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 * For policy-specific per-blkcg data:
14 * Copyright (C) 2015 Paolo Valente <paolo.valente@unimore.it>
15 * Arianna Avanzini <avanzini.arianna@gmail.com>
17 #include <linux/ioprio.h>
18 #include <linux/kdev_t.h>
19 #include <linux/module.h>
20 #include <linux/err.h>
21 #include <linux/blkdev.h>
22 #include <linux/backing-dev.h>
23 #include <linux/slab.h>
24 #include <linux/genhd.h>
25 #include <linux/delay.h>
26 #include <linux/atomic.h>
27 #include <linux/ctype.h>
28 #include <linux/blk-cgroup.h>
31 #define MAX_KEY_LEN 100
34 * blkcg_pol_mutex protects blkcg_policy[] and policy [de]activation.
35 * blkcg_pol_register_mutex nests outside of it and synchronizes entire
36 * policy [un]register operations including cgroup file additions /
37 * removals. Putting cgroup file registration outside blkcg_pol_mutex
38 * allows grabbing it from cgroup callbacks.
40 static DEFINE_MUTEX(blkcg_pol_register_mutex);
41 static DEFINE_MUTEX(blkcg_pol_mutex);
43 struct blkcg blkcg_root;
44 EXPORT_SYMBOL_GPL(blkcg_root);
46 struct cgroup_subsys_state * const blkcg_root_css = &blkcg_root.css;
48 static struct blkcg_policy *blkcg_policy[BLKCG_MAX_POLS];
50 static LIST_HEAD(all_blkcgs); /* protected by blkcg_pol_mutex */
52 static bool blkcg_policy_enabled(struct request_queue *q,
53 const struct blkcg_policy *pol)
55 return pol && test_bit(pol->plid, q->blkcg_pols);
59 * blkg_free - free a blkg
62 * Free @blkg which may be partially allocated.
64 static void blkg_free(struct blkcg_gq *blkg)
71 for (i = 0; i < BLKCG_MAX_POLS; i++)
73 blkcg_policy[i]->pd_free_fn(blkg->pd[i]);
75 if (blkg->blkcg != &blkcg_root)
76 blk_exit_rl(&blkg->rl);
78 blkg_rwstat_exit(&blkg->stat_ios);
79 blkg_rwstat_exit(&blkg->stat_bytes);
84 * blkg_alloc - allocate a blkg
85 * @blkcg: block cgroup the new blkg is associated with
86 * @q: request_queue the new blkg is associated with
87 * @gfp_mask: allocation mask to use
89 * Allocate a new blkg assocating @blkcg and @q.
91 static struct blkcg_gq *blkg_alloc(struct blkcg *blkcg, struct request_queue *q,
94 struct blkcg_gq *blkg;
97 /* alloc and init base part */
98 blkg = kzalloc_node(sizeof(*blkg), gfp_mask, q->node);
102 if (blkg_rwstat_init(&blkg->stat_bytes, gfp_mask) ||
103 blkg_rwstat_init(&blkg->stat_ios, gfp_mask))
107 INIT_LIST_HEAD(&blkg->q_node);
109 atomic_set(&blkg->refcnt, 1);
111 /* root blkg uses @q->root_rl, init rl only for !root blkgs */
112 if (blkcg != &blkcg_root) {
113 if (blk_init_rl(&blkg->rl, q, gfp_mask))
115 blkg->rl.blkg = blkg;
118 for (i = 0; i < BLKCG_MAX_POLS; i++) {
119 struct blkcg_policy *pol = blkcg_policy[i];
120 struct blkg_policy_data *pd;
122 if (!blkcg_policy_enabled(q, pol))
125 /* alloc per-policy data and attach it to blkg */
126 pd = pol->pd_alloc_fn(gfp_mask, q->node);
142 struct blkcg_gq *blkg_lookup_slowpath(struct blkcg *blkcg,
143 struct request_queue *q, bool update_hint)
145 struct blkcg_gq *blkg;
148 * Hint didn't match. Look up from the radix tree. Note that the
149 * hint can only be updated under queue_lock as otherwise @blkg
150 * could have already been removed from blkg_tree. The caller is
151 * responsible for grabbing queue_lock if @update_hint.
153 blkg = radix_tree_lookup(&blkcg->blkg_tree, q->id);
154 if (blkg && blkg->q == q) {
156 lockdep_assert_held(q->queue_lock);
157 rcu_assign_pointer(blkcg->blkg_hint, blkg);
164 EXPORT_SYMBOL_GPL(blkg_lookup_slowpath);
167 * If @new_blkg is %NULL, this function tries to allocate a new one as
168 * necessary using %GFP_NOWAIT. @new_blkg is always consumed on return.
