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/sched/signal.h>
21 #include <linux/err.h>
22 #include <linux/blkdev.h>
23 #include <linux/backing-dev.h>
24 #include <linux/slab.h>
25 #include <linux/genhd.h>
26 #include <linux/delay.h>
27 #include <linux/atomic.h>
28 #include <linux/ctype.h>
29 #include <linux/blk-cgroup.h>
32 #define MAX_KEY_LEN 100
35 * blkcg_pol_mutex protects blkcg_policy[] and policy [de]activation.
36 * blkcg_pol_register_mutex nests outside of it and synchronizes entire
37 * policy [un]register operations including cgroup file additions /
38 * removals. Putting cgroup file registration outside blkcg_pol_mutex
39 * allows grabbing it from cgroup callbacks.
41 static DEFINE_MUTEX(blkcg_pol_register_mutex);
42 static DEFINE_MUTEX(blkcg_pol_mutex);
44 struct blkcg blkcg_root;
45 EXPORT_SYMBOL_GPL(blkcg_root);
47 struct cgroup_subsys_state * const blkcg_root_css = &blkcg_root.css;
49 static struct blkcg_policy *blkcg_policy[BLKCG_MAX_POLS];
51 static LIST_HEAD(all_blkcgs); /* protected by blkcg_pol_mutex */
53 static bool blkcg_policy_enabled(struct request_queue *q,
54 const struct blkcg_policy *pol)
56 return pol && test_bit(pol->plid, q->blkcg_pols);
60 * blkg_free - free a blkg
63 * Free @blkg which may be partially allocated.
65 static void blkg_free(struct blkcg_gq *blkg)
72 for (i = 0; i < BLKCG_MAX_POLS; i++)
74 blkcg_policy[i]->pd_free_fn(blkg->pd[i]);
76 if (blkg->blkcg != &blkcg_root)
77 blk_exit_rl(blkg->q, &blkg->rl);
79 blkg_rwstat_exit(&blkg->stat_ios);
80 blkg_rwstat_exit(&blkg->stat_bytes);
85 * blkg_alloc - allocate a blkg
86 * @blkcg: block cgroup the new blkg is associated with
87 * @q: request_queue the new blkg is associated with
88 * @gfp_mask: allocation mask to use
90 * Allocate a new blkg assocating @blkcg and @q.
92 static struct blkcg_gq *blkg_alloc(struct blkcg *blkcg, struct request_queue *q,
95 struct blkcg_gq *blkg;
98 /* alloc and init base part */
99 blkg = kzalloc_node(sizeof(*blkg), gfp_mask, q->node);
103 if (blkg_rwstat_init(&blkg->stat_bytes, gfp_mask) ||
104 blkg_rwstat_init(&blkg->stat_ios, gfp_mask))
108 INIT_LIST_HEAD(&blkg->q_node);
110 atomic_set(&blkg->refcnt, 1);
112 /* root blkg uses @q->root_rl, init rl only for !root blkgs */
113 if (blkcg != &blkcg_root) {
114 if (blk_init_rl(&blkg->rl, q, gfp_mask))
116 blkg->rl.blkg = blkg;
119 for (i = 0; i < BLKCG_MAX_POLS; i++) {
120 struct blkcg_policy *pol = blkcg_policy[i];
121 struct blkg_policy_data *pd;
123 if (!blkcg_policy_enabled(q, pol))
126 /* alloc per-policy data and attach it to blkg */
127 pd = pol->pd_alloc_fn(gfp_mask, q->node);
143 struct blkcg_gq *blkg_lookup_slowpath(struct blkcg *blkcg,
144 struct request_queue *q, bool update_hint)
146 struct blkcg_gq *blkg;
149 * Hint didn't match. Look up from the radix tree. Note that the
150 * hint can only be updated under queue_lock as otherwise @blkg
151 * could have already been removed from blkg_tree. The caller is
152 * responsible for grabbing queue_lock if @update_hint.
154 blkg = radix_tree_lookup(&blkcg->blkg_tree, q->id);
155 if (blkg && blkg->q == q) {
157 lockdep_assert_held(q->queue_lock);
158 rcu_assign_pointer(blkcg->blkg_hint, blkg);
165 EXPORT_SYMBOL_GPL(blkg_lookup_slowpath);
168 * If @new_blkg is %NULL, this function tries to allocate a new one as
169 * necessary using %GFP_NOWAIT. @new_blkg is always consumed on return.
