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);
375 q->root_rl.blkg = NULL;
379 * A group is RCU protected, but having an rcu lock does not mean that one
380 * can access all the fields of blkg and assume these are valid. For
381 * example, don't try to follow throtl_data and request queue links.
383 * Having a reference to blkg under an rcu allows accesses to only values
384 * local to groups like group stats and group rate limits.
386 void __blkg_release_rcu(struct rcu_head *rcu_head)
388 struct blkcg_gq *blkg = container_of(rcu_head, struct blkcg_gq, rcu_head);
390 /* release the blkcg and parent blkg refs this blkg has been holding */
391 css_put(&blkg->blkcg->css);
393 blkg_put(blkg->parent);
395 wb_congested_put(blkg->wb_congested);
399 EXPORT_SYMBOL_GPL(__blkg_release_rcu);
402 * The next function used by blk_queue_for_each_rl(). It's a bit tricky
403 * because the root blkg uses @q->root_rl instead of its own rl.
405 struct request_list *__blk_queue_next_rl(struct request_list *rl,
406 struct request_queue *q)
408 struct list_head *ent;
409 struct blkcg_gq *blkg;
412 * Determine the current blkg list_head. The first entry is
413 * root_rl which is off @q->blkg_list and mapped to the head.
415 if (rl == &q->root_rl) {
417 /* There are no more block groups, hence no request lists */
421 blkg = container_of(rl, struct blkcg_gq, rl);
425 /* walk to the next list_head, skip root blkcg */
427 if (ent == &q->root_blkg->q_node)
429 if (ent == &q->blkg_list)
432 blkg = container_of(ent, struct blkcg_gq, q_node);
436 static int blkcg_reset_stats(struct cgroup_subsys_state *css,
437 struct cftype *cftype, u64 val)
439 struct blkcg *blkcg = css_to_blkcg(css);
440 struct blkcg_gq *blkg;
443 mutex_lock(&blkcg_pol_mutex);
444 spin_lock_irq(&blkcg->lock);
447 * Note that stat reset is racy - it doesn't synchronize against
448 * stat updates. This is a debug feature which shouldn't exist
449 * anyway. If you get hit by a race, retry.
451 hlist_for_each_entry(blkg, &blkcg->blkg_list, blkcg_node) {
452 blkg_rwstat_reset(&blkg->stat_bytes);
453 blkg_rwstat_reset(&blkg->stat_ios);
455 for (i = 0; i < BLKCG_MAX_POLS; i++) {
456 struct blkcg_policy *pol = blkcg_policy[i];
458 if (blkg->pd[i] && pol->pd_reset_stats_fn)
459 pol->pd_reset_stats_fn(blkg->pd[i]);
463 spin_unlock_irq(&blkcg->lock);
464 mutex_unlock(&blkcg_pol_mutex);
468 const char *blkg_dev_name(struct blkcg_gq *blkg)
470 /* some drivers (floppy) instantiate a queue w/o disk registered */
471 if (blkg->q->backing_dev_info.dev)
472 return dev_name(blkg->q->backing_dev_info.dev);
475 EXPORT_SYMBOL_GPL(blkg_dev_name);
478 * blkcg_print_blkgs - helper for printing per-blkg data
479 * @sf: seq_file to print to
480 * @blkcg: blkcg of interest
481 * @prfill: fill function to print out a blkg
482 * @pol: policy in question
483 * @data: data to be passed to @prfill
484 * @show_total: to print out sum of prfill return values or not
486 * This function invokes @prfill on each blkg of @blkcg if pd for the
487 * policy specified by @pol exists. @prfill is invoked with @sf, the
488 * policy data and @data and the matching queue lock held. If @show_total
489 * is %true, the sum of the return values from @prfill is printed with
490 * "Total" label at the end.
492 * This is to be used to construct print functions for
493 * cftype->read_seq_string method.
