On 4/7/20 4:50 AM, Damien Le Moal wrote:
On 2020/04/07 3:24, Hannes Reinecke wrote:
Instead of having each function to reference the device for
themselves add it as an argument to the function.
And replace the 'target' pointer in the bio context with the
device pointer.
Signed-off-by: Hannes Reinecke <hare@xxxxxxx>
---
drivers/md/dm-zoned-target.c | 88 +++++++++++++++++++++---------------
1 file changed, 52 insertions(+), 36 deletions(-)
diff --git a/drivers/md/dm-zoned-target.c b/drivers/md/dm-zoned-target.c
index 02ee0ca1f0b0..8ed6d9f2df25 100644
--- a/drivers/md/dm-zoned-target.c
+++ b/drivers/md/dm-zoned-target.c
@@ -17,7 +17,7 @@
* Zone BIO context.
*/
struct dmz_bioctx {
- struct dmz_target *target;
+ struct dmz_dev *dev;
struct dm_zone *zone;
struct bio *bio;
refcount_t ref;
@@ -71,6 +71,11 @@ struct dmz_target {
*/
#define DMZ_FLUSH_PERIOD (10 * HZ)
+struct dmz_dev *dmz_sect_to_dev(struct dmz_target *dmz, sector_t sect)
+{
+ return &dmz->dev[0];
+}
I do not get it. Regardless of the chunk sector specified, always returning the
first device seems wrong. If the sector belongs to a chunk mapped to a zone on
the second device, things will break, no ?
This is just a stub for now, so that the actual patch introducing
several devices can fold into here and the code churn for the final
patch is reduced.
+
/*
* Target BIO completion.
*/
@@ -81,7 +86,7 @@ static inline void dmz_bio_endio(struct bio *bio, blk_status_t status)
if (status != BLK_STS_OK && bio->bi_status == BLK_STS_OK)
bio->bi_status = status;
if (bio->bi_status != BLK_STS_OK)
- bioctx->target->dev->flags |= DMZ_CHECK_BDEV;
+ bioctx->dev->flags |= DMZ_CHECK_BDEV;
if (refcount_dec_and_test(&bioctx->ref)) {
struct dm_zone *zone = bioctx->zone;
@@ -114,18 +119,20 @@ static void dmz_clone_endio(struct bio *clone)
* Issue a clone of a target BIO. The clone may only partially process the
* original target BIO.
*/
-static int dmz_submit_bio(struct dmz_target *dmz, struct dm_zone *zone,
- struct bio *bio, sector_t chunk_block,
- unsigned int nr_blocks)
+static int dmz_submit_bio(struct dmz_target *dmz, struct dmz_dev *dev,
+ struct dm_zone *zone, struct bio *bio,
+ sector_t chunk_block, unsigned int nr_blocks)
dev could be inferred from the zone with dmz_zone_to_dev(), no ?
Yes, it could, but then I'll find myself calling dmz_zone_to_dev() on
every function in the call chain.
So I thought to pass in 'dev' directly to avoid that.
{
- struct dmz_bioctx *bioctx = dm_per_bio_data(bio, sizeof(struct dmz_bioctx));
+ struct dmz_bioctx *bioctx =
+ dm_per_bio_data(bio, sizeof(struct dmz_bioctx));
struct bio *clone;
clone = bio_clone_fast(bio, GFP_NOIO, &dmz->bio_set);
if (!clone)
return -ENOMEM;
- bio_set_dev(clone, dmz->dev->bdev);
+ bio_set_dev(clone, dev->bdev);
+ bioctx->dev = dev;
clone->bi_iter.bi_sector =
dmz_start_sect(dmz->metadata, zone) + dmz_blk2sect(chunk_block);
clone->bi_iter.bi_size = dmz_blk2sect(nr_blocks) << SECTOR_SHIFT;
@@ -162,8 +169,8 @@ static void dmz_handle_read_zero(struct dmz_target *dmz, struct bio *bio,
/*
* Process a read BIO.
*/
-static int dmz_handle_read(struct dmz_target *dmz, struct dm_zone *zone,
- struct bio *bio)
+static int dmz_handle_read(struct dmz_target *dmz, struct dmz_dev *dev,
+ struct dm_zone *zone, struct bio *bio)
Same comment as above.
