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/kdev_t.h>
15 #include <linux/module.h>
16 #include <linux/err.h>
17 #include <linux/blkdev.h>
18 #include <linux/slab.h>
19 #include <linux/genhd.h>
20 #include <linux/delay.h>
21 #include <linux/atomic.h>
22 #include "blk-cgroup.h"
25 #define MAX_KEY_LEN 100
27 static DEFINE_MUTEX(blkcg_pol_mutex);
29 struct blkcg blkcg_root = { .cfq_weight = 2 * CFQ_WEIGHT_DEFAULT,
30 .cfq_leaf_weight = 2 * CFQ_WEIGHT_DEFAULT, };
31 EXPORT_SYMBOL_GPL(blkcg_root);
33 static struct blkcg_policy *blkcg_policy[BLKCG_MAX_POLS];
35 static bool blkcg_policy_enabled(struct request_queue *q,
36 const struct blkcg_policy *pol)
38 return pol && test_bit(pol->plid, q->blkcg_pols);
42 * blkg_free - free a blkg
45 * Free @blkg which may be partially allocated.
47 static void blkg_free(struct blkcg_gq *blkg)
54 for (i = 0; i < BLKCG_MAX_POLS; i++)
57 blk_exit_rl(&blkg->rl);
62 * blkg_alloc - allocate a blkg
63 * @blkcg: block cgroup the new blkg is associated with
64 * @q: request_queue the new blkg is associated with
65 * @gfp_mask: allocation mask to use
67 * Allocate a new blkg assocating @blkcg and @q.
69 static struct blkcg_gq *blkg_alloc(struct blkcg *blkcg, struct request_queue *q,
72 struct blkcg_gq *blkg;
75 /* alloc and init base part */
76 blkg = kzalloc_node(sizeof(*blkg), gfp_mask, q->node);
81 INIT_LIST_HEAD(&blkg->q_node);
83 atomic_set(&blkg->refcnt, 1);
85 /* root blkg uses @q->root_rl, init rl only for !root blkgs */
86 if (blkcg != &blkcg_root) {
87 if (blk_init_rl(&blkg->rl, q, gfp_mask))
92 for (i = 0; i < BLKCG_MAX_POLS; i++) {
93 struct blkcg_policy *pol = blkcg_policy[i];
94 struct blkg_policy_data *pd;
96 if (!blkcg_policy_enabled(q, pol))
99 /* alloc per-policy data and attach it to blkg */
100 pd = kzalloc_node(pol->pd_size, gfp_mask, q->node);
117 * __blkg_lookup - internal version of blkg_lookup()
118 * @blkcg: blkcg of interest
119 * @q: request_queue of interest
120 * @update_hint: whether to update lookup hint with the result or not
122 * This is internal version and shouldn't be used by policy
123 * implementations. Looks up blkgs for the @blkcg - @q pair regardless of
124 * @q's bypass state. If @update_hint is %true, the caller should be
125 * holding @q->queue_lock and lookup hint is updated on success.
127 struct blkcg_gq *__blkg_lookup(struct blkcg *blkcg, struct request_queue *q,
130 struct blkcg_gq *blkg;
132 blkg = rcu_dereference(blkcg->blkg_hint);
133 if (blkg && blkg->q == q)
137 * Hint didn't match. Look up from the radix tree. Note that the
138 * hint can only be updated under queue_lock as otherwise @blkg
139 * could have already been removed from blkg_tree. The caller is
140 * responsible for grabbing queue_lock if @update_hint.
142 blkg = radix_tree_lookup(&blkcg->blkg_tree, q->id);
143 if (blkg && blkg->q == q) {
145 lockdep_assert_held(q->queue_lock);
146 rcu_assign_pointer(blkcg->blkg_hint, blkg);
155 * blkg_lookup - lookup blkg for the specified blkcg - q pair
156 * @blkcg: blkcg of interest
157 * @q: request_queue of interest
159 * Lookup blkg for the @blkcg - @q pair. This function should be called
160 * under RCU read lock and is guaranteed to return %NULL if @q is bypassing
161 * - see blk_queue_bypass_start() for details.
