1 // SPDX-License-Identifier: GPL-2.0
3 * Common Block IO controller cgroup interface
5 * Based on ideas and code from CFQ, CFS and BFQ:
6 * Copyright (C) 2003 Jens Axboe <axboe@kernel.dk>
8 * Copyright (C) 2008 Fabio Checconi <fabio@gandalf.sssup.it>
9 * Paolo Valente <paolo.valente@unimore.it>
11 * Copyright (C) 2009 Vivek Goyal <vgoyal@redhat.com>
12 * Nauman Rafique <nauman@google.com>
14 * For policy-specific per-blkcg data:
15 * Copyright (C) 2015 Paolo Valente <paolo.valente@unimore.it>
16 * Arianna Avanzini <avanzini.arianna@gmail.com>
18 #include <linux/ioprio.h>
19 #include <linux/kdev_t.h>
20 #include <linux/module.h>
21 #include <linux/sched/signal.h>
22 #include <linux/err.h>
23 #include <linux/blkdev.h>
24 #include <linux/backing-dev.h>
25 #include <linux/slab.h>
26 #include <linux/genhd.h>
27 #include <linux/delay.h>
28 #include <linux/atomic.h>
29 #include <linux/ctype.h>
30 #include <linux/blk-cgroup.h>
31 #include <linux/tracehook.h>
32 #include <linux/psi.h>
35 #define MAX_KEY_LEN 100
38 * blkcg_pol_mutex protects blkcg_policy[] and policy [de]activation.
39 * blkcg_pol_register_mutex nests outside of it and synchronizes entire
40 * policy [un]register operations including cgroup file additions /
41 * removals. Putting cgroup file registration outside blkcg_pol_mutex
42 * allows grabbing it from cgroup callbacks.
44 static DEFINE_MUTEX(blkcg_pol_register_mutex);
45 static DEFINE_MUTEX(blkcg_pol_mutex);
47 struct blkcg blkcg_root;
48 EXPORT_SYMBOL_GPL(blkcg_root);
50 struct cgroup_subsys_state * const blkcg_root_css = &blkcg_root.css;
52 static struct blkcg_policy *blkcg_policy[BLKCG_MAX_POLS];
54 static LIST_HEAD(all_blkcgs); /* protected by blkcg_pol_mutex */
56 static bool blkcg_debug_stats = false;
58 static bool blkcg_policy_enabled(struct request_queue *q,
59 const struct blkcg_policy *pol)
61 return pol && test_bit(pol->plid, q->blkcg_pols);
65 * blkg_free - free a blkg
68 * Free @blkg which may be partially allocated.
70 static void blkg_free(struct blkcg_gq *blkg)
77 for (i = 0; i < BLKCG_MAX_POLS; i++)
79 blkcg_policy[i]->pd_free_fn(blkg->pd[i]);
81 blkg_rwstat_exit(&blkg->stat_ios);
82 blkg_rwstat_exit(&blkg->stat_bytes);
83 percpu_ref_exit(&blkg->refcnt);
87 static void __blkg_release(struct rcu_head *rcu)
89 struct blkcg_gq *blkg = container_of(rcu, struct blkcg_gq, rcu_head);
91 /* release the blkcg and parent blkg refs this blkg has been holding */
92 css_put(&blkg->blkcg->css);
94 blkg_put(blkg->parent);
96 wb_congested_put(blkg->wb_congested);
102 * A group is RCU protected, but having an rcu lock does not mean that one
103 * can access all the fields of blkg and assume these are valid. For
104 * example, don't try to follow throtl_data and request queue links.
106 * Having a reference to blkg under an rcu allows accesses to only values
107 * local to groups like group stats and group rate limits.
109 static void blkg_release(struct percpu_ref *ref)
111 struct blkcg_gq *blkg = container_of(ref, struct blkcg_gq, refcnt);
113 call_rcu(&blkg->rcu_head, __blkg_release);
117 * blkg_alloc - allocate a blkg
118 * @blkcg: block cgroup the new blkg is associated with
119 * @q: request_queue the new blkg is associated with
120 * @gfp_mask: allocation mask to use
122 * Allocate a new blkg assocating @blkcg and @q.
124 static struct blkcg_gq *blkg_alloc(struct blkcg *blkcg, struct request_queue *q,
127 struct blkcg_gq *blkg;
130 /* alloc and init base part */
131 blkg = kzalloc_node(sizeof(*blkg), gfp_mask, q->node);
135 if (percpu_ref_init(&blkg->refcnt, blkg_release, 0, gfp_mask))
138 if (blkg_rwstat_init(&blkg->stat_bytes, gfp_mask) ||
139 blkg_rwstat_init(&blkg->stat_ios, gfp_mask))
143 INIT_LIST_HEAD(&blkg->q_node);
146 for (i = 0; i < BLKCG_MAX_POLS; i++) {
147 struct blkcg_policy *pol = blkcg_policy[i];
148 struct blkg_policy_data *pd;
150 if (!blkcg_policy_enabled(q, pol))
153 /* alloc per-policy data and attach it to blkg */
154 pd = pol->pd_alloc_fn(gfp_mask, q->node);
170 struct blkcg_gq *blkg_lookup_slowpath(struct blkcg *blkcg,
171 struct request_queue *q, bool update_hint)
173 struct blkcg_gq *blkg;
176 * Hint didn't match. Look up from the radix tree. Note that the
177 * hint can only be updated under queue_lock as otherwise @blkg
178 * could have already been removed from blkg_tree. The caller is
179 * responsible for grabbing queue_lock if @update_hint.
181 blkg = radix_tree_lookup(&blkcg->blkg_tree, q->id);
182 if (blkg && blkg->q == q) {
184 lockdep_assert_held(&q->queue_lock);
185 rcu_assign_pointer(blkcg->blkg_hint, blkg);
192 EXPORT_SYMBOL_GPL(blkg_lookup_slowpath);
195 * If @new_blkg is %NULL, this function tries to allocate a new one as
196 * necessary using %GFP_NOWAIT. @new_blkg is always consumed on return.
