multimodrpcs_client(exp))
+		atomic_inc(&tgt->lut_num_clients);
+
+	RETURN(0);
 }
 EXPORT_SYMBOL(tgt_client_new);
 
-/* Add client data to the MDS.  We use a bitmap to locate a free space
- * in the last_rcvd file if cl_off is -1 (i.e. a new client).
- * Otherwise, we just have to read the data from the last_rcvd file and
- * we know its offset.
+/* Add an existing client to the MDS in-memory state based on
+ * a client that was previously found in the last_rcvd file and
+ * already has an assigned slot (idx >= 0).
  *
  * It should not be possible to fail adding an existing client - otherwise
  * mdt_init_server_data() callsite needs to be fixed.
@@ -600,7 +1142,7 @@ int tgt_client_add(const struct lu_env *env,  struct obd_export *exp, int idx)
 
 	ENTRY;
 
-	LASSERT(tgt->lut_client_bitmap != NULL);
+	LASSERT(tgt && tgt->lut_client_bitmap != NULL);
 	LASSERTF(idx >= 0, "%d\n", idx);
 
 	if (!strcmp(ted->ted_lcd->lcd_uuid, tgt->lut_obd->obd_uuid.uuid) ||
@@ -613,8 +1155,10 @@ int tgt_client_add(const struct lu_env *env,  struct obd_export *exp, int idx)
 		LBUG();
 	}
 
-	CDEBUG(D_INFO, "%s: client at idx %d with UUID '%s' added\n",
-	       tgt->lut_obd->obd_name, idx, ted->ted_lcd->lcd_uuid);
+	CDEBUG(D_INFO, "%s: client at idx %d with UUID '%s' added, "
+	       "generation %d\n",
+	       tgt->lut_obd->obd_name, idx, ted->ted_lcd->lcd_uuid,
+	       ted->ted_lcd->lcd_generation);
 
 	ted->ted_lr_idx = idx;
 	ted->ted_lr_off = tgt->lut_lsd.lsd_client_start +
@@ -637,11 +1181,26 @@ int tgt_client_del(const struct lu_env *env, struct obd_export *exp)
 
 	LASSERT(ted->ted_lcd);
 
+	if (unlikely(tgt == NULL)) {
+		CDEBUG(D_ERROR, "%s: No target for connected export\n",
+		       class_exp2obd(exp)->obd_name);
+		RETURN(-EINVAL);
+	}
+
 	/* XXX if lcd_uuid were a real obd_uuid, I could use obd_uuid_equals */
 	if (!strcmp((char *)ted->ted_lcd->lcd_uuid,
 		    (char *)tgt->lut_obd->obd_uuid.uuid) ||
-	    exp_connect_flags(exp) & OBD_CONNECT_LIGHTWEIGHT)
+	    exp_connect_flags(exp) & OBD_CONNECT_LIGHTWEIGHT ||
+	    exp->exp_no_recovery)
+		RETURN(0);
+
+	/* Slot may be not yet assigned, use case is race between Client
+	 * reconnect and forced eviction */
+	if (ted->ted_lr_idx < 0) {
+		CWARN("%s: client with UUID '%s' not in bitmap\n",
+		      tgt->lut_obd->obd_name, ted->ted_lcd->lcd_uuid);
 		RETURN(0);
+	}
 
 	CDEBUG(D_INFO, "%s: del client at idx %u, off %lld, UUID '%s'\n",
 	       tgt->lut_obd->obd_name, ted->ted_lr_idx, ted->ted_lr_off,
@@ -659,6 +1218,9 @@ int tgt_client_del(const struct lu_env *env, struct obd_export *exp)
 	if (exp->exp_flags & OBD_OPT_FAILOVER)
 		RETURN(0);
 
+	if (CFS_FAIL_CHECK(OBD_FAIL_TGT_CLIENT_DEL))
+		RETURN(0);
+
 	/* Make sure the server's last_transno is up to date.
 	 * This should be done before zeroing client slot so last_transno will
 	 * be in server data or in client data in case of failure */
@@ -670,11 +1232,21 @@ int tgt_client_del(const struct lu_env *env, struct obd_export *exp)
 		RETURN(rc);
 	}
 
+	/* Race between an eviction and a disconnection ?*/
 	mutex_lock(&ted->ted_lcd_lock);
+	if (ted->ted_lcd->lcd_uuid[0] == '\0') {
+		mutex_unlock(&ted->ted_lcd_lock);
+		RETURN(rc);
+	}
+
 	memset(ted->ted_lcd->lcd_uuid, 0, sizeof ted->ted_lcd->lcd_uuid);
-	rc = tgt_client_data_update(env, exp);
 	mutex_unlock(&ted->ted_lcd_lock);
 
+	rc = tgt_client_data_update(env, exp);
+
+	if (!rc && tgt_is_multimodrpcs_record(tgt, ted->ted_lcd))
+		atomic_dec(&tgt->lut_num_clients);
+
 	CDEBUG(rc == 0 ? D_INFO : D_ERROR,
 	       "%s: zeroing out client %s at idx %u (%llu), rc %d\n",
 	       tgt->lut_obd->obd_name, ted->ted_lcd->lcd_uuid,
@@ -683,6 +1255,215 @@ int tgt_client_del(const struct lu_env *env, struct obd_export *exp)
 }
 EXPORT_SYMBOL(tgt_client_del);
 
