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->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 gfp mask allows blocking, this function temporarily drops rcu and queue
169 * locks to allocate memory.
171 static struct blkcg_gq *blkg_create(struct blkcg *blkcg,
172 struct request_queue *q, gfp_t gfp,
173 const struct blkcg_policy *pol)
175 struct blkcg_gq *blkg = NULL;
176 struct bdi_writeback_congested *wb_congested;
178 const bool drop_locks = gfpflags_allow_blocking(gfp);
179 bool preloaded = false;
181 WARN_ON_ONCE(!rcu_read_lock_held());
182 lockdep_assert_held(q->queue_lock);
184 /* blkg holds a reference to blkcg */
185 if (!css_tryget_online(&blkcg->css)) {
191 spin_unlock_irq(q->queue_lock);
195 wb_congested = wb_congested_get_create(q->backing_dev_info,
197 blkg = blkg_alloc(blkcg, q, gfp);
200 preloaded = !radix_tree_preload(gfp);
202 spin_lock_irq(q->queue_lock);
205 if (unlikely(!wb_congested || !blkg)) {
210 blkg->wb_congested = wb_congested;
213 WARN_ON(!drop_locks);
215 if (!blkcg_policy_enabled(q, pol)) {
221 * This could be the first entry point of blkcg implementation
222 * and we shouldn't allow anything to go through for a bypassing
225 if (unlikely(blk_queue_bypass(q))) {
226 ret = blk_queue_dying(q) ? -ENODEV : -EBUSY;
232 if (blkcg_parent(blkcg)) {
233 blkg->parent = __blkg_lookup(blkcg_parent(blkcg), q, false);
234 if (WARN_ON_ONCE(!blkg->parent)) {
238 blkg_get(blkg->parent);
241 /* invoke per-policy init */
242 for (i = 0; i < BLKCG_MAX_POLS; i++) {
243 struct blkcg_policy *pol = blkcg_policy[i];
245 if (blkg->pd[i] && pol->pd_init_fn)
246 pol->pd_init_fn(blkg->pd[i]);
250 spin_lock(&blkcg->lock);
251 ret = radix_tree_insert(&blkcg->blkg_tree, q->id, blkg);
253 hlist_add_head_rcu(&blkg->blkcg_node, &blkcg->blkg_list);
254 list_add(&blkg->q_node, &q->blkg_list);
256 for (i = 0; i < BLKCG_MAX_POLS; i++) {
257 struct blkcg_policy *pol = blkcg_policy[i];
259 if (blkg->pd[i] && pol->pd_online_fn)
260 pol->pd_online_fn(blkg->pd[i]);
265 radix_tree_preload_end();
267 spin_unlock(&blkcg->lock);
272 /* @blkg failed fully initialized, use the usual release path */
278 radix_tree_preload_end();
280 wb_congested_put(wb_congested);
281 css_put(&blkcg->css);
288 * __blkg_lookup_create - lookup blkg, try to create one if not there
289 * @blkcg: blkcg of interest
290 * @q: request_queue of interest
292 * @pol: blkcg policy (optional)
294 * Lookup blkg for the @blkcg - @q pair. If it doesn't exist, try to
295 * create one. blkg creation is performed recursively from blkcg_root such
296 * that all non-root blkg's have access to the parent blkg. This function
297 * should be called under RCU read lock and @q->queue_lock.
299 * When gfp mask allows blocking, rcu and queue locks may be dropped for
300 * allocating memory. In this case, the locks will be reacquired on return.
302 * Returns pointer to the looked up or created blkg on success, ERR_PTR()
303 * value on error. If @q is dead, returns ERR_PTR(-EINVAL). If @q is not
304 * dead and bypassing, returns ERR_PTR(-EBUSY).
306 struct blkcg_gq *__blkg_lookup_create(struct blkcg *blkcg,
307 struct request_queue *q, gfp_t gfp,
308 const struct blkcg_policy *pol)
310 struct blkcg_gq *blkg;
312 WARN_ON_ONCE(!rcu_read_lock_held());
313 lockdep_assert_held(q->queue_lock);
315 blkg = __blkg_lookup(blkcg, q, true);
320 * Create blkgs walking down from blkcg_root to @blkcg, so that all
321 * non-root blkgs have access to their parents.
324 struct blkcg *pos = blkcg;
325 struct blkcg *parent = blkcg_parent(blkcg);
327 while (parent && !__blkg_lookup(parent, q, false)) {
329 parent = blkcg_parent(parent);
332 blkg = blkg_create(pos, q, gfp, pol);
333 if (pos == blkcg || IS_ERR(blkg))
339 * blkg_lookup_create - lookup blkg, try to create one if not there
341 * Performs an initial queue bypass check and then passes control to
342 * __blkg_lookup_create().
344 struct blkcg_gq *blkg_lookup_create(struct blkcg *blkcg,
345 struct request_queue *q, gfp_t gfp,
346 const struct blkcg_policy *pol)
348 WARN_ON_ONCE(!rcu_read_lock_held());
349 lockdep_assert_held(q->queue_lock);
352 * This could be the first entry point of blkcg implementation and
353 * we shouldn't allow anything to go through for a bypassing queue.