198 static struct blkcg_gq *blkg_create(struct blkcg *blkcg,
199 struct request_queue *q,
200 struct blkcg_gq *new_blkg)
202 struct blkcg_gq *blkg;
203 struct bdi_writeback_congested *wb_congested;
206 WARN_ON_ONCE(!rcu_read_lock_held());
207 lockdep_assert_held(&q->queue_lock);
209 /* request_queue is dying, do not create/recreate a blkg */
210 if (blk_queue_dying(q)) {
215 /* blkg holds a reference to blkcg */
216 if (!css_tryget_online(&blkcg->css)) {
221 wb_congested = wb_congested_get_create(q->backing_dev_info,
223 GFP_NOWAIT | __GFP_NOWARN);
231 new_blkg = blkg_alloc(blkcg, q, GFP_NOWAIT | __GFP_NOWARN);
232 if (unlikely(!new_blkg)) {
234 goto err_put_congested;
238 blkg->wb_congested = wb_congested;
241 if (blkcg_parent(blkcg)) {
242 blkg->parent = __blkg_lookup(blkcg_parent(blkcg), q, false);
243 if (WARN_ON_ONCE(!blkg->parent)) {
245 goto err_put_congested;
247 blkg_get(blkg->parent);
250 /* invoke per-policy init */
251 for (i = 0; i < BLKCG_MAX_POLS; i++) {
252 struct blkcg_policy *pol = blkcg_policy[i];
254 if (blkg->pd[i] && pol->pd_init_fn)
255 pol->pd_init_fn(blkg->pd[i]);
259 spin_lock(&blkcg->lock);
260 ret = radix_tree_insert(&blkcg->blkg_tree, q->id, blkg);
262 hlist_add_head_rcu(&blkg->blkcg_node, &blkcg->blkg_list);
263 list_add(&blkg->q_node, &q->blkg_list);
265 for (i = 0; i < BLKCG_MAX_POLS; i++) {
266 struct blkcg_policy *pol = blkcg_policy[i];
268 if (blkg->pd[i] && pol->pd_online_fn)
269 pol->pd_online_fn(blkg->pd[i]);
273 spin_unlock(&blkcg->lock);
278 /* @blkg failed fully initialized, use the usual release path */
283 wb_congested_put(wb_congested);
285 css_put(&blkcg->css);
292 * __blkg_lookup_create - lookup blkg, try to create one if not there
293 * @blkcg: blkcg of interest
294 * @q: request_queue of interest
296 * Lookup blkg for the @blkcg - @q pair. If it doesn't exist, try to
297 * create one. blkg creation is performed recursively from blkcg_root such
298 * that all non-root blkg's have access to the parent blkg. This function
299 * should be called under RCU read lock and @q->queue_lock.
301 * Returns the blkg or the closest blkg if blkg_create() fails as it walks
304 struct blkcg_gq *__blkg_lookup_create(struct blkcg *blkcg,
305 struct request_queue *q)
307 struct blkcg_gq *blkg;
309 WARN_ON_ONCE(!rcu_read_lock_held());
310 lockdep_assert_held(&q->queue_lock);
312 blkg = __blkg_lookup(blkcg, q, true);
317 * Create blkgs walking down from blkcg_root to @blkcg, so that all
318 * non-root blkgs have access to their parents. Returns the closest
319 * blkg to the intended blkg should blkg_create() fail.
322 struct blkcg *pos = blkcg;
323 struct blkcg *parent = blkcg_parent(blkcg);
324 struct blkcg_gq *ret_blkg = q->root_blkg;
327 blkg = __blkg_lookup(parent, q, false);
329 /* remember closest blkg */
334 parent = blkcg_parent(parent);
337 blkg = blkg_create(pos, q, NULL);
346 * blkg_lookup_create - find or create a blkg
347 * @blkcg: target block cgroup
348 * @q: target request_queue
350 * This looks up or creates the blkg representing the unique pair
351 * of the blkcg and the request_queue.
353 struct blkcg_gq *blkg_lookup_create(struct blkcg *blkcg,
354 struct request_queue *q)
356 struct blkcg_gq *blkg = blkg_lookup(blkcg, q);
358 if (unlikely(!blkg)) {
361 spin_lock_irqsave(&q->queue_lock, flags);
362 blkg = __blkg_lookup_create(blkcg, q);
363 spin_unlock_irqrestore(&q->queue_lock, flags);
369 static void blkg_destroy(struct blkcg_gq *blkg)
371 struct blkcg *blkcg = blkg->blkcg;
372 struct blkcg_gq *parent = blkg->parent;
375 lockdep_assert_held(&blkg->q->queue_lock);
376 lockdep_assert_held(&blkcg->lock);
378 /* Something wrong if we are trying to remove same group twice */
379 WARN_ON_ONCE(list_empty(&blkg->q_node));
380 WARN_ON_ONCE(hlist_unhashed(&blkg->blkcg_node));
382 for (i = 0; i < BLKCG_MAX_POLS; i++) {
383 struct blkcg_policy *pol = blkcg_policy[i];
385 if (blkg->pd[i] && pol->pd_offline_fn)
386 pol->pd_offline_fn(blkg->pd[i]);
390 blkg_rwstat_add_aux(&parent->stat_bytes, &blkg->stat_bytes);
391 blkg_rwstat_add_aux(&parent->stat_ios, &blkg->stat_ios);
394 blkg->online = false;
396 radix_tree_delete(&blkcg->blkg_tree, blkg->q->id);
397 list_del_init(&blkg->q_node);
398 hlist_del_init_rcu(&blkg->blkcg_node);
401 * Both setting lookup hint to and clearing it from @blkg are done
402 * under queue_lock. If it's not pointing to @blkg now, it never
403 * will. Hint assignment itself can race safely.
405 if (rcu_access_pointer(blkcg->blkg_hint) == blkg)
406 rcu_assign_pointer(blkcg->blkg_hint, NULL);
409 * Put the reference taken at the time of creation so that when all
410 * queues are gone, group can be destroyed.
412 percpu_ref_kill(&blkg->refcnt);
416 * blkg_destroy_all - destroy all blkgs associated with a request_queue
417 * @q: request_queue of interest
419 * Destroy all blkgs associated with @q.
421 static void blkg_destroy_all(struct request_queue *q)
423 struct blkcg_gq *blkg, *n;
425 spin_lock_irq(&q->queue_lock);
426 list_for_each_entry_safe(blkg, n, &q->blkg_list, q_node) {
427 struct blkcg *blkcg = blkg->blkcg;
429 spin_lock(&blkcg->lock);
431 spin_unlock(&blkcg->lock);
435 spin_unlock_irq(&q->queue_lock);
438 static int blkcg_reset_stats(struct cgroup_subsys_state *css,
439 struct cftype *cftype, u64 val)
441 struct blkcg *blkcg = css_to_blkcg(css);
442 struct blkcg_gq *blkg;
445 mutex_lock(&blkcg_pol_mutex);
446 spin_lock_irq(&blkcg->lock);
449 * Note that stat reset is racy - it doesn't synchronize against
450 * stat updates. This is a debug feature which shouldn't exist
451 * anyway. If you get hit by a race, retry.