170 static struct blkcg_gq *blkg_create(struct blkcg *blkcg,
171 struct request_queue *q,
172 struct blkcg_gq *new_blkg)
174 struct blkcg_gq *blkg;
175 struct bdi_writeback_congested *wb_congested;
178 WARN_ON_ONCE(!rcu_read_lock_held());
179 lockdep_assert_held(q->queue_lock);
181 /* blkg holds a reference to blkcg */
182 if (!css_tryget_online(&blkcg->css)) {
187 wb_congested = wb_congested_get_create(&q->backing_dev_info,
188 blkcg->css.id, GFP_NOWAIT);
196 new_blkg = blkg_alloc(blkcg, q, GFP_NOWAIT);
197 if (unlikely(!new_blkg)) {
199 goto err_put_congested;
203 blkg->wb_congested = wb_congested;
206 if (blkcg_parent(blkcg)) {
207 blkg->parent = __blkg_lookup(blkcg_parent(blkcg), q, false);
208 if (WARN_ON_ONCE(!blkg->parent)) {
210 goto err_put_congested;
212 blkg_get(blkg->parent);
215 /* invoke per-policy init */
216 for (i = 0; i < BLKCG_MAX_POLS; i++) {
217 struct blkcg_policy *pol = blkcg_policy[i];
219 if (blkg->pd[i] && pol->pd_init_fn)
220 pol->pd_init_fn(blkg->pd[i]);
224 spin_lock(&blkcg->lock);
225 ret = radix_tree_insert(&blkcg->blkg_tree, q->id, blkg);
227 hlist_add_head_rcu(&blkg->blkcg_node, &blkcg->blkg_list);
228 list_add(&blkg->q_node, &q->blkg_list);
230 for (i = 0; i < BLKCG_MAX_POLS; i++) {
231 struct blkcg_policy *pol = blkcg_policy[i];
233 if (blkg->pd[i] && pol->pd_online_fn)
234 pol->pd_online_fn(blkg->pd[i]);
238 spin_unlock(&blkcg->lock);
243 /* @blkg failed fully initialized, use the usual release path */
248 wb_congested_put(wb_congested);
250 css_put(&blkcg->css);
257 * blkg_lookup_create - lookup blkg, try to create one if not there
258 * @blkcg: blkcg of interest
259 * @q: request_queue of interest
261 * Lookup blkg for the @blkcg - @q pair. If it doesn't exist, try to
262 * create one. blkg creation is performed recursively from blkcg_root such
263 * that all non-root blkg's have access to the parent blkg. This function
264 * should be called under RCU read lock and @q->queue_lock.
266 * Returns pointer to the looked up or created blkg on success, ERR_PTR()
267 * value on error. If @q is dead, returns ERR_PTR(-EINVAL). If @q is not
268 * dead and bypassing, returns ERR_PTR(-EBUSY).
270 struct blkcg_gq *blkg_lookup_create(struct blkcg *blkcg,
271 struct request_queue *q)
273 struct blkcg_gq *blkg;
275 WARN_ON_ONCE(!rcu_read_lock_held());
276 lockdep_assert_held(q->queue_lock);
279 * This could be the first entry point of blkcg implementation and
280 * we shouldn't allow anything to go through for a bypassing queue.
282 if (unlikely(blk_queue_bypass(q)))
283 return ERR_PTR(blk_queue_dying(q) ? -ENODEV : -EBUSY);
285 blkg = __blkg_lookup(blkcg, q, true);
290 * Create blkgs walking down from blkcg_root to @blkcg, so that all
291 * non-root blkgs have access to their parents.
294 struct blkcg *pos = blkcg;
295 struct blkcg *parent = blkcg_parent(blkcg);
297 while (parent && !__blkg_lookup(parent, q, false)) {
299 parent = blkcg_parent(parent);
302 blkg = blkg_create(pos, q, NULL);
303 if (pos == blkcg || IS_ERR(blkg))
308 static void blkg_destroy(struct blkcg_gq *blkg)
310 struct blkcg *blkcg = blkg->blkcg;
311 struct blkcg_gq *parent = blkg->parent;
314 lockdep_assert_held(blkg->q->queue_lock);
315 lockdep_assert_held(&blkcg->lock);
317 /* Something wrong if we are trying to remove same group twice */
318 WARN_ON_ONCE(list_empty(&blkg->q_node));
319 WARN_ON_ONCE(hlist_unhashed(&blkg->blkcg_node));
321 for (i = 0; i < BLKCG_MAX_POLS; i++) {
322 struct blkcg_policy *pol = blkcg_policy[i];
324 if (blkg->pd[i] && pol->pd_offline_fn)
325 pol->pd_offline_fn(blkg->pd[i]);
329 blkg_rwstat_add_aux(&parent->stat_bytes, &blkg->stat_bytes);
330 blkg_rwstat_add_aux(&parent->stat_ios, &blkg->stat_ios);
333 blkg->online = false;
335 radix_tree_delete(&blkcg->blkg_tree, blkg->q->id);
336 list_del_init(&blkg->q_node);
337 hlist_del_init_rcu(&blkg->blkcg_node);
340 * Both setting lookup hint to and clearing it from @blkg are done
341 * under queue_lock. If it's not pointing to @blkg now, it never
342 * will. Hint assignment itself can race safely.
344 if (rcu_access_pointer(blkcg->blkg_hint) == blkg)
345 rcu_assign_pointer(blkcg->blkg_hint, NULL);
348 * Put the reference taken at the time of creation so that when all
349 * queues are gone, group can be destroyed.
