dm zoned: reduce overhead of backing device checks
Commit 75d66ffb48efb3 added backing device health checks and as a part
of these checks, check_events() block ops template call is invoked in
dm-zoned mapping path as well as in reclaim and flush path. Calling
check_events() with ATA or SCSI backing devices introduces a blocking
scsi_test_unit_ready() call being made in sd_check_events(). Even though
the overhead of calling scsi_test_unit_ready() is small for ATA zoned
devices, it is much larger for SCSI and it affects performance in a very
negative way.
Fix this performance regression by executing check_events() only in case
of any I/O errors. The function dmz_bdev_is_dying() is modified to call
only blk_queue_dying(), while calls to check_events() are made in a new
helper function, dmz_check_bdev().
Reported-by: zhangxiaoxu <zhangxiaoxu5@huawei.com>
Fixes: 75d66ffb48efb3 ("dm zoned: properly handle backing device failure")
Cc: stable@vger.kernel.org
Signed-off-by: Dmitry Fomichev <dmitry.fomichev@wdc.com>
Signed-off-by: Mike Snitzer <snitzer@redhat.com>
diff --git a/drivers/md/dm-zoned-metadata.c b/drivers/md/dm-zoned-metadata.c
index 595a731..ac1179c 100644
--- a/drivers/md/dm-zoned-metadata.c
+++ b/drivers/md/dm-zoned-metadata.c
@@ -554,6 +554,7 @@ static struct dmz_mblock *dmz_get_mblock(struct dmz_metadata *zmd,
TASK_UNINTERRUPTIBLE);
if (test_bit(DMZ_META_ERROR, &mblk->state)) {
dmz_release_mblock(zmd, mblk);
+ dmz_check_bdev(zmd->dev);
return ERR_PTR(-EIO);
}
@@ -625,6 +626,8 @@ static int dmz_rdwr_block(struct dmz_metadata *zmd, int op, sector_t block,
ret = submit_bio_wait(bio);
bio_put(bio);
+ if (ret)
+ dmz_check_bdev(zmd->dev);
return ret;
}
@@ -691,6 +694,7 @@ static int dmz_write_dirty_mblocks(struct dmz_metadata *zmd,
TASK_UNINTERRUPTIBLE);
if (test_bit(DMZ_META_ERROR, &mblk->state)) {
clear_bit(DMZ_META_ERROR, &mblk->state);
+ dmz_check_bdev(zmd->dev);
ret = -EIO;
}
nr_mblks_submitted--;
@@ -768,7 +772,7 @@ int dmz_flush_metadata(struct dmz_metadata *zmd)
/* If there are no dirty metadata blocks, just flush the device cache */
if (list_empty(&write_list)) {
ret = blkdev_issue_flush(zmd->dev->bdev, GFP_NOIO, NULL);
- goto out;
+ goto err;
}
/*
@@ -778,7 +782,7 @@ int dmz_flush_metadata(struct dmz_metadata *zmd)
*/
ret = dmz_log_dirty_mblocks(zmd, &write_list);
if (ret)
- goto out;
+ goto err;
/*
* The log is on disk. It is now safe to update in place
@@ -786,11 +790,11 @@ int dmz_flush_metadata(struct dmz_metadata *zmd)
*/
ret = dmz_write_dirty_mblocks(zmd, &write_list, zmd->mblk_primary);
if (ret)
- goto out;
+ goto err;
ret = dmz_write_sb(zmd, zmd->mblk_primary);
if (ret)
- goto out;
+ goto err;
while (!list_empty(&write_list)) {
mblk = list_first_entry(&write_list, struct dmz_mblock, link);
@@ -805,16 +809,20 @@ int dmz_flush_metadata(struct dmz_metadata *zmd)
zmd->sb_gen++;
out:
- if (ret && !list_empty(&write_list)) {
- spin_lock(&zmd->mblk_lock);
- list_splice(&write_list, &zmd->mblk_dirty_list);
- spin_unlock(&zmd->mblk_lock);
- }
-
dmz_unlock_flush(zmd);
up_write(&zmd->mblk_sem);
return ret;
+
+err:
+ if (!list_empty(&write_list)) {
+ spin_lock(&zmd->mblk_lock);
+ list_splice(&write_list, &zmd->mblk_dirty_list);
+ spin_unlock(&zmd->mblk_lock);
+ }
+ if (!dmz_check_bdev(zmd->dev))
+ ret = -EIO;
+ goto out;
}
/*
@@ -1244,6 +1252,7 @@ static int dmz_update_zone(struct dmz_metadata *zmd, struct dm_zone *zone)
if (ret) {
dmz_dev_err(zmd->dev, "Get zone %u report failed",
dmz_id(zmd, zone));
+ dmz_check_bdev(zmd->dev);
return ret;
}