453 hlist_for_each_entry(blkg, &blkcg->blkg_list, blkcg_node) {
454 blkg_rwstat_reset(&blkg->stat_bytes);
455 blkg_rwstat_reset(&blkg->stat_ios);
457 for (i = 0; i < BLKCG_MAX_POLS; i++) {
458 struct blkcg_policy *pol = blkcg_policy[i];
460 if (blkg->pd[i] && pol->pd_reset_stats_fn)
461 pol->pd_reset_stats_fn(blkg->pd[i]);
465 spin_unlock_irq(&blkcg->lock);
466 mutex_unlock(&blkcg_pol_mutex);
470 const char *blkg_dev_name(struct blkcg_gq *blkg)
472 /* some drivers (floppy) instantiate a queue w/o disk registered */
473 if (blkg->q->backing_dev_info->dev)
474 return dev_name(blkg->q->backing_dev_info->dev);
479 * blkcg_print_blkgs - helper for printing per-blkg data
480 * @sf: seq_file to print to
481 * @blkcg: blkcg of interest
482 * @prfill: fill function to print out a blkg
483 * @pol: policy in question
484 * @data: data to be passed to @prfill
485 * @show_total: to print out sum of prfill return values or not
487 * This function invokes @prfill on each blkg of @blkcg if pd for the
488 * policy specified by @pol exists. @prfill is invoked with @sf, the
489 * policy data and @data and the matching queue lock held. If @show_total
490 * is %true, the sum of the return values from @prfill is printed with
491 * "Total" label at the end.
493 * This is to be used to construct print functions for
494 * cftype->read_seq_string method.
496 void blkcg_print_blkgs(struct seq_file *sf, struct blkcg *blkcg,
497 u64 (*prfill)(struct seq_file *,
498 struct blkg_policy_data *, int),
499 const struct blkcg_policy *pol, int data,
502 struct blkcg_gq *blkg;
506 hlist_for_each_entry_rcu(blkg, &blkcg->blkg_list, blkcg_node) {
507 spin_lock_irq(&blkg->q->queue_lock);
508 if (blkcg_policy_enabled(blkg->q, pol))
509 total += prfill(sf, blkg->pd[pol->plid], data);
510 spin_unlock_irq(&blkg->q->queue_lock);
515 seq_printf(sf, "Total %llu\n", (unsigned long long)total);
517 EXPORT_SYMBOL_GPL(blkcg_print_blkgs);
520 * __blkg_prfill_u64 - prfill helper for a single u64 value
521 * @sf: seq_file to print to
522 * @pd: policy private data of interest
525 * Print @v to @sf for the device assocaited with @pd.
527 u64 __blkg_prfill_u64(struct seq_file *sf, struct blkg_policy_data *pd, u64 v)
529 const char *dname = blkg_dev_name(pd->blkg);
534 seq_printf(sf, "%s %llu\n", dname, (unsigned long long)v);
537 EXPORT_SYMBOL_GPL(__blkg_prfill_u64);
540 * __blkg_prfill_rwstat - prfill helper for a blkg_rwstat
541 * @sf: seq_file to print to
542 * @pd: policy private data of interest
543 * @rwstat: rwstat to print
545 * Print @rwstat to @sf for the device assocaited with @pd.
547 u64 __blkg_prfill_rwstat(struct seq_file *sf, struct blkg_policy_data *pd,
548 const struct blkg_rwstat_sample *rwstat)
550 static const char *rwstr[] = {
551 [BLKG_RWSTAT_READ] = "Read",
552 [BLKG_RWSTAT_WRITE] = "Write",
553 [BLKG_RWSTAT_SYNC] = "Sync",
554 [BLKG_RWSTAT_ASYNC] = "Async",
555 [BLKG_RWSTAT_DISCARD] = "Discard",
557 const char *dname = blkg_dev_name(pd->blkg);
564 for (i = 0; i < BLKG_RWSTAT_NR; i++)
565 seq_printf(sf, "%s %s %llu\n", dname, rwstr[i],
568 v = rwstat->cnt[BLKG_RWSTAT_READ] +
569 rwstat->cnt[BLKG_RWSTAT_WRITE] +
570 rwstat->cnt[BLKG_RWSTAT_DISCARD];
571 seq_printf(sf, "%s Total %llu\n", dname, v);
574 EXPORT_SYMBOL_GPL(__blkg_prfill_rwstat);
577 * blkg_prfill_rwstat - prfill callback for blkg_rwstat
578 * @sf: seq_file to print to
579 * @pd: policy private data of interest
580 * @off: offset to the blkg_rwstat in @pd
582 * prfill callback for printing a blkg_rwstat.
584 u64 blkg_prfill_rwstat(struct seq_file *sf, struct blkg_policy_data *pd,
587 struct blkg_rwstat_sample rwstat = { };
589 blkg_rwstat_read((void *)pd + off, &rwstat);
590 return __blkg_prfill_rwstat(sf, pd, &rwstat);
592 EXPORT_SYMBOL_GPL(blkg_prfill_rwstat);
594 static u64 blkg_prfill_rwstat_field(struct seq_file *sf,
595 struct blkg_policy_data *pd, int off)
597 struct blkg_rwstat_sample rwstat = { };
599 blkg_rwstat_read((void *)pd->blkg + off, &rwstat);
600 return __blkg_prfill_rwstat(sf, pd, &rwstat);
604 * blkg_print_stat_bytes - seq_show callback for blkg->stat_bytes
605 * @sf: seq_file to print to
608 * To be used as cftype->seq_show to print blkg->stat_bytes.
609 * cftype->private must be set to the blkcg_policy.
611 int blkg_print_stat_bytes(struct seq_file *sf, void *v)
613 blkcg_print_blkgs(sf, css_to_blkcg(seq_css(sf)),
614 blkg_prfill_rwstat_field, (void *)seq_cft(sf)->private,
615 offsetof(struct blkcg_gq, stat_bytes), true);
618 EXPORT_SYMBOL_GPL(blkg_print_stat_bytes);
621 * blkg_print_stat_bytes - seq_show callback for blkg->stat_ios
622 * @sf: seq_file to print to
625 * To be used as cftype->seq_show to print blkg->stat_ios. cftype->private
626 * must be set to the blkcg_policy.