163 struct blkcg_gq *blkg_lookup(struct blkcg *blkcg, struct request_queue *q)
165 WARN_ON_ONCE(!rcu_read_lock_held());
167 if (unlikely(blk_queue_bypass(q)))
169 return __blkg_lookup(blkcg, q, false);
171 EXPORT_SYMBOL_GPL(blkg_lookup);
174 * If @new_blkg is %NULL, this function tries to allocate a new one as
175 * necessary using %GFP_ATOMIC. @new_blkg is always consumed on return.
177 static struct blkcg_gq *blkg_create(struct blkcg *blkcg,
178 struct request_queue *q,
179 struct blkcg_gq *new_blkg)
181 struct blkcg_gq *blkg;
184 WARN_ON_ONCE(!rcu_read_lock_held());
185 lockdep_assert_held(q->queue_lock);
187 /* blkg holds a reference to blkcg */
188 if (!css_tryget_online(&blkcg->css)) {
195 new_blkg = blkg_alloc(blkcg, q, GFP_ATOMIC);
196 if (unlikely(!new_blkg)) {
204 if (blkcg_parent(blkcg)) {
205 blkg->parent = __blkg_lookup(blkcg_parent(blkcg), q, false);
206 if (WARN_ON_ONCE(!blkg->parent)) {
210 blkg_get(blkg->parent);
213 /* invoke per-policy init */
214 for (i = 0; i < BLKCG_MAX_POLS; i++) {
215 struct blkcg_policy *pol = blkcg_policy[i];
217 if (blkg->pd[i] && pol->pd_init_fn)
218 pol->pd_init_fn(blkg);
222 spin_lock(&blkcg->lock);
223 ret = radix_tree_insert(&blkcg->blkg_tree, q->id, blkg);
225 hlist_add_head_rcu(&blkg->blkcg_node, &blkcg->blkg_list);
226 list_add(&blkg->q_node, &q->blkg_list);
228 for (i = 0; i < BLKCG_MAX_POLS; i++) {
229 struct blkcg_policy *pol = blkcg_policy[i];
231 if (blkg->pd[i] && pol->pd_online_fn)
232 pol->pd_online_fn(blkg);
236 spin_unlock(&blkcg->lock);
239 if (blkcg == &blkcg_root) {
241 q->root_rl.blkg = blkg;
246 /* @blkg failed fully initialized, use the usual release path */
251 css_put(&blkcg->css);
258 * blkg_lookup_create - lookup blkg, try to create one if not there
259 * @blkcg: blkcg of interest
260 * @q: request_queue of interest
262 * Lookup blkg for the @blkcg - @q pair. If it doesn't exist, try to
263 * create one. blkg creation is performed recursively from blkcg_root such
264 * that all non-root blkg's have access to the parent blkg. This function
265 * should be called under RCU read lock and @q->queue_lock.
267 * Returns pointer to the looked up or created blkg on success, ERR_PTR()
268 * value on error. If @q is dead, returns ERR_PTR(-EINVAL). If @q is not
269 * dead and bypassing, returns ERR_PTR(-EBUSY).
271 struct blkcg_gq *blkg_lookup_create(struct blkcg *blkcg,
272 struct request_queue *q)
274 struct blkcg_gq *blkg;
276 WARN_ON_ONCE(!rcu_read_lock_held());
277 lockdep_assert_held(q->queue_lock);
280 * This could be the first entry point of blkcg implementation and
281 * we shouldn't allow anything to go through for a bypassing queue.
283 if (unlikely(blk_queue_bypass(q)))
284 return ERR_PTR(blk_queue_dying(q) ? -EINVAL : -EBUSY);
286 blkg = __blkg_lookup(blkcg, q, true);
291 * Create blkgs walking down from blkcg_root to @blkcg, so that all
292 * non-root blkgs have access to their parents.
