diff --git a/drivers/target/target_core_file.c b/drivers/target/target_core_file.c
index a984964ab70..bbcedcfc068 100644
--- a/drivers/target/target_core_file.c
+++ b/drivers/target/target_core_file.c
@@ -244,53 +244,33 @@ static void fd_free_device(void *p)
 	kfree(fd_dev);
 }
 
-static inline struct fd_request *FILE_REQ(struct se_task *task)
+static int fd_do_readv(struct se_cmd *cmd, struct scatterlist *sgl,
+		u32 sgl_nents)
 {
-	return container_of(task, struct fd_request, fd_task);
-}
-
-
-static struct se_task *
-fd_alloc_task(unsigned char *cdb)
-{
-	struct fd_request *fd_req;
-
-	fd_req = kzalloc(sizeof(struct fd_request), GFP_KERNEL);
-	if (!fd_req) {
-		pr_err("Unable to allocate struct fd_request\n");
-		return NULL;
-	}
-
-	return &fd_req->fd_task;
-}
-
-static int fd_do_readv(struct se_task *task)
-{
-	struct fd_request *req = FILE_REQ(task);
-	struct se_device *se_dev = req->fd_task.task_se_cmd->se_dev;
+	struct se_device *se_dev = cmd->se_dev;
 	struct fd_dev *dev = se_dev->dev_ptr;
 	struct file *fd = dev->fd_file;
-	struct scatterlist *sg = task->task_sg;
+	struct scatterlist *sg;
 	struct iovec *iov;
 	mm_segment_t old_fs;
-	loff_t pos = (task->task_se_cmd->t_task_lba *
+	loff_t pos = (cmd->t_task_lba *
 		      se_dev->se_sub_dev->se_dev_attrib.block_size);
 	int ret = 0, i;
 
-	iov = kzalloc(sizeof(struct iovec) * task->task_sg_nents, GFP_KERNEL);
+	iov = kzalloc(sizeof(struct iovec) * sgl_nents, GFP_KERNEL);
 	if (!iov) {
 		pr_err("Unable to allocate fd_do_readv iov[]\n");
 		return -ENOMEM;
 	}
 
-	for_each_sg(task->task_sg, sg, task->task_sg_nents, i) {
+	for_each_sg(sgl, sg, sgl_nents, i) {
 		iov[i].iov_len = sg->length;
 		iov[i].iov_base = sg_virt(sg);
 	}
 
 	old_fs = get_fs();
 	set_fs(get_ds());
-	ret = vfs_readv(fd, &iov[0], task->task_sg_nents, &pos);
+	ret = vfs_readv(fd, &iov[0], sgl_nents, &pos);
 	set_fs(old_fs);
 
 	kfree(iov);
@@ -300,10 +280,10 @@ static int fd_do_readv(struct se_task *task)
 	 * block_device.
 	 */
 	if (S_ISBLK(fd->f_dentry->d_inode->i_mode)) {
-		if (ret < 0 || ret != task->task_se_cmd->data_length) {
+		if (ret < 0 || ret != cmd->data_length) {
 			pr_err("vfs_readv() returned %d,"
 				" expecting %d for S_ISBLK\n", ret,
-				(int)task->task_se_cmd->data_length);
+				(int)cmd->data_length);
 			return (ret < 0 ? ret : -EINVAL);
 		}
 	} else {
@@ -317,38 +297,38 @@ static int fd_do_readv(struct se_task *task)
 	return 1;
 }
 
-static int fd_do_writev(struct se_task *task)
+static int fd_do_writev(struct se_cmd *cmd, struct scatterlist *sgl,
+		u32 sgl_nents)
 {
-	struct fd_request *req = FILE_REQ(task);
-	struct se_device *se_dev = req->fd_task.task_se_cmd->se_dev;
+	struct se_device *se_dev = cmd->se_dev;
 	struct fd_dev *dev = se_dev->dev_ptr;
 	struct file *fd = dev->fd_file;
-	struct scatterlist *sg = task->task_sg;
+	struct scatterlist *sg;
 	struct iovec *iov;
 	mm_segment_t old_fs;
-	loff_t pos = (task->task_se_cmd->t_task_lba *
+	loff_t pos = (cmd->t_task_lba *
 		      se_dev->se_sub_dev->se_dev_attrib.block_size);
 	int ret, i = 0;
 
-	iov = kzalloc(sizeof(struct iovec) * task->task_sg_nents, GFP_KERNEL);
+	iov = kzalloc(sizeof(struct iovec) * sgl_nents, GFP_KERNEL);
 	if (!iov) {
 		pr_err("Unable to allocate fd_do_writev iov[]\n");
 		return -ENOMEM;
 	}
 
-	for_each_sg(task->task_sg, sg, task->task_sg_nents, i) {
+	for_each_sg(sgl, sg, sgl_nents, i) {
 		iov[i].iov_len = sg->length;
 		iov[i].iov_base = sg_virt(sg);
 	}
 
 	old_fs = get_fs();
 	set_fs(get_ds());
-	ret = vfs_writev(fd, &iov[0], task->task_sg_nents, &pos);
+	ret = vfs_writev(fd, &iov[0], sgl_nents, &pos);
 	set_fs(old_fs);
 
 	kfree(iov);
 
-	if (ret < 0 || ret != task->task_se_cmd->data_length) {
+	if (ret < 0 || ret != cmd->data_length) {
 		pr_err("vfs_writev() returned %d\n", ret);
 		return (ret < 0 ? ret : -EINVAL);
 	}
@@ -369,7 +349,7 @@ static void fd_emulate_sync_cache(struct se_cmd *cmd)
 	 * for this SYNCHRONIZE_CACHE op
 	 */
 	if (immed)
-		transport_complete_sync_cache(cmd, 1);
+		target_complete_cmd(cmd, SAM_STAT_GOOD);
 
 	/*
 	 * Determine if we will be flushing the entire device.
@@ -389,35 +369,37 @@ static void fd_emulate_sync_cache(struct se_cmd *cmd)
 	if (ret != 0)
 		pr_err("FILEIO: vfs_fsync_range() failed: %d\n", ret);
 
-	if (!immed)
-		transport_complete_sync_cache(cmd, ret == 0);
+	if (immed)
+		return;
+
+	if (ret) {
+		cmd->scsi_sense_reason = TCM_LOGICAL_UNIT_COMMUNICATION_FAILURE;
+		target_complete_cmd(cmd, SAM_STAT_CHECK_CONDITION);
+	} else {
+		target_complete_cmd(cmd, SAM_STAT_GOOD);
+	}
 }
 
-/*
- * WRITE Force Unit Access (FUA) emulation on a per struct se_task
- * LBA range basis..
- */
-static void fd_emulate_write_fua(struct se_cmd *cmd, struct se_task *task)
+static void fd_emulate_write_fua(struct se_cmd *cmd)
 {
 	struct se_device *dev = cmd->se_dev;
 	struct fd_dev *fd_dev = dev->dev_ptr;
-	loff_t start = task->task_se_cmd->t_task_lba *
+	loff_t start = cmd->t_task_lba *
 		dev->se_sub_dev->se_dev_attrib.block_size;
-	loff_t end = start + task->task_se_cmd->data_length;
+	loff_t end = start + cmd->data_length;
 	int ret;
 
 	pr_debug("FILEIO: FUA WRITE LBA: %llu, bytes: %u\n",
-			task->task_se_cmd->t_task_lba,
-			task->task_se_cmd->data_length);
+		cmd->t_task_lba, cmd->data_length);
 
 	ret = vfs_fsync_range(fd_dev->fd_file, start, end, 1);
 	if (ret != 0)
 		pr_err("FILEIO: vfs_fsync_range() failed: %d\n", ret);
 }
 
