Re: [PATCH v2 1/2] tcm: Add support for BIDI-COMMANDS and XDWRITE_READ_10 emulation

[Boaz Harrosh <bharrosh@panasas.com>]

On 09/16/2010 12:41 AM, Nicholas A. Bellinger wrote:
> From: Nicholas Bellinger <nab@linux-iscsi.org>
> 

Hi dear Nicholas

I still have a few reservations regarding the use of the:
+	int			t_tasks_bidi:1;

at struct se_transport_task at minimum I'd use the t_tasks_se_bidi_num
as a look ahead. But I hate that as well. I suspect none of this is
needed. But

At http://git.kernel.org/?p=linux/kernel/git/nab/lio-4.0.git I still get
an old head without these or the CDB32 stuff.

Where can I find a git web. With latest bits? I'd like to have a
closer look.

(You know that I have a vested interest in all this I need it to be solid)

> This v2 patch series adds BIDI-COMMANDS support to TCM Core.  This
> includes the following for struct se_cmd to handle BIDI READ payloads using a new
> struct se_transport_task->t_mem_bidi_list and struct se_transport_task->t_tasks_se_bidi_num.
> The model employed is to keep the WRITE payload at struct se_transport_task->t_mem_list,
> and add a new BIDI READ payload memory list at struct se_transport_task->t_mem_bidi_list.
> 
> *) descriptor setup:
> 
> This patch adds support for XDWRITEREAD_10 within transport_generic_cmd_sequencer(),
> and sets up the new struct se_cmd->transport_xor_callback() completion function
> at transport_xor_callback().  It then updates transport_new_cmd_obj() to handle the
> BIDI READ case.
> 
> It also adds support for checking BIDI READ data poaylods in transport_generic_map_mem_to_cmd()
> to signal the allocation of struct se_transport_task->t_mem_bidi_list in the same function.
> 
> *) memory mapping:
> 
> This patch updates transport_generic_get_cdb_count() to accept a enum dma_data_direction
> parameter to handle the BIDI and the existing WRITE/READ cases for SCF_SCSI_DATA_SG_IO_CDB.
> This patch also updates transport_generic_map_mem_to_cmd() to accept a 'void *mem_bidi_in'
> and 'u32 se_mem_bidi_num' from BIDI capable TCM fabric modules.
> 
> It then updates transport_generic_get_task() and struct se_cmd->transport_get_task()
> to accept a enum dma_data_direction function parameter.
> 
> *) descriptor callback:
> 
> For the struct se_cmd callback, it updates transport_generic_complete_ok() to support
> BIDI-COMMANDS and looks for the new generic struct se_cmd->transport_complete_callback() in
> transport_generic_cmd_sequencer() to perform the post READ/WRITE XOR emulation.
> This also includes the additon of transport_memcpy_se_mem_read_contig() used to copy the
> WRITE scatterlists into a local contigious buffer for the XOR instructions within
> transport_xor_callback();
> 
> *) descriptor release:
> 
> Update transport_free_pages() to walk the new T_TASK(cmd)->t_mem_bidi_list (when available)
> and release struct se_mem and pages.
> 
> So far this has been tested with TCM_Loop using BSG w/ userspace code generating
> BIDI XDWRITE_READ_10 CDBs.
> 
> Signed-off-by: Nicholas A. Bellinger <nab@linux-iscsi.org>
> ---
>  drivers/target/target_core_pr.c        |    3 +-
>  drivers/target/target_core_transport.c |  278 +++++++++++++++++++++++++++-----
>  include/target/target_core_base.h      |   10 +-
>  include/target/target_core_transport.h |   10 +-
>  4 files changed, 254 insertions(+), 47 deletions(-)
> 
> diff --git a/drivers/target/target_core_pr.c b/drivers/target/target_core_pr.c
> index 4c408cb..141ce48 100644