495 void blkcg_print_blkgs(struct seq_file *sf, struct blkcg *blkcg,
496 u64 (*prfill)(struct seq_file *,
497 struct blkg_policy_data *, int),
498 const struct blkcg_policy *pol, int data,
501 struct blkcg_gq *blkg;
505 hlist_for_each_entry_rcu(blkg, &blkcg->blkg_list, blkcg_node) {
506 spin_lock_irq(blkg->q->queue_lock);
507 if (blkcg_policy_enabled(blkg->q, pol))
508 total += prfill(sf, blkg->pd[pol->plid], data);
509 spin_unlock_irq(blkg->q->queue_lock);
514 seq_printf(sf, "Total %llu\n", (unsigned long long)total);
516 EXPORT_SYMBOL_GPL(blkcg_print_blkgs);
519 * __blkg_prfill_u64 - prfill helper for a single u64 value
520 * @sf: seq_file to print to
521 * @pd: policy private data of interest
524 * Print @v to @sf for the device assocaited with @pd.
526 u64 __blkg_prfill_u64(struct seq_file *sf, struct blkg_policy_data *pd, u64 v)
528 const char *dname = blkg_dev_name(pd->blkg);
533 seq_printf(sf, "%s %llu\n", dname, (unsigned long long)v);
536 EXPORT_SYMBOL_GPL(__blkg_prfill_u64);
539 * __blkg_prfill_rwstat - prfill helper for a blkg_rwstat
540 * @sf: seq_file to print to
541 * @pd: policy private data of interest
542 * @rwstat: rwstat to print
544 * Print @rwstat to @sf for the device assocaited with @pd.
546 u64 __blkg_prfill_rwstat(struct seq_file *sf, struct blkg_policy_data *pd,
547 const struct blkg_rwstat *rwstat)
549 static const char *rwstr[] = {
550 [BLKG_RWSTAT_READ] = "Read",
551 [BLKG_RWSTAT_WRITE] = "Write",
552 [BLKG_RWSTAT_SYNC] = "Sync",
553 [BLKG_RWSTAT_ASYNC] = "Async",
555 const char *dname = blkg_dev_name(pd->blkg);
562 for (i = 0; i < BLKG_RWSTAT_NR; i++)
563 seq_printf(sf, "%s %s %llu\n", dname, rwstr[i],
564 (unsigned long long)atomic64_read(&rwstat->aux_cnt[i]));
566 v = atomic64_read(&rwstat->aux_cnt[BLKG_RWSTAT_READ]) +
567 atomic64_read(&rwstat->aux_cnt[BLKG_RWSTAT_WRITE]);
568 seq_printf(sf, "%s Total %llu\n", dname, (unsigned long long)v);
571 EXPORT_SYMBOL_GPL(__blkg_prfill_rwstat);
574 * blkg_prfill_stat - prfill callback for blkg_stat
575 * @sf: seq_file to print to
576 * @pd: policy private data of interest
577 * @off: offset to the blkg_stat in @pd
579 * prfill callback for printing a blkg_stat.
581 u64 blkg_prfill_stat(struct seq_file *sf, struct blkg_policy_data *pd, int off)
583 return __blkg_prfill_u64(sf, pd, blkg_stat_read((void *)pd + off));
585 EXPORT_SYMBOL_GPL(blkg_prfill_stat);
588 * blkg_prfill_rwstat - prfill callback for blkg_rwstat
589 * @sf: seq_file to print to
590 * @pd: policy private data of interest
591 * @off: offset to the blkg_rwstat in @pd
593 * prfill callback for printing a blkg_rwstat.
595 u64 blkg_prfill_rwstat(struct seq_file *sf, struct blkg_policy_data *pd,
598 struct blkg_rwstat rwstat = blkg_rwstat_read((void *)pd + off);
600 return __blkg_prfill_rwstat(sf, pd, &rwstat);
602 EXPORT_SYMBOL_GPL(blkg_prfill_rwstat);
604 static u64 blkg_prfill_rwstat_field(struct seq_file *sf,
605 struct blkg_policy_data *pd, int off)
607 struct blkg_rwstat rwstat = blkg_rwstat_read((void *)pd->blkg + off);
609 return __blkg_prfill_rwstat(sf, pd, &rwstat);
613 * blkg_print_stat_bytes - seq_show callback for blkg->stat_bytes
614 * @sf: seq_file to print to
617 * To be used as cftype->seq_show to print blkg->stat_bytes.
618 * cftype->private must be set to the blkcg_policy.