171 static struct blkcg_gq *blkg_create(struct blkcg *blkcg,
172 struct request_queue *q,
173 struct blkcg_gq *new_blkg)
175 struct blkcg_gq *blkg;
176 struct bdi_writeback_congested *wb_congested;
179 WARN_ON_ONCE(!rcu_read_lock_held());
180 lockdep_assert_held(q->queue_lock);
182 /* blkg holds a reference to blkcg */
183 if (!css_tryget_online(&blkcg->css)) {
188 wb_congested = wb_congested_get_create(q->backing_dev_info,
190 GFP_NOWAIT | __GFP_NOWARN);
198 new_blkg = blkg_alloc(blkcg, q, GFP_NOWAIT | __GFP_NOWARN);
199 if (unlikely(!new_blkg)) {
201 goto err_put_congested;
205 blkg->wb_congested = wb_congested;
208 if (blkcg_parent(blkcg)) {
209 blkg->parent = __blkg_lookup(blkcg_parent(blkcg), q, false);
210 if (WARN_ON_ONCE(!blkg->parent)) {
212 goto err_put_congested;
214 blkg_get(blkg->parent);
217 /* invoke per-policy init */
218 for (i = 0; i < BLKCG_MAX_POLS; i++) {
219 struct blkcg_policy *pol = blkcg_policy[i];
221 if (blkg->pd[i] && pol->pd_init_fn)
222 pol->pd_init_fn(blkg->pd[i]);
226 spin_lock(&blkcg->lock);
227 ret = radix_tree_insert(&blkcg->blkg_tree, q->id, blkg);
229 hlist_add_head_rcu(&blkg->blkcg_node, &blkcg->blkg_list);
230 list_add(&blkg->q_node, &q->blkg_list);
232 for (i = 0; i < BLKCG_MAX_POLS; i++) {
233 struct blkcg_policy *pol = blkcg_policy[i];
235 if (blkg->pd[i] && pol->pd_online_fn)
236 pol->pd_online_fn(blkg->pd[i]);
240 spin_unlock(&blkcg->lock);
245 /* @blkg failed fully initialized, use the usual release path */
250 wb_congested_put(wb_congested);
252 css_put(&blkcg->css);
259 * blkg_lookup_create - lookup blkg, try to create one if not there
260 * @blkcg: blkcg of interest
261 * @q: request_queue of interest
263 * Lookup blkg for the @blkcg - @q pair. If it doesn't exist, try to
264 * create one. blkg creation is performed recursively from blkcg_root such
265 * that all non-root blkg's have access to the parent blkg. This function
266 * should be called under RCU read lock and @q->queue_lock.
268 * Returns pointer to the looked up or created blkg on success, ERR_PTR()
269 * value on error. If @q is dead, returns ERR_PTR(-EINVAL). If @q is not
270 * dead and bypassing, returns ERR_PTR(-EBUSY).
272 struct blkcg_gq *blkg_lookup_create(struct blkcg *blkcg,
273 struct request_queue *q)
275 struct blkcg_gq *blkg;
277 WARN_ON_ONCE(!rcu_read_lock_held());
278 lockdep_assert_held(q->queue_lock);
281 * This could be the first entry point of blkcg implementation and
282 * we shouldn't allow anything to go through for a bypassing queue.
284 if (unlikely(blk_queue_bypass(q)))
285 return ERR_PTR(blk_queue_dying(q) ? -ENODEV : -EBUSY);
287 blkg = __blkg_lookup(blkcg, q, true);
292 * Create blkgs walking down from blkcg_root to @blkcg, so that all
293 * non-root blkgs have access to their parents.
296 struct blkcg *pos = blkcg;
297 struct blkcg *parent = blkcg_parent(blkcg);
299 while (parent && !__blkg_lookup(parent, q, false)) {
301 parent = blkcg_parent(parent);
304 blkg = blkg_create(pos, q, NULL);
305 if (pos == blkcg || IS_ERR(blkg))
310 static void blkg_destroy(struct blkcg_gq *blkg)
312 struct blkcg *blkcg = blkg->blkcg;
313 struct blkcg_gq *parent = blkg->parent;
316 lockdep_assert_held(blkg->q->queue_lock);
317 lockdep_assert_held(&blkcg->lock);
319 /* Something wrong if we are trying to remove same group twice */
320 WARN_ON_ONCE(list_empty(&blkg->q_node));
321 WARN_ON_ONCE(hlist_unhashed(&blkg->blkcg_node));
323 for (i = 0; i < BLKCG_MAX_POLS; i++) {
324 struct blkcg_policy *pol = blkcg_policy[i];
326 if (blkg->pd[i] && pol->pd_offline_fn)
327 pol->pd_offline_fn(blkg->pd[i]);
331 blkg_rwstat_add_aux(&parent->stat_bytes, &blkg->stat_bytes);
332 blkg_rwstat_add_aux(&parent->stat_ios, &blkg->stat_ios);
335 blkg->online = false;
337 radix_tree_delete(&blkcg->blkg_tree, blkg->q->id);
338 list_del_init(&blkg->q_node);
339 hlist_del_init_rcu(&blkg->blkcg_node);
342 * Both setting lookup hint to and clearing it from @blkg are done
343 * under queue_lock. If it's not pointing to @blkg now, it never
344 * will. Hint assignment itself can race safely.
346 if (rcu_access_pointer(blkcg->blkg_hint) == blkg)
347 rcu_assign_pointer(blkcg->blkg_hint, NULL);
350 * Put the reference taken at the time of creation so that when all
351 * queues are gone, group can be destroyed.
357 * blkg_destroy_all - destroy all blkgs associated with a request_queue
358 * @q: request_queue of interest
360 * Destroy all blkgs associated with @q.
362 static void blkg_destroy_all(struct request_queue *q)
364 struct blkcg_gq *blkg, *n;
366 lockdep_assert_held(q->queue_lock);
368 list_for_each_entry_safe(blkg, n, &q->blkg_list, q_node) {
369 struct blkcg *blkcg = blkg->blkcg;
371 spin_lock(&blkcg->lock);
373 spin_unlock(&blkcg->lock);
377 q->root_rl.blkg = NULL;
381 * A group is RCU protected, but having an rcu lock does not mean that one
382 * can access all the fields of blkg and assume these are valid. For
383 * example, don't try to follow throtl_data and request queue links.
385 * Having a reference to blkg under an rcu allows accesses to only values
386 * local to groups like group stats and group rate limits.
