Commit de701c74 authored by Vivek Goyal's avatar Vivek Goyal Committed by Jens Axboe
Browse files

blk-throttle: Some cleanups and race fixes in limit update code 



When throttle group limits are updated through cgroups, a thread is
woken up to process these updates. While reviewing that code, oleg noted
couple of race conditions existed in the code and he also suggested that
code can be simplified.

This patch fixes the races simplifies the code based on Oleg's suggestions:

	- Use xchg().
	- Introduced a common function throtl_update_blkio_group_common()
          which is shared now by all iops/bps update functions.

Reviewed-by: default avatarOleg Nesterov <oleg@redhat.com>
Reviewed-by: default avatarPaul E. McKenney <paulmck@linux.vnet.ibm.com>
Signed-off-by: default avatarVivek Goyal <vgoyal@redhat.com>

Fixed a merge issue, throtl_schedule_delayed_work() takes throtl_data
as the argument now, not the queue.

Signed-off-by: default avatarJens Axboe <jaxboe@fusionio.com>
parent 231d704b

block/blk-throttle.c

+40 −56
Original line number Diff line number Diff line
@@ -102,7 +102,7 @@ struct throtl_data 
	/* Work for dispatching throttled bios */

	struct delayed_work throtl_work;



	atomic_t limits_changed;

	bool limits_changed;

};



enum tg_state_flags {
@@ -201,6 +201,7 @@ static struct throtl_grp * throtl_find_alloc_tg(struct throtl_data *td, 
	RB_CLEAR_NODE(&tg->rb_node);

	bio_list_init(&tg->bio_lists[0]);

	bio_list_init(&tg->bio_lists[1]);

	td->limits_changed = false;



	/*

	 * Take the initial reference that will be released on destroy
@@ -737,34 +738,27 @@ static void throtl_process_limit_change(struct throtl_data *td) 
	struct throtl_grp *tg;

	struct hlist_node *pos, *n;



	if (!atomic_read(&td->limits_changed))

	if (!td->limits_changed)

		return;



	throtl_log(td, "limit changed =%d", atomic_read(&td->limits_changed));

	xchg(&td->limits_changed, false);



	/*

	 * Make sure updates from throtl_update_blkio_group_read_bps() group

	 * of functions to tg->limits_changed are visible. We do not

	 * want update td->limits_changed to be visible but update to

	 * tg->limits_changed not being visible yet on this cpu. Hence

	 * the read barrier.

	 */

	smp_rmb();

	throtl_log(td, "limits changed");



	hlist_for_each_entry_safe(tg, pos, n, &td->tg_list, tg_node) {

		if (throtl_tg_on_rr(tg) && tg->limits_changed) {

		if (!tg->limits_changed)

			continue;



		if (!xchg(&tg->limits_changed, false))

			continue;



		throtl_log_tg(td, tg, "limit change rbps=%llu wbps=%llu"

				" riops=%u wiops=%u", tg->bps[READ],

				tg->bps[WRITE], tg->iops[READ],

				tg->iops[WRITE]);

			" riops=%u wiops=%u", tg->bps[READ], tg->bps[WRITE],

			tg->iops[READ], tg->iops[WRITE]);



		if (throtl_tg_on_rr(tg))

			tg_update_disptime(td, tg);

			tg->limits_changed = false;

		}

	}



	smp_mb__before_atomic_dec();

	atomic_dec(&td->limits_changed);

	smp_mb__after_atomic_dec();

}



/* Dispatch throttled bios. Should be called without queue lock held. */
@@ -898,6 +892,15 @@ void throtl_unlink_blkio_group(void *key, struct blkio_group *blkg) 
	spin_unlock_irqrestore(td->queue->queue_lock, flags);

}



static void throtl_update_blkio_group_common(struct throtl_data *td,

				struct throtl_grp *tg)

{

	xchg(&tg->limits_changed, true);

	xchg(&td->limits_changed, true);

	/* Schedule a work now to process the limit change */

	throtl_schedule_delayed_work(td, 0);