355 * blkg_destroy_all - destroy all blkgs associated with a request_queue
356 * @q: request_queue of interest
358 * Destroy all blkgs associated with @q.
360 static void blkg_destroy_all(struct request_queue *q)
362 struct blkcg_gq *blkg, *n;
364 lockdep_assert_held(q->queue_lock);
366 list_for_each_entry_safe(blkg, n, &q->blkg_list, q_node) {
367 struct blkcg *blkcg = blkg->blkcg;
369 spin_lock(&blkcg->lock);
371 spin_unlock(&blkcg->lock);
376 * A group is RCU protected, but having an rcu lock does not mean that one
377 * can access all the fields of blkg and assume these are valid. For
378 * example, don't try to follow throtl_data and request queue links.
380 * Having a reference to blkg under an rcu allows accesses to only values
381 * local to groups like group stats and group rate limits.
383 void __blkg_release_rcu(struct rcu_head *rcu_head)
385 struct blkcg_gq *blkg = container_of(rcu_head, struct blkcg_gq, rcu_head);
387 /* release the blkcg and parent blkg refs this blkg has been holding */
388 css_put(&blkg->blkcg->css);
390 blkg_put(blkg->parent);
392 wb_congested_put(blkg->wb_congested);
396 EXPORT_SYMBOL_GPL(__blkg_release_rcu);
399 * The next function used by blk_queue_for_each_rl(). It's a bit tricky
400 * because the root blkg uses @q->root_rl instead of its own rl.
402 struct request_list *__blk_queue_next_rl(struct request_list *rl,
403 struct request_queue *q)
405 struct list_head *ent;
406 struct blkcg_gq *blkg;
409 * Determine the current blkg list_head. The first entry is
410 * root_rl which is off @q->blkg_list and mapped to the head.
412 if (rl == &q->root_rl) {
414 /* There are no more block groups, hence no request lists */
418 blkg = container_of(rl, struct blkcg_gq, rl);
422 /* walk to the next list_head, skip root blkcg */
424 if (ent == &q->root_blkg->q_node)
426 if (ent == &q->blkg_list)
429 blkg = container_of(ent, struct blkcg_gq, q_node);
433 static int blkcg_reset_stats(struct cgroup_subsys_state *css,
434 struct cftype *cftype, u64 val)
436 struct blkcg *blkcg = css_to_blkcg(css);
437 struct blkcg_gq *blkg;
440 mutex_lock(&blkcg_pol_mutex);
441 spin_lock_irq(&blkcg->lock);
444 * Note that stat reset is racy - it doesn't synchronize against
445 * stat updates. This is a debug feature which shouldn't exist
446 * anyway. If you get hit by a race, retry.
448 hlist_for_each_entry(blkg, &blkcg->blkg_list, blkcg_node) {
449 blkg_rwstat_reset(&blkg->stat_bytes);
450 blkg_rwstat_reset(&blkg->stat_ios);
452 for (i = 0; i < BLKCG_MAX_POLS; i++) {
453 struct blkcg_policy *pol = blkcg_policy[i];
455 if (blkg->pd[i] && pol->pd_reset_stats_fn)
456 pol->pd_reset_stats_fn(blkg->pd[i]);
460 spin_unlock_irq(&blkcg->lock);
461 mutex_unlock(&blkcg_pol_mutex);
465 static const char *blkg_dev_name(struct blkcg_gq *blkg)
467 /* some drivers (floppy) instantiate a queue w/o disk registered */
468 if (blkg->q->backing_dev_info.dev)
469 return dev_name(blkg->q->backing_dev_info.dev);
474 * blkcg_print_blkgs - helper for printing per-blkg data
475 * @sf: seq_file to print to
476 * @blkcg: blkcg of interest
477 * @prfill: fill function to print out a blkg
478 * @pol: policy in question
479 * @data: data to be passed to @prfill
480 * @show_total: to print out sum of prfill return values or not
482 * This function invokes @prfill on each blkg of @blkcg if pd for the
483 * policy specified by @pol exists. @prfill is invoked with @sf, the
484 * policy data and @data and the matching queue lock held. If @show_total
485 * is %true, the sum of the return values from @prfill is printed with
486 * "Total" label at the end.
488 * This is to be used to construct print functions for
489 * cftype->read_seq_string method.
491 void blkcg_print_blkgs(struct seq_file *sf, struct blkcg *blkcg,
492 u64 (*prfill)(struct seq_file *,
493 struct blkg_policy_data *, int),
494 const struct blkcg_policy *pol, int data,
497 struct blkcg_gq *blkg;
501 hlist_for_each_entry_rcu(blkg, &blkcg->blkg_list, blkcg_node) {
502 spin_lock_irq(blkg->q->queue_lock);
503 if (blkcg_policy_enabled(blkg->q, pol))
504 total += prfill(sf, blkg->pd[pol->plid], data);
505 spin_unlock_irq(blkg->q->queue_lock);
510 seq_printf(sf, "Total %llu\n", (unsigned long long)total);
512 EXPORT_SYMBOL_GPL(blkcg_print_blkgs);
515 * __blkg_prfill_u64 - prfill helper for a single u64 value
516 * @sf: seq_file to print to
517 * @pd: policy private data of interest
520 * Print @v to @sf for the device assocaited with @pd.
