[openib-general] [PATCH 3/6] [RFC] iser initiator

Or Gerlitz ogerlitz at voltaire.com
Wed Feb 22 06:32:56 PST 2006


+ the main entry points to this code are iser_send_control/command/dataout
  for flow coming from iscsi_iser.c and iser_rcv_completion for handilg 
  completions towards iscsi_iser.c

--- /ulp/iser-x/iser_initiator.c	2006-02-22 15:06:56.000000000 +0200
+++ /ulp/iser/iser_initiator.c	2006-02-22 13:48:55.000000000 +0200
@@ -1 +1,743 @@
+/*
+ * Copyright (c) 2004, 2005, 2006 Voltaire, Inc. All rights reserved.
+ *
+ * This software is available to you under a choice of one of two
+ * licenses.  You may choose to be licensed under the terms of the GNU
+ * General Public License (GPL) Version 2, available from the file
+ * COPYING in the main directory of this source tree, or the
+ * OpenIB.org BSD license below:
+ *
+ *     Redistribution and use in source and binary forms, with or
+ *     without modification, are permitted provided that the following
+ *     conditions are met:
+ *
+ *	- Redistributions of source code must retain the above
+ *	  copyright notice, this list of conditions and the following
+ *	  disclaimer.
+ *
+ *	- Redistributions in binary form must reproduce the above
+ *	  copyright notice, this list of conditions and the following
+ *	  disclaimer in the documentation and/or other materials
+ *	  provided with the distribution.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ *
+ * $Id: iser_initiator.c 5459 2006-02-22 11:00:48Z ogerlitz $
+ */
+#include <linux/kernel.h>
+#include <linux/slab.h>
+#include <linux/mm.h>
+#include <asm/io.h>
+#include <asm/scatterlist.h>
+#include <scsi/scsi_cmnd.h>
+
+#include "iscsi_iser.h"
+
+/* Constant PDU lengths calculations */
+#define ISER_HDR_LEN            sizeof (struct iser_hdr)
+#define ISER_PDU_BHS_LENGTH     sizeof (struct iscsi_hdr)
+#define ISER_TOTAL_HEADERS_LEN  (ISER_HDR_LEN + ISER_PDU_BHS_LENGTH)
+
+#define USE_OFFSET(offset)      (offset)
+#define USE_NO_OFFSET           0
+#define USE_SIZE(size)          (size)
+#define USE_ENTIRE_SIZE         0
+
+/* iser_dto_add_regd_buff - increments the reference count for *
+ * the registered buffer & adds it to the DTO object           */
+static void iser_dto_add_regd_buff(struct iser_dto *p_dto,
+				   struct iser_regd_buf *p_regd_buf,
+				   unsigned long use_offset,
+				   unsigned long use_size)
+{
+	int add_idx;
+
+	atomic_inc(&p_regd_buf->ref_count);
+
+	add_idx = p_dto->regd_vector_len;
+	p_dto->regd[add_idx] = p_regd_buf;
+	p_dto->used_sz[add_idx] = use_size;
+	p_dto->offset[add_idx] = use_offset;
+
+	p_dto->regd_vector_len++;
+}
+
+static int iser_dma_map_task_data(struct iscsi_iser_cmd_task *p_iser_task,
+				  struct iser_data_buf       *p_data,
+				  enum   iser_data_dir       iser_dir,
+				  enum   dma_data_direction  dma_dir)
+{
+	struct device *dma_device;
+	dma_addr_t     dma_addr;
+	int            dma_nents;
+
+	p_iser_task->dir[iser_dir] = 1;
+	dma_device = p_iser_task->conn->ib_conn->p_adaptor->device->dma_device;
+
+	if (p_data->type == ISER_BUF_TYPE_SINGLE) {