628 int blkg_print_stat_ios(struct seq_file *sf, void *v)
630 blkcg_print_blkgs(sf, css_to_blkcg(seq_css(sf)),
631 blkg_prfill_rwstat_field, (void *)seq_cft(sf)->private,
632 offsetof(struct blkcg_gq, stat_ios), true);
635 EXPORT_SYMBOL_GPL(blkg_print_stat_ios);
637 static u64 blkg_prfill_rwstat_field_recursive(struct seq_file *sf,
638 struct blkg_policy_data *pd,
641 struct blkg_rwstat_sample rwstat;
643 blkg_rwstat_recursive_sum(pd->blkg, NULL, off, &rwstat);
644 return __blkg_prfill_rwstat(sf, pd, &rwstat);
648 * blkg_print_stat_bytes_recursive - recursive version of blkg_print_stat_bytes
649 * @sf: seq_file to print to
652 int blkg_print_stat_bytes_recursive(struct seq_file *sf, void *v)
654 blkcg_print_blkgs(sf, css_to_blkcg(seq_css(sf)),
655 blkg_prfill_rwstat_field_recursive,
656 (void *)seq_cft(sf)->private,
657 offsetof(struct blkcg_gq, stat_bytes), true);
660 EXPORT_SYMBOL_GPL(blkg_print_stat_bytes_recursive);
663 * blkg_print_stat_ios_recursive - recursive version of blkg_print_stat_ios
664 * @sf: seq_file to print to
667 int blkg_print_stat_ios_recursive(struct seq_file *sf, void *v)
669 blkcg_print_blkgs(sf, css_to_blkcg(seq_css(sf)),
670 blkg_prfill_rwstat_field_recursive,
671 (void *)seq_cft(sf)->private,
672 offsetof(struct blkcg_gq, stat_ios), true);
675 EXPORT_SYMBOL_GPL(blkg_print_stat_ios_recursive);
678 * blkg_rwstat_recursive_sum - collect hierarchical blkg_rwstat
679 * @blkg: blkg of interest
680 * @pol: blkcg_policy which contains the blkg_rwstat
681 * @off: offset to the blkg_rwstat in blkg_policy_data or @blkg
682 * @sum: blkg_rwstat_sample structure containing the results
684 * Collect the blkg_rwstat specified by @blkg, @pol and @off and all its
685 * online descendants and their aux counts. The caller must be holding the
686 * queue lock for online tests.
688 * If @pol is NULL, blkg_rwstat is at @off bytes into @blkg; otherwise, it
689 * is at @off bytes into @blkg's blkg_policy_data of the policy.
691 void blkg_rwstat_recursive_sum(struct blkcg_gq *blkg, struct blkcg_policy *pol,
692 int off, struct blkg_rwstat_sample *sum)
694 struct blkcg_gq *pos_blkg;
695 struct cgroup_subsys_state *pos_css;
698 lockdep_assert_held(&blkg->q->queue_lock);
701 blkg_for_each_descendant_pre(pos_blkg, pos_css, blkg) {
702 struct blkg_rwstat *rwstat;
704 if (!pos_blkg->online)
708 rwstat = (void *)blkg_to_pd(pos_blkg, pol) + off;
710 rwstat = (void *)pos_blkg + off;
712 for (i = 0; i < BLKG_RWSTAT_NR; i++)
713 sum->cnt[i] = blkg_rwstat_read_counter(rwstat, i);
717 EXPORT_SYMBOL_GPL(blkg_rwstat_recursive_sum);
719 /* Performs queue bypass and policy enabled checks then looks up blkg. */
720 static struct blkcg_gq *blkg_lookup_check(struct blkcg *blkcg,
721 const struct blkcg_policy *pol,
722 struct request_queue *q)
724 WARN_ON_ONCE(!rcu_read_lock_held());
725 lockdep_assert_held(&q->queue_lock);
727 if (!blkcg_policy_enabled(q, pol))
728 return ERR_PTR(-EOPNOTSUPP);
729 return __blkg_lookup(blkcg, q, true /* update_hint */);
733 * blkg_conf_prep - parse and prepare for per-blkg config update
734 * @blkcg: target block cgroup
735 * @pol: target policy
736 * @input: input string
737 * @ctx: blkg_conf_ctx to be filled
739 * Parse per-blkg config update from @input and initialize @ctx with the
740 * result. @ctx->blkg points to the blkg to be updated and @ctx->body the
741 * part of @input following MAJ:MIN. This function returns with RCU read
742 * lock and queue lock held and must be paired with blkg_conf_finish().
744 int blkg_conf_prep(struct blkcg *blkcg, const struct blkcg_policy *pol,
745 char *input, struct blkg_conf_ctx *ctx)
746 __acquires(rcu) __acquires(&disk->queue->queue_lock)
748 struct gendisk *disk;
749 struct request_queue *q;
750 struct blkcg_gq *blkg;
751 unsigned int major, minor;
752 int key_len, part, ret;
755 if (sscanf(input, "%u:%u%n", &major, &minor, &key_len) != 2)
758 body = input + key_len;
761 body = skip_spaces(body);
763 disk = get_gendisk(MKDEV(major, minor), &part);
774 spin_lock_irq(&q->queue_lock);
776 blkg = blkg_lookup_check(blkcg, pol, q);
786 * Create blkgs walking down from blkcg_root to @blkcg, so that all
787 * non-root blkgs have access to their parents.
790 struct blkcg *pos = blkcg;
791 struct blkcg *parent;
792 struct blkcg_gq *new_blkg;
794 parent = blkcg_parent(blkcg);
795 while (parent && !__blkg_lookup(parent, q, false)) {
797 parent = blkcg_parent(parent);
800 /* Drop locks to do new blkg allocation with GFP_KERNEL. */
801 spin_unlock_irq(&q->queue_lock);
804 new_blkg = blkg_alloc(pos, q, GFP_KERNEL);
805 if (unlikely(!new_blkg)) {
811 spin_lock_irq(&q->queue_lock);
813 blkg = blkg_lookup_check(pos, pol, q);
822 blkg = blkg_create(pos, q, new_blkg);
839 spin_unlock_irq(&q->queue_lock);
842 put_disk_and_module(disk);
844 * If queue was bypassing, we should retry. Do so after a
845 * short msleep(). It isn't strictly necessary but queue
846 * can be bypassing for some time and it's always nice to
847 * avoid busy looping.
851 ret = restart_syscall();
857 * blkg_conf_finish - finish up per-blkg config update
858 * @ctx: blkg_conf_ctx intiailized by blkg_conf_prep()
860 * Finish up after per-blkg config update. This function must be paired
861 * with blkg_conf_prep().
863 void blkg_conf_finish(struct blkg_conf_ctx *ctx)
864 __releases(&ctx->disk->queue->queue_lock) __releases(rcu)
866 spin_unlock_irq(&ctx->disk->queue->queue_lock);
868 put_disk_and_module(ctx->disk);
871 static int blkcg_print_stat(struct seq_file *sf, void *v)
873 struct blkcg *blkcg = css_to_blkcg(seq_css(sf));
874 struct blkcg_gq *blkg;
878 hlist_for_each_entry_rcu(blkg, &blkcg->blkg_list, blkcg_node) {
881 struct blkg_rwstat_sample rwstat;
882 u64 rbytes, wbytes, rios, wios, dbytes, dios;
883 size_t size = seq_get_buf(sf, &buf), off = 0;
885 bool has_stats = false;
887 dname = blkg_dev_name(blkg);
892 * Hooray string manipulation, count is the size written NOT
893 * INCLUDING THE \0, so size is now count+1 less than what we
894 * had before, but we want to start writing the next bit from
895 * the \0 so we only add count to buf.
