[PATCH 2/5] Add devices/data_alignment_detection to lvm.conf.

Mike Snitzer <[email protected]>
Newsgroups dev.linux.lists.lvm-devel
Message-ID <[email protected]> 
Adds 'data_alignment_detection' config option to the devices section of
lvm.conf.  If your kernel provides topology information in sysfs (linux
>= 2.6.31) for the Physical Volume, the start of data area will be
aligned on a multiple of the ?minimum_io_size? or ?optimal_io_size?
exposed in sysfs.

minimum_io_size is used if optimal_io_size is undefined (0).  If both
md_chunk_alignment and data_alignment_detection are enabled the result
of data_alignment_detection is used.

Signed-off-by: Mike Snitzer <[email protected]>
---
 WHATS_NEW               |    1 +
 doc/example.conf        |   12 ++++++-
 lib/config/defaults.h   |    1 +
 lib/device/device.c     |   75 +++++++++++++++++++++++++++++++++++++++++++---
 lib/device/device.h     |    6 ++++
 lib/metadata/metadata.c |   19 ++++++++++++
 man/lvm.conf.5.in       |   13 +++++++-
 7 files changed, 118 insertions(+), 9 deletions(-)

diff --git a/WHATS_NEW b/WHATS_NEW
index 8709a3a..55d2df5 100644
--- a/WHATS_NEW
+++ b/WHATS_NEW
@@ -1,5 +1,6 @@
 Version 2.02.51 - 
 ================================
+  Add devices/data_alignment_detection to lvm.conf.
   Add devices/data_alignment_offset_detection to lvm.conf.
   Add --dataalignmentoffset to pvcreate to shift start of aligned data area.
   Fix _mda_setup() to not check first mda's size before pe_align rounding.
diff --git a/doc/example.conf b/doc/example.conf
index f8528fd..7512b24 100644
--- a/doc/example.conf
+++ b/doc/example.conf
@@ -98,9 +98,17 @@ devices {
     # 1 enables; 0 disables.
     md_chunk_alignment = 1
 
+    # By default, the start of a PV's data area will be aligned with
+    # the 'minimum_io_size' or 'optimal_io_size' exposed in sysfs.
+    # minimum_io_size is used if optimal_io_size is undefined (0).
+    # If md_chunk_alignment is enabled, that detects the optimal_io_size.
+    # This setting takes precedence over md_chunk_alignment.
+    # 1 enables; 0 disables.
+    data_alignment_detection = 1
+
     # Alignment (in KB) of start of data area when creating a new PV.
-    # If a PV is placed directly upon an md device and md_chunk_alignment is
-    # enabled this parameter is ignored.
+    # If a PV is placed directly upon an md device and md_chunk_alignment or
+    # data_alignment_detection is enabled this parameter is ignored.
     # Set to 0 for the default alignment of 64KB or page size, if larger.
     data_alignment = 0
 
diff --git a/lib/config/defaults.h b/lib/config/defaults.h
index 227ec34..95d8769 100644
--- a/lib/config/defaults.h
+++ b/lib/config/defaults.h
@@ -35,6 +35,7 @@
 #define DEFAULT_MD_CHUNK_ALIGNMENT 1
 #define DEFAULT_IGNORE_SUSPENDED_DEVICES 1
 #define DEFAULT_DATA_ALIGNMENT_OFFSET_DETECTION 1
+#define DEFAULT_DATA_ALIGNMENT_DETECTION 1
 
 #define DEFAULT_LOCK_DIR "/var/lock/lvm"
 #define DEFAULT_LOCKING_LIB "liblvm2clusterlock.so"
diff --git a/lib/device/device.c b/lib/device/device.c
index ebd8141..e627517 100644
--- a/lib/device/device.c
+++ b/lib/device/device.c
@@ -285,13 +285,36 @@ int _get_partition_type(struct dev_mgr *dm, struct device *d)
 
