2 * Common Block IO controller cgroup interface
4 * Based on ideas and code from CFQ, CFS and BFQ:
5 * Copyright (C) 2003 Jens Axboe <axboe@kernel.dk>
7 * Copyright (C) 2008 Fabio Checconi <fabio@gandalf.sssup.it>
8 * Paolo Valente <paolo.valente@unimore.it>
10 * Copyright (C) 2009 Vivek Goyal <vgoyal@redhat.com>
11 * Nauman Rafique <nauman@google.com>
13 * For policy-specific per-blkcg data:
14 * Copyright (C) 2015 Paolo Valente <paolo.valente@unimore.it>
15 * Arianna Avanzini <avanzini.arianna@gmail.com>
17 #include <linux/ioprio.h>
18 #include <linux/kdev_t.h>
19 #include <linux/module.h>
20 #include <linux/err.h>
21 #include <linux/blkdev.h>
22 #include <linux/backing-dev.h>
23 #include <linux/slab.h>
24 #include <linux/genhd.h>
25 #include <linux/delay.h>
26 #include <linux/atomic.h>
27 #include <linux/ctype.h>
28 #include <linux/blk-cgroup.h>
31 #define MAX_KEY_LEN 100
34 * blkcg_pol_mutex protects blkcg_policy[] and policy [de]activation.
35 * blkcg_pol_register_mutex nests outside of it and synchronizes entire
36 * policy [un]register operations including cgroup file additions /
37 * removals. Putting cgroup file registration outside blkcg_pol_mutex
38 * allows grabbing it from cgroup callbacks.
40 static DEFINE_MUTEX(blkcg_pol_register_mutex);
41 static DEFINE_MUTEX(blkcg_pol_mutex);
43 struct blkcg blkcg_root;
44 EXPORT_SYMBOL_GPL(blkcg_root);
46 struct cgroup_subsys_state * const blkcg_root_css = &blkcg_root.css;
48 static struct blkcg_policy *blkcg_policy[BLKCG_MAX_POLS];
50 static LIST_HEAD(all_blkcgs); /* protected by blkcg_pol_mutex */
52 static bool blkcg_policy_enabled(struct request_queue *q,
53 const struct blkcg_policy *pol)
55 return pol && test_bit(pol->plid, q->blkcg_pols);
59 * blkg_free - free a blkg
62 * Free @blkg which may be partially allocated.
64 static void blkg_free(struct blkcg_gq *blkg)
71 for (i = 0; i < BLKCG_MAX_POLS; i++)
73 blkcg_policy[i]->pd_free_fn(blkg->pd[i]);
75 if (blkg->blkcg != &blkcg_root)
76 blk_exit_rl(&blkg->rl);
78 blkg_rwstat_exit(&blkg->stat_ios);
79 blkg_rwstat_exit(&blkg->stat_bytes);
84 * blkg_alloc - allocate a blkg
85 * @blkcg: block cgroup the new blkg is associated with
86 * @q: request_queue the new blkg is associated with
87 * @gfp_mask: allocation mask to use
89 * Allocate a new blkg assocating @blkcg and @q.
91 static struct blkcg_gq *blkg_alloc(struct blkcg *blkcg, struct request_queue *q,
94 struct blkcg_gq *blkg;
97 /* alloc and init base part */
98 blkg = kzalloc_node(sizeof(*blkg), gfp_mask, q->node);
102 if (blkg_rwstat_init(&blkg->stat_bytes, gfp_mask) ||
103 blkg_rwstat_init(&blkg->stat_ios, gfp_mask))
107 INIT_LIST_HEAD(&blkg->q_node);
109 atomic_set(&blkg->refcnt, 1);
111 /* root blkg uses @q->root_rl, init rl only for !root blkgs */
112 if (blkcg != &blkcg_root) {
113 if (blk_init_rl(&blkg->rl, q, gfp_mask))
115 blkg->rl.blkg = blkg;
118 for (i = 0; i < BLKCG_MAX_POLS; i++) {
119 struct blkcg_policy *pol = blkcg_policy[i];
120 struct blkg_policy_data *pd;
122 if (!blkcg_policy_enabled(q, pol))
125 /* alloc per-policy data and attach it to blkg */
126 pd = pol->pd_alloc_fn(gfp_mask, q->node);
142 struct blkcg_gq *blkg_lookup_slowpath(struct blkcg *blkcg,
143 struct request_queue *q, bool update_hint)
145 struct blkcg_gq *blkg;
148 * Hint didn't match. Look up from the radix tree. Note that the
149 * hint can only be updated under queue_lock as otherwise @blkg
150 * could have already been removed from blkg_tree. The caller is
151 * responsible for grabbing queue_lock if @update_hint.
153 blkg = radix_tree_lookup(&blkcg->blkg_tree, q->id);
154 if (blkg && blkg->q == q) {
156 lockdep_assert_held(q->queue_lock);
157 rcu_assign_pointer(blkcg->blkg_hint, blkg);
164 EXPORT_SYMBOL_GPL(blkg_lookup_slowpath);
167 * If @new_blkg is %NULL, this function tries to allocate a new one as
168 * necessary using %GFP_NOWAIT. @new_blkg is always consumed on return.