522 u64 __blkg_prfill_u64(struct seq_file *sf, struct blkg_policy_data *pd, u64 v)
524 const char *dname = blkg_dev_name(pd->blkg);
529 seq_printf(sf, "%s %llu\n", dname, (unsigned long long)v);
532 EXPORT_SYMBOL_GPL(__blkg_prfill_u64);
535 * __blkg_prfill_rwstat - prfill helper for a blkg_rwstat
536 * @sf: seq_file to print to
537 * @pd: policy private data of interest
538 * @rwstat: rwstat to print
540 * Print @rwstat to @sf for the device assocaited with @pd.
542 u64 __blkg_prfill_rwstat(struct seq_file *sf, struct blkg_policy_data *pd,
543 const struct blkg_rwstat *rwstat)
545 static const char *rwstr[] = {
546 [BLKG_RWSTAT_READ] = "Read",
547 [BLKG_RWSTAT_WRITE] = "Write",
548 [BLKG_RWSTAT_SYNC] = "Sync",
549 [BLKG_RWSTAT_ASYNC] = "Async",
551 const char *dname = blkg_dev_name(pd->blkg);
558 for (i = 0; i < BLKG_RWSTAT_NR; i++)
559 seq_printf(sf, "%s %s %llu\n", dname, rwstr[i],
560 (unsigned long long)atomic64_read(&rwstat->aux_cnt[i]));
562 v = atomic64_read(&rwstat->aux_cnt[BLKG_RWSTAT_READ]) +
563 atomic64_read(&rwstat->aux_cnt[BLKG_RWSTAT_WRITE]);
564 seq_printf(sf, "%s Total %llu\n", dname, (unsigned long long)v);
567 EXPORT_SYMBOL_GPL(__blkg_prfill_rwstat);
570 * blkg_prfill_stat - prfill callback for blkg_stat
571 * @sf: seq_file to print to
572 * @pd: policy private data of interest
573 * @off: offset to the blkg_stat in @pd
575 * prfill callback for printing a blkg_stat.
577 u64 blkg_prfill_stat(struct seq_file *sf, struct blkg_policy_data *pd, int off)
579 return __blkg_prfill_u64(sf, pd, blkg_stat_read((void *)pd + off));
581 EXPORT_SYMBOL_GPL(blkg_prfill_stat);
584 * blkg_prfill_rwstat - prfill callback for blkg_rwstat
585 * @sf: seq_file to print to
586 * @pd: policy private data of interest
587 * @off: offset to the blkg_rwstat in @pd
589 * prfill callback for printing a blkg_rwstat.
591 u64 blkg_prfill_rwstat(struct seq_file *sf, struct blkg_policy_data *pd,
594 struct blkg_rwstat rwstat = blkg_rwstat_read((void *)pd + off);
596 return __blkg_prfill_rwstat(sf, pd, &rwstat);
598 EXPORT_SYMBOL_GPL(blkg_prfill_rwstat);
600 static u64 blkg_prfill_rwstat_field(struct seq_file *sf,
601 struct blkg_policy_data *pd, int off)
603 struct blkg_rwstat rwstat = blkg_rwstat_read((void *)pd->blkg + off);
605 return __blkg_prfill_rwstat(sf, pd, &rwstat);
609 * blkg_print_stat_bytes - seq_show callback for blkg->stat_bytes
610 * @sf: seq_file to print to
613 * To be used as cftype->seq_show to print blkg->stat_bytes.
614 * cftype->private must be set to the blkcg_policy.
616 int blkg_print_stat_bytes(struct seq_file *sf, void *v)
618 blkcg_print_blkgs(sf, css_to_blkcg(seq_css(sf)),
619 blkg_prfill_rwstat_field, (void *)seq_cft(sf)->private,
620 offsetof(struct blkcg_gq, stat_bytes), true);
623 EXPORT_SYMBOL_GPL(blkg_print_stat_bytes);
626 * blkg_print_stat_bytes - seq_show callback for blkg->stat_ios
627 * @sf: seq_file to print to
630 * To be used as cftype->seq_show to print blkg->stat_ios. cftype->private
631 * must be set to the blkcg_policy.