-static int fd_do_task(struct se_task *task)
+static int fd_execute_cmd(struct se_cmd *cmd, struct scatterlist *sgl,
+		u32 sgl_nents, enum dma_data_direction data_direction)
 {
-	struct se_cmd *cmd = task->task_se_cmd;
 	struct se_device *dev = cmd->se_dev;
 	int ret = 0;
 
@@ -425,10 +407,10 @@ static int fd_do_task(struct se_task *task)
 	 * Call vectorized fileio functions to map struct scatterlist
 	 * physical memory addresses to struct iovec virtual memory.
 	 */
-	if (task->task_data_direction == DMA_FROM_DEVICE) {
-		ret = fd_do_readv(task);
+	if (data_direction == DMA_FROM_DEVICE) {
+		ret = fd_do_readv(cmd, sgl, sgl_nents);
 	} else {
-		ret = fd_do_writev(task);
+		ret = fd_do_writev(cmd, sgl, sgl_nents);
 
 		if (ret > 0 &&
 		    dev->se_sub_dev->se_dev_attrib.emulate_write_cache > 0 &&
@@ -439,7 +421,7 @@ static int fd_do_task(struct se_task *task)
 			 * and return some sense data to let the initiator
 			 * know the FUA WRITE cache sync failed..?
 			 */
-			fd_emulate_write_fua(cmd, task);
+			fd_emulate_write_fua(cmd);
 		}
 
 	}
@@ -448,24 +430,11 @@ static int fd_do_task(struct se_task *task)
 		cmd->scsi_sense_reason = TCM_LOGICAL_UNIT_COMMUNICATION_FAILURE;
 		return ret;
 	}
-	if (ret) {
-		task->task_scsi_status = GOOD;
-		transport_complete_task(task, 1);
-	}
+	if (ret)
+		target_complete_cmd(cmd, SAM_STAT_GOOD);
 	return 0;
 }
 
-/*	fd_free_task(): (Part of se_subsystem_api_t template)
- *
- *
- */
-static void fd_free_task(struct se_task *task)
-{
-	struct fd_request *req = FILE_REQ(task);
-
-	kfree(req);
-}
-
 enum {
 	Opt_fd_dev_name, Opt_fd_dev_size, Opt_fd_buffered_io, Opt_err
 };
@@ -618,10 +587,8 @@ static struct se_subsystem_api fileio_template = {
 	.allocate_virtdevice	= fd_allocate_virtdevice,
 	.create_virtdevice	= fd_create_virtdevice,
 	.free_device		= fd_free_device,
-	.alloc_task		= fd_alloc_task,
-	.do_task		= fd_do_task,
+	.execute_cmd		= fd_execute_cmd,
 	.do_sync_cache		= fd_emulate_sync_cache,
-	.free_task		= fd_free_task,
 	.check_configfs_dev_params = fd_check_configfs_dev_params,
 	.set_configfs_dev_params = fd_set_configfs_dev_params,
 	.show_configfs_dev_params = fd_show_configfs_dev_params,
diff --git a/drivers/target/target_core_file.h b/drivers/target/target_core_file.h
index 59e6e73106c..fbd59ef7d8b 100644
--- a/drivers/target/target_core_file.h
+++ b/drivers/target/target_core_file.h
@@ -12,10 +12,6 @@
 #define RRF_EMULATE_CDB		0x01
 #define RRF_GOT_LBA		0x02
 
-struct fd_request {
-	struct se_task	fd_task;
-};
-
 #define FBDF_HAS_PATH		0x01
 #define FBDF_HAS_SIZE		0x02
 #define FDBD_USE_BUFFERED_IO	0x04
diff --git a/drivers/target/target_core_iblock.c b/drivers/target/target_core_iblock.c
index 82ec0d3fa9d..fd47950727b 100644
--- a/drivers/target/target_core_iblock.c
+++ b/drivers/target/target_core_iblock.c
@@ -189,26 +189,6 @@ static void iblock_free_device(void *p)
 	kfree(ib_dev);
 }
 
-static inline struct iblock_req *IBLOCK_REQ(struct se_task *task)
-{
-	return container_of(task, struct iblock_req, ib_task);
-}
-
-static struct se_task *
-iblock_alloc_task(unsigned char *cdb)
-{
-	struct iblock_req *ib_req;
-
-	ib_req = kzalloc(sizeof(struct iblock_req), GFP_KERNEL);
-	if (!ib_req) {
-		pr_err("Unable to allocate memory for struct iblock_req\n");
-		return NULL;
-	}
-
-	atomic_set(&ib_req->pending, 1);
-	return &ib_req->ib_task;
-}
-
 static unsigned long long iblock_emulate_read_cap_with_block_size(
 	struct se_device *dev,
 	struct block_device *bd,
@@ -295,8 +275,16 @@ static void iblock_end_io_flush(struct bio *bio, int err)
 	if (err)
 		pr_err("IBLOCK: cache flush failed: %d\n", err);
 
-	if (cmd)
-		transport_complete_sync_cache(cmd, err == 0);
+	if (cmd) {
+		if (err) {
+			cmd->scsi_sense_reason =
+				TCM_LOGICAL_UNIT_COMMUNICATION_FAILURE;
+			target_complete_cmd(cmd, SAM_STAT_CHECK_CONDITION);
+		} else {
+			target_complete_cmd(cmd, SAM_STAT_GOOD);
+		}
+	}
+
 	bio_put(bio);
 }
 
@@ -315,7 +303,7 @@ static void iblock_emulate_sync_cache(struct se_cmd *cmd)
 	 * for this SYNCHRONIZE_CACHE op.
 	 */
 	if (immed)
-		transport_complete_sync_cache(cmd, 1);
+		target_complete_cmd(cmd, SAM_STAT_GOOD);
 
 	bio = bio_alloc(GFP_KERNEL, 0);
 	bio->bi_end_io = iblock_end_io_flush;
@@ -334,11 +322,6 @@ static int iblock_do_discard(struct se_device *dev, sector_t lba, u32 range)
 	return blkdev_issue_discard(bd, lba, range, GFP_KERNEL, barrier);
 }
 
-static void iblock_free_task(struct se_task *task)
-{
-	kfree(IBLOCK_REQ(task));
-}
-
 enum {
 	Opt_udev_path, Opt_force, Opt_err
 };
@@ -447,19 +430,35 @@ static ssize_t iblock_show_configfs_dev_params(
 	return bl;
 }
 
+static void iblock_complete_cmd(struct se_cmd *cmd)
+{
+	struct iblock_req *ibr = cmd->priv;
+	u8 status;
+
+	if (!atomic_dec_and_test(&ibr->pending))
+		return;
+
+	if (atomic_read(&ibr->ib_bio_err_cnt))
+		status = SAM_STAT_CHECK_CONDITION;
+	else
+		status = SAM_STAT_GOOD;
+
+	target_complete_cmd(cmd, status);
+	kfree(ibr);
+}
+
 static void iblock_bio_destructor(struct bio *bio)
 {
-	struct se_task *task = bio->bi_private;
-	struct iblock_dev *ib_dev = task->task_se_cmd->se_dev->dev_ptr;
+	struct se_cmd *cmd = bio->bi_private;
+	struct iblock_dev *ib_dev = cmd->se_dev->dev_ptr;
 
 	bio_free(bio, ib_dev->ibd_bio_set);
 }
 
 static struct bio *
-iblock_get_bio(struct se_task *task, sector_t lba, u32 sg_num)
+iblock_get_bio(struct se_cmd *cmd, sector_t lba, u32 sg_num)
 {
-	struct iblock_dev *ib_dev = task->task_se_cmd->se_dev->dev_ptr;
-	struct iblock_req *ib_req = IBLOCK_REQ(task);
+	struct iblock_dev *ib_dev = cmd->se_dev->dev_ptr;
 	struct bio *bio;
 