897 off += scnprintf(buf+off, size-off, "%s ", dname);
899 spin_lock_irq(&blkg->q->queue_lock);
901 blkg_rwstat_recursive_sum(blkg, NULL,
902 offsetof(struct blkcg_gq, stat_bytes), &rwstat);
903 rbytes = rwstat.cnt[BLKG_RWSTAT_READ];
904 wbytes = rwstat.cnt[BLKG_RWSTAT_WRITE];
905 dbytes = rwstat.cnt[BLKG_RWSTAT_DISCARD];
907 blkg_rwstat_recursive_sum(blkg, NULL,
908 offsetof(struct blkcg_gq, stat_ios), &rwstat);
909 rios = rwstat.cnt[BLKG_RWSTAT_READ];
910 wios = rwstat.cnt[BLKG_RWSTAT_WRITE];
911 dios = rwstat.cnt[BLKG_RWSTAT_DISCARD];
913 spin_unlock_irq(&blkg->q->queue_lock);
915 if (rbytes || wbytes || rios || wios) {
917 off += scnprintf(buf+off, size-off,
918 "rbytes=%llu wbytes=%llu rios=%llu wios=%llu dbytes=%llu dios=%llu",
919 rbytes, wbytes, rios, wios,
923 if (!blkcg_debug_stats)
926 if (atomic_read(&blkg->use_delay)) {
928 off += scnprintf(buf+off, size-off,
929 " use_delay=%d delay_nsec=%llu",
930 atomic_read(&blkg->use_delay),
931 (unsigned long long)atomic64_read(&blkg->delay_nsec));
934 for (i = 0; i < BLKCG_MAX_POLS; i++) {
935 struct blkcg_policy *pol = blkcg_policy[i];
938 if (!blkg->pd[i] || !pol->pd_stat_fn)
941 written = pol->pd_stat_fn(blkg->pd[i], buf+off, size-off);
948 if (off < size - 1) {
949 off += scnprintf(buf+off, size-off, "\n");
961 static struct cftype blkcg_files[] = {
964 .flags = CFTYPE_NOT_ON_ROOT,
965 .seq_show = blkcg_print_stat,
970 static struct cftype blkcg_legacy_files[] = {
972 .name = "reset_stats",
973 .write_u64 = blkcg_reset_stats,
979 * blkcg destruction is a three-stage process.
981 * 1. Destruction starts. The blkcg_css_offline() callback is invoked
982 * which offlines writeback. Here we tie the next stage of blkg destruction
983 * to the completion of writeback associated with the blkcg. This lets us
984 * avoid punting potentially large amounts of outstanding writeback to root
985 * while maintaining any ongoing policies. The next stage is triggered when
986 * the nr_cgwbs count goes to zero.
988 * 2. When the nr_cgwbs count goes to zero, blkcg_destroy_blkgs() is called
989 * and handles the destruction of blkgs. Here the css reference held by
990 * the blkg is put back eventually allowing blkcg_css_free() to be called.
991 * This work may occur in cgwb_release_workfn() on the cgwb_release
992 * workqueue. Any submitted ios that fail to get the blkg ref will be
993 * punted to the root_blkg.
995 * 3. Once the blkcg ref count goes to zero, blkcg_css_free() is called.
996 * This finally frees the blkcg.
1000 * blkcg_css_offline - cgroup css_offline callback
1001 * @css: css of interest
1003 * This function is called when @css is about to go away. Here the cgwbs are
1004 * offlined first and only once writeback associated with the blkcg has
1005 * finished do we start step 2 (see above).
1007 static void blkcg_css_offline(struct cgroup_subsys_state *css)
1009 struct blkcg *blkcg = css_to_blkcg(css);
1011 /* this prevents anyone from attaching or migrating to this blkcg */
1012 wb_blkcg_offline(blkcg);
1014 /* put the base cgwb reference allowing step 2 to be triggered */
1015 blkcg_cgwb_put(blkcg);
1019 * blkcg_destroy_blkgs - responsible for shooting down blkgs
1020 * @blkcg: blkcg of interest
1022 * blkgs should be removed while holding both q and blkcg locks. As blkcg lock
1023 * is nested inside q lock, this function performs reverse double lock dancing.
1024 * Destroying the blkgs releases the reference held on the blkcg's css allowing
1025 * blkcg_css_free to eventually be called.
1027 * This is the blkcg counterpart of ioc_release_fn().
1029 void blkcg_destroy_blkgs(struct blkcg *blkcg)
1031 spin_lock_irq(&blkcg->lock);
1033 while (!hlist_empty(&blkcg->blkg_list)) {
1034 struct blkcg_gq *blkg = hlist_entry(blkcg->blkg_list.first,
1035 struct blkcg_gq, blkcg_node);
1036 struct request_queue *q = blkg->q;
1038 if (spin_trylock(&q->queue_lock)) {
1040 spin_unlock(&q->queue_lock);
1042 spin_unlock_irq(&blkcg->lock);
1044 spin_lock_irq(&blkcg->lock);
1048 spin_unlock_irq(&blkcg->lock);
1051 static void blkcg_css_free(struct cgroup_subsys_state *css)
1053 struct blkcg *blkcg = css_to_blkcg(css);
1056 mutex_lock(&blkcg_pol_mutex);
1058 list_del(&blkcg->all_blkcgs_node);
1060 for (i = 0; i < BLKCG_MAX_POLS; i++)
1062 blkcg_policy[i]->cpd_free_fn(blkcg->cpd[i]);
1064 mutex_unlock(&blkcg_pol_mutex);
1069 static struct cgroup_subsys_state *
1070 blkcg_css_alloc(struct cgroup_subsys_state *parent_css)
1072 struct blkcg *blkcg;
1073 struct cgroup_subsys_state *ret;
1076 mutex_lock(&blkcg_pol_mutex);
1079 blkcg = &blkcg_root;
1081 blkcg = kzalloc(sizeof(*blkcg), GFP_KERNEL);
1083 ret = ERR_PTR(-ENOMEM);
1088 for (i = 0; i < BLKCG_MAX_POLS ; i++) {
1089 struct blkcg_policy *pol = blkcg_policy[i];
1090 struct blkcg_policy_data *cpd;
1093 * If the policy hasn't been attached yet, wait for it
1094 * to be attached before doing anything else. Otherwise,
1095 * check if the policy requires any specific per-cgroup
1096 * data: if it does, allocate and initialize it.