633 int blkg_print_stat_ios(struct seq_file *sf, void *v)
635 blkcg_print_blkgs(sf, css_to_blkcg(seq_css(sf)),
636 blkg_prfill_rwstat_field, (void *)seq_cft(sf)->private,
637 offsetof(struct blkcg_gq, stat_ios), true);
640 EXPORT_SYMBOL_GPL(blkg_print_stat_ios);
642 static u64 blkg_prfill_rwstat_field_recursive(struct seq_file *sf,
643 struct blkg_policy_data *pd,
646 struct blkg_rwstat rwstat = blkg_rwstat_recursive_sum(pd->blkg,
648 return __blkg_prfill_rwstat(sf, pd, &rwstat);
652 * blkg_print_stat_bytes_recursive - recursive version of blkg_print_stat_bytes
653 * @sf: seq_file to print to
656 int blkg_print_stat_bytes_recursive(struct seq_file *sf, void *v)
658 blkcg_print_blkgs(sf, css_to_blkcg(seq_css(sf)),
659 blkg_prfill_rwstat_field_recursive,
660 (void *)seq_cft(sf)->private,
661 offsetof(struct blkcg_gq, stat_bytes), true);
664 EXPORT_SYMBOL_GPL(blkg_print_stat_bytes_recursive);
667 * blkg_print_stat_ios_recursive - recursive version of blkg_print_stat_ios
668 * @sf: seq_file to print to
671 int blkg_print_stat_ios_recursive(struct seq_file *sf, void *v)
673 blkcg_print_blkgs(sf, css_to_blkcg(seq_css(sf)),
674 blkg_prfill_rwstat_field_recursive,
675 (void *)seq_cft(sf)->private,
676 offsetof(struct blkcg_gq, stat_ios), true);
679 EXPORT_SYMBOL_GPL(blkg_print_stat_ios_recursive);
682 * blkg_stat_recursive_sum - collect hierarchical blkg_stat
683 * @blkg: blkg of interest
684 * @pol: blkcg_policy which contains the blkg_stat
685 * @off: offset to the blkg_stat in blkg_policy_data or @blkg
687 * Collect the blkg_stat specified by @blkg, @pol and @off and all its
688 * online descendants and their aux counts. The caller must be holding the
689 * queue lock for online tests.
691 * If @pol is NULL, blkg_stat is at @off bytes into @blkg; otherwise, it is
692 * at @off bytes into @blkg's blkg_policy_data of the policy.
694 u64 blkg_stat_recursive_sum(struct blkcg_gq *blkg,
695 struct blkcg_policy *pol, int off)
697 struct blkcg_gq *pos_blkg;
698 struct cgroup_subsys_state *pos_css;
701 lockdep_assert_held(blkg->q->queue_lock);
704 blkg_for_each_descendant_pre(pos_blkg, pos_css, blkg) {
705 struct blkg_stat *stat;
707 if (!pos_blkg->online)
711 stat = (void *)blkg_to_pd(pos_blkg, pol) + off;
713 stat = (void *)blkg + off;
715 sum += blkg_stat_read(stat) + atomic64_read(&stat->aux_cnt);
721 EXPORT_SYMBOL_GPL(blkg_stat_recursive_sum);
724 * blkg_rwstat_recursive_sum - collect hierarchical blkg_rwstat
725 * @blkg: blkg of interest
726 * @pol: blkcg_policy which contains the blkg_rwstat
727 * @off: offset to the blkg_rwstat in blkg_policy_data or @blkg
729 * Collect the blkg_rwstat specified by @blkg, @pol and @off and all its
730 * online descendants and their aux counts. The caller must be holding the
731 * queue lock for online tests.
733 * If @pol is NULL, blkg_rwstat is at @off bytes into @blkg; otherwise, it
734 * is at @off bytes into @blkg's blkg_policy_data of the policy.
736 struct blkg_rwstat blkg_rwstat_recursive_sum(struct blkcg_gq *blkg,
737 struct blkcg_policy *pol, int off)
739 struct blkcg_gq *pos_blkg;
740 struct cgroup_subsys_state *pos_css;
741 struct blkg_rwstat sum = { };
744 lockdep_assert_held(blkg->q->queue_lock);
747 blkg_for_each_descendant_pre(pos_blkg, pos_css, blkg) {
748 struct blkg_rwstat *rwstat;
750 if (!pos_blkg->online)
754 rwstat = (void *)blkg_to_pd(pos_blkg, pol) + off;
756 rwstat = (void *)pos_blkg + off;
758 for (i = 0; i < BLKG_RWSTAT_NR; i++)
759 atomic64_add(atomic64_read(&rwstat->aux_cnt[i]) +
760 percpu_counter_sum_positive(&rwstat->cpu_cnt[i]),
767 EXPORT_SYMBOL_GPL(blkg_rwstat_recursive_sum);
770 * blkg_conf_prep - parse and prepare for per-blkg config update
771 * @blkcg: target block cgroup
772 * @pol: target policy
773 * @input: input string
774 * @ctx: blkg_conf_ctx to be filled
776 * Parse per-blkg config update from @input and initialize @ctx with the
777 * result. @ctx->blkg points to the blkg to be updated and @ctx->body the
778 * part of @input following MAJ:MIN. This function returns with RCU read
779 * lock and queue lock held and must be paired with blkg_conf_finish().
781 int blkg_conf_prep(struct blkcg *blkcg, const struct blkcg_policy *pol,
782 char *input, struct blkg_conf_ctx *ctx)
783 __acquires(rcu) __acquires(disk->queue->queue_lock)
785 struct gendisk *disk;
786 struct blkcg_gq *blkg;
787 unsigned int major, minor;
788 int key_len, part, ret;
791 if (sscanf(input, "%u:%u%n", &major, &minor, &key_len) != 2)
794 body = input + key_len;
797 body = skip_spaces(body);
799 disk = get_gendisk(MKDEV(major, minor), &part);
808 spin_lock_irq(disk->queue->queue_lock);
810 if (blkcg_policy_enabled(disk->queue, pol))
811 blkg = blkg_lookup_create(blkcg, disk->queue);
813 blkg = ERR_PTR(-EOPNOTSUPP);
818 spin_unlock_irq(disk->queue->queue_lock);
821 * If queue was bypassing, we should retry. Do so after a
822 * short msleep(). It isn't strictly necessary but queue
823 * can be bypassing for some time and it's always nice to
824 * avoid busy looping.