355 if (unlikely(blk_queue_bypass(q)))
356 return ERR_PTR(blk_queue_dying(q) ? -ENODEV : -EBUSY);
358 return __blkg_lookup_create(blkcg, q, gfp, pol);
361 static void blkg_destroy(struct blkcg_gq *blkg)
363 struct blkcg *blkcg = blkg->blkcg;
364 struct blkcg_gq *parent = blkg->parent;
367 lockdep_assert_held(blkg->q->queue_lock);
368 lockdep_assert_held(&blkcg->lock);
370 /* Something wrong if we are trying to remove same group twice */
371 WARN_ON_ONCE(list_empty(&blkg->q_node));
372 WARN_ON_ONCE(hlist_unhashed(&blkg->blkcg_node));
374 for (i = 0; i < BLKCG_MAX_POLS; i++) {
375 struct blkcg_policy *pol = blkcg_policy[i];
377 if (blkg->pd[i] && pol->pd_offline_fn)
378 pol->pd_offline_fn(blkg->pd[i]);
382 blkg_rwstat_add_aux(&parent->stat_bytes, &blkg->stat_bytes);
383 blkg_rwstat_add_aux(&parent->stat_ios, &blkg->stat_ios);
386 blkg->online = false;
388 radix_tree_delete(&blkcg->blkg_tree, blkg->q->id);
389 list_del_init(&blkg->q_node);
390 hlist_del_init_rcu(&blkg->blkcg_node);
393 * Both setting lookup hint to and clearing it from @blkg are done
394 * under queue_lock. If it's not pointing to @blkg now, it never
395 * will. Hint assignment itself can race safely.
397 if (rcu_access_pointer(blkcg->blkg_hint) == blkg)
398 rcu_assign_pointer(blkcg->blkg_hint, NULL);
401 * Put the reference taken at the time of creation so that when all
402 * queues are gone, group can be destroyed.
408 * blkg_destroy_all - destroy all blkgs associated with a request_queue
409 * @q: request_queue of interest
411 * Destroy all blkgs associated with @q.
413 static void blkg_destroy_all(struct request_queue *q)
415 struct blkcg_gq *blkg, *n;
417 lockdep_assert_held(q->queue_lock);
419 list_for_each_entry_safe(blkg, n, &q->blkg_list, q_node) {
420 struct blkcg *blkcg = blkg->blkcg;
422 spin_lock(&blkcg->lock);
424 spin_unlock(&blkcg->lock);
428 q->root_rl.blkg = NULL;
432 * A group is RCU protected, but having an rcu lock does not mean that one
433 * can access all the fields of blkg and assume these are valid. For
434 * example, don't try to follow throtl_data and request queue links.
436 * Having a reference to blkg under an rcu allows accesses to only values
437 * local to groups like group stats and group rate limits.
439 void __blkg_release_rcu(struct rcu_head *rcu_head)
441 struct blkcg_gq *blkg = container_of(rcu_head, struct blkcg_gq, rcu_head);
443 /* release the blkcg and parent blkg refs this blkg has been holding */
444 css_put(&blkg->blkcg->css);
446 blkg_put(blkg->parent);
448 wb_congested_put(blkg->wb_congested);
452 EXPORT_SYMBOL_GPL(__blkg_release_rcu);
455 * The next function used by blk_queue_for_each_rl(). It's a bit tricky
456 * because the root blkg uses @q->root_rl instead of its own rl.
458 struct request_list *__blk_queue_next_rl(struct request_list *rl,
459 struct request_queue *q)
461 struct list_head *ent;
462 struct blkcg_gq *blkg;
465 * Determine the current blkg list_head. The first entry is
466 * root_rl which is off @q->blkg_list and mapped to the head.
468 if (rl == &q->root_rl) {
470 /* There are no more block groups, hence no request lists */
474 blkg = container_of(rl, struct blkcg_gq, rl);
478 /* walk to the next list_head, skip root blkcg */
480 if (ent == &q->root_blkg->q_node)
482 if (ent == &q->blkg_list)
485 blkg = container_of(ent, struct blkcg_gq, q_node);
489 static int blkcg_reset_stats(struct cgroup_subsys_state *css,
490 struct cftype *cftype, u64 val)
492 struct blkcg *blkcg = css_to_blkcg(css);
493 struct blkcg_gq *blkg;
496 mutex_lock(&blkcg_pol_mutex);
497 spin_lock_irq(&blkcg->lock);
500 * Note that stat reset is racy - it doesn't synchronize against
501 * stat updates. This is a debug feature which shouldn't exist
502 * anyway. If you get hit by a race, retry.
