Commit 5a85071c authored by NeilBrown's avatar NeilBrown Committed by Shaohua Li

md: use a separate bio_set for synchronous IO.

md devices allocate a bio_set and use it for two
distinct purposes.
mddev->bio_set is used to clone bios as part of sending
upper level requests down to lower level devices,
and it is also use for synchronous IO such as superblock
and bitmap updates, and for correcting read errors.

This multiple usage can lead to deadlocks.  It is likely
that cloned bios might be queued for write and to be
waiting for a metadata update before the write can be permitted.
If the cloning exhausted mddev->bio_set, the metadata update
may not be able to proceed.

This scenario has been seen during heavy testing, with lots of IO and
lots of memory pressure.

Address this by adding a new bio_set specifically for synchronous IO.
All synchronous IO goes directly to the underlying device and is not
queued at the md level, so request using entries from the new
mddev->sync_set will complete in a timely fashion.
Requests that use mddev->bio_set will sometimes need to wait
for synchronous IO, but will no longer risk deadlocking that iO.

Also: small simplification in mddev_put(): there is no need to
wait until the spinlock is released before calling bioset_free().
Signed-off-by: default avatarNeilBrown <neilb@suse.com>
Signed-off-by: default avatarShaohua Li <shli@fb.com>
parent 8df72024
...@@ -203,6 +203,14 @@ struct bio *bio_alloc_mddev(gfp_t gfp_mask, int nr_iovecs, ...@@ -203,6 +203,14 @@ struct bio *bio_alloc_mddev(gfp_t gfp_mask, int nr_iovecs,
} }
EXPORT_SYMBOL_GPL(bio_alloc_mddev); EXPORT_SYMBOL_GPL(bio_alloc_mddev);
static struct bio *md_bio_alloc_sync(struct mddev *mddev)
{
if (!mddev->sync_set)
return bio_alloc(GFP_NOIO, 1);
return bio_alloc_bioset(GFP_NOIO, 1, mddev->sync_set);
}
/* /*
* We have a system wide 'event count' that is incremented * We have a system wide 'event count' that is incremented
* on any 'interesting' event, and readers of /proc/mdstat * on any 'interesting' event, and readers of /proc/mdstat
...@@ -467,8 +475,6 @@ static void mddev_delayed_delete(struct work_struct *ws); ...@@ -467,8 +475,6 @@ static void mddev_delayed_delete(struct work_struct *ws);
static void mddev_put(struct mddev *mddev) static void mddev_put(struct mddev *mddev)
{ {
struct bio_set *bs = NULL;
if (!atomic_dec_and_lock(&mddev->active, &all_mddevs_lock)) if (!atomic_dec_and_lock(&mddev->active, &all_mddevs_lock))
return; return;
if (!mddev->raid_disks && list_empty(&mddev->disks) && if (!mddev->raid_disks && list_empty(&mddev->disks) &&
...@@ -476,8 +482,12 @@ static void mddev_put(struct mddev *mddev) ...@@ -476,8 +482,12 @@ static void mddev_put(struct mddev *mddev)
/* Array is not configured at all, and not held active, /* Array is not configured at all, and not held active,
* so destroy it */ * so destroy it */
list_del_init(&mddev->all_mddevs); list_del_init(&mddev->all_mddevs);
bs = mddev->bio_set; if (mddev->bio_set)
bioset_free(mddev->bio_set);
if (mddev->sync_set)
bioset_free(mddev->sync_set);
mddev->bio_set = NULL; mddev->bio_set = NULL;
mddev->sync_set = NULL;
if (mddev->gendisk) { if (mddev->gendisk) {
/* We did a probe so need to clean up. Call /* We did a probe so need to clean up. Call
* queue_work inside the spinlock so that * queue_work inside the spinlock so that
...@@ -490,8 +500,6 @@ static void mddev_put(struct mddev *mddev) ...@@ -490,8 +500,6 @@ static void mddev_put(struct mddev *mddev)
kfree(mddev); kfree(mddev);
} }
spin_unlock(&all_mddevs_lock); spin_unlock(&all_mddevs_lock);
if (bs)
bioset_free(bs);
} }
static void md_safemode_timeout(unsigned long data); static void md_safemode_timeout(unsigned long data);
...@@ -756,7 +764,7 @@ void md_super_write(struct mddev *mddev, struct md_rdev *rdev, ...@@ -756,7 +764,7 @@ void md_super_write(struct mddev *mddev, struct md_rdev *rdev,
if (test_bit(Faulty, &rdev->flags)) if (test_bit(Faulty, &rdev->flags))
return; return;
bio = bio_alloc_mddev(GFP_NOIO, 1, mddev); bio = md_bio_alloc_sync(mddev);
atomic_inc(&rdev->nr_pending); atomic_inc(&rdev->nr_pending);
...@@ -788,7 +796,7 @@ int md_super_wait(struct mddev *mddev) ...@@ -788,7 +796,7 @@ int md_super_wait(struct mddev *mddev)
int sync_page_io(struct md_rdev *rdev, sector_t sector, int size, int sync_page_io(struct md_rdev *rdev, sector_t sector, int size,
struct page *page, int op, int op_flags, bool metadata_op) struct page *page, int op, int op_flags, bool metadata_op)
{ {
struct bio *bio = bio_alloc_mddev(GFP_NOIO, 1, rdev->mddev); struct bio *bio = md_bio_alloc_sync(rdev->mddev);
int ret; int ret;
bio->bi_bdev = (metadata_op && rdev->meta_bdev) ? bio->bi_bdev = (metadata_op && rdev->meta_bdev) ?
...@@ -5437,6 +5445,11 @@ int md_run(struct mddev *mddev) ...@@ -5437,6 +5445,11 @@ int md_run(struct mddev *mddev)
if (!mddev->bio_set) if (!mddev->bio_set)
return -ENOMEM; return -ENOMEM;
} }
if (mddev->sync_set == NULL) {
mddev->sync_set = bioset_create(BIO_POOL_SIZE, 0);
if (!mddev->sync_set)
return -ENOMEM;
}
spin_lock(&pers_lock); spin_lock(&pers_lock);
pers = find_pers(mddev->level, mddev->clevel); pers = find_pers(mddev->level, mddev->clevel);
......
...@@ -444,6 +444,9 @@ struct mddev { ...@@ -444,6 +444,9 @@ struct mddev {
struct attribute_group *to_remove; struct attribute_group *to_remove;
struct bio_set *bio_set; struct bio_set *bio_set;
struct bio_set *sync_set; /* for sync operations like
* metadata and bitmap writes
*/
/* Generic flush handling. /* Generic flush handling.
* The last to finish preflush schedules a worker to submit * The last to finish preflush schedules a worker to submit
......
Markdown is supported
0%
or
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment