2 * Common Block IO controller cgroup interface
4 * Based on ideas and code from CFQ, CFS and BFQ:
5 * Copyright (C) 2003 Jens Axboe <axboe@kernel.dk>
7 * Copyright (C) 2008 Fabio Checconi <fabio@gandalf.sssup.it>
8 * Paolo Valente <paolo.valente@unimore.it>
10 * Copyright (C) 2009 Vivek Goyal <vgoyal@redhat.com>
11 * Nauman Rafique <nauman@google.com>
13 #include <linux/ioprio.h>
14 #include <linux/seq_file.h>
15 #include <linux/kdev_t.h>
16 #include <linux/module.h>
17 #include <linux/err.h>
18 #include <linux/blkdev.h>
19 #include <linux/slab.h>
20 #include <linux/genhd.h>
21 #include <linux/delay.h>
22 #include <linux/atomic.h>
23 #include "blk-cgroup.h"
26 #define MAX_KEY_LEN 100
28 static DEFINE_SPINLOCK(blkio_list_lock);
29 static LIST_HEAD(blkio_list);
31 static DEFINE_MUTEX(all_q_mutex);
32 static LIST_HEAD(all_q_list);
34 /* List of groups pending per cpu stats allocation */
35 static DEFINE_SPINLOCK(alloc_list_lock);
36 static LIST_HEAD(alloc_list);
38 static void blkio_stat_alloc_fn(struct work_struct *);
39 static DECLARE_DELAYED_WORK(blkio_stat_alloc_work, blkio_stat_alloc_fn);
41 struct blkio_cgroup blkio_root_cgroup = { .weight = 2*BLKIO_WEIGHT_DEFAULT };
42 EXPORT_SYMBOL_GPL(blkio_root_cgroup);
44 static struct blkio_policy_type *blkio_policy[BLKIO_NR_POLICIES];
46 /* for encoding cft->private value on file */
47 #define BLKIOFILE_PRIVATE(x, val) (((x) << 16) | (val))
48 /* What policy owns the file, proportional or throttle */
49 #define BLKIOFILE_POLICY(val) (((val) >> 16) & 0xffff)
50 #define BLKIOFILE_ATTR(val) ((val) & 0xffff)
52 struct blkio_cgroup *cgroup_to_blkio_cgroup(struct cgroup *cgroup)
54 return container_of(cgroup_subsys_state(cgroup, blkio_subsys_id),
55 struct blkio_cgroup, css);
57 EXPORT_SYMBOL_GPL(cgroup_to_blkio_cgroup);
59 static struct blkio_cgroup *task_blkio_cgroup(struct task_struct *tsk)
61 return container_of(task_subsys_state(tsk, blkio_subsys_id),
62 struct blkio_cgroup, css);
65 struct blkio_cgroup *bio_blkio_cgroup(struct bio *bio)
67 if (bio && bio->bi_css)
68 return container_of(bio->bi_css, struct blkio_cgroup, css);
69 return task_blkio_cgroup(current);
71 EXPORT_SYMBOL_GPL(bio_blkio_cgroup);
73 static inline void blkio_update_group_weight(struct blkio_group *blkg,
74 int plid, unsigned int weight)
76 struct blkio_policy_type *blkiop;
78 list_for_each_entry(blkiop, &blkio_list, list) {
79 /* If this policy does not own the blkg, do not send updates */
80 if (blkiop->plid != plid)
82 if (blkiop->ops.blkio_update_group_weight_fn)
83 blkiop->ops.blkio_update_group_weight_fn(blkg->q,
88 static inline void blkio_update_group_bps(struct blkio_group *blkg, int plid,
91 struct blkio_policy_type *blkiop;
93 list_for_each_entry(blkiop, &blkio_list, list) {
95 /* If this policy does not own the blkg, do not send updates */
96 if (blkiop->plid != plid)
99 if (fileid == BLKIO_THROTL_read_bps_device
100 && blkiop->ops.blkio_update_group_read_bps_fn)
101 blkiop->ops.blkio_update_group_read_bps_fn(blkg->q,
104 if (fileid == BLKIO_THROTL_write_bps_device
105 && blkiop->ops.blkio_update_group_write_bps_fn)
106 blkiop->ops.blkio_update_group_write_bps_fn(blkg->q,
111 static inline void blkio_update_group_iops(struct blkio_group *blkg,
112 int plid, unsigned int iops,
115 struct blkio_policy_type *blkiop;
117 list_for_each_entry(blkiop, &blkio_list, list) {
119 /* If this policy does not own the blkg, do not send updates */
120 if (blkiop->plid != plid)
123 if (fileid == BLKIO_THROTL_read_iops_device
124 && blkiop->ops.blkio_update_group_read_iops_fn)
125 blkiop->ops.blkio_update_group_read_iops_fn(blkg->q,
128 if (fileid == BLKIO_THROTL_write_iops_device
129 && blkiop->ops.blkio_update_group_write_iops_fn)
130 blkiop->ops.blkio_update_group_write_iops_fn(blkg->q,
136 * Add to the appropriate stat variable depending on the request type.
137 * This should be called with queue_lock held.
139 static void blkio_add_stat(uint64_t *stat, uint64_t add, bool direction,
143 stat[BLKIO_STAT_WRITE] += add;
145 stat[BLKIO_STAT_READ] += add;
147 stat[BLKIO_STAT_SYNC] += add;
149 stat[BLKIO_STAT_ASYNC] += add;
153 * Decrements the appropriate stat variable if non-zero depending on the
154 * request type. Panics on value being zero.
155 * This should be called with the queue_lock held.