170 static struct blkcg_gq *blkg_create(struct blkcg *blkcg,
171 struct request_queue *q,
172 struct blkcg_gq *new_blkg)
174 struct blkcg_gq *blkg;
175 struct bdi_writeback_congested *wb_congested;
178 WARN_ON_ONCE(!rcu_read_lock_held());
179 lockdep_assert_held(q->queue_lock);
181 /* blkg holds a reference to blkcg */
182 if (!css_tryget_online(&blkcg->css)) {
187 wb_congested = wb_congested_get_create(&q->backing_dev_info,
189 GFP_NOWAIT | __GFP_NOWARN);
197 new_blkg = blkg_alloc(blkcg, q, GFP_NOWAIT | __GFP_NOWARN);
198 if (unlikely(!new_blkg)) {
200 goto err_put_congested;
204 blkg->wb_congested = wb_congested;
207 if (blkcg_parent(blkcg)) {
208 blkg->parent = __blkg_lookup(blkcg_parent(blkcg), q, false);
209 if (WARN_ON_ONCE(!blkg->parent)) {
211 goto err_put_congested;
213 blkg_get(blkg->parent);
216 /* invoke per-policy init */
217 for (i = 0; i < BLKCG_MAX_POLS; i++) {
218 struct blkcg_policy *pol = blkcg_policy[i];
220 if (blkg->pd[i] && pol->pd_init_fn)
221 pol->pd_init_fn(blkg->pd[i]);
225 spin_lock(&blkcg->lock);
226 ret = radix_tree_insert(&blkcg->blkg_tree, q->id, blkg);
228 hlist_add_head_rcu(&blkg->blkcg_node, &blkcg->blkg_list);
229 list_add(&blkg->q_node, &q->blkg_list);
231 for (i = 0; i < BLKCG_MAX_POLS; i++) {
232 struct blkcg_policy *pol = blkcg_policy[i];
234 if (blkg->pd[i] && pol->pd_online_fn)
235 pol->pd_online_fn(blkg->pd[i]);
239 spin_unlock(&blkcg->lock);
244 /* @blkg failed fully initialized, use the usual release path */
249 wb_congested_put(wb_congested);
251 css_put(&blkcg->css);
258 * blkg_lookup_create - lookup blkg, try to create one if not there
259 * @blkcg: blkcg of interest
260 * @q: request_queue of interest
262 * Lookup blkg for the @blkcg - @q pair. If it doesn't exist, try to
263 * create one. blkg creation is performed recursively from blkcg_root such
264 * that all non-root blkg's have access to the parent blkg. This function
265 * should be called under RCU read lock and @q->queue_lock.
267 * Returns pointer to the looked up or created blkg on success, ERR_PTR()
268 * value on error. If @q is dead, returns ERR_PTR(-EINVAL). If @q is not
269 * dead and bypassing, returns ERR_PTR(-EBUSY).
271 struct blkcg_gq *blkg_lookup_create(struct blkcg *blkcg,
272 struct request_queue *q)
274 struct blkcg_gq *blkg;
276 WARN_ON_ONCE(!rcu_read_lock_held());
277 lockdep_assert_held(q->queue_lock);
280 * This could be the first entry point of blkcg implementation and
281 * we shouldn't allow anything to go through for a bypassing queue.
283 if (unlikely(blk_queue_bypass(q)))
284 return ERR_PTR(blk_queue_dying(q) ? -ENODEV : -EBUSY);
286 blkg = __blkg_lookup(blkcg, q, true);
291 * Create blkgs walking down from blkcg_root to @blkcg, so that all
292 * non-root blkgs have access to their parents.
295 struct blkcg *pos = blkcg;
296 struct blkcg *parent = blkcg_parent(blkcg);
298 while (parent && !__blkg_lookup(parent, q, false)) {
300 parent = blkcg_parent(parent);
303 blkg = blkg_create(pos, q, NULL);
304 if (pos == blkcg || IS_ERR(blkg))
309 static void blkg_destroy(struct blkcg_gq *blkg)
311 struct blkcg *blkcg = blkg->blkcg;
312 struct blkcg_gq *parent = blkg->parent;
315 lockdep_assert_held(blkg->q->queue_lock);
316 lockdep_assert_held(&blkcg->lock);
318 /* Something wrong if we are trying to remove same group twice */
319 WARN_ON_ONCE(list_empty(&blkg->q_node));
320 WARN_ON_ONCE(hlist_unhashed(&blkg->blkcg_node));
322 for (i = 0; i < BLKCG_MAX_POLS; i++) {
323 struct blkcg_policy *pol = blkcg_policy[i];
325 if (blkg->pd[i] && pol->pd_offline_fn)
326 pol->pd_offline_fn(blkg->pd[i]);
330 blkg_rwstat_add_aux(&parent->stat_bytes, &blkg->stat_bytes);
331 blkg_rwstat_add_aux(&parent->stat_ios, &blkg->stat_ios);
334 blkg->online = false;
336 radix_tree_delete(&blkcg->blkg_tree, blkg->q->id);
337 list_del_init(&blkg->q_node);
338 hlist_del_init_rcu(&blkg->blkcg_node);
341 * Both setting lookup hint to and clearing it from @blkg are done
342 * under queue_lock. If it's not pointing to @blkg now, it never
343 * will. Hint assignment itself can race safely.
345 if (rcu_access