1098 if (!pol || !pol->cpd_alloc_fn)
1101 cpd = pol->cpd_alloc_fn(GFP_KERNEL);
1103 ret = ERR_PTR(-ENOMEM);
1106 blkcg->cpd[i] = cpd;
1109 if (pol->cpd_init_fn)
1110 pol->cpd_init_fn(cpd);
1113 spin_lock_init(&blkcg->lock);
1114 INIT_RADIX_TREE(&blkcg->blkg_tree, GFP_NOWAIT | __GFP_NOWARN);
1115 INIT_HLIST_HEAD(&blkcg->blkg_list);
1116 #ifdef CONFIG_CGROUP_WRITEBACK
1117 INIT_LIST_HEAD(&blkcg->cgwb_list);
1118 refcount_set(&blkcg->cgwb_refcnt, 1);
1120 list_add_tail(&blkcg->all_blkcgs_node, &all_blkcgs);
1122 mutex_unlock(&blkcg_pol_mutex);
1126 for (i--; i >= 0; i--)
1128 blkcg_policy[i]->cpd_free_fn(blkcg->cpd[i]);
1130 if (blkcg != &blkcg_root)
1133 mutex_unlock(&blkcg_pol_mutex);
1138 * blkcg_init_queue - initialize blkcg part of request queue
1139 * @q: request_queue to initialize
1141 * Called from blk_alloc_queue_node(). Responsible for initializing blkcg
1142 * part of new request_queue @q.
1145 * 0 on success, -errno on failure.
1147 int blkcg_init_queue(struct request_queue *q)
1149 struct blkcg_gq *new_blkg, *blkg;
1153 new_blkg = blkg_alloc(&blkcg_root, q, GFP_KERNEL);
1157 preloaded = !radix_tree_preload(GFP_KERNEL);
1159 /* Make sure the root blkg exists. */
1161 spin_lock_irq(&q->queue_lock);
1162 blkg = blkg_create(&blkcg_root, q, new_blkg);
1165 q->root_blkg = blkg;
1166 spin_unlock_irq(&q->queue_lock);
1170 radix_tree_preload_end();
1172 ret = blk_iolatency_init(q);
1174 goto err_destroy_all;
1176 ret = blk_throtl_init(q);
1178 goto err_destroy_all;
1182 blkg_destroy_all(q);
1185 spin_unlock_irq(&q->queue_lock);
1188 radix_tree_preload_end();
1189 return PTR_ERR(blkg);
1193 * blkcg_drain_queue - drain blkcg part of request_queue
1194 * @q: request_queue to drain
1196 * Called from blk_drain_queue(). Responsible for draining blkcg part.
1198 void blkcg_drain_queue(struct request_queue *q)
1200 lockdep_assert_held(&q->queue_lock);
1203 * @q could be exiting and already have destroyed all blkgs as
1204 * indicated by NULL root_blkg. If so, don't confuse policies.
1209 blk_throtl_drain(q);
1213 * blkcg_exit_queue - exit and release blkcg part of request_queue
1214 * @q: request_queue being released
1216 * Called from blk_exit_queue(). Responsible for exiting blkcg part.
1218 void blkcg_exit_queue(struct request_queue *q)
1220 blkg_destroy_all(q);
1225 * We cannot support shared io contexts, as we have no mean to support
1226 * two tasks with the same ioc in two different groups without major rework
1227 * of the main cic data structures. For now we allow a task to change
1228 * its cgroup only if it's the only owner of its ioc.
1230 static int blkcg_can_attach(struct cgroup_taskset *tset)
1232 struct task_struct *task;
1233 struct cgroup_subsys_state *dst_css;
1234 struct io_context *ioc;
1237 /* task_lock() is needed to avoid races with exit_io_context() */
1238 cgroup_taskset_for_each(task, dst_css, tset) {
1240 ioc = task->io_context;
1241 if (ioc && atomic_read(&ioc->nr_tasks) > 1)
1250 static void blkcg_bind(struct cgroup_subsys_state *root_css)
1254 mutex_lock(&blkcg_pol_mutex);
1256 for (i = 0; i < BLKCG_MAX_POLS; i++) {
1257 struct blkcg_policy *pol = blkcg_policy[i];
1258 struct blkcg *blkcg;
1260 if (!pol || !pol->cpd_bind_fn)
1263 list_for_each_entry(blkcg, &all_blkcgs, all_blkcgs_node)
1264 if (blkcg->cpd[pol->plid])
1265 pol->cpd_bind_fn(blkcg->cpd[pol->plid]);
1267 mutex_unlock(&blkcg_pol_mutex);
1270 static void blkcg_exit(struct task_struct *tsk)
1272 if (tsk->throttle_queue)
1273 blk_put_queue(tsk->throttle_queue);
1274 tsk->throttle_queue = NULL;
1277 struct cgroup_subsys io_cgrp_subsys = {
1278 .css_alloc = blkcg_css_alloc,
1279 .css_offline = blkcg_css_offline,
1280 .css_free = blkcg_css_free,
1281 .can_attach = blkcg_can_attach,
1283 .dfl_cftypes = blkcg_files,
1284 .legacy_cftypes = blkcg_legacy_files,
1285 .legacy_name = "blkio",
1289 * This ensures that, if available, memcg is automatically enabled
1290 * together on the default hierarchy so that the owner cgroup can
1291 * be retrieved from writeback pages.
1293 .depends_on = 1 << memory_cgrp_id,
1296 EXPORT_SYMBOL_GPL(io_cgrp_subsys);
1299 * blkcg_activate_policy - activate a blkcg policy on a request_queue
1300 * @q: request_queue of interest
1301 * @pol: blkcg policy to activate
1303 * Activate @pol on @q. Requires %GFP_KERNEL context. @q goes through
1304 * bypass mode to populate its blkgs with policy_data for @pol.
1306 * Activation happens with @q bypassed, so nobody would be accessing blkgs
1307 * from IO path. Update of each blkg is protected by both queue and blkcg
1308 * locks so that holding either lock and testing blkcg_policy_enabled() is
1309 * always enough for dereferencing policy data.
1311 * The caller is responsible for synchronizing [de]activations and policy
1312 * [un]registerations. Returns 0 on success, -errno on failure.