828 ret = restart_syscall();
838 EXPORT_SYMBOL_GPL(blkg_conf_prep);
841 * blkg_conf_finish - finish up per-blkg config update
842 * @ctx: blkg_conf_ctx intiailized by blkg_conf_prep()
844 * Finish up after per-blkg config update. This function must be paired
845 * with blkg_conf_prep().
847 void blkg_conf_finish(struct blkg_conf_ctx *ctx)
848 __releases(ctx->disk->queue->queue_lock) __releases(rcu)
850 spin_unlock_irq(ctx->disk->queue->queue_lock);
854 EXPORT_SYMBOL_GPL(blkg_conf_finish);
856 struct cftype blkcg_legacy_files[] = {
858 .name = "reset_stats",
859 .write_u64 = blkcg_reset_stats,
865 * blkcg_css_offline - cgroup css_offline callback
866 * @css: css of interest
868 * This function is called when @css is about to go away and responsible
869 * for shooting down all blkgs associated with @css. blkgs should be
870 * removed while holding both q and blkcg locks. As blkcg lock is nested
871 * inside q lock, this function performs reverse double lock dancing.
873 * This is the blkcg counterpart of ioc_release_fn().
875 static void blkcg_css_offline(struct cgroup_subsys_state *css)
877 struct blkcg *blkcg = css_to_blkcg(css);
879 spin_lock_irq(&blkcg->lock);
881 while (!hlist_empty(&blkcg->blkg_list)) {
882 struct blkcg_gq *blkg = hlist_entry(blkcg->blkg_list.first,
883 struct blkcg_gq, blkcg_node);
884 struct request_queue *q = blkg->q;
886 if (spin_trylock(q->queue_lock)) {
888 spin_unlock(q->queue_lock);
890 spin_unlock_irq(&blkcg->lock);
892 spin_lock_irq(&blkcg->lock);
896 spin_unlock_irq(&blkcg->lock);
898 wb_blkcg_offline(blkcg);
901 static void blkcg_css_free(struct cgroup_subsys_state *css)
903 struct blkcg *blkcg = css_to_blkcg(css);
906 mutex_lock(&blkcg_pol_mutex);
908 list_del(&blkcg->all_blkcgs_node);
910 for (i = 0; i < BLKCG_MAX_POLS; i++)
912 blkcg_policy[i]->cpd_free_fn(blkcg->cpd[i]);
914 mutex_unlock(&blkcg_pol_mutex);
919 static struct cgroup_subsys_state *
920 blkcg_css_alloc(struct cgroup_subsys_state *parent_css)
923 struct cgroup_subsys_state *ret;
926 mutex_lock(&blkcg_pol_mutex);
931 blkcg = kzalloc(sizeof(*blkcg), GFP_KERNEL);
933 ret = ERR_PTR(-ENOMEM);
938 for (i = 0; i < BLKCG_MAX_POLS ; i++) {
939 struct blkcg_policy *pol = blkcg_policy[i];
940 struct blkcg_policy_data *cpd;
943 * If the policy hasn't been attached yet, wait for it
944 * to be attached before doing anything else. Otherwise,
945 * check if the policy requires any specific per-cgroup
946 * data: if it does, allocate and initialize it.
948 if (!pol || !pol->cpd_alloc_fn)
951 cpd = pol->cpd_alloc_fn(GFP_KERNEL);
953 ret = ERR_PTR(-ENOMEM);
959 if (pol->cpd_init_fn)
960 pol->cpd_init_fn(cpd);
963 spin_lock_init(&blkcg->lock);
964 INIT_RADIX_TREE(&blkcg->blkg_tree, GFP_NOWAIT);
965 INIT_HLIST_HEAD(&blkcg->blkg_list);
966 #ifdef CONFIG_CGROUP_WRITEBACK
967 INIT_LIST_HEAD(&blkcg->cgwb_list);
969 list_add_tail(&blkcg->all_blkcgs_node, &all_blkcgs);
971 mutex_unlock(&blkcg_pol_mutex);
975 for (i--; i >= 0; i--)
977 blkcg_policy[i]->cpd_free_fn(blkcg->cpd[i]);
980 mutex_unlock(&blkcg_pol_mutex);
985 * blkcg_init_queue - initialize blkcg part of request queue
986 * @q: request_queue to initialize
988 * Called from blk_alloc_queue_node(). Responsible for initializing blkcg
989 * part of new request_queue @q.
992 * 0 on success, -errno on failure.