+static void tgt_clean_by_tag(struct obd_export *exp, __u64 xid, __u16 tag)
+{
+	struct tg_export_data	*ted = &exp->exp_target_data;
+	struct lu_target	*lut = class_exp2tgt(exp);
+	struct tg_reply_data	*trd, *tmp;
+
+	if (tag == 0)
+		return;
+
+	list_for_each_entry_safe(trd, tmp, &ted->ted_reply_list, trd_list) {
+		if (trd->trd_tag != tag)
+			continue;
+
+		LASSERT(ergo(tgt_is_increasing_xid_client(exp),
+			     trd->trd_reply.lrd_xid <= xid));
+
+		ted->ted_release_tag++;
+		tgt_release_reply_data(lut, ted, trd);
+	}
+}
+
+static int tgt_add_reply_data(const struct lu_env *env, struct lu_target *tgt,
+		       struct tg_export_data *ted, struct tg_reply_data *trd,
+		       struct ptlrpc_request *req,
+		       struct thandle *th, bool update_lrd_file)
+{
+	struct tgt_session_info *tsi = NULL;
+	struct lsd_reply_data *lrd;
+	int i = -1;
+	int rc;
+
+	lrd = &trd->trd_reply;
+	/* update export last transno */
+	mutex_lock(&ted->ted_lcd_lock);
+	if (lrd->lrd_transno > ted->ted_lcd->lcd_last_transno)
+		ted->ted_lcd->lcd_last_transno = lrd->lrd_transno;
+	mutex_unlock(&ted->ted_lcd_lock);
+
+	if (!tgt) {
+		trd->trd_index = TRD_INDEX_MEMORY;
+		GOTO(add_reply_data, rc = 0);
+	}
+
+	if (env) {
+		tsi = tgt_ses_info(env);
+		if (tsi->tsi_batch_trd) {
+			LASSERT(tsi->tsi_batch_env);
+			trd = tsi->tsi_batch_trd;
+			i = trd->trd_index;
+		}
+	}
+
+	if (i == -1) {
+		/* find a empty slot */
+		i = tgt_find_free_reply_slot(tgt);
+		if (unlikely(i < 0)) {
+			CERROR("%s: couldn't find a slot for reply data: rc = %d\n",
+			       tgt_name(tgt), i);
+			RETURN(i);
+		}
+		trd->trd_index = i;
+	}
+
+	if (update_lrd_file) {
+		struct lsd_reply_header *lrh = &tgt->lut_reply_header;
+		loff_t	off;
+
+		/* write reply data to disk */
+		off = lrh->lrh_header_size + lrh->lrh_reply_size * i;
+		rc = tgt_reply_data_write(env, tgt, lrd, off, th);
+		if (unlikely(rc != 0)) {
+			CERROR("%s: can't update %s file: rc = %d\n",
+			       tgt_name(tgt), REPLY_DATA, rc);
+			GOTO(free_slot, rc);
+		}
+	}
+
+add_reply_data:
+	/* add reply data to target export's reply list */
+	mutex_lock(&ted->ted_lcd_lock);
+	if (req != NULL) {
+		int exclude = tgt_is_increasing_xid_client(req->rq_export) ?
+			      MSG_REPLAY : MSG_REPLAY|MSG_RESENT;
+
+		if (req->rq_obsolete) {
+			CDEBUG(D_INFO,
+			       "drop reply data update for obsolete req xid=%llu,"
+			       "transno=%llu, tag=%hu\n", req->rq_xid,
+			       lrd->lrd_transno, trd->trd_tag);
+			mutex_unlock(&ted->ted_lcd_lock);
+			GOTO(free_slot, rc = -EBADR);
+		}
+
+		if (!(lustre_msg_get_flags(req->rq_reqmsg) & exclude) &&
+		    !(tsi && tsi->tsi_batch_env &&
+		      trd->trd_reply.lrd_batch_idx > 0))
+			tgt_clean_by_tag(req->rq_export, req->rq_xid,
+					 trd->trd_tag);
+	}
+
+	/*
+	 * For the batched RPC, all sub requests use one common @trd for the
+	 * reply data.
+	 */
+	if (list_empty(&trd->trd_list)) {
+		list_add(&trd->trd_list, &ted->ted_reply_list);
+		ted->ted_reply_cnt++;
+		if (ted->ted_reply_cnt > ted->ted_reply_max)
+			ted->ted_reply_max = ted->ted_reply_cnt;
+	}
+	mutex_unlock(&ted->ted_lcd_lock);
+
+	CDEBUG(D_TRACE, "add reply %p: xid %llu, transno %llu, "
+	       "tag %hu, client gen %u, slot idx %d\n",
+	       trd, lrd->lrd_xid, lrd->lrd_transno,
+	       trd->trd_tag, lrd->lrd_client_gen, trd->trd_index);
+
+	RETURN(0);
+
+free_slot:
+	if (tgt != NULL)
+		tgt_clear_reply_slot(tgt, trd->trd_index);
+	return rc;
+}
+
+int tgt_mk_reply_data(const struct lu_env *env,
+		      struct lu_target *tgt,
+		      struct tg_export_data *ted,
+		      struct ptlrpc_request *req,
+		      __u64 opdata,
+		      struct thandle *th,
+		      bool write_update,
+		      __u64 transno)
+{
+	struct tg_reply_data *trd = NULL;
+	struct lsd_reply_data *lrd;
+	__u64 *pre_versions = NULL;
+	struct tgt_session_info *tsi = NULL;
+	int rc;
+
+	if (env != NULL) {
+		tsi = tgt_ses_info(env);
+		if (tsi->tsi_batch_trd) {
+			LASSERT(tsi->tsi_batch_env);
+			trd = tsi->tsi_batch_trd;
+		}
+	}
+
+	if (trd == NULL) {
+		OBD_ALLOC_PTR(trd);
+		if (unlikely(trd == NULL))
+			RETURN(-ENOMEM);
+
+		INIT_LIST_HEAD(&trd->trd_list);
+	}
+
+	/* fill reply data information */
+	lrd = &trd->trd_reply;
+	lrd->lrd_transno = transno;
+	if (tsi && tsi->tsi_batch_env) {
+		if (tsi->tsi_batch_idx == 0) {
+			LASSERT(req != NULL);
+			tsi->tsi_batch_trd = trd;
+			trd->trd_index = -1;
+			lrd->lrd_xid = req->rq_xid;
+			trd->trd_tag = lustre_msg_get_tag(req->rq_reqmsg);
+			lrd->lrd_client_gen = ted->ted_lcd->lcd_generation;
+		}
+		lrd->lrd_batch_idx = tsi->tsi_batch_idx;
+	} else if (req != NULL) {
+		lrd->lrd_xid = req->rq_xid;
+		trd->trd_tag = lustre_msg_get_tag(req->rq_reqmsg);
+		lrd->lrd_client_gen = ted->ted_lcd->lcd_generation;
+		if (write_update) {
+			pre_versions = lustre_msg_get_versions(req->rq_repmsg);
+			lrd->lrd_result = th->th_result;
+		}
+	} else {
+		LASSERT(env != NULL);
+		LASSERT(tsi->tsi_xid != 0);
+
+		lrd->lrd_xid = tsi->tsi_xid;
+		lrd->lrd_result = tsi->tsi_result;
+		lrd->lrd_client_gen = tsi->tsi_client_gen;
+	}
+
+	lrd->lrd_data = opdata;
+	if (pre_versions) {
+		trd->trd_pre_versions[0] = pre_versions[0];
+		trd->trd_pre_versions[1] = pre_versions[1];
+		trd->trd_pre_versions[2] = pre_versions[2];
+		trd->trd_pre_versions[3] = pre_versions[3];
+	}
+
+	if (tsi && tsi->tsi_open_obj)
+		trd->trd_object = *lu_object_fid(&tsi->tsi_open_obj->do_lu);
+
+	rc = tgt_add_reply_data(env, tgt, ted, trd, req,
+				th, write_update);
+	if (rc < 0) {
+		OBD_FREE_PTR(trd);
+		if (rc == -EBADR)
+			rc = 0;
+	}
+	return rc;
+
+}
+EXPORT_SYMBOL(tgt_mk_reply_data);
+
 /*
  * last_rcvd & last_committed update callbacks
  */
@@ -691,27 +1472,39 @@ static int tgt_last_rcvd_update(const struct lu_env *env, struct lu_target *tgt,
 				struct thandle *th, struct ptlrpc_request *req)
 {
 	struct tgt_thread_info	*tti = tgt_th_info(env);
-	struct tg_export_data	*ted;
-	__u64			*transno_p;
-	int			 rc = 0;
-	bool			 lw_client, update = false;
+	struct tgt_session_info *tsi = tgt_ses_info(env);
+	struct obd_export *exp = tsi->tsi_exp;
+	struct tg_export_data *ted;
+	__u64 *transno_p;
+	bool nolcd = false;
+	int rc = 0;
 