620 int blkg_print_stat_bytes(struct seq_file *sf, void *v)
622 blkcg_print_blkgs(sf, css_to_blkcg(seq_css(sf)),
623 blkg_prfill_rwstat_field, (void *)seq_cft(sf)->private,
624 offsetof(struct blkcg_gq, stat_bytes), true);
627 EXPORT_SYMBOL_GPL(blkg_print_stat_bytes);
630 * blkg_print_stat_bytes - seq_show callback for blkg->stat_ios
631 * @sf: seq_file to print to
634 * To be used as cftype->seq_show to print blkg->stat_ios. cftype->private
635 * must be set to the blkcg_policy.
637 int blkg_print_stat_ios(struct seq_file *sf, void *v)
639 blkcg_print_blkgs(sf, css_to_blkcg(seq_css(sf)),
640 blkg_prfill_rwstat_field, (void *)seq_cft(sf)->private,
641 offsetof(struct blkcg_gq, stat_ios), true);
644 EXPORT_SYMBOL_GPL(blkg_print_stat_ios);
646 static u64 blkg_prfill_rwstat_field_recursive(struct seq_file *sf,
647 struct blkg_policy_data *pd,
650 struct blkg_rwstat rwstat = blkg_rwstat_recursive_sum(pd->blkg,
652 return __blkg_prfill_rwstat(sf, pd, &rwstat);
656 * blkg_print_stat_bytes_recursive - recursive version of blkg_print_stat_bytes
657 * @sf: seq_file to print to
660 int blkg_print_stat_bytes_recursive(struct seq_file *sf, void *v)
662 blkcg_print_blkgs(sf, css_to_blkcg(seq_css(sf)),
663 blkg_prfill_rwstat_field_recursive,
664 (void *)seq_cft(sf)->private,
665 offsetof(struct blkcg_gq, stat_bytes), true);
668 EXPORT_SYMBOL_GPL(blkg_print_stat_bytes_recursive);
671 * blkg_print_stat_ios_recursive - recursive version of blkg_print_stat_ios
672 * @sf: seq_file to print to
675 int blkg_print_stat_ios_recursive(struct seq_file *sf, void *v)
677 blkcg_print_blkgs(sf, css_to_blkcg(seq_css(sf)),
678 blkg_prfill_rwstat_field_recursive,
679 (void *)seq_cft(sf)->private,
680 offsetof(struct blkcg_gq, stat_ios), true);
683 EXPORT_SYMBOL_GPL(blkg_print_stat_ios_recursive);
686 * blkg_stat_recursive_sum - collect hierarchical blkg_stat
687 * @blkg: blkg of interest
688 * @pol: blkcg_policy which contains the blkg_stat
689 * @off: offset to the blkg_stat in blkg_policy_data or @blkg
691 * Collect the blkg_stat specified by @blkg, @pol and @off and all its
692 * online descendants and their aux counts. The caller must be holding the
693 * queue lock for online tests.
695 * If @pol is NULL, blkg_stat is at @off bytes into @blkg; otherwise, it is
696 * at @off bytes into @blkg's blkg_policy_data of the policy.
698 u64 blkg_stat_recursive_sum(struct blkcg_gq *blkg,
699 struct blkcg_policy *pol, int off)
701 struct blkcg_gq *pos_blkg;
702 struct cgroup_subsys_state *pos_css;
705 lockdep_assert_held(blkg->q->queue_lock);
708 blkg_for_each_descendant_pre(pos_blkg, pos_css, blkg) {
709 struct blkg_stat *stat;
711 if (!pos_blkg->online)
715 stat = (void *)blkg_to_pd(pos_blkg, pol) + off;
717 stat = (void *)blkg + off;
719 sum += blkg_stat_read(stat) + atomic64_read(&stat->aux_cnt);
725 EXPORT_SYMBOL_GPL(blkg_stat_recursive_sum);
728 * blkg_rwstat_recursive_sum - collect hierarchical blkg_rwstat
729 * @blkg: blkg of interest
730 * @pol: blkcg_policy which contains the blkg_rwstat
731 * @off: offset to the blkg_rwstat in blkg_policy_data or @blkg
733 * Collect the blkg_rwstat specified by @blkg, @pol and @off and all its
734 * online descendants and their aux counts. The caller must be holding the
735 * queue lock for online tests.