1314 int blkcg_activate_policy(struct request_queue *q,
1315 const struct blkcg_policy *pol)
1317 struct blkg_policy_data *pd_prealloc = NULL;
1318 struct blkcg_gq *blkg;
1321 if (blkcg_policy_enabled(q, pol))
1325 blk_mq_freeze_queue(q);
1328 pd_prealloc = pol->pd_alloc_fn(GFP_KERNEL, q->node);
1331 goto out_bypass_end;
1335 spin_lock_irq(&q->queue_lock);
1337 /* blkg_list is pushed at the head, reverse walk to init parents first */
1338 list_for_each_entry_reverse(blkg, &q->blkg_list, q_node) {
1339 struct blkg_policy_data *pd;
1341 if (blkg->pd[pol->plid])
1344 pd = pol->pd_alloc_fn(GFP_NOWAIT | __GFP_NOWARN, q->node);
1346 swap(pd, pd_prealloc);
1348 spin_unlock_irq(&q->queue_lock);
1352 blkg->pd[pol->plid] = pd;
1354 pd->plid = pol->plid;
1355 if (pol->pd_init_fn)
1356 pol->pd_init_fn(pd);
1359 __set_bit(pol->plid, q->blkcg_pols);
1362 spin_unlock_irq(&q->queue_lock);
1365 blk_mq_unfreeze_queue(q);
1367 pol->pd_free_fn(pd_prealloc);
1370 EXPORT_SYMBOL_GPL(blkcg_activate_policy);
1373 * blkcg_deactivate_policy - deactivate a blkcg policy on a request_queue
1374 * @q: request_queue of interest
1375 * @pol: blkcg policy to deactivate
1377 * Deactivate @pol on @q. Follows the same synchronization rules as
1378 * blkcg_activate_policy().
1380 void blkcg_deactivate_policy(struct request_queue *q,
1381 const struct blkcg_policy *pol)
1383 struct blkcg_gq *blkg;
1385 if (!blkcg_policy_enabled(q, pol))
1389 blk_mq_freeze_queue(q);
1391 spin_lock_irq(&q->queue_lock);
1393 __clear_bit(pol->plid, q->blkcg_pols);
1395 list_for_each_entry(blkg, &q->blkg_list, q_node) {
1396 if (blkg->pd[pol->plid]) {
1397 if (pol->pd_offline_fn)
1398 pol->pd_offline_fn(blkg->pd[pol->plid]);
1399 pol->pd_free_fn(blkg->pd[pol->plid]);
1400 blkg->pd[pol->plid] = NULL;
1404 spin_unlock_irq(&q->queue_lock);
1407 blk_mq_unfreeze_queue(q);
1409 EXPORT_SYMBOL_GPL(blkcg_deactivate_policy);
1412 * blkcg_policy_register - register a blkcg policy
1413 * @pol: blkcg policy to register
1415 * Register @pol with blkcg core. Might sleep and @pol may be modified on
1416 * successful registration. Returns 0 on success and -errno on failure.
1418 int blkcg_policy_register(struct blkcg_policy *pol)
1420 struct blkcg *blkcg;
1423 mutex_lock(&blkcg_pol_register_mutex);
1424 mutex_lock(&blkcg_pol_mutex);
1426 /* find an empty slot */
1428 for (i = 0; i < BLKCG_MAX_POLS; i++)
1429 if (!blkcg_policy[i])
1431 if (i >= BLKCG_MAX_POLS) {
1432 pr_warn("blkcg_policy_register: BLKCG_MAX_POLS too small\n");
1436 /* Make sure cpd/pd_alloc_fn and cpd/pd_free_fn in pairs */
1437 if ((!pol->cpd_alloc_fn ^ !pol->cpd_free_fn) ||
1438 (!pol->pd_alloc_fn ^ !pol->pd_free_fn))
1443 blkcg_policy[pol->plid] = pol;
1445 /* allocate and install cpd's */
1446 if (pol->cpd_alloc_fn) {
1447 list_for_each_entry(blkcg, &all_blkcgs, all_blkcgs_node) {
1448 struct blkcg_policy_data *cpd;
1450 cpd = pol->cpd_alloc_fn(GFP_KERNEL);
1454 blkcg->cpd[pol->plid] = cpd;
1456 cpd->plid = pol->plid;
1457 pol->cpd_init_fn(cpd);
1461 mutex_unlock(&blkcg_pol_mutex);
1463 /* everything is in place, add intf files for the new policy */
1464 if (pol->dfl_cftypes)
1465 WARN_ON(cgroup_add_dfl_cftypes(&io_cgrp_subsys,
1467 if (pol->legacy_cftypes)
1468 WARN_ON(cgroup_add_legacy_cftypes(&io_cgrp_subsys,
1469 pol->legacy_cftypes));
1470 mutex_unlock(&blkcg_pol_register_mutex);
1474 if (pol->cpd_free_fn) {
1475 list_for_each_entry(blkcg, &all_blkcgs, all_blkcgs_node) {
1476 if (blkcg->cpd[pol->plid]) {
1477 pol->cpd_free_fn(blkcg->cpd[pol->plid]);
1478 blkcg->cpd[pol->plid] = NULL;
1482 blkcg_policy[pol->plid] = NULL;
1484 mutex_unlock(&blkcg_pol_mutex);
1485 mutex_unlock(&blkcg_pol_register_mutex);
1488 EXPORT_SYMBOL_GPL(blkcg_policy_register);
1491 * blkcg_policy_unregister - unregister a blkcg policy
1492 * @pol: blkcg policy to unregister
1494 * Undo blkcg_policy_register(@pol). Might sleep.
1496 void blkcg_policy_unregister(struct blkcg_policy *pol)
1498 struct blkcg *blkcg;
1500 mutex_lock(&blkcg_pol_register_mutex);
1502 if (WARN_ON(blkcg_policy[pol->plid] != pol))
1505 /* kill the intf files first */
1506 if (pol->dfl_cftypes)
1507 cgroup_rm_cftypes(pol->dfl_cftypes);
1508 if (pol->legacy_cftypes)
1509 cgroup_rm_cftypes(pol->legacy_cftypes);
1511 /* remove cpds and unregister */
1512 mutex_lock(&blkcg_pol_mutex);
1514 if (pol->cpd_free_fn) {
1515 list_for_each_entry(blkcg, &all_blkcgs, all_blkcgs_node) {
1516 if (blkcg->cpd[pol->plid]) {
1517 pol->cpd_free_fn(blkcg->cpd[pol->plid]);
1518 blkcg->cpd[pol->plid] = NULL;
1522 blkcg_policy[pol->plid] = NULL;
1524 mutex_unlock(&blkcg_pol_mutex);
1526 mutex_unlock(&blkcg_pol_register_mutex);
1528 EXPORT_SYMBOL_GPL(blkcg_policy_unregister);
1531 * Scale the accumulated delay based on how long it has been since we updated
1532 * the delay. We only call this when we are adding delay, in case it's been a
1533 * while since we added delay, and when we are checking to see if we need to
1534 * delay a task, to account for any delays that may have occurred.