> --- a/drivers/target/target_core_pr.c
> +++ b/drivers/target/target_core_pr.c
> @@ -497,8 +497,7 @@ static int core_scsi3_pr_seq_non_holder(
>  	 * WRITE_EXCLUSIVE_* reservation.
>  	 */
>  	if ((we) && !(registered_nexus)) {
> -		if ((cmd->data_direction == DMA_TO_DEVICE) ||
> -		    (cmd->data_direction == DMA_BIDIRECTIONAL)) {
> +		if (cmd->data_direction == DMA_TO_DEVICE) {
>  			/*
>  			 * Conflict for write exclusive
>  			 */
> diff --git a/drivers/target/target_core_transport.c b/drivers/target/target_core_transport.c
> index 517a59c..e709771 100644
> --- a/drivers/target/target_core_transport.c
> +++ b/drivers/target/target_core_transport.c
> @@ -2529,7 +2529,8 @@ static inline int transport_check_device_cdb_sector_count(
>  static struct se_task *transport_generic_get_task(
>  	struct se_transform_info *ti,
>  	struct se_cmd *cmd,
> -	void *se_obj_ptr)
> +	void *se_obj_ptr,
> +	enum dma_data_direction data_direction)
>  {
>  	struct se_task *task;
>  	struct se_device *dev = SE_DEV(cmd);
> @@ -2625,7 +2626,8 @@ static int transport_process_control_sg_transform(
>  		return -1;
>  	}
>  
> -	task = cmd->transport_get_task(ti, cmd, ti->se_obj_ptr);
> +	task = cmd->transport_get_task(ti, cmd, ti->se_obj_ptr,
> +				cmd->data_direction);
>  	if (!(task))
>  		return -1;
>  
> @@ -2665,7 +2667,8 @@ static int transport_process_control_nonsg_transform(
>  	unsigned char *cdb;
>  	struct se_task *task;
>  
> -	task = cmd->transport_get_task(ti, cmd, ti->se_obj_ptr);
> +	task = cmd->transport_get_task(ti, cmd, ti->se_obj_ptr,
> +				cmd->data_direction);
>  	if (!(task))
>  		return -1;
>  
> @@ -2699,7 +2702,8 @@ static int transport_process_non_data_transform(
>  	unsigned char *cdb;
>  	struct se_task *task;
>  
> -	task = cmd->transport_get_task(ti, cmd, ti->se_obj_ptr);
> +	task = cmd->transport_get_task(ti, cmd, ti->se_obj_ptr,
> +				cmd->data_direction);
>  	if (!(task))
>  		return -1;
>  
> @@ -2741,11 +2745,6 @@ struct se_cmd *__transport_alloc_se_cmd(
>  	unsigned char *sense_buffer;
>  	int gfp_type = (in_interrupt()) ? GFP_ATOMIC : GFP_KERNEL;
>  
> -	if (data_direction == DMA_BIDIRECTIONAL) {
> -		printk(KERN_ERR "SCSI BiDirectional mode not supported yet\n");
> -		return ERR_PTR(-ENOSYS);
> -	}
> -
>  	cmd = kmem_cache_zalloc(se_cmd_cache, gfp_type);
>  	if (!(cmd)) {
>  		printk(KERN_ERR "kmem_cache_alloc() failed for se_cmd_cache\n");
> @@ -5183,6 +5182,54 @@ int transport_generic_emulate_request_sense(
>  }
>  EXPORT_SYMBOL(transport_generic_emulate_request_sense);
>  
> +static void transport_xor_callback(struct se_cmd *cmd)
> +{
> +	unsigned char *buf, *addr;
> +	struct se_mem *se_mem;
> +	unsigned int offset;
> +	int i;
> +	/*
> +	 * From sbc3r22.pdf section 5.48 XDWRITEREAD (10) command
> +	 *
> +	 * 1) read the specified logical block(s);
> +	 * 2) transfer logical blocks from the data-out buffer;
> +	 * 3) XOR the logical blocks transferred from the data-out buffer with
> +	 *    the logical blocks read, storing the resulting XOR data in a buffer;
> +	 * 4) if the DISABLE WRITE bit is set to zero, then write the logical
> +	 *    blocks transferred from the data-out buffer; and
> +	 * 5) transfer the resulting XOR data to the data-in buffer.
> +	 */
> +	buf = kmalloc(cmd->data_length, GFP_KERNEL);
> +	if (!(buf)) {
> +		printk(KERN_ERR "Unable to allocate xor_callback buf\n");
> +		return;
> +	}
> +	/*
> +	 * Copy the scatterlist WRITE buffer located at T_TASK(cmd)->t_mem_list
> +	 * into the locally allocated *buf
