On Wed, May 31, 2023 at 3:55 PM Yu Kuai <yukuai1@xxxxxxxxxxxxxxx> wrote:
>
> Hi,
>
> 在 2023/05/31 15:50, Xiao Ni 写道:
> > On Mon, May 29, 2023 at 9:14 PM Yu Kuai <yukuai1@xxxxxxxxxxxxxxx> wrote:
> >>
> >> From: Yu Kuai <yukuai3@xxxxxxxxxx>
> >>
> >> current->bio_list will be set under submit_bio() context, in this case
> >> bitmap io will be added to the list and wait for current io submission to
> >> finish, while current io submission must wait for bitmap io to be done.
> >> commit 874807a83139 ("md/raid1{,0}: fix deadlock in bitmap_unplug.") fix
> >> the deadlock by handling plugged bio by daemon thread.
> >
> > Thanks for the historic introduction. I did a test and printed the
> > logs in raid10_unplug. The tools I used are dd and mkfs. from_schedule
> > is always true during I/O and it's 0 when io finishes. So I have a
> > question here, how can I trigger the condition that from_schedule is 0
> > and current->list is not NULL? In other words, is there really a
> > deadlock here? Before your patch it looks like all bios are merged
> > into conf->pending_bio_list and are handled by raid10d. It can't
> > submit bio directly in the originating process which mentioned in
> > 57c67df48866
> >
> As I mentioned below, after commit a214b949d8e3, this deadlock doesn't
> exist anymore, and without this patch, patch 7 will introduce this
> scenario again.
>
> Thanks,
> Kuai
> >>
> >> On the one hand, the deadlock won't exist after commit a214b949d8e3
> >> ("blk-mq: only flush requests from the plug in blk_mq_submit_bio"). On
> >> the other hand, current solution makes it impossible to flush plugged bio
> >> in raid1/10_make_request(), because this will cause that all the writes
> >> will goto daemon thread.
> >>
> >> In order to limit the number of plugged bio, commit 874807a83139
> >> ("md/raid1{,0}: fix deadlock in bitmap_unplug.") is reverted, and the
> >> deadlock is fixed by handling bitmap io asynchronously.
> >>
> >> Signed-off-by: Yu Kuai <yukuai3@xxxxxxxxxx>
> >> ---
> >> drivers/md/raid1-10.c | 14 ++++++++++++++
> >> drivers/md/raid1.c | 4 ++--
> >> drivers/md/raid10.c | 8 +++-----
> >> 3 files changed, 19 insertions(+), 7 deletions(-)
> >>
> >> diff --git a/drivers/md/raid1-10.c b/drivers/md/raid1-10.c
> >> index 73cc3cb9154d..17e55c1fd5a1 100644
> >> --- a/drivers/md/raid1-10.c
> >> +++ b/drivers/md/raid1-10.c
> >> @@ -151,3 +151,17 @@ static inline bool raid1_add_bio_to_plug(struct mddev *mddev, struct bio *bio,
> >>
> >> return true;
> >> }
> >> +
> >> +/*
> >> + * current->bio_list will be set under submit_bio() context, in this case bitmap
> >> + * io will be added to the list and wait for current io submission to finish,
> >> + * while current io submission must wait for bitmap io to be done. In order to
> >> + * avoid such deadlock, submit bitmap io asynchronously.
> >> + */
> >> +static inline void raid1_prepare_flush_writes(struct bitmap *bitmap)
> >> +{
> >> + if (current->bio_list)
> >> + md_bitmap_unplug_async(bitmap);
> >> + else
> >> + md_bitmap_unplug(bitmap);
> >> +}
> >> diff --git a/drivers/md/raid1.c b/drivers/md/raid1.c
> >> index 0778e398584c..006620fed595 100644
> >> --- a/drivers/md/raid1.c
> >> +++ b/drivers/md/raid1.c
> >> @@ -794,7 +794,7 @@ static int read_balance(struct r1conf *conf, struct r1bio *r1_bio, int *max_sect
> >> static void flush_bio_list(struct r1conf *conf, struct bio *bio)
> >> {
> >> /* flush any pending bitmap writes to disk before proceeding w/ I/O */
> >> - md_bitmap_unplug(conf->mddev->bitmap);
> >> + raid1_prepare_flush_writes(conf->mddev->bitmap);
> >
> > If we unplug bitmap asynchronously, can we make sure the bitmap are
> > flushed before the corresponding data?
Could you explain this question?
Regards
Xiao
> >
> > Regards
> > Xiao
> >
> >> wake_up(&conf->wait_barrier);
> >>
> >> while (bio) { /* submit pending writes */
> >> @@ -1166,7 +1166,7 @@ static void raid1_unplug(struct blk_plug_cb *cb, bool from_schedule)
> >> struct r1conf *conf = mddev->private;
> >> struct bio *bio;
> >>
> >> - if (from_schedule || current->bio_list) {
> >> + if (from_schedule) {
> >> spin_lock_irq(&conf->device_lock);
> >> bio_list_merge(&conf->pending_bio_list, &plug->pending);
> >> spin_unlock_irq(&conf->device_lock);
> >> diff --git a/drivers/md/raid10.c b/drivers/md/raid10.c
> >> index 6640507ecb0d..fb22cfe94d32 100644
> >> --- a/drivers/md/raid10.c
> >> +++ b/drivers/md/raid10.c
> >> @@ -902,9 +902,7 @@ static void flush_pending_writes(struct r10conf *conf)
> >> __set_current_state(TASK_RUNNING);
> >>
> >> blk_start_plug(&plug);
> >> - /* flush any pending bitmap writes to disk
> >> - * before proceeding w/ I/O */
> >> - md_bitmap_unplug(conf->mddev->bitmap);
> >> + raid1_prepare_flush_writes(conf->mddev->bitmap);
> >> wake_up(&conf->wait_barrier);
> >>
> >> while (bio) { /* submit pending writes */
> >> @@ -1108,7 +1106,7 @@ static void raid10_unplug(struct blk_plug_cb *cb, bool from_schedule)
> >> struct r10conf *conf = mddev->private;
> >> struct bio *bio;
> >>
> >> - if (from_schedule || current->bio_list) {
> >> + if (from_schedule) {
> >> spin_lock_irq(&conf->device_lock);
> >> bio_list_merge(&conf->pending_bio_list, &plug->pending);
> >> spin_unlock_irq(&conf->device_lock);
> >> @@ -1120,7 +1118,7 @@ static void raid10_unplug(struct blk_plug_cb *cb, bool from_schedule)
> >>
> >> /* we aren't scheduling, so we can do the write-out directly. */
> >> bio = bio_list_get(&plug->pending);
> >> - md_bitmap_unplug(mddev->bitmap);
> >> + raid1_prepare_flush_writes(mddev->bitmap);
> >> wake_up(&conf->wait_barrier);
> >>
> >> while (bio) { /* submit pending writes */
> >> --
> >> 2.39.2
> >>
> >
> > .
> >
>