504 hlist_for_each_entry(blkg, &blkcg->blkg_list, blkcg_node) {
505 blkg_rwstat_reset(&blkg->stat_bytes);
506 blkg_rwstat_reset(&blkg->stat_ios);
508 for (i = 0; i < BLKCG_MAX_POLS; i++) {
509 struct blkcg_policy *pol = blkcg_policy[i];
511 if (blkg->pd[i] && pol->pd_reset_stats_fn)
512 pol->pd_reset_stats_fn(blkg->pd[i]);
516 spin_unlock_irq(&blkcg->lock);
517 mutex_unlock(&blkcg_pol_mutex);
521 const char *blkg_dev_name(struct blkcg_gq *blkg)
523 /* some drivers (floppy) instantiate a queue w/o disk registered */
524 if (blkg->q->backing_dev_info->dev)
525 return dev_name(blkg->q->backing_dev_info->dev);
528 EXPORT_SYMBOL_GPL(blkg_dev_name);
531 * blkcg_print_blkgs - helper for printing per-blkg data
532 * @sf: seq_file to print to
533 * @blkcg: blkcg of interest
534 * @prfill: fill function to print out a blkg
535 * @pol: policy in question
536 * @data: data to be passed to @prfill
537 * @show_total: to print out sum of prfill return values or not
539 * This function invokes @prfill on each blkg of @blkcg if pd for the
540 * policy specified by @pol exists. @prfill is invoked with @sf, the
541 * policy data and @data and the matching queue lock held. If @show_total
542 * is %true, the sum of the return values from @prfill is printed with
543 * "Total" label at the end.
545 * This is to be used to construct print functions for
546 * cftype->read_seq_string method.
548 void blkcg_print_blkgs(struct seq_file *sf, struct blkcg *blkcg,
549 u64 (*prfill)(struct seq_file *,
550 struct blkg_policy_data *, int),
551 const struct blkcg_policy *pol, int data,
554 struct blkcg_gq *blkg;
558 hlist_for_each_entry_rcu(blkg, &blkcg->blkg_list, blkcg_node) {
559 spin_lock_irq(blkg->q->queue_lock);
560 if (blkcg_policy_enabled(blkg->q, pol))
561 total += prfill(sf, blkg->pd[pol->plid], data);
562 spin_unlock_irq(blkg->q->queue_lock);
567 seq_printf(sf, "Total %llu\n", (unsigned long long)total);
569 EXPORT_SYMBOL_GPL(blkcg_print_blkgs);
572 * __blkg_prfill_u64 - prfill helper for a single u64 value
573 * @sf: seq_file to print to
574 * @pd: policy private data of interest
577 * Print @v to @sf for the device assocaited with @pd.
579 u64 __blkg_prfill_u64(struct seq_file *sf, struct blkg_policy_data *pd, u64 v)
581 const char *dname = blkg_dev_name(pd->blkg);
586 seq_printf(sf, "%s %llu\n", dname, (unsigned long long)v);
589 EXPORT_SYMBOL_GPL(__blkg_prfill_u64);
592 * __blkg_prfill_rwstat - prfill helper for a blkg_rwstat
593 * @sf: seq_file to print to
594 * @pd: policy private data of interest
595 * @rwstat: rwstat to print
597 * Print @rwstat to @sf for the device assocaited with @pd.
599 u64 __blkg_prfill_rwstat(struct seq_file *sf, struct blkg_policy_data *pd,
600 const struct blkg_rwstat *rwstat)
602 static const char *rwstr[] = {
603 [BLKG_RWSTAT_READ] = "Read",
604 [BLKG_RWSTAT_WRITE] = "Write",
605 [BLKG_RWSTAT_SYNC] = "Sync",
606 [BLKG_RWSTAT_ASYNC] = "Async",
608 const char *dname = blkg_dev_name(pd->blkg);
615 for (i = 0; i < BLKG_RWSTAT_NR; i++)
616 seq_printf(sf, "%s %s %llu\n", dname, rwstr[i],
617 (unsigned long long)atomic64_read(&rwstat->aux_cnt[i]));
619 v = atomic64_read(&rwstat->aux_cnt[BLKG_RWSTAT_READ]) +
620 atomic64_read(&rwstat->aux_cnt[BLKG_RWSTAT_WRITE]);
621 seq_printf(sf, "%s Total %llu\n", dname, (unsigned long long)v);
624 EXPORT_SYMBOL_GPL(__blkg_prfill_rwstat);
627 * blkg_prfill_stat - prfill callback for blkg_stat
628 * @sf: seq_file to print to
629 * @pd: policy private data of interest
630 * @off: offset to the blkg_stat in @pd
632 * prfill callback for printing a blkg_stat.
634 u64 blkg_prfill_stat(struct seq_file *sf, struct blkg_policy_data *pd, int off)
636 return __blkg_prfill_u64(sf, pd, blkg_stat_read((void *)pd + off));
638 EXPORT_SYMBOL_GPL(blkg_prfill_stat);
641 * blkg_prfill_rwstat - prfill callback for blkg_rwstat
642 * @sf: seq_file to print to
643 * @pd: policy private data of interest
644 * @off: offset to the blkg_rwstat in @pd
646 * prfill callback for printing a blkg_rwstat.