> +	 */
> +	transport_memcpy_se_mem_read_contig(cmd, buf, T_TASK(cmd)->t_mem_list);
> +	/*
> +	 * Now perform the XOR against the BIDI read memory located at
> +	 * T_TASK(cmd)->t_mem_bidi_list
> +	 */
> +
> +	offset = 0;
> +	list_for_each_entry(se_mem, T_TASK(cmd)->t_mem_bidi_list, se_list) {
> +		addr = (unsigned char *)kmap_atomic(se_mem->se_page, KM_USER0);
> +		if (!(addr))
> +			goto out;
> +
> +		for (i = 0; i < se_mem->se_len; i++)
> +			*(addr + se_mem->se_off + i) ^= *(buf + offset + i);
> +
> +		offset += se_mem->se_len;
> +		kunmap_atomic(addr, KM_USER0);
> +	}
> +out:
> +	kfree(buf);
> +}
> +
>  /*
>   * Used to obtain Sense Data from underlying Linux/SCSI struct scsi_cmnd
>   */
> @@ -5472,6 +5519,26 @@ static int transport_generic_cmd_sequencer(
>  		T_TASK(cmd)->t_tasks_fua = (cdb[1] & 0x8);
>  		ret = TGCS_DATA_SG_IO_CDB;
>  		break;
> +	case XDWRITEREAD_10:
> +		SET_GENERIC_TRANSPORT_FUNCTIONS(cmd);
> +		if ((cmd->data_direction != DMA_TO_DEVICE) ||
> +		    !(T_TASK(cmd)->t_tasks_bidi))
> +			return TGCS_INVALID_CDB_FIELD;
> +		sectors = transport_get_sectors_10(cdb, cmd, &sector_ret);
> +		if (sector_ret)
> +			return TGCS_UNSUPPORTED_CDB;
> +		size = transport_get_size(sectors, cdb, cmd);
> +		transport_dev_get_mem_SG(cmd->se_orig_obj_ptr, cmd);
> +		transport_get_maps(cmd);
> +		cmd->transport_split_cdb = &split_cdb_XX_10;
> +		cmd->transport_get_lba = &transport_lba_32;
> +		/*
> +		 * Setup BIDI XOR callback to be run during transport_generic_complete_ok()
> +		 */
> +		cmd->transport_complete_callback = &transport_xor_callback;
> +		T_TASK(cmd)->t_tasks_fua = (cdb[1] & 0x8);
> +		ret = TGCS_DATA_SG_IO_CDB;
> +		break;
>  	case 0xa3:
>  		SET_GENERIC_TRANSPORT_FUNCTIONS(cmd);
>  		if (TRANSPORT(dev)->get_device_type(dev) != TYPE_ROM) {
> @@ -6096,6 +6163,33 @@ void transport_memcpy_read_contig(
>  }
>  EXPORT_SYMBOL(transport_memcpy_read_contig);
>  
> +void transport_memcpy_se_mem_read_contig(
> +	struct se_cmd *cmd,
> +	unsigned char *dst,
> +	struct list_head *se_mem_list)
> +{
> +	struct se_mem *se_mem;
> +	void *src;
> +	u32 length = 0, total_length = cmd->data_length;
> +
> +	list_for_each_entry(se_mem, se_mem_list, se_list) {
> +		length = se_mem->se_len;
> +
> +		if (length > total_length)
> +			length = total_length;
> +
> +		src = page_address(se_mem->se_page) + se_mem->se_off;
> +
> +		memcpy(dst, src, length);
> +
> +		if (!(total_length -= length))
> +			return;
> +
> +		dst += length;
> +	}
> +}
> +
> +
>  /*     transport_generic_passthrough():
>   *
>   *
> @@ -6249,6 +6343,12 @@ void transport_generic_complete_ok(struct se_cmd *cmd)
>  			return;
>  		}
>  	}
> +	/*
> +	 * Check for a callback, used by amoungst other things
> +	 * XDWRITE_READ_10 emulation.
> +	 */
> +	if (cmd->transport_complete_callback)
> +		cmd->transport_complete_callback(cmd);
>  
>  	switch (cmd->data_direction) {
>  	case DMA_FROM_DEVICE:
> @@ -6267,6 +6367,19 @@ void transport_generic_complete_ok(struct se_cmd *cmd)
>  				cmd->data_length;
>  		}
>  		spin_unlock(&cmd->se_lun->lun_sep_lock);
> +		/*
> +		 * Check if we need to send READ payload for BIDI-COMMAND
> +		 */
> +		if (T_TASK(cmd)->t_mem_bidi_list != NULL) {
> +			spin_lock(&cmd->se_lun->lun_sep_lock);
> +			if (SE_LUN(cmd)->lun_sep) {
> +				SE_LUN(cmd)->lun_sep->sep_stats.tx_data_octets +=
> +					cmd->data_length;
> +			}
> +			spin_unlock(&cmd->se_lun->lun_sep_lock);
> +			CMD_TFO(cmd)->queue_data_in(cmd);
> +			break;