 	ENTRY;
 
-	ted = &req->rq_export->exp_target_data;
 
-	lw_client = exp_connect_flags(req->rq_export) & OBD_CONNECT_LIGHTWEIGHT;
-	if (ted->ted_lr_idx < 0 && !lw_client)
-		/* ofd connect may cause transaction before export has
-		 * last_rcvd slot */
-		RETURN(0);
+	LASSERT(exp != NULL);
+	ted = &exp->exp_target_data;
 
-	tti->tti_transno = lustre_msg_get_transno(req->rq_reqmsg);
+	/* Some clients don't support recovery, and they don't have last_rcvd
+	 * client data:
+	 * 1. lightweight clients.
+	 * 2. local clients on MDS which doesn't enable "localrecov".
+	 * 3. OFD connect may cause transaction before export has last_rcvd
+	 *    slot.
+	 */
+	if (ted->ted_lr_idx < 0)
+		nolcd = true;
+
+	if (req != NULL)
+		tti->tti_transno = lustre_msg_get_transno(req->rq_reqmsg);
+	else
+		/* From update replay, tti_transno should be set already */
+		LASSERT(tti->tti_transno != 0);
 
 	spin_lock(&tgt->lut_translock);
 	if (th->th_result != 0) {
 		if (tti->tti_transno != 0) {
-			CERROR("%s: replay transno "LPU64" failed: rc = %d\n",
+			CERROR("%s: replay transno %llu failed: rc = %d\n",
 			       tgt_name(tgt), tti->tti_transno, th->th_result);
 		}
 	} else if (tti->tti_transno == 0) {
@@ -724,51 +1517,59 @@ static int tgt_last_rcvd_update(const struct lu_env *env, struct lu_target *tgt,
 	spin_unlock(&tgt->lut_translock);
 
 	/** VBR: set new versions */
-	if (th->th_result == 0 && obj != NULL)
-		dt_version_set(env, obj, tti->tti_transno, th);
+	if (th->th_result == 0 && obj != NULL) {
+		struct dt_object *dto = dt_object_locate(obj, th->th_dev);
+
+		dt_version_set(env, dto, tti->tti_transno, th);
+		if (unlikely(tsi->tsi_dv_update))
+			dt_data_version_set(env, dto, tti->tti_transno, th);
+	}
 
 	/* filling reply data */
-	CDEBUG(D_INODE, "transno = "LPU64", last_committed = "LPU64"\n",
+	CDEBUG(D_INODE, "transno = %llu, last_committed = %llu\n",
 	       tti->tti_transno, tgt->lut_obd->obd_last_committed);
 
-	req->rq_transno = tti->tti_transno;
-	lustre_msg_set_transno(req->rq_repmsg, tti->tti_transno);
+	if (req != NULL) {
+		req->rq_transno = tti->tti_transno;
+		lustre_msg_set_transno(req->rq_repmsg, tti->tti_transno);
+	}
 
 	/* if can't add callback, do sync write */
-	th->th_sync |= !!tgt_last_commit_cb_add(th, tgt, req->rq_export,
-						tti->tti_transno);
+	th->th_sync |= !!tgt_last_commit_cb_add(th, tgt, exp, tti->tti_transno);
 
-	if (lw_client) {
-		/* All operations performed by LW clients are synchronous and
-		 * we store the committed transno in the last_rcvd header */
+	if (nolcd) {
+		/* store transno in the last_rcvd header */
 		spin_lock(&tgt->lut_translock);
 		if (tti->tti_transno > tgt->lut_lsd.lsd_last_transno) {
 			tgt->lut_lsd.lsd_last_transno = tti->tti_transno;
-			update = true;
+			spin_unlock(&tgt->lut_translock);
+			/* Although current connection doesn't have slot
+			 * in the last_rcvd, we still want to maintain
+			 * the in-memory lsd_client_data structure in order to
+			 * properly handle reply reconstruction. */
+			rc = tgt_server_data_write(env, tgt, th);
+		} else {
+			spin_unlock(&tgt->lut_translock);
 		}
-		spin_unlock(&tgt->lut_translock);
-		/* Although lightweight (LW) connections have no slot in
-		 * last_rcvd, we still want to maintain the in-memory
-		 * lsd_client_data structure in order to properly handle reply
-		 * reconstruction. */
 	} else if (ted->ted_lr_off == 0) {
 		CERROR("%s: client idx %d has offset %lld\n",
 		       tgt_name(tgt), ted->ted_lr_idx, ted->ted_lr_off);
 		RETURN(-EINVAL);
 	}
 