648 u64 blkg_prfill_rwstat(struct seq_file *sf, struct blkg_policy_data *pd,
651 struct blkg_rwstat rwstat = blkg_rwstat_read((void *)pd + off);
653 return __blkg_prfill_rwstat(sf, pd, &rwstat);
655 EXPORT_SYMBOL_GPL(blkg_prfill_rwstat);
657 static u64 blkg_prfill_rwstat_field(struct seq_file *sf,
658 struct blkg_policy_data *pd, int off)
660 struct blkg_rwstat rwstat = blkg_rwstat_read((void *)pd->blkg + off);
662 return __blkg_prfill_rwstat(sf, pd, &rwstat);
666 * blkg_print_stat_bytes - seq_show callback for blkg->stat_bytes
667 * @sf: seq_file to print to
670 * To be used as cftype->seq_show to print blkg->stat_bytes.
671 * cftype->private must be set to the blkcg_policy.
673 int blkg_print_stat_bytes(struct seq_file *sf, void *v)
675 blkcg_print_blkgs(sf, css_to_blkcg(seq_css(sf)),
676 blkg_prfill_rwstat_field, (void *)seq_cft(sf)->private,
677 offsetof(struct blkcg_gq, stat_bytes), true);
680 EXPORT_SYMBOL_GPL(blkg_print_stat_bytes);
683 * blkg_print_stat_bytes - seq_show callback for blkg->stat_ios
684 * @sf: seq_file to print to
687 * To be used as cftype->seq_show to print blkg->stat_ios. cftype->private
688 * must be set to the blkcg_policy.
690 int blkg_print_stat_ios(struct seq_file *sf, void *v)
692 blkcg_print_blkgs(sf, css_to_blkcg(seq_css(sf)),
693 blkg_prfill_rwstat_field, (void *)seq_cft(sf)->private,
694 offsetof(struct blkcg_gq, stat_ios), true);
697 EXPORT_SYMBOL_GPL(blkg_print_stat_ios);
699 static u64 blkg_prfill_rwstat_field_recursive(struct seq_file *sf,
700 struct blkg_policy_data *pd,
703 struct blkg_rwstat rwstat = blkg_rwstat_recursive_sum(pd->blkg,
705 return __blkg_prfill_rwstat(sf, pd, &rwstat);
709 * blkg_print_stat_bytes_recursive - recursive version of blkg_print_stat_bytes
710 * @sf: seq_file to print to
713 int blkg_print_stat_bytes_recursive(struct seq_file *sf, void *v)
715 blkcg_print_blkgs(sf, css_to_blkcg(seq_css(sf)),
716 blkg_prfill_rwstat_field_recursive,
717 (void *)seq_cft(sf)->private,
718 offsetof(struct blkcg_gq, stat_bytes), true);
721 EXPORT_SYMBOL_GPL(blkg_print_stat_bytes_recursive);
724 * blkg_print_stat_ios_recursive - recursive version of blkg_print_stat_ios
725 * @sf: seq_file to print to
728 int blkg_print_stat_ios_recursive(struct seq_file *sf, void *v)
730 blkcg_print_blkgs(sf, css_to_blkcg(seq_css(sf)),
731 blkg_prfill_rwstat_field_recursive,
732 (void *)seq_cft(sf)->private,
733 offsetof(struct blkcg_gq, stat_ios), true);
736 EXPORT_SYMBOL_GPL(blkg_print_stat_ios_recursive);
739 * blkg_stat_recursive_sum - collect hierarchical blkg_stat
740 * @blkg: blkg of interest
741 * @pol: blkcg_policy which contains the blkg_stat
742 * @off: offset to the blkg_stat in blkg_policy_data or @blkg
744 * Collect the blkg_stat specified by @blkg, @pol and @off and all its
745 * online descendants and their aux counts. The caller must be holding the
746 * queue lock for online tests.
748 * If @pol is NULL, blkg_stat is at @off bytes into @blkg; otherwise, it is
749 * at @off bytes into @blkg's blkg_policy_data of the policy.
751 u64 blkg_stat_recursive_sum(struct blkcg_gq *blkg,
752 struct blkcg_policy *pol, int off)
754 struct blkcg_gq *pos_blkg;
755 struct cgroup_subsys_state *pos_css;
758 lockdep_assert_held(blkg->q->queue_lock);
761 blkg_for_each_descendant_pre(pos_blkg, pos_css, blkg) {
762 struct blkg_stat *stat;
764 if (!pos_blkg->online)
768 stat = (void *)blkg_to_pd(pos_blkg, pol) + off;
770 stat = (void *)blkg + off;
772 sum += blkg_stat_read(stat) + atomic64_read(&stat->aux_cnt);
778 EXPORT_SYMBOL_GPL(blkg_stat_recursive_sum);
781 * blkg_rwstat_recursive_sum - collect hierarchical blkg_rwstat
782 * @blkg: blkg of interest
783 * @pol: blkcg_policy which contains the blkg_rwstat
784 * @off: offset to the blkg_rwstat in blkg_policy_data or @blkg
786 * Collect the blkg_rwstat specified by @blkg, @pol and @off and all its
787 * online descendants and their aux counts. The caller must be holding the
788 * queue lock for online tests.
790 * If @pol is NULL, blkg_rwstat is at @off bytes into @blkg; otherwise, it
791 * is at @off bytes into @blkg's blkg_policy_data of the policy.