Again this looks fishy to me. How do you know that this command needs a queue_data_in()
now? Are you not assuming XOR operations? can you not move this to transport_xor_callback
above?

> +		}
>  		/* Fall through for DMA_TO_DEVICE */
>  	case DMA_NONE:
>  		CMD_TFO(cmd)->queue_status(cmd);
> @@ -6347,6 +6460,23 @@ static inline void transport_free_pages(struct se_cmd *cmd)
>  		kmem_cache_free(se_mem_cache, se_mem);
>  	}
>  

Again the request for an update tree. (Might be missing some code here)

The above does a break; in bidi case, but are you freeing this here regardless?

> +	if (T_TASK(cmd)->t_mem_bidi_list && T_TASK(cmd)->t_tasks_se_bidi_num) {
> +		list_for_each_entry_safe(se_mem, se_mem_tmp,
> +				T_TASK(cmd)->t_mem_bidi_list, se_list) {
> +			/*
> +			 * We only release call __free_page(struct se_mem->se_page) when
> +			 * SCF_PASSTHROUGH_SG_TO_MEM_NOALLOC is NOT in use,
> +			 */
> +			if (free_page)
> +				__free_page(se_mem->se_page);
> +
> +			list_del(&se_mem->se_list);
> +			kmem_cache_free(se_mem_cache, se_mem);
> +		}
> +	}
> +
> +	kfree(T_TASK(cmd)->t_mem_bidi_list);
> +	T_TASK(cmd)->t_mem_bidi_list = NULL;
>  	kfree(T_TASK(cmd)->t_mem_list);
>  	T_TASK(cmd)->t_mem_list = NULL;
>  	T_TASK(cmd)->t_tasks_se_num = 0;
> @@ -6477,7 +6607,9 @@ release_cmd:
>  int transport_generic_map_mem_to_cmd(
>  	struct se_cmd *cmd,
>  	void *mem,
> -	u32 se_mem_num)
> +	u32 se_mem_num,
> +	void *mem_bidi_in,
> +	u32 se_mem_bidi_num)
>  {
>  	u32 se_mem_cnt_out = 0;
>  	int ret;
> @@ -6489,6 +6621,12 @@ int transport_generic_map_mem_to_cmd(
>  	 * struct se_mem elements...
>  	 */
>  	if (!(cmd->se_cmd_flags & SCF_PASSTHROUGH_SG_TO_MEM)) {
> +		if ((mem_bidi_in) || (se_mem_bidi_num)) {
> +			printk(KERN_ERR "SCF_CMD_PASSTHROUGH_NOALLOC not supported"
> +				" with BIDI-COMMAND\n");
> +			return -ENOSYS;
> +		}
> +
>  		T_TASK(cmd)->t_mem_list = (struct list_head *)mem;
>  		T_TASK(cmd)->t_tasks_se_num = se_mem_num;
>  		cmd->se_cmd_flags |= SCF_CMD_PASSTHROUGH_NOALLOC;
> @@ -6507,14 +6645,35 @@ int transport_generic_map_mem_to_cmd(
>  		 */ 
>  		T_TASK(cmd)->t_mem_list = transport_init_se_mem_list();
>  		if (!(T_TASK(cmd)->t_mem_list))
> -			return -1;
> +			return -ENOMEM;
>  
>  		ret = transport_map_sg_to_mem(cmd,
>  			T_TASK(cmd)->t_mem_list, mem, &se_mem_cnt_out);
>  		if (ret < 0)
> -			return -1;
> +			return -ENOMEM;
>  
>  		T_TASK(cmd)->t_tasks_se_num = se_mem_cnt_out;
> +		/*
> +		 * Setup BIDI READ list of struct se_mem elements
> +		 */
> +		if ((mem_bidi_in) && (se_mem_bidi_num)) {
> +			T_TASK(cmd)->t_mem_bidi_list = transport_init_se_mem_list();
