#include <linux/bio.h>
#include <linux/blktrace_api.h>
#include "blk-cgroup.h"
+#include "blk.h"
/* Max dispatch from a group in 1 round */
static int throtl_grp_quantum = 8;
unsigned long slice_end[2];
/* Some throttle limits got updated for the group */
- bool limits_changed;
+ int limits_changed;
+
+ struct rcu_head rcu_head;
};
struct throtl_data
/* service tree for active throtl groups */
struct throtl_rb_root tg_service_tree;
- struct throtl_grp root_tg;
+ struct throtl_grp *root_tg;
struct request_queue *queue;
/* Total Number of queued bios on READ and WRITE lists */
/* Work for dispatching throttled bios */
struct delayed_work throtl_work;
- bool limits_changed;
+ int limits_changed;
};
enum tg_state_flags {
return NULL;
}
-static inline int total_nr_queued(struct throtl_data *td)
+static inline unsigned int total_nr_queued(struct throtl_data *td)
{
- return (td->nr_queued[0] + td->nr_queued[1]);
+ return td->nr_queued[0] + td->nr_queued[1];
}
static inline struct throtl_grp *throtl_ref_get_tg(struct throtl_grp *tg)
return tg;
}
-static void throtl_put_tg(struct throtl_grp *tg)
+static void throtl_free_tg(struct rcu_head *head)
{
- BUG_ON(atomic_read(&tg->ref) <= 0);
- if (!atomic_dec_and_test(&tg->ref))
- return;
+ struct throtl_grp *tg;
+
+ tg = container_of(head, struct throtl_grp, rcu_head);
+ free_percpu(tg->blkg.stats_cpu);
kfree(tg);
}
-static struct throtl_grp * throtl_find_alloc_tg(struct throtl_data *td,
- struct cgroup *cgroup)
+static void throtl_put_tg(struct throtl_grp *tg)
{
- struct blkio_cgroup *blkcg = cgroup_to_blkio_cgroup(cgroup);
- struct throtl_grp *tg = NULL;
- void *key = td;
- struct backing_dev_info *bdi = &td->queue->backing_dev_info;
- unsigned int major, minor;
+ BUG_ON(atomic_read(&tg->ref) <= 0);
+ if (!atomic_dec_and_test(&tg->ref))
+ return;
/*
- * TODO: Speed up blkiocg_lookup_group() by maintaining a radix
- * tree of blkg (instead of traversing through hash list all
- * the time.
+ * A group is freed in rcu manner. But having an rcu lock does not
+ * mean that one can access all the fields of blkg and assume these
+ * are valid. For example, don't try to follow throtl_data and
+ * request queue links.
+ *
+ * Having a reference to blkg under an rcu allows acess to only
+ * values local to groups like group stats and group rate limits
*/
+ call_rcu(&tg->rcu_head, throtl_free_tg);
+}
- /*
- * This is the common case when there are no blkio cgroups.
- * Avoid lookup in this case
- */
- if (blkcg == &blkio_root_cgroup)
- tg = &td->root_tg;
- else
- tg = tg_of_blkg(blkiocg_lookup_group(blkcg, key));
-
- /* Fill in device details for root group */
- if (tg && !tg->blkg.dev && bdi->dev && dev_name(bdi->dev)) {
- sscanf(dev_name(bdi->dev), "%u:%u", &major, &minor);
- tg->blkg.dev = MKDEV(major, minor);
- goto done;
- }
-
- if (tg)
- goto done;
-
- tg = kzalloc_node(sizeof(*tg), GFP_ATOMIC, td->queue->node);
- if (!tg)
- goto done;
-
+static void throtl_init_group(struct throtl_grp *tg)
+{
INIT_HLIST_NODE(&tg->tg_node);
RB_CLEAR_NODE(&tg->rb_node);
bio_list_init(&tg->bio_lists[0]);
bio_list_init(&tg->bio_lists[1]);
- td->limits_changed = false;
+ tg->limits_changed = false;
+
+ /* Practically unlimited BW */
+ tg->bps[0] = tg->bps[1] = -1;
+ tg->iops[0] = tg->iops[1] = -1;
/*
* Take the initial reference that will be released on destroy
* exit or cgroup deletion path depending on who is exiting first.