 	/*
@@ -475,20 +474,11 @@ iblock_get_bio(struct se_task *task, sector_t lba, u32 sg_num)
 		return NULL;
 	}
 
-	pr_debug("Allocated bio: %p task_sg_nents: %u using ibd_bio_set:"
-		" %p\n", bio, task->task_sg_nents, ib_dev->ibd_bio_set);
-	pr_debug("Allocated bio: %p task_size: %u\n", bio,
-		task->task_se_cmd->data_length);
-
 	bio->bi_bdev = ib_dev->ibd_bd;
-	bio->bi_private = task;
+	bio->bi_private = cmd;
 	bio->bi_destructor = iblock_bio_destructor;
 	bio->bi_end_io = &iblock_bio_done;
 	bio->bi_sector = lba;
-	atomic_inc(&ib_req->pending);
-
-	pr_debug("Set bio->bi_sector: %llu\n", (unsigned long long)bio->bi_sector);
-	pr_debug("Set ib_req->pending: %d\n", atomic_read(&ib_req->pending));
 	return bio;
 }
 
@@ -503,20 +493,21 @@ static void iblock_submit_bios(struct bio_list *list, int rw)
 	blk_finish_plug(&plug);
 }
 
-static int iblock_do_task(struct se_task *task)
+static int iblock_execute_cmd(struct se_cmd *cmd, struct scatterlist *sgl,
+		u32 sgl_nents, enum dma_data_direction data_direction)
 {
-	struct se_cmd *cmd = task->task_se_cmd;
 	struct se_device *dev = cmd->se_dev;
-	struct iblock_req *ibr = IBLOCK_REQ(task);
+	struct iblock_req *ibr;
 	struct bio *bio;
 	struct bio_list list;
 	struct scatterlist *sg;
-	u32 i, sg_num = task->task_sg_nents;
+	u32 sg_num = sgl_nents;
 	sector_t block_lba;
 	unsigned bio_cnt;
 	int rw;
+	int i;
 
-	if (task->task_data_direction == DMA_TO_DEVICE) {
+	if (data_direction == DMA_TO_DEVICE) {
 		/*
 		 * Force data to disk if we pretend to not have a volatile
 		 * write cache, or the initiator set the Force Unit Access bit.
@@ -532,8 +523,8 @@ static int iblock_do_task(struct se_task *task)
 	}
 
 	/*
-	 * Do starting conversion up from non 512-byte blocksize with
-	 * struct se_task SCSI blocksize into Linux/Block 512 units for BIO.
+	 * Convert the blocksize advertised to the initiator to the 512 byte
+	 * units unconditionally used by the Linux block layer.
 	 */
 	if (dev->se_sub_dev->se_dev_attrib.block_size == 4096)
 		block_lba = (cmd->t_task_lba << 3);
@@ -550,17 +541,22 @@ static int iblock_do_task(struct se_task *task)
 		return -ENOSYS;
 	}
 
-	bio = iblock_get_bio(task, block_lba, sg_num);
-	if (!bio) {
-		cmd->scsi_sense_reason = TCM_LOGICAL_UNIT_COMMUNICATION_FAILURE;
-		return -ENOMEM;
-	}
+	ibr = kzalloc(sizeof(struct iblock_req), GFP_KERNEL);
+	if (!ibr)
+		goto fail;
+	cmd->priv = ibr;
+
+	bio = iblock_get_bio(cmd, block_lba, sgl_nents);
+	if (!bio)
+		goto fail_free_ibr;
 
 	bio_list_init(&list);
 	bio_list_add(&list, bio);
+
+	atomic_set(&ibr->pending, 2);
 	bio_cnt = 1;
 
-	for_each_sg(task->task_sg, sg, task->task_sg_nents, i) {
+	for_each_sg(sgl, sg, sgl_nents, i) {
 		/*
 		 * XXX: if the length the device accepts is shorter than the
 		 *	length of the S/G list entry this will cause and
@@ -573,9 +569,11 @@ static int iblock_do_task(struct se_task *task)
 				bio_cnt = 0;
 			}
 
-			bio = iblock_get_bio(task, block_lba, sg_num);
+			bio = iblock_get_bio(cmd, block_lba, sg_num);
 			if (!bio)
-				goto fail;
+				goto fail_put_bios;
+
+			atomic_inc(&ibr->pending);
 			bio_list_add(&list, bio);
 			bio_cnt++;
 		}
@@ -586,17 +584,16 @@ static int iblock_do_task(struct se_task *task)
 	}
 
 	iblock_submit_bios(&list, rw);
-
-	if (atomic_dec_and_test(&ibr->pending)) {
-		transport_complete_task(task,
-				!atomic_read(&ibr->ib_bio_err_cnt));
-	}
+	iblock_complete_cmd(cmd);
 	return 0;
 
-fail:
+fail_put_bios:
 	while ((bio = bio_list_pop(&list)))
 		bio_put(bio);
+fail_free_ibr:
+	kfree(ibr);
 	cmd->scsi_sense_reason = TCM_LOGICAL_UNIT_COMMUNICATION_FAILURE;
+fail:
 	return -ENOMEM;
 }
 
@@ -621,8 +618,8 @@ static sector_t iblock_get_blocks(struct se_device *dev)
 
 static void iblock_bio_done(struct bio *bio, int err)
 {
-	struct se_task *task = bio->bi_private;
-	struct iblock_req *ibr = IBLOCK_REQ(task);
+	struct se_cmd *cmd = bio->bi_private;
+	struct iblock_req *ibr = cmd->priv;
 
 	/*
 	 * Set -EIO if !BIO_UPTODATE and the passed is still err=0
@@ -642,14 +639,7 @@ static void iblock_bio_done(struct bio *bio, int err)
 
 	bio_put(bio);
 
-	if (!atomic_dec_and_test(&ibr->pending))
-		return;
-
-	pr_debug("done[%p] bio: %p task_lba: %llu bio_lba: %llu err=%d\n",
-		 task, bio, task->task_se_cmd->t_task_lba,
-		 (unsigned long long)bio->bi_sector, err);
-
-	transport_complete_task(task, !atomic_read(&ibr->ib_bio_err_cnt));
+	iblock_complete_cmd(cmd);
 }
 
 static struct se_subsystem_api iblock_template = {
@@ -663,11 +653,9 @@ static struct se_subsystem_api iblock_template = {
 	.allocate_virtdevice	= iblock_allocate_virtdevice,
 	.create_virtdevice	= iblock_create_virtdevice,
 	.free_device		= iblock_free_device,
-	.alloc_task		= iblock_alloc_task,
-	.do_task		= iblock_do_task,
+	.execute_cmd		= iblock_execute_cmd,
 	.do_discard		= iblock_do_discard,
 	.do_sync_cache		= iblock_emulate_sync_cache,
-	.free_task		= iblock_free_task,
 	.check_configfs_dev_params = iblock_check_configfs_dev_params,
 	.set_configfs_dev_params = iblock_set_configfs_dev_params,
 	.show_configfs_dev_params = iblock_show_configfs_dev_params,
diff --git a/drivers/target/target_core_iblock.h b/drivers/target/target_core_iblock.h
index e929370b6fd..66cf7b9e205 100644
--- a/drivers/target/target_core_iblock.h
+++ b/drivers/target/target_core_iblock.h
@@ -7,7 +7,6 @@
 #define IBLOCK_LBA_SHIFT	9
 
 struct iblock_req {
-	struct se_task ib_task;
 	atomic_t pending;
 	atomic_t ib_bio_err_cnt;
 } ____cacheline_aligned;
diff --git a/drivers/target/target_core_pscsi.c b/drivers/target/target_core_pscsi.c
index 46eb017287a..4ce2cf642fc 100644
--- a/drivers/target/target_core_pscsi.c
+++ b/drivers/target/target_core_pscsi.c
@@ -663,22 +663,12 @@ static void pscsi_free_device(void *p)
 	kfree(pdv);
 }
 