157 static void blkio_check_and_dec_stat(uint64_t *stat, bool direction, bool sync)
160 BUG_ON(stat[BLKIO_STAT_WRITE] == 0);
161 stat[BLKIO_STAT_WRITE]--;
163 BUG_ON(stat[BLKIO_STAT_READ] == 0);
164 stat[BLKIO_STAT_READ]--;
167 BUG_ON(stat[BLKIO_STAT_SYNC] == 0);
168 stat[BLKIO_STAT_SYNC]--;
170 BUG_ON(stat[BLKIO_STAT_ASYNC] == 0);
171 stat[BLKIO_STAT_ASYNC]--;
175 #ifdef CONFIG_DEBUG_BLK_CGROUP
176 /* This should be called with the queue_lock held. */
177 static void blkio_set_start_group_wait_time(struct blkio_group *blkg,
178 struct blkio_policy_type *pol,
179 struct blkio_group *curr_blkg)
181 struct blkg_policy_data *pd = blkg->pd[pol->plid];
183 if (blkio_blkg_waiting(&pd->stats))
185 if (blkg == curr_blkg)
187 pd->stats.start_group_wait_time = sched_clock();
188 blkio_mark_blkg_waiting(&pd->stats);
191 /* This should be called with the queue_lock held. */
192 static void blkio_update_group_wait_time(struct blkio_group_stats *stats)
194 unsigned long long now;
196 if (!blkio_blkg_waiting(stats))
200 if (time_after64(now, stats->start_group_wait_time))
201 stats->group_wait_time += now - stats->start_group_wait_time;
202 blkio_clear_blkg_waiting(stats);
205 /* This should be called with the queue_lock held. */
206 static void blkio_end_empty_time(struct blkio_group_stats *stats)
208 unsigned long long now;
210 if (!blkio_blkg_empty(stats))
214 if (time_after64(now, stats->start_empty_time))
215 stats->empty_time += now - stats->start_empty_time;
216 blkio_clear_blkg_empty(stats);
219 void blkiocg_update_set_idle_time_stats(struct blkio_group *blkg,
220 struct blkio_policy_type *pol)
222 struct blkio_group_stats *stats = &blkg->pd[pol->plid]->stats;
224 lockdep_assert_held(blkg->q->queue_lock);
225 BUG_ON(blkio_blkg_idling(stats));
227 stats->start_idle_time = sched_clock();
228 blkio_mark_blkg_idling(stats);
230 EXPORT_SYMBOL_GPL(blkiocg_update_set_idle_time_stats);
232 void blkiocg_update_idle_time_stats(struct blkio_group *blkg,
233 struct blkio_policy_type *pol)
235 struct blkio_group_stats *stats = &blkg->pd[pol->plid]->stats;
237 lockdep_assert_held(blkg->q->queue_lock);
239 if (blkio_blkg_idling(stats)) {
240 unsigned long long now = sched_clock();
242 if (time_after64(now, stats->start_idle_time)) {
243 u64_stats_update_begin(&stats->syncp);
244 stats->idle_time += now - stats->start_idle_time;
245 u64_stats_update_end(&stats->syncp);
247 blkio_clear_blkg_idling(stats);
250 EXPORT_SYMBOL_GPL(blkiocg_update_idle_time_stats);
252 void blkiocg_update_avg_queue_size_stats(struct blkio_group *blkg,
253 struct blkio_policy_type *pol)
255 struct blkio_group_stats *stats = &blkg->pd[pol->plid]->stats;
257 lockdep_assert_held(blkg->q->queue_lock);
259 u64_stats_update_begin(&stats->syncp);
260 stats->avg_queue_size_sum +=
261 stats->stat_arr[BLKIO_STAT_QUEUED][BLKIO_STAT_READ] +
262 stats->stat_arr[BLKIO_STAT_QUEUED][BLKIO_STAT_WRITE];
263 stats->avg_queue_size_samples++;
264 blkio_update_group_wait_time(stats);
265 u64_stats_update_end(&stats->syncp);
267 EXPORT_SYMBOL_GPL(blkiocg_update_avg_queue_size_stats);
269 void blkiocg_set_start_empty_time(struct blkio_group *blkg,
270 struct blkio_policy_type *pol)
272 struct blkio_group_stats *stats = &blkg->pd[pol->plid]->stats;
274 lockdep_assert_held(blkg->q->queue_lock);
276 if (stats->stat_arr[BLKIO_STAT_QUEUED][BLKIO_STAT_READ] ||
277 stats->stat_arr[BLKIO_STAT_QUEUED][BLKIO_STAT_WRITE])
281 * group is already marked empty. This can happen if cfqq got new
282 * request in parent group and moved to this group while being added
283 * to service tree. Just ignore the event and move on.
285 if (blkio_blkg_empty(stats))
288 stats->start_empty_time = sched_clock();
289 blkio_mark_blkg_empty(stats);
291 EXPORT_SYMBOL_GPL(blkiocg_set_start_empty_time);
293 void blkiocg_update_dequeue_stats(struct blkio_group *blkg,
294 struct blkio_policy_type *pol,
295 unsigned long dequeue)
297 struct blkg_policy_data *pd = blkg->pd[pol->plid];
299 lockdep_assert_held(blkg->q->queue_lock);
301 pd->stats.dequeue += dequeue;
303 EXPORT_SYMBOL_GPL(blkiocg_update_dequeue_stats);
305 static inline void blkio_set_start_group_wait_time(struct blkio_group *blkg,
306 struct blkio_policy_type *pol,
307 struct blkio_group *curr_blkg) { }
308 static inline void blkio_end_empty_time(struct blkio_group_stats *stats) { }
311 void blkiocg_update_io_add_stats(struct blkio_group *blkg,
312 struct blkio_policy_type *pol,
313 struct blkio_group *curr_blkg, bool direction,
316 struct blkio_group_stats *stats = &blkg->pd[pol->plid]->stats;
318 lockdep_assert_held(blkg->q->queue_lock);
320 u64_stats_update_begin(&stats->syncp);
321 blkio_add_stat(stats->stat_arr[BLKIO_STAT_QUEUED], 1, direction, sync);
322 blkio_end_empty_time(stats);
323 u64_stats_update_end(&stats->syncp);
325 blkio_set_start_group_wait_time(blkg, pol, curr_blkg);
327 EXPORT_SYMBOL_GPL(blkiocg_update_io_add_stats);
329 void blkiocg_update_io_remove_stats(struct blkio_group *blkg,
330 struct blkio_policy_type *pol,
331 bool direction, bool sync)
333 struct blkio_group_stats *stats = &blkg->pd[pol->plid]->stats;
335 lockdep_assert_held(blkg->q->queue_lock);
337 u64_stats_update_begin(&stats->syncp);
338 blkio_check_and_dec_stat(stats->stat_arr[BLKIO_STAT_QUEUED], direction,
340 u64_stats_update_end(&stats->syncp);
342 EXPORT_SYMBOL_GPL(blkiocg_update_io_remove_stats);
344 void blkiocg_update_timeslice_used(struct blkio_group *blkg,
345 struct blkio_policy_type *pol,
347 unsigned long unaccounted_time)
349 struct blkio_group_stats *stats = &blkg->pd[pol->plid]->stats;
351 lockdep_assert_held(blkg->q->queue_lock);
353 u64_stats_update_begin(&stats->syncp);
355 #ifdef CONFIG_DEBUG_BLK_CGROUP
356 stats->unaccounted_time += unaccounted_time;
358 u64_stats_update_end(&stats->syncp);
360 EXPORT_SYMBOL_GPL(blkiocg_update_timeslice_used);
363 * should be called under rcu read lock or queue lock to make sure blkg pointer
366 void blkiocg_update_dispatch_stats(struct blkio_group *blkg,
367 struct blkio_policy_type *pol,
368 uint64_t bytes, bool direction, bool sync)
370 struct blkg_policy_data *pd = blkg->pd[pol->plid];
371 struct blkio_group_stats_cpu *stats_cpu;
374 /* If per cpu stats are not allocated yet, don't do any accounting. */
375 if (pd->stats_cpu == NULL)
379 * Disabling interrupts to provide mutual exclusion between two
380 * writes on same cpu. It probably is not needed for 64bit. Not
381 * optimizing that case yet.