> +			if (!(T_TASK(cmd)->t_mem_bidi_list)) {
> +				kfree(T_TASK(cmd)->t_mem_list);
> +				return -ENOMEM;
> +			}
> +			se_mem_cnt_out = 0;
> +
> +			ret = transport_map_sg_to_mem(cmd,
> +				T_TASK(cmd)->t_mem_bidi_list, mem_bidi_in,
> +				&se_mem_cnt_out);
> +			if (ret < 0) {
> +				kfree(T_TASK(cmd)->t_mem_list);
> +				return -ENOMEM;
> +			}
> +
> +			T_TASK(cmd)->t_tasks_se_bidi_num = se_mem_cnt_out;
> +		}
>  		cmd->se_cmd_flags |= SCF_PASSTHROUGH_SG_TO_MEM_NOALLOC;
>  
>  	} else if (cmd->se_cmd_flags & SCF_SCSI_CONTROL_NONSG_IO_CDB) {
> @@ -6610,6 +6769,11 @@ non_scsi_data:
>   */
>  int transport_generic_do_transform(struct se_cmd *cmd, struct se_transform_info *ti)
>  {
> +	if (!(cmd->transport_cdb_transform)) {
> +		dump_stack();
> +		return -1;
> +	}
> +
>  	if (cmd->transport_cdb_transform(cmd, ti) < 0)
>  		return -1;
>  
> @@ -6656,9 +6820,8 @@ int transport_new_cmd_obj(
>  	struct se_transform_info *ti,
>  	int post_execute)
>  {
> -	u32 task_cdbs = 0;
> -	struct se_mem *se_mem_out = NULL;
>  	struct se_device *dev = SE_DEV(cmd);
> +	u32 task_cdbs = 0, rc;
>  
>  	if (!(cmd->se_cmd_flags & SCF_SCSI_DATA_SG_IO_CDB)) {
>  		task_cdbs++;
> @@ -6666,11 +6829,32 @@ int transport_new_cmd_obj(
>  	} else {
>  		ti->ti_set_counts = 1;
>  		ti->ti_dev = dev;
> -
> +		/*
> +		 * Setup any BIDI READ tasks and memory from
> +		 * T_TASK(cmd)->t_mem_bidi_list so the READ struct se_tasks
> +		 * are queued first..
> +		 */
> +		if (T_TASK(cmd)->t_mem_bidi_list != NULL) {
> +			rc = transport_generic_get_cdb_count(cmd, ti,
> +				T_TASK(cmd)->t_task_lba,
> +				T_TASK(cmd)->t_tasks_sectors,
> +				DMA_FROM_DEVICE, T_TASK(cmd)->t_mem_bidi_list);
> +			if (!(rc)) {
> +				cmd->se_cmd_flags |= SCF_SCSI_CDB_EXCEPTION;
> +				cmd->scsi_sense_reason =
> +					TCM_LOGICAL_UNIT_COMMUNICATION_FAILURE;
> +				return PYX_TRANSPORT_LU_COMM_FAILURE;
> +			}
> +			ti->ti_set_counts = 0;
> +		}
> +		/*
> +		 * Setup the tasks and memory from T_TASK(cmd)->t_mem_list
> +		 * Note for BIDI transfers this will contain the WRITE payload
> +		 */
>  		task_cdbs = transport_generic_get_cdb_count(cmd, ti,
>  				T_TASK(cmd)->t_task_lba,
>  				T_TASK(cmd)->t_tasks_sectors,
> -				NULL, &se_mem_out);
> +				cmd->data_direction, T_TASK(cmd)->t_mem_list);
>  		if (!(task_cdbs)) {
>  			cmd->se_cmd_flags |= SCF_SCSI_CDB_EXCEPTION;
>  			cmd->scsi_sense_reason =
> @@ -6743,7 +6927,17 @@ int transport_generic_get_mem(struct se_cmd *cmd, u32 length, u32 dma_size)
>  
>  	T_TASK(cmd)->t_mem_list = transport_init_se_mem_list();
>  	if (!(T_TASK(cmd)->t_mem_list))
> -		return -1;
> +		return -ENOMEM;
> +	/*