793 struct blkg_rwstat blkg_rwstat_recursive_sum(struct blkcg_gq *blkg,
794 struct blkcg_policy *pol, int off)
796 struct blkcg_gq *pos_blkg;
797 struct cgroup_subsys_state *pos_css;
798 struct blkg_rwstat sum = { };
801 lockdep_assert_held(blkg->q->queue_lock);
804 blkg_for_each_descendant_pre(pos_blkg, pos_css, blkg) {
805 struct blkg_rwstat *rwstat;
807 if (!pos_blkg->online)
811 rwstat = (void *)blkg_to_pd(pos_blkg, pol) + off;
813 rwstat = (void *)pos_blkg + off;
815 for (i = 0; i < BLKG_RWSTAT_NR; i++)
816 atomic64_add(atomic64_read(&rwstat->aux_cnt[i]) +
817 percpu_counter_sum_positive(&rwstat->cpu_cnt[i]),
824 EXPORT_SYMBOL_GPL(blkg_rwstat_recursive_sum);
827 * blkg_conf_prep - parse and prepare for per-blkg config update
828 * @blkcg: target block cgroup
829 * @pol: target policy
830 * @input: input string
831 * @ctx: blkg_conf_ctx to be filled
833 * Parse per-blkg config update from @input and initialize @ctx with the
834 * result. @ctx->blkg points to the blkg to be updated and @ctx->body the
835 * part of @input following MAJ:MIN. This function returns with RCU read
836 * lock and queue lock held and must be paired with blkg_conf_finish().
838 int blkg_conf_prep(struct blkcg *blkcg, const struct blkcg_policy *pol,
839 char *input, struct blkg_conf_ctx *ctx)
840 __acquires(rcu) __acquires(disk->queue->queue_lock)
842 struct gendisk *disk;
843 struct blkcg_gq *blkg;
844 struct module *owner;
845 unsigned int major, minor;
846 int key_len, part, ret;
849 if (sscanf(input, "%u:%u%n", &major, &minor, &key_len) != 2)
852 body = input + key_len;
855 body = skip_spaces(body);
857 disk = get_gendisk(MKDEV(major, minor), &part);
861 owner = disk->fops->owner;
868 spin_lock_irq(disk->queue->queue_lock);
870 if (blkcg_policy_enabled(disk->queue, pol))
871 blkg = blkg_lookup_create(blkcg, disk->queue, GFP_KERNEL, pol);
873 blkg = ERR_PTR(-EOPNOTSUPP);
878 spin_unlock_irq(disk->queue->queue_lock);
879 owner = disk->fops->owner;
883 * If queue was bypassing, we should retry. Do so after a
884 * short msleep(). It isn't strictly necessary but queue
885 * can be bypassing for some time and it's always nice to
886 * avoid busy looping.
890 ret = restart_syscall();
900 EXPORT_SYMBOL_GPL(blkg_conf_prep);
903 * blkg_conf_finish - finish up per-blkg config update
904 * @ctx: blkg_conf_ctx intiailized by blkg_conf_prep()
906 * Finish up after per-blkg config update. This function must be paired
907 * with blkg_conf_prep().
909 void blkg_conf_finish(struct blkg_conf_ctx *ctx)
910 __releases(ctx->disk->queue->queue_lock) __releases(rcu)
912 struct module *owner;
914 spin_unlock_irq(ctx->disk->queue->queue_lock);
916 owner = ctx->disk->fops->owner;
920 EXPORT_SYMBOL_GPL(blkg_conf_finish);
922 static int blkcg_print_stat(struct seq_file *sf, void *v)
924 struct blkcg *blkcg = css_to_blkcg(seq_css(sf));
925 struct blkcg_gq *blkg;
929 hlist_for_each_entry_rcu(blkg, &blkcg->blkg_list, blkcg_node) {
931 struct blkg_rwstat rwstat;
932 u64 rbytes, wbytes, rios, wios;
934 dname = blkg_dev_name(blkg);
938 spin_lock_irq(blkg->q->queue_lock);
940 rwstat = blkg_rwstat_recursive_sum(blkg, NULL,
941 offsetof(struct blkcg_gq, stat_bytes));
942 rbytes = atomic64_read(&rwstat.aux_cnt[BLKG_RWSTAT_READ]);
943 wbytes = atomic64_read(&rwstat.aux_cnt[BLKG_RWSTAT_WRITE]);
945 rwstat = blkg_rwstat_recursive_sum(blkg, NULL,
946 offsetof(struct blkcg_gq, stat_ios));
947 rios = atomic64_read(&rwstat.aux_cnt[BLKG_RWSTAT_READ]);
948 wios = atomic64_read(&rwstat.aux_cnt[BLKG_RWSTAT_WRITE]);
950 spin_unlock_irq(blkg->q->queue_lock);
952 if (rbytes || wbytes || rios || wios)
953 seq_printf(sf, "%s rbytes=%llu wbytes=%llu rios=%llu wios=%llu\n",
954 dname, rbytes, wbytes, rios, wios);
961 static struct cftype blkcg_files[] = {
964 .flags = CFTYPE_NOT_ON_ROOT,
965 .seq_show = blkcg_print_stat,
970 static struct cftype blkcg_legacy_files[] = {
972 .name = "reset_stats",
973 .write_u64 = blkcg_reset_stats,
979 * blkcg_css_offline - cgroup css_offline callback
980 * @css: css of interest
982 * This function is called when @css is about to go away and responsible
983 * for shooting down all blkgs associated with @css. blkgs should be
984 * removed while holding both q and blkcg locks. As blkcg lock is nested
985 * inside q lock, this function performs reverse double lock dancing.
