drbd: remove BIOSET_NEED_RESCUER flag from drbd_{md_,}io_bio_set

Careful analysis shows that this flag is not needed.

The RESCUER flag is only needed when a make_request_fn might:
 - allocate a bio from the bioset
 - submit it with generic_make_request() or similar
 - allocate another bio from the bioset

The second allocation can block until the first bio is processed, so
a rescuer is needed to ensure the first bio does get processed.  With
a rescuer it will only get processed when the make_request_fn completes.

In drbd, allocations from drbd_io_bio_set happen from drbd_new_req()
or w_restart_disk_io() which is only called to handle
RESTART_FROZEN_DISK_IO.

In former is called precisely once from the make_request_fn.
The later is never called by within the make_request_fn.

So there cannot be two allocations in the same call to the
make_request_fn, so a rescuer is not needed.

Allocations from drbd_md_io_bio_set are used for IO to the bitmap and
the activity log.  There are only accessed from worker threads and
workqueues, never directly from make_request_fn.
Again, the rescuer isn't needed.

Signed-off-by: NeilBrown <neilb@suse.com>
Signed-off-by: Philipp Reisner <philipp.reisner@linbit.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
This commit is contained in:
NeilBrown 2017-08-30 13:47:12 +02:00 committed by Jens Axboe
parent 5fc1efd5de
commit 974c58566e
1 changed files with 2 additions and 3 deletions

View File

@ -2178,13 +2178,12 @@ static int drbd_create_mempools(void)
goto Enomem;
/* mempools */
drbd_io_bio_set = bioset_create(BIO_POOL_SIZE, 0, BIOSET_NEED_RESCUER);
drbd_io_bio_set = bioset_create(BIO_POOL_SIZE, 0, 0);
if (drbd_io_bio_set == NULL)
goto Enomem;
drbd_md_io_bio_set = bioset_create(DRBD_MIN_POOL_PAGES, 0,
BIOSET_NEED_BVECS |
BIOSET_NEED_RESCUER);
BIOSET_NEED_BVECS);
if (drbd_md_io_bio_set == NULL)
goto Enomem;