On Tue, Feb 08, 2022 at 07:38:08PM +0800, Yu Kuai wrote:
> Throttled bios can't be issued after del_gendisk() is done, thus
> it's better to cancel them immediately rather than waiting for
> throttle is done.
>
> For example, if user thread is throttled with low bps while it's
> issuing large io, and the device is deleted. The user thread will
> wait for a long time for io to return.
>
> Signed-off-by: Yu Kuai <yukuai3@xxxxxxxxxx>
> ---
> Changes in v8:
> - fold two patches into one
> Changes in v7:
> - use the new solution as suggested by Ming.
>
> block/blk-throttle.c | 49 ++++++++++++++++++++++++++++++++++++++++----
> block/blk-throttle.h | 2 ++
> block/genhd.c | 2 ++
> 3 files changed, 49 insertions(+), 4 deletions(-)
>
> diff --git a/block/blk-throttle.c b/block/blk-throttle.c
> index 7c462c006b26..557d20796157 100644
> --- a/block/blk-throttle.c
> +++ b/block/blk-throttle.c
> @@ -43,8 +43,12 @@
> static struct workqueue_struct *kthrotld_workqueue;
>
> enum tg_state_flags {
> - THROTL_TG_PENDING = 1 << 0, /* on parent's pending tree */
> - THROTL_TG_WAS_EMPTY = 1 << 1, /* bio_lists[] became non-empty */
> + /* on parent's pending tree */
> + THROTL_TG_PENDING = 1 << 0,
> + /* bio_lists[] became non-empty */
> + THROTL_TG_WAS_EMPTY = 1 << 1,
> + /* starts to cancel all bios, will be set if the disk is deleted */
> + THROTL_TG_CANCELING = 1 << 2,
> };
>
> #define rb_entry_tg(node) rb_entry((node), struct throtl_grp, rb_node)
> @@ -871,7 +875,8 @@ static bool tg_may_dispatch(struct throtl_grp *tg, struct bio *bio,
> bio != throtl_peek_queued(&tg->service_queue.queued[rw]));
>
> /* If tg->bps = -1, then BW is unlimited */
> - if (bps_limit == U64_MAX && iops_limit == UINT_MAX) {
> + if ((bps_limit == U64_MAX && iops_limit == UINT_MAX) ||
> + tg->flags & THROTL_TG_CANCELING) {
> if (wait)
> *wait = 0;
> return true;
> @@ -974,6 +979,9 @@ static void tg_update_disptime(struct throtl_grp *tg)
> unsigned long read_wait = -1, write_wait = -1, min_wait = -1, disptime;
> struct bio *bio;
>
> + if (tg->flags & THROTL_TG_CANCELING)
> + goto update;
> +
> bio = throtl_peek_queued(&sq->queued[READ]);
> if (bio)
> tg_may_dispatch(tg, bio, &read_wait);
> @@ -983,9 +991,10 @@ static void tg_update_disptime(struct throtl_grp *tg)
> tg_may_dispatch(tg, bio, &write_wait);
>
> min_wait = min(read_wait, write_wait);
> - disptime = jiffies + min_wait;
>
> +update:
> /* Update dispatch time */
> + disptime = jiffies + min_wait;
As I mentioned on V7, the change in tg_update_disptime() isn't needed, please
drop it.
> throtl_dequeue_tg(tg);
> tg->disptime = disptime;
> throtl_enqueue_tg(tg);
> @@ -1763,6 +1772,38 @@ static bool throtl_hierarchy_can_upgrade(struct throtl_grp *tg)
> return false;
> }
>
> +void blk_throtl_cancel_bios(struct request_queue *q)
> +{
> + struct cgroup_subsys_state *pos_css;
> + struct blkcg_gq *blkg;
> +
> + spin_lock_irq(&q->queue_lock);
> + /*
> + * queue_lock is held, rcu lock is not needed here technically.
> + * However, rcu lock is still held to emphasize that following
> + * path need RCU protection and to prevent warning from lockdep.
> + */
> + rcu_read_lock();
> + blkg_for_each_descendant_post(blkg, pos_css, q->root_blkg) {
> + struct throtl_grp *tg = blkg_to_tg(blkg);
> + struct throtl_service_queue *sq = &tg->service_queue;
> +
> + /*
> + * Set disptime in the past to make sure
> + * throtl_select_dispatch() won't exit without dispatching.
> + */
> + tg->disptime = jiffies - 1;
It might be better to replace the above line with tg_update_disptime().
Otherwise, the patch looks good.
Thanks,
Ming