383 local_irq_save(flags);
385 stats_cpu = this_cpu_ptr(pd->stats_cpu);
387 u64_stats_update_begin(&stats_cpu->syncp);
388 stats_cpu->sectors += bytes >> 9;
389 blkio_add_stat(stats_cpu->stat_arr_cpu[BLKIO_STAT_CPU_SERVICED],
391 blkio_add_stat(stats_cpu->stat_arr_cpu[BLKIO_STAT_CPU_SERVICE_BYTES],
392 bytes, direction, sync);
393 u64_stats_update_end(&stats_cpu->syncp);
394 local_irq_restore(flags);
396 EXPORT_SYMBOL_GPL(blkiocg_update_dispatch_stats);
398 void blkiocg_update_completion_stats(struct blkio_group *blkg,
399 struct blkio_policy_type *pol,
401 uint64_t io_start_time, bool direction,
404 struct blkio_group_stats *stats = &blkg->pd[pol->plid]->stats;
405 unsigned long long now = sched_clock();
407 lockdep_assert_held(blkg->q->queue_lock);
409 u64_stats_update_begin(&stats->syncp);
410 if (time_after64(now, io_start_time))
411 blkio_add_stat(stats->stat_arr[BLKIO_STAT_SERVICE_TIME],
412 now - io_start_time, direction, sync);
413 if (time_after64(io_start_time, start_time))
414 blkio_add_stat(stats->stat_arr[BLKIO_STAT_WAIT_TIME],
415 io_start_time - start_time, direction, sync);
416 u64_stats_update_end(&stats->syncp);
418 EXPORT_SYMBOL_GPL(blkiocg_update_completion_stats);
420 /* Merged stats are per cpu. */
421 void blkiocg_update_io_merged_stats(struct blkio_group *blkg,
422 struct blkio_policy_type *pol,
423 bool direction, bool sync)
425 struct blkio_group_stats *stats = &blkg->pd[pol->plid]->stats;
427 lockdep_assert_held(blkg->q->queue_lock);
429 u64_stats_update_begin(&stats->syncp);
430 blkio_add_stat(stats->stat_arr[BLKIO_STAT_MERGED], 1, direction, sync);
431 u64_stats_update_end(&stats->syncp);
433 EXPORT_SYMBOL_GPL(blkiocg_update_io_merged_stats);
436 * Worker for allocating per cpu stat for blk groups. This is scheduled on
437 * the system_nrt_wq once there are some groups on the alloc_list waiting
440 static void blkio_stat_alloc_fn(struct work_struct *work)
442 static void *pcpu_stats[BLKIO_NR_POLICIES];
443 struct delayed_work *dwork = to_delayed_work(work);
444 struct blkio_group *blkg;
449 for (i = 0; i < BLKIO_NR_POLICIES; i++) {
450 if (pcpu_stats[i] != NULL)
453 pcpu_stats[i] = alloc_percpu(struct blkio_group_stats_cpu);
455 /* Allocation failed. Try again after some time. */
456 if (pcpu_stats[i] == NULL) {
457 queue_delayed_work(system_nrt_wq, dwork,
458 msecs_to_jiffies(10));
463 spin_lock_irq(&blkio_list_lock);
464 spin_lock(&alloc_list_lock);
466 /* cgroup got deleted or queue exited. */
467 if (!list_empty(&alloc_list)) {
468 blkg = list_first_entry(&alloc_list, struct blkio_group,
470 for (i = 0; i < BLKIO_NR_POLICIES; i++) {
471 struct blkg_policy_data *pd = blkg->pd[i];
473 if (blkio_policy[i] && pd && !pd->stats_cpu)
474 swap(pd->stats_cpu, pcpu_stats[i]);
477 list_del_init(&blkg->alloc_node);
480 empty = list_empty(&alloc_list);
482 spin_unlock(&alloc_list_lock);
483 spin_unlock_irq(&blkio_list_lock);
490 * blkg_free - free a blkg
491 * @blkg: blkg to free
493 * Free @blkg which may be partially allocated.
495 static void blkg_free(struct blkio_group *blkg)
502 for (i = 0; i < BLKIO_NR_POLICIES; i++) {
503 struct blkg_policy_data *pd = blkg->pd[i];
506 free_percpu(pd->stats_cpu);
515 * blkg_alloc - allocate a blkg
516 * @blkcg: block cgroup the new blkg is associated with
517 * @q: request_queue the new blkg is associated with
519 * Allocate a new blkg assocating @blkcg and @q.
521 static struct blkio_group *blkg_alloc(struct blkio_cgroup *blkcg,
522 struct request_queue *q)
524 struct blkio_group *blkg;
527 /* alloc and init base part */
528 blkg = kzalloc_node(sizeof(*blkg), GFP_ATOMIC, q->node);
533 INIT_LIST_HEAD(&blkg->q_node);
534 INIT_LIST_HEAD(&blkg->alloc_node);
537 cgroup_path(blkcg->css.cgroup, blkg->path, sizeof(blkg->path));
539 for (i = 0; i < BLKIO_NR_POLICIES; i++) {
540 struct blkio_policy_type *pol = blkio_policy[i];
541 struct blkg_policy_data *pd;
546 /* alloc per-policy data and attach it to blkg */
547 pd = kzalloc_node(sizeof(*pd) + pol->pdata_size, GFP_ATOMIC,
558 /* invoke per-policy init */
559 for (i = 0; i < BLKIO_NR_POLICIES; i++) {
560 struct blkio_policy_type *pol = blkio_policy[i];
563 pol->ops.blkio_init_group_fn(blkg);
569 struct blkio_group *blkg_lookup_create(struct blkio_cgroup *blkcg,
570 struct request_queue *q,
571 enum blkio_policy_id plid,
573 __releases(q->queue_lock) __acquires(q->queue_lock)
575 struct blkio_group *blkg;
577 WARN_ON_ONCE(!rcu_read_lock_held());
578 lockdep_assert_held(q->queue_lock);
581 * This could be the first entry point of blkcg implementation and
582 * we shouldn't allow anything to go through for a bypassing queue.
583 * The following can be removed if blkg lookup is guaranteed to
584 * fail on a bypassing queue.
586 if (unlikely(blk_queue_bypass(q)) && !for_root)
587 return ERR_PTR(blk_queue_dead(q) ? -EINVAL : -EBUSY);
589 blkg = blkg_lookup(blkcg, q);
593 /* blkg holds a reference to blkcg */
594 if (!css_tryget(&blkcg->css))
595 return ERR_PTR(-EINVAL);
598 * Allocate and initialize.