> +	 * Setup BIDI-COMMAND READ list of struct se_mem elements
> +	 */
> +	if (T_TASK(cmd)->t_tasks_bidi) {
> +		T_TASK(cmd)->t_mem_bidi_list = transport_init_se_mem_list();
> +		if (!(T_TASK(cmd)->t_mem_bidi_list)) {
> +			kfree(T_TASK(cmd)->t_mem_list);
> +			return -ENOMEM;
> +		}
> +	}
>  
>  	while (length) {
>  		se_mem = kmem_cache_zalloc(se_mem_cache, GFP_KERNEL);
> @@ -7240,28 +7434,28 @@ u32 transport_generic_get_cdb_count(
>  	struct se_transform_info *ti,
>  	unsigned long long starting_lba,
>  	u32 sectors,
> -	struct se_mem *se_mem_in,
> -	struct se_mem **se_mem_out)
> +	enum dma_data_direction data_direction,
> +	struct list_head *mem_list)
>  {
>  	unsigned char *cdb = NULL;
>  	struct se_task *task;
> -	struct se_mem *se_mem, *se_mem_lout = NULL;
> +	struct se_mem *se_mem = NULL, *se_mem_lout = NULL;
>  	struct se_device *dev = SE_DEV(cmd);
>  	int max_sectors_set = 0, ret;
>  	u32 task_offset_in = 0, se_mem_cnt = 0, task_cdbs = 0;
>  	unsigned long long lba;
>  
> -	if (!se_mem_in) {
> -		list_for_each_entry(se_mem_in, T_TASK(cmd)->t_mem_list, se_list)
> -			break;
> -
> -		if (!se_mem_in) {
> -			printk(KERN_ERR "se_mem_in is NULL\n");
> -			return 0;
> -		}
> +	if (!mem_list) {
> +		printk(KERN_ERR "mem_list is NULL in transport_generic_get"
> +				"_cdb_count()\n");
> +		return 0;
>  	}
> -	se_mem = se_mem_in;
> -
> +	/*
> +	 * While using RAMDISK_DR backstores is the only case where
> +	 * mem_list will ever be empty at this point.
> +	 */
> +	if (!(list_empty(mem_list)))
> +		se_mem = list_entry(mem_list->next, struct se_mem, se_list);
>  	/*
>  	 * Locate the start volume segment in which the received LBA will be
>  	 * executed upon.
> @@ -7280,7 +7474,12 @@ u32 transport_generic_get_cdb_count(
>  			CMD_TFO(cmd)->get_task_tag(cmd), lba, sectors,
>  			transport_dev_end_lba(dev));
>  
> -		task = cmd->transport_get_task(ti, cmd, dev);
> +		if (!(cmd->transport_get_task)) {
> +			dump_stack();
> +			goto out;
> +		}
> +
> +		task = cmd->transport_get_task(ti, cmd, dev, data_direction);
>  		if (!(task))
>  			goto out;
>  
> @@ -7293,7 +7492,7 @@ u32 transport_generic_get_cdb_count(
>  		task->task_size = (task->task_sectors *
>  				   DEV_ATTRIB(dev)->block_size);
>  		task->transport_map_task = transport_dev_get_map_SG(dev,
> -					cmd->data_direction);
> +					data_direction);
>  
>  		cdb = TRANSPORT(dev)->get_cdb(task);
>  		if ((cdb)) {
> @@ -7306,14 +7505,13 @@ u32 transport_generic_get_cdb_count(
>  		 * Perform the SE OBJ plugin and/or Transport plugin specific
>  		 * mapping for T_TASK(cmd)->t_mem_list.
>  		 */
> -		ret = transport_do_se_mem_map(dev, task,
