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_pd_offline(struct blkcg_gq *blkg)
314 lockdep_assert_held(blkg->q->queue_lock);
315 lockdep_assert_held(&blkg->blkcg->lock);
317 for (i = 0; i < BLKCG_MAX_POLS; i++) {
318 struct blkcg_policy *pol = blkcg_policy[i];
320 if (blkg->pd[i] && !blkg->pd[i]->offline &&
321 pol->pd_offline_fn) {
322 pol->pd_offline_fn(blkg->pd[i]);
323 blkg->pd[i]->offline = true;
328 static void blkg_destroy(struct blkcg_gq *blkg)
330 struct blkcg *blkcg = blkg->blkcg;
331 struct blkcg_gq *parent = blkg->parent;
333 lockdep_assert_held(blkg->q->queue_lock);
334 lockdep_assert_held(&blkcg->lock);
336 /* Something wrong if we are trying to remove same group twice */
337 WARN_ON_ONCE(list_empty(&blkg->q_node));
338 WARN_ON_ONCE(hlist_unhashed(&blkg->blkcg_node));
341 blkg_rwstat_add_aux(&parent->stat_bytes, &blkg->stat_bytes);
342 blkg_rwstat_add_aux(&parent->stat_ios, &blkg->stat_ios);
345 blkg->online = false;
347 radix_tree_delete(&blkcg->blkg_tree, blkg->q->id);
348 list_del_init(&blkg->q_node);
349 hlist_del_init_rcu(&blkg->blkcg_node);
352 * Both setting lookup hint to and clearing it from @blkg are done
353 * under queue_lock. If it's not pointing to @blkg now, it never
354 * will. Hint assignment itself can race safely.
356 if (rcu_access_pointer(blkcg->blkg_hint) == blkg)
357 rcu_assign_pointer(blkcg->blkg_hint, NULL);
360 * Put the reference taken at the time of creation so that when all
361 * queues are gone, group can be destroyed.
367 * blkg_destroy_all - destroy all blkgs associated with a request_queue
368 * @q: request_queue of interest
370 * Destroy all blkgs associated with @q.
372 static void blkg_destroy_all(struct request_queue *q)
374 struct blkcg_gq *blkg, *n;
376 lockdep_assert_held(q->queue_lock);
378 list_for_each_entry_safe(blkg, n, &q->blkg_list, q_node) {
379 struct blkcg *blkcg = blkg->blkcg;
381 spin_lock(&blkcg->lock);
382 blkg_pd_offline(blkg);
384 spin_unlock(&blkcg->lock);
388 q->root_rl.blkg = NULL;
392 * A group is RCU protected, but having an rcu lock does not mean that one
393 * can access all the fields of blkg and assume these are valid. For
394 * example, don't try to follow throtl_data and request queue links.
396 * Having a reference to blkg under an rcu allows accesses to only values
397 * local to groups like group stats and group rate limits.
399 void __blkg_release_rcu(struct rcu_head *rcu_head)
401 struct blkcg_gq *blkg = container_of(rcu_head, struct blkcg_gq, rcu_head);
403 /* release the blkcg and parent blkg refs this blkg has been holding */
404 css_put(&blkg->blkcg->css);
406 blkg_put(blkg->parent);
408 wb_congested_put(blkg->wb_congested);
412 EXPORT_SYMBOL_GPL(__blkg_release_rcu);
415 * The next function used by blk_queue_for_each_rl(). It's a bit tricky
416 * because the root blkg uses @q->root_rl instead of its own rl.
418 struct request_list *__blk_queue_next_rl(struct request_list *rl,
419 struct request_queue *q)
421 struct list_head *ent;
422 struct blkcg_gq *blkg;
425 * Determine the current blkg list_head. The first entry is
426 * root_rl which is off @q->blkg_list and mapped to the head.
428 if (rl == &q->root_rl) {
430 /* There are no more block groups, hence no request lists */
434 blkg = container_of(rl, struct blkcg_gq, rl);
438 /* walk to the next list_head, skip root blkcg */
440 if (ent == &q->root_blkg->q_node)
442 if (ent == &q->blkg_list)
445 blkg = container_of(ent, struct blkcg_gq, q_node);
449 static int blkcg_reset_stats(struct cgroup_subsys_state *css,
450 struct cftype *cftype, u64 val)
452 struct blkcg *blkcg = css_to_blkcg(css);
453 struct blkcg_gq *blkg;
456 mutex_lock(&blkcg_pol_mutex);
457 spin_lock_irq(&blkcg->lock);
460 * Note that stat reset is racy - it doesn't synchronize against
461 * stat updates. This is a debug feature which shouldn't exist
462 * anyway. If you get hit by a race, retry.