388 void __blkg_release_rcu(struct rcu_head *rcu_head)
390 struct blkcg_gq *blkg = container_of(rcu_head, struct blkcg_gq, rcu_head);
392 /* release the blkcg and parent blkg refs this blkg has been holding */
393 css_put(&blkg->blkcg->css);
395 blkg_put(blkg->parent);
397 wb_congested_put(blkg->wb_congested);
401 EXPORT_SYMBOL_GPL(__blkg_release_rcu);
404 * The next function used by blk_queue_for_each_rl(). It's a bit tricky
405 * because the root blkg uses @q->root_rl instead of its own rl.
407 struct request_list *__blk_queue_next_rl(struct request_list *rl,
408 struct request_queue *q)
410 struct list_head *ent;
411 struct blkcg_gq *blkg;
414 * Determine the current blkg list_head. The first entry is
415 * root_rl which is off @q->blkg_list and mapped to the head.
417 if (rl == &q->root_rl) {
419 /* There are no more block groups, hence no request lists */
423 blkg = container_of(rl, struct blkcg_gq, rl);
427 /* walk to the next list_head, skip root blkcg */
429 if (ent == &q->root_blkg->q_node)
431 if (ent == &q->blkg_list)
434 blkg = container_of(ent, struct blkcg_gq, q_node);
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);
477 EXPORT_SYMBOL_GPL(blkg_dev_name);
480 * blkcg_print_blkgs - helper for printing per-blkg data
481 * @sf: seq_file to print to
482 * @blkcg: blkcg of interest
483 * @prfill: fill function to print out a blkg
484 * @pol: policy in question
485 * @data: data to be passed to @prfill
486 * @show_total: to print out sum of prfill return values or not
488 * This function invokes @prfill on each blkg of @blkcg if pd for the
489 * policy specified by @pol exists. @prfill is invoked with @sf, the
490 * policy data and @data and the matching queue lock held. If @show_total
491 * is %true, the sum of the return values from @prfill is printed with
492 * "Total" label at the end.
494 * This is to be used to construct print functions for
495 * cftype->read_seq_string method.
497 void blkcg_print_blkgs(struct seq_file *sf, struct blkcg *blkcg,
498 u64 (*prfill)(struct seq_file *,
499 struct blkg_policy_data *, int),
500 const struct blkcg_policy *pol, int data,
503 struct blkcg_gq *blkg;
507 hlist_for_each_entry_rcu(blkg, &blkcg->blkg_list, blkcg_node) {
508 spin_lock_irq(blkg->q->queue_lock);
509 if (blkcg_policy_enabled(blkg->q, pol))
510 total += prfill(sf, blkg->pd[pol->plid], data);
511 spin_unlock_irq(blkg->q->queue_lock);
516 seq_printf(sf, "Total %llu\n", (unsigned long long)total);
518 EXPORT_SYMBOL_GPL(blkcg_print_blkgs);
521 * __blkg_prfill_u64 - prfill helper for a single u64 value
522 * @sf: seq_file to print to
523 * @pd: policy private data of interest
526 * Print @v to @sf for the device assocaited with @pd.
528 u64 __blkg_prfill_u64(struct seq_file *sf, struct blkg_policy_data *pd, u64 v)
530 const char *dname = blkg_dev_name(pd->blkg);
535 seq_printf(sf, "%s %llu\n", dname, (unsigned long long)v);
538 EXPORT_SYMBOL_GPL(__blkg_prfill_u64);
541 * __blkg_prfill_rwstat - prfill helper for a blkg_rwstat
542 * @sf: seq_file to print to
543 * @pd: policy private data of interest
544 * @rwstat: rwstat to print
546 * Print @rwstat to @sf for the device assocaited with @pd.
548 u64 __blkg_prfill_rwstat(struct seq_file *sf, struct blkg_policy_data *pd,
549 const struct blkg_rwstat *rwstat)
551 static const char *rwstr[] = {
552 [BLKG_RWSTAT_READ] = "Read",
553 [BLKG_RWSTAT_WRITE] = "Write",
554 [BLKG_RWSTAT_SYNC] = "Sync",
555 [BLKG_RWSTAT_ASYNC] = "Async",
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],
566 (unsigned long long)atomic64_read(&rwstat->aux_cnt[i]));
568 v = atomic64_read(&rwstat->aux_cnt[BLKG_RWSTAT_READ]) +
569 atomic64_read(&rwstat->aux_cnt[BLKG_RWSTAT_WRITE]);
570 seq_printf(sf, "%s Total %llu\n", dname, (unsigned long long)v);
573 EXPORT_SYMBOL_GPL(__blkg_prfill_rwstat);
576 * blkg_prfill_stat - prfill callback for blkg_stat
577 * @sf: seq_file to print to
578 * @pd: policy private data of interest
579 * @off: offset to the blkg_stat in @pd
581 * prfill callback for printing a blkg_stat.
583 u64 blkg_prfill_stat(struct seq_file *sf, struct blkg_policy_data *pd, int off)
585 return __blkg_prfill_u64(sf, pd, blkg_stat_read((void *)pd + off));
587 EXPORT_SYMBOL_GPL(blkg_prfill_stat);
590 * blkg_prfill_rwstat - prfill callback for blkg_rwstat
591 * @sf: seq_file to print to
592 * @pd: policy private data of interest
593 * @off: offset to the blkg_rwstat in @pd
595 * prfill callback for printing a blkg_rwstat.