737 * If @pol is NULL, blkg_rwstat is at @off bytes into @blkg; otherwise, it
738 * is at @off bytes into @blkg's blkg_policy_data of the policy.
740 struct blkg_rwstat blkg_rwstat_recursive_sum(struct blkcg_gq *blkg,
741 struct blkcg_policy *pol, int off)
743 struct blkcg_gq *pos_blkg;
744 struct cgroup_subsys_state *pos_css;
745 struct blkg_rwstat sum = { };
748 lockdep_assert_held(blkg->q->queue_lock);
751 blkg_for_each_descendant_pre(pos_blkg, pos_css, blkg) {
752 struct blkg_rwstat *rwstat;
754 if (!pos_blkg->online)
758 rwstat = (void *)blkg_to_pd(pos_blkg, pol) + off;
760 rwstat = (void *)pos_blkg + off;
762 for (i = 0; i < BLKG_RWSTAT_NR; i++)
763 atomic64_add(atomic64_read(&rwstat->aux_cnt[i]) +
764 percpu_counter_sum_positive(&rwstat->cpu_cnt[i]),
771 EXPORT_SYMBOL_GPL(blkg_rwstat_recursive_sum);
774 * blkg_conf_prep - parse and prepare for per-blkg config update
775 * @blkcg: target block cgroup
776 * @pol: target policy
777 * @input: input string
778 * @ctx: blkg_conf_ctx to be filled
780 * Parse per-blkg config update from @input and initialize @ctx with the
781 * result. @ctx->blkg points to the blkg to be updated and @ctx->body the
782 * part of @input following MAJ:MIN. This function returns with RCU read
783 * lock and queue lock held and must be paired with blkg_conf_finish().
785 int blkg_conf_prep(struct blkcg *blkcg, const struct blkcg_policy *pol,
786 char *input, struct blkg_conf_ctx *ctx)
787 __acquires(rcu) __acquires(disk->queue->queue_lock)
789 struct gendisk *disk;
790 struct blkcg_gq *blkg;
791 struct module *owner;
792 unsigned int major, minor;
793 int key_len, part, ret;
796 if (sscanf(input, "%u:%u%n", &major, &minor, &key_len) != 2)
799 body = input + key_len;
802 body = skip_spaces(body);
804 disk = get_gendisk(MKDEV(major, minor), &part);
808 owner = disk->fops->owner;
815 spin_lock_irq(disk->queue->queue_lock);
817 if (blkcg_policy_enabled(disk->queue, pol))
818 blkg = blkg_lookup_create(blkcg, disk->queue);
820 blkg = ERR_PTR(-EOPNOTSUPP);
825 spin_unlock_irq(disk->queue->queue_lock);
826 owner = disk->fops->owner;
830 * If queue was bypassing, we should retry. Do so after a
831 * short msleep(). It isn't strictly necessary but queue
832 * can be bypassing for some time and it's always nice to
833 * avoid busy looping.
837 ret = restart_syscall();
847 EXPORT_SYMBOL_GPL(blkg_conf_prep);
850 * blkg_conf_finish - finish up per-blkg config update
851 * @ctx: blkg_conf_ctx intiailized by blkg_conf_prep()
853 * Finish up after per-blkg config update. This function must be paired
854 * with blkg_conf_prep().