*/
atomic_set(&tg->ref, 1);
+}
+
+/* Should be called with rcu read lock held (needed for blkcg) */
+static void
+throtl_add_group_to_td_list(struct throtl_data *td, struct throtl_grp *tg)
+{
+ hlist_add_head(&tg->tg_node, &td->tg_list);
+ td->nr_undestroyed_grps++;
+}
+
+static void
+__throtl_tg_fill_dev_details(struct throtl_data *td, struct throtl_grp *tg)
+{
+ struct backing_dev_info *bdi = &td->queue->backing_dev_info;
+ unsigned int major, minor;
+
+ if (!tg || tg->blkg.dev)
+ return;
+
+ /*
+ * Fill in device details for a group which might not have been
+ * filled at group creation time as queue was being instantiated
+ * and driver had not attached a device yet
+ */
+ if (bdi->dev && dev_name(bdi->dev)) {
+ sscanf(dev_name(bdi->dev), "%u:%u", &major, &minor);
+ tg->blkg.dev = MKDEV(major, minor);
+ }
+}
+
+/*
+ * Should be called with without queue lock held. Here queue lock will be
+ * taken rarely. It will be taken only once during life time of a group
+ * if need be
+ */
+static void
+throtl_tg_fill_dev_details(struct throtl_data *td, struct throtl_grp *tg)
+{
+ if (!tg || tg->blkg.dev)
+ return;
+
+ spin_lock_irq(td->queue->queue_lock);
+ __throtl_tg_fill_dev_details(td, tg);
+ spin_unlock_irq(td->queue->queue_lock);
+}
+
+static void throtl_init_add_tg_lists(struct throtl_data *td,
+ struct throtl_grp *tg, struct blkio_cgroup *blkcg)
+{
+ __throtl_tg_fill_dev_details(td, tg);
/* Add group onto cgroup list */
- sscanf(dev_name(bdi->dev), "%u:%u", &major, &minor);
blkiocg_add_blkio_group(blkcg, &tg->blkg, (void *)td,
- MKDEV(major, minor), BLKIO_POLICY_THROTL);
+ tg->blkg.dev, BLKIO_POLICY_THROTL);
tg->bps[READ] = blkcg_get_read_bps(blkcg, tg->blkg.dev);
tg->bps[WRITE] = blkcg_get_write_bps(blkcg, tg->blkg.dev);
tg->iops[READ] = blkcg_get_read_iops(blkcg, tg->blkg.dev);
tg->iops[WRITE] = blkcg_get_write_iops(blkcg, tg->blkg.dev);
- hlist_add_head(&tg->tg_node, &td->tg_list);
- td->nr_undestroyed_grps++;
-done:
+ throtl_add_group_to_td_list(td, tg);
+}
+
+/* Should be called without queue lock and outside of rcu period */
+static struct throtl_grp *throtl_alloc_tg(struct throtl_data *td)
+{
+ struct throtl_grp *tg = NULL;
+ int ret;
+
+ tg = kzalloc_node(sizeof(*tg), GFP_ATOMIC, td->queue->node);
+ if (!tg)
+ return NULL;
+
+ ret = blkio_alloc_blkg_stats(&tg->blkg);
+
+ if (ret) {
+ kfree(tg);
+ return NULL;
+ }
+
+ throtl_init_group(tg);
return tg;
}
-static struct throtl_grp * throtl_get_tg(struct throtl_data *td)
+static struct
+throtl_grp *throtl_find_tg(struct throtl_data *td, struct blkio_cgroup *blkcg)
{
- struct cgroup *cgroup;
struct throtl_grp *tg = NULL;
+ void *key = td;
+
+ /*
+ * This is the common case when there are no blkio cgroups.