295 struct blkcg *pos = blkcg;
296 struct blkcg *parent = blkcg_parent(blkcg);
298 while (parent && !__blkg_lookup(parent, q, false)) {
300 parent = blkcg_parent(parent);
303 blkg = blkg_create(pos, q, NULL);
304 if (pos == blkcg || IS_ERR(blkg))
308 EXPORT_SYMBOL_GPL(blkg_lookup_create);
310 static void blkg_destroy(struct blkcg_gq *blkg)
312 struct blkcg *blkcg = blkg->blkcg;
315 lockdep_assert_held(blkg->q->queue_lock);
316 lockdep_assert_held(&blkcg->lock);
318 /* Something wrong if we are trying to remove same group twice */
319 WARN_ON_ONCE(list_empty(&blkg->q_node));
320 WARN_ON_ONCE(hlist_unhashed(&blkg->blkcg_node));
322 for (i = 0; i < BLKCG_MAX_POLS; i++) {
323 struct blkcg_policy *pol = blkcg_policy[i];
325 if (blkg->pd[i] && pol->pd_offline_fn)
326 pol->pd_offline_fn(blkg);
328 blkg->online = false;
330 radix_tree_delete(&blkcg->blkg_tree, blkg->q->id);
331 list_del_init(&blkg->q_node);
332 hlist_del_init_rcu(&blkg->blkcg_node);
335 * Both setting lookup hint to and clearing it from @blkg are done
336 * under queue_lock. If it's not pointing to @blkg now, it never
337 * will. Hint assignment itself can race safely.
339 if (rcu_access_pointer(blkcg->blkg_hint) == blkg)
340 rcu_assign_pointer(blkcg->blkg_hint, NULL);
343 * If root blkg is destroyed. Just clear the pointer since root_rl
344 * does not take reference on root blkg.
346 if (blkcg == &blkcg_root) {
347 blkg->q->root_blkg = NULL;
348 blkg->q->root_rl.blkg = NULL;
352 * Put the reference taken at the time of creation so that when all
353 * queues are gone, group can be destroyed.
359 * blkg_destroy_all - destroy all blkgs associated with a request_queue
360 * @q: request_queue of interest
362 * Destroy all blkgs associated with @q.
364 static void blkg_destroy_all(struct request_queue *q)
366 struct blkcg_gq *blkg, *n;
368 lockdep_assert_held(q->queue_lock);
370 list_for_each_entry_safe(blkg, n, &q->blkg_list, q_node) {
371 struct blkcg *blkcg = blkg->blkcg;
373 spin_lock(&blkcg->lock);
375 spin_unlock(&blkcg->lock);
380 * A group is RCU protected, but having an rcu lock does not mean that one
381 * can access all the fields of blkg and assume these are valid. For
382 * example, don't try to follow throtl_data and request queue links.
384 * Having a reference to blkg under an rcu allows accesses to only values
385 * local to groups like group stats and group rate limits.
387 void __blkg_release_rcu(struct rcu_head *rcu_head)
389 struct blkcg_gq *blkg = container_of(rcu_head, struct blkcg_gq, rcu_head);
392 /* tell policies that this one is being freed */
393 for (i = 0; i < BLKCG_MAX_POLS; i++) {
394 struct blkcg_policy *pol = blkcg_policy[i];
396 if (blkg->pd[i] && pol->pd_exit_fn)
397 pol->pd_exit_fn(blkg);
400 /* release the blkcg and parent blkg refs this blkg has been holding */
401 css_put(&blkg->blkcg->css);
403 blkg_put(blkg->parent);
407 EXPORT_SYMBOL_GPL(__blkg_release_rcu);
410 * The next function used by blk_queue_for_each_rl(). It's a bit tricky
411 * because the root blkg uses @q->root_rl instead of its own rl.
413 struct request_list *__blk_queue_next_rl(struct request_list *rl,
414 struct request_queue *q)
416 struct list_head *ent;
417 struct blkcg_gq *blkg;
420 * Determine the current blkg list_head. The first entry is
421 * root_rl which is off @q->blkg_list and mapped to the head.
423 if (rl == &q->root_rl) {
425 /* There are no more block groups, hence no request lists */
429 blkg = container_of(rl, struct blkcg_gq, rl);
433 /* walk to the next list_head, skip root blkcg */
435 if (ent == &q->root_blkg->q_node)
437 if (ent == &q->blkg_list)
440 blkg = container_of(ent, struct blkcg_gq, q_node);
444 static int blkcg_reset_stats(struct cgroup_subsys_state *css,
445 struct cftype *cftype, u64 val)
447 struct blkcg *blkcg = css_to_blkcg(css);
448 struct blkcg_gq *blkg;
452 * XXX: We invoke cgroup_add/rm_cftypes() under blkcg_pol_mutex
453 * which ends up putting cgroup's internal cgroup_tree_mutex under
454 * it; however, cgroup_tree_mutex is nested above cgroup file
455 * active protection and grabbing blkcg_pol_mutex from a cgroup
456 * file operation creates a possible circular dependency. cgroup
457 * internal locking is planned to go through further simplification
458 * and this issue should go away soon. For now, let's trylock
459 * blkcg_pol_mutex and restart the write on failure.