600 blkg = blkg_alloc(blkcg, q);
602 /* did alloc fail? */
603 if (unlikely(!blkg)) {
604 blkg = ERR_PTR(-ENOMEM);
609 spin_lock(&blkcg->lock);
610 hlist_add_head_rcu(&blkg->blkcg_node, &blkcg->blkg_list);
611 list_add(&blkg->q_node, &q->blkg_list);
612 spin_unlock(&blkcg->lock);
614 spin_lock(&alloc_list_lock);
615 list_add(&blkg->alloc_node, &alloc_list);
616 /* Queue per cpu stat allocation from worker thread. */
617 queue_delayed_work(system_nrt_wq, &blkio_stat_alloc_work, 0);
618 spin_unlock(&alloc_list_lock);
622 EXPORT_SYMBOL_GPL(blkg_lookup_create);
624 /* called under rcu_read_lock(). */
625 struct blkio_group *blkg_lookup(struct blkio_cgroup *blkcg,
626 struct request_queue *q)
628 struct blkio_group *blkg;
629 struct hlist_node *n;
631 hlist_for_each_entry_rcu(blkg, n, &blkcg->blkg_list, blkcg_node)
636 EXPORT_SYMBOL_GPL(blkg_lookup);
638 static void blkg_destroy(struct blkio_group *blkg)
640 struct request_queue *q = blkg->q;
641 struct blkio_cgroup *blkcg = blkg->blkcg;
643 lockdep_assert_held(q->queue_lock);
644 lockdep_assert_held(&blkcg->lock);
646 /* Something wrong if we are trying to remove same group twice */
647 WARN_ON_ONCE(list_empty(&blkg->q_node));
648 WARN_ON_ONCE(hlist_unhashed(&blkg->blkcg_node));
649 list_del_init(&blkg->q_node);
650 hlist_del_init_rcu(&blkg->blkcg_node);
652 spin_lock(&alloc_list_lock);
653 list_del_init(&blkg->alloc_node);
654 spin_unlock(&alloc_list_lock);
657 * Put the reference taken at the time of creation so that when all
658 * queues are gone, group can be destroyed.
664 * XXX: This updates blkg policy data in-place for root blkg, which is
665 * necessary across elevator switch and policy registration as root blkgs
666 * aren't shot down. This broken and racy implementation is temporary.
667 * Eventually, blkg shoot down will be replaced by proper in-place update.
669 void update_root_blkg_pd(struct request_queue *q, enum blkio_policy_id plid)
671 struct blkio_policy_type *pol = blkio_policy[plid];
672 struct blkio_group *blkg = blkg_lookup(&blkio_root_cgroup, q);
673 struct blkg_policy_data *pd;
678 kfree(blkg->pd[plid]);
679 blkg->pd[plid] = NULL;
684 pd = kzalloc(sizeof(*pd) + pol->pdata_size, GFP_KERNEL);
687 pd->stats_cpu = alloc_percpu(struct blkio_group_stats_cpu);
688 WARN_ON_ONCE(!pd->stats_cpu);
692 pol->ops.blkio_init_group_fn(blkg);
694 EXPORT_SYMBOL_GPL(update_root_blkg_pd);
697 * blkg_destroy_all - destroy all blkgs associated with a request_queue
698 * @q: request_queue of interest
699 * @destroy_root: whether to destroy root blkg or not
701 * Destroy blkgs associated with @q. If @destroy_root is %true, all are
702 * destroyed; otherwise, root blkg is left alone.
704 void blkg_destroy_all(struct request_queue *q, bool destroy_root)
706 struct blkio_group *blkg, *n;
708 spin_lock_irq(q->queue_lock);
710 list_for_each_entry_safe(blkg, n, &q->blkg_list, q_node) {
711 struct blkio_cgroup *blkcg = blkg->blkcg;
714 if (!destroy_root && blkg->blkcg == &blkio_root_cgroup)
717 spin_lock(&blkcg->lock);
719 spin_unlock(&blkcg->lock);
722 spin_unlock_irq(q->queue_lock);
724 EXPORT_SYMBOL_GPL(blkg_destroy_all);
726 static void blkg_rcu_free(struct rcu_head *rcu_head)
728 blkg_free(container_of(rcu_head, struct blkio_group, rcu_head));
731 void __blkg_release(struct blkio_group *blkg)
733 /* release the extra blkcg reference this blkg has been holding */
734 css_put(&blkg->blkcg->css);
737 * A group is freed in rcu manner. But having an rcu lock does not
738 * mean that one can access all the fields of blkg and assume these
739 * are valid. For example, don't try to follow throtl_data and
740 * request queue links.
742 * Having a reference to blkg under an rcu allows acess to only
743 * values local to groups like group stats and group rate limits
745 call_rcu(&blkg->rcu_head, blkg_rcu_free);
747 EXPORT_SYMBOL_GPL(__blkg_release);
749 static void blkio_reset_stats_cpu(struct blkio_group *blkg, int plid)
751 struct blkg_policy_data *pd = blkg->pd[plid];
754 if (pd->stats_cpu == NULL)
757 for_each_possible_cpu(cpu) {
758 struct blkio_group_stats_cpu *sc =
759 per_cpu_ptr(pd->stats_cpu, cpu);
762 memset(sc->stat_arr_cpu, 0, sizeof(sc->stat_arr_cpu));
767 blkiocg_reset_stats(struct cgroup *cgroup, struct cftype *cftype, u64 val)
769 struct blkio_cgroup *blkcg = cgroup_to_blkio_cgroup(cgroup);
770 struct blkio_group *blkg;
771 struct hlist_node *n;
774 spin_lock(&blkio_list_lock);
775 spin_lock_irq(&blkcg->lock);
778 * Note that stat reset is racy - it doesn't synchronize against
779 * stat updates. This is a debug feature which shouldn't exist
780 * anyway. If you get hit by a race, retry.