 #ifdef linux
 
+static int _primary_dev(const char *sysfs_dir,
+			struct device *dev, dev_t *result)
+{
+	char path[PATH_MAX+1];
+	struct stat info;
+
+	/* check if dev is a partition */
+	if (dm_snprintf(path, PATH_MAX, "%s/dev/block/%d:%d/partition",
+			sysfs_dir, (int)MAJOR(dev->dev), (int)MINOR(dev->dev)) < 0) {
+		log_error("dm_snprintf partition failed");
+		return 0;
+	}
+
+	if (stat(path, &info) < 0)
+		return 0;
+
+	*result = dev->dev -
+		(MINOR(dev->dev) % max_partitions(MAJOR(dev->dev)));
+	return 1;
+}
+
 static unsigned long _dev_topology_attribute(const char *attribute,
 					     const char *sysfs_dir,
 					     struct device *dev)
 {
+	const char *sysfs_fmt_str = "%s/dev/block/%d:%d/%s";
 	char path[PATH_MAX+1], buffer[64];
 	FILE *fp;
 	struct stat info;
+	dev_t uninitialized_var(primary);
 	unsigned long result = 0UL;
 
 	if (!attribute || !*attribute)
@@ -300,16 +323,32 @@ static unsigned long _dev_topology_attribute(const char *attribute,
 	if (!sysfs_dir || !*sysfs_dir)
 		return_0;
 
-	if (dm_snprintf(path, PATH_MAX, "%s/dev/block/%d:%d/%s",
-			sysfs_dir, (int)MAJOR(dev->dev), (int)MINOR(dev->dev),
+	if (dm_snprintf(path, PATH_MAX, sysfs_fmt_str, sysfs_dir,
+			(int)MAJOR(dev->dev), (int)MINOR(dev->dev),
 			attribute) < 0) {
 		log_error("dm_snprintf %s failed", attribute);
 		return 0;
 	}
 
-	/* check if the desired sysfs attribute exists */
-	if (stat(path, &info) < 0)
-		return 0;
+	/*
+	 * check if the desired sysfs attribute exists
+	 * - if not: either the kernel doesn't have topology support
+	 *   or the device could be a partition
+	 */
+	if (stat(path, &info) < 0) {
+		if (!_primary_dev(sysfs_dir, dev, &primary))
+			return 0;
+
+		/* get attribute from partition's primary device */
+		if (dm_snprintf(path, PATH_MAX, sysfs_fmt_str, sysfs_dir,
+				(int)MAJOR(primary), (int)MINOR(primary),
+				attribute) < 0) {
+			log_error("pri dm_snprintf %s failed", attribute);
+			return 0;
+		}
+		if (stat(path, &info) < 0)
+			return 0;
+	}
 
 	if (!(fp = fopen(path, "r"))) {
 		log_sys_error("fopen", path);
@@ -344,6 +383,20 @@ unsigned long dev_alignment_offset(const char *sysfs_dir,
 				       sysfs_dir, dev);
 }
 
+unsigned long dev_minimum_io_size(const char *sysfs_dir,
+				  struct device *dev)
+{
+	return _dev_topology_attribute("queue/minimum_io_size",
+				       sysfs_dir, dev);
+}
+
+unsigned long dev_optimal_io_size(const char *sysfs_dir,
+				  struct device *dev)
+{
+	return _dev_topology_attribute("queue/optimal_io_size",
+				       sysfs_dir, dev);
+}
+
 #else
 
 unsigned long dev_alignment_offset(const char *sysfs_dir,
@@ -352,4 +405,16 @@ unsigned long dev_alignment_offset(const char *sysfs_dir,
 	return 0UL;
 }
 