+		p_iser_task->data_len[iser_dir] = p_data->size;
+		dma_addr = dma_map_single(dma_device,p_data->p_buf, p_data->size,
+					  dma_dir);
+		if (dma_mapping_error(dma_addr)) {
+			iser_err("dma_map_single failed at %p\n", p_data->p_buf);
+			return -EINVAL;
+		}
+		p_data->dma_addr = dma_addr;
+	} else {
+		dma_nents = dma_map_sg(dma_device, p_data->p_buf, p_data->size,
+				       dma_dir);
+		if (dma_nents == 0) {
+			iser_err("dma_map_sg failed!!!\n");
+			return -EINVAL;
+		}
+		p_data->dma_nents = dma_nents;
+		p_iser_task->data_len[iser_dir] = iser_sg_size(p_data);
+	}
+	return 0;
+}
+
+static void iser_dma_unmap_task_data(struct iscsi_iser_cmd_task *p_iser_task)
+{
+	struct device  *dma_device;
+	struct iser_data_buf *p_data;
+
+	dma_device = p_iser_task->conn->ib_conn->p_adaptor->device->dma_device;
+
+	p_data = &p_iser_task->data[ISER_DIR_IN];
+	if (p_data->p_buf != NULL && p_data->type == ISER_BUF_TYPE_SCATTERLIST)
+		dma_unmap_sg(dma_device, p_data->p_buf, p_data->size,
+			     DMA_FROM_DEVICE);
+	else if (p_data->p_buf != NULL) /* p_data->type == ISER_BUF_TYPE_SINGLE */
+		dma_unmap_single(dma_device, p_data->dma_addr, p_data->size,
+				 DMA_FROM_DEVICE);
+
+	p_data = &p_iser_task->data[ISER_DIR_OUT];
+	if (p_data->p_buf != NULL && p_data->type == ISER_BUF_TYPE_SCATTERLIST)
+		dma_unmap_sg(dma_device, p_data->p_buf, p_data->size,
+			     DMA_TO_DEVICE);
+	else if (p_data->p_buf != NULL) /* p_data->type == ISER_BUF_TYPE_SINGLE */
+		dma_unmap_single(dma_device, p_data->dma_addr, p_data->size,
+				 DMA_TO_DEVICE);
+}
+
+/* Register user buffer memory and initialize passive rdma
+ *  dto descriptor. Total data size is stored in
+ *  p_iser_task->data_len[ISER_DIR_IN].
+ */
+static int iser_prepare_read_cmd(struct iscsi_iser_cmd_task *p_iser_task,
+				 struct iser_data_buf *buf_in,
+				 unsigned int edtl)
+
+{
+	struct iser_regd_buf *p_regd_buf;
+	int err;
+	struct iser_hdr *hdr = &p_iser_task->desc.iser_header;
+
+	err = iser_dma_map_task_data(p_iser_task,
+				     buf_in,
+				     ISER_DIR_IN,
+				     DMA_FROM_DEVICE);
+	if (err)
+		return err;
+
+	if (edtl > p_iser_task->data_len[ISER_DIR_IN]) {
+		iser_err("Total data length: %ld, less than EDTL: "
+			 "%d, in READ cmd BHS itt: %d, p_conn: 0x%p\n",
+			 p_iser_task->data_len[ISER_DIR_IN], edtl,
+			 p_iser_task->itt, p_iser_task->conn);
+		return -EINVAL;
+	}
+
+	memcpy(&p_iser_task->data[ISER_DIR_IN], buf_in,
+	       sizeof(struct iser_data_buf));
+
+	err = iser_reg_rdma_mem(p_iser_task,ISER_DIR_IN);
+	if (err) {
+		iser_err("Failed to set up Data-IN RDMA\n");
+		return err;
+	}
+	p_regd_buf = &p_iser_task->rdma_regd[ISER_DIR_IN];
+
+	hdr->flags    |= ISER_RSV;
+	hdr->read_stag = cpu_to_be32(p_regd_buf->reg.rkey);
+	hdr->read_va   = cpu_to_be64(p_regd_buf->reg.va);
+
+	iser_dbg("Cmd itt:%d READ tags RKEY:%#.4X VA:%#llX\n",
+		 p_iser_task->itt, p_regd_buf->reg.rkey,
+		 (unsigned long long)p_regd_buf->reg.va);
+
+	return 0;
+}
+
+/* Register user buffer memory and initialize passive rdma
+ *  dto descriptor. Total data size is stored in
+ *  p_iser_task->data_len[ISER_DIR_OUT].