856 void blkg_conf_finish(struct blkg_conf_ctx *ctx)
857 __releases(ctx->disk->queue->queue_lock) __releases(rcu)
859 struct module *owner;
861 spin_unlock_irq(ctx->disk->queue->queue_lock);
863 owner = ctx->disk->fops->owner;
867 EXPORT_SYMBOL_GPL(blkg_conf_finish);
869 static int blkcg_print_stat(struct seq_file *sf, void *v)
871 struct blkcg *blkcg = css_to_blkcg(seq_css(sf));
872 struct blkcg_gq *blkg;
876 hlist_for_each_entry_rcu(blkg, &blkcg->blkg_list, blkcg_node) {
878 struct blkg_rwstat rwstat;
879 u64 rbytes, wbytes, rios, wios;
881 dname = blkg_dev_name(blkg);
885 spin_lock_irq(blkg->q->queue_lock);
887 rwstat = blkg_rwstat_recursive_sum(blkg, NULL,
888 offsetof(struct blkcg_gq, stat_bytes));
889 rbytes = atomic64_read(&rwstat.aux_cnt[BLKG_RWSTAT_READ]);
890 wbytes = atomic64_read(&rwstat.aux_cnt[BLKG_RWSTAT_WRITE]);
892 rwstat = blkg_rwstat_recursive_sum(blkg, NULL,
893 offsetof(struct blkcg_gq, stat_ios));
894 rios = atomic64_read(&rwstat.aux_cnt[BLKG_RWSTAT_READ]);
895 wios = atomic64_read(&rwstat.aux_cnt[BLKG_RWSTAT_WRITE]);
897 spin_unlock_irq(blkg->q->queue_lock);
899 if (rbytes || wbytes || rios || wios)
900 seq_printf(sf, "%s rbytes=%llu wbytes=%llu rios=%llu wios=%llu\n",
901 dname, rbytes, wbytes, rios, wios);
908 static struct cftype blkcg_files[] = {
911 .flags = CFTYPE_NOT_ON_ROOT,
912 .seq_show = blkcg_print_stat,
917 static struct cftype blkcg_legacy_files[] = {
919 .name = "reset_stats",
920 .write_u64 = blkcg_reset_stats,
926 * blkcg_css_offline - cgroup css_offline callback
927 * @css: css of interest
929 * This function is called when @css is about to go away and responsible
930 * for shooting down all blkgs associated with @css. blkgs should be
931 * removed while holding both q and blkcg locks. As blkcg lock is nested
932 * inside q lock, this function performs reverse double lock dancing.
934 * This is the blkcg counterpart of ioc_release_fn().
936 static void blkcg_css_offline(struct cgroup_subsys_state *css)
938 struct blkcg *blkcg = css_to_blkcg(css);
940 spin_lock_irq(&blkcg->lock);
942 while (!hlist_empty(&blkcg->blkg_list)) {
943 struct blkcg_gq *blkg = hlist_entry(blkcg->blkg_list.first,
944 struct blkcg_gq, blkcg_node);
945 struct request_queue *q = blkg->q;
947 if (spin_trylock(q->queue_lock)) {
949 spin_unlock(q->queue_lock);
951 spin_unlock_irq(&blkcg->lock);
953 spin_lock_irq(&blkcg->lock);
957 spin_unlock_irq(&blkcg->lock);
959 wb_blkcg_offline(blkcg);
962 static void blkcg_css_free(struct cgroup_subsys_state *css)
964 struct blkcg *blkcg = css_to_blkcg(css);
967 mutex_lock(&blkcg_pol_mutex);
969 list_del(&blkcg->all_blkcgs_node);
971 for (i = 0; i < BLKCG_MAX_POLS; i++)
973 blkcg_policy[i]->cpd_free_fn(blkcg->cpd[i]);
975 mutex_unlock(&blkcg_pol_mutex);
980 static struct cgroup_subsys_state *
981 blkcg_css_alloc(struct cgroup_subsys_state *parent_css)
984 struct cgroup_subsys_state *ret;
987 mutex_lock(&blkcg_pol_mutex);
992 blkcg = kzalloc(sizeof(*blkcg), GFP_KERNEL);
994 ret = ERR_PTR(-ENOMEM);
999 for (i = 0; i < BLKCG_MAX_POLS ; i++) {
1000 struct blkcg_policy *pol = blkcg_policy[i];
1001 struct blkcg_policy_data *cpd;
1004 * If the policy hasn't been attached yet, wait for it
1005 * to be attached before doing anything else. Otherwise,
1006 * check if the policy requires any specific per-cgroup
1007 * data: if it does, allocate and initialize it.
