scsi: target: Allow userspace to request direct submissions

This allows userspace to request the fabric drivers do direct submissions
if they support it. With the new device file, submit_type, users can
write 0 - 2 to control how commands are submitted to the backend:

 0 - TARGET_FABRIC_DEFAULT_SUBMIT - LIO will use the fabric's default
     submission type. This is the default for compat.

 1 - TARGET_DIRECT_SUBMIT - LIO will submit the cmd to the backend from the
     calling context if the fabric the cmd was received on supports it,
     else it will use the fabric's default type.

 2 - TARGET_QUEUE_SUBMIT - LIO will queue the cmd to the LIO submission
     workqueue which will pass it to the backend.

When using an NVMe drive and vhost-scsi with direct submission we see
around a 20% improvement in 4K I/Os:

fio jobs        1       2       4       8       10
--------------------------------------------------
defer           94K     190K    394K    770K    890K
direct          128K    252K    488K    950K    -

And when using the queueing mode, we now no longer see issues like where
the iSCSI tx thread is blocked in the block layer waiting on a tag so it
can't respond to a nop or perform I/Os for other LUs.

Signed-off-by: Mike Christie <[email protected]>
Link: https://lore.kernel.org/r/[email protected]
Signed-off-by: Martin K. Petersen <[email protected]>

Author: mikechristie

drivers/target/loopback/tcm_loop.c

@@ -154,7 +154,7 @@ static void tcm_loop_target_queue_cmd(struct tcm_loop_cmd *tl_cmd)
 			       GFP_ATOMIC))
 		return;
 
-	target_queue_submission(se_cmd);
+	target_submit(se_cmd);
 	return;
 
 out_done:

drivers/target/target_core_configfs.c

@@ -577,6 +577,7 @@ DEF_CONFIGFS_ATTRIB_SHOW(unmap_granularity_alignment);
 DEF_CONFIGFS_ATTRIB_SHOW(unmap_zeroes_data);
 DEF_CONFIGFS_ATTRIB_SHOW(max_write_same_len);
 DEF_CONFIGFS_ATTRIB_SHOW(emulate_rsoc);
+DEF_CONFIGFS_ATTRIB_SHOW(submit_type);
 
 #define DEF_CONFIGFS_ATTRIB_STORE_U32(_name)				\
 static ssize_t _name##_store(struct config_item *item, const char *page,\
@@ -1231,6 +1232,24 @@ static ssize_t emulate_rsoc_store(struct config_item *item,
 	return count;
 }
 
+static ssize_t submit_type_store(struct config_item *item, const char *page,
+				 size_t count)
+{
+	struct se_dev_attrib *da = to_attrib(item);
+	int ret;
+	u8 val;
+
+	ret = kstrtou8(page, 0, &val);
+	if (ret < 0)
+		return ret;
+
+	if (val > TARGET_QUEUE_SUBMIT)
+		return -EINVAL;
+
+	da->submit_type = val;
+	return count;
+}
+
 CONFIGFS_ATTR(, emulate_model_alias);
 CONFIGFS_ATTR(, emulate_dpo);
 CONFIGFS_ATTR(, emulate_fua_write);
@@ -1266,6 +1285,7 @@ CONFIGFS_ATTR(, unmap_zeroes_data);
 CONFIGFS_ATTR(, max_write_same_len);
 CONFIGFS_ATTR(, alua_support);
 CONFIGFS_ATTR(, pgr_support);
+CONFIGFS_ATTR(, submit_type);
 
 /*
  * dev_attrib attributes for devices using the target core SBC/SPC
@@ -1308,6 +1328,7 @@ struct configfs_attribute *sbc_attrib_attrs[] = {
 	&attr_alua_support,
 	&attr_pgr_support,
 	&attr_emulate_rsoc,
+	&attr_submit_type,
 	NULL,
 };
 EXPORT_SYMBOL(sbc_attrib_attrs);
@@ -1325,6 +1346,7 @@ struct configfs_attribute *passthrough_attrib_attrs[] = {
 	&attr_emulate_pr,
 	&attr_alua_support,
 	&attr_pgr_support,
+	&attr_submit_type,
 	NULL,
 };
 EXPORT_SYMBOL(passthrough_attrib_attrs);