+ */
+static int
+iser_prepare_write_cmd(struct iscsi_iser_cmd_task *p_iser_task,
+		       struct iser_data_buf *buf_out,
+		       unsigned int imm_sz,
+		       unsigned int unsol_sz,
+		       unsigned int edtl)
+{
+	struct iser_regd_buf *p_regd_buf;
+	int err;
+	struct iser_dto *p_send_dto = &p_iser_task->desc.dto;
+	struct iser_hdr *hdr = &p_iser_task->desc.iser_header;
+
+	err = iser_dma_map_task_data(p_iser_task,
+				     buf_out,
+				     ISER_DIR_OUT,
+				     DMA_TO_DEVICE);
+	if (err)
+		return err;
+
+	if (edtl > p_iser_task->data_len[ISER_DIR_OUT]) {
+		iser_err("Total data length: %ld, less than EDTL: %d, "
+			 "in WRITE cmd BHS itt: %d, p_conn: 0x%p\n",
+			 p_iser_task->data_len[ISER_DIR_OUT],
+			 edtl, p_iser_task->itt, p_iser_task->conn);
+		return -EINVAL;
+	}
+
+	memcpy(&p_iser_task->data[ISER_DIR_OUT], buf_out,
+	       sizeof(struct iser_data_buf));
+
+	err = iser_reg_rdma_mem(p_iser_task,ISER_DIR_OUT);
+	if (err != 0) {
+		iser_err("Failed to register write cmd RDMA mem\n");
+		return err;
+	}
+
+	p_regd_buf = &p_iser_task->rdma_regd[ISER_DIR_OUT];
+
+	if (unsol_sz < edtl) {
+		hdr->flags     |= ISER_WSV;
+		hdr->write_stag = cpu_to_be32(p_regd_buf->reg.rkey);
+		hdr->write_va   = cpu_to_be64(p_regd_buf->reg.va + unsol_sz);
+
+		iser_dbg("Cmd itt:%d, WRITE tags, RKEY:%#.4X "
+			 "VA:%#llX + unsol:%d\n",
+			 p_iser_task->itt, p_regd_buf->reg.rkey,
+			 (unsigned long long)p_regd_buf->reg.va, unsol_sz);
+	}
+
+	if (imm_sz > 0) {
+		iser_dbg("Cmd itt:%d, WRITE, adding imm.data sz: %d\n",
+			 p_iser_task->itt, imm_sz);
+		iser_dto_add_regd_buff(p_send_dto,
+				       p_regd_buf,
+				       USE_NO_OFFSET,
+				       USE_SIZE(imm_sz));
+	}
+
+	return 0;
+}
+
+/**
+ * iser_post_receive_control - allocates, initializes and posts receive DTO.
+ */
+static int iser_post_receive_control(struct iscsi_iser_conn *p_iser_conn)
+{
+	struct iser_desc     *rx_desc;
+	struct iser_regd_buf *p_regd_hdr;
+	struct iser_regd_buf *p_regd_data;
+	struct iser_dto      *p_recv_dto = NULL;
+	struct iser_adaptor  *p_iser_adaptor = p_iser_conn->ib_conn->p_adaptor;
+	int rx_data_size, err = 0;
+
+	rx_desc = kmem_cache_alloc(ig.desc_cache,
+				      GFP_KERNEL | __GFP_NOFAIL);
+	if (rx_desc == NULL) {
+		iser_err("Failed to alloc desc for post recv\n");
+		err = -ENOMEM;
+		goto post_receive_control_exit;
+	}
+	rx_desc->type = ISCSI_RX;
+
+	/* for the login sequence we must support rx of upto 8K */
+	if (p_iser_conn->c_stage == ISCSI_CONN_INITIAL_STAGE)
+		rx_data_size = DEFAULT_MAX_RECV_DATA_SEGMENT_LENGTH;
+	else /* FIXME till user space sets conn->max_recv_dlength correctly */
+		rx_data_size = 128;
+
+	rx_desc->data = kmalloc(rx_data_size, GFP_KERNEL | __GFP_NOFAIL);
+
+	if (rx_desc->data == NULL) {
