On Wed, 30 Oct 2013 13:20:22 +0100 Pawel Baldysiak
[off-list ref] wrote:

quoted

In case of reshape of raid0 through raid4 a value of
discard_granularity will be set to stripe size. MD driver should
re-set this value to correct one when migration will be finished.
Otherwise array will be left with wrong value and discard operations will not

work properly.

quoted

Signed-off-by: Pawel Baldysiak <redacted>
Cc: Shaohua Li <shli@kernel.org>
---
 drivers/md/raid0.c |    2 ++
 1 file changed, 2 insertions(+)

diff --git a/drivers/md/raid0.c b/drivers/md/raid0.c index

c4d420b..807ca3a 100644

--- a/drivers/md/raid0.c
+++ b/drivers/md/raid0.c

@@ -266,6 +266,8 @@ static int create_strip_zones(struct mddev *mddev,

struct r0conf **private_conf)

quoted

 	}
 	mddev->queue->backing_dev_info.congested_fn = raid0_congested;
 	mddev->queue->backing_dev_info.congested_data = mddev;
+	mddev->queue->limits.discard_granularity =
+			queue_logical_block_size(mddev->queue);

 	/*
 	 * now since we have the hard sector sizes, we can make sure

Thanks, but this doesn't seem like the right sort of fix.  It is to specific to the
symptom rather than trying to address the underlying problem.

Maybe something like this?  Can you review and test?

Thanks,
NeilBrown

diff --git a/drivers/md/md.c b/drivers/md/md.c index

628cd529343f..740b6340f980 100644

--- a/drivers/md/md.c
+++ b/drivers/md/md.c

@@ -3620,6 +3620,7 @@ level_store(struct mddev *mddev, const char

*buf, size_t len)
 		mddev->in_sync = 1;
 		del_timer_sync(&mddev->safemode_timer);
 	}
+	blk_set_stacking_limit(&mddev->queue->limits);
 	pers->run(mddev);
 	set_bit(MD_CHANGE_DEVS, &mddev->flags);
 	mddev_resume(mddev);