dm: support non power of two target max_io_len

Remove the restriction that limits a target's specified maximum incoming
I/O size to be a power of 2.

Rename this setting from 'split_io' to the less-ambiguous 'max_io_len'.
Change it from sector_t to uint32_t, which is plenty big enough, and
introduce a wrapper function dm_set_target_max_io_len() to set it.
Use sector_div() to process it now that it is not necessarily a power of 2.

Signed-off-by: Mike Snitzer <snitzer@redhat.com>
Signed-off-by: Alasdair G Kergon <agk@redhat.com>

drivers/md/dm-raid.c

@@ -353,6 +353,7 @@ static int parse_raid_params(struct raid_set *rs, char **argv,
 {
 	unsigned i, rebuild_cnt = 0;
 	unsigned long value, region_size = 0;
+	sector_t max_io_len;
 	char *key;
 
 	/*
@@ -522,14 +523,12 @@ static int parse_raid_params(struct raid_set *rs, char **argv,
 		return -EINVAL;
 
 	if (rs->md.chunk_sectors)
-		rs->ti->split_io = rs->md.chunk_sectors;
+		max_io_len = rs->md.chunk_sectors;
 	else
-		rs->ti->split_io = region_size;
+		max_io_len = region_size;
 
-	if (rs->md.chunk_sectors)
+	if (dm_set_target_max_io_len(rs->ti, max_io_len))
-		rs->ti->split_io = rs->md.chunk_sectors;
+		return -EINVAL;
-	else
-		rs->ti->split_io = region_size;
 
 	/* Assume there are no metadata devices until the drives are parsed */
 	rs->md.persistent = 0;

drivers/md/dm-raid1.c

@@ -1081,7 +1081,11 @@ static int mirror_ctr(struct dm_target *ti, unsigned int argc, char **argv)
 	}
 
 	ti->private = ms;
-	ti->split_io = dm_rh_get_region_size(ms->rh);
+
+	r = dm_set_target_max_io_len(ti, dm_rh_get_region_size(ms->rh));
+	if (r)
+		goto err_free_context;
+
 	ti->num_flush_requests = 1;
 	ti->num_discard_requests = 1;
 	ti->discard_zeroes_data_unsupported = 1;

drivers/md/dm-snap.c

@@ -691,7 +691,7 @@ static int dm_add_exception(void *context, chunk_t old, chunk_t new)
  * Return a minimum chunk size of all snapshots that have the specified origin.
  * Return zero if the origin has no snapshots.
  */
-static sector_t __minimum_chunk_size(struct origin *o)
+static uint32_t __minimum_chunk_size(struct origin *o)
 {
 	struct dm_snapshot *snap;
 	unsigned chunk_size = 0;
@@ -701,7 +701,7 @@ static sector_t __minimum_chunk_size(struct origin *o)
 			chunk_size = min_not_zero(chunk_size,
 						  snap->store->chunk_size);
 
-	return chunk_size;
+	return (uint32_t) chunk_size;
 }
 
 /*
@@ -1172,7 +1172,10 @@ static int snapshot_ctr(struct dm_target *ti, unsigned int argc, char **argv)
 		ti->error = "Chunk size not set";
 		goto bad_read_metadata;
 	}
-	ti->split_io = s->store->chunk_size;
+
+	r = dm_set_target_max_io_len(ti, s->store->chunk_size);
+	if (r)
+		goto bad_read_metadata;
 
 	return 0;
 
@@ -1239,7 +1242,7 @@ static void __handover_exceptions(struct dm_snapshot *snap_src,
 	snap_dest->store->snap = snap_dest;
 	snap_src->store->snap = snap_src;
 
-	snap_dest->ti->split_io = snap_dest->store->chunk_size;
+	snap_dest->ti->max_io_len = snap_dest->store->chunk_size;
 	snap_dest->valid = snap_src->valid;
 
 	/*
@@ -1817,9 +1820,9 @@ static void snapshot_resume(struct dm_target *ti)
 	up_write(&s->lock);
 }
 
-static sector_t get_origin_minimum_chunksize(struct block_device *bdev)
+static uint32_t get_origin_minimum_chunksize(struct block_device *bdev)
 {
-	sector_t min_chunksize;
+	uint32_t min_chunksize;
 
 	down_read(&_origins_lock);
 	min_chunksize = __minimum_chunk_size(__lookup_origin(bdev));
@@ -1838,9 +1841,9 @@ static void snapshot_merge_resume(struct dm_target *ti)
 	snapshot_resume(ti);
 
 	/*
-	 * snapshot-merge acts as an origin, so set ti->split_io
+	 * snapshot-merge acts as an origin, so set ti->max_io_len
 	 */
-	ti->split_io = get_origin_minimum_chunksize(s->origin->bdev);
+	ti->max_io_len = get_origin_minimum_chunksize(s->origin->bdev);
 
 	start_merge(s);
 }
@@ -2073,12 +2076,12 @@ static int origin_write_extent(struct dm_snapshot *merging_snap,
 	struct origin *o;
 