+		iser_err("Failed to alloc data buf for post recv\n");
+		err = -ENOMEM;
+		goto post_receive_control_exit;
+
+	}
+
+	p_recv_dto = &rx_desc->dto;
+	p_recv_dto->p_conn          = p_iser_conn;
+	p_recv_dto->regd_vector_len = 0;
+
+	p_regd_hdr = &rx_desc->hdr_regd_buf;
+	memset(p_regd_hdr, 0, sizeof(struct iser_regd_buf));
+	p_regd_hdr->p_adaptor  = p_iser_adaptor;
+	p_regd_hdr->virt_addr  = rx_desc; /* == &rx_desc->iser_header */
+	p_regd_hdr->data_size  = ISER_TOTAL_HEADERS_LEN;
+
+	iser_reg_single(p_iser_adaptor, p_regd_hdr, DMA_FROM_DEVICE);
+
+	iser_dto_add_regd_buff(p_recv_dto, p_regd_hdr, USE_NO_OFFSET,
+			       USE_ENTIRE_SIZE);
+
+	p_regd_data = &rx_desc->data_regd_buf;
+	memset(p_regd_data, 0, sizeof(struct iser_regd_buf));
+	p_regd_data->p_adaptor  = p_iser_adaptor;
+	p_regd_data->virt_addr  = rx_desc->data;
+	p_regd_data->data_size  = rx_data_size;
+
+	iser_reg_single(p_iser_adaptor, p_regd_data, DMA_FROM_DEVICE);
+
+	iser_dto_add_regd_buff(p_recv_dto, p_regd_data,
+			       USE_NO_OFFSET, USE_ENTIRE_SIZE);
+
+	err = iser_post_recv(rx_desc);
+
+post_receive_control_exit:
+	if (err && rx_desc) {
+		iser_dto_buffs_release(p_recv_dto);
+		if (rx_desc->data != NULL)
+			kfree(rx_desc->data);
+		kmem_cache_free(ig.desc_cache, rx_desc);
+	}
+	return err;
+}
+
+/* creates a new tx descriptor and adds header regd buffer */
+static void iser_create_send_desc(struct iscsi_iser_conn *p_iser_conn,
+				  struct iser_desc       *tx_desc)
+{
+	struct iser_regd_buf *p_regd_hdr = &tx_desc->hdr_regd_buf;
+	struct iser_dto      *p_send_dto = &tx_desc->dto;
+
+	memset(p_regd_hdr, 0, sizeof(struct iser_regd_buf));
+	p_regd_hdr->p_adaptor  = p_iser_conn->ib_conn->p_adaptor;
+	p_regd_hdr->virt_addr  = tx_desc; /* == &tx_desc->iser_header */
+	p_regd_hdr->data_size  = ISER_TOTAL_HEADERS_LEN;
+
+	p_send_dto->p_conn          = p_iser_conn;
+	p_send_dto->notify_enable   = 1;
+	p_send_dto->regd_vector_len = 0;
+
+	memset(&tx_desc->iser_header, 0, ISER_HDR_LEN);
+	tx_desc->iser_header.flags = ISER_VER;
+
+	iser_dto_add_regd_buff(p_send_dto, p_regd_hdr,
+			       USE_NO_OFFSET, USE_ENTIRE_SIZE);
+}
+
+/**
+ *  iser_conn_set_full_featured_mode - (iSER API)
+ */
+int iser_conn_set_full_featured_mode(struct iscsi_iser_conn *p_iser_conn)
+{
+	int i, err =  0;
+	/* no need to keep it in a var, we are after login so if this should
+	 * be negotiated, by now the result should be available here */
+	int initial_post_recv_bufs_num = ISER_MAX_RX_MISC_PDUS;
+
+	iser_dbg("Initially post: %d\n", initial_post_recv_bufs_num);
+
+	/* Check that there is no posted recv or send buffers left - */
+	/* they must be consumed during the login phase */
+	if (atomic_read(&p_iser_conn->ib_conn->post_recv_buf_count) != 0)