> -				T_TASK(cmd)->t_mem_list, NULL, se_mem,
> -				&se_mem_lout, &se_mem_cnt, &task_offset_in);
> +		ret = transport_do_se_mem_map(dev, task, mem_list,
> +				NULL, se_mem, &se_mem_lout, &se_mem_cnt,
> +				&task_offset_in);
>  		if (ret < 0)
>  			goto out;
>  
>  		se_mem = se_mem_lout;
> -		*se_mem_out = se_mem_lout;
>  		task_cdbs++;
>  
>  		DEBUG_VOL("Incremented task_cdbs(%u) task->task_sg_num(%u)\n",
> @@ -7333,8 +7531,9 @@ u32 transport_generic_get_cdb_count(
>  		atomic_inc(&T_TASK(cmd)->t_se_count);
>  	}
>  
> -	DEBUG_VOL("ITT[0x%08x] total cdbs(%u)\n",
> -		CMD_TFO(cmd)->get_task_tag(cmd), task_cdbs);
> +	DEBUG_VOL("ITT[0x%08x] total %s cdbs(%u)\n",
> +		CMD_TFO(cmd)->get_task_tag(cmd), (data_direction == DMA_TO_DEVICE)
> +		? "DMA_TO_DEVICE" : "DMA_FROM_DEVICE", task_cdbs);
>  
>  	return task_cdbs;
>  out:
> @@ -8129,8 +8328,7 @@ void transport_send_task_abort(struct se_cmd *cmd)
>  	 * response.  This response with TASK_ABORTED status will be
>  	 * queued back to fabric module by transport_check_aborted_status().
>  	 */
> -	if ((cmd->data_direction == DMA_TO_DEVICE) ||
> -	    (cmd->data_direction == DMA_BIDIRECTIONAL)) {
> +	if (cmd->data_direction == DMA_TO_DEVICE) {
>  		if (CMD_TFO(cmd)->write_pending_status(cmd) != 0) {
>  			atomic_inc(&T_TASK(cmd)->t_transport_aborted);
>  			smp_mb__after_atomic_inc();
> diff --git a/include/target/target_core_base.h b/include/target/target_core_base.h
> index e9b98e9..97e9715 100644
> --- a/include/target/target_core_base.h
> +++ b/include/target/target_core_base.h
> @@ -414,11 +414,13 @@ struct se_transport_task {
>  	unsigned long long	t_task_lba;
>  	int			t_tasks_failed;
>  	int			t_tasks_fua;
> +	int			t_tasks_bidi:1;
>  	u32			t_task_cdbs;
>  	u32			t_tasks_check;
>  	u32			t_tasks_no;
>  	u32			t_tasks_sectors;
>  	u32			t_tasks_se_num;
> +	u32			t_tasks_se_bidi_num;
>  	u32			t_tasks_sg_chained_no;
>  	atomic_t		t_fe_count;
>  	atomic_t		t_se_count;
> @@ -447,8 +449,10 @@ struct se_transport_task {
>  	struct scatterlist	t_tasks_sg_bounce;
>  	void			*t_task_buf;
>  	void			*t_task_pt_buf;
> -	struct list_head	t_task_list;
>  	struct list_head	*t_mem_list;
> +	/* Used for BIDI READ */
> +	struct list_head	*t_mem_bidi_list;
> +	struct list_head	t_task_list;
>  } ____cacheline_aligned;
>  
>  struct se_task {
> @@ -598,7 +602,8 @@ struct se_cmd {
>  	u32 (*transport_get_lba)(unsigned char *);
>  	unsigned long long (*transport_get_long_lba)(unsigned char *);
>  	struct se_task *(*transport_get_task)(struct se_transform_info *,
> -					struct se_cmd *, void *);
> +					struct se_cmd *, void *,