994 int blkcg_init_queue(struct request_queue *q)
996 struct blkcg_gq *new_blkg, *blkg;
1000 new_blkg = blkg_alloc(&blkcg_root, q, GFP_KERNEL);
1004 preloaded = !radix_tree_preload(GFP_KERNEL);
1007 * Make sure the root blkg exists and count the existing blkgs. As
1008 * @q is bypassing at this point, blkg_lookup_create() can't be
1009 * used. Open code insertion.
1012 spin_lock_irq(q->queue_lock);
1013 blkg = blkg_create(&blkcg_root, q, new_blkg);
1014 spin_unlock_irq(q->queue_lock);
1018 radix_tree_preload_end();
1021 blkg_free(new_blkg);
1022 return PTR_ERR(blkg);
1025 q->root_blkg = blkg;
1026 q->root_rl.blkg = blkg;
1028 ret = blk_throtl_init(q);
1030 spin_lock_irq(q->queue_lock);
1031 blkg_destroy_all(q);
1032 spin_unlock_irq(q->queue_lock);
1038 * blkcg_drain_queue - drain blkcg part of request_queue
1039 * @q: request_queue to drain
1041 * Called from blk_drain_queue(). Responsible for draining blkcg part.
1043 void blkcg_drain_queue(struct request_queue *q)
1045 lockdep_assert_held(q->queue_lock);
1048 * @q could be exiting and already have destroyed all blkgs as
1049 * indicated by NULL root_blkg. If so, don't confuse policies.
1054 blk_throtl_drain(q);
1058 * blkcg_exit_queue - exit and release blkcg part of request_queue
1059 * @q: request_queue being released
1061 * Called from blk_release_queue(). Responsible for exiting blkcg part.
1063 void blkcg_exit_queue(struct request_queue *q)
1065 spin_lock_irq(q->queue_lock);
1066 blkg_destroy_all(q);
1067 spin_unlock_irq(q->queue_lock);
1073 * We cannot support shared io contexts, as we have no mean to support
1074 * two tasks with the same ioc in two different groups without major rework
1075 * of the main cic data structures. For now we allow a task to change
1076 * its cgroup only if it's the only owner of its ioc.
1078 static int blkcg_can_attach(struct cgroup_subsys_state *css,
1079 struct cgroup_taskset *tset)
1081 struct task_struct *task;
1082 struct io_context *ioc;
1085 /* task_lock() is needed to avoid races with exit_io_context() */
1086 cgroup_taskset_for_each(task, tset) {
1088 ioc = task->io_context;
1089 if (ioc && atomic_read(&ioc->nr_tasks) > 1)
1098 struct cgroup_subsys io_cgrp_subsys = {
1099 .css_alloc = blkcg_css_alloc,
1100 .css_offline = blkcg_css_offline,
1101 .css_free = blkcg_css_free,
1102 .can_attach = blkcg_can_attach,
1103 .legacy_cftypes = blkcg_legacy_files,
1104 .legacy_name = "blkio",
1107 * This ensures that, if available, memcg is automatically enabled
1108 * together on the default hierarchy so that the owner cgroup can
1109 * be retrieved from writeback pages.
1111 .depends_on = 1 << memory_cgrp_id,
1114 EXPORT_SYMBOL_GPL(io_cgrp_subsys);
1117 * blkcg_activate_policy - activate a blkcg policy on a request_queue
1118 * @q: request_queue of interest
1119 * @pol: blkcg policy to activate
1121 * Activate @pol on @q. Requires %GFP_KERNEL context. @q goes through
1122 * bypass mode to populate its blkgs with policy_data for @pol.
1124 * Activation happens with @q bypassed, so nobody would be accessing blkgs
1125 * from IO path. Update of each blkg is protected by both queue and blkcg
1126 * locks so that holding either lock and testing blkcg_policy_enabled() is
1127 * always enough for dereferencing policy data.
1129 * The caller is responsible for synchronizing [de]activations and policy
1130 * [un]registerations. Returns 0 on success, -errno on failure.
1132 int blkcg_activate_policy(struct request_queue *q,
1133 const struct blkcg_policy *pol)
1135 struct blkg_policy_data *pd_prealloc = NULL;
1136 struct blkcg_gq *blkg;
1139 if (blkcg_policy_enabled(q, pol))
1142 blk_queue_bypass_start(q);
1145 pd_prealloc = pol->pd_alloc_fn(GFP_KERNEL, q->node);
1148 goto out_bypass_end;
1152 spin_lock_irq(q->queue_lock);
1154 list_for_each_entry(blkg, &q->blkg_list, q_node) {
1155 struct blkg_policy_data *pd;
1157 if (blkg->pd[pol->plid])
1160 pd = pol->pd_alloc_fn(GFP_NOWAIT, q->node);
1162 swap(pd, pd_prealloc);
1164 spin_unlock_irq(q->queue_lock);
1168 blkg->pd[pol->plid] = pd;
1170 pd->plid = pol->plid;
1171 if (pol->pd_init_fn)
1172 pol->pd_init_fn(pd);
1175 __set_bit(pol->plid, q->blkcg_pols);
1178 spin_unlock_irq(q->queue_lock);
1180 blk_queue_bypass_end(q);
1182 pol->pd_free_fn(pd_prealloc);
1185 EXPORT_SYMBOL_GPL(blkcg_activate_policy);
1188 * blkcg_deactivate_policy - deactivate a blkcg policy on a request_queue
1189 * @q: request_queue of interest
1190 * @pol: blkcg policy to deactivate
1192 * Deactivate @pol on @q. Follows the same synchronization rules as
1193 * blkcg_activate_policy().