-static inline struct pscsi_plugin_task *PSCSI_TASK(struct se_task *task)
+static int pscsi_transport_complete(struct se_cmd *cmd, struct scatterlist *sg)
 {
-	return container_of(task, struct pscsi_plugin_task, pscsi_task);
-}
-
-
-/*	pscsi_transport_complete():
- *
- *
- */
-static int pscsi_transport_complete(struct se_task *task)
-{
-	struct pscsi_dev_virt *pdv = task->task_se_cmd->se_dev->dev_ptr;
+	struct pscsi_dev_virt *pdv = cmd->se_dev->dev_ptr;
 	struct scsi_device *sd = pdv->pdv_sd;
 	int result;
-	struct pscsi_plugin_task *pt = PSCSI_TASK(task);
+	struct pscsi_plugin_task *pt = cmd->priv;
 	unsigned char *cdb = &pt->pscsi_cdb[0];
 
 	result = pt->pscsi_result;
@@ -688,12 +678,11 @@ static int pscsi_transport_complete(struct se_task *task)
 	 */
 	if (((cdb[0] == MODE_SENSE) || (cdb[0] == MODE_SENSE_10)) &&
 	     (status_byte(result) << 1) == SAM_STAT_GOOD) {
-		if (!task->task_se_cmd->se_deve)
+		if (!cmd->se_deve)
 			goto after_mode_sense;
 
-		if (task->task_se_cmd->se_deve->lun_flags &
-				TRANSPORT_LUNFLAGS_READ_ONLY) {
-			unsigned char *buf = transport_kmap_data_sg(task->task_se_cmd);
+		if (cmd->se_deve->lun_flags & TRANSPORT_LUNFLAGS_READ_ONLY) {
+			unsigned char *buf = transport_kmap_data_sg(cmd);
 
 			if (cdb[0] == MODE_SENSE_10) {
 				if (!(buf[3] & 0x80))
@@ -703,7 +692,7 @@ static int pscsi_transport_complete(struct se_task *task)
 					buf[2] |= 0x80;
 			}
 
-			transport_kunmap_data_sg(task->task_se_cmd);
+			transport_kunmap_data_sg(cmd);
 		}
 	}
 after_mode_sense:
@@ -722,7 +711,6 @@ after_mode_sense:
 	if (((cdb[0] == MODE_SELECT) || (cdb[0] == MODE_SELECT_10)) &&
 	      (status_byte(result) << 1) == SAM_STAT_GOOD) {
 		unsigned char *buf;
-		struct scatterlist *sg = task->task_sg;
 		u16 bdl;
 		u32 blocksize;
 
@@ -757,35 +745,6 @@ after_mode_select:
 	return 0;
 }
 
-static struct se_task *
-pscsi_alloc_task(unsigned char *cdb)
-{
-	struct pscsi_plugin_task *pt;
-
-	/*
-	 * Dynamically alloc cdb space, since it may be larger than
-	 * TCM_MAX_COMMAND_SIZE
-	 */
-	pt = kzalloc(sizeof(*pt) + scsi_command_size(cdb), GFP_KERNEL);
-	if (!pt) {
-		pr_err("Unable to allocate struct pscsi_plugin_task\n");
-		return NULL;
-	}
-
-	return &pt->pscsi_task;
-}
-
-static void pscsi_free_task(struct se_task *task)
-{
-	struct pscsi_plugin_task *pt = PSCSI_TASK(task);
-
-	/*
-	 * We do not release the bio(s) here associated with this task, as
-	 * this is handled by bio_put() and pscsi_bi_endio().
-	 */
-	kfree(pt);
-}
-
 enum {
 	Opt_scsi_host_id, Opt_scsi_channel_id, Opt_scsi_target_id,
 	Opt_scsi_lun_id, Opt_err
@@ -958,26 +917,25 @@ static inline struct bio *pscsi_get_bio(int sg_num)
 	return bio;
 }
 
-static int pscsi_map_sg(struct se_task *task, struct scatterlist *task_sg,
+static int pscsi_map_sg(struct se_cmd *cmd, struct scatterlist *sgl,
+		u32 sgl_nents, enum dma_data_direction data_direction,
 		struct bio **hbio)
 {
-	struct se_cmd *cmd = task->task_se_cmd;
-	struct pscsi_dev_virt *pdv = task->task_se_cmd->se_dev->dev_ptr;
-	u32 task_sg_num = task->task_sg_nents;
+	struct pscsi_dev_virt *pdv = cmd->se_dev->dev_ptr;
 	struct bio *bio = NULL, *tbio = NULL;
 	struct page *page;
 	struct scatterlist *sg;
 	u32 data_len = cmd->data_length, i, len, bytes, off;
-	int nr_pages = (cmd->data_length + task_sg[0].offset +
+	int nr_pages = (cmd->data_length + sgl[0].offset +
 			PAGE_SIZE - 1) >> PAGE_SHIFT;
 	int nr_vecs = 0, rc;
-	int rw = (task->task_data_direction == DMA_TO_DEVICE);
+	int rw = (data_direction == DMA_TO_DEVICE);
 
 	*hbio = NULL;
 
 	pr_debug("PSCSI: nr_pages: %d\n", nr_pages);
 
-	for_each_sg(task_sg, sg, task_sg_num, i) {
+	for_each_sg(sgl, sg, sgl_nents, i) {
 		page = sg_page(sg);
 		off = sg->offset;
 		len = sg->length;
@@ -1009,7 +967,7 @@ static int pscsi_map_sg(struct se_task *task, struct scatterlist *task_sg,
 				 * Set *hbio pointer to handle the case:
 				 * nr_pages > BIO_MAX_PAGES, where additional
 				 * bios need to be added to complete a given
-				 * struct se_task
+				 * command.
 				 */
 				if (!*hbio)
 					*hbio = tbio = bio;
@@ -1049,7 +1007,7 @@ static int pscsi_map_sg(struct se_task *task, struct scatterlist *task_sg,
 		}
 	}
 
-	return task->task_sg_nents;
+	return sgl_nents;
 fail:
 	while (*hbio) {
 		bio = *hbio;
@@ -1061,53 +1019,61 @@ fail:
 	return -ENOMEM;
 }
 