drivers/target/target_core_device.c

@@ -779,6 +779,7 @@ struct se_device *target_alloc_device(struct se_hba *hba, const char *name)
 	dev->dev_attrib.unmap_zeroes_data =
 				DA_UNMAP_ZEROES_DATA_DEFAULT;
 	dev->dev_attrib.max_write_same_len = DA_MAX_WRITE_SAME_LEN;
+	dev->dev_attrib.submit_type = TARGET_FABRIC_DEFAULT_SUBMIT;
 
 	xcopy_lun = &dev->xcopy_lun;
 	rcu_assign_pointer(xcopy_lun->lun_se_dev, dev);

drivers/target/target_core_transport.c

@@ -1575,14 +1575,7 @@ target_cmd_parse_cdb(struct se_cmd *cmd)
 }
 EXPORT_SYMBOL(target_cmd_parse_cdb);
 
-/**
- * target_submit - perform final initialization and submit cmd to LIO core
- * @cmd: command descriptor to submit
- *
- * target_submit_prep or something similar must have been called on the cmd,
- * and this must be called from process context.
- */
-int target_submit(struct se_cmd *cmd)
+static int __target_submit(struct se_cmd *cmd)
 {
 	sense_reason_t ret;
 
@@ -1642,7 +1635,6 @@ int target_submit(struct se_cmd *cmd)
 		transport_generic_request_failure(cmd, ret);
 	return 0;
 }
-EXPORT_SYMBOL_GPL(target_submit);
 
 sense_reason_t
 transport_generic_map_mem_to_cmd(struct se_cmd *cmd, struct scatterlist *sgl,
@@ -1904,7 +1896,7 @@ void target_queued_submit_work(struct work_struct *work)
 			se_plug = target_plug_device(se_dev);
 		}
 
-		target_submit(se_cmd);
+		__target_submit(se_cmd);
 	}
 
 	if (se_plug)
@@ -1927,6 +1919,35 @@ void target_queue_submission(struct se_cmd *se_cmd)
 }
 EXPORT_SYMBOL_GPL(target_queue_submission);
 
+/**
+ * target_submit - perform final initialization and submit cmd to LIO core
+ * @cmd: command descriptor to submit
+ *
+ * target_submit_prep or something similar must have been called on the cmd,
+ * and this must be called from process context.
+ */
+int target_submit(struct se_cmd *se_cmd)
+{
+	const struct target_core_fabric_ops *tfo = se_cmd->se_sess->se_tpg->se_tpg_tfo;
+	struct se_dev_attrib *da = &se_cmd->se_dev->dev_attrib;
+	u8 submit_type;
+
+	if (da->submit_type == TARGET_FABRIC_DEFAULT_SUBMIT)
+		submit_type = tfo->default_submit_type;
+	else if (da->submit_type == TARGET_DIRECT_SUBMIT &&
+		 tfo->direct_submit_supp)
+		submit_type = TARGET_DIRECT_SUBMIT;
+	else
+		submit_type = TARGET_QUEUE_SUBMIT;
+
+	if (submit_type == TARGET_DIRECT_SUBMIT)
+		return __target_submit(se_cmd);
+
+	target_queue_submission(se_cmd);
+	return 0;
+}
+EXPORT_SYMBOL_GPL(target_submit);
+
 static void target_complete_tmr_failure(struct work_struct *work)
 {
 	struct se_cmd *se_cmd = container_of(work, struct se_cmd, work);

drivers/vhost/scsi.c

@@ -909,7 +909,7 @@ static void vhost_scsi_target_queue_cmd(struct vhost_scsi_cmd *cmd)
 			       cmd->tvc_prot_sgl_count, GFP_KERNEL))
 		return;
 
-	target_queue_submission(se_cmd);
+	target_submit(se_cmd);
 }
 
 static void

include/target/target_core_base.h

@@ -726,6 +726,7 @@ struct se_dev_attrib {
 	u32		unmap_granularity;
 	u32		unmap_granularity_alignment;
 	u32		max_write_same_len;
+	u8		submit_type;
 	struct se_device *da_dev;
 	struct config_group da_group;
 };