987 * This is the blkcg counterpart of ioc_release_fn().
989 static void blkcg_css_offline(struct cgroup_subsys_state *css)
991 struct blkcg *blkcg = css_to_blkcg(css);
993 spin_lock_irq(&blkcg->lock);
995 while (!hlist_empty(&blkcg->blkg_list)) {
996 struct blkcg_gq *blkg = hlist_entry(blkcg->blkg_list.first,
997 struct blkcg_gq, blkcg_node);
998 struct request_queue *q = blkg->q;
1000 if (spin_trylock(q->queue_lock)) {
1002 spin_unlock(q->queue_lock);
1004 spin_unlock_irq(&blkcg->lock);
1006 spin_lock_irq(&blkcg->lock);
1010 spin_unlock_irq(&blkcg->lock);
1012 wb_blkcg_offline(blkcg);
1015 static void blkcg_css_free(struct cgroup_subsys_state *css)
1017 struct blkcg *blkcg = css_to_blkcg(css);
1020 mutex_lock(&blkcg_pol_mutex);
1022 list_del(&blkcg->all_blkcgs_node);
1024 for (i = 0; i < BLKCG_MAX_POLS; i++)
1026 blkcg_policy[i]->cpd_free_fn(blkcg->cpd[i]);
1028 mutex_unlock(&blkcg_pol_mutex);
1033 static struct cgroup_subsys_state *
1034 blkcg_css_alloc(struct cgroup_subsys_state *parent_css)
1036 struct blkcg *blkcg;
1037 struct cgroup_subsys_state *ret;
1040 mutex_lock(&blkcg_pol_mutex);
1043 blkcg = &blkcg_root;
1045 blkcg = kzalloc(sizeof(*blkcg), GFP_KERNEL);
1047 ret = ERR_PTR(-ENOMEM);
1052 for (i = 0; i < BLKCG_MAX_POLS ; i++) {
1053 struct blkcg_policy *pol = blkcg_policy[i];
1054 struct blkcg_policy_data *cpd;
1057 * If the policy hasn't been attached yet, wait for it
1058 * to be attached before doing anything else. Otherwise,
1059 * check if the policy requires any specific per-cgroup
1060 * data: if it does, allocate and initialize it.
1062 if (!pol || !pol->cpd_alloc_fn)
1065 cpd = pol->cpd_alloc_fn(GFP_KERNEL);
1067 ret = ERR_PTR(-ENOMEM);
1070 blkcg->cpd[i] = cpd;
1073 if (pol->cpd_init_fn)
1074 pol->cpd_init_fn(cpd);
1077 spin_lock_init(&blkcg->lock);
1078 INIT_RADIX_TREE(&blkcg->blkg_tree, GFP_NOWAIT | __GFP_NOWARN);
1079 INIT_HLIST_HEAD(&blkcg->blkg_list);
1080 #ifdef CONFIG_CGROUP_WRITEBACK
1081 INIT_LIST_HEAD(&blkcg->cgwb_list);
1083 list_add_tail(&blkcg->all_blkcgs_node, &all_blkcgs);
1085 mutex_unlock(&blkcg_pol_mutex);
1089 for (i--; i >= 0; i--)
1091 blkcg_policy[i]->cpd_free_fn(blkcg->cpd[i]);
1094 mutex_unlock(&blkcg_pol_mutex);
1099 * blkcg_init_queue - initialize blkcg part of request queue
1100 * @q: request_queue to initialize
1102 * Called from blk_alloc_queue_node(). Responsible for initializing blkcg
1103 * part of new request_queue @q.
1106 * 0 on success, -errno on failure.
1108 int blkcg_init_queue(struct request_queue *q)
1110 struct blkcg_gq *blkg;
1114 spin_lock_irq(q->queue_lock);
1115 blkg = __blkg_lookup_create(&blkcg_root, q, GFP_KERNEL, NULL);
1116 spin_unlock_irq(q->queue_lock);
1120 return PTR_ERR(blkg);
1122 q->root_blkg = blkg;
1123 q->root_rl.blkg = blkg;
1125 ret = blk_throtl_init(q);
1127 spin_lock_irq(q->queue_lock);
1128 blkg_destroy_all(q);
1129 spin_unlock_irq(q->queue_lock);
1135 * blkcg_drain_queue - drain blkcg part of request_queue
1136 * @q: request_queue to drain
1138 * Called from blk_drain_queue(). Responsible for draining blkcg part.