1009 if (!pol || !pol->cpd_alloc_fn)
1012 cpd = pol->cpd_alloc_fn(GFP_KERNEL);
1014 ret = ERR_PTR(-ENOMEM);
1017 blkcg->cpd[i] = cpd;
1020 if (pol->cpd_init_fn)
1021 pol->cpd_init_fn(cpd);
1024 spin_lock_init(&blkcg->lock);
1025 INIT_RADIX_TREE(&blkcg->blkg_tree, GFP_NOWAIT);
1026 INIT_HLIST_HEAD(&blkcg->blkg_list);
1027 #ifdef CONFIG_CGROUP_WRITEBACK
1028 INIT_LIST_HEAD(&blkcg->cgwb_list);
1030 list_add_tail(&blkcg->all_blkcgs_node, &all_blkcgs);
1032 mutex_unlock(&blkcg_pol_mutex);
1036 for (i--; i >= 0; i--)
1038 blkcg_policy[i]->cpd_free_fn(blkcg->cpd[i]);
1041 mutex_unlock(&blkcg_pol_mutex);
1046 * blkcg_init_queue - initialize blkcg part of request queue
1047 * @q: request_queue to initialize
1049 * Called from blk_alloc_queue_node(). Responsible for initializing blkcg
1050 * part of new request_queue @q.
1053 * 0 on success, -errno on failure.
1055 int blkcg_init_queue(struct request_queue *q)
1057 struct blkcg_gq *new_blkg, *blkg;
1061 new_blkg = blkg_alloc(&blkcg_root, q, GFP_KERNEL);
1065 preloaded = !radix_tree_preload(GFP_KERNEL);
1068 * Make sure the root blkg exists and count the existing blkgs. As
1069 * @q is bypassing at this point, blkg_lookup_create() can't be
1070 * used. Open code insertion.
1073 spin_lock_irq(q->queue_lock);
1074 blkg = blkg_create(&blkcg_root, q, new_blkg);
1075 spin_unlock_irq(q->queue_lock);
1079 radix_tree_preload_end();
1082 blkg_free(new_blkg);
1083 return PTR_ERR(blkg);
1086 q->root_blkg = blkg;
1087 q->root_rl.blkg = blkg;
1089 ret = blk_throtl_init(q);
1091 spin_lock_irq(q->queue_lock);
1092 blkg_destroy_all(q);
1093 spin_unlock_irq(q->queue_lock);
1099 * blkcg_drain_queue - drain blkcg part of request_queue
1100 * @q: request_queue to drain
1102 * Called from blk_drain_queue(). Responsible for draining blkcg part.
1104 void blkcg_drain_queue(struct request_queue *q)
1106 lockdep_assert_held(q->queue_lock);
1109 * @q could be exiting and already have destroyed all blkgs as
1110 * indicated by NULL root_blkg. If so, don't confuse policies.
1115 blk_throtl_drain(q);
1119 * blkcg_exit_queue - exit and release blkcg part of request_queue
1120 * @q: request_queue being released
1122 * Called from blk_release_queue(). Responsible for exiting blkcg part.
1124 void blkcg_exit_queue(struct request_queue *q)
1126 spin_lock_irq(q->queue_lock);
1127 blkg_destroy_all(q);
1128 spin_unlock_irq(q->queue_lock);
1134 * We cannot support shared io contexts, as we have no mean to support
1135 * two tasks with the same ioc in two different groups without major rework
1136 * of the main cic data structures. For now we allow a task to change
1137 * its cgroup only if it's the only owner of its ioc.
1139 static int blkcg_can_attach(struct cgroup_taskset *tset)
1141 struct task_struct *task;
1142 struct cgroup_subsys_state *dst_css;
1143 struct io_context *ioc;
1146 /* task_lock() is needed to avoid races with exit_io_context() */
1147 cgroup_taskset_for_each(task, dst_css, tset) {
1149 ioc = task->io_context;
1150 if (ioc && atomic_read(&ioc->nr_tasks) > 1)
1159 static void blkcg_bind(struct cgroup_subsys_state *root_css)
1163 mutex_lock(&blkcg_pol_mutex);
1165 for (i = 0; i < BLKCG_MAX_POLS; i++) {
1166 struct blkcg_policy *pol = blkcg_policy[i];
1167 struct blkcg *blkcg;
1169 if (!pol || !pol->cpd_bind_fn)
1172 list_for_each_entry(blkcg, &all_blkcgs, all_blkcgs_node)
1173 if (blkcg->cpd[pol->plid])
1174 pol->cpd_bind_fn(blkcg->cpd[pol->plid]);
1176 mutex_unlock(&blkcg_pol_mutex);
1179 struct cgroup_subsys io_cgrp_subsys = {
1180 .css_alloc = blkcg_css_alloc,
1181 .css_offline = blkcg_css_offline,
1182 .css_free = blkcg_css_free,
1183 .can_attach = blkcg_can_attach,
1185 .dfl_cftypes = blkcg_files,
1186 .legacy_cftypes = blkcg_legacy_files,
1187 .legacy_name = "blkio",
1190 * This ensures that, if available, memcg is automatically enabled
1191 * together on the default hierarchy so that the owner cgroup can
1192 * be retrieved from writeback pages.