597 u64 blkg_prfill_rwstat(struct seq_file *sf, struct blkg_policy_data *pd,
600 struct blkg_rwstat rwstat = blkg_rwstat_read((void *)pd + off);
602 return __blkg_prfill_rwstat(sf, pd, &rwstat);
604 EXPORT_SYMBOL_GPL(blkg_prfill_rwstat);
606 static u64 blkg_prfill_rwstat_field(struct seq_file *sf,
607 struct blkg_policy_data *pd, int off)
609 struct blkg_rwstat rwstat = blkg_rwstat_read((void *)pd->blkg + off);
611 return __blkg_prfill_rwstat(sf, pd, &rwstat);
615 * blkg_print_stat_bytes - seq_show callback for blkg->stat_bytes
616 * @sf: seq_file to print to
619 * To be used as cftype->seq_show to print blkg->stat_bytes.
620 * cftype->private must be set to the blkcg_policy.
622 int blkg_print_stat_bytes(struct seq_file *sf, void *v)
624 blkcg_print_blkgs(sf, css_to_blkcg(seq_css(sf)),
625 blkg_prfill_rwstat_field, (void *)seq_cft(sf)->private,
626 offsetof(struct blkcg_gq, stat_bytes), true);
629 EXPORT_SYMBOL_GPL(blkg_print_stat_bytes);
632 * blkg_print_stat_bytes - seq_show callback for blkg->stat_ios
633 * @sf: seq_file to print to
636 * To be used as cftype->seq_show to print blkg->stat_ios. cftype->private
637 * must be set to the blkcg_policy.
639 int blkg_print_stat_ios(struct seq_file *sf, void *v)
641 blkcg_print_blkgs(sf, css_to_blkcg(seq_css(sf)),
642 blkg_prfill_rwstat_field, (void *)seq_cft(sf)->private,
643 offsetof(struct blkcg_gq, stat_ios), true);
646 EXPORT_SYMBOL_GPL(blkg_print_stat_ios);
648 static u64 blkg_prfill_rwstat_field_recursive(struct seq_file *sf,
649 struct blkg_policy_data *pd,
652 struct blkg_rwstat rwstat = blkg_rwstat_recursive_sum(pd->blkg,
654 return __blkg_prfill_rwstat(sf, pd, &rwstat);
658 * blkg_print_stat_bytes_recursive - recursive version of blkg_print_stat_bytes
659 * @sf: seq_file to print to
662 int blkg_print_stat_bytes_recursive(struct seq_file *sf, void *v)
664 blkcg_print_blkgs(sf, css_to_blkcg(seq_css(sf)),
665 blkg_prfill_rwstat_field_recursive,
666 (void *)seq_cft(sf)->private,
667 offsetof(struct blkcg_gq, stat_bytes), true);
670 EXPORT_SYMBOL_GPL(blkg_print_stat_bytes_recursive);
673 * blkg_print_stat_ios_recursive - recursive version of blkg_print_stat_ios
674 * @sf: seq_file to print to
677 int blkg_print_stat_ios_recursive(struct seq_file *sf, void *v)
679 blkcg_print_blkgs(sf, css_to_blkcg(seq_css(sf)),
680 blkg_prfill_rwstat_field_recursive,
681 (void *)seq_cft(sf)->private,
682 offsetof(struct blkcg_gq, stat_ios), true);
685 EXPORT_SYMBOL_GPL(blkg_print_stat_ios_recursive);
688 * blkg_stat_recursive_sum - collect hierarchical blkg_stat
689 * @blkg: blkg of interest
690 * @pol: blkcg_policy which contains the blkg_stat
691 * @off: offset to the blkg_stat in blkg_policy_data or @blkg
693 * Collect the blkg_stat specified by @blkg, @pol and @off and all its
694 * online descendants and their aux counts. The caller must be holding the
695 * queue lock for online tests.
697 * If @pol is NULL, blkg_stat is at @off bytes into @blkg; otherwise, it is
698 * at @off bytes into @blkg's blkg_policy_data of the policy.
700 u64 blkg_stat_recursive_sum(struct blkcg_gq *blkg,
701 struct blkcg_policy *pol, int off)
703 struct blkcg_gq *pos_blkg;
704 struct cgroup_subsys_state *pos_css;
707 lockdep_assert_held(blkg->q->queue_lock);
710 blkg_for_each_descendant_pre(pos_blkg, pos_css, blkg) {
711 struct blkg_stat *stat;
713 if (!pos_blkg->online)
717 stat = (void *)blkg_to_pd(pos_blkg, pol) + off;
719 stat = (void *)blkg + off;
721 sum += blkg_stat_read(stat) + atomic64_read(&stat->aux_cnt);
727 EXPORT_SYMBOL_GPL(blkg_stat_recursive_sum);
730 * blkg_rwstat_recursive_sum - collect hierarchical blkg_rwstat
731 * @blkg: blkg of interest
732 * @pol: blkcg_policy which contains the blkg_rwstat
733 * @off: offset to the blkg_rwstat in blkg_policy_data or @blkg
735 * Collect the blkg_rwstat specified by @blkg, @pol and @off and all its
736 * online descendants and their aux counts. The caller must be holding the
737 * queue lock for online tests.
739 * If @pol is NULL, blkg_rwstat is at @off bytes into @blkg; otherwise, it
740 * is at @off bytes into @blkg's blkg_policy_data of the policy.