1140 void blkcg_drain_queue(struct request_queue *q)
1142 lockdep_assert_held(q->queue_lock);
1145 * @q could be exiting and already have destroyed all blkgs as
1146 * indicated by NULL root_blkg. If so, don't confuse policies.
1151 blk_throtl_drain(q);
1155 * blkcg_exit_queue - exit and release blkcg part of request_queue
1156 * @q: request_queue being released
1158 * Called from blk_release_queue(). Responsible for exiting blkcg part.
1160 void blkcg_exit_queue(struct request_queue *q)
1162 spin_lock_irq(q->queue_lock);
1163 blkg_destroy_all(q);
1164 spin_unlock_irq(q->queue_lock);
1170 * We cannot support shared io contexts, as we have no mean to support
1171 * two tasks with the same ioc in two different groups without major rework
1172 * of the main cic data structures. For now we allow a task to change
1173 * its cgroup only if it's the only owner of its ioc.
1175 static int blkcg_can_attach(struct cgroup_taskset *tset)
1177 struct task_struct *task;
1178 struct cgroup_subsys_state *dst_css;
1179 struct io_context *ioc;
1182 /* task_lock() is needed to avoid races with exit_io_context() */
1183 cgroup_taskset_for_each(task, dst_css, tset) {
1185 ioc = task->io_context;
1186 if (ioc && atomic_read(&ioc->nr_tasks) > 1)
1195 static void blkcg_bind(struct cgroup_subsys_state *root_css)
1199 mutex_lock(&blkcg_pol_mutex);
1201 for (i = 0; i < BLKCG_MAX_POLS; i++) {
1202 struct blkcg_policy *pol = blkcg_policy[i];
1203 struct blkcg *blkcg;
1205 if (!pol || !pol->cpd_bind_fn)
1208 list_for_each_entry(blkcg, &all_blkcgs, all_blkcgs_node)
1209 if (blkcg->cpd[pol->plid])
1210 pol->cpd_bind_fn(blkcg->cpd[pol->plid]);
1212 mutex_unlock(&blkcg_pol_mutex);
1215 struct cgroup_subsys io_cgrp_subsys = {
1216 .css_alloc = blkcg_css_alloc,
1217 .css_offline = blkcg_css_offline,
1218 .css_free = blkcg_css_free,
1219 .can_attach = blkcg_can_attach,
1221 .dfl_cftypes = blkcg_files,
1222 .legacy_cftypes = blkcg_legacy_files,
1223 .legacy_name = "blkio",
1226 * This ensures that, if available, memcg is automatically enabled
1227 * together on the default hierarchy so that the owner cgroup can
1228 * be retrieved from writeback pages.
1230 .depends_on = 1 << memory_cgrp_id,
1233 EXPORT_SYMBOL_GPL(io_cgrp_subsys);
1236 * blkcg_activate_policy - activate a blkcg policy on a request_queue
1237 * @q: request_queue of interest
1238 * @pol: blkcg policy to activate
1240 * Activate @pol on @q. Requires %GFP_KERNEL context. @q goes through
1241 * bypass mode to populate its blkgs with policy_data for @pol.
1243 * Activation happens with @q bypassed, so nobody would be accessing blkgs
1244 * from IO path. Update of each blkg is protected by both queue and blkcg
1245 * locks so that holding either lock and testing blkcg_policy_enabled() is
1246 * always enough for dereferencing policy data.
1248 * The caller is responsible for synchronizing [de]activations and policy
1249 * [un]registerations. Returns 0 on success, -errno on failure.
1251 int blkcg_activate_policy(struct request_queue *q,
1252 const struct blkcg_policy *pol)
1254 struct blkg_policy_data *pd_prealloc = NULL;
1255 struct blkcg_gq *blkg;
1258 if (blkcg_policy_enabled(q, pol))
1262 blk_mq_freeze_queue(q);
1264 blk_queue_bypass_start(q);
1267 pd_prealloc = pol->pd_alloc_fn(GFP_KERNEL, q->node);
1270 goto out_bypass_end;
1274 spin_lock_irq(q->queue_lock);
1276 list_for_each_entry(blkg, &q->blkg_list, q_node) {
1277 struct blkg_policy_data *pd;
1279 if (blkg->pd[pol->plid])
1282 pd = pol->pd_alloc_fn(GFP_NOWAIT | __GFP_NOWARN, q->node);
1284 swap(pd, pd_prealloc);
1286 spin_unlock_irq(q->queue_lock);
1290 blkg->pd[pol->plid] = pd;
1292 pd->plid = pol->plid;
1293 if (pol->pd_init_fn)
1294 pol->pd_init_fn(pd);
1297 __set_bit(pol->plid, q->blkcg_pols);
1300 spin_unlock_irq(q->queue_lock);
1303 blk_mq_unfreeze_queue(q);
1305 blk_queue_bypass_end(q);
1307 pol->pd_free_fn(pd_prealloc);
1310 EXPORT_SYMBOL_GPL(blkcg_activate_policy);
1313 * blkcg_deactivate_policy - deactivate a blkcg policy on a request_queue
1314 * @q: request_queue of interest
1315 * @pol: blkcg policy to deactivate
1317 * Deactivate @pol on @q. Follows the same synchronization rules as
1318 * blkcg_activate_policy().