+ * Avoid lookup in this case
+ */
+ if (blkcg == &blkio_root_cgroup)
+ tg = td->root_tg;
+ else
+ tg = tg_of_blkg(blkiocg_lookup_group(blkcg, key));
+
+ __throtl_tg_fill_dev_details(td, tg);
+ return tg;
+}
+
+static struct throtl_grp * throtl_get_tg(struct throtl_data *td)
+{
+ struct throtl_grp *tg = NULL, *__tg = NULL;
+ struct blkio_cgroup *blkcg;
+ struct request_queue *q = td->queue;
+
+ /* no throttling for dead queue */
+ if (unlikely(blk_queue_dead(q)))
+ return NULL;
rcu_read_lock();
- cgroup = task_cgroup(current, blkio_subsys_id);
- tg = throtl_find_alloc_tg(td, cgroup);
- if (!tg)
- tg = &td->root_tg;
+ blkcg = task_blkio_cgroup(current);
+ tg = throtl_find_tg(td, blkcg);
+ if (tg) {
+ rcu_read_unlock();
+ return tg;
+ }
+
+ /*
+ * Need to allocate a group. Allocation of group also needs allocation
+ * of per cpu stats which in-turn takes a mutex() and can block. Hence
+ * we need to drop rcu lock and queue_lock before we call alloc.
+ */
+ rcu_read_unlock();
+ spin_unlock_irq(q->queue_lock);
+
+ tg = throtl_alloc_tg(td);
+
+ /* Group allocated and queue is still alive. take the lock */
+ spin_lock_irq(q->queue_lock);
+
+ /* Make sure @q is still alive */
+ if (unlikely(blk_queue_dead(q))) {
+ kfree(tg);
+ return NULL;
+ }
+
+ /*
+ * Initialize the new group. After sleeping, read the blkcg again.
+ */
+ rcu_read_lock();
+ blkcg = task_blkio_cgroup(current);
+
+ /*
+ * If some other thread already allocated the group while we were
+ * not holding queue lock, free up the group
+ */
+ __tg = throtl_find_tg(td, blkcg);
+
+ if (__tg) {
+ kfree(tg);
+ rcu_read_unlock();
+ return __tg;
+ }
+
+ /* Group allocation failed. Account the IO to root group */
+ if (!tg) {
+ tg = td->root_tg;
+ return tg;
+ }
+
+ throtl_init_add_tg_lists(td, tg, blkcg);
rcu_read_unlock();
return tg;
}
return 0;
}
+static bool tg_no_rule_group(struct throtl_grp *tg, bool rw) {
+ if (tg->bps[rw] == -1 && tg->iops[rw] == -1)
+ return 1;
+ return 0;
+}
+
/*
* Returns whether one can dispatch a bio or not. Also returns approx number
* of jiffies to wait before this bio is with-in IO rate and can be dispatched
static void throtl_charge_bio(struct throtl_grp *tg, struct bio *bio)
{
bool rw = bio_data_dir(bio);
- bool sync = bio->bi_rw & REQ_SYNC;
+ bool sync = rw_is_sync(bio->bi_rw);
/* Charge the bio to the group */
tg->bytes_disp[rw] += bio->bi_size;
tg->io_disp[rw]++;
- /*
- * TODO: This will take blkg->stats_lock. Figure out a way
- * to avoid this cost.