+		iser_bug("Number of currently posted recv bufs non-zero\n");
+	if (atomic_read(&p_iser_conn->ib_conn->post_send_buf_count) != 0)
+		iser_bug("Number of currently posted send bufs non-zero\n");
+
+	/* Initial post receive buffers */
+	for (i = 0; i < initial_post_recv_bufs_num; i++) {
+		if (iser_post_receive_control(p_iser_conn) != 0) {
+			iser_err("Failed to post recv bufs at:%d conn:0x%p\n",
+				 i, p_iser_conn);
+			err = -ENOMEM;
+			goto ffeatured_mode_failure;
+		}
+	}
+	iser_dbg("Posted %d post recv bufs, conn:0x%p\n", i, p_iser_conn);
+	return 0;
+
+ffeatured_mode_failure:
+	return err;
+}
+
+static int
+iser_check_xmit(struct iscsi_iser_conn  *conn, void *task)
+{
+	int rc = 0;
+
+	spin_lock_bh(&conn->lock);
+	if (atomic_read(&conn->ib_conn->post_send_buf_count) ==
+	    ISER_QP_MAX_REQ_DTOS) {
+		iser_dbg("%ld can't xmit task %p, suspending tx\n",jiffies,task);
+		set_bit(SUSPEND_BIT, &conn->suspend_tx);
+		rc = -EAGAIN;
+	}
+	spin_unlock_bh(&conn->lock);
+	return rc;
+}
+
+
+/**
+ * iser_send_command - send command PDU
+ */
+int iser_send_command(struct iscsi_iser_conn     *p_iser_conn,
+		      struct iscsi_iser_cmd_task *p_ctask)
+{
+	struct iser_dto *p_send_dto = NULL;
+	unsigned long edtl;
+	int err = 0;
+	struct iser_data_buf data_buf;
+
+	struct iscsi_cmd *hdr =  p_ctask->hdr;
+	struct scsi_cmnd *sc  =  p_ctask->sc;
+
+	if (atomic_read(&p_iser_conn->ib_conn->state) != ISER_CONN_UP) {
+		iser_err("Failed to send, conn: 0x%p is not up\n", p_iser_conn->ib_conn);
+		return -EPERM;
+	}
+	if (iser_check_xmit(p_iser_conn, p_ctask))
+		return -EAGAIN;
+
+	edtl = ntohl(hdr->data_length);
+
+	/* build the tx desc regd header and add it to the tx desc dto */
+	p_ctask->desc.type = ISCSI_TX_SCSI_COMMAND;
+	p_send_dto = &p_ctask->desc.dto;
+	p_send_dto->p_task = p_ctask;
+	iser_create_send_desc(p_iser_conn, &p_ctask->desc);
+
+	if (sc->use_sg) { /* using a scatter list */
+		data_buf.p_buf = sc->request_buffer;
+		data_buf.size  = sc->use_sg;
+		data_buf.type  = ISER_BUF_TYPE_SCATTERLIST;
+	} else { /* using a single buffer */
+		data_buf.p_buf = sc->request_buffer;
+		data_buf.size  = sc->request_bufflen;
+		data_buf.type  = ISER_BUF_TYPE_SINGLE;
+	}
+
+	if (hdr->flags & ISCSI_FLAG_CMD_READ) {
+		err = iser_prepare_read_cmd(p_ctask, &data_buf, edtl);
+		if (err)
+			goto send_command_error;
+	}
+	if (hdr->flags & ISCSI_FLAG_CMD_WRITE) {
+		err = iser_prepare_write_cmd(p_ctask, &data_buf,
+					     p_ctask->imm_count,
+				             p_ctask->imm_count +
+					     p_ctask->unsol_count,
+					     edtl);
+		if (err)
+			goto send_command_error;
+	}
+
+	iser_reg_single(p_iser_conn->ib_conn->p_adaptor,
+			p_send_dto->regd[0], DMA_TO_DEVICE);
+
+	if (iser_post_receive_control(p_iser_conn) != 0) {
+		iser_err("post_recv failed!\n");