-static int pscsi_do_task(struct se_task *task)
+static int pscsi_execute_cmd(struct se_cmd *cmd, struct scatterlist *sgl,
+		u32 sgl_nents, enum dma_data_direction data_direction)
 {
-	struct se_cmd *cmd = task->task_se_cmd;
-	struct pscsi_dev_virt *pdv = task->task_se_cmd->se_dev->dev_ptr;
-	struct pscsi_plugin_task *pt = PSCSI_TASK(task);
+	struct pscsi_dev_virt *pdv = cmd->se_dev->dev_ptr;
+	struct pscsi_plugin_task *pt;
 	struct request *req;
 	struct bio *hbio;
 	int ret;
 
+	/*
+	 * Dynamically alloc cdb space, since it may be larger than
+	 * TCM_MAX_COMMAND_SIZE
+	 */
+	pt = kzalloc(sizeof(*pt) + scsi_command_size(cmd->t_task_cdb), GFP_KERNEL);
+	if (!pt) {
+		cmd->scsi_sense_reason = TCM_LOGICAL_UNIT_COMMUNICATION_FAILURE;
+		return -ENOMEM;
+	}
+	cmd->priv = pt;
+
 	memcpy(pt->pscsi_cdb, cmd->t_task_cdb,
 		scsi_command_size(cmd->t_task_cdb));
 
-	if (task->task_se_cmd->se_cmd_flags & SCF_SCSI_NON_DATA_CDB) {
+	if (cmd->se_cmd_flags & SCF_SCSI_NON_DATA_CDB) {
 		req = blk_get_request(pdv->pdv_sd->request_queue,
-				(task->task_data_direction == DMA_TO_DEVICE),
+				(data_direction == DMA_TO_DEVICE),
 				GFP_KERNEL);
 		if (!req || IS_ERR(req)) {
 			pr_err("PSCSI: blk_get_request() failed: %ld\n",
 					req ? IS_ERR(req) : -ENOMEM);
 			cmd->scsi_sense_reason =
 				TCM_LOGICAL_UNIT_COMMUNICATION_FAILURE;
-			return -ENODEV;
+			goto fail;
 		}
 	} else {
 		BUG_ON(!cmd->data_length);
 
-		/*
-		 * Setup the main struct request for the task->task_sg[] payload
-		 */
-		ret = pscsi_map_sg(task, task->task_sg, &hbio);
+		ret = pscsi_map_sg(cmd, sgl, sgl_nents, data_direction, &hbio);
 		if (ret < 0) {
 			cmd->scsi_sense_reason =
 				TCM_LOGICAL_UNIT_COMMUNICATION_FAILURE;
-			return ret;
+			goto fail;
 		}
 
 		req = blk_make_request(pdv->pdv_sd->request_queue, hbio,
 				       GFP_KERNEL);
 		if (IS_ERR(req)) {
 			pr_err("pSCSI: blk_make_request() failed\n");
-			goto fail;
+			goto fail_free_bio;
 		}
 	}
 
 	req->cmd_type = REQ_TYPE_BLOCK_PC;
 	req->end_io = pscsi_req_done;
-	req->end_io_data = task;
+	req->end_io_data = cmd;
 	req->cmd_len = scsi_command_size(pt->pscsi_cdb);
 	req->cmd = &pt->pscsi_cdb[0];
 	req->sense = &pt->pscsi_sense[0];
@@ -1119,12 +1085,12 @@ static int pscsi_do_task(struct se_task *task)
 	req->retries = PS_RETRY;
 
 	blk_execute_rq_nowait(pdv->pdv_sd->request_queue, NULL, req,
-			(task->task_se_cmd->sam_task_attr == MSG_HEAD_TAG),
+			(cmd->sam_task_attr == MSG_HEAD_TAG),
 			pscsi_req_done);
 
 	return 0;
 
-fail:
+fail_free_bio:
 	while (hbio) {
 		struct bio *bio = hbio;
 		hbio = hbio->bi_next;
@@ -1132,16 +1098,14 @@ fail:
 		bio_endio(bio, 0);	/* XXX: should be error */
 	}
 	cmd->scsi_sense_reason = TCM_LOGICAL_UNIT_COMMUNICATION_FAILURE;
+fail:
+	kfree(pt);
 	return -ENOMEM;
 }
 
-/*	pscsi_get_sense_buffer():
- *
- *
- */
-static unsigned char *pscsi_get_sense_buffer(struct se_task *task)
+static unsigned char *pscsi_get_sense_buffer(struct se_cmd *cmd)
 {
-	struct pscsi_plugin_task *pt = PSCSI_TASK(task);
+	struct pscsi_plugin_task *pt = cmd->priv;
 
 	return pt->pscsi_sense;
 }
@@ -1181,48 +1145,36 @@ static sector_t pscsi_get_blocks(struct se_device *dev)
 	return 0;
 }
 
-/*	pscsi_handle_SAM_STATUS_failures():
- *
- *
- */
-static inline void pscsi_process_SAM_status(
-	struct se_task *task,
-	struct pscsi_plugin_task *pt)
+static void pscsi_req_done(struct request *req, int uptodate)
 {
-	task->task_scsi_status = status_byte(pt->pscsi_result);
-	if (task->task_scsi_status) {
-		task->task_scsi_status <<= 1;
-		pr_debug("PSCSI Status Byte exception at task: %p CDB:"
-			" 0x%02x Result: 0x%08x\n", task, pt->pscsi_cdb[0],
+	struct se_cmd *cmd = req->end_io_data;
+	struct pscsi_plugin_task *pt = cmd->priv;
+
+	pt->pscsi_result = req->errors;
+	pt->pscsi_resid = req->resid_len;
+
+	cmd->scsi_status = status_byte(pt->pscsi_result) << 1;
+	if (cmd->scsi_status) {
+		pr_debug("PSCSI Status Byte exception at cmd: %p CDB:"
+			" 0x%02x Result: 0x%08x\n", cmd, pt->pscsi_cdb[0],
 			pt->pscsi_result);
 	}
 
 	switch (host_byte(pt->pscsi_result)) {
 	case DID_OK:
-		transport_complete_task(task, (!task->task_scsi_status));
+		target_complete_cmd(cmd, cmd->scsi_status);
 		break;
 	default:
-		pr_debug("PSCSI Host Byte exception at task: %p CDB:"
-			" 0x%02x Result: 0x%08x\n", task, pt->pscsi_cdb[0],
+		pr_debug("PSCSI Host Byte exception at cmd: %p CDB:"
+			" 0x%02x Result: 0x%08x\n", cmd, pt->pscsi_cdb[0],
 			pt->pscsi_result);
-		task->task_scsi_status = SAM_STAT_CHECK_CONDITION;
-		task->task_se_cmd->scsi_sense_reason =
-					TCM_UNSUPPORTED_SCSI_OPCODE;
-		transport_complete_task(task, 0);
+		cmd->scsi_sense_reason = TCM_UNSUPPORTED_SCSI_OPCODE;
+		target_complete_cmd(cmd, SAM_STAT_CHECK_CONDITION);
 		break;
 	}
-}
 