-	/* if the export has already been disconnected, we have no last_rcvd
-	 * slot, update server data with latest transno then */
-	if (ted->ted_lcd == NULL) {
-		CWARN("commit transaction for disconnected client %s: rc %d\n",
-		      req->rq_export->exp_client_uuid.uuid, rc);
-		GOTO(srv_update, rc = 0);
+	/* Target that supports multiple reply data */
+	if (tgt_is_multimodrpcs_client(exp)) {
+		return tgt_mk_reply_data(env, tgt, ted, req, opdata, th,
+					 !!(req != NULL), tti->tti_transno);
 	}
 
+	/* Enough for update replay, let's return */
+	if (req == NULL)
+		RETURN(rc);
+
 	mutex_lock(&ted->ted_lcd_lock);
 	LASSERT(ergo(tti->tti_transno == 0, th->th_result != 0));
-	if (lustre_msg_get_opc(req->rq_reqmsg) == MDS_CLOSE ||
-	    lustre_msg_get_opc(req->rq_reqmsg) == MDS_DONE_WRITING) {
+	if (lustre_msg_get_opc(req->rq_reqmsg) == MDS_CLOSE) {
 		transno_p = &ted->ted_lcd->lcd_last_close_transno;
 		ted->ted_lcd->lcd_last_close_xid = req->rq_xid;
 		ted->ted_lcd->lcd_last_close_result = th->th_result;
@@ -792,37 +1593,43 @@ static int tgt_last_rcvd_update(const struct lu_env *env, struct lu_target *tgt,
 
 	/* Update transno in slot only if non-zero number, i.e. no errors */
 	if (likely(tti->tti_transno != 0)) {
-		if (*transno_p > tti->tti_transno &&
-		    !tgt->lut_no_reconstruct) {
-			CERROR("%s: trying to overwrite bigger transno:"
-			       "on-disk: "LPU64", new: "LPU64" replay: %d. "
-			       "see LU-617.\n", tgt_name(tgt), *transno_p,
-			       tti->tti_transno, req_is_replay(req));
-			if (req_is_replay(req)) {
-				spin_lock(&req->rq_export->exp_lock);
-				req->rq_export->exp_vbr_failed = 1;
-				spin_unlock(&req->rq_export->exp_lock);
+		/* Don't overwrite bigger transaction number with lower one.
+		 * That is not sign of problem in all cases, but in any case
+		 * this value should be monotonically increased only. */
+		if (*transno_p > tti->tti_transno) {
+			if (!tgt->lut_no_reconstruct) {
+				CERROR("%s: trying to overwrite bigger transno:"
+				       "on-disk: %llu, new: %llu replay: "
+				       "%d. See LU-617.\n", tgt_name(tgt),
+				       *transno_p, tti->tti_transno,
+				       req_is_replay(req));
+				if (req_is_replay(req)) {
+					spin_lock(&req->rq_export->exp_lock);
+					req->rq_export->exp_vbr_failed = 1;
+					spin_unlock(&req->rq_export->exp_lock);
+				}
+				mutex_unlock(&ted->ted_lcd_lock);
+				RETURN(req_is_replay(req) ? -EOVERFLOW : 0);
 			}
-			mutex_unlock(&ted->ted_lcd_lock);
-			RETURN(req_is_replay(req) ? -EOVERFLOW : 0);
+		} else {
+			*transno_p = tti->tti_transno;
 		}
-		*transno_p = tti->tti_transno;
 	}
 
-	if (!lw_client) {
+	if (!nolcd) {
 		tti->tti_off = ted->ted_lr_off;
-		rc = tgt_client_data_write(env, tgt, ted->ted_lcd, &tti->tti_off, th);
+		if (CFS_FAIL_CHECK(OBD_FAIL_TGT_RCVD_EIO))
+			rc = -EIO;
+		else
+			rc = tgt_client_data_write(env, tgt, ted->ted_lcd,
+						   &tti->tti_off, th);
 		if (rc < 0) {
 			mutex_unlock(&ted->ted_lcd_lock);
 			RETURN(rc);
 		}
 	}
 	mutex_unlock(&ted->ted_lcd_lock);
-	EXIT;
-srv_update:
-	if (update)
-		rc = tgt_server_data_write(env, tgt, th);
-	return rc;
+	RETURN(rc);
 }
 
 /*
@@ -881,16 +1688,25 @@ static int tgt_clients_data_init(const struct lu_env *env,
 	int			 cl_idx;
 	int			 rc = 0;
 	loff_t			 off = lsd->lsd_client_start;
+	__u32			 generation = 0;
+	struct cfs_hash		*hash = NULL;
 
 	ENTRY;
 
-	CLASSERT(offsetof(struct lsd_client_data, lcd_padding) +
-		 sizeof(lcd->lcd_padding) == LR_CLIENT_SIZE);
+	if (tgt->lut_bottom->dd_rdonly)
+		RETURN(0);
+
+	BUILD_BUG_ON(offsetof(struct lsd_client_data, lcd_padding) +
+		     sizeof(lcd->lcd_padding) != LR_CLIENT_SIZE);
 
 	OBD_ALLOC_PTR(lcd);
 	if (lcd == NULL)
 		RETURN(-ENOMEM);
 
+	hash = cfs_hash_getref(tgt->lut_obd->obd_gen_hash);
+	if (hash == NULL)
+		GOTO(err_out, rc = -ENODEV);
+
 	for (cl_idx = 0; off < last_size; cl_idx++) {
 		struct obd_export	*exp;
 		__u64			 last_transno;
@@ -919,9 +1735,10 @@ static int tgt_clients_data_init(const struct lu_env *env,
 		/* These exports are cleaned up by disconnect, so they
 		 * need to be set up like real exports as connect does.
 		 */
-		CDEBUG(D_HA, "RCVRNG CLIENT uuid: %s idx: %d lr: "LPU64
-		       " srv lr: "LPU64" lx: "LPU64"\n", lcd->lcd_uuid, cl_idx,
-		       last_transno, lsd->lsd_last_transno, lcd_last_xid(lcd));
+		CDEBUG(D_HA, "RCVRNG CLIENT uuid: %s idx: %d lr: %llu"
+		       " srv lr: %llu lx: %llu gen %u\n", lcd->lcd_uuid,
+		       cl_idx, last_transno, lsd->lsd_last_transno,
+		       lcd_last_xid(lcd), lcd->lcd_generation);
 