 	/*
-	 * The origin's __minimum_chunk_size() got stored in split_io
+	 * The origin's __minimum_chunk_size() got stored in max_io_len
 	 * by snapshot_merge_resume().
 	 */
 	down_read(&_origins_lock);
 	o = __lookup_origin(merging_snap->origin->bdev);
-	for (n = 0; n < size; n += merging_snap->ti->split_io)
+	for (n = 0; n < size; n += merging_snap->ti->max_io_len)
 		if (__origin_write(&o->snapshots, sector + n, NULL) ==
 		    DM_MAPIO_SUBMITTED)
 			must_wait = 1;
@@ -2138,14 +2141,14 @@ static int origin_map(struct dm_target *ti, struct bio *bio,
 }
 
 /*
- * Set the target "split_io" field to the minimum of all the snapshots'
+ * Set the target "max_io_len" field to the minimum of all the snapshots'
  * chunk sizes.
  */
 static void origin_resume(struct dm_target *ti)
 {
 	struct dm_dev *dev = ti->private;
 
-	ti->split_io = get_origin_minimum_chunksize(dev->bdev);
+	ti->max_io_len = get_origin_minimum_chunksize(dev->bdev);
 }
 
 static int origin_status(struct dm_target *ti, status_type_t type, char *result,

drivers/md/dm-stripe.c

@@ -165,7 +165,10 @@ static int stripe_ctr(struct dm_target *ti, unsigned int argc, char **argv)
 	else
 		sc->stripes_shift = __ffs(stripes);
 
-	ti->split_io = chunk_size;
+	r = dm_set_target_max_io_len(ti, chunk_size);
+	if (r)
+		return r;
+
 	ti->num_flush_requests = stripes;
 	ti->num_discard_requests = stripes;
 

drivers/md/dm-thin.c

@@ -2628,7 +2628,10 @@ static int thin_ctr(struct dm_target *ti, unsigned argc, char **argv)
 		goto bad_thin_open;
 	}
 
-	ti->split_io = tc->pool->sectors_per_block;
+	r = dm_set_target_max_io_len(ti, tc->pool->sectors_per_block);
+	if (r)
+		goto bad_thin_open;
+
 	ti->num_flush_requests = 1;
 
 	/* In case the pool supports discards, pass them on. */

drivers/md/dm.c

@@ -968,22 +968,41 @@ static sector_t max_io_len_target_boundary(sector_t sector, struct dm_target *ti
 static sector_t max_io_len(sector_t sector, struct dm_target *ti)
 {
 	sector_t len = max_io_len_target_boundary(sector, ti);
+	sector_t offset, max_len;
 
 	/*
-	 * Does the target need to split even further ?
+	 * Does the target need to split even further?
 	 */
-	if (ti->split_io) {
+	if (ti->max_io_len) {
-		sector_t boundary;
+		offset = dm_target_offset(ti, sector);
-		sector_t offset = dm_target_offset(ti, sector);
+		if (unlikely(ti->max_io_len & (ti->max_io_len - 1)))
-		boundary = ((offset + ti->split_io) & ~(ti->split_io - 1))
+			max_len = sector_div(offset, ti->max_io_len);
-			   - offset;
+		else
-		if (len > boundary)
+			max_len = offset & (ti->max_io_len - 1);
-			len = boundary;
+		max_len = ti->max_io_len - max_len;
+
+		if (len > max_len)
+			len = max_len;
 	}
 
 	return len;
 }
 
+int dm_set_target_max_io_len(struct dm_target *ti, sector_t len)
+{
+	if (len > UINT_MAX) {
+		DMERR("Specified maximum size of target IO (%llu) exceeds limit (%u)",
+		      (unsigned long long)len, UINT_MAX);
+		ti->error = "Maximum size of target IO is too large";
+		return -EINVAL;
+	}
+
+	ti->max_io_len = (uint32_t) len;
+
+	return 0;
+}
+EXPORT_SYMBOL_GPL(dm_set_target_max_io_len);
+
 static void __map_bio(struct dm_target *ti, struct bio *clone,
 		      struct dm_target_io *tio)
 {

include/linux/device-mapper.h

@@ -186,8 +186,8 @@ struct dm_target {
 	sector_t begin;
 	sector_t len;
 
-	/* Always a power of 2 */
+	/* If non-zero, maximum size of I/O submitted to a target. */
-	sector_t split_io;
+	uint32_t max_io_len;
 
 	/*
 	 * A number of zero-length barrier requests that will be submitted
@@ -357,6 +357,11 @@ void dm_table_add_target_callbacks(struct dm_table *t, struct dm_target_callback
  */
 int dm_table_complete(struct dm_table *t);
 
+/*
+ * Target may require that it is never sent I/O larger than len.
+ */
+int __must_check dm_set_target_max_io_len(struct dm_target *ti, sector_t len);
+
 /*
  * Table reference counting.
  */

include/linux/dm-ioctl.h

@@ -268,8 +268,8 @@ enum {
 
 #define DM_VERSION_MAJOR	4
 #define DM_VERSION_MINOR	22
-#define DM_VERSION_PATCHLEVEL	0
+#define DM_VERSION_PATCHLEVEL	1
-#define DM_VERSION_EXTRA	"-ioctl (2011-10-19)"
+#define DM_VERSION_EXTRA	"-ioctl (2012-06-01)"
 
 /* Status bits */
 #define DM_READONLY_FLAG	(1 << 0) /* In/Out */