461 * http://lkml.kernel.org/g/5363C04B.4010400@oracle.com
463 if (!mutex_trylock(&blkcg_pol_mutex))
464 return restart_syscall();
465 spin_lock_irq(&blkcg->lock);
468 * Note that stat reset is racy - it doesn't synchronize against
469 * stat updates. This is a debug feature which shouldn't exist
470 * anyway. If you get hit by a race, retry.
472 hlist_for_each_entry(blkg, &blkcg->blkg_list, blkcg_node) {
473 for (i = 0; i < BLKCG_MAX_POLS; i++) {
474 struct blkcg_policy *pol = blkcg_policy[i];
476 if (blkcg_policy_enabled(blkg->q, pol) &&
477 pol->pd_reset_stats_fn)
478 pol->pd_reset_stats_fn(blkg);
482 spin_unlock_irq(&blkcg->lock);
483 mutex_unlock(&blkcg_pol_mutex);
487 static const char *blkg_dev_name(struct blkcg_gq *blkg)
489 /* some drivers (floppy) instantiate a queue w/o disk registered */
490 if (blkg->q->backing_dev_info.dev)
491 return dev_name(blkg->q->backing_dev_info.dev);
496 * blkcg_print_blkgs - helper for printing per-blkg data
497 * @sf: seq_file to print to
498 * @blkcg: blkcg of interest
499 * @prfill: fill function to print out a blkg
500 * @pol: policy in question
501 * @data: data to be passed to @prfill
502 * @show_total: to print out sum of prfill return values or not
504 * This function invokes @prfill on each blkg of @blkcg if pd for the
505 * policy specified by @pol exists. @prfill is invoked with @sf, the
506 * policy data and @data and the matching queue lock held. If @show_total
507 * is %true, the sum of the return values from @prfill is printed with
508 * "Total" label at the end.
510 * This is to be used to construct print functions for
511 * cftype->read_seq_string method.
513 void blkcg_print_blkgs(struct seq_file *sf, struct blkcg *blkcg,
514 u64 (*prfill)(struct seq_file *,
515 struct blkg_policy_data *, int),
516 const struct blkcg_policy *pol, int data,
519 struct blkcg_gq *blkg;
523 hlist_for_each_entry_rcu(blkg, &blkcg->blkg_list, blkcg_node) {
524 spin_lock_irq(blkg->q->queue_lock);
525 if (blkcg_policy_enabled(blkg->q, pol))
526 total += prfill(sf, blkg->pd[pol->plid], data);
527 spin_unlock_irq(blkg->q->queue_lock);
532 seq_printf(sf, "Total %llu\n", (unsigned long long)total);
534 EXPORT_SYMBOL_GPL(blkcg_print_blkgs);
537 * __blkg_prfill_u64 - prfill helper for a single u64 value
538 * @sf: seq_file to print to
539 * @pd: policy private data of interest
542 * Print @v to @sf for the device assocaited with @pd.
544 u64 __blkg_prfill_u64(struct seq_file *sf, struct blkg_policy_data *pd, u64 v)
546 const char *dname = blkg_dev_name(pd->blkg);
551 seq_printf(sf, "%s %llu\n", dname, (unsigned long long)v);
554 EXPORT_SYMBOL_GPL(__blkg_prfill_u64);
557 * __blkg_prfill_rwstat - prfill helper for a blkg_rwstat
558 * @sf: seq_file to print to
559 * @pd: policy private data of interest
560 * @rwstat: rwstat to print
562 * Print @rwstat to @sf for the device assocaited with @pd.
564 u64 __blkg_prfill_rwstat(struct seq_file *sf, struct blkg_policy_data *pd,
565 const struct blkg_rwstat *rwstat)
567 static const char *rwstr[] = {
568 [BLKG_RWSTAT_READ] = "Read",
569 [BLKG_RWSTAT_WRITE] = "Write",
570 [BLKG_RWSTAT_SYNC] = "Sync",
571 [BLKG_RWSTAT_ASYNC] = "Async",
573 const char *dname = blkg_dev_name(pd->blkg);
580 for (i = 0; i < BLKG_RWSTAT_NR; i++)
581 seq_printf(sf, "%s %s %llu\n", dname, rwstr[i],
582 (unsigned long long)rwstat->cnt[i]);
584 v = rwstat->cnt[BLKG_RWSTAT_READ] + rwstat->cnt[BLKG_RWSTAT_WRITE];
585 seq_printf(sf, "%s Total %llu\n", dname, (unsigned long long)v);
588 EXPORT_SYMBOL_GPL(__blkg_prfill_rwstat);
591 * blkg_prfill_stat - prfill callback for blkg_stat
592 * @sf: seq_file to print to
593 * @pd: policy private data of interest
594 * @off: offset to the blkg_stat in @pd
596 * prfill callback for printing a blkg_stat.