1194 .depends_on = 1 << memory_cgrp_id,
1197 EXPORT_SYMBOL_GPL(io_cgrp_subsys);
1200 * blkcg_activate_policy - activate a blkcg policy on a request_queue
1201 * @q: request_queue of interest
1202 * @pol: blkcg policy to activate
1204 * Activate @pol on @q. Requires %GFP_KERNEL context. @q goes through
1205 * bypass mode to populate its blkgs with policy_data for @pol.
1207 * Activation happens with @q bypassed, so nobody would be accessing blkgs
1208 * from IO path. Update of each blkg is protected by both queue and blkcg
1209 * locks so that holding either lock and testing blkcg_policy_enabled() is
1210 * always enough for dereferencing policy data.
1212 * The caller is responsible for synchronizing [de]activations and policy
1213 * [un]registerations. Returns 0 on success, -errno on failure.
1215 int blkcg_activate_policy(struct request_queue *q,
1216 const struct blkcg_policy *pol)
1218 struct blkg_policy_data *pd_prealloc = NULL;
1219 struct blkcg_gq *blkg;
1222 if (blkcg_policy_enabled(q, pol))
1225 blk_queue_bypass_start(q);
1228 pd_prealloc = pol->pd_alloc_fn(GFP_KERNEL, q->node);
1231 goto out_bypass_end;
1235 spin_lock_irq(q->queue_lock);
1237 list_for_each_entry(blkg, &q->blkg_list, q_node) {
1238 struct blkg_policy_data *pd;
1240 if (blkg->pd[pol->plid])
1243 pd = pol->pd_alloc_fn(GFP_NOWAIT, q->node);
1245 swap(pd, pd_prealloc);
1247 spin_unlock_irq(q->queue_lock);
1251 blkg->pd[pol->plid] = pd;
1253 pd->plid = pol->plid;
1254 if (pol->pd_init_fn)
1255 pol->pd_init_fn(pd);
1258 __set_bit(pol->plid, q->blkcg_pols);
1261 spin_unlock_irq(q->queue_lock);
1263 blk_queue_bypass_end(q);
1265 pol->pd_free_fn(pd_prealloc);
1268 EXPORT_SYMBOL_GPL(blkcg_activate_policy);
1271 * blkcg_deactivate_policy - deactivate a blkcg policy on a request_queue
1272 * @q: request_queue of interest
1273 * @pol: blkcg policy to deactivate
1275 * Deactivate @pol on @q. Follows the same synchronization rules as
1276 * blkcg_activate_policy().
1278 void blkcg_deactivate_policy(struct request_queue *q,
1279 const struct blkcg_policy *pol)
1281 struct blkcg_gq *blkg;
1283 if (!blkcg_policy_enabled(q, pol))
1286 blk_queue_bypass_start(q);
1287 spin_lock_irq(q->queue_lock);
1289 __clear_bit(pol->plid, q->blkcg_pols);
1291 list_for_each_entry(blkg, &q->blkg_list, q_node) {
1292 /* grab blkcg lock too while removing @pd from @blkg */
1293 spin_lock(&blkg->blkcg->lock);
1295 if (blkg->pd[pol->plid]) {
1296 if (pol->pd_offline_fn)
1297 pol->pd_offline_fn(blkg->pd[pol->plid]);
1298 pol->pd_free_fn(blkg->pd[pol->plid]);
1299 blkg->pd[pol->plid] = NULL;
1302 spin_unlock(&blkg->blkcg->lock);
1305 spin_unlock_irq(q->queue_lock);
1306 blk_queue_bypass_end(q);
1308 EXPORT_SYMBOL_GPL(blkcg_deactivate_policy);
1311 * blkcg_policy_register - register a blkcg policy
1312 * @pol: blkcg policy to register
1314 * Register @pol with blkcg core. Might sleep and @pol may be modified on
1315 * successful registration. Returns 0 on success and -errno on failure.
