The maximum size of a bio is restricted by DM to BLK_DEF_MAX_SECTORS
although the I/O size can be increased for each member device above this
value by using the /sys/block/queue/max_sectors interface.
IMHO the limitation when loading the dm table is unnecessary, the values
for sub-devices are checked while set through the sysfs interface and
the restrictions of the sub-devices are combined, the minimal values are
used for the DM queue. Chopping this additionally to BLK_DEF_MAX_SECTORS
makes it impossible to use IOs bigger than 512KB on big RAIDs accessed
through multipath, even if the devices may support them. The trivial patch
removes this limitation. Sub-devices' max_sectors need to be set before
creating the DM device.
Patch is against 2.6.24. For older kernels max_hw_sectors needs to be set
additionally.
Signed-off-by: Erich Focht <[email protected]>
---
dm-table.c | 4 ++--
1 file changed, 2 insertions(+), 2 deletions(-)
--- linux-2.6.24/drivers/md/dm-table.c.orig 2008-01-31 17:06:30.000000000 +0100
+++ linux-2.6.24/drivers/md/dm-table.c 2008-01-31 17:34:03.000000000 +0100
@@ -901,10 +901,10 @@
void dm_table_set_restrictions(struct dm_table *t, struct request_queue *q)
{
/*
- * Make sure we obey the optimistic sub devices
+ * Make sure we obey the minimum of the sub devices
* restrictions.
*/
- blk_queue_max_sectors(q, t->limits.max_sectors);
+ q->max_sectors = t->limits.max_sectors;
q->max_phys_segments = t->limits.max_phys_segments;
q->max_hw_segments = t->limits.max_hw_segments;
q->hardsect_size = t->limits.hardsect_size;