464 hlist_for_each_entry(blkg, &blkcg->blkg_list, blkcg_node) {
465 blkg_rwstat_reset(&blkg->stat_bytes);
466 blkg_rwstat_reset(&blkg->stat_ios);
468 for (i = 0; i < BLKCG_MAX_POLS; i++) {
469 struct blkcg_policy *pol = blkcg_policy[i];
471 if (blkg->pd[i] && pol->pd_reset_stats_fn)
472 pol->pd_reset_stats_fn(blkg->pd[i]);
476 spin_unlock_irq(&blkcg->lock);
477 mutex_unlock(&blkcg_pol_mutex);
481 const char *blkg_dev_name(struct blkcg_gq *blkg)
483 /* some drivers (floppy) instantiate a queue w/o disk registered */
484 if (blkg->q->backing_dev_info->dev)
485 return dev_name(blkg->q->backing_dev_info->dev);
488 EXPORT_SYMBOL_GPL(blkg_dev_name);
491 * blkcg_print_blkgs - helper for printing per-blkg data
492 * @sf: seq_file to print to
493 * @blkcg: blkcg of interest
494 * @prfill: fill function to print out a blkg
495 * @pol: policy in question
496 * @data: data to be passed to @prfill
497 * @show_total: to print out sum of prfill return values or not
499 * This function invokes @prfill on each blkg of @blkcg if pd for the
500 * policy specified by @pol exists. @prfill is invoked with @sf, the
501 * policy data and @data and the matching queue lock held. If @show_total
502 * is %true, the sum of the return values from @prfill is printed with
503 * "Total" label at the end.
505 * This is to be used to construct print functions for
506 * cftype->read_seq_string method.
508 void blkcg_print_blkgs(struct seq_file *sf, struct blkcg *blkcg,
509 u64 (*prfill)(struct seq_file *,
510 struct blkg_policy_data *, int),
511 const struct blkcg_policy *pol, int data,
514 struct blkcg_gq *blkg;
518 hlist_for_each_entry_rcu(blkg, &blkcg->blkg_list, blkcg_node) {
519 spin_lock_irq(blkg->q->queue_lock);
520 if (blkcg_policy_enabled(blkg->q, pol))
521 total += prfill(sf, blkg->pd[pol->plid], data);
522 spin_unlock_irq(blkg->q->queue_lock);
527 seq_printf(sf, "Total %llu\n", (unsigned long long)total);
529 EXPORT_SYMBOL_GPL(blkcg_print_blkgs);
532 * __blkg_prfill_u64 - prfill helper for a single u64 value
533 * @sf: seq_file to print to
534 * @pd: policy private data of interest
537 * Print @v to @sf for the device assocaited with @pd.
539 u64 __blkg_prfill_u64(struct seq_file *sf, struct blkg_policy_data *pd, u64 v)
541 const char *dname = blkg_dev_name(pd->blkg);
546 seq_printf(sf, "%s %llu\n", dname, (unsigned long long)v);
549 EXPORT_SYMBOL_GPL(__blkg_prfill_u64);
552 * __blkg_prfill_rwstat - prfill helper for a blkg_rwstat
553 * @sf: seq_file to print to
554 * @pd: policy private data of interest
555 * @rwstat: rwstat to print
557 * Print @rwstat to @sf for the device assocaited with @pd.
559 u64 __blkg_prfill_rwstat(struct seq_file *sf, struct blkg_policy_data *pd,
560 const struct blkg_rwstat *rwstat)
562 static const char *rwstr[] = {
563 [BLKG_RWSTAT_READ] = "Read",
564 [BLKG_RWSTAT_WRITE] = "Write",
565 [BLKG_RWSTAT_SYNC] = "Sync",
566 [BLKG_RWSTAT_ASYNC] = "Async",
568 const char *dname = blkg_dev_name(pd->blkg);
575 for (i = 0; i < BLKG_RWSTAT_NR; i++)
576 seq_printf(sf, "%s %s %llu\n", dname, rwstr[i],
577 (unsigned long long)atomic64_read(&rwstat->aux_cnt[i]));
579 v = atomic64_read(&rwstat->aux_cnt[BLKG_RWSTAT_READ]) +
580 atomic64_read(&rwstat->aux_cnt[BLKG_RWSTAT_WRITE]);
581 seq_printf(sf, "%s Total %llu\n", dname, (unsigned long long)v);
584 EXPORT_SYMBOL_GPL(__blkg_prfill_rwstat);
587 * blkg_prfill_stat - prfill callback for blkg_stat
588 * @sf: seq_file to print to
589 * @pd: policy private data of interest
590 * @off: offset to the blkg_stat in @pd
592 * prfill callback for printing a blkg_stat.
594 u64 blkg_prfill_stat(struct seq_file *sf, struct blkg_policy_data *pd, int off)
596 return __blkg_prfill_u64(sf, pd, blkg_stat_read((void *)pd + off));
598 EXPORT_SYMBOL_GPL(blkg_prfill_stat);
601 * blkg_prfill_rwstat - prfill callback for blkg_rwstat
602 * @sf: seq_file to print to
603 * @pd: policy private data of interest
604 * @off: offset to the blkg_rwstat in @pd
606 * prfill callback for printing a blkg_rwstat.