1320 void blkcg_deactivate_policy(struct request_queue *q,
1321 const struct blkcg_policy *pol)
1323 struct blkcg_gq *blkg;
1325 if (!blkcg_policy_enabled(q, pol))
1329 blk_mq_freeze_queue(q);
1331 blk_queue_bypass_start(q);
1333 spin_lock_irq(q->queue_lock);
1335 __clear_bit(pol->plid, q->blkcg_pols);
1337 list_for_each_entry(blkg, &q->blkg_list, q_node) {
1338 /* grab blkcg lock too while removing @pd from @blkg */
1339 spin_lock(&blkg->blkcg->lock);
1341 if (blkg->pd[pol->plid]) {
1342 if (pol->pd_offline_fn)
1343 pol->pd_offline_fn(blkg->pd[pol->plid]);
1344 pol->pd_free_fn(blkg->pd[pol->plid]);
1345 blkg->pd[pol->plid] = NULL;
1348 spin_unlock(&blkg->blkcg->lock);
1351 spin_unlock_irq(q->queue_lock);
1354 blk_mq_unfreeze_queue(q);
1356 blk_queue_bypass_end(q);
1358 EXPORT_SYMBOL_GPL(blkcg_deactivate_policy);
1361 * blkcg_policy_register - register a blkcg policy
1362 * @pol: blkcg policy to register
1364 * Register @pol with blkcg core. Might sleep and @pol may be modified on
1365 * successful registration. Returns 0 on success and -errno on failure.
1367 int blkcg_policy_register(struct blkcg_policy *pol)
1369 struct blkcg *blkcg;
1372 mutex_lock(&blkcg_pol_register_mutex);
1373 mutex_lock(&blkcg_pol_mutex);
1375 /* find an empty slot */
1377 for (i = 0; i < BLKCG_MAX_POLS; i++)
1378 if (!blkcg_policy[i])
1380 if (i >= BLKCG_MAX_POLS)
1385 blkcg_policy[pol->plid] = pol;
1387 /* allocate and install cpd's */
1388 if (pol->cpd_alloc_fn) {
1389 list_for_each_entry(blkcg, &all_blkcgs, all_blkcgs_node) {
1390 struct blkcg_policy_data *cpd;
1392 cpd = pol->cpd_alloc_fn(GFP_KERNEL);
1396 blkcg->cpd[pol->plid] = cpd;
1398 cpd->plid = pol->plid;
1399 pol->cpd_init_fn(cpd);
1403 mutex_unlock(&blkcg_pol_mutex);
1405 /* everything is in place, add intf files for the new policy */
1406 if (pol->dfl_cftypes)
1407 WARN_ON(cgroup_add_dfl_cftypes(&io_cgrp_subsys,
1409 if (pol->legacy_cftypes)
1410 WARN_ON(cgroup_add_legacy_cftypes(&io_cgrp_subsys,
1411 pol->legacy_cftypes));
1412 mutex_unlock(&blkcg_pol_register_mutex);
1416 if (pol->cpd_alloc_fn) {
1417 list_for_each_entry(blkcg, &all_blkcgs, all_blkcgs_node) {
1418 if (blkcg->cpd[pol->plid]) {
1419 pol->cpd_free_fn(blkcg->cpd[pol->plid]);
1420 blkcg->cpd[pol->plid] = NULL;
1424 blkcg_policy[pol->plid] = NULL;
1426 mutex_unlock(&blkcg_pol_mutex);
1427 mutex_unlock(&blkcg_pol_register_mutex);
1430 EXPORT_SYMBOL_GPL(blkcg_policy_register);
1433 * blkcg_policy_unregister - unregister a blkcg policy
1434 * @pol: blkcg policy to unregister
1436 * Undo blkcg_policy_register(@pol). Might sleep.
1438 void blkcg_policy_unregister(struct blkcg_policy *pol)
1440 struct blkcg *blkcg;
1442 mutex_lock(&blkcg_pol_register_mutex);
1444 if (WARN_ON(blkcg_policy[pol->plid] != pol))
1447 /* kill the intf files first */
1448 if (pol->dfl_cftypes)
1449 cgroup_rm_cftypes(pol->dfl_cftypes);
1450 if (pol->legacy_cftypes)
1451 cgroup_rm_cftypes(pol->legacy_cftypes);
1453 /* remove cpds and unregister */
1454 mutex_lock(&blkcg_pol_mutex);
1456 if (pol->cpd_alloc_fn) {
1457 list_for_each_entry(blkcg, &all_blkcgs, all_blkcgs_node) {
1458 if (blkcg->cpd[pol->plid]) {
1459 pol->cpd_free_fn(blkcg->cpd[pol->plid]);
1460 blkcg->cpd[pol->plid] = NULL;
1464 blkcg_policy[pol->plid] = NULL;
1466 mutex_unlock(&blkcg_pol_mutex);
1468 mutex_unlock(&blkcg_pol_register_mutex);
1470 EXPORT_SYMBOL_GPL(blkcg_policy_unregister);