- */
blkiocg_update_dispatch_stats(&tg->blkg, bio->bi_size, rw, sync);
}
" riops=%u wiops=%u", tg->bps[READ], tg->bps[WRITE],
tg->iops[READ], tg->iops[WRITE]);
+ /*
+ * Restart the slices for both READ and WRITES. It
+ * might happen that a group's limit are dropped
+ * suddenly and we don't want to account recently
+ * dispatched IO with new low rate
+ */
+ throtl_start_new_slice(td, tg, 0);
+ throtl_start_new_slice(td, tg, 1);
+
if (throtl_tg_on_rr(tg))
tg_update_disptime(td, tg);
}
unsigned int nr_disp = 0;
struct bio_list bio_list_on_stack;
struct bio *bio;
+ struct blk_plug plug;
spin_lock_irq(q->queue_lock);
bio_list_init(&bio_list_on_stack);
- throtl_log(td, "dispatch nr_queued=%lu read=%u write=%u",
+ throtl_log(td, "dispatch nr_queued=%u read=%u write=%u",
total_nr_queued(td), td->nr_queued[READ],
td->nr_queued[WRITE]);
* immediate dispatch
*/
if (nr_disp) {
+ blk_start_plug(&plug);
while((bio = bio_list_pop(&bio_list_on_stack)))
generic_make_request(bio);
- blk_unplug(q);
+ blk_finish_plug(&plug);
}
return nr_disp;
}
struct delayed_work *dwork = &td->throtl_work;
- if (total_nr_queued(td) > 0) {
+ /* schedule work if limits changed even if no bio is queued */
+ if (total_nr_queued(td) || td->limits_changed) {
/*
* We might have a work scheduled to be executed in future.
* Cancel that and schedule a new one.
}
}
-static void throtl_td_free(struct throtl_data *td)
-{
- kfree(td);
-}
-
/*
* Blk cgroup controller notification saying that blkio_group object is being
* delinked as associated cgroup object is going away. That also means that
/*
* For all update functions, key should be a valid pointer because these
* update functions are called under blkcg_lock, that means, blkg is
- * valid and in turn key is valid. queue exit path can not race becuase
+ * valid and in turn key is valid. queue exit path can not race because
* of blkcg_lock
*
* Can not take queue lock in update functions as queue lock under blkcg_lock
.plid = BLKIO_POLICY_THROTL,
};
-int blk_throtl_bio(struct request_queue *q, struct bio **biop)
+bool blk_throtl_bio(struct request_queue *q, struct bio *bio)
{
struct throtl_data *td = q->td;
struct throtl_grp *tg;
- struct bio *bio = *biop;
bool rw = bio_data_dir(bio), update_disptime = true;
+ struct blkio_cgroup *blkcg;
+ bool throttled = false;
if (bio->bi_rw & REQ_THROTTLED) {
bio->bi_rw &= ~REQ_THROTTLED;
- return 0;
+ goto out;
}
+ /*
+ * A throtl_grp pointer retrieved under rcu can be used to access
+ * basic fields like stats and io rates. If a group has no rules,
+ * just update the dispatch stats in lockless manner and return.
+ */
+
+ rcu_read_lock();
+ blkcg = task_blkio_cgroup(current);
+ tg = throtl_find_tg(td, blkcg);
+ if (tg) {
+ throtl_tg_fill_dev_details(td, tg);
+
+ if (tg_no_rule_group(tg, rw)) {
+ blkiocg_update_dispatch_stats(&tg->blkg, bio->bi_size,
+ rw, rw_is_sync(bio->bi_rw));
+ rcu_read_unlock();
+ goto out;
+ }
+ }
+ rcu_read_unlock();
+
+ /*
+ * Either group has not been allocated yet or it is not an unlimited
+ * IO group
+ */
spin_lock_irq(q->queue_lock);
tg = throtl_get_tg(td);
+ if (unlikely(!tg))
+ goto out_unlock;
if (tg->nr_queued[rw]) {
/*
/* Bio is with-in rate limit of group */
if (tg_may_dispatch(td, tg, bio, NULL)) {
throtl_charge_bio(tg, bio);
- goto out;
+
+ /*
+ * We need to trim slice even when bios are not being queued
+ * otherwise it might happen that a bio is not queued for
+ * a long time and slice keeps on extending and trim is not
+ * called for a long time. Now if limits are reduced suddenly
+ * we take into account all the IO dispatched so far at new
+ * low rate and * newly queued IO gets a really long dispatch
+ * time.
+ *
+ * So keep on trimming slice even if bio is not queued.
+ */
+ throtl_trim_slice(td, tg, rw);
+ goto out_unlock;
}
queue_bio:
- throtl_log_tg(td, tg, "[%c] bio. bdisp=%u sz=%u bps=%llu"
+ throtl_log_tg(td, tg, "[%c] bio. bdisp=%llu sz=%u bps=%llu"
" iodisp=%u iops=%u queued=%d/%d",
rw == READ ? 'R' : 'W',
tg->bytes_disp[rw], bio->bi_size, tg->bps[rw],
tg->nr_queued[READ], tg->nr_queued[WRITE]);
throtl_add_bio_tg(q->td, tg, bio);
- *biop = NULL;
+ throttled = true;
if (update_disptime) {
tg_update_disptime(td, tg);
throtl_schedule_next_dispatch(td);
}
+out_unlock:
+ spin_unlock_irq(q->queue_lock);
out:
+ return throttled;
+}
+
+/**
+ * blk_throtl_drain - drain throttled bios
+ * @q: request_queue to drain throttled bios for
+ *
+ * Dispatch all currently throttled bios on @q through ->make_request_fn().
+ */
+void blk_throtl_drain(struct request_queue *q)
+ __releases(q->queue_lock) __acquires(q->queue_lock)
+{
+ struct throtl_data *td = q->td;
+ struct throtl_rb_root *st = &td->tg_service_tree;
+ struct throtl_grp *tg;
+ struct bio_list bl;
+ struct bio *bio;
+
+ queue_lockdep_assert_held(q);
+
+ bio_list_init(&bl);
+
+ while ((tg = throtl_rb_first(st))) {
+ throtl_dequeue_tg(td, tg);
+
+ while ((bio = bio_list_peek(&tg->bio_lists[READ])))
+ tg_dispatch_one_bio(td, tg, bio_data_dir(bio), &bl);
+ while ((bio = bio_list_peek(&tg->bio_lists[WRITE])))
+ tg_dispatch_one_bio(td, tg, bio_data_dir(bio), &bl);
+ }
spin_unlock_irq(q->queue_lock);
- return 0;
+
+ while ((bio = bio_list_pop(&bl)))
+ generic_make_request(bio);
+
+ spin_lock_irq(q->queue_lock);
}
int blk_throtl_init(struct request_queue *q)
INIT_HLIST_HEAD(&td->tg_list);
td->tg_service_tree = THROTL_RB_ROOT;
td->limits_changed = false;
+ INIT_DELAYED_WORK(&td->throtl_work, blk_throtl_work);
- /* Init root group */
- tg = &td->root_tg;
- INIT_HLIST_NODE(&tg->tg_node);
- RB_CLEAR_NODE(&tg->rb_node);
- bio_list_init(&tg->bio_lists[0]);
- bio_list_init(&tg->bio_lists[1]);
-
- /* Practically unlimited BW */
- tg->bps[0] = tg->bps[1] = -1;
- tg->iops[0] = tg->iops[1] = -1;
- td->limits_changed = false;
+ /* alloc and Init root group. */
+ td->queue = q;
+ tg = throtl_alloc_tg(td);
- /*
- * Set root group reference to 2. One reference will be dropped when
- * all groups on tg_list are being deleted during queue exit. Other
- * reference will remain there as we don't want to delete this group
- * as it is statically allocated and gets destroyed when throtl_data
- * goes away.
- */
- atomic_set(&tg->ref, 2);
- hlist_add_head(&tg->tg_node, &td->tg_list);
- td->nr_undestroyed_grps++;
+ if (!tg) {
+ kfree(td);
+ return -ENOMEM;
+ }
- INIT_DELAYED_WORK(&td->throtl_work, blk_throtl_work);
+ td->root_tg = tg;
rcu_read_lock();
- blkiocg_add_blkio_group(&blkio_root_cgroup, &tg->blkg, (void *)td,
- 0, BLKIO_POLICY_THROTL);
+ throtl_init_add_tg_lists(td, tg, &blkio_root_cgroup);
rcu_read_unlock();
/* Attach throtl data to request queue */
- td->queue = q;
q->td = td;
return 0;
}
* it.
*/
throtl_shutdown_wq(q);
- throtl_td_free(td);
+}
+
+void blk_throtl_release(struct request_queue *q)
+{
+ kfree(q->td);
}
static int __init throtl_init(void)