608 u64 blkg_prfill_rwstat(struct seq_file *sf, struct blkg_policy_data *pd,
611 struct blkg_rwstat rwstat = blkg_rwstat_read((void *)pd + off);
613 return __blkg_prfill_rwstat(sf, pd, &rwstat);
615 EXPORT_SYMBOL_GPL(blkg_prfill_rwstat);
617 static u64 blkg_prfill_rwstat_field(struct seq_file *sf,
618 struct blkg_policy_data *pd, int off)
620 struct blkg_rwstat rwstat = blkg_rwstat_read((void *)pd->blkg + off);
622 return __blkg_prfill_rwstat(sf, pd, &rwstat);
626 * blkg_print_stat_bytes - seq_show callback for blkg->stat_bytes
627 * @sf: seq_file to print to
630 * To be used as cftype->seq_show to print blkg->stat_bytes.
631 * cftype->private must be set to the blkcg_policy.
633 int blkg_print_stat_bytes(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_bytes), true);
640 EXPORT_SYMBOL_GPL(blkg_print_stat_bytes);
643 * blkg_print_stat_bytes - seq_show callback for blkg->stat_ios
644 * @sf: seq_file to print to
647 * To be used as cftype->seq_show to print blkg->stat_ios. cftype->private
648 * must be set to the blkcg_policy.
650 int blkg_print_stat_ios(struct seq_file *sf, void *v)
652 blkcg_print_blkgs(sf, css_to_blkcg(seq_css(sf)),
653 blkg_prfill_rwstat_field, (void *)seq_cft(sf)->private,
654 offsetof(struct blkcg_gq, stat_ios), true);
657 EXPORT_SYMBOL_GPL(blkg_print_stat_ios);
659 static u64 blkg_prfill_rwstat_field_recursive(struct seq_file *sf,
660 struct blkg_policy_data *pd,
663 struct blkg_rwstat rwstat = blkg_rwstat_recursive_sum(pd->blkg,
665 return __blkg_prfill_rwstat(sf, pd, &rwstat);
669 * blkg_print_stat_bytes_recursive - recursive version of blkg_print_stat_bytes
670 * @sf: seq_file to print to
673 int blkg_print_stat_bytes_recursive(struct seq_file *sf, void *v)
675 blkcg_print_blkgs(sf, css_to_blkcg(seq_css(sf)),
676 blkg_prfill_rwstat_field_recursive,
677 (void *)seq_cft(sf)->private,
678 offsetof(struct blkcg_gq, stat_bytes), true);
681 EXPORT_SYMBOL_GPL(blkg_print_stat_bytes_recursive);
684 * blkg_print_stat_ios_recursive - recursive version of blkg_print_stat_ios
685 * @sf: seq_file to print to
688 int blkg_print_stat_ios_recursive(struct seq_file *sf, void *v)
690 blkcg_print_blkgs(sf, css_to_blkcg(seq_css(sf)),
691 blkg_prfill_rwstat_field_recursive,
692 (void *)seq_cft(sf)->private,
693 offsetof(struct blkcg_gq, stat_ios), true);
696 EXPORT_SYMBOL_GPL(blkg_print_stat_ios_recursive);
699 * blkg_stat_recursive_sum - collect hierarchical blkg_stat
700 * @blkg: blkg of interest
701 * @pol: blkcg_policy which contains the blkg_stat
702 * @off: offset to the blkg_stat in blkg_policy_data or @blkg
704 * Collect the blkg_stat specified by @blkg, @pol and @off and all its
705 * online descendants and their aux counts. The caller must be holding the
706 * queue lock for online tests.
708 * If @pol is NULL, blkg_stat is at @off bytes into @blkg; otherwise, it is
709 * at @off bytes into @blkg's blkg_policy_data of the policy.
711 u64 blkg_stat_recursive_sum(struct blkcg_gq *blkg,
712 struct blkcg_policy *pol, int off)
714 struct blkcg_gq *pos_blkg;
715 struct cgroup_subsys_state *pos_css;
718 lockdep_assert_held(blkg->q->queue_lock);
721 blkg_for_each_descendant_pre(pos_blkg, pos_css, blkg) {
722 struct blkg_stat *stat;
724 if (!pos_blkg->online)
728 stat = (void *)blkg_to_pd(pos_blkg, pol) + off;
730 stat = (void *)blkg + off;
732 sum += blkg_stat_read(stat) + atomic64_read(&stat->aux_cnt);
738 EXPORT_SYMBOL_GPL(blkg_stat_recursive_sum);
741 * blkg_rwstat_recursive_sum - collect hierarchical blkg_rwstat
742 * @blkg: blkg of interest
743 * @pol: blkcg_policy which contains the blkg_rwstat
744 * @off: offset to the blkg_rwstat in blkg_policy_data or @blkg
746 * Collect the blkg_rwstat specified by @blkg, @pol and @off and all its
747 * online descendants and their aux counts. The caller must be holding the
748 * queue lock for online tests.
