target: Pass through I/O topology for block backstores

In addition to block size (already implemented), passing through alignment offset, logical-to-phys block exponent, I/O granularity and optimal I/O length will allow initiators to properly handle layout on LUNs with 4K block sizes. Tested with various weird values via scsi_debug module. One thing to look at with this patch is the new block limits values -- instead of granularity 1 optimal 8192, Lio will now be returning whatever the block device says, which may affect performance. Signed-off-by: Andy Grover <agrover@redhat.com> Acked-by: Martin K. Petersen <martin.petersen@oracle.com> Signed-off-by: Nicholas Bellinger <nab@linux-iscsi.org>

target: Pass through I/O topology for block backstores
In addition to block size (already implemented), passing through alignment offset, logical-to-phys block exponent, I/O granularity and optimal I/O length will allow initiators to properly handle layout on LUNs with 4K block sizes. Tested with various weird values via scsi_debug module. One thing to look at with this patch is the new block limits values -- instead of granularity 1 optimal 8192, Lio will now be returning whatever the block device says, which may affect performance. Signed-off-by: Andy Grover <agrover@redhat.com> Acked-by: Martin K. Petersen <martin.petersen@oracle.com> Signed-off-by: Nicholas Bellinger <nab@linux-iscsi.org>
Andy Grover · Nicholas Bellinger
1 parent f01b9f7339
Showing 4 changed files with 68 additions and 3 deletions Side-by-side Diff
drivers/target/target_core_iblock.c
drivers/target/target_core_sbc.c
drivers/target/target_core_spc.c
include/target/target_core_backend.h
@@ -710,6 +710,45 @@
 	return iblock_emulate_read_cap_with_block_size(dev, bd, q);
 }
  
+static sector_t iblock_get_alignment_offset_lbas(struct se_device *dev)
+{
+	struct iblock_dev *ib_dev = IBLOCK_DEV(dev);
+	struct block_device *bd = ib_dev->ibd_bd;
+	int ret;
+
+	ret = bdev_alignment_offset(bd);
+	if (ret == -1)
+		return 0;
+
+	/* convert offset-bytes to offset-lbas */
+	return ret / bdev_logical_block_size(bd);
+}
+
+static unsigned int iblock_get_lbppbe(struct se_device *dev)
+{
+	struct iblock_dev *ib_dev = IBLOCK_DEV(dev);
+	struct block_device *bd = ib_dev->ibd_bd;
+	int logs_per_phys = bdev_physical_block_size(bd) / bdev_logical_block_size(bd);
+
+	return ilog2(logs_per_phys);
+}
+
+static unsigned int iblock_get_io_min(struct se_device *dev)
+{
+	struct iblock_dev *ib_dev = IBLOCK_DEV(dev);
+	struct block_device *bd = ib_dev->ibd_bd;
+
+	return bdev_io_min(bd);
+}
+
+static unsigned int iblock_get_io_opt(struct se_device *dev)
+{
+	struct iblock_dev *ib_dev = IBLOCK_DEV(dev);
+	struct block_device *bd = ib_dev->ibd_bd;
+
+	return bdev_io_opt(bd);
+}
+
 static struct sbc_ops iblock_sbc_ops = {
 	.execute_rw		= iblock_execute_rw,
 	.execute_sync_cache	= iblock_execute_sync_cache,
@@ -749,6 +788,10 @@
 	.show_configfs_dev_params = iblock_show_configfs_dev_params,
 	.get_device_type	= sbc_get_device_type,
 	.get_blocks		= iblock_get_blocks,
+	.get_alignment_offset_lbas = iblock_get_alignment_offset_lbas,
+	.get_lbppbe		= iblock_get_lbppbe,
+	.get_io_min		= iblock_get_io_min,
+	.get_io_opt		= iblock_get_io_opt,
 	.get_write_cache	= iblock_get_write_cache,
 };
  