598 u64 blkg_prfill_stat(struct seq_file *sf, struct blkg_policy_data *pd, int off)
600 return __blkg_prfill_u64(sf, pd, blkg_stat_read((void *)pd + off));
602 EXPORT_SYMBOL_GPL(blkg_prfill_stat);
605 * blkg_prfill_rwstat - prfill callback for blkg_rwstat
606 * @sf: seq_file to print to
607 * @pd: policy private data of interest
608 * @off: offset to the blkg_rwstat in @pd
610 * prfill callback for printing a blkg_rwstat.
612 u64 blkg_prfill_rwstat(struct seq_file *sf, struct blkg_policy_data *pd,
615 struct blkg_rwstat rwstat = blkg_rwstat_read((void *)pd + off);
617 return __blkg_prfill_rwstat(sf, pd, &rwstat);
619 EXPORT_SYMBOL_GPL(blkg_prfill_rwstat);
622 * blkg_stat_recursive_sum - collect hierarchical blkg_stat
623 * @pd: policy private data of interest
624 * @off: offset to the blkg_stat in @pd
626 * Collect the blkg_stat specified by @off from @pd and all its online
627 * descendants and return the sum. The caller must be holding the queue
628 * lock for online tests.
630 u64 blkg_stat_recursive_sum(struct blkg_policy_data *pd, int off)
632 struct blkcg_policy *pol = blkcg_policy[pd->plid];
633 struct blkcg_gq *pos_blkg;
634 struct cgroup_subsys_state *pos_css;
637 lockdep_assert_held(pd->blkg->q->queue_lock);
640 blkg_for_each_descendant_pre(pos_blkg, pos_css, pd_to_blkg(pd)) {
641 struct blkg_policy_data *pos_pd = blkg_to_pd(pos_blkg, pol);
642 struct blkg_stat *stat = (void *)pos_pd + off;
644 if (pos_blkg->online)
645 sum += blkg_stat_read(stat);
651 EXPORT_SYMBOL_GPL(blkg_stat_recursive_sum);
654 * blkg_rwstat_recursive_sum - collect hierarchical blkg_rwstat
655 * @pd: policy private data of interest
656 * @off: offset to the blkg_stat in @pd
658 * Collect the blkg_rwstat specified by @off from @pd and all its online
659 * descendants and return the sum. The caller must be holding the queue
660 * lock for online tests.
662 struct blkg_rwstat blkg_rwstat_recursive_sum(struct blkg_policy_data *pd,
665 struct blkcg_policy *pol = blkcg_policy[pd->plid];
666 struct blkcg_gq *pos_blkg;
667 struct cgroup_subsys_state *pos_css;
668 struct blkg_rwstat sum = { };
671 lockdep_assert_held(pd->blkg->q->queue_lock);
674 blkg_for_each_descendant_pre(pos_blkg, pos_css, pd_to_blkg(pd)) {
675 struct blkg_policy_data *pos_pd = blkg_to_pd(pos_blkg, pol);
676 struct blkg_rwstat *rwstat = (void *)pos_pd + off;
677 struct blkg_rwstat tmp;
679 if (!pos_blkg->online)
682 tmp = blkg_rwstat_read(rwstat);
684 for (i = 0; i < BLKG_RWSTAT_NR; i++)
685 sum.cnt[i] += tmp.cnt[i];
691 EXPORT_SYMBOL_GPL(blkg_rwstat_recursive_sum);
694 * blkg_conf_prep - parse and prepare for per-blkg config update
695 * @blkcg: target block cgroup
696 * @pol: target policy
697 * @input: input string
698 * @ctx: blkg_conf_ctx to be filled
700 * Parse per-blkg config update from @input and initialize @ctx with the
701 * result. @ctx->blkg points to the blkg to be updated and @ctx->v the new
702 * value. This function returns with RCU read lock and queue lock held and
703 * must be paired with blkg_conf_finish().