 		exp = class_new_export(obd, (struct obd_uuid *)lcd->lcd_uuid);
 		if (IS_ERR(exp)) {
@@ -945,11 +1762,35 @@ static int tgt_clients_data_init(const struct lu_env *env,
 		exp->exp_connecting = 0;
 		exp->exp_in_recovery = 0;
 		spin_unlock(&exp->exp_lock);
-		obd->obd_max_recoverable_clients++;
+		atomic_inc(&obd->obd_max_recoverable_clients);
+
+		if (tgt_is_multimodrpcs_record(tgt, lcd)) {
+			atomic_inc(&tgt->lut_num_clients);
+
+			/* compute the highest valid client generation */
+			generation = max(generation, lcd->lcd_generation);
+			/* fill client_generation <-> export hash table */
+			rc = cfs_hash_add_unique(hash, &lcd->lcd_generation,
+						 &exp->exp_gen_hash);
+			if (rc != 0) {
+				CERROR("%s: duplicate export for client "
+				       "generation %u\n",
+				       tgt_name(tgt), lcd->lcd_generation);
+				class_export_put(exp);
+				GOTO(err_out, rc);
+			}
+		}
+
 		class_export_put(exp);
 
+		rc = rev_import_init(exp);
+		if (rc != 0) {
+			class_unlink_export(exp);
+			GOTO(err_out, rc);
+		}
+
 		/* Need to check last_rcvd even for duplicated exports. */
-		CDEBUG(D_OTHER, "client at idx %d has last_transno = "LPU64"\n",
+		CDEBUG(D_OTHER, "client at idx %d has last_transno = %llu\n",
 		       cl_idx, last_transno);
 
 		spin_lock(&tgt->lut_translock);
@@ -958,7 +1799,12 @@ static int tgt_clients_data_init(const struct lu_env *env,
 		spin_unlock(&tgt->lut_translock);
 	}
 
+	/* record highest valid client generation */
+	atomic_set(&tgt->lut_client_generation, generation);
+
 err_out:
+	if (hash != NULL)
+		cfs_hash_putref(hash);
 	OBD_FREE_PTR(lcd);
 	RETURN(rc);
 }
@@ -975,16 +1821,17 @@ static struct server_compat_data tgt_scd[] = {
 		.rocompat = OBD_ROCOMPAT_LOVOBJID,
 		.incompat = OBD_INCOMPAT_MDT | OBD_INCOMPAT_COMMON_LR |
 			    OBD_INCOMPAT_FID | OBD_INCOMPAT_IAM_DIR |
-			    OBD_INCOMPAT_LMM_VER | OBD_INCOMPAT_MULTI_OI,
+			    OBD_INCOMPAT_LMM_VER | OBD_INCOMPAT_MULTI_OI |
+			    OBD_INCOMPAT_MULTI_RPCS,
 		.rocinit = OBD_ROCOMPAT_LOVOBJID,
 		.incinit = OBD_INCOMPAT_MDT | OBD_INCOMPAT_COMMON_LR |
 			   OBD_INCOMPAT_MULTI_OI,
 	},
 	[LDD_F_SV_TYPE_OST] = {
-		.rocompat = 0,
+		.rocompat = OBD_ROCOMPAT_IDX_IN_IDIF,
 		.incompat = OBD_INCOMPAT_OST | OBD_INCOMPAT_COMMON_LR |
 			    OBD_INCOMPAT_FID,
-		.rocinit = 0,
+		.rocinit = OBD_ROCOMPAT_IDX_IN_IDIF,
 		.incinit = OBD_INCOMPAT_OST | OBD_INCOMPAT_COMMON_LR,
 	}
 };
@@ -997,15 +1844,15 @@ int tgt_server_data_init(const struct lu_env *env, struct lu_target *tgt)
 	__u32				 index;
 	int				 rc, type;
 
-	rc = dt_attr_get(env, tgt->lut_last_rcvd, &tti->tti_attr, BYPASS_CAPA);
+	rc = dt_attr_get(env, tgt->lut_last_rcvd, &tti->tti_attr);
 	if (rc)
 		RETURN(rc);
 
 	last_rcvd_size = (unsigned long)tti->tti_attr.la_size;
 
 	/* ensure padding in the struct is the correct size */
-	CLASSERT(offsetof(struct lr_server_data, lsd_padding) +
-		 sizeof(lsd->lsd_padding) == LR_SERVER_SIZE);
+	BUILD_BUG_ON(offsetof(struct lr_server_data, lsd_padding) +
+		     sizeof(lsd->lsd_padding) != LR_SERVER_SIZE);
 
 	rc = server_name2index(tgt_name(tgt), &index, NULL);
 	if (rc < 0) {
@@ -1048,19 +1895,29 @@ int tgt_server_data_init(const struct lu_env *env, struct lu_target *tgt)
 			RETURN(rc);
 		}
 		if (strcmp(lsd->lsd_uuid, tgt->lut_obd->obd_uuid.uuid)) {
-			LCONSOLE_ERROR_MSG(0x157, "Trying to start OBD %s "
-					   "using the wrong disk %s. Were the"
-					   " /dev/ assignments rearranged?\n",
-					   tgt->lut_obd->obd_uuid.uuid,
-					   lsd->lsd_uuid);
-			RETURN(-EINVAL);
+			if (tgt->lut_bottom->dd_rdonly) {
+				/* Such difference may be caused by mounting
+				 * up snapshot with new fsname under rd_only
+				 * mode. But even if it was NOT, it will not
+				 * damage the system because of "rd_only". */
+				memcpy(lsd->lsd_uuid,
+				       tgt->lut_obd->obd_uuid.uuid,
+				       sizeof(lsd->lsd_uuid));
+			} else {
+				LCONSOLE_ERROR_MSG(0x157, "Trying to start "
+						   "OBD %s using the wrong "
+						   "disk %s. Were the /dev/ "
+						   "assignments rearranged?\n",
+						   tgt->lut_obd->obd_uuid.uuid,
+						   lsd->lsd_uuid);
+				RETURN(-EINVAL);
+			}
 		}
 
 		if (lsd->lsd_osd_index != index) {
-			LCONSOLE_ERROR_MSG(0x157, "%s: index %d in last rcvd "
-					   "is different with the index %d in"
-					   "config log, It might be disk"
-					   "corruption!\n", tgt_name(tgt),
+			LCONSOLE_ERROR_MSG(0x157,
+					   "%s: index %d in last rcvd is different with the index %d in config log, It might be disk corruption!\n",
+					   tgt_name(tgt),
 					   lsd->lsd_osd_index, index);
 			RETURN(-EINVAL);
 		}
@@ -1089,8 +1946,8 @@ int tgt_server_data_init(const struct lu_env *env, struct lu_target *tgt)
 			LCONSOLE_WARN("%s: mounting at first time on 1.8 FS, "
 				      "remove all clients for interop needs\n",
 				      tgt_name(tgt));
-			rc = tgt_truncate_last_rcvd(env, tgt,
-						    lsd->lsd_client_start);
+			rc = tgt_truncate_object(env, tgt, tgt->lut_last_rcvd,
+						 lsd->lsd_client_start);
 			if (rc)
 				RETURN(rc);
 			last_rcvd_size = lsd->lsd_client_start;
@@ -1106,9 +1963,9 @@ int tgt_server_data_init(const struct lu_env *env, struct lu_target *tgt)
 	lsd->lsd_mount_count++;
 