750 * If @pol is NULL, blkg_rwstat is at @off bytes into @blkg; otherwise, it
751 * is at @off bytes into @blkg's blkg_policy_data of the policy.
753 struct blkg_rwstat blkg_rwstat_recursive_sum(struct blkcg_gq *blkg,
754 struct blkcg_policy *pol, int off)
756 struct blkcg_gq *pos_blkg;
757 struct cgroup_subsys_state *pos_css;
758 struct blkg_rwstat sum = { };
761 lockdep_assert_held(blkg->q->queue_lock);
764 blkg_for_each_descendant_pre(pos_blkg, pos_css, blkg) {
765 struct blkg_rwstat *rwstat;
767 if (!pos_blkg->online)
771 rwstat = (void *)blkg_to_pd(pos_blkg, pol) + off;
773 rwstat = (void *)pos_blkg + off;
775 for (i = 0; i < BLKG_RWSTAT_NR; i++)
776 atomic64_add(atomic64_read(&rwstat->aux_cnt[i]) +
777 percpu_counter_sum_positive(&rwstat->cpu_cnt[i]),
784 EXPORT_SYMBOL_GPL(blkg_rwstat_recursive_sum);
786 /* Performs queue bypass and policy enabled checks then looks up blkg. */
787 static struct blkcg_gq *blkg_lookup_check(struct blkcg *blkcg,
788 const struct blkcg_policy *pol,
789 struct request_queue *q)
791 WARN_ON_ONCE(!rcu_read_lock_held());
792 lockdep_assert_held(q->queue_lock);
794 if (!blkcg_policy_enabled(q, pol))
795 return ERR_PTR(-EOPNOTSUPP);
798 * This could be the first entry point of blkcg implementation and
799 * we shouldn't allow anything to go through for a bypassing queue.
801 if (unlikely(blk_queue_bypass(q)))
802 return ERR_PTR(blk_queue_dying(q) ? -ENODEV : -EBUSY);
804 return __blkg_lookup(blkcg, q, true /* update_hint */);
808 * blkg_conf_prep - parse and prepare for per-blkg config update
809 * @blkcg: target block cgroup
810 * @pol: target policy
811 * @input: input string
812 * @ctx: blkg_conf_ctx to be filled
814 * Parse per-blkg config update from @input and initialize @ctx with the
815 * result. @ctx->blkg points to the blkg to be updated and @ctx->body the
816 * part of @input following MAJ:MIN. This function returns with RCU read
817 * lock and queue lock held and must be paired with blkg_conf_finish().
819 int blkg_conf_prep(struct blkcg *blkcg, const struct blkcg_policy *pol,
820 char *input, struct blkg_conf_ctx *ctx)
821 __acquires(rcu) __acquires(disk->queue->queue_lock)
823 struct gendisk *disk;
824 struct request_queue *q;
825 struct blkcg_gq *blkg;
826 unsigned int major, minor;
827 int key_len, part, ret;
830 if (sscanf(input, "%u:%u%n", &major, &minor, &key_len) != 2)
833 body = input + key_len;
836 body = skip_spaces(body);
838 disk = get_gendisk(MKDEV(major, minor), &part);
849 spin_lock_irq(q->queue_lock);
851 blkg = blkg_lookup_check(blkcg, pol, q);
861 * Create blkgs walking down from blkcg_root to @blkcg, so that all
862 * non-root blkgs have access to their parents.
865 struct blkcg *pos = blkcg;
866 struct blkcg *parent;
867 struct blkcg_gq *new_blkg;
869 parent = blkcg_parent(blkcg);
870 while (parent && !__blkg_lookup(parent, q, false)) {
872 parent = blkcg_parent(parent);
875 /* Drop locks to do new blkg allocation with GFP_KERNEL. */
876 spin_unlock_irq(q->queue_lock);
879 new_blkg = blkg_alloc(pos, q, GFP_KERNEL);
880 if (unlikely(!new_blkg)) {
886 spin_lock_irq(q->queue_lock);
888 blkg = blkg_lookup_check(pos, pol, q);
897 blkg = blkg_create(pos, q, new_blkg);
898 if (unlikely(IS_ERR(blkg))) {
914 spin_unlock_irq(q->queue_lock);
917 put_disk_and_module(disk);
919 * If queue was bypassing, we should retry. Do so after a
920 * short msleep(). It isn't strictly necessary but queue
921 * can be bypassing for some time and it's always nice to
922 * avoid busy looping.