705 int blkg_conf_prep(struct blkcg *blkcg, const struct blkcg_policy *pol,
706 const char *input, struct blkg_conf_ctx *ctx)
707 __acquires(rcu) __acquires(disk->queue->queue_lock)
709 struct gendisk *disk;
710 struct blkcg_gq *blkg;
711 unsigned int major, minor;
712 unsigned long long v;
715 if (sscanf(input, "%u:%u %llu", &major, &minor, &v) != 3)
718 disk = get_gendisk(MKDEV(major, minor), &part);
723 spin_lock_irq(disk->queue->queue_lock);
725 if (blkcg_policy_enabled(disk->queue, pol))
726 blkg = blkg_lookup_create(blkcg, disk->queue);
728 blkg = ERR_PTR(-EINVAL);
733 spin_unlock_irq(disk->queue->queue_lock);
736 * If queue was bypassing, we should retry. Do so after a
737 * short msleep(). It isn't strictly necessary but queue
738 * can be bypassing for some time and it's always nice to
739 * avoid busy looping.
743 ret = restart_syscall();
753 EXPORT_SYMBOL_GPL(blkg_conf_prep);
756 * blkg_conf_finish - finish up per-blkg config update
757 * @ctx: blkg_conf_ctx intiailized by blkg_conf_prep()
759 * Finish up after per-blkg config update. This function must be paired
760 * with blkg_conf_prep().
762 void blkg_conf_finish(struct blkg_conf_ctx *ctx)
763 __releases(ctx->disk->queue->queue_lock) __releases(rcu)
765 spin_unlock_irq(ctx->disk->queue->queue_lock);
769 EXPORT_SYMBOL_GPL(blkg_conf_finish);
771 struct cftype blkcg_files[] = {
773 .name = "reset_stats",
774 .write_u64 = blkcg_reset_stats,
780 * blkcg_css_offline - cgroup css_offline callback
781 * @css: css of interest
783 * This function is called when @css is about to go away and responsible
784 * for shooting down all blkgs associated with @css. blkgs should be
785 * removed while holding both q and blkcg locks. As blkcg lock is nested
786 * inside q lock, this function performs reverse double lock dancing.
788 * This is the blkcg counterpart of ioc_release_fn().
790 static void blkcg_css_offline(struct cgroup_subsys_state *css)
792 struct blkcg *blkcg = css_to_blkcg(css);
794 spin_lock_irq(&blkcg->lock);
796 while (!hlist_empty(&blkcg->blkg_list)) {
797 struct blkcg_gq *blkg = hlist_entry(blkcg->blkg_list.first,
798 struct blkcg_gq, blkcg_node);
799 struct request_queue *q = blkg->q;
801 if (spin_trylock(q->queue_lock)) {
803 spin_unlock(q->queue_lock);
805 spin_unlock_irq(&blkcg->lock);
807 spin_lock_irq(&blkcg->lock);
811 spin_unlock_irq(&blkcg->lock);
814 static void blkcg_css_free(struct cgroup_subsys_state *css)
816 struct blkcg *blkcg = css_to_blkcg(css);
818 if (blkcg != &blkcg_root)
822 static struct cgroup_subsys_state *
823 blkcg_css_alloc(struct cgroup_subsys_state *parent_css)
825 static atomic64_t id_seq = ATOMIC64_INIT(0);
833 blkcg = kzalloc(sizeof(*blkcg), GFP_KERNEL);
835 return ERR_PTR(-ENOMEM);
837 blkcg->cfq_weight = CFQ_WEIGHT_DEFAULT;
838 blkcg->cfq_leaf_weight = CFQ_WEIGHT_DEFAULT;
839 blkcg->id = atomic64_inc_return(&id_seq); /* root is 0, start from 1 */
841 spin_lock_init(&blkcg->lock);
842 INIT_RADIX_TREE(&blkcg->blkg_tree, GFP_ATOMIC);
843 INIT_HLIST_HEAD(&blkcg->blkg_list);
849 * blkcg_init_queue - initialize blkcg part of request queue
850 * @q: request_queue to initialize
852 * Called from blk_alloc_queue_node(). Responsible for initializing blkcg
853 * part of new request_queue @q.