 	CDEBUG(D_INODE, "=======,=BEGIN DUMPING LAST_RCVD========\n");
-	CDEBUG(D_INODE, "%s: server last_transno: "LPU64"\n",
+	CDEBUG(D_INODE, "%s: server last_transno: %llu\n",
 	       tgt_name(tgt), tgt->lut_last_transno);
-	CDEBUG(D_INODE, "%s: server mount_count: "LPU64"\n",
+	CDEBUG(D_INODE, "%s: server mount_count: %llu\n",
 	       tgt_name(tgt), lsd->lsd_mount_count);
 	CDEBUG(D_INODE, "%s: server data size: %u\n",
 	       tgt_name(tgt), lsd->lsd_server_size);
@@ -1145,8 +2002,8 @@ int tgt_server_data_init(const struct lu_env *env, struct lu_target *tgt)
 	tgt->lut_obd->obd_last_committed = tgt->lut_last_transno;
 	spin_unlock(&tgt->lut_translock);
 
-	tgt->lut_obd->u.obt.obt_mount_count = lsd->lsd_mount_count;
-	tgt->lut_obd->u.obt.obt_instance = (__u32)lsd->lsd_mount_count;
+	obd2obt(tgt->lut_obd)->obt_mount_count = lsd->lsd_mount_count;
+	obd2obt(tgt->lut_obd)->obt_instance = (__u32)lsd->lsd_mount_count;
 
 	/* save it, so mount count and last_transno is current */
 	rc = tgt_server_data_update(env, tgt, 0);
@@ -1167,8 +2024,17 @@ int tgt_txn_start_cb(const struct lu_env *env, struct thandle *th,
 	struct lu_target	*tgt = cookie;
 	struct tgt_session_info	*tsi;
 	struct tgt_thread_info	*tti = tgt_th_info(env);
+	struct dt_object	*dto;
 	int			 rc;
 
+	/* For readonly case, the caller should have got failure
+	 * when start the transaction. If the logic comes here,
+	 * there must be something wrong. */
+	if (unlikely(tgt->lut_bottom->dd_rdonly)) {
+		dump_stack();
+		LBUG();
+	}
+
 	/* if there is no session, then this transaction is not result of
 	 * request processing but some local operation */
 	if (env->le_ses == NULL)
@@ -1180,23 +2046,37 @@ int tgt_txn_start_cb(const struct lu_env *env, struct thandle *th,
 	if (tsi->tsi_exp == NULL)
 		return 0;
 
-	tti_buf_lcd(tti);
-	rc = dt_declare_record_write(env, tgt->lut_last_rcvd,
-				     &tti->tti_buf,
-				     tsi->tsi_exp->exp_target_data.ted_lr_off,
-				     th);
-	if (rc)
-		return rc;
-
-	tti_buf_lsd(tti);
-	rc = dt_declare_record_write(env, tgt->lut_last_rcvd,
-				     &tti->tti_buf, 0, th);
-	if (rc)
-		return rc;
+	if (tgt_is_multimodrpcs_client(tsi->tsi_exp)) {
+		/*
+		 * Use maximum possible file offset for declaration to ensure
+		 * ZFS will reserve enough credits for a write anywhere in this
+		 * file, since we don't know where in the file the write will be
+		 * because a replay slot has not been assigned.  This should be
+		 * replaced by dmu_tx_hold_append() when available.
+		 */
+		tti->tti_buf.lb_buf = NULL;
+		tti->tti_buf.lb_len = sizeof(struct lsd_reply_data);
+		dto = dt_object_locate(tgt->lut_reply_data, th->th_dev);
+		rc = dt_declare_record_write(env, dto, &tti->tti_buf, -1, th);
+		if (rc)
+			return rc;
+	} else {
+		dto = dt_object_locate(tgt->lut_last_rcvd, th->th_dev);
+		tti_buf_lcd(tti);
+		tti->tti_off = tsi->tsi_exp->exp_target_data.ted_lr_off;
+		rc = dt_declare_record_write(env, dto, &tti->tti_buf,
+					     tti->tti_off, th);
+		if (rc)
+			return rc;
+	}
 
 	if (tsi->tsi_vbr_obj != NULL &&
-	    !lu_object_remote(&tsi->tsi_vbr_obj->do_lu))
-		rc = dt_declare_version_set(env, tsi->tsi_vbr_obj, th);
+	    !lu_object_remote(&tsi->tsi_vbr_obj->do_lu)) {
+		dto = dt_object_locate(tsi->tsi_vbr_obj, th->th_dev);
+		rc = dt_declare_version_set(env, dto, th);
+		if (!rc && tsi->tsi_dv_update)
+			rc = dt_declare_data_version_set(env, dto, th);
+	}
 
 	return rc;
 }
@@ -1220,20 +2100,31 @@ int tgt_txn_stop_cb(const struct lu_env *env, struct thandle *th,
 	if (tsi->tsi_exp == NULL)
 		return 0;
 
-	echo_client = (tgt_ses_req(tsi) == NULL);
+	echo_client = (tgt_ses_req(tsi) == NULL && tsi->tsi_xid == 0);
 