1536 static void blkcg_scale_delay(struct blkcg_gq *blkg, u64 now)
1538 u64 old = atomic64_read(&blkg->delay_start);
1541 * We only want to scale down every second. The idea here is that we
1542 * want to delay people for min(delay_nsec, NSEC_PER_SEC) in a certain
1543 * time window. We only want to throttle tasks for recent delay that
1544 * has occurred, in 1 second time windows since that's the maximum
1545 * things can be throttled. We save the current delay window in
1546 * blkg->last_delay so we know what amount is still left to be charged
1547 * to the blkg from this point onward. blkg->last_use keeps track of
1548 * the use_delay counter. The idea is if we're unthrottling the blkg we
1549 * are ok with whatever is happening now, and we can take away more of
1550 * the accumulated delay as we've already throttled enough that
1551 * everybody is happy with their IO latencies.
1553 if (time_before64(old + NSEC_PER_SEC, now) &&
1554 atomic64_cmpxchg(&blkg->delay_start, old, now) == old) {
1555 u64 cur = atomic64_read(&blkg->delay_nsec);
1556 u64 sub = min_t(u64, blkg->last_delay, now - old);
1557 int cur_use = atomic_read(&blkg->use_delay);
1560 * We've been unthrottled, subtract a larger chunk of our
1561 * accumulated delay.
1563 if (cur_use < blkg->last_use)
1564 sub = max_t(u64, sub, blkg->last_delay >> 1);
1567 * This shouldn't happen, but handle it anyway. Our delay_nsec
1568 * should only ever be growing except here where we subtract out
1569 * min(last_delay, 1 second), but lord knows bugs happen and I'd
1570 * rather not end up with negative numbers.
1572 if (unlikely(cur < sub)) {
1573 atomic64_set(&blkg->delay_nsec, 0);
1574 blkg->last_delay = 0;
1576 atomic64_sub(sub, &blkg->delay_nsec);
1577 blkg->last_delay = cur - sub;
1579 blkg->last_use = cur_use;
1584 * This is called when we want to actually walk up the hierarchy and check to
1585 * see if we need to throttle, and then actually throttle if there is some
1586 * accumulated delay. This should only be called upon return to user space so
1587 * we're not holding some lock that would induce a priority inversion.
1589 static void blkcg_maybe_throttle_blkg(struct blkcg_gq *blkg, bool use_memdelay)
1591 unsigned long pflags;
1592 u64 now = ktime_to_ns(ktime_get());
1597 while (blkg->parent) {
1598 if (atomic_read(&blkg->use_delay)) {
1599 blkcg_scale_delay(blkg, now);
1600 delay_nsec = max_t(u64, delay_nsec,
1601 atomic64_read(&blkg->delay_nsec));
1603 blkg = blkg->parent;
1610 * Let's not sleep for all eternity if we've amassed a huge delay.
1611 * Swapping or metadata IO can accumulate 10's of seconds worth of
1612 * delay, and we want userspace to be able to do _something_ so cap the
1613 * delays at 1 second. If there's 10's of seconds worth of delay then
1614 * the tasks will be delayed for 1 second for every syscall.
1616 delay_nsec = min_t(u64, delay_nsec, 250 * NSEC_PER_MSEC);
1619 psi_memstall_enter(&pflags);
1621 exp = ktime_add_ns(now, delay_nsec);
1622 tok = io_schedule_prepare();
1624 __set_current_state(TASK_KILLABLE);
1625 if (!schedule_hrtimeout(&exp, HRTIMER_MODE_ABS))
1627 } while (!fatal_signal_pending(current));
1628 io_schedule_finish(tok);
1631 psi_memstall_leave(&pflags);
1635 * blkcg_maybe_throttle_current - throttle the current task if it has been marked
1637 * This is only called if we've been marked with set_notify_resume(). Obviously
1638 * we can be set_notify_resume() for reasons other than blkcg throttling, so we
1639 * check to see if current->throttle_queue is set and if not this doesn't do
1640 * anything. This should only ever be called by the resume code, it's not meant
1641 * to be called by people willy-nilly as it will actually do the work to
1642 * throttle the task if it is setup for throttling.
1644 void blkcg_maybe_throttle_current(void)
1646 struct request_queue *q = current->throttle_queue;
1647 struct cgroup_subsys_state *css;
1648 struct blkcg *blkcg;
1649 struct blkcg_gq *blkg;
1650 bool use_memdelay = current->use_memdelay;
1655 current->throttle_queue = NULL;
1656 current->use_memdelay = false;
1659 css = kthread_blkcg();
1661 blkcg = css_to_blkcg(css);
1663 blkcg = css_to_blkcg(task_css(current, io_cgrp_id));
1667 blkg = blkg_lookup(blkcg, q);
1670 if (!blkg_tryget(blkg))
1674 blkcg_maybe_throttle_blkg(blkg, use_memdelay);
1684 * blkcg_schedule_throttle - this task needs to check for throttling
1685 * @q: the request queue IO was submitted on
1686 * @use_memdelay: do we charge this to memory delay for PSI
1688 * This is called by the IO controller when we know there's delay accumulated
1689 * for the blkg for this task. We do not pass the blkg because there are places
1690 * we call this that may not have that information, the swapping code for
1691 * instance will only have a request_queue at that point. This set's the
1692 * notify_resume for the task to check and see if it requires throttling before
1693 * returning to user space.
1695 * We will only schedule once per syscall. You can call this over and over
1696 * again and it will only do the check once upon return to user space, and only
1697 * throttle once. If the task needs to be throttled again it'll need to be
1698 * re-set at the next time we see the task.
1700 void blkcg_schedule_throttle(struct request_queue *q, bool use_memdelay)
1702 if (unlikely(current->flags & PF_KTHREAD))
1705 if (!blk_get_queue(q))
1708 if (current->throttle_queue)
1709 blk_put_queue(current->throttle_queue);
1710 current->throttle_queue = q;
1712 current->use_memdelay = use_memdelay;
1713 set_notify_resume(current);
1717 * blkcg_add_delay - add delay to this blkg
1718 * @blkg: blkg of interest
1719 * @now: the current time in nanoseconds
1720 * @delta: how many nanoseconds of delay to add
1722 * Charge @delta to the blkg's current delay accumulation. This is used to
1723 * throttle tasks if an IO controller thinks we need more throttling.
1725 void blkcg_add_delay(struct blkcg_gq *blkg, u64 now, u64 delta)
1727 blkcg_scale_delay(blkg, now);
1728 atomic64_add(delta, &blkg->delay_nsec);
1731 module_param(blkcg_debug_stats, bool, 0644);
1732 MODULE_PARM_DESC(blkcg_debug_stats, "True if you want debug stats, false if not");