742 struct blkg_rwstat blkg_rwstat_recursive_sum(struct blkcg_gq *blkg,
743 struct blkcg_policy *pol, int off)
745 struct blkcg_gq *pos_blkg;
746 struct cgroup_subsys_state *pos_css;
747 struct blkg_rwstat sum = { };
750 lockdep_assert_held(blkg->q->queue_lock);
753 blkg_for_each_descendant_pre(pos_blkg, pos_css, blkg) {
754 struct blkg_rwstat *rwstat;
756 if (!pos_blkg->online)
760 rwstat = (void *)blkg_to_pd(pos_blkg, pol) + off;
762 rwstat = (void *)pos_blkg + off;
764 for (i = 0; i < BLKG_RWSTAT_NR; i++)
765 atomic64_add(atomic64_read(&rwstat->aux_cnt[i]) +
766 percpu_counter_sum_positive(&rwstat->cpu_cnt[i]),
773 EXPORT_SYMBOL_GPL(blkg_rwstat_recursive_sum);
775 /* Performs queue bypass and policy enabled checks then looks up blkg. */
776 static struct blkcg_gq *blkg_lookup_check(struct blkcg *blkcg,
777 const struct blkcg_policy *pol,
778 struct request_queue *q)
780 WARN_ON_ONCE(!rcu_read_lock_held());
781 lockdep_assert_held(q->queue_lock);
783 if (!blkcg_policy_enabled(q, pol))
784 return ERR_PTR(-EOPNOTSUPP);
787 * This could be the first entry point of blkcg implementation and
788 * we shouldn't allow anything to go through for a bypassing queue.
790 if (unlikely(blk_queue_bypass(q)))
791 return ERR_PTR(blk_queue_dying(q) ? -ENODEV : -EBUSY);
793 return __blkg_lookup(blkcg, q, true /* update_hint */);
797 * blkg_conf_prep - parse and prepare for per-blkg config update
798 * @blkcg: target block cgroup
799 * @pol: target policy
800 * @input: input string
801 * @ctx: blkg_conf_ctx to be filled
803 * Parse per-blkg config update from @input and initialize @ctx with the
804 * result. @ctx->blkg points to the blkg to be updated and @ctx->body the
805 * part of @input following MAJ:MIN. This function returns with RCU read
806 * lock and queue lock held and must be paired with blkg_conf_finish().
808 int blkg_conf_prep(struct blkcg *blkcg, const struct blkcg_policy *pol,
809 char *input, struct blkg_conf_ctx *ctx)
810 __acquires(rcu) __acquires(disk->queue->queue_lock)
812 struct gendisk *disk;
813 struct request_queue *q;
814 struct blkcg_gq *blkg;
815 struct module *owner;
816 unsigned int major, minor;
817 int key_len, part, ret;
820 if (sscanf(input, "%u:%u%n", &major, &minor, &key_len) != 2)
823 body = input + key_len;
826 body = skip_spaces(body);
828 disk = get_gendisk(MKDEV(major, minor), &part);
839 spin_lock_irq(q->queue_lock);
841 blkg = blkg_lookup_check(blkcg, pol, q);
851 * Create blkgs walking down from blkcg_root to @blkcg, so that all
852 * non-root blkgs have access to their parents.
855 struct blkcg *pos = blkcg;
856 struct blkcg *parent;
857 struct blkcg_gq *new_blkg;
859 parent = blkcg_parent(blkcg);
860 while (parent && !__blkg_lookup(parent, q, false)) {
862 parent = blkcg_parent(parent);
865 /* Drop locks to do new blkg allocation with GFP_KERNEL. */
866 spin_unlock_irq(q->queue_lock);
869 new_blkg = blkg_alloc(pos, q, GFP_KERNEL);
870 if (unlikely(!new_blkg)) {
876 spin_lock_irq(q->queue_lock);
878 blkg = blkg_lookup_check(pos, pol, q);
887 blkg = blkg_create(pos, q, new_blkg);
888 if (unlikely(IS_ERR(blkg))) {
904 spin_unlock_irq(q->queue_lock);
907 owner = disk->fops->owner;
911 * If queue was bypassing, we should retry. Do so after a
912 * short msleep(). It isn't strictly necessary but queue
913 * can be bypassing for some time and it's always nice to
914 * avoid busy looping.
918 ret = restart_syscall();
922 EXPORT_SYMBOL_GPL(blkg_conf_prep);
925 * blkg_conf_finish - finish up per-blkg config update
926 * @ctx: blkg_conf_ctx intiailized by blkg_conf_prep()
928 * Finish up after per-blkg config update. This function must be paired
929 * with blkg_conf_prep().