}



/*

 * For all update functions, key should be a valid pointer because these

 * update functions are called under blkcg_lock, that means, blkg is
@@ -911,61 +914,40 @@ static void throtl_update_blkio_group_read_bps(void *key, 
				struct blkio_group *blkg, u64 read_bps)

{

	struct throtl_data *td = key;

	struct throtl_grp *tg = tg_of_blkg(blkg);



	tg_of_blkg(blkg)->bps[READ] = read_bps;

	/* Make sure read_bps is updated before setting limits_changed */

	smp_wmb();

	tg_of_blkg(blkg)->limits_changed = true;



	/* Make sure tg->limits_changed is updated before td->limits_changed */

	smp_mb__before_atomic_inc();

	atomic_inc(&td->limits_changed);

	smp_mb__after_atomic_inc();



	/* Schedule a work now to process the limit change */

	throtl_schedule_delayed_work(td, 0);

	tg->bps[READ] = read_bps;

	throtl_update_blkio_group_common(td, tg);

}



static void throtl_update_blkio_group_write_bps(void *key,

				struct blkio_group *blkg, u64 write_bps)

{

	struct throtl_data *td = key;

	struct throtl_grp *tg = tg_of_blkg(blkg);



	tg_of_blkg(blkg)->bps[WRITE] = write_bps;

	smp_wmb();

	tg_of_blkg(blkg)->limits_changed = true;

	smp_mb__before_atomic_inc();

	atomic_inc(&td->limits_changed);

	smp_mb__after_atomic_inc();

	throtl_schedule_delayed_work(td, 0);

	tg->bps[WRITE] = write_bps;

	throtl_update_blkio_group_common(td, tg);

}



static void throtl_update_blkio_group_read_iops(void *key,

			struct blkio_group *blkg, unsigned int read_iops)

{

	struct throtl_data *td = key;

	struct throtl_grp *tg = tg_of_blkg(blkg);



	tg_of_blkg(blkg)->iops[READ] = read_iops;

	smp_wmb();

	tg_of_blkg(blkg)->limits_changed = true;

	smp_mb__before_atomic_inc();

	atomic_inc(&td->limits_changed);

	smp_mb__after_atomic_inc();

	throtl_schedule_delayed_work(td, 0);

	tg->iops[READ] = read_iops;

	throtl_update_blkio_group_common(td, tg);

}



static void throtl_update_blkio_group_write_iops(void *key,

			struct blkio_group *blkg, unsigned int write_iops)

{

	struct throtl_data *td = key;

	struct throtl_grp *tg = tg_of_blkg(blkg);



	tg_of_blkg(blkg)->iops[WRITE] = write_iops;

	smp_wmb();

	tg_of_blkg(blkg)->limits_changed = true;

	smp_mb__before_atomic_inc();

	atomic_inc(&td->limits_changed);

	smp_mb__after_atomic_inc();

	throtl_schedule_delayed_work(td, 0);

	tg->iops[WRITE] = write_iops;

	throtl_update_blkio_group_common(td, tg);

}



static void throtl_shutdown_wq(struct request_queue *q)
@@ -1012,6 +994,7 @@ int blk_throtl_bio(struct request_queue *q, struct bio **biop) 
		 */

		update_disptime = false;

		goto queue_bio;



	}



	/* Bio is with-in rate limit of group */
@@ -1052,7 +1035,7 @@ int blk_throtl_init(struct request_queue *q) 


	INIT_HLIST_HEAD(&td->tg_list);

	td->tg_service_tree = THROTL_RB_ROOT;

	atomic_set(&td->limits_changed, 0);

	td->limits_changed = false;



	/* Init root group */

	tg = &td->root_tg;
@@ -1064,6 +1047,7 @@ int blk_throtl_init(struct request_queue *q) 
	/* Practically unlimited BW */

	tg->bps[0] = tg->bps[1] = -1;

	tg->iops[0] = tg->iops[1] = -1;

	td->limits_changed = false;



	/*

	 * Set root group reference to 2. One reference will be dropped when

CRA Git | Maintained and supported by SUSTech CRA and CCSE