-	if (tti->tti_has_trans && !echo_client) {
-		if (tti->tti_mult_trans == 0) {
-			CDEBUG(D_HA, "More than one transaction "LPU64"\n",
+	if (tsi->tsi_has_trans && !echo_client && !tsi->tsi_batch_env) {
+		if (!tsi->tsi_mult_trans) {
+			CDEBUG(D_HA, "More than one transaction %llu\n",
 			       tti->tti_transno);
-			RETURN(0);
+			/**
+			 * if RPC handler sees unexpected multiple last_rcvd
+			 * updates with transno, then it is better to return
+			 * the latest transaction number to the client.
+			 * In that case replay may fail if part of operation
+			 * was committed and can't be re-applied easily. But
+			 * that is better than report the first transno, in
+			 * which case partially committed operation would be
+			 * considered as finished so never replayed causing
+			 * data loss.
+			 */
 		}
-		/* we need another transno to be assigned */
+		/* we need new transno to be assigned */
 		tti->tti_transno = 0;
-	} else if (th->th_result == 0) {
-		tti->tti_has_trans = 1;
 	}
 
+	if (!th->th_result)
+		tsi->tsi_has_trans++;
+
 	if (tsi->tsi_vbr_obj != NULL &&
 	    !lu_object_remote(&tsi->tsi_vbr_obj->do_lu)) {
 		obj = tsi->tsi_vbr_obj;
@@ -1247,3 +2138,276 @@ int tgt_txn_stop_cb(const struct lu_env *env, struct thandle *th,
 					  tgt_ses_req(tsi));
 	return rc;
 }
+
+int tgt_reply_data_init(const struct lu_env *env, struct lu_target *tgt)
+{
+	struct tgt_thread_info	*tti = tgt_th_info(env);
+	struct lsd_reply_data	*lrd = &tti->tti_lrd;
+	unsigned long		 reply_data_size;
+	int			 rc;
+	struct lsd_reply_header	*lrh = &tgt->lut_reply_header;
+	struct tg_reply_data	*trd = NULL;
+	int                      idx;
+	loff_t			 off;
+	struct cfs_hash		*hash = NULL;
+	struct obd_export	*exp;
+	struct tg_export_data   *ted;
+	int			 reply_data_recovered = 0;
+
+	rc = dt_attr_get(env, tgt->lut_reply_data, &tti->tti_attr);
+	if (rc)
+		GOTO(out, rc);
+	reply_data_size = (unsigned long)tti->tti_attr.la_size;
+
+	if (reply_data_size == 0) {
+		CDEBUG(D_INFO, "%s: new reply_data file, initializing\n",
+		       tgt_name(tgt));
+		lrh->lrh_magic = LRH_MAGIC;
+		lrh->lrh_header_size = sizeof(struct lsd_reply_header);
+		if (lrh->lrh_magic == LRH_MAGIC_V1)
+			lrh->lrh_reply_size = sizeof(struct lsd_reply_data_v1);
+		else
+			lrh->lrh_reply_size = sizeof(struct lsd_reply_data_v2);
+		rc = tgt_reply_header_write(env, tgt, lrh);
+		if (rc) {
+			CERROR("%s: error writing %s: rc = %d\n",
+			       tgt_name(tgt), REPLY_DATA, rc);
+			GOTO(out, rc);
+		}
+	} else {
+		__u32 recsz = sizeof(*lrd);
+		const char *lrd_ver = "v2";
+
+		rc = tgt_reply_header_read(env, tgt, lrh);
+		if (rc) {
+			CERROR("%s: error reading %s: rc = %d\n",
+			       tgt_name(tgt), REPLY_DATA, rc);
+			GOTO(out, rc);
+		}
+
+		switch (lrh->lrh_magic) {
+#if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(3, 5, 53, 0)
+		/* The old reply_data is replaced on the first mount after
+		 * an upgrade, so no need to keep this interop code forever.
+		 */
+		case LRH_MAGIC_V1:
+			recsz = sizeof(struct lsd_reply_data_v1);
+			lrd_ver = "v1";
+
+			if (lrh->lrh_magic != LRH_MAGIC)
+				CWARN("%s: %s record size will be %s\n",
+				      tgt_name(tgt), REPLY_DATA,
+				      lrh->lrh_magic < LRH_MAGIC ?
+				      "upgraded" : "downgraded");
+			fallthrough;
+#endif
+		case LRH_MAGIC_V2:
+			if (lrh->lrh_header_size != sizeof(*lrh)) {
+				CERROR("%s: bad %s %s header size: %u != %lu\n",
+				       tgt_name(tgt), REPLY_DATA, lrd_ver,
+				       lrh->lrh_header_size, sizeof(*lrh));
+				GOTO(out, rc = -EINVAL);
+			}
+			if (lrh->lrh_reply_size != recsz) {
+				CERROR("%s: bad %s %s reply size: %u != %u\n",
+				tgt_name(tgt), REPLY_DATA, lrd_ver,
+				lrh->lrh_reply_size, recsz);
+				GOTO(out, rc = -EINVAL);
+			}
+			break;
+		default:
+			CERROR("%s: invalid %s magic: %x != %x/%x\n",
+			       tgt_name(tgt), REPLY_DATA,
+			       lrh->lrh_magic, LRH_MAGIC_V1, LRH_MAGIC_V2);
+			GOTO(out, rc = -EINVAL);
+		}
+
+		hash = cfs_hash_getref(tgt->lut_obd->obd_gen_hash);
+		if (hash == NULL)
+			GOTO(out, rc = -ENODEV);
+
+		OBD_ALLOC_PTR(trd);
+		if (trd == NULL)
+			GOTO(out, rc = -ENOMEM);
+
+		/* Load reply_data from disk */
+		for (idx = 0, off = lrh->lrh_header_size;
+		     off < reply_data_size; idx++, off += recsz) {
+			rc = tgt_reply_data_read(env, tgt, lrd, off, lrh);
+			if (rc) {
+				CERROR("%s: error reading %s: rc = %d\n",
+				       tgt_name(tgt), REPLY_DATA, rc);
+				GOTO(out, rc);
+			}
+
+			exp = cfs_hash_lookup(hash, &lrd->lrd_client_gen);
+			if (exp == NULL) {
+				/* old reply data from a disconnected client */
+				continue;
+			}
+			ted = &exp->exp_target_data;
+			mutex_lock(&ted->ted_lcd_lock);
+
+			/* create in-memory reply_data and link it to
+			 * target export's reply list */
+			rc = tgt_set_reply_slot(tgt, idx);