+		err = -ENOMEM;
+		goto send_command_error;
+	}
+
+	p_ctask->status = ISER_TASK_STATUS_STARTED;
+
+	err = iser_post_send(&p_ctask->desc);
+	if (!err)
+		return 0;
+
+send_command_error:
+	if (p_send_dto != NULL)
+		iser_dto_buffs_release(p_send_dto);
+	iser_err("conn %p failed err %d\n",p_iser_conn, err);
+	return err;
+}
+
+/**
+ * iser_send_data_out - send data out PDU
+ */
+int iser_send_data_out(struct iscsi_iser_conn     *p_iser_conn,
+		       struct iscsi_iser_cmd_task *p_ctask,
+		       struct iscsi_data *hdr)
+{
+	struct iser_desc *tx_desc = NULL;
+	struct iser_dto *p_send_dto = NULL;
+	unsigned long buf_offset;
+	unsigned long data_seg_len;
+	unsigned int itt;
+	int err = 0;
+
+	if (atomic_read(&p_iser_conn->ib_conn->state) != ISER_CONN_UP) {
+		iser_err("Failed to send, conn: 0x%p is not up\n", p_iser_conn->ib_conn);
+		return -EPERM;
+	}
+
+	if (iser_check_xmit(p_iser_conn, p_ctask))
+		return -EAGAIN;
+
+	itt = ntohl(hdr->itt);
+	data_seg_len = ntoh24(hdr->dlength);
+	buf_offset   = ntohl(hdr->offset);
+
+	iser_dbg("%s itt %d dseg_len %d offset %d\n",
+		 __func__,(int)itt,(int)data_seg_len,(int)buf_offset);
+
+	tx_desc = kmem_cache_alloc(ig.desc_cache, GFP_KERNEL | __GFP_NOFAIL);
+	if (tx_desc == NULL) {
+		iser_err("Failed to alloc desc for post dataout\n");
+		err = -ENOMEM;
+		goto send_data_out_error;
+	}
+
+	tx_desc->type = ISCSI_TX_DATAOUT;
+	memcpy(&tx_desc->iscsi_header, hdr, sizeof(struct iscsi_hdr));
+
+	/* build the tx desc regd header and add it to the tx desc dto */
+	p_send_dto = &tx_desc->dto;
+	p_send_dto->p_task = p_ctask;
+	iser_create_send_desc(p_iser_conn, tx_desc);
+
+	iser_reg_single(p_iser_conn->ib_conn->p_adaptor,
+			p_send_dto->regd[0], DMA_TO_DEVICE);
+
+	/* all data was registered for RDMA, we can use the lkey */
+	iser_dto_add_regd_buff(p_send_dto,
+			       &p_ctask->rdma_regd[ISER_DIR_OUT],
+			       USE_OFFSET(buf_offset),
+			       USE_SIZE(data_seg_len));
+
+	if (buf_offset + data_seg_len > p_ctask->data_len[ISER_DIR_OUT]) {
+		iser_err("Offset:%ld & DSL:%ld in Data-Out "
+			 "inconsistent with total len:%ld, itt:%d\n",
+			 buf_offset, data_seg_len,
+			 p_ctask->data_len[ISER_DIR_OUT], itt);
+		err = -EINVAL;
+		goto send_data_out_error;
+	}
+	iser_dbg("data-out itt: %d, offset: %ld, sz: %ld\n",
+		 itt, buf_offset, data_seg_len);
+
+
+	err = iser_post_send(tx_desc);
+	if (!err)
+		return 0;
+
+send_data_out_error:
+	if (p_send_dto != NULL)
+		iser_dto_buffs_release(p_send_dto);
+	if (tx_desc != NULL)
+		kmem_cache_free(ig.desc_cache, tx_desc);
+	iser_err("conn %p failed err %d\n",p_iser_conn, err);
+	return err;
+}
+
+int iser_send_control(struct iscsi_iser_conn *p_iser_conn,
+		      struct iscsi_iser_mgmt_task *p_mtask)
+{
+	struct iser_dto *p_send_dto = NULL;