1195 void blkcg_deactivate_policy(struct request_queue *q,
1196 const struct blkcg_policy *pol)
1198 struct blkcg_gq *blkg;
1200 if (!blkcg_policy_enabled(q, pol))
1203 blk_queue_bypass_start(q);
1204 spin_lock_irq(q->queue_lock);
1206 __clear_bit(pol->plid, q->blkcg_pols);
1208 list_for_each_entry(blkg, &q->blkg_list, q_node) {
1209 /* grab blkcg lock too while removing @pd from @blkg */
1210 spin_lock(&blkg->blkcg->lock);
1212 if (blkg->pd[pol->plid]) {
1213 if (pol->pd_offline_fn)
1214 pol->pd_offline_fn(blkg->pd[pol->plid]);
1215 pol->pd_free_fn(blkg->pd[pol->plid]);
1216 blkg->pd[pol->plid] = NULL;
1219 spin_unlock(&blkg->blkcg->lock);
1222 spin_unlock_irq(q->queue_lock);
1223 blk_queue_bypass_end(q);
1225 EXPORT_SYMBOL_GPL(blkcg_deactivate_policy);
1228 * blkcg_policy_register - register a blkcg policy
1229 * @pol: blkcg policy to register
1231 * Register @pol with blkcg core. Might sleep and @pol may be modified on
1232 * successful registration. Returns 0 on success and -errno on failure.
1234 int blkcg_policy_register(struct blkcg_policy *pol)
1236 struct blkcg *blkcg;
1239 mutex_lock(&blkcg_pol_register_mutex);
1240 mutex_lock(&blkcg_pol_mutex);
1242 /* find an empty slot */
1244 for (i = 0; i < BLKCG_MAX_POLS; i++)
1245 if (!blkcg_policy[i])
1247 if (i >= BLKCG_MAX_POLS)
1252 blkcg_policy[pol->plid] = pol;
1254 /* allocate and install cpd's */
1255 if (pol->cpd_alloc_fn) {
1256 list_for_each_entry(blkcg, &all_blkcgs, all_blkcgs_node) {
1257 struct blkcg_policy_data *cpd;
1259 cpd = pol->cpd_alloc_fn(GFP_KERNEL);
1261 mutex_unlock(&blkcg_pol_mutex);
1265 blkcg->cpd[pol->plid] = cpd;
1267 cpd->plid = pol->plid;
1268 pol->cpd_init_fn(cpd);
1272 mutex_unlock(&blkcg_pol_mutex);
1274 /* everything is in place, add intf files for the new policy */
1275 if (pol->legacy_cftypes)
1276 WARN_ON(cgroup_add_legacy_cftypes(&io_cgrp_subsys,
1277 pol->legacy_cftypes));
1278 mutex_unlock(&blkcg_pol_register_mutex);
1282 if (pol->cpd_alloc_fn) {
1283 list_for_each_entry(blkcg, &all_blkcgs, all_blkcgs_node) {
1284 if (blkcg->cpd[pol->plid]) {
1285 pol->cpd_free_fn(blkcg->cpd[pol->plid]);
1286 blkcg->cpd[pol->plid] = NULL;
1290 blkcg_policy[pol->plid] = NULL;
1292 mutex_unlock(&blkcg_pol_mutex);
1293 mutex_unlock(&blkcg_pol_register_mutex);
1296 EXPORT_SYMBOL_GPL(blkcg_policy_register);
1299 * blkcg_policy_unregister - unregister a blkcg policy
1300 * @pol: blkcg policy to unregister
1302 * Undo blkcg_policy_register(@pol). Might sleep.
1304 void blkcg_policy_unregister(struct blkcg_policy *pol)
1306 struct blkcg *blkcg;
1308 mutex_lock(&blkcg_pol_register_mutex);
1310 if (WARN_ON(blkcg_policy[pol->plid] != pol))
1313 /* kill the intf files first */
1314 if (pol->legacy_cftypes)
1315 cgroup_rm_cftypes(pol->legacy_cftypes);
1317 /* remove cpds and unregister */
1318 mutex_lock(&blkcg_pol_mutex);
1320 if (pol->cpd_alloc_fn) {
1321 list_for_each_entry(blkcg, &all_blkcgs, all_blkcgs_node) {
1322 if (blkcg->cpd[pol->plid]) {
1323 pol->cpd_free_fn(blkcg->cpd[pol->plid]);
1324 blkcg->cpd[pol->plid] = NULL;
1328 blkcg_policy[pol->plid] = NULL;
1330 mutex_unlock(&blkcg_pol_mutex);
1332 mutex_unlock(&blkcg_pol_register_mutex);
1334 EXPORT_SYMBOL_GPL(blkcg_policy_unregister);