1317 int blkcg_policy_register(struct blkcg_policy *pol)
1319 struct blkcg *blkcg;
1322 mutex_lock(&blkcg_pol_register_mutex);
1323 mutex_lock(&blkcg_pol_mutex);
1325 /* find an empty slot */
1327 for (i = 0; i < BLKCG_MAX_POLS; i++)
1328 if (!blkcg_policy[i])
1330 if (i >= BLKCG_MAX_POLS)
1335 blkcg_policy[pol->plid] = pol;
1337 /* allocate and install cpd's */
1338 if (pol->cpd_alloc_fn) {
1339 list_for_each_entry(blkcg, &all_blkcgs, all_blkcgs_node) {
1340 struct blkcg_policy_data *cpd;
1342 cpd = pol->cpd_alloc_fn(GFP_KERNEL);
1344 mutex_unlock(&blkcg_pol_mutex);
1348 blkcg->cpd[pol->plid] = cpd;
1350 cpd->plid = pol->plid;
1351 pol->cpd_init_fn(cpd);
1355 mutex_unlock(&blkcg_pol_mutex);
1357 /* everything is in place, add intf files for the new policy */
1358 if (pol->dfl_cftypes)
1359 WARN_ON(cgroup_add_dfl_cftypes(&io_cgrp_subsys,
1361 if (pol->legacy_cftypes)
1362 WARN_ON(cgroup_add_legacy_cftypes(&io_cgrp_subsys,
1363 pol->legacy_cftypes));
1364 mutex_unlock(&blkcg_pol_register_mutex);
1368 if (pol->cpd_alloc_fn) {
1369 list_for_each_entry(blkcg, &all_blkcgs, all_blkcgs_node) {
1370 if (blkcg->cpd[pol->plid]) {
1371 pol->cpd_free_fn(blkcg->cpd[pol->plid]);
1372 blkcg->cpd[pol->plid] = NULL;
1376 blkcg_policy[pol->plid] = NULL;
1378 mutex_unlock(&blkcg_pol_mutex);
1379 mutex_unlock(&blkcg_pol_register_mutex);
1382 EXPORT_SYMBOL_GPL(blkcg_policy_register);
1385 * blkcg_policy_unregister - unregister a blkcg policy
1386 * @pol: blkcg policy to unregister
1388 * Undo blkcg_policy_register(@pol). Might sleep.
1390 void blkcg_policy_unregister(struct blkcg_policy *pol)
1392 struct blkcg *blkcg;
1394 mutex_lock(&blkcg_pol_register_mutex);
1396 if (WARN_ON(blkcg_policy[pol->plid] != pol))
1399 /* kill the intf files first */
1400 if (pol->dfl_cftypes)
1401 cgroup_rm_cftypes(pol->dfl_cftypes);
1402 if (pol->legacy_cftypes)
1403 cgroup_rm_cftypes(pol->legacy_cftypes);
1405 /* remove cpds and unregister */
1406 mutex_lock(&blkcg_pol_mutex);
1408 if (pol->cpd_alloc_fn) {
1409 list_for_each_entry(blkcg, &all_blkcgs, all_blkcgs_node) {
1410 if (blkcg->cpd[pol->plid]) {
1411 pol->cpd_free_fn(blkcg->cpd[pol->plid]);
1412 blkcg->cpd[pol->plid] = NULL;
1416 blkcg_policy[pol->plid] = NULL;
1418 mutex_unlock(&blkcg_pol_mutex);
1420 mutex_unlock(&blkcg_pol_register_mutex);
1422 EXPORT_SYMBOL_GPL(blkcg_policy_unregister);