> +					enum dma_data_direction);
>  	int (*transport_map_buffers_to_tasks)(struct se_cmd *);
>  	void (*transport_map_SG_segments)(struct se_unmap_sg *);
>  	void (*transport_passthrough_done)(struct se_cmd *);
> @@ -607,6 +612,7 @@ struct se_cmd {
>  					struct se_unmap_sg *);
>  	void (*transport_split_cdb)(unsigned long long, u32 *, unsigned char *);
>  	void (*transport_wait_for_tasks)(struct se_cmd *, int, int);
> +	void (*transport_complete_callback)(struct se_cmd *);
>  	void (*callback)(struct se_cmd *cmd, void *callback_arg,
>  			int complete_status);
>  	void *callback_arg;
> diff --git a/include/target/target_core_transport.h b/include/target/target_core_transport.h
> index ef4c084..20702d7 100644
> --- a/include/target/target_core_transport.h
> +++ b/include/target/target_core_transport.h
> @@ -232,6 +232,8 @@ extern void transport_memcpy_write_contig(struct se_cmd *, struct scatterlist *,
>  				unsigned char *);
>  extern void transport_memcpy_read_contig(struct se_cmd *, unsigned char *,
>  				struct scatterlist *);
> +extern void transport_memcpy_se_mem_read_contig(struct se_cmd *,
> +				unsigned char *, struct list_head *);
>  extern int transport_generic_passthrough_async(struct se_cmd *cmd,
>  				void(*callback)(struct se_cmd *cmd,
>  				void *callback_arg, int complete_status),
> @@ -242,7 +244,8 @@ extern void transport_generic_complete_ok(struct se_cmd *);
>  extern void transport_free_dev_tasks(struct se_cmd *);
>  extern void transport_release_fe_cmd(struct se_cmd *);
>  extern int transport_generic_remove(struct se_cmd *, int, int);
> -extern int transport_generic_map_mem_to_cmd(struct se_cmd *cmd, void *, u32);
> +extern int transport_generic_map_mem_to_cmd(struct se_cmd *cmd, void *, u32,
> +				void *, u32);
>  extern int transport_lun_wait_for_tasks(struct se_cmd *, struct se_lun *);
>  extern int transport_clear_lun_from_sessions(struct se_lun *);
>  extern int transport_check_aborted_status(struct se_cmd *, int);
> @@ -271,8 +274,9 @@ extern int transport_map_mem_to_sg(struct se_task *, struct list_head *,
>  extern void transport_do_task_sg_chain(struct se_cmd *);
>  extern u32 transport_generic_get_cdb_count(struct se_cmd *,
>  					struct se_transform_info *,
> -					unsigned long long, u32, struct se_mem *,
> -					struct se_mem **);
> +					unsigned long long, u32,
> +					enum dma_data_direction,
> +					struct list_head *);
>  extern int transport_generic_new_cmd(struct se_cmd *);
>  extern void transport_generic_process_write(struct se_cmd *);
>  extern int transport_generic_do_tmr(struct se_cmd *);

Thanks in advance
Boaz