782 hlist_for_each_entry(blkg, n, &blkcg->blkg_list, blkcg_node) {
783 struct blkio_policy_type *pol;
785 list_for_each_entry(pol, &blkio_list, list) {
786 struct blkg_policy_data *pd = blkg->pd[pol->plid];
787 struct blkio_group_stats *stats = &pd->stats;
789 /* queued stats shouldn't be cleared */
790 for (i = 0; i < ARRAY_SIZE(stats->stat_arr); i++)
791 if (i != BLKIO_STAT_QUEUED)
792 memset(stats->stat_arr[i], 0,
793 sizeof(stats->stat_arr[i]));
795 #ifdef CONFIG_DEBUG_BLK_CGROUP
796 memset((void *)stats + BLKG_STATS_DEBUG_CLEAR_START, 0,
797 BLKG_STATS_DEBUG_CLEAR_SIZE);
799 blkio_reset_stats_cpu(blkg, pol->plid);
803 spin_unlock_irq(&blkcg->lock);
804 spin_unlock(&blkio_list_lock);
808 static void blkio_get_key_name(enum stat_sub_type type, const char *dname,
809 char *str, int chars_left, bool diskname_only)
811 snprintf(str, chars_left, "%s", dname);
812 chars_left -= strlen(str);
813 if (chars_left <= 0) {
815 "Possibly incorrect cgroup stat display format");
821 case BLKIO_STAT_READ:
822 strlcat(str, " Read", chars_left);
824 case BLKIO_STAT_WRITE:
825 strlcat(str, " Write", chars_left);
827 case BLKIO_STAT_SYNC:
828 strlcat(str, " Sync", chars_left);
830 case BLKIO_STAT_ASYNC:
831 strlcat(str, " Async", chars_left);
833 case BLKIO_STAT_TOTAL:
834 strlcat(str, " Total", chars_left);
837 strlcat(str, " Invalid", chars_left);
841 static uint64_t blkio_read_stat_cpu(struct blkio_group *blkg, int plid,
842 enum stat_type_cpu type, enum stat_sub_type sub_type)
844 struct blkg_policy_data *pd = blkg->pd[plid];
846 struct blkio_group_stats_cpu *stats_cpu;
849 if (pd->stats_cpu == NULL)
852 for_each_possible_cpu(cpu) {
854 stats_cpu = per_cpu_ptr(pd->stats_cpu, cpu);
857 start = u64_stats_fetch_begin(&stats_cpu->syncp);
858 if (type == BLKIO_STAT_CPU_SECTORS)
859 tval = stats_cpu->sectors;
861 tval = stats_cpu->stat_arr_cpu[type][sub_type];
862 } while(u64_stats_fetch_retry(&stats_cpu->syncp, start));
870 static uint64_t blkio_get_stat_cpu(struct blkio_group *blkg, int plid,
871 struct cgroup_map_cb *cb, const char *dname,
872 enum stat_type_cpu type)
874 uint64_t disk_total, val;
875 char key_str[MAX_KEY_LEN];
876 enum stat_sub_type sub_type;
878 if (type == BLKIO_STAT_CPU_SECTORS) {
879 val = blkio_read_stat_cpu(blkg, plid, type, 0);
880 blkio_get_key_name(0, dname, key_str, MAX_KEY_LEN, true);
881 cb->fill(cb, key_str, val);
885 for (sub_type = BLKIO_STAT_READ; sub_type < BLKIO_STAT_TOTAL;
887 blkio_get_key_name(sub_type, dname, key_str, MAX_KEY_LEN,
889 val = blkio_read_stat_cpu(blkg, plid, type, sub_type);
890 cb->fill(cb, key_str, val);
893 disk_total = blkio_read_stat_cpu(blkg, plid, type, BLKIO_STAT_READ) +
894 blkio_read_stat_cpu(blkg, plid, type, BLKIO_STAT_WRITE);
896 blkio_get_key_name(BLKIO_STAT_TOTAL, dname, key_str, MAX_KEY_LEN,
898 cb->fill(cb, key_str, disk_total);
902 static uint64_t blkio_get_stat(struct blkio_group *blkg, int plid,
903 struct cgroup_map_cb *cb, const char *dname,
906 struct blkio_group_stats *stats = &blkg->pd[plid]->stats;
907 uint64_t v = 0, disk_total = 0;
908 char key_str[MAX_KEY_LEN];
909 unsigned int sync_start;
912 if (type >= BLKIO_STAT_ARR_NR) {
914 sync_start = u64_stats_fetch_begin(&stats->syncp);
916 case BLKIO_STAT_TIME:
919 #ifdef CONFIG_DEBUG_BLK_CGROUP
920 case BLKIO_STAT_UNACCOUNTED_TIME:
921 v = stats->unaccounted_time;
923 case BLKIO_STAT_AVG_QUEUE_SIZE: {
924 uint64_t samples = stats->avg_queue_size_samples;
927 v = stats->avg_queue_size_sum;
932 case BLKIO_STAT_IDLE_TIME:
933 v = stats->idle_time;
935 case BLKIO_STAT_EMPTY_TIME:
936 v = stats->empty_time;
938 case BLKIO_STAT_DEQUEUE:
941 case BLKIO_STAT_GROUP_WAIT_TIME:
942 v = stats->group_wait_time;
948 } while (u64_stats_fetch_retry(&stats->syncp, sync_start));
950 blkio_get_key_name(0, dname, key_str, MAX_KEY_LEN, true);
951 cb->fill(cb, key_str, v);
955 for (st = BLKIO_STAT_READ; st < BLKIO_STAT_TOTAL; st++) {
957 sync_start = u64_stats_fetch_begin(&stats->syncp);
958 v = stats->stat_arr[type][st];
959 } while (u64_stats_fetch_retry(&stats->syncp, sync_start));
961 blkio_get_key_name(st, dname, key_str, MAX_KEY_LEN, false);
962 cb->fill(cb, key_str, v);
963 if (st == BLKIO_STAT_READ || st == BLKIO_STAT_WRITE)
967 blkio_get_key_name(BLKIO_STAT_TOTAL, dname, key_str, MAX_KEY_LEN,
969 cb->fill(cb, key_str, disk_total);
973 static int blkio_policy_parse_and_set(char *buf, enum blkio_policy_id plid,
974 int fileid, struct blkio_cgroup *blkcg)
976 struct gendisk *disk = NULL;
977 struct blkio_group *blkg = NULL;
978 struct blkg_policy_data *pd;
979 char *s[4], *p, *major_s = NULL, *minor_s = NULL;
980 unsigned long major, minor;
981 int i = 0, ret = -EINVAL;
986 memset(s, 0, sizeof(s));
988 while ((p = strsep(&buf, " ")) != NULL) {
994 /* Prevent from inputing too many things */
1002 p = strsep(&s[0], ":");
1012 if (strict_strtoul(major_s, 10, &major))
1015 if (strict_strtoul(minor_s, 10, &minor))
1018 dev = MKDEV(major, minor);
1020 if (strict_strtoull(s[1], 10, &temp))
1023 disk = get_gendisk(dev, &part);
1029 spin_lock_irq(disk->queue->queue_lock);
1030 blkg = blkg_lookup_create(blkcg, disk->queue, plid, false);
1031 spin_unlock_irq(disk->queue->queue_lock);
1034 ret = PTR_ERR(blkg);
1038 pd = blkg->pd[plid];
1041 case BLKIO_POLICY_PROP:
1042 if ((temp < BLKIO_WEIGHT_MIN && temp > 0) ||
1043 temp > BLKIO_WEIGHT_MAX)
1046 pd->conf.weight = temp;
1047 blkio_update_group_weight(blkg, plid, temp ?: blkcg->weight);
1049 case BLKIO_POLICY_THROTL:
1051 case BLKIO_THROTL_read_bps_device:
1052 pd->conf.bps[READ] = temp;
1053 blkio_update_group_bps(blkg, plid, temp ?: -1, fileid);
1055 case BLKIO_THROTL_write_bps_device:
1056 pd->conf.bps[WRITE] = temp;
1057 blkio_update_group_bps(blkg, plid, temp ?: -1, fileid);
1059 case BLKIO_THROTL_read_iops_device:
1060 if (temp > THROTL_IOPS_MAX)
1062 pd->conf.iops[READ] = temp;
1063 blkio_update_group_iops(blkg, plid, temp ?: -1, fileid);