926 ret = restart_syscall();
930 EXPORT_SYMBOL_GPL(blkg_conf_prep);
933 * blkg_conf_finish - finish up per-blkg config update
934 * @ctx: blkg_conf_ctx intiailized by blkg_conf_prep()
936 * Finish up after per-blkg config update. This function must be paired
937 * with blkg_conf_prep().
939 void blkg_conf_finish(struct blkg_conf_ctx *ctx)
940 __releases(ctx->disk->queue->queue_lock) __releases(rcu)
942 spin_unlock_irq(ctx->disk->queue->queue_lock);
944 put_disk_and_module(ctx->disk);
946 EXPORT_SYMBOL_GPL(blkg_conf_finish);
948 static int blkcg_print_stat(struct seq_file *sf, void *v)
950 struct blkcg *blkcg = css_to_blkcg(seq_css(sf));
951 struct blkcg_gq *blkg;
955 hlist_for_each_entry_rcu(blkg, &blkcg->blkg_list, blkcg_node) {
957 struct blkg_rwstat rwstat;
958 u64 rbytes, wbytes, rios, wios;
960 dname = blkg_dev_name(blkg);
964 spin_lock_irq(blkg->q->queue_lock);
966 rwstat = blkg_rwstat_recursive_sum(blkg, NULL,
967 offsetof(struct blkcg_gq, stat_bytes));
968 rbytes = atomic64_read(&rwstat.aux_cnt[BLKG_RWSTAT_READ]);
969 wbytes = atomic64_read(&rwstat.aux_cnt[BLKG_RWSTAT_WRITE]);
971 rwstat = blkg_rwstat_recursive_sum(blkg, NULL,
972 offsetof(struct blkcg_gq, stat_ios));
973 rios = atomic64_read(&rwstat.aux_cnt[BLKG_RWSTAT_READ]);
974 wios = atomic64_read(&rwstat.aux_cnt[BLKG_RWSTAT_WRITE]);
976 spin_unlock_irq(blkg->q->queue_lock);
978 if (rbytes || wbytes || rios || wios)
979 seq_printf(sf, "%s rbytes=%llu wbytes=%llu rios=%llu wios=%llu\n",
980 dname, rbytes, wbytes, rios, wios);
987 static struct cftype blkcg_files[] = {
990 .flags = CFTYPE_NOT_ON_ROOT,
991 .seq_show = blkcg_print_stat,
996 static struct cftype blkcg_legacy_files[] = {
998 .name = "reset_stats",
999 .write_u64 = blkcg_reset_stats,
1005 * blkcg_css_offline - cgroup css_offline callback
1006 * @css: css of interest
1008 * This function is called when @css is about to go away and responsible
1009 * for offlining all blkgs pd and killing all wbs associated with @css.
1010 * blkgs pd offline should be done while holding both q and blkcg locks.
1011 * As blkcg lock is nested inside q lock, this function performs reverse
1012 * double lock dancing.
1014 * This is the blkcg counterpart of ioc_release_fn().
1016 static void blkcg_css_offline(struct cgroup_subsys_state *css)
1018 struct blkcg *blkcg = css_to_blkcg(css);
1019 struct blkcg_gq *blkg;
1021 spin_lock_irq(&blkcg->lock);
1023 hlist_for_each_entry(blkg, &blkcg->blkg_list, blkcg_node) {
1024 struct request_queue *q = blkg->q;
1026 if (spin_trylock(q->queue_lock)) {
1027 blkg_pd_offline(blkg);
1028 spin_unlock(q->queue_lock);
1030 spin_unlock_irq(&blkcg->lock);
1032 spin_lock_irq(&blkcg->lock);
1036 spin_unlock_irq(&blkcg->lock);
1038 wb_blkcg_offline(blkcg);
1042 * blkcg_destroy_all_blkgs - destroy all blkgs associated with a blkcg
1043 * @blkcg: blkcg of interest
1045 * This function is called when blkcg css is about to free and responsible for
1046 * destroying all blkgs associated with @blkcg.
1047 * blkgs should be removed while holding both q and blkcg locks. As blkcg lock
1048 * is nested inside q lock, this function performs reverse double lock dancing.
