blk-throttle: make blk_throtl_drain() ready for hierarchy
authorTejun Heo <tj@kernel.org>
Tue, 14 May 2013 20:52:37 +0000 (13:52 -0700)
committerTejun Heo <tj@kernel.org>
Tue, 14 May 2013 20:52:37 +0000 (13:52 -0700)
The current blk_throtl_drain() assumes that all active throtl_grps are
queued on throtl_data->service_queue, which won't be true once
hierarchy support is implemented.

This patch makes blk_throtl_drain() perform post-order walk of the
blkg hierarchy draining each associated throtl_grp, which guarantees
that all bios will eventually be pushed to the top-level service_queue
in throtl_data.

Signed-off-by: Tejun Heo <tj@kernel.org>
Acked-by: Vivek Goyal <vgoyal@redhat.com>
block/blk-throttle.c

index 918d222408560dc413c7c47cecf758a10ef5939a..8c6e13359781c605f1146cc827a1c00e92c1d87f 100644 (file)
@@ -1299,6 +1299,28 @@ out:
        return throttled;
 }
 
+/*
+ * Dispatch all bios from all children tg's queued on @parent_sq.  On
+ * return, @parent_sq is guaranteed to not have any active children tg's
+ * and all bios from previously active tg's are on @parent_sq->bio_lists[].
+ */
+static void tg_drain_bios(struct throtl_service_queue *parent_sq)
+{
+       struct throtl_grp *tg;
+
+       while ((tg = throtl_rb_first(parent_sq))) {
+               struct throtl_service_queue *sq = &tg->service_queue;
+               struct bio *bio;
+
+               throtl_dequeue_tg(tg);
+
+               while ((bio = bio_list_peek(&sq->bio_lists[READ])))
+                       tg_dispatch_one_bio(tg, bio_data_dir(bio));
+               while ((bio = bio_list_peek(&sq->bio_lists[WRITE])))
+                       tg_dispatch_one_bio(tg, bio_data_dir(bio));
+       }
+}
+
 /**
  * blk_throtl_drain - drain throttled bios
  * @q: request_queue to drain throttled bios for
@@ -1309,27 +1331,34 @@ void blk_throtl_drain(struct request_queue *q)
        __releases(q->queue_lock) __acquires(q->queue_lock)
 {
        struct throtl_data *td = q->td;
-       struct throtl_service_queue *parent_sq = &td->service_queue;
-       struct throtl_grp *tg;
+       struct blkcg_gq *blkg;
+       struct cgroup *pos_cgrp;
        struct bio *bio;
        int rw;
 
        queue_lockdep_assert_held(q);
+       rcu_read_lock();
 
-       while ((tg = throtl_rb_first(parent_sq))) {
-               struct throtl_service_queue *sq = &tg->service_queue;
+       /*
+        * Drain each tg while doing post-order walk on the blkg tree, so
+        * that all bios are propagated to td->service_queue.  It'd be
+        * better to walk service_queue tree directly but blkg walk is
+        * easier.
+        */
+       blkg_for_each_descendant_post(blkg, pos_cgrp, td->queue->root_blkg)
+               tg_drain_bios(&blkg_to_tg(blkg)->service_queue);
 
-               throtl_dequeue_tg(tg);
+       tg_drain_bios(&td_root_tg(td)->service_queue);
 
-               while ((bio = bio_list_peek(&sq->bio_lists[READ])))
-                       tg_dispatch_one_bio(tg, bio_data_dir(bio));
-               while ((bio = bio_list_peek(&sq->bio_lists[WRITE])))
-                       tg_dispatch_one_bio(tg, bio_data_dir(bio));
-       }
+       /* finally, transfer bios from top-level tg's into the td */
+       tg_drain_bios(&td->service_queue);
+
+       rcu_read_unlock();
        spin_unlock_irq(q->queue_lock);
 
+       /* all bios now should be in td->service_queue, issue them */
        for (rw = READ; rw <= WRITE; rw++)
-               while ((bio = bio_list_pop(&parent_sq->bio_lists[rw])))
+               while ((bio = bio_list_pop(&td->service_queue.bio_lists[rw])))
                        generic_make_request(bio);
 
        spin_lock_irq(q->queue_lock);