... which is why I did it ...
{
struct dmz_metadata *zmd = dmz->metadata;
sector_t chunk_block = dmz_chunk_block(zmd, dmz_bio_block(bio));
@@ -178,7 +185,7 @@ static int dmz_handle_read(struct dmz_target *dmz, struct dm_zone *zone,
return 0;
}
- dmz_dev_debug(dmz->dev, "READ chunk %llu -> %s zone %u, block %llu, %u blocks",
+ dmz_dev_debug(dev, "READ chunk %llu -> %s zone %u, block %llu, %u blocks",
(unsigned long long)dmz_bio_chunk(zmd, bio),
(dmz_is_rnd(zone) ? "RND" : "SEQ"),
dmz_id(zmd, zone),
DMDEBUG to print the label ? or we could also have dmz_dev_debug() print format
changed to something like: "%s (%s): ...", label_name, dev->bdev_name
As you've seen, I've not been very consistent when to use the device
name or the label. But indeed, using both as you suggested is a good idea.
I'll be modifying it for the next round.
@@ -218,7 +225,8 @@ static int dmz_handle_read(struct dmz_target *dmz, struct dm_zone *zone,
if (nr_blocks) {
/* Valid blocks found: read them */
nr_blocks = min_t(unsigned int, nr_blocks, end_block - chunk_block);
- ret = dmz_submit_bio(dmz, rzone, bio, chunk_block, nr_blocks);
+ ret = dmz_submit_bio(dmz, dev, rzone, bio,
+ chunk_block, nr_blocks);
if (ret)
return ret;
chunk_block += nr_blocks;
@@ -238,6 +246,7 @@ static int dmz_handle_read(struct dmz_target *dmz, struct dm_zone *zone,
* in place.
*/
static int dmz_handle_direct_write(struct dmz_target *dmz,
+ struct dmz_dev *dev,
Use dmz_zone_to_dev() ?
struct dm_zone *zone, struct bio *bio,
sector_t chunk_block,
unsigned int nr_blocks)
@@ -250,7 +259,7 @@ static int dmz_handle_direct_write(struct dmz_target *dmz,
return -EROFS;
/* Submit write */
- ret = dmz_submit_bio(dmz, zone, bio, chunk_block, nr_blocks);
+ ret = dmz_submit_bio(dmz, dev, zone, bio, chunk_block, nr_blocks);
if (ret)
return ret;
@@ -271,6 +280,7 @@ static int dmz_handle_direct_write(struct dmz_target *dmz,
* Called with @zone write locked.
*/
static int dmz_handle_buffered_write(struct dmz_target *dmz,
+ struct dmz_dev *dev,
struct dm_zone *zone, struct bio *bio,
sector_t chunk_block,
unsigned int nr_blocks)
@@ -288,7 +298,7 @@ static int dmz_handle_buffered_write(struct dmz_target *dmz,
return -EROFS;
/* Submit write */
- ret = dmz_submit_bio(dmz, bzone, bio, chunk_block, nr_blocks);
+ ret = dmz_submit_bio(dmz, dev, bzone, bio, chunk_block, nr_blocks);
Yes, I think it would be far better to use dmz_zone_to_dev() instead of passing
directly dev. That will avoid bugs like passing a wrong dev for a zone or wrong
zone for a dev.
As mentioned above, yes, we could.
In fact, that's what I did originally.
But then I thought it easier to pass the device precisely to avoid
having to call dmz_zone_to_dev() in every function.
However, I'll see how things pan out and will be modifying it for the
next round.
Cheers,
Hannes
--
Dr. Hannes Reinecke Teamlead Storage & Networking
hare@xxxxxxx +49 911 74053 688
SUSE Software Solutions GmbH, Maxfeldstr. 5, 90409 Nürnberg
HRB 36809 (AG Nürnberg), Geschäftsführer: Felix Imendörffer
--
dm-devel mailing list
dm-devel@xxxxxxxxxx
https://www.redhat.com/mailman/listinfo/dm-devel