1050 static void blkcg_destroy_all_blkgs(struct blkcg *blkcg)
1052 spin_lock_irq(&blkcg->lock);
1053 while (!hlist_empty(&blkcg->blkg_list)) {
1054 struct blkcg_gq *blkg = hlist_entry(blkcg->blkg_list.first,
1057 struct request_queue *q = blkg->q;
1059 if (spin_trylock(q->queue_lock)) {
1061 spin_unlock(q->queue_lock);
1063 spin_unlock_irq(&blkcg->lock);
1065 spin_lock_irq(&blkcg->lock);
1068 spin_unlock_irq(&blkcg->lock);
1071 static void blkcg_css_free(struct cgroup_subsys_state *css)
1073 struct blkcg *blkcg = css_to_blkcg(css);
1076 blkcg_destroy_all_blkgs(blkcg);
1078 mutex_lock(&blkcg_pol_mutex);
1080 list_del(&blkcg->all_blkcgs_node);
1082 for (i = 0; i < BLKCG_MAX_POLS; i++)
1084 blkcg_policy[i]->cpd_free_fn(blkcg->cpd[i]);
1086 mutex_unlock(&blkcg_pol_mutex);
1091 static struct cgroup_subsys_state *
1092 blkcg_css_alloc(struct cgroup_subsys_state *parent_css)
1094 struct blkcg *blkcg;
1095 struct cgroup_subsys_state *ret;
1098 mutex_lock(&blkcg_pol_mutex);
1101 blkcg = &blkcg_root;
1103 blkcg = kzalloc(sizeof(*blkcg), GFP_KERNEL);
1105 ret = ERR_PTR(-ENOMEM);
1110 for (i = 0; i < BLKCG_MAX_POLS ; i++) {
1111 struct blkcg_policy *pol = blkcg_policy[i];
1112 struct blkcg_policy_data *cpd;
1115 * If the policy hasn't been attached yet, wait for it
1116 * to be attached before doing anything else. Otherwise,
1117 * check if the policy requires any specific per-cgroup
1118 * data: if it does, allocate and initialize it.
1120 if (!pol || !pol->cpd_alloc_fn)
1123 cpd = pol->cpd_alloc_fn(GFP_KERNEL);
1125 ret = ERR_PTR(-ENOMEM);
1128 blkcg->cpd[i] = cpd;
1131 if (pol->cpd_init_fn)
1132 pol->cpd_init_fn(cpd);
1135 spin_lock_init(&blkcg->lock);
1136 INIT_RADIX_TREE(&blkcg->blkg_tree, GFP_NOWAIT | __GFP_NOWARN);
1137 INIT_HLIST_HEAD(&blkcg->blkg_list);
1138 #ifdef CONFIG_CGROUP_WRITEBACK
1139 INIT_LIST_HEAD(&blkcg->cgwb_list);
1141 list_add_tail(&blkcg->all_blkcgs_node, &all_blkcgs);
1143 mutex_unlock(&blkcg_pol_mutex);
1147 for (i--; i >= 0; i--)
1149 blkcg_policy[i]->cpd_free_fn(blkcg->cpd[i]);
1151 if (blkcg != &blkcg_root)
1154 mutex_unlock(&blkcg_pol_mutex);
1159 * blkcg_init_queue - initialize blkcg part of request queue
1160 * @q: request_queue to initialize
1162 * Called from blk_alloc_queue_node(). Responsible for initializing blkcg
1163 * part of new request_queue @q.
1166 * 0 on success, -errno on failure.
1168 int blkcg_init_queue(struct request_queue *q)
1170 struct blkcg_gq *new_blkg, *blkg;
1174 new_blkg = blkg_alloc(&blkcg_root, q, GFP_KERNEL);
1178 preloaded = !radix_tree_preload(GFP_KERNEL);
1181 * Make sure the root blkg exists and count the existing blkgs. As
1182 * @q is bypassing at this point, blkg_lookup_create() can't be
1183 * used. Open code insertion.
1186 spin_lock_irq(q->queue_lock);
1187 blkg = blkg_create(&blkcg_root, q, new_blkg);
1188 spin_unlock_irq(q->queue_lock);
1192 radix_tree_preload_end();
1195 return PTR_ERR(blkg);
1197 q->root_blkg = blkg;
1198 q->root_rl.blkg = blkg;
1200 ret = blk_throtl_init(q);
1202 spin_lock_irq(q->queue_lock);
1203 blkg_destroy_all(q);
1204 spin_unlock_irq(q->queue_lock);
1210 * blkcg_drain_queue - drain blkcg part of request_queue
1211 * @q: request_queue to drain
1213 * Called from blk_drain_queue(). Responsible for draining blkcg part.
1215 void blkcg_drain_queue(struct request_queue *q)
1217 lockdep_assert_held(q->queue_lock);
1220 * @q could be exiting and already have destroyed all blkgs as
1221 * indicated by NULL root_blkg. If so, don't confuse policies.
1226 blk_throtl_drain(q);
1230 * blkcg_exit_queue - exit and release blkcg part of request_queue
1231 * @q: request_queue being released
1233 * Called from blk_release_queue(). Responsible for exiting blkcg part.
1235 void blkcg_exit_queue(struct request_queue *q)
1237 spin_lock_irq(q->queue_lock);
1238 blkg_destroy_all(q);
1239 spin_unlock_irq(q->queue_lock);
1245 * We cannot support shared io contexts, as we have no mean to support
1246 * two tasks with the same ioc in two different groups without major rework
1247 * of the main cic data structures. For now we allow a task to change
1248 * its cgroup only if it's the only owner of its ioc.