-static void pscsi_req_done(struct request *req, int uptodate)
-{
-	struct se_task *task = req->end_io_data;
-	struct pscsi_plugin_task *pt = PSCSI_TASK(task);
-
-	pt->pscsi_result = req->errors;
-	pt->pscsi_resid = req->resid_len;
-
-	pscsi_process_SAM_status(task, pt);
 	__blk_put_request(req->q, req);
+	kfree(pt);
 }
 
 static struct se_subsystem_api pscsi_template = {
@@ -1236,9 +1188,7 @@ static struct se_subsystem_api pscsi_template = {
 	.create_virtdevice	= pscsi_create_virtdevice,
 	.free_device		= pscsi_free_device,
 	.transport_complete	= pscsi_transport_complete,
-	.alloc_task		= pscsi_alloc_task,
-	.do_task		= pscsi_do_task,
-	.free_task		= pscsi_free_task,
+	.execute_cmd		= pscsi_execute_cmd,
 	.check_configfs_dev_params = pscsi_check_configfs_dev_params,
 	.set_configfs_dev_params = pscsi_set_configfs_dev_params,
 	.show_configfs_dev_params = pscsi_show_configfs_dev_params,
diff --git a/drivers/target/target_core_pscsi.h b/drivers/target/target_core_pscsi.h
index 43f1c419e8e..bc1e5e11eca 100644
--- a/drivers/target/target_core_pscsi.h
+++ b/drivers/target/target_core_pscsi.h
@@ -22,7 +22,6 @@
 #include <linux/kobject.h>
 
 struct pscsi_plugin_task {
-	struct se_task pscsi_task;
 	unsigned char pscsi_sense[SCSI_SENSE_BUFFERSIZE];
 	int	pscsi_direction;
 	int	pscsi_result;
diff --git a/drivers/target/target_core_rd.c b/drivers/target/target_core_rd.c
index c6ce46891be..d0ceb873c0e 100644
--- a/drivers/target/target_core_rd.c
+++ b/drivers/target/target_core_rd.c
@@ -266,12 +266,6 @@ static void rd_free_device(void *p)
 	kfree(rd_dev);
 }
 
-static struct se_task *
-rd_alloc_task(unsigned char *cdb)
-{
-	return kzalloc(sizeof(struct se_task), GFP_KERNEL);
-}
-
 static struct rd_dev_sg_table *rd_get_sg_table(struct rd_dev *rd_dev, u32 page)
 {
 	u32 i;
@@ -290,9 +284,10 @@ static struct rd_dev_sg_table *rd_get_sg_table(struct rd_dev *rd_dev, u32 page)
 	return NULL;
 }
 
-static int rd_do_task(struct se_task *task)
+static int rd_execute_cmd(struct se_cmd *cmd, struct scatterlist *sgl,
+		u32 sgl_nents, enum dma_data_direction data_direction)
 {
-	struct se_device *se_dev = task->task_se_cmd->se_dev;
+	struct se_device *se_dev = cmd->se_dev;
 	struct rd_dev *dev = se_dev->dev_ptr;
 	struct rd_dev_sg_table *table;
 	struct scatterlist *rd_sg;
@@ -303,11 +298,10 @@ static int rd_do_task(struct se_task *task)
 	u32 src_len;
 	u64 tmp;
 
-	tmp = task->task_se_cmd->t_task_lba *
-		se_dev->se_sub_dev->se_dev_attrib.block_size;
+	tmp = cmd->t_task_lba * se_dev->se_sub_dev->se_dev_attrib.block_size;
 	rd_offset = do_div(tmp, PAGE_SIZE);
 	rd_page = tmp;
-	rd_size = task->task_se_cmd->data_length;
+	rd_size = cmd->data_length;
 
 	table = rd_get_sg_table(dev, rd_page);
 	if (!table)
@@ -317,14 +311,12 @@ static int rd_do_task(struct se_task *task)
 
 	pr_debug("RD[%u]: %s LBA: %llu, Size: %u Page: %u, Offset: %u\n",
 			dev->rd_dev_id,
-			task->task_data_direction == DMA_FROM_DEVICE ?
-				"Read" : "Write",
-			task->task_se_cmd->t_task_lba,
-			rd_size, rd_page, rd_offset);
+			data_direction == DMA_FROM_DEVICE ? "Read" : "Write",
+			cmd->t_task_lba, rd_size, rd_page, rd_offset);
 
 	src_len = PAGE_SIZE - rd_offset;
-	sg_miter_start(&m, task->task_sg, task->task_sg_nents,
-			task->task_data_direction == DMA_FROM_DEVICE ?
+	sg_miter_start(&m, sgl, sgl_nents,
+			data_direction == DMA_FROM_DEVICE ?
 				SG_MITER_TO_SG : SG_MITER_FROM_SG);
 	while (rd_size) {
 		u32 len;
@@ -336,7 +328,7 @@ static int rd_do_task(struct se_task *task)
 
 		rd_addr = sg_virt(rd_sg) + rd_offset;
 
-		if (task->task_data_direction == DMA_FROM_DEVICE)
+		if (data_direction == DMA_FROM_DEVICE)
 			memcpy(m.addr, rd_addr, len);
 		else
 			memcpy(rd_addr, m.addr, len);
@@ -371,16 +363,10 @@ static int rd_do_task(struct se_task *task)
 	}
 	sg_miter_stop(&m);
 
-	task->task_scsi_status = GOOD;
-	transport_complete_task(task, 1);
+	target_complete_cmd(cmd, SAM_STAT_GOOD);
 	return 0;
 }
 
-static void rd_free_task(struct se_task *task)
-{
-	kfree(task);
-}
-
 enum {
 	Opt_rd_pages, Opt_err
 };
@@ -482,9 +468,7 @@ static struct se_subsystem_api rd_mcp_template = {
 	.allocate_virtdevice	= rd_allocate_virtdevice,
 	.create_virtdevice	= rd_create_virtdevice,
 	.free_device		= rd_free_device,
-	.alloc_task		= rd_alloc_task,
-	.do_task		= rd_do_task,
-	.free_task		= rd_free_task,
+	.execute_cmd		= rd_execute_cmd,
 	.check_configfs_dev_params = rd_check_configfs_dev_params,
 	.set_configfs_dev_params = rd_set_configfs_dev_params,
 	.show_configfs_dev_params = rd_show_configfs_dev_params,
diff --git a/drivers/target/target_core_tmr.c b/drivers/target/target_core_tmr.c
index e0ffbdc4a48..6e2378a88bd 100644
--- a/drivers/target/target_core_tmr.c
+++ b/drivers/target/target_core_tmr.c
@@ -374,13 +374,11 @@ static void core_tmr_drain_cmd_list(
 	struct se_queue_obj *qobj = &dev->dev_queue_obj;
 	struct se_cmd *cmd, *tcmd;
 	unsigned long flags;
+
 	/*
-	 * Release all commands remaining in the struct se_device cmd queue.
+	 * Release all commands remaining in the per-device command queue.
 	 *
-	 * This follows the same logic as above for the struct se_device
-	 * struct se_task state list, where commands are returned with
-	 * TASK_ABORTED status, if there is an outstanding $FABRIC_MOD
-	 * reference, otherwise the struct se_cmd is released.
+	 * This follows the same logic as above for the state list.
 	 */
 	spin_lock_irqsave(&qobj->cmd_queue_lock, flags);
 	list_for_each_entry_safe(cmd, tcmd, &qobj->qobj_list, se_queue_node) {
diff --git a/drivers/target/target_core_transport.c b/drivers/target/target_core_transport.c
index 4ee33954697..47669c6d917 100644
--- a/drivers/target/target_core_transport.c
+++ b/drivers/target/target_core_transport.c
@@ -72,7 +72,6 @@ static int __transport_execute_tasks(struct se_device *dev, struct se_cmd *);
 static void transport_complete_task_attr(struct se_cmd *cmd);
 static void transport_handle_queue_full(struct se_cmd *cmd,
 		struct se_device *dev);
-static void transport_free_dev_tasks(struct se_cmd *cmd);
 static int transport_generic_get_mem(struct se_cmd *cmd);
 static void transport_put_cmd(struct se_cmd *cmd);
 static void transport_remove_cmd_from_queue(struct se_cmd *cmd);
@@ -662,28 +661,6 @@ static void transport_remove_cmd_from_queue(struct se_cmd *cmd)
 	spin_unlock_irqrestore(&qobj->cmd_queue_lock, flags);
 }
 