1065 case BLKIO_THROTL_write_iops_device:
1066 if (temp > THROTL_IOPS_MAX)
1068 pd->conf.iops[WRITE] = temp;
1069 blkio_update_group_iops(blkg, plid, temp ?: -1, fileid);
1083 * If queue was bypassing, we should retry. Do so after a short
1084 * msleep(). It isn't strictly necessary but queue can be
1085 * bypassing for some time and it's always nice to avoid busy
1088 if (ret == -EBUSY) {
1090 return restart_syscall();
1095 static int blkiocg_file_write(struct cgroup *cgrp, struct cftype *cft,
1100 struct blkio_cgroup *blkcg = cgroup_to_blkio_cgroup(cgrp);
1101 enum blkio_policy_id plid = BLKIOFILE_POLICY(cft->private);
1102 int fileid = BLKIOFILE_ATTR(cft->private);
1104 buf = kstrdup(buffer, GFP_KERNEL);
1108 ret = blkio_policy_parse_and_set(buf, plid, fileid, blkcg);
1113 static const char *blkg_dev_name(struct blkio_group *blkg)
1115 /* some drivers (floppy) instantiate a queue w/o disk registered */
1116 if (blkg->q->backing_dev_info.dev)
1117 return dev_name(blkg->q->backing_dev_info.dev);
1121 static void blkio_print_group_conf(struct cftype *cft, struct blkio_group *blkg,
1124 int plid = BLKIOFILE_POLICY(cft->private);
1125 int fileid = BLKIOFILE_ATTR(cft->private);
1126 struct blkg_policy_data *pd = blkg->pd[plid];
1127 const char *dname = blkg_dev_name(blkg);
1134 case BLKIO_POLICY_PROP:
1135 if (pd->conf.weight)
1136 seq_printf(m, "%s\t%u\n",
1137 dname, pd->conf.weight);
1139 case BLKIO_POLICY_THROTL:
1141 case BLKIO_THROTL_read_bps_device:
1143 case BLKIO_THROTL_write_bps_device:
1144 if (pd->conf.bps[rw])
1145 seq_printf(m, "%s\t%llu\n",
1146 dname, pd->conf.bps[rw]);
1148 case BLKIO_THROTL_read_iops_device:
1150 case BLKIO_THROTL_write_iops_device:
1151 if (pd->conf.iops[rw])
1152 seq_printf(m, "%s\t%u\n",
1153 dname, pd->conf.iops[rw]);
1162 /* cgroup files which read their data from policy nodes end up here */
1163 static void blkio_read_conf(struct cftype *cft, struct blkio_cgroup *blkcg,
1166 struct blkio_group *blkg;
1167 struct hlist_node *n;
1169 spin_lock_irq(&blkcg->lock);
1170 hlist_for_each_entry(blkg, n, &blkcg->blkg_list, blkcg_node)
1171 blkio_print_group_conf(cft, blkg, m);
1172 spin_unlock_irq(&blkcg->lock);
1175 static int blkiocg_file_read(struct cgroup *cgrp, struct cftype *cft,
1178 struct blkio_cgroup *blkcg;
1179 enum blkio_policy_id plid = BLKIOFILE_POLICY(cft->private);
1180 int name = BLKIOFILE_ATTR(cft->private);
1182 blkcg = cgroup_to_blkio_cgroup(cgrp);
1185 case BLKIO_POLICY_PROP:
1187 case BLKIO_PROP_weight_device:
1188 blkio_read_conf(cft, blkcg, m);
1194 case BLKIO_POLICY_THROTL:
1196 case BLKIO_THROTL_read_bps_device:
1197 case BLKIO_THROTL_write_bps_device:
1198 case BLKIO_THROTL_read_iops_device:
1199 case BLKIO_THROTL_write_iops_device:
1200 blkio_read_conf(cft, blkcg, m);
1213 static int blkio_read_blkg_stats(struct blkio_cgroup *blkcg,
1214 struct cftype *cft, struct cgroup_map_cb *cb,
1215 enum stat_type type, bool show_total, bool pcpu)
1217 struct blkio_group *blkg;
1218 struct hlist_node *n;
1219 uint64_t cgroup_total = 0;
1221 spin_lock_irq(&blkcg->lock);
1223 hlist_for_each_entry(blkg, n, &blkcg->blkg_list, blkcg_node) {
1224 const char *dname = blkg_dev_name(blkg);
1225 int plid = BLKIOFILE_POLICY(cft->private);
1230 cgroup_total += blkio_get_stat_cpu(blkg, plid,
1233 cgroup_total += blkio_get_stat(blkg, plid,
1237 cb->fill(cb, "Total", cgroup_total);
1239 spin_unlock_irq(&blkcg->lock);
1243 /* All map kind of cgroup file get serviced by this function */
1244 static int blkiocg_file_read_map(struct cgroup *cgrp, struct cftype *cft,
1245 struct cgroup_map_cb *cb)
1247 struct blkio_cgroup *blkcg;
1248 enum blkio_policy_id plid = BLKIOFILE_POLICY(cft->private);
1249 int name = BLKIOFILE_ATTR(cft->private);
1251 blkcg = cgroup_to_blkio_cgroup(cgrp);
1254 case BLKIO_POLICY_PROP:
1256 case BLKIO_PROP_time:
1257 return blkio_read_blkg_stats(blkcg, cft, cb,
1258 BLKIO_STAT_TIME, 0, 0);
1259 case BLKIO_PROP_sectors:
1260 return blkio_read_blkg_stats(blkcg, cft, cb,
1261 BLKIO_STAT_CPU_SECTORS, 0, 1);
1262 case BLKIO_PROP_io_service_bytes:
1263 return blkio_read_blkg_stats(blkcg, cft, cb,
1264 BLKIO_STAT_CPU_SERVICE_BYTES, 1, 1);
1265 case BLKIO_PROP_io_serviced:
1266 return blkio_read_blkg_stats(blkcg, cft, cb,
1267 BLKIO_STAT_CPU_SERVICED, 1, 1);
1268 case BLKIO_PROP_io_service_time:
1269 return blkio_read_blkg_stats(blkcg, cft, cb,
1270 BLKIO_STAT_SERVICE_TIME, 1, 0);
1271 case BLKIO_PROP_io_wait_time:
1272 return blkio_read_blkg_stats(blkcg, cft, cb,
1273 BLKIO_STAT_WAIT_TIME, 1, 0);
1274 case BLKIO_PROP_io_merged:
1275 return blkio_read_blkg_stats(blkcg, cft, cb,
1276 BLKIO_STAT_MERGED, 1, 0);
1277 case BLKIO_PROP_io_queued:
1278 return blkio_read_blkg_stats(blkcg, cft, cb,
1279 BLKIO_STAT_QUEUED, 1, 0);
1280 #ifdef CONFIG_DEBUG_BLK_CGROUP
1281 case BLKIO_PROP_unaccounted_time:
1282 return blkio_read_blkg_stats(blkcg, cft, cb,
1283 BLKIO_STAT_UNACCOUNTED_TIME, 0, 0);
1284 case BLKIO_PROP_dequeue:
1285 return blkio_read_blkg_stats(blkcg, cft, cb,
1286 BLKIO_STAT_DEQUEUE, 0, 0);
1287 case BLKIO_PROP_avg_queue_size:
1288 return blkio_read_blkg_stats(blkcg, cft, cb,
1289 BLKIO_STAT_AVG_QUEUE_SIZE, 0, 0);
1290 case BLKIO_PROP_group_wait_time:
1291 return blkio_read_blkg_stats(blkcg, cft, cb,
1292 BLKIO_STAT_GROUP_WAIT_TIME, 0, 0);
1293 case BLKIO_PROP_idle_time:
1294 return blkio_read_blkg_stats(blkcg, cft, cb,
1295 BLKIO_STAT_IDLE_TIME, 0, 0);
1296 case BLKIO_PROP_empty_time:
1297 return blkio_read_blkg_stats(blkcg, cft, cb,
1298 BLKIO_STAT_EMPTY_TIME, 0, 0);
1304 case BLKIO_POLICY_THROTL:
1306 case BLKIO_THROTL_io_service_bytes:
1307 return blkio_read_blkg_stats(blkcg, cft, cb,
1308 BLKIO_STAT_CPU_SERVICE_BYTES, 1, 1);
1309 case BLKIO_THROTL_io_serviced:
1310 return blkio_read_blkg_stats(blkcg, cft, cb,
1311 BLKIO_STAT_CPU_SERVICED, 1, 1);
1323 static int blkio_weight_write(struct blkio_cgroup *blkcg, int plid, u64 val)