1250 static int blkcg_can_attach(struct cgroup_taskset *tset)
1252 struct task_struct *task;
1253 struct cgroup_subsys_state *dst_css;
1254 struct io_context *ioc;
1257 /* task_lock() is needed to avoid races with exit_io_context() */
1258 cgroup_taskset_for_each(task, dst_css, tset) {
1260 ioc = task->io_context;
1261 if (ioc && atomic_read(&ioc->nr_tasks) > 1)
1270 static void blkcg_bind(struct cgroup_subsys_state *root_css)
1274 mutex_lock(&blkcg_pol_mutex);
1276 for (i = 0; i < BLKCG_MAX_POLS; i++) {
1277 struct blkcg_policy *pol = blkcg_policy[i];
1278 struct blkcg *blkcg;
1280 if (!pol || !pol->cpd_bind_fn)
1283 list_for_each_entry(blkcg, &all_blkcgs, all_blkcgs_node)
1284 if (blkcg->cpd[pol->plid])
1285 pol->cpd_bind_fn(blkcg->cpd[pol->plid]);
1287 mutex_unlock(&blkcg_pol_mutex);
1290 struct cgroup_subsys io_cgrp_subsys = {
1291 .css_alloc = blkcg_css_alloc,
1292 .css_offline = blkcg_css_offline,
1293 .css_free = blkcg_css_free,
1294 .can_attach = blkcg_can_attach,
1296 .dfl_cftypes = blkcg_files,
1297 .legacy_cftypes = blkcg_legacy_files,
1298 .legacy_name = "blkio",
1301 * This ensures that, if available, memcg is automatically enabled
1302 * together on the default hierarchy so that the owner cgroup can
1303 * be retrieved from writeback pages.
1305 .depends_on = 1 << memory_cgrp_id,
1308 EXPORT_SYMBOL_GPL(io_cgrp_subsys);
1311 * blkcg_activate_policy - activate a blkcg policy on a request_queue
1312 * @q: request_queue of interest
1313 * @pol: blkcg policy to activate
1315 * Activate @pol on @q. Requires %GFP_KERNEL context. @q goes through
1316 * bypass mode to populate its blkgs with policy_data for @pol.
1318 * Activation happens with @q bypassed, so nobody would be accessing blkgs
1319 * from IO path. Update of each blkg is protected by both queue and blkcg
1320 * locks so that holding either lock and testing blkcg_policy_enabled() is
1321 * always enough for dereferencing policy data.
1323 * The caller is responsible for synchronizing [de]activations and policy
1324 * [un]registerations. Returns 0 on success, -errno on failure.
1326 int blkcg_activate_policy(struct request_queue *q,
1327 const struct blkcg_policy *pol)
1329 struct blkg_policy_data *pd_prealloc = NULL;
1330 struct blkcg_gq *blkg;
1333 if (blkcg_policy_enabled(q, pol))
1337 blk_mq_freeze_queue(q);
1339 blk_queue_bypass_start(q);
1342 pd_prealloc = pol->pd_alloc_fn(GFP_KERNEL, q->node);
1345 goto out_bypass_end;
1349 spin_lock_irq(q->queue_lock);
1351 list_for_each_entry(blkg, &q->blkg_list, q_node) {
1352 struct blkg_policy_data *pd;
1354 if (blkg->pd[pol->plid])
1357 pd = pol->pd_alloc_fn(GFP_NOWAIT | __GFP_NOWARN, q->node);
1359 swap(pd, pd_prealloc);
1361 spin_unlock_irq(q->queue_lock);
1365 blkg->pd[pol->plid] = pd;
1367 pd->plid = pol->plid;
1368 if (pol->pd_init_fn)
1369 pol->pd_init_fn(pd);
1372 __set_bit(pol->plid, q->blkcg_pols);
1375 spin_unlock_irq(q->queue_lock);
1378 blk_mq_unfreeze_queue(q);
1380 blk_queue_bypass_end(q);
1382 pol->pd_free_fn(pd_prealloc);
1385 EXPORT_SYMBOL_GPL(blkcg_activate_policy);
1388 * blkcg_deactivate_policy - deactivate a blkcg policy on a request_queue
1389 * @q: request_queue of interest
1390 * @pol: blkcg policy to deactivate
1392 * Deactivate @pol on @q. Follows the same synchronization rules as
1393 * blkcg_activate_policy().