+unsigned long dev_minimum_io_size(const char *sysfs_dir,
+				  struct device *dev)
+{
+	return 0UL;
+}
+
+unsigned long dev_optimal_io_size(const char *sysfs_dir,
+				  struct device *dev)
+{
+	return 0UL;
+}
+
 #endif
diff --git a/lib/device/device.h b/lib/device/device.h
index 32aee41..454fd0b 100644
--- a/lib/device/device.h
+++ b/lib/device/device.h
@@ -103,4 +103,10 @@ int is_partitioned_dev(struct device *dev);
 unsigned long dev_alignment_offset(const char *sysfs_dir,
 				   struct device *dev);
 
+unsigned long dev_minimum_io_size(const char *sysfs_dir,
+				  struct device *dev);
+
+unsigned long dev_optimal_io_size(const char *sysfs_dir,
+				  struct device *dev);
+
 #endif
diff --git a/lib/metadata/metadata.c b/lib/metadata/metadata.c
index d050b6c..a3d64a1 100644
--- a/lib/metadata/metadata.c
+++ b/lib/metadata/metadata.c
@@ -86,6 +86,25 @@ unsigned long set_pe_align(struct physical_volume *pv, unsigned long data_alignm
 				   dev_md_stripe_width(pv->fmt->cmd->sysfs_dir,
 						       pv->dev));
 
+	/*
+	 * Align to topology's minimum_io_size or optimal_io_size if present
+	 * - minimum_io_size - the smallest request the device can perform
+	 *   w/o incurring a read-modify-write penalty (e.g. MD's chunk size)
+	 * - optimal_io_size - the device's preferred unit of receiving I/O
+	 *   (e.g. MD's stripe width)
+	 */
+	if (find_config_tree_bool(pv->fmt->cmd,
+				  "devices/data_alignment_detection",
+				  DEFAULT_DATA_ALIGNMENT_DETECTION)) {
+		pv->pe_align = MAX(pv->pe_align,
+				   dev_minimum_io_size(pv->fmt->cmd->sysfs_dir,
+						       pv->dev));
+
+		pv->pe_align = MAX(pv->pe_align,
+				   dev_optimal_io_size(pv->fmt->cmd->sysfs_dir,
+						       pv->dev));
+	}
+
 	log_very_verbose("%s: Setting PE alignment to %lu sectors.",
 			 dev_name(pv->dev), pv->pe_align);
 
diff --git a/man/lvm.conf.5.in b/man/lvm.conf.5.in
index 7b3ee22..82f51ec 100644
--- a/man/lvm.conf.5.in
+++ b/man/lvm.conf.5.in
@@ -137,11 +137,20 @@ has been reused without wiping the md superblocks first.
 directly upon an md device, LVM2 will align its data blocks with the
 md device's stripe-width.
 .IP
+\fBdata_alignment_detection\fP \(em If set to 1, and your kernel provides
+topology information in sysfs for the Physical Volume, the start of data
+area will be aligned on a multiple of the ?minimum_io_size? or
+?optimal_io_size? exposed in sysfs.  minimum_io_size is used if
+optimal_io_size is undefined (0).  If both \fBmd_chunk_alignment\fP and
+\fBdata_alignment_detection\fP are enabled the result of
+\fBdata_alignment_detection\fP is used.
+.IP
 \fBdata_alignment\fP \(em Default alignment (in KB) of start of data area
 when creating a new Physical Volume using the \fBlvm2\fP format.
 If a Physical Volume is placed directly upon an md device and
-\fBmd_chunk_alignment\fP is enabled this parameter is ignored.
-Set to 0 to use the default alignment of 64KB or the page size, if larger.
+\fBmd_chunk_alignment\fP or \fBdata_alignment_detection\fP is enabled
+this parameter is ignored.  Set to 0 to use the default alignment of
+64KB or the page size, if larger.
 .IP
 \fBdata_alignment_offset_detection\fP \(em If set to 1, and your kernel
 provides topology information in sysfs for the Physical Volume, the
-- 
1.6.2.5
lmpx.com only provides a reader for public news (NNTP) servers. It is not affiliated with the servers or forums shown here and is not responsible for the content of articles, which is written by their respective authors.