On Thu, Aug 30, 2018 at 09:53:44PM -0400, Dennis Zhou wrote:
> From: "Dennis Zhou (Facebook)" <dennisszhou@xxxxxxxxx>
>
> There is a very small change a bio gets caught up in a really
> unfortunate race between a task migration, cgroup exiting, and itself
> trying to associate with a blkg. This is due to css offlining being
> performed after the css->refcnt is killed which triggers removal of
> blkgs that reach their blkg->refcnt of 0.
>
> To avoid this, association with a blkg should use tryget and fallback to
> using the root_blkg.
>
> Fixes: 08e18eab0c579 ("block: add bi_blkg to the bio for cgroups")
> Signed-off-by: Dennis Zhou <dennisszhou@xxxxxxxxx>
> Cc: Jiufei Xue <jiufei.xue@xxxxxxxxxxxxxxxxx>
> Cc: Joseph Qi <joseph.qi@xxxxxxxxxxxxxxxxx>
> Cc: Tejun Heo <tj@xxxxxxxxxx>
> Cc: Josef Bacik <josef@xxxxxxxxxxxxxx>
> Cc: Jens Axboe <axboe@xxxxxxxxx>
> ---
> block/bio.c | 3 ++-
> block/blk-throttle.c | 5 +++--
> 2 files changed, 5 insertions(+), 3 deletions(-)
>
> diff --git a/block/bio.c b/block/bio.c
> index 04969b392c72..4473ccd22987 100644
> --- a/block/bio.c
> +++ b/block/bio.c
> @@ -1987,7 +1987,8 @@ int bio_associate_blkg(struct bio *bio, struct blkcg_gq *blkg)
> {
> if (unlikely(bio->bi_blkg))
> return -EBUSY;
> - blkg_get(blkg);
> + if (!blkg_try_get(blkg))
> + return -ENODEV;
> bio->bi_blkg = blkg;
> return 0;
> }
> diff --git a/block/blk-throttle.c b/block/blk-throttle.c
> index a3eede00d302..c626e1f7cdcd 100644
> --- a/block/blk-throttle.c
> +++ b/block/blk-throttle.c
> @@ -2129,8 +2129,9 @@ static inline void throtl_update_latency_buckets(struct throtl_data *td)
> static void blk_throtl_assoc_bio(struct throtl_grp *tg, struct bio *bio)
> {
> #ifdef CONFIG_BLK_DEV_THROTTLING_LOW
> - if (bio->bi_css)
> - bio_associate_blkg(bio, tg_to_blkg(tg));
> + /* fallback to root_blkg if we fail to get a blkg ref */
> + if (bio->bi_css && bio_associate_blkg(bio, tg_to_blkg(tg)))
> + bio_associate_blkg(bio, bio->bi_disk->queue->root_blkg);
Except if we've already assocated a blkg this is just extra, can we do
if (bio->bi_css && (bio_associate_blkg(bio, tg_to_blkg(tg)) == -ENODEV))
to make it clear that we're only attaching it to the root if we failed to attach
a blkg at all? Thanks,
Josef