-/*
- * Completion function used by TCM subsystem plugins (such as FILEIO)
- * for queueing up response from struct se_subsystem_api->do_task()
- */
-void transport_complete_sync_cache(struct se_cmd *cmd, int good)
-{
-	struct se_task *task = cmd->t_task;
-
-	if (good) {
-		cmd->scsi_status = SAM_STAT_GOOD;
-		task->task_scsi_status = GOOD;
-	} else {
-		task->task_scsi_status = SAM_STAT_CHECK_CONDITION;
-		task->task_se_cmd->scsi_sense_reason =
-				TCM_LOGICAL_UNIT_COMMUNICATION_FAILURE;
-
-	}
-
-	transport_complete_task(task, good);
-}
-EXPORT_SYMBOL(transport_complete_sync_cache);
-
 static void target_complete_failure_work(struct work_struct *work)
 {
 	struct se_cmd *cmd = container_of(work, struct se_cmd, work);
@@ -691,35 +668,28 @@ static void target_complete_failure_work(struct work_struct *work)
 	transport_generic_request_failure(cmd);
 }
 
-/*	transport_complete_task():
- *
- *	Called from interrupt and non interrupt context depending
- *	on the transport plugin.
- */
-void transport_complete_task(struct se_task *task, int success)
+void target_complete_cmd(struct se_cmd *cmd, u8 scsi_status)
 {
-	struct se_cmd *cmd = task->task_se_cmd;
 	struct se_device *dev = cmd->se_dev;
+	int success = scsi_status == GOOD;
 	unsigned long flags;
 
+	cmd->scsi_status = scsi_status;
+
+
 	spin_lock_irqsave(&cmd->t_state_lock, flags);
 	cmd->transport_state &= ~CMD_T_BUSY;
 
-	/*
-	 * See if any sense data exists, if so set the TASK_SENSE flag.
-	 * Also check for any other post completion work that needs to be
-	 * done by the plugins.
-	 */
 	if (dev && dev->transport->transport_complete) {
-		if (dev->transport->transport_complete(task) != 0) {
+		if (dev->transport->transport_complete(cmd,
+				cmd->t_data_sg) != 0) {
 			cmd->se_cmd_flags |= SCF_TRANSPORT_TASK_SENSE;
 			success = 1;
 		}
 	}
 
 	/*
-	 * See if we are waiting for outstanding struct se_task
-	 * to complete for an exception condition
+	 * See if we are waiting to complete for an exception condition.
 	 */
 	if (cmd->transport_state & CMD_T_REQUEST_STOP) {
 		spin_unlock_irqrestore(&cmd->t_state_lock, flags);
@@ -730,15 +700,11 @@ void transport_complete_task(struct se_task *task, int success)
 	if (!success)
 		cmd->transport_state |= CMD_T_FAILED;
 
-	/*
-	 * Decrement the outstanding t_task_cdbs_left count.  The last
-	 * struct se_task from struct se_cmd will complete itself into the
-	 * device queue depending upon int success.
-	 */
 	if (!atomic_dec_and_test(&cmd->t_task_cdbs_left)) {
 		spin_unlock_irqrestore(&cmd->t_state_lock, flags);
 		return;
 	}
+
 	/*
 	 * Check for case where an explict ABORT_TASK has been received
 	 * and transport_wait_for_tasks() will be waiting for completion..
@@ -761,15 +727,6 @@ void transport_complete_task(struct se_task *task, int success)
 
 	queue_work(target_completion_wq, &cmd->work);
 }
-EXPORT_SYMBOL(transport_complete_task);
-
-void target_complete_cmd(struct se_cmd *cmd, u8 scsi_status)
-{
-	struct se_task *task = cmd->t_task;
-
-	task->task_scsi_status = scsi_status;
-	transport_complete_task(task, scsi_status == GOOD);
-}
 EXPORT_SYMBOL(target_complete_cmd);
 
 static void target_add_to_state_list(struct se_cmd *cmd)
@@ -2076,8 +2033,10 @@ check_depth:
 
 	if (cmd->execute_cmd)
 		error = cmd->execute_cmd(cmd);
-	else
-		error = dev->transport->do_task(cmd->t_task);
+	else {
+		error = dev->transport->execute_cmd(cmd, cmd->t_data_sg,
+				cmd->t_data_nents, cmd->data_direction);
+	}
 
 	if (error != 0) {
 		spin_lock_irqsave(&cmd->t_state_lock, flags);
@@ -2312,7 +2271,6 @@ static int transport_get_sense_data(struct se_cmd *cmd)
 {
 	unsigned char *buffer = cmd->sense_buffer, *sense_buffer = NULL;
 	struct se_device *dev = cmd->se_dev;
-	struct se_task *task = NULL;
 	unsigned long flags;
 	u32 offset = 0;
 
@@ -2327,9 +2285,6 @@ static int transport_get_sense_data(struct se_cmd *cmd)
 		return 0;
 	}
 
-	if (!cmd->t_task)
-		goto out;
-
 	if (!(cmd->se_cmd_flags & SCF_TRANSPORT_TASK_SENSE))
 		goto out;
 
@@ -2338,19 +2293,19 @@ static int transport_get_sense_data(struct se_cmd *cmd)
 		goto out;
 	}
 
-	sense_buffer = dev->transport->get_sense_buffer(task);
+	sense_buffer = dev->transport->get_sense_buffer(cmd);
 	if (!sense_buffer) {
-		pr_err("ITT[0x%08x]_TASK[%p]: Unable to locate"
+		pr_err("ITT 0x%08x cmd %p: Unable to locate"
 			" sense buffer for task with sense\n",
-			cmd->se_tfo->get_task_tag(cmd), task);
+			cmd->se_tfo->get_task_tag(cmd), cmd);
 		goto out;
 	}
+
 	spin_unlock_irqrestore(&cmd->t_state_lock, flags);
 
 	offset = cmd->se_tfo->set_fabric_sense_len(cmd, TRANSPORT_SENSE_BUFFER);
 
 	memcpy(&buffer[offset], sense_buffer, TRANSPORT_SENSE_BUFFER);
-	cmd->scsi_status = task->task_scsi_status;
 
 	/* Automatically padded */
 	cmd->scsi_sense_length = TRANSPORT_SENSE_BUFFER + offset;
@@ -3199,10 +3154,6 @@ static void target_complete_ok_work(struct work_struct *work)
 		if (transport_get_sense_data(cmd) < 0)
 			reason = TCM_NON_EXISTENT_LUN;
 
-		/*
-		 * Only set when an struct se_task->task_scsi_status returned
-		 * a non GOOD status.
-		 */
 		if (cmd->scsi_status) {
 			ret = transport_send_check_condition_and_sense(
 					cmd, reason, 1);
@@ -3277,15 +3228,6 @@ queue_full:
 	transport_handle_queue_full(cmd, cmd->se_dev);
 }
 