+			if (rc != 0) {
+				mutex_unlock(&ted->ted_lcd_lock);
+				GOTO(out, rc);
+			}
+			trd->trd_reply = *lrd;
+			trd->trd_pre_versions[0] = 0;
+			trd->trd_pre_versions[1] = 0;
+			trd->trd_pre_versions[2] = 0;
+			trd->trd_pre_versions[3] = 0;
+			trd->trd_index = idx;
+			trd->trd_tag = 0;
+			fid_zero(&trd->trd_object);
+			list_add(&trd->trd_list, &ted->ted_reply_list);
+			ted->ted_reply_cnt++;
+			if (ted->ted_reply_cnt > ted->ted_reply_max)
+				ted->ted_reply_max = ted->ted_reply_cnt;
+
+			CDEBUG(D_HA, "%s: restore reply %p: xid %llu, "
+			       "transno %llu, client gen %u, slot idx %d\n",
+			       tgt_name(tgt), trd, lrd->lrd_xid,
+			       lrd->lrd_transno, lrd->lrd_client_gen,
+			       trd->trd_index);
+
+			/* update export last committed transation */
+			exp->exp_last_committed = max(exp->exp_last_committed,
+						      lrd->lrd_transno);
+			/* Update lcd_last_transno as well for check in
+			 * tgt_release_reply_data() or the latest client
+			 * transno can be lost.
+			 */
+			ted->ted_lcd->lcd_last_transno =
+				max(ted->ted_lcd->lcd_last_transno,
+				    exp->exp_last_committed);
+
+			mutex_unlock(&ted->ted_lcd_lock);
+			class_export_put(exp);
+
+			/* update target last committed transaction */
+			spin_lock(&tgt->lut_translock);
+			tgt->lut_last_transno = max(tgt->lut_last_transno,
+						    lrd->lrd_transno);
+			spin_unlock(&tgt->lut_translock);
+
+			reply_data_recovered++;
+
+			OBD_ALLOC_PTR(trd);
+			if (trd == NULL)
+				GOTO(out, rc = -ENOMEM);
+		}
+		CDEBUG(D_INFO, "%s: %d reply data have been recovered\n",
+		       tgt_name(tgt), reply_data_recovered);
+	}
+
+	spin_lock(&tgt->lut_translock);
+	/* obd_last_committed is used for compatibility
+	 * with other lustre recovery code */
+	tgt->lut_obd->obd_last_committed = tgt->lut_last_transno;
+	spin_unlock(&tgt->lut_translock);
+
+	rc = 0;
+
+out:
+	if (hash != NULL)
+		cfs_hash_putref(hash);
+	if (trd != NULL)
+		OBD_FREE_PTR(trd);
+	return rc;
+}
+
+static int tgt_check_lookup_req(struct ptlrpc_request *req, int lookup,
+				struct tg_reply_data *trd)
+{
+	struct tg_export_data *ted = &req->rq_export->exp_target_data;
+	struct lu_target *lut = class_exp2tgt(req->rq_export);
+	__u16 tag = lustre_msg_get_tag(req->rq_reqmsg);
+	int rc = 0;
+	struct tg_reply_data *reply;
+	bool check_increasing;
+
+	if (tag == 0)
+		return 0;
+
+	check_increasing = tgt_is_increasing_xid_client(req->rq_export) &&
+			   !(lustre_msg_get_flags(req->rq_reqmsg) & MSG_REPLAY);
+	if (!lookup && !check_increasing)
+		return 0;
+
+	list_for_each_entry(reply, &ted->ted_reply_list, trd_list) {
+		if (lookup && reply->trd_reply.lrd_xid == req->rq_xid) {
+			rc = 1;
+			if (trd != NULL)
+				*trd = *reply;
+			break;
+		} else if (check_increasing && reply->trd_tag == tag &&
+			   reply->trd_reply.lrd_xid > req->rq_xid) {
+			rc = -EPROTO;
+			CERROR("%s: busy tag=%u req_xid=%llu, trd=%p: xid=%llu transno=%llu client_gen=%u slot_idx=%d: rc = %d\n",
+			       tgt_name(lut), tag, req->rq_xid, trd,
+			       reply->trd_reply.lrd_xid,
+			       reply->trd_reply.lrd_transno,
+			       reply->trd_reply.lrd_client_gen,
+			       reply->trd_index, rc);
+			break;
+		}
+	}
+
+	return rc;
+}
+
+/* Look for a reply data matching specified request @req
+ * A copy is returned in @trd if the pointer is not NULL
+ */
+int tgt_lookup_reply(struct ptlrpc_request *req, struct tg_reply_data *trd)
+{
+	struct tg_export_data *ted = &req->rq_export->exp_target_data;
+	int found = 0;
+	bool not_replay = !(lustre_msg_get_flags(req->rq_reqmsg) & MSG_REPLAY);
+
+	mutex_lock(&ted->ted_lcd_lock);
+	if (not_replay && req->rq_xid <= req->rq_export->exp_last_xid) {
+		/* A check for the last_xid is needed here in case there is
+		 * no reply data is left in the list. It may happen if another
+		 * RPC on another slot increased the last_xid between our
+		 * process_req_last_xid & tgt_lookup_reply calls */
+		found = -EPROTO;
+	} else {
+		found = tgt_check_lookup_req(req, 1, trd);
+	}
+	mutex_unlock(&ted->ted_lcd_lock);
+
+	CDEBUG(D_TRACE, "%s: lookup reply xid %llu, found %d last_xid %llu\n",
+	       tgt_name(class_exp2tgt(req->rq_export)), req->rq_xid, found,
+	       req->rq_export->exp_last_xid);
+
+	return found;
+}
+EXPORT_SYMBOL(tgt_lookup_reply);
+
+int tgt_handle_received_xid(struct obd_export *exp, __u64 rcvd_xid)
+{
+	struct tg_export_data	*ted = &exp->exp_target_data;
+	struct lu_target	*lut = class_exp2tgt(exp);
+	struct tg_reply_data	*trd, *tmp;
+
+
+	list_for_each_entry_safe(trd, tmp, &ted->ted_reply_list, trd_list) {
+		if (trd->trd_reply.lrd_xid > rcvd_xid)
+			continue;
+		ted->ted_release_xid++;
+		tgt_release_reply_data(lut, ted, trd);
+	}
+
+	return 0;
+}
+
+int tgt_handle_tag(struct ptlrpc_request *req)
+{
+	return tgt_check_lookup_req(req, 0, NULL);
+}
+