1325 struct blkio_group *blkg;
1326 struct hlist_node *n;
1328 if (val < BLKIO_WEIGHT_MIN || val > BLKIO_WEIGHT_MAX)
1331 spin_lock(&blkio_list_lock);
1332 spin_lock_irq(&blkcg->lock);
1333 blkcg->weight = (unsigned int)val;
1335 hlist_for_each_entry(blkg, n, &blkcg->blkg_list, blkcg_node) {
1336 struct blkg_policy_data *pd = blkg->pd[plid];
1338 if (!pd->conf.weight)
1339 blkio_update_group_weight(blkg, plid, blkcg->weight);
1342 spin_unlock_irq(&blkcg->lock);
1343 spin_unlock(&blkio_list_lock);
1347 static u64 blkiocg_file_read_u64 (struct cgroup *cgrp, struct cftype *cft) {
1348 struct blkio_cgroup *blkcg;
1349 enum blkio_policy_id plid = BLKIOFILE_POLICY(cft->private);
1350 int name = BLKIOFILE_ATTR(cft->private);
1352 blkcg = cgroup_to_blkio_cgroup(cgrp);
1355 case BLKIO_POLICY_PROP:
1357 case BLKIO_PROP_weight:
1358 return (u64)blkcg->weight;
1368 blkiocg_file_write_u64(struct cgroup *cgrp, struct cftype *cft, u64 val)
1370 struct blkio_cgroup *blkcg;
1371 enum blkio_policy_id plid = BLKIOFILE_POLICY(cft->private);
1372 int name = BLKIOFILE_ATTR(cft->private);
1374 blkcg = cgroup_to_blkio_cgroup(cgrp);
1377 case BLKIO_POLICY_PROP:
1379 case BLKIO_PROP_weight:
1380 return blkio_weight_write(blkcg, plid, val);
1390 struct cftype blkio_files[] = {
1392 .name = "weight_device",
1393 .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_PROP,
1394 BLKIO_PROP_weight_device),
1395 .read_seq_string = blkiocg_file_read,
1396 .write_string = blkiocg_file_write,
1397 .max_write_len = 256,
1401 .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_PROP,
1403 .read_u64 = blkiocg_file_read_u64,
1404 .write_u64 = blkiocg_file_write_u64,
1408 .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_PROP,
1410 .read_map = blkiocg_file_read_map,
1414 .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_PROP,
1415 BLKIO_PROP_sectors),
1416 .read_map = blkiocg_file_read_map,
1419 .name = "io_service_bytes",
1420 .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_PROP,
1421 BLKIO_PROP_io_service_bytes),
1422 .read_map = blkiocg_file_read_map,
1425 .name = "io_serviced",
1426 .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_PROP,
1427 BLKIO_PROP_io_serviced),
1428 .read_map = blkiocg_file_read_map,
1431 .name = "io_service_time",
1432 .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_PROP,
1433 BLKIO_PROP_io_service_time),
1434 .read_map = blkiocg_file_read_map,
1437 .name = "io_wait_time",
1438 .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_PROP,
1439 BLKIO_PROP_io_wait_time),
1440 .read_map = blkiocg_file_read_map,
1443 .name = "io_merged",
1444 .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_PROP,
1445 BLKIO_PROP_io_merged),
1446 .read_map = blkiocg_file_read_map,
1449 .name = "io_queued",
1450 .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_PROP,
1451 BLKIO_PROP_io_queued),
1452 .read_map = blkiocg_file_read_map,
1455 .name = "reset_stats",
1456 .write_u64 = blkiocg_reset_stats,
1458 #ifdef CONFIG_BLK_DEV_THROTTLING
1460 .name = "throttle.read_bps_device",
1461 .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_THROTL,
1462 BLKIO_THROTL_read_bps_device),
1463 .read_seq_string = blkiocg_file_read,
1464 .write_string = blkiocg_file_write,
1465 .max_write_len = 256,
1469 .name = "throttle.write_bps_device",
1470 .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_THROTL,
1471 BLKIO_THROTL_write_bps_device),
1472 .read_seq_string = blkiocg_file_read,
1473 .write_string = blkiocg_file_write,
1474 .max_write_len = 256,
1478 .name = "throttle.read_iops_device",
1479 .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_THROTL,
1480 BLKIO_THROTL_read_iops_device),
1481 .read_seq_string = blkiocg_file_read,
1482 .write_string = blkiocg_file_write,
1483 .max_write_len = 256,
1487 .name = "throttle.write_iops_device",
1488 .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_THROTL,
1489 BLKIO_THROTL_write_iops_device),
1490 .read_seq_string = blkiocg_file_read,
1491 .write_string = blkiocg_file_write,
1492 .max_write_len = 256,
1495 .name = "throttle.io_service_bytes",
1496 .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_THROTL,
1497 BLKIO_THROTL_io_service_bytes),
1498 .read_map = blkiocg_file_read_map,
1501 .name = "throttle.io_serviced",
1502 .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_THROTL,
1503 BLKIO_THROTL_io_serviced),
1504 .read_map = blkiocg_file_read_map,
1506 #endif /* CONFIG_BLK_DEV_THROTTLING */
1508 #ifdef CONFIG_DEBUG_BLK_CGROUP
1510 .name = "avg_queue_size",
1511 .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_PROP,
1512 BLKIO_PROP_avg_queue_size),
1513 .read_map = blkiocg_file_read_map,
1516 .name = "group_wait_time",
1517 .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_PROP,
1518 BLKIO_PROP_group_wait_time),
1519 .read_map = blkiocg_file_read_map,
1522 .name = "idle_time",
1523 .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_PROP,
1524 BLKIO_PROP_idle_time),
1525 .read_map = blkiocg_file_read_map,
1528 .name = "empty_time",
1529 .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_PROP,
1530 BLKIO_PROP_empty_time),
1531 .read_map = blkiocg_file_read_map,
1535 .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_PROP,
1536 BLKIO_PROP_dequeue),
1537 .read_map = blkiocg_file_read_map,
1540 .name = "unaccounted_time",
1541 .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_PROP,
1542 BLKIO_PROP_unaccounted_time),
1543 .read_map = blkiocg_file_read_map,
1550 * blkiocg_pre_destroy - cgroup pre_destroy callback
1551 * @cgroup: cgroup of interest
1553 * This function is called when @cgroup is about to go away and responsible
1554 * for shooting down all blkgs associated with @cgroup. blkgs should be
1555 * removed while holding both q and blkcg locks. As blkcg lock is nested
1556 * inside q lock, this function performs reverse double lock dancing.
