diff --git a/lib/iscsi/conn.c b/lib/iscsi/conn.c
index 148cc4f8f..0f236e4c3 100644
--- a/lib/iscsi/conn.c
+++ b/lib/iscsi/conn.c
@@ -1100,13 +1100,13 @@ spdk_iscsi_conn_flush_pdus_internal(struct spdk_iscsi_conn *conn)
 						if (pdu->task->scsi.offset > 0) {
 							conn->data_in_cnt--;
 							if (pdu->bhs.flags & ISCSI_DATAIN_STATUS) {
+								/* Free the primary task after the last subtask done */
+								conn->data_in_cnt--;
 								spdk_iscsi_task_put(spdk_iscsi_task_get_primary(pdu->task));
 							}
+							spdk_iscsi_conn_handle_queued_tasks(conn);
 						}
-
-						spdk_iscsi_conn_handle_queued_datain(conn);
 					}
-
 					spdk_iscsi_task_put(pdu->task);
 				}
 				spdk_put_pdu(pdu);
diff --git a/lib/iscsi/iscsi.c b/lib/iscsi/iscsi.c
index 6ab957ea0..193933735 100644
--- a/lib/iscsi/iscsi.c
+++ b/lib/iscsi/iscsi.c
@@ -2838,15 +2838,21 @@ static void spdk_iscsi_queue_mgmt_task(struct spdk_iscsi_conn *conn,
 	spdk_scsi_dev_queue_mgmt_task(conn->dev, &task->scsi);
 }
 
-int spdk_iscsi_conn_handle_queued_datain(struct spdk_iscsi_conn *conn)
+int spdk_iscsi_conn_handle_queued_tasks(struct spdk_iscsi_conn *conn)
 {
 	struct spdk_iscsi_task *task;
 
 	while (!TAILQ_EMPTY(&conn->queued_datain_tasks) &&
-	       conn->data_in_cnt < MAX_EXTRA_DATAIN_PER_CONNECTION) {
+	       conn->data_in_cnt < MAX_LARGE_DATAIN_PER_CONNECTION) {
 		task = TAILQ_FIRST(&conn->queued_datain_tasks);
 		assert(task->current_datain_offset <= task->scsi.transfer_len);
 
+		if (task->current_datain_offset == 0) {
+			task->current_datain_offset = task->scsi.length;
+			conn->data_in_cnt++;
+			spdk_iscsi_queue_task(conn, task);
+			continue;
+		}
 		if (task->current_datain_offset < task->scsi.transfer_len) {
 			struct spdk_iscsi_task *subtask;
 			uint32_t remaining_size = 0;
@@ -2879,16 +2885,18 @@ static int spdk_iscsi_op_scsi_read(struct spdk_iscsi_conn *conn,
 	task->scsi.offset = 0;
 	task->scsi.length = DMIN32(SPDK_BDEV_LARGE_RBUF_MAX_SIZE, task->scsi.transfer_len);
 	task->scsi.rbuf = NULL;
-	spdk_iscsi_queue_task(conn, task);
 
 	remaining_size = task->scsi.transfer_len - task->scsi.length;
-	task->current_datain_offset = task->scsi.length;
+	task->current_datain_offset = 0;
 
-	if (remaining_size > 0) {
-		TAILQ_INSERT_TAIL(&conn->queued_datain_tasks, task, link);
-		return spdk_iscsi_conn_handle_queued_datain(conn);
+	if (remaining_size == 0) {
+		spdk_iscsi_queue_task(conn, task);
+		return 0;
 	}
-	return 0;
+
+	TAILQ_INSERT_TAIL(&conn->queued_datain_tasks, task, link);
+
+	return spdk_iscsi_conn_handle_queued_tasks(conn);
 }
 
 static int
diff --git a/lib/iscsi/iscsi.h b/lib/iscsi/iscsi.h
index 8c6a1d479..a101c34fa 100644
--- a/lib/iscsi/iscsi.h
+++ b/lib/iscsi/iscsi.h
@@ -107,13 +107,12 @@
 
 /*
  * Defines maximum number of data in buffers each connection can have in
- *  use at any given time.  An "extra data in buffer" means any buffer after
- *  the first for the iSCSI I/O command.  So this limit does not affect I/O
- *  smaller than SPDK_ISCSI_MAX_SEND_DATA_SEGMENT_LENGTH.
+ *  use at any given time. So this limit does not affect I/O smaller than
+ *  SPDK_BDEV_SMALL_RBUF_MAX_SIZE.
  */
-#define MAX_EXTRA_DATAIN_PER_CONNECTION 64
+#define MAX_LARGE_DATAIN_PER_CONNECTION 64
 
-#define NUM_PDU_PER_CONNECTION	(2 * (SPDK_ISCSI_MAX_QUEUE_DEPTH + MAX_EXTRA_DATAIN_PER_CONNECTION + 8))
+#define NUM_PDU_PER_CONNECTION	(2 * (SPDK_ISCSI_MAX_QUEUE_DEPTH + MAX_LARGE_DATAIN_PER_CONNECTION + 8))
 
 #define SPDK_ISCSI_MAX_BURST_LENGTH	\
 		(SPDK_ISCSI_MAX_RECV_DATA_SEGMENT_LENGTH * MAX_DATA_OUT_PER_CONNECTION)
@@ -358,7 +357,7 @@ void process_task_mgmt_completion(spdk_event_t event);
 /* Memory management */
 void spdk_put_pdu(struct spdk_iscsi_pdu *pdu);
 struct spdk_iscsi_pdu *spdk_get_pdu(void);
-int spdk_iscsi_conn_handle_queued_datain(struct spdk_iscsi_conn *conn);
+int spdk_iscsi_conn_handle_queued_tasks(struct spdk_iscsi_conn *conn);
 
 static inline int
 spdk_get_immediate_data_buffer_size(void)