Re: [PATCH -next v2] md: synchronize flush io with array reconfiguration

From: Song Liu
Date: Mon Nov 27 2023 - 22:47:39 EST


On Mon, Nov 27, 2023 at 6:12 PM Yu Kuai <yukuai1@xxxxxxxxxxxxxxx> wrote:
>
> Hi,
>
> 在 2023/11/28 7:32, Song Liu 写道:
> > On Mon, Nov 27, 2023 at 2:16 PM Song Liu <song@xxxxxxxxxx> wrote:
> >>
> >> On Fri, Nov 24, 2023 at 10:54 PM Yu Kuai <yukuai1@xxxxxxxxxxxxxxx> wrote:
> >>>
> >>> From: Yu Kuai <yukuai3@xxxxxxxxxx>
> >>>
> >>> Currently rcu is used to protect iterating rdev from submit_flushes():
> >>>
> >>> submit_flushes remove_and_add_spares
> >>> synchronize_rcu
> >>> pers->hot_remove_disk()
> >>> rcu_read_lock()
> >>> rdev_for_each_rcu
> >>> if (rdev->raid_disk >= 0)
> >>> rdev->radi_disk = -1;
> >>> atomic_inc(&rdev->nr_pending)
> >>> rcu_read_unlock()
> >>> bi = bio_alloc_bioset()
> >>> bi->bi_end_io = md_end_flush
> >>> bi->private = rdev
> >>> submit_bio
> >>> // issue io for removed rdev
> >>>
> >>> Fix this problem by grabbing 'acive_io' before iterating rdev, make sure
> >>> that remove_and_add_spares() won't concurrent with submit_flushes().
> >>>
> >>> Fixes: a2826aa92e2e ("md: support barrier requests on all personalities.")
> >>> Signed-off-by: Yu Kuai <yukuai3@xxxxxxxxxx>
> >>> ---
> >>> Changes v2:
> >>> - Add WARN_ON in case md_flush_request() is not called from
> >>> md_handle_request() in future.
> >>>
> >>> drivers/md/md.c | 22 ++++++++++++++++------
> >>> 1 file changed, 16 insertions(+), 6 deletions(-)
> >>>
> >>> diff --git a/drivers/md/md.c b/drivers/md/md.c
> >>> index 86efc9c2ae56..2ffedc39edd6 100644
> >>> --- a/drivers/md/md.c
> >>> +++ b/drivers/md/md.c
> >>> @@ -538,6 +538,9 @@ static void md_end_flush(struct bio *bio)
> >>> rdev_dec_pending(rdev, mddev);
> >>>
> >>> if (atomic_dec_and_test(&mddev->flush_pending)) {
> >>> + /* The pair is percpu_ref_tryget() from md_flush_request() */
> >>> + percpu_ref_put(&mddev->active_io);
> >>> +
> >>> /* The pre-request flush has finished */
> >>> queue_work(md_wq, &mddev->flush_work);
> >>> }
> >>> @@ -557,12 +560,8 @@ static void submit_flushes(struct work_struct *ws)
> >>> rdev_for_each_rcu(rdev, mddev)
> >>> if (rdev->raid_disk >= 0 &&
> >>> !test_bit(Faulty, &rdev->flags)) {
> >>> - /* Take two references, one is dropped
> >>> - * when request finishes, one after
> >>> - * we reclaim rcu_read_lock
> >>> - */
> >>> struct bio *bi;
> >>> - atomic_inc(&rdev->nr_pending);
> >>> +
> >>> atomic_inc(&rdev->nr_pending);
> >>> rcu_read_unlock();
> >>> bi = bio_alloc_bioset(rdev->bdev, 0,
> >>> @@ -573,7 +572,6 @@ static void submit_flushes(struct work_struct *ws)
> >>> atomic_inc(&mddev->flush_pending);
> >>> submit_bio(bi);
> >>> rcu_read_lock();
> >>> - rdev_dec_pending(rdev, mddev);
> >>> }
> >>> rcu_read_unlock();
> >>> if (atomic_dec_and_test(&mddev->flush_pending))
> >>> @@ -626,6 +624,18 @@ bool md_flush_request(struct mddev *mddev, struct bio *bio)
> >>> /* new request after previous flush is completed */
> >>> if (ktime_after(req_start, mddev->prev_flush_start)) {
> >>> WARN_ON(mddev->flush_bio);
> >>> + /*
> >>> + * Grab a reference to make sure mddev_suspend() will wait for
> >>> + * this flush to be done.
> >>> + *
> >>> + * md_flush_reqeust() is called under md_handle_request() and
> >>> + * 'active_io' is already grabbed, hence percpu_ref_tryget()
> >>> + * won't fail, percpu_ref_tryget_live() can't be used because
> >>> + * percpu_ref_kill() can be called by mddev_suspend()
> >>> + * concurrently.
> >>> + */
> >>> + if (WARN_ON(percpu_ref_tryget(&mddev->active_io)))
> >>
> >> This should be "if (!WARN_ON(..))", right?
>
> Sorry for the mistake, this actually should be:
>
> if (WARN_ON(!percpu_ref_tryget(...))
> >>
> >> Song
> >>
> >>> + percpu_ref_get(&mddev->active_io);
> >
> > Actually, we can just use percpu_ref_get(), no?
>
> Yes, we can, but if someone else doesn't call md_flush_request() under
> md_handle_request() in the fulture, there will be problem and
> percpu_ref_get() can't catch this, do you think it'll make sense to
> prevent such case?

This combination is really weird

+ if (WARN_ON(percpu_ref_tryget(&mddev->active_io)))
+ percpu_ref_get(&mddev->active_io);

We can use percpu_ref_get() here, and add
WARN_ON(percpu_ref_is_zero()) earlier in the function. Does this
make sense?

Thanks,
Song