931 void blkg_conf_finish(struct blkg_conf_ctx *ctx)
932 __releases(ctx->disk->queue->queue_lock) __releases(rcu)
934 struct module *owner;
936 spin_unlock_irq(ctx->disk->queue->queue_lock);
938 owner = ctx->disk->fops->owner;
942 EXPORT_SYMBOL_GPL(blkg_conf_finish);
944 static int blkcg_print_stat(struct seq_file *sf, void *v)
946 struct blkcg *blkcg = css_to_blkcg(seq_css(sf));
947 struct blkcg_gq *blkg;
951 hlist_for_each_entry_rcu(blkg, &blkcg->blkg_list, blkcg_node) {
953 struct blkg_rwstat rwstat;
954 u64 rbytes, wbytes, rios, wios;
956 dname = blkg_dev_name(blkg);
960 spin_lock_irq(blkg->q->queue_lock);
962 rwstat = blkg_rwstat_recursive_sum(blkg, NULL,
963 offsetof(struct blkcg_gq, stat_bytes));
964 rbytes = atomic64_read(&rwstat.aux_cnt[BLKG_RWSTAT_READ]);
965 wbytes = atomic64_read(&rwstat.aux_cnt[BLKG_RWSTAT_WRITE]);
967 rwstat = blkg_rwstat_recursive_sum(blkg, NULL,
968 offsetof(struct blkcg_gq, stat_ios));
969 rios = atomic64_read(&rwstat.aux_cnt[BLKG_RWSTAT_READ]);
970 wios = atomic64_read(&rwstat.aux_cnt[BLKG_RWSTAT_WRITE]);
972 spin_unlock_irq(blkg->q->queue_lock);
974 if (rbytes || wbytes || rios || wios)
975 seq_printf(sf, "%s rbytes=%llu wbytes=%llu rios=%llu wios=%llu\n",
976 dname, rbytes, wbytes, rios, wios);
983 static struct cftype blkcg_files[] = {
986 .flags = CFTYPE_NOT_ON_ROOT,
987 .seq_show = blkcg_print_stat,
992 static struct cftype blkcg_legacy_files[] = {
994 .name = "reset_stats",
995 .write_u64 = blkcg_reset_stats,
1001 * blkcg_css_offline - cgroup css_offline callback
1002 * @css: css of interest
1004 * This function is called when @css is about to go away and responsible
1005 * for shooting down all blkgs associated with @css. blkgs should be
1006 * removed while holding both q and blkcg locks. As blkcg lock is nested
1007 * inside q lock, this function performs reverse double lock dancing.
1009 * This is the blkcg counterpart of ioc_release_fn().
1011 static void blkcg_css_offline(struct cgroup_subsys_state *css)
1013 struct blkcg *blkcg = css_to_blkcg(css);
1015 spin_lock_irq(&blkcg->lock);
1017 while (!hlist_empty(&blkcg->blkg_list)) {
1018 struct blkcg_gq *blkg = hlist_entry(blkcg->blkg_list.first,
1019 struct blkcg_gq, blkcg_node);
1020 struct request_queue *q = blkg->q;
1022 if (spin_trylock(q->queue_lock)) {
1024 spin_unlock(q->queue_lock);
1026 spin_unlock_irq(&blkcg->lock);
1028 spin_lock_irq(&blkcg->lock);
1032 spin_unlock_irq(&blkcg->lock);
1034 wb_blkcg_offline(blkcg);
1037 static void blkcg_css_free(struct cgroup_subsys_state *css)
1039 struct blkcg *blkcg = css_to_blkcg(css);
1042 mutex_lock(&blkcg_pol_mutex);
1044 list_del(&blkcg->all_blkcgs_node);
1046 for (i = 0; i < BLKCG_MAX_POLS; i++)
1048 blkcg_policy[i]->cpd_free_fn(blkcg->cpd[i]);
1050 mutex_unlock(&blkcg_pol_mutex);
1055 static struct cgroup_subsys_state *
1056 blkcg_css_alloc(struct cgroup_subsys_state *parent_css)
1058 struct blkcg *blkcg;
1059 struct cgroup_subsys_state *ret;
1062 mutex_lock(&blkcg_pol_mutex);
1065 blkcg = &blkcg_root;
1067 blkcg = kzalloc(sizeof(*blkcg), GFP_KERNEL);
1069 ret = ERR_PTR(-ENOMEM);
1074 for (i = 0; i < BLKCG_MAX_POLS ; i++) {
1075 struct blkcg_policy *pol = blkcg_policy[i];
1076 struct blkcg_policy_data *cpd;
1079 * If the policy hasn't been attached yet, wait for it
1080 * to be attached before doing anything else. Otherwise,
1081 * check if the policy requires any specific per-cgroup
1082 * data: if it does, allocate and initialize it.
1084 if (!pol || !pol->cpd_alloc_fn)
1087 cpd = pol->cpd_alloc_fn(GFP_KERNEL);
1089 ret = ERR_PTR(-ENOMEM);
1092 blkcg->cpd[i] = cpd;
1095 if (pol->cpd_init_fn)
1096 pol->cpd_init_fn(cpd);
1099 spin_lock_init(&blkcg->lock);
1100 INIT_RADIX_TREE(&blkcg->blkg_tree, GFP_NOWAIT | __GFP_NOWARN);
1101 INIT_HLIST_HEAD(&blkcg->blkg_list);
1102 #ifdef CONFIG_CGROUP_WRITEBACK
1103 INIT_LIST_HEAD(&blkcg->cgwb_list);
1105 list_add_tail(&blkcg->all_blkcgs_node, &all_blkcgs);
1107 mutex_unlock(&blkcg_pol_mutex);
1111 for (i--; i >= 0; i--)
1113 blkcg_policy[i]->cpd_free_fn(blkcg->cpd[i]);
1116 mutex_unlock(&blkcg_pol_mutex);
1121 * blkcg_init_queue - initialize blkcg part of request queue
1122 * @q: request_queue to initialize
1124 * Called from blk_alloc_queue_node(). Responsible for initializing blkcg
1125 * part of new request_queue @q.
1128 * 0 on success, -errno on failure.
1130 int blkcg_init_queue(struct request_queue *q)
1132 struct blkcg_gq *new_blkg, *blkg;
1136 new_blkg = blkg_alloc(&blkcg_root, q, GFP_KERNEL);
1140 preloaded = !radix_tree_preload(GFP_KERNEL);
1143 * Make sure the root blkg exists and count the existing blkgs. As
1144 * @q is bypassing at this point, blkg_lookup_create() can't be
1145 * used. Open code insertion.