856 * 0 on success, -errno on failure.
858 int blkcg_init_queue(struct request_queue *q)
862 return blk_throtl_init(q);
866 * blkcg_drain_queue - drain blkcg part of request_queue
867 * @q: request_queue to drain
869 * Called from blk_drain_queue(). Responsible for draining blkcg part.
871 void blkcg_drain_queue(struct request_queue *q)
873 lockdep_assert_held(q->queue_lock);
876 * @q could be exiting and already have destroyed all blkgs as
877 * indicated by NULL root_blkg. If so, don't confuse policies.
886 * blkcg_exit_queue - exit and release blkcg part of request_queue
887 * @q: request_queue being released
889 * Called from blk_release_queue(). Responsible for exiting blkcg part.
891 void blkcg_exit_queue(struct request_queue *q)
893 spin_lock_irq(q->queue_lock);
895 spin_unlock_irq(q->queue_lock);
901 * We cannot support shared io contexts, as we have no mean to support
902 * two tasks with the same ioc in two different groups without major rework
903 * of the main cic data structures. For now we allow a task to change
904 * its cgroup only if it's the only owner of its ioc.
906 static int blkcg_can_attach(struct cgroup_subsys_state *css,
907 struct cgroup_taskset *tset)
909 struct task_struct *task;
910 struct io_context *ioc;
913 /* task_lock() is needed to avoid races with exit_io_context() */
914 cgroup_taskset_for_each(task, tset) {
916 ioc = task->io_context;
917 if (ioc && atomic_read(&ioc->nr_tasks) > 1)
926 struct cgroup_subsys blkio_cgrp_subsys = {
927 .css_alloc = blkcg_css_alloc,
928 .css_offline = blkcg_css_offline,
929 .css_free = blkcg_css_free,
930 .can_attach = blkcg_can_attach,
931 .legacy_cftypes = blkcg_files,
934 * This ensures that, if available, memcg is automatically enabled
935 * together on the default hierarchy so that the owner cgroup can
936 * be retrieved from writeback pages.
938 .depends_on = 1 << memory_cgrp_id,
941 EXPORT_SYMBOL_GPL(blkio_cgrp_subsys);
944 * blkcg_activate_policy - activate a blkcg policy on a request_queue
945 * @q: request_queue of interest
946 * @pol: blkcg policy to activate
948 * Activate @pol on @q. Requires %GFP_KERNEL context. @q goes through
949 * bypass mode to populate its blkgs with policy_data for @pol.
951 * Activation happens with @q bypassed, so nobody would be accessing blkgs
952 * from IO path. Update of each blkg is protected by both queue and blkcg
953 * locks so that holding either lock and testing blkcg_policy_enabled() is
954 * always enough for dereferencing policy data.
956 * The caller is responsible for synchronizing [de]activations and policy
957 * [un]registerations. Returns 0 on success, -errno on failure.
959 int blkcg_activate_policy(struct request_queue *q,
960 const struct blkcg_policy *pol)
963 struct blkcg_gq *blkg, *new_blkg;
964 struct blkg_policy_data *pd, *n;
968 if (blkcg_policy_enabled(q, pol))
971 /* preallocations for root blkg */
972 new_blkg = blkg_alloc(&blkcg_root, q, GFP_KERNEL);
976 blk_queue_bypass_start(q);
978 preloaded = !radix_tree_preload(GFP_KERNEL);
981 * Make sure the root blkg exists and count the existing blkgs. As
982 * @q is bypassing at this point, blkg_lookup_create() can't be
983 * used. Open code it.
985 spin_lock_irq(q->queue_lock);
988 blkg = __blkg_lookup(&blkcg_root, q, false);
992 blkg = blkg_create(&blkcg_root, q, new_blkg);
996 radix_tree_preload_end();
1003 list_for_each_entry(blkg, &q->blkg_list, q_node)
1006 spin_unlock_irq(q->queue_lock);
1008 /* allocate policy_data for all existing blkgs */
1010 pd = kzalloc_node(pol->pd_size, GFP_KERNEL, q->node);
1015 list_add_tail(&pd->alloc_node, &pds);
1019 * Install the allocated pds. With @q bypassing, no new blkg
1020 * should have been created while the queue lock was dropped.