+	unsigned int itt;
+	unsigned long data_seg_len;
+	int err = 0;
+	unsigned char opcode;
+	struct iser_regd_buf *p_regd_buf;
+	struct iser_adaptor *p_iser_adaptor;
+
+	if (atomic_read(&p_iser_conn->ib_conn->state) != ISER_CONN_UP) {
+		iser_err("Failed to send, conn: 0x%p is not up\n", p_iser_conn->ib_conn);
+		return -EPERM;
+	}
+
+	if (iser_check_xmit(p_iser_conn,p_mtask))
+		return -EAGAIN;
+
+	/* build the tx desc regd header and add it to the tx desc dto */
+	p_mtask->desc.type = ISCSI_TX_CONTROL;
+	p_send_dto = &p_mtask->desc.dto;
+	p_send_dto->p_task = NULL;
+	iser_create_send_desc(p_iser_conn, &p_mtask->desc);
+
+	p_iser_adaptor = p_iser_conn->ib_conn->p_adaptor;
+
+	iser_reg_single(p_iser_adaptor, p_send_dto->regd[0], DMA_TO_DEVICE);
+
+	itt = ntohl(p_mtask->hdr->itt);
+	opcode = p_mtask->hdr->opcode & ISCSI_OPCODE_MASK;
+	data_seg_len = ntoh24(p_mtask->hdr->dlength);
+
+	if (data_seg_len > 0) {
+		p_regd_buf = &p_mtask->desc.data_regd_buf;
+		memset(p_regd_buf, 0, sizeof(struct iser_regd_buf));
+		p_regd_buf->p_adaptor = p_iser_adaptor;
+		p_regd_buf->virt_addr = p_mtask->data;
+		p_regd_buf->data_size = p_mtask->data_count;
+		iser_reg_single(p_iser_adaptor, p_regd_buf,
+				DMA_TO_DEVICE);
+		iser_dto_add_regd_buff(p_send_dto, p_regd_buf,
+				       USE_NO_OFFSET,
+				       USE_SIZE(data_seg_len));
+	}
+
+	if (iser_post_receive_control(p_iser_conn) != 0) {
+		iser_err("post_rcv_buff failed!\n");
+		err = -ENOMEM;
+		goto send_control_error;
+	}
+
+	err = iser_post_send(&p_mtask->desc);
+	if (!err)
+		return 0;
+
+send_control_error:
+	if (p_send_dto != NULL)
+		iser_dto_buffs_release(p_send_dto);
+	iser_err("conn %p failed err %d\n",p_iser_conn, err);
+	return err;
+}
+
+/**
+ * iser_rcv_dto_completion - recv DTO completion
+ */
+void iser_rcv_completion(struct iser_desc *p_rx_desc,
+			 unsigned long dto_xfer_len)
+{
+	struct iscsi_iser_session *p_session;
+	struct iser_dto        *p_dto = &p_rx_desc->dto;
+	struct iscsi_iser_conn *p_iser_conn = p_dto->p_conn;
+	struct iscsi_iser_cmd_task *p_iser_task = NULL;
+	struct iscsi_hdr *p_hdr;
+	char   *rx_data = NULL;
+	int     rc, rx_data_size = 0;
+	unsigned int itt;
+	unsigned char opcode;
+
+	p_hdr = &p_rx_desc->iscsi_header;
+
+	iser_dbg("op 0x%x itt 0x%x\n", p_hdr->opcode,p_hdr->itt);
+
+	if (dto_xfer_len > ISER_TOTAL_HEADERS_LEN) { /* we have data */
+		rx_data_size = dto_xfer_len - ISER_TOTAL_HEADERS_LEN;
+		rx_data      = p_dto->regd[1]->virt_addr;
+		rx_data     += p_dto->offset[1];
+	}
+
+	opcode = p_hdr->opcode & ISCSI_OPCODE_MASK;
+
+	if (opcode == ISCSI_OP_SCSI_CMD_RSP) {
+		p_session = p_iser_conn->session;
+	        itt = p_hdr->itt;
+		if (!(itt < p_session->cmds_max))
+			iser_bug("itt can't be matched to task!!!"