1148 spin_lock_irq(q->queue_lock);
1149 blkg = blkg_create(&blkcg_root, q, new_blkg);
1150 spin_unlock_irq(q->queue_lock);
1154 radix_tree_preload_end();
1157 return PTR_ERR(blkg);
1159 q->root_blkg = blkg;
1160 q->root_rl.blkg = blkg;
1162 ret = blk_throtl_init(q);
1164 spin_lock_irq(q->queue_lock);
1165 blkg_destroy_all(q);
1166 spin_unlock_irq(q->queue_lock);
1172 * blkcg_drain_queue - drain blkcg part of request_queue
1173 * @q: request_queue to drain
1175 * Called from blk_drain_queue(). Responsible for draining blkcg part.
1177 void blkcg_drain_queue(struct request_queue *q)
1179 lockdep_assert_held(q->queue_lock);
1182 * @q could be exiting and already have destroyed all blkgs as
1183 * indicated by NULL root_blkg. If so, don't confuse policies.
1188 blk_throtl_drain(q);
1192 * blkcg_exit_queue - exit and release blkcg part of request_queue
1193 * @q: request_queue being released
1195 * Called from blk_release_queue(). Responsible for exiting blkcg part.
1197 void blkcg_exit_queue(struct request_queue *q)
1199 spin_lock_irq(q->queue_lock);
1200 blkg_destroy_all(q);
1201 spin_unlock_irq(q->queue_lock);
1207 * We cannot support shared io contexts, as we have no mean to support
1208 * two tasks with the same ioc in two different groups without major rework
1209 * of the main cic data structures. For now we allow a task to change
1210 * its cgroup only if it's the only owner of its ioc.
1212 static int blkcg_can_attach(struct cgroup_taskset *tset)
1214 struct task_struct *task;
1215 struct cgroup_subsys_state *dst_css;
1216 struct io_context *ioc;
1219 /* task_lock() is needed to avoid races with exit_io_context() */
1220 cgroup_taskset_for_each(task, dst_css, tset) {
1222 ioc = task->io_context;
1223 if (ioc && atomic_read(&ioc->nr_tasks) > 1)
1232 static void blkcg_bind(struct cgroup_subsys_state *root_css)
1236 mutex_lock(&blkcg_pol_mutex);
1238 for (i = 0; i < BLKCG_MAX_POLS; i++) {
1239 struct blkcg_policy *pol = blkcg_policy[i];
1240 struct blkcg *blkcg;
1242 if (!pol || !pol->cpd_bind_fn)
1245 list_for_each_entry(blkcg, &all_blkcgs, all_blkcgs_node)
1246 if (blkcg->cpd[pol->plid])
1247 pol->cpd_bind_fn(blkcg->cpd[pol->plid]);
1249 mutex_unlock(&blkcg_pol_mutex);
1252 struct cgroup_subsys io_cgrp_subsys = {
1253 .css_alloc = blkcg_css_alloc,
1254 .css_offline = blkcg_css_offline,
1255 .css_free = blkcg_css_free,
1256 .can_attach = blkcg_can_attach,
1258 .dfl_cftypes = blkcg_files,
1259 .legacy_cftypes = blkcg_legacy_files,
1260 .legacy_name = "blkio",
1263 * This ensures that, if available, memcg is automatically enabled
1264 * together on the default hierarchy so that the owner cgroup can
1265 * be retrieved from writeback pages.
1267 .depends_on = 1 << memory_cgrp_id,
1270 EXPORT_SYMBOL_GPL(io_cgrp_subsys);
1273 * blkcg_activate_policy - activate a blkcg policy on a request_queue
1274 * @q: request_queue of interest
1275 * @pol: blkcg policy to activate
1277 * Activate @pol on @q. Requires %GFP_KERNEL context. @q goes through
1278 * bypass mode to populate its blkgs with policy_data for @pol.
1280 * Activation happens with @q bypassed, so nobody would be accessing blkgs
1281 * from IO path. Update of each blkg is protected by both queue and blkcg
1282 * locks so that holding either lock and testing blkcg_policy_enabled() is
1283 * always enough for dereferencing policy data.
1285 * The caller is responsible for synchronizing [de]activations and policy
1286 * [un]registerations. Returns 0 on success, -errno on failure.
1288 int blkcg_activate_policy(struct request_queue *q,
1289 const struct blkcg_policy *pol)
1291 struct blkg_policy_data *pd_prealloc = NULL;
1292 struct blkcg_gq *blkg;
1295 if (blkcg_policy_enabled(q, pol))
1299 blk_mq_freeze_queue(q);
1301 blk_queue_bypass_start(q);
1304 pd_prealloc = pol->pd_alloc_fn(GFP_KERNEL, q->node);
1307 goto out_bypass_end;
1311 spin_lock_irq(q->queue_lock);
1313 list_for_each_entry(blkg, &q->blkg_list, q_node) {
1314 struct blkg_policy_data *pd;
1316 if (blkg->pd[pol->plid])
1319 pd = pol->pd_alloc_fn(GFP_NOWAIT | __GFP_NOWARN, q->node);
1321 swap(pd, pd_prealloc);
1323 spin_unlock_irq(q->queue_lock);
1327 blkg->pd[pol->plid] = pd;
1329 pd->plid = pol->plid;
1330 if (pol->pd_init_fn)
1331 pol->pd_init_fn(pd);
1334 __set_bit(pol->plid, q->blkcg_pols);
1337 spin_unlock_irq(q->queue_lock);
1340 blk_mq_unfreeze_queue(q);
1342 blk_queue_bypass_end(q);
1344 pol->pd_free_fn(pd_prealloc);
1347 EXPORT_SYMBOL_GPL(blkcg_activate_policy);
1350 * blkcg_deactivate_policy - deactivate a blkcg policy on a request_queue
1351 * @q: request_queue of interest
1352 * @pol: blkcg policy to deactivate
1354 * Deactivate @pol on @q. Follows the same synchronization rules as
1355 * blkcg_activate_policy().