@@ -105,12 +105,22 @@
 	buf[9] = (dev->dev_attrib.block_size >> 16) & 0xff;
 	buf[10] = (dev->dev_attrib.block_size >> 8) & 0xff;
 	buf[11] = dev->dev_attrib.block_size & 0xff;
+
+	if (dev->transport->get_lbppbe)
+		buf[13] = dev->transport->get_lbppbe(dev) & 0x0f;
+
+	if (dev->transport->get_alignment_offset_lbas) {
+		u16 lalba = dev->transport->get_alignment_offset_lbas(dev);
+		buf[14] = (lalba >> 8) & 0x3f;
+		buf[15] = lalba & 0xff;
+	}
+
 	/*
 	 * Set Thin Provisioning Enable bit following sbc3r22 in section
 	 * READ CAPACITY (16) byte 14 if emulate_tpu or emulate_tpws is enabled.
 	 */
 	if (dev->dev_attrib.emulate_tpu || dev->dev_attrib.emulate_tpws)
-		buf[14] = 0x80;
+		buf[14] |= 0x80;
  
 	rbuf = transport_kmap_data_sg(cmd);
 	if (rbuf) {
@@ -452,6 +452,7 @@
 	struct se_device *dev = cmd->se_dev;
 	u32 max_sectors;
 	int have_tp = 0;
+	int opt, min;
  
 	/*
 	 * Following spc3r22 section 6.5.3 Block Limits VPD page, when
@@ -475,7 +476,10 @@
 	/*
 	 * Set OPTIMAL TRANSFER LENGTH GRANULARITY
 	 */
-	put_unaligned_be16(1, &buf[6]);
+	if (dev->transport->get_io_min && (min = dev->transport->get_io_min(dev)))
+		put_unaligned_be16(min / dev->dev_attrib.block_size, &buf[6]);
+	else
+		put_unaligned_be16(1, &buf[6]);
  
 	/*
 	 * Set MAXIMUM TRANSFER LENGTH
@@ -487,7 +491,10 @@
 	/*
 	 * Set OPTIMAL TRANSFER LENGTH
 	 */
-	put_unaligned_be32(dev->dev_attrib.optimal_sectors, &buf[12]);
+	if (dev->transport->get_io_opt && (opt = dev->transport->get_io_opt(dev)))
+		put_unaligned_be32(opt / dev->dev_attrib.block_size, &buf[12]);
+	else
+		put_unaligned_be32(dev->dev_attrib.optimal_sectors, &buf[12]);
  