+				 "conn %p opcode %d cmds_max %d itt %d\n",
+				 p_iser_conn,opcode,p_session->cmds_max,itt);
+		/* use the mapping given with the cmds array indexed by itt */
+		p_iser_task = (struct iscsi_iser_cmd_task *)p_session->cmds[itt];
+		iser_dbg("itt %d p_iser_task %p\n",itt,p_iser_task);
+		if (p_iser_task != NULL) {
+			if (p_iser_task->data_copy[ISER_DIR_IN].p_buf != NULL ||
+			    p_iser_task->data_copy[ISER_DIR_OUT].p_buf != NULL)
+				/* if we were reading, copy back to unaligned *
+				 * sglist, anyway dma_unmap and free the copy */
+				iser_finalize_rdma_unaligned_sg(p_iser_task);
+
+			p_iser_task->status = ISER_TASK_STATUS_COMPLETED;
+			iser_ctask_rdma_finalize(p_iser_task);
+		}
+	}
+
+	rc = iscsi_iser_hdr_recv(p_iser_conn, p_hdr, rx_data);
+	if (rc)
+		iscsi_iser_conn_failure(p_iser_conn, rc);
+
+	iser_dto_buffs_release(p_dto);
+	kfree(p_rx_desc->data);
+	kmem_cache_free(ig.desc_cache, p_rx_desc);
+
+	/* decrementing conn->post_recv_buf_count only --after-- freeing the   *
+	 * task eliminates the need to worry on tasks which are completed in   *
+	 * parallel to the execution of iser_conn_term. So the code that waits *
+         * for the posted rx bufs refcount to become zero handles everything   */
+	atomic_dec(&p_iser_conn->ib_conn->post_recv_buf_count);
+}
+
+void iser_snd_completion(struct iser_desc *p_tx_desc)
+{
+	struct iser_dto        *p_dto = &p_tx_desc->dto;
+	struct iscsi_iser_conn *p_iser_conn = p_dto->p_conn;
+
+	iser_dbg("Initiator, Data sent p_dto=0x%p\n", p_dto);
+
+	iser_dto_buffs_release(p_dto);
+
+	if (p_tx_desc->type == ISCSI_TX_DATAOUT)
+		kmem_cache_free(ig.desc_cache, p_tx_desc);
+
+	atomic_dec(&p_iser_conn->ib_conn->post_send_buf_count);
+
+	spin_lock(&p_iser_conn->lock);
+	if (p_iser_conn->suspend_tx) {
+		iser_dbg("%ld resuming tx\n",jiffies);
+		clear_bit(SUSPEND_BIT, &p_iser_conn->suspend_tx);
+		schedule_work(&p_iser_conn->xmitwork);
+	}
+	spin_unlock(&p_iser_conn->lock);
+}
+
+void iser_ctask_rdma_init(struct iscsi_iser_cmd_task *p_iser_task)
+
+{
+	p_iser_task->status = ISER_TASK_STATUS_INIT;
+
+	p_iser_task->dir[ISER_DIR_IN] = 0;
+	p_iser_task->dir[ISER_DIR_OUT] = 0;
+
+	p_iser_task->data_len[ISER_DIR_IN] = 0;
+	p_iser_task->data_len[ISER_DIR_OUT] = 0;
+
+	memset(&p_iser_task->rdma_regd[ISER_DIR_IN], 0,
+	       sizeof(struct iser_regd_buf));
+	memset(&p_iser_task->rdma_regd[ISER_DIR_OUT], 0,
+	       sizeof(struct iser_regd_buf));
+}
+
+void iser_ctask_rdma_finalize(struct iscsi_iser_cmd_task *p_iser_task)
+{
+	int deferred;
+
+	if (p_iser_task->dir[ISER_DIR_IN]) {
+		deferred = iser_regd_buff_release
+			(&p_iser_task->rdma_regd[ISER_DIR_IN]);
+		if (deferred)
+			iser_bug("References remain for BUF-IN rdma reg\n");
+	}
+
+	if (p_iser_task->dir[ISER_DIR_OUT]) {
+		deferred = iser_regd_buff_release
+			(&p_iser_task->rdma_regd[ISER_DIR_OUT]);
+		if (deferred)
+			iser_bug("References remain for BUF-OUT rdma reg\n");
+	}
+
+	iser_dma_unmap_task_data(p_iser_task);
+}
+
+void iser_dto_buffs_release(struct iser_dto *p_dto)
+{
+	int i;
+
+	for (i = 0; i < p_dto->regd_vector_len; i++)
+		iser_regd_buff_release(p_dto->regd[i]);
+}
 




More information about the general mailing list