1558 * This is the blkcg counterpart of ioc_release_fn().
1560 static int blkiocg_pre_destroy(struct cgroup *cgroup)
1562 struct blkio_cgroup *blkcg = cgroup_to_blkio_cgroup(cgroup);
1564 spin_lock_irq(&blkcg->lock);
1566 while (!hlist_empty(&blkcg->blkg_list)) {
1567 struct blkio_group *blkg = hlist_entry(blkcg->blkg_list.first,
1568 struct blkio_group, blkcg_node);
1569 struct request_queue *q = blkg->q;
1571 if (spin_trylock(q->queue_lock)) {
1573 spin_unlock(q->queue_lock);
1575 spin_unlock_irq(&blkcg->lock);
1577 spin_lock_irq(&blkcg->lock);
1581 spin_unlock_irq(&blkcg->lock);
1585 static void blkiocg_destroy(struct cgroup *cgroup)
1587 struct blkio_cgroup *blkcg = cgroup_to_blkio_cgroup(cgroup);
1589 if (blkcg != &blkio_root_cgroup)
1593 static struct cgroup_subsys_state *blkiocg_create(struct cgroup *cgroup)
1595 static atomic64_t id_seq = ATOMIC64_INIT(0);
1596 struct blkio_cgroup *blkcg;
1597 struct cgroup *parent = cgroup->parent;
1600 blkcg = &blkio_root_cgroup;
1604 blkcg = kzalloc(sizeof(*blkcg), GFP_KERNEL);
1606 return ERR_PTR(-ENOMEM);
1608 blkcg->weight = BLKIO_WEIGHT_DEFAULT;
1609 blkcg->id = atomic64_inc_return(&id_seq); /* root is 0, start from 1 */
1611 spin_lock_init(&blkcg->lock);
1612 INIT_HLIST_HEAD(&blkcg->blkg_list);
1618 * blkcg_init_queue - initialize blkcg part of request queue
1619 * @q: request_queue to initialize
1621 * Called from blk_alloc_queue_node(). Responsible for initializing blkcg
1622 * part of new request_queue @q.
1625 * 0 on success, -errno on failure.
1627 int blkcg_init_queue(struct request_queue *q)
1633 ret = blk_throtl_init(q);
1637 mutex_lock(&all_q_mutex);
1638 INIT_LIST_HEAD(&q->all_q_node);
1639 list_add_tail(&q->all_q_node, &all_q_list);
1640 mutex_unlock(&all_q_mutex);
1646 * blkcg_drain_queue - drain blkcg part of request_queue
1647 * @q: request_queue to drain
1649 * Called from blk_drain_queue(). Responsible for draining blkcg part.
1651 void blkcg_drain_queue(struct request_queue *q)
1653 lockdep_assert_held(q->queue_lock);
1655 blk_throtl_drain(q);
1659 * blkcg_exit_queue - exit and release blkcg part of request_queue
1660 * @q: request_queue being released
1662 * Called from blk_release_queue(). Responsible for exiting blkcg part.
1664 void blkcg_exit_queue(struct request_queue *q)
1666 mutex_lock(&all_q_mutex);
1667 list_del_init(&q->all_q_node);
1668 mutex_unlock(&all_q_mutex);
1670 blkg_destroy_all(q, true);
1676 * We cannot support shared io contexts, as we have no mean to support
1677 * two tasks with the same ioc in two different groups without major rework
1678 * of the main cic data structures. For now we allow a task to change
1679 * its cgroup only if it's the only owner of its ioc.
1681 static int blkiocg_can_attach(struct cgroup *cgrp, struct cgroup_taskset *tset)
1683 struct task_struct *task;
1684 struct io_context *ioc;
1687 /* task_lock() is needed to avoid races with exit_io_context() */
1688 cgroup_taskset_for_each(task, cgrp, tset) {
1690 ioc = task->io_context;
1691 if (ioc && atomic_read(&ioc->nr_tasks) > 1)
1700 static void blkcg_bypass_start(void)
1701 __acquires(&all_q_mutex)
1703 struct request_queue *q;
1705 mutex_lock(&all_q_mutex);
1707 list_for_each_entry(q, &all_q_list, all_q_node) {
1708 blk_queue_bypass_start(q);
1709 blkg_destroy_all(q, false);
1713 static void blkcg_bypass_end(void)
1714 __releases(&all_q_mutex)
1716 struct request_queue *q;
1718 list_for_each_entry(q, &all_q_list, all_q_node)
1719 blk_queue_bypass_end(q);
1721 mutex_unlock(&all_q_mutex);
1724 struct cgroup_subsys blkio_subsys = {
1726 .create = blkiocg_create,
1727 .can_attach = blkiocg_can_attach,
1728 .pre_destroy = blkiocg_pre_destroy,
1729 .destroy = blkiocg_destroy,
1730 .subsys_id = blkio_subsys_id,
1731 .base_cftypes = blkio_files,
1732 .module = THIS_MODULE,
1734 EXPORT_SYMBOL_GPL(blkio_subsys);
1736 void blkio_policy_register(struct blkio_policy_type *blkiop)
1738 struct request_queue *q;
1740 blkcg_bypass_start();
1741 spin_lock(&blkio_list_lock);
1743 BUG_ON(blkio_policy[blkiop->plid]);
1744 blkio_policy[blkiop->plid] = blkiop;
1745 list_add_tail(&blkiop->list, &blkio_list);
1747 spin_unlock(&blkio_list_lock);
1748 list_for_each_entry(q, &all_q_list, all_q_node)
1749 update_root_blkg_pd(q, blkiop->plid);
1752 EXPORT_SYMBOL_GPL(blkio_policy_register);
1754 void blkio_policy_unregister(struct blkio_policy_type *blkiop)
1756 struct request_queue *q;
1758 blkcg_bypass_start();
1759 spin_lock(&blkio_list_lock);
1761 BUG_ON(blkio_policy[blkiop->plid] != blkiop);
1762 blkio_policy[blkiop->plid] = NULL;
1763 list_del_init(&blkiop->list);
1765 spin_unlock(&blkio_list_lock);
1766 list_for_each_entry(q, &all_q_list, all_q_node)
1767 update_root_blkg_pd(q, blkiop->plid);
1770 EXPORT_SYMBOL_GPL(blkio_policy_unregister);