1395 void blkcg_deactivate_policy(struct request_queue *q,
1396 const struct blkcg_policy *pol)
1398 struct blkcg_gq *blkg;
1400 if (!blkcg_policy_enabled(q, pol))
1404 blk_mq_freeze_queue(q);
1406 blk_queue_bypass_start(q);
1408 spin_lock_irq(q->queue_lock);
1410 __clear_bit(pol->plid, q->blkcg_pols);
1412 list_for_each_entry(blkg, &q->blkg_list, q_node) {
1413 /* grab blkcg lock too while removing @pd from @blkg */
1414 spin_lock(&blkg->blkcg->lock);
1416 if (blkg->pd[pol->plid]) {
1417 if (!blkg->pd[pol->plid]->offline &&
1418 pol->pd_offline_fn) {
1419 pol->pd_offline_fn(blkg->pd[pol->plid]);
1420 blkg->pd[pol->plid]->offline = true;
1422 pol->pd_free_fn(blkg->pd[pol->plid]);
1423 blkg->pd[pol->plid] = NULL;
1426 spin_unlock(&blkg->blkcg->lock);
1429 spin_unlock_irq(q->queue_lock);
1432 blk_mq_unfreeze_queue(q);
1434 blk_queue_bypass_end(q);
1436 EXPORT_SYMBOL_GPL(blkcg_deactivate_policy);
1439 * blkcg_policy_register - register a blkcg policy
1440 * @pol: blkcg policy to register
1442 * Register @pol with blkcg core. Might sleep and @pol may be modified on
1443 * successful registration. Returns 0 on success and -errno on failure.
1445 int blkcg_policy_register(struct blkcg_policy *pol)
1447 struct blkcg *blkcg;
1450 mutex_lock(&blkcg_pol_register_mutex);
1451 mutex_lock(&blkcg_pol_mutex);
1453 /* find an empty slot */
1455 for (i = 0; i < BLKCG_MAX_POLS; i++)
1456 if (!blkcg_policy[i])
1458 if (i >= BLKCG_MAX_POLS)
1461 /* Make sure cpd/pd_alloc_fn and cpd/pd_free_fn in pairs */
1462 if ((!pol->cpd_alloc_fn ^ !pol->cpd_free_fn) ||
1463 (!pol->pd_alloc_fn ^ !pol->pd_free_fn))
1468 blkcg_policy[pol->plid] = pol;
1470 /* allocate and install cpd's */
1471 if (pol->cpd_alloc_fn) {
1472 list_for_each_entry(blkcg, &all_blkcgs, all_blkcgs_node) {
1473 struct blkcg_policy_data *cpd;
1475 cpd = pol->cpd_alloc_fn(GFP_KERNEL);
1479 blkcg->cpd[pol->plid] = cpd;
1481 cpd->plid = pol->plid;
1482 pol->cpd_init_fn(cpd);
1486 mutex_unlock(&blkcg_pol_mutex);
1488 /* everything is in place, add intf files for the new policy */
1489 if (pol->dfl_cftypes)
1490 WARN_ON(cgroup_add_dfl_cftypes(&io_cgrp_subsys,
1492 if (pol->legacy_cftypes)
1493 WARN_ON(cgroup_add_legacy_cftypes(&io_cgrp_subsys,
1494 pol->legacy_cftypes));
1495 mutex_unlock(&blkcg_pol_register_mutex);
1499 if (pol->cpd_free_fn) {
1500 list_for_each_entry(blkcg, &all_blkcgs, all_blkcgs_node) {
1501 if (blkcg->cpd[pol->plid]) {
1502 pol->cpd_free_fn(blkcg->cpd[pol->plid]);
1503 blkcg->cpd[pol->plid] = NULL;
1507 blkcg_policy[pol->plid] = NULL;
1509 mutex_unlock(&blkcg_pol_mutex);
1510 mutex_unlock(&blkcg_pol_register_mutex);
1513 EXPORT_SYMBOL_GPL(blkcg_policy_register);
1516 * blkcg_policy_unregister - unregister a blkcg policy
1517 * @pol: blkcg policy to unregister
1519 * Undo blkcg_policy_register(@pol). Might sleep.
1521 void blkcg_policy_unregister(struct blkcg_policy *pol)
1523 struct blkcg *blkcg;
1525 mutex_lock(&blkcg_pol_register_mutex);
1527 if (WARN_ON(blkcg_policy[pol->plid] != pol))
1530 /* kill the intf files first */
1531 if (pol->dfl_cftypes)
1532 cgroup_rm_cftypes(pol->dfl_cftypes);
1533 if (pol->legacy_cftypes)
1534 cgroup_rm_cftypes(pol->legacy_cftypes);
1536 /* remove cpds and unregister */
1537 mutex_lock(&blkcg_pol_mutex);
1539 if (pol->cpd_free_fn) {
1540 list_for_each_entry(blkcg, &all_blkcgs, all_blkcgs_node) {
1541 if (blkcg->cpd[pol->plid]) {
1542 pol->cpd_free_fn(blkcg->cpd[pol->plid]);
1543 blkcg->cpd[pol->plid] = NULL;
1547 blkcg_policy[pol->plid] = NULL;
1549 mutex_unlock(&blkcg_pol_mutex);
1551 mutex_unlock(&blkcg_pol_register_mutex);
1553 EXPORT_SYMBOL_GPL(blkcg_policy_unregister);