1357 void blkcg_deactivate_policy(struct request_queue *q,
1358 const struct blkcg_policy *pol)
1360 struct blkcg_gq *blkg;
1362 if (!blkcg_policy_enabled(q, pol))
1366 blk_mq_freeze_queue(q);
1368 blk_queue_bypass_start(q);
1370 spin_lock_irq(q->queue_lock);
1372 __clear_bit(pol->plid, q->blkcg_pols);
1374 list_for_each_entry(blkg, &q->blkg_list, q_node) {
1375 /* grab blkcg lock too while removing @pd from @blkg */
1376 spin_lock(&blkg->blkcg->lock);
1378 if (blkg->pd[pol->plid]) {
1379 if (pol->pd_offline_fn)
1380 pol->pd_offline_fn(blkg->pd[pol->plid]);
1381 pol->pd_free_fn(blkg->pd[pol->plid]);
1382 blkg->pd[pol->plid] = NULL;
1385 spin_unlock(&blkg->blkcg->lock);
1388 spin_unlock_irq(q->queue_lock);
1391 blk_mq_unfreeze_queue(q);
1393 blk_queue_bypass_end(q);
1395 EXPORT_SYMBOL_GPL(blkcg_deactivate_policy);
1398 * blkcg_policy_register - register a blkcg policy
1399 * @pol: blkcg policy to register
1401 * Register @pol with blkcg core. Might sleep and @pol may be modified on
1402 * successful registration. Returns 0 on success and -errno on failure.
1404 int blkcg_policy_register(struct blkcg_policy *pol)
1406 struct blkcg *blkcg;
1409 mutex_lock(&blkcg_pol_register_mutex);
1410 mutex_lock(&blkcg_pol_mutex);
1412 /* find an empty slot */
1414 for (i = 0; i < BLKCG_MAX_POLS; i++)
1415 if (!blkcg_policy[i])
1417 if (i >= BLKCG_MAX_POLS)
1422 blkcg_policy[pol->plid] = pol;
1424 /* allocate and install cpd's */
1425 if (pol->cpd_alloc_fn) {
1426 list_for_each_entry(blkcg, &all_blkcgs, all_blkcgs_node) {
1427 struct blkcg_policy_data *cpd;
1429 cpd = pol->cpd_alloc_fn(GFP_KERNEL);
1433 blkcg->cpd[pol->plid] = cpd;
1435 cpd->plid = pol->plid;
1436 pol->cpd_init_fn(cpd);
1440 mutex_unlock(&blkcg_pol_mutex);
1442 /* everything is in place, add intf files for the new policy */
1443 if (pol->dfl_cftypes)
1444 WARN_ON(cgroup_add_dfl_cftypes(&io_cgrp_subsys,
1446 if (pol->legacy_cftypes)
1447 WARN_ON(cgroup_add_legacy_cftypes(&io_cgrp_subsys,
1448 pol->legacy_cftypes));
1449 mutex_unlock(&blkcg_pol_register_mutex);
1453 if (pol->cpd_alloc_fn) {
1454 list_for_each_entry(blkcg, &all_blkcgs, all_blkcgs_node) {
1455 if (blkcg->cpd[pol->plid]) {
1456 pol->cpd_free_fn(blkcg->cpd[pol->plid]);
1457 blkcg->cpd[pol->plid] = NULL;
1461 blkcg_policy[pol->plid] = NULL;
1463 mutex_unlock(&blkcg_pol_mutex);
1464 mutex_unlock(&blkcg_pol_register_mutex);
1467 EXPORT_SYMBOL_GPL(blkcg_policy_register);
1470 * blkcg_policy_unregister - unregister a blkcg policy
1471 * @pol: blkcg policy to unregister
1473 * Undo blkcg_policy_register(@pol). Might sleep.
1475 void blkcg_policy_unregister(struct blkcg_policy *pol)
1477 struct blkcg *blkcg;
1479 mutex_lock(&blkcg_pol_register_mutex);
1481 if (WARN_ON(blkcg_policy[pol->plid] != pol))
1484 /* kill the intf files first */
1485 if (pol->dfl_cftypes)
1486 cgroup_rm_cftypes(pol->dfl_cftypes);
1487 if (pol->legacy_cftypes)
1488 cgroup_rm_cftypes(pol->legacy_cftypes);
1490 /* remove cpds and unregister */
1491 mutex_lock(&blkcg_pol_mutex);
1493 if (pol->cpd_alloc_fn) {
1494 list_for_each_entry(blkcg, &all_blkcgs, all_blkcgs_node) {
1495 if (blkcg->cpd[pol->plid]) {
1496 pol->cpd_free_fn(blkcg->cpd[pol->plid]);
1497 blkcg->cpd[pol->plid] = NULL;
1501 blkcg_policy[pol->plid] = NULL;
1503 mutex_unlock(&blkcg_pol_mutex);
1505 mutex_unlock(&blkcg_pol_register_mutex);
1507 EXPORT_SYMBOL_GPL(blkcg_policy_unregister);