1022 spin_lock_irq(q->queue_lock);
1024 list_for_each_entry(blkg, &q->blkg_list, q_node) {
1025 if (WARN_ON(list_empty(&pds))) {
1026 /* umm... this shouldn't happen, just abort */
1030 pd = list_first_entry(&pds, struct blkg_policy_data, alloc_node);
1031 list_del_init(&pd->alloc_node);
1033 /* grab blkcg lock too while installing @pd on @blkg */
1034 spin_lock(&blkg->blkcg->lock);
1036 blkg->pd[pol->plid] = pd;
1038 pd->plid = pol->plid;
1039 pol->pd_init_fn(blkg);
1041 spin_unlock(&blkg->blkcg->lock);
1044 __set_bit(pol->plid, q->blkcg_pols);
1047 spin_unlock_irq(q->queue_lock);
1049 blk_queue_bypass_end(q);
1050 list_for_each_entry_safe(pd, n, &pds, alloc_node)
1054 EXPORT_SYMBOL_GPL(blkcg_activate_policy);
1057 * blkcg_deactivate_policy - deactivate a blkcg policy on a request_queue
1058 * @q: request_queue of interest
1059 * @pol: blkcg policy to deactivate
1061 * Deactivate @pol on @q. Follows the same synchronization rules as
1062 * blkcg_activate_policy().
1064 void blkcg_deactivate_policy(struct request_queue *q,
1065 const struct blkcg_policy *pol)
1067 struct blkcg_gq *blkg;
1069 if (!blkcg_policy_enabled(q, pol))
1072 blk_queue_bypass_start(q);
1073 spin_lock_irq(q->queue_lock);
1075 __clear_bit(pol->plid, q->blkcg_pols);
1077 /* if no policy is left, no need for blkgs - shoot them down */
1078 if (bitmap_empty(q->blkcg_pols, BLKCG_MAX_POLS))
1079 blkg_destroy_all(q);
1081 list_for_each_entry(blkg, &q->blkg_list, q_node) {
1082 /* grab blkcg lock too while removing @pd from @blkg */
1083 spin_lock(&blkg->blkcg->lock);
1085 if (pol->pd_offline_fn)
1086 pol->pd_offline_fn(blkg);
1087 if (pol->pd_exit_fn)
1088 pol->pd_exit_fn(blkg);
1090 kfree(blkg->pd[pol->plid]);
1091 blkg->pd[pol->plid] = NULL;
1093 spin_unlock(&blkg->blkcg->lock);
1096 spin_unlock_irq(q->queue_lock);
1097 blk_queue_bypass_end(q);
1099 EXPORT_SYMBOL_GPL(blkcg_deactivate_policy);
1102 * blkcg_policy_register - register a blkcg policy
1103 * @pol: blkcg policy to register
1105 * Register @pol with blkcg core. Might sleep and @pol may be modified on
1106 * successful registration. Returns 0 on success and -errno on failure.
1108 int blkcg_policy_register(struct blkcg_policy *pol)
1112 if (WARN_ON(pol->pd_size < sizeof(struct blkg_policy_data)))
1115 mutex_lock(&blkcg_pol_mutex);
1117 /* find an empty slot */
1119 for (i = 0; i < BLKCG_MAX_POLS; i++)
1120 if (!blkcg_policy[i])
1122 if (i >= BLKCG_MAX_POLS)
1125 /* register and update blkgs */
1127 blkcg_policy[i] = pol;
1129 /* everything is in place, add intf files for the new policy */
1131 WARN_ON(cgroup_add_legacy_cftypes(&blkio_cgrp_subsys,
1135 mutex_unlock(&blkcg_pol_mutex);
1138 EXPORT_SYMBOL_GPL(blkcg_policy_register);
1141 * blkcg_policy_unregister - unregister a blkcg policy
1142 * @pol: blkcg policy to unregister
1144 * Undo blkcg_policy_register(@pol). Might sleep.
1146 void blkcg_policy_unregister(struct blkcg_policy *pol)
1148 mutex_lock(&blkcg_pol_mutex);
1150 if (WARN_ON(blkcg_policy[pol->plid] != pol))
1153 /* kill the intf files first */
1155 cgroup_rm_cftypes(pol->cftypes);
1157 /* unregister and update blkgs */
1158 blkcg_policy[pol->plid] = NULL;
1160 mutex_unlock(&blkcg_pol_mutex);
1162 EXPORT_SYMBOL_GPL(blkcg_policy_unregister);