 	/*
 	 * Exit now if we don't support TP.
@@ -34,6 +34,11 @@
 	sense_reason_t (*parse_cdb)(struct se_cmd *cmd);
 	u32 (*get_device_type)(struct se_device *);
 	sector_t (*get_blocks)(struct se_device *);
+	sector_t (*get_alignment_offset_lbas)(struct se_device *);
+	/* lbppbe = logical blocks per physical block exponent. see SBC-3 */
+	unsigned int (*get_lbppbe)(struct se_device *);
+	unsigned int (*get_io_min)(struct se_device *);
+	unsigned int (*get_io_opt)(struct se_device *);
 	unsigned char *(*get_sense_buffer)(struct se_cmd *);
 	bool (*get_write_cache)(struct se_device *);
 };
...	...	@@ -710,6 +710,45 @@
710	710	return iblock_emulate_read_cap_with_block_size(dev, bd, q);
711	711	}
712	712
	713	+static sector_t iblock_get_alignment_offset_lbas(struct se_device *dev)
	714	+{
	715	+ struct iblock_dev *ib_dev = IBLOCK_DEV(dev);
	716	+ struct block_device *bd = ib_dev->ibd_bd;
	717	+ int ret;
	718	+
	719	+ ret = bdev_alignment_offset(bd);
	720	+ if (ret == -1)
	721	+ return 0;
	722	+
	723	+ /* convert offset-bytes to offset-lbas */
	724	+ return ret / bdev_logical_block_size(bd);
	725	+}
	726	+
	727	+static unsigned int iblock_get_lbppbe(struct se_device *dev)
	728	+{
	729	+ struct iblock_dev *ib_dev = IBLOCK_DEV(dev);
	730	+ struct block_device *bd = ib_dev->ibd_bd;
	731	+ int logs_per_phys = bdev_physical_block_size(bd) / bdev_logical_block_size(bd);
	732	+
	733	+ return ilog2(logs_per_phys);
	734	+}
	735	+
	736	+static unsigned int iblock_get_io_min(struct se_device *dev)
	737	+{
	738	+ struct iblock_dev *ib_dev = IBLOCK_DEV(dev);
	739	+ struct block_device *bd = ib_dev->ibd_bd;
	740	+
	741	+ return bdev_io_min(bd);
	742	+}
	743	+
	744	+static unsigned int iblock_get_io_opt(struct se_device *dev)
	745	+{
	746	+ struct iblock_dev *ib_dev = IBLOCK_DEV(dev);
	747	+ struct block_device *bd = ib_dev->ibd_bd;
	748	+
	749	+ return bdev_io_opt(bd);
	750	+}
	751	+
713	752	static struct sbc_ops iblock_sbc_ops = {
714	753	.execute_rw = iblock_execute_rw,
715	754	.execute_sync_cache = iblock_execute_sync_cache,
...	...	@@ -749,6 +788,10 @@
749	788	.show_configfs_dev_params = iblock_show_configfs_dev_params,
750	789	.get_device_type = sbc_get_device_type,
751	790	.get_blocks = iblock_get_blocks,
	791	+ .get_alignment_offset_lbas = iblock_get_alignment_offset_lbas,
	792	+ .get_lbppbe = iblock_get_lbppbe,
	793	+ .get_io_min = iblock_get_io_min,
	794	+ .get_io_opt = iblock_get_io_opt,
752	795	.get_write_cache = iblock_get_write_cache,
753	796	};
754	797
...	...	@@ -105,12 +105,22 @@
105	105	buf[9] = (dev->dev_attrib.block_size >> 16) & 0xff;
106	106	buf[10] = (dev->dev_attrib.block_size >> 8) & 0xff;
107	107	buf[11] = dev->dev_attrib.block_size & 0xff;
	108	+
	109	+ if (dev->transport->get_lbppbe)
	110	+ buf[13] = dev->transport->get_lbppbe(dev) & 0x0f;
	111	+
	112	+ if (dev->transport->get_alignment_offset_lbas) {
	113	+ u16 lalba = dev->transport->get_alignment_offset_lbas(dev);
	114	+ buf[14] = (lalba >> 8) & 0x3f;
	115	+ buf[15] = lalba & 0xff;
	116	+ }
	117	+
108	118	/*
109	119	* Set Thin Provisioning Enable bit following sbc3r22 in section
110	120	* READ CAPACITY (16) byte 14 if emulate_tpu or emulate_tpws is enabled.
111	121	*/
112	122	if (dev->dev_attrib.emulate_tpu \|\| dev->dev_attrib.emulate_tpws)
113		- buf[14] = 0x80;
	123	+ buf[14] \|= 0x80;
114	124
115	125	rbuf = transport_kmap_data_sg(cmd);
116	126	if (rbuf) {
...	...	@@ -452,6 +452,7 @@
452	452	struct se_device *dev = cmd->se_dev;
453	453	u32 max_sectors;
454	454	int have_tp = 0;
	455	+ int opt, min;
455	456
456	457	/*
457	458	* Following spc3r22 section 6.5.3 Block Limits VPD page, when
...	...	@@ -475,7 +476,10 @@
475	476	/*
476	477	* Set OPTIMAL TRANSFER LENGTH GRANULARITY
477	478	*/
478		- put_unaligned_be16(1, &buf[6]);
	479	+ if (dev->transport->get_io_min && (min = dev->transport->get_io_min(dev)))
	480	+ put_unaligned_be16(min / dev->dev_attrib.block_size, &buf[6]);
	481	+ else
	482	+ put_unaligned_be16(1, &buf[6]);
479	483
480	484	/*
481	485	* Set MAXIMUM TRANSFER LENGTH
...	...	@@ -487,7 +491,10 @@
487	491	/*
488	492	* Set OPTIMAL TRANSFER LENGTH
489	493	*/
490		- put_unaligned_be32(dev->dev_attrib.optimal_sectors, &buf[12]);
	494	+ if (dev->transport->get_io_opt && (opt = dev->transport->get_io_opt(dev)))
	495	+ put_unaligned_be32(opt / dev->dev_attrib.block_size, &buf[12]);
	496	+ else
	497	+ put_unaligned_be32(dev->dev_attrib.optimal_sectors, &buf[12]);
491	498
492	499	/*
493	500	* Exit now if we don't support TP.
...	...	@@ -34,6 +34,11 @@
34	34	sense_reason_t (parse_cdb)(struct se_cmd cmd);
35	35	u32 (get_device_type)(struct se_device );
36	36	sector_t (get_blocks)(struct se_device );
	37	+ sector_t (get_alignment_offset_lbas)(struct se_device );
	38	+ /* lbppbe = logical blocks per physical block exponent. see SBC-3 */
	39	+ unsigned int (get_lbppbe)(struct se_device );
	40	+ unsigned int (get_io_min)(struct se_device );
	41	+ unsigned int (get_io_opt)(struct se_device );
37	42	unsigned char (get_sense_buffer)(struct se_cmd *);
38	43	bool (get_write_cache)(struct se_device );
39	44	};