-static void transport_free_dev_tasks(struct se_cmd *cmd)
-{
-	struct se_task *task;
-
-	task = cmd->t_task;
-	if (task && !(cmd->transport_state & CMD_T_BUSY))
-		cmd->se_dev->transport->free_task(task);
-}
-
 static inline void transport_free_sgl(struct scatterlist *sgl, int nents)
 {
 	struct scatterlist *sg;
@@ -3346,7 +3288,6 @@ static void transport_release_cmd(struct se_cmd *cmd)
 static void transport_put_cmd(struct se_cmd *cmd)
 {
 	unsigned long flags;
-	int free_tasks = 0;
 
 	spin_lock_irqsave(&cmd->t_state_lock, flags);
 	if (atomic_read(&cmd->t_fe_count)) {
@@ -3362,13 +3303,9 @@ static void transport_put_cmd(struct se_cmd *cmd)
 	if (cmd->transport_state & CMD_T_DEV_ACTIVE) {
 		cmd->transport_state &= ~CMD_T_DEV_ACTIVE;
 		target_remove_from_state_list(cmd);
-		free_tasks = 1;
 	}
 	spin_unlock_irqrestore(&cmd->t_state_lock, flags);
 
-	if (free_tasks != 0)
-		transport_free_dev_tasks(cmd);
-
 	transport_free_pages(cmd);
 	transport_release_cmd(cmd);
 	return;
@@ -3526,7 +3463,6 @@ out:
 int transport_generic_new_cmd(struct se_cmd *cmd)
 {
 	struct se_device *dev = cmd->se_dev;
-	struct se_task *task;
 	int ret = 0;
 
 	/*
@@ -3572,19 +3508,6 @@ int transport_generic_new_cmd(struct se_cmd *cmd)
 			attr->max_sectors);
 	}
 
-	task = dev->transport->alloc_task(cmd->t_task_cdb);
-	if (!task) {
-		pr_err("Unable to allocate struct se_task\n");
-		goto out_fail;
-	}
-
-	task->task_se_cmd = cmd;
-	task->task_data_direction = cmd->data_direction;
-	task->task_sg = cmd->t_data_sg;
-	task->task_sg_nents = cmd->t_data_nents;
-
-	cmd->t_task = task;
-
 	atomic_inc(&cmd->t_fe_count);
 	atomic_inc(&cmd->t_se_count);
 
@@ -3592,19 +3515,17 @@ int transport_generic_new_cmd(struct se_cmd *cmd)
 	atomic_set(&cmd->t_task_cdbs_ex_left, 1);
 
 	/*
-	 * For WRITEs, let the fabric know its buffer is ready..
-	 * This WRITE struct se_cmd (and all of its associated struct se_task's)
-	 * will be added to the struct se_device execution queue after its WRITE
-	 * data has arrived. (ie: It gets handled by the transport processing
-	 * thread a second time)
+	 * For WRITEs, let the fabric know its buffer is ready.
+	 *
+	 * The command will be added to the execution queue after its write
+	 * data has arrived.
 	 */
 	if (cmd->data_direction == DMA_TO_DEVICE) {
 		target_add_to_state_list(cmd);
 		return transport_generic_write_pending(cmd);
 	}
 	/*
-	 * Everything else but a WRITE, add the struct se_cmd's struct se_task's
-	 * to the execution queue.
+	 * Everything else but a WRITE, add the command to the execution queue.
 	 */
 	transport_execute_tasks(cmd);
 	return 0;
@@ -3691,8 +3612,6 @@ void transport_generic_free_cmd(struct se_cmd *cmd, int wait_for_tasks)
 		if (cmd->se_lun)
 			transport_lun_remove_cmd(cmd);
 
-		transport_free_dev_tasks(cmd);
-
 		transport_put_cmd(cmd);
 	}
 }
@@ -3832,7 +3751,6 @@ EXPORT_SYMBOL(target_wait_for_sess_cmds);
  */
 static int transport_lun_wait_for_tasks(struct se_cmd *cmd, struct se_lun *lun)
 {
-	struct se_task *task = cmd->t_task;
 	unsigned long flags;
 	int ret = 0;
 
@@ -3944,7 +3862,6 @@ static void __transport_clear_lun_from_sessions(struct se_lun *lun)
 		target_remove_from_state_list(cmd);
 		spin_unlock_irqrestore(&cmd->t_state_lock, cmd_flags);
 
-		transport_free_dev_tasks(cmd);
 		/*
 		 * The Storage engine stopped this struct se_cmd before it was
 		 * send to the fabric frontend for delivery back to the
diff --git a/include/target/target_core_backend.h b/include/target/target_core_backend.h
index 2f54649dc02..2d7db85e93a 100644
--- a/include/target/target_core_backend.h
+++ b/include/target/target_core_backend.h
@@ -23,12 +23,11 @@ struct se_subsystem_api {
 	struct se_device *(*create_virtdevice)(struct se_hba *,
 				struct se_subsystem_dev *, void *);
 	void (*free_device)(void *);
-	int (*transport_complete)(struct se_task *task);
-	struct se_task *(*alloc_task)(unsigned char *cdb);
-	int (*do_task)(struct se_task *);
+	int (*transport_complete)(struct se_cmd *cmd, struct scatterlist *);
+	int (*execute_cmd)(struct se_cmd *, struct scatterlist *, u32,
+			enum dma_data_direction);
 	int (*do_discard)(struct se_device *, sector_t, u32);
 	void (*do_sync_cache)(struct se_cmd *);
-	void (*free_task)(struct se_task *);
 	ssize_t (*check_configfs_dev_params)(struct se_hba *,
 			struct se_subsystem_dev *);
 	ssize_t (*set_configfs_dev_params)(struct se_hba *,
@@ -38,7 +37,7 @@ struct se_subsystem_api {
 	u32 (*get_device_rev)(struct se_device *);
 	u32 (*get_device_type)(struct se_device *);
 	sector_t (*get_blocks)(struct se_device *);
-	unsigned char *(*get_sense_buffer)(struct se_task *);
+	unsigned char *(*get_sense_buffer)(struct se_cmd *);
 };
 
 int	transport_subsystem_register(struct se_subsystem_api *);
@@ -48,8 +47,6 @@ struct se_device *transport_add_device_to_core_hba(struct se_hba *,
 		struct se_subsystem_api *, struct se_subsystem_dev *, u32,
 		void *, struct se_dev_limits *, const char *, const char *);
 
-void	transport_complete_sync_cache(struct se_cmd *, int);
-void	transport_complete_task(struct se_task *, int);
 void	target_complete_cmd(struct se_cmd *, u8);
 
 void	transport_set_vpd_proto_id(struct t10_vpd *, unsigned char *);
diff --git a/include/target/target_core_base.h b/include/target/target_core_base.h
index 9ea87e68275..55a80bd5070 100644
--- a/include/target/target_core_base.h
+++ b/include/target/target_core_base.h
@@ -477,14 +477,6 @@ struct se_queue_obj {
 	wait_queue_head_t	thread_wq;
 };
 
-struct se_task {
-	struct se_cmd		*task_se_cmd;
-	struct scatterlist	*task_sg;
-	u32			task_sg_nents;
-	u8			task_scsi_status;
-	enum dma_data_direction	task_data_direction;
-};
-
 struct se_tmr_req {
 	/* Task Management function to be performed */
 	u8			function;
@@ -592,7 +584,8 @@ struct se_cmd {
 	/* old task stop completion, consider merging with some of the above */
 	struct completion	task_stop_comp;
 
-	struct se_task		*t_task;
+	/* backend private data */
+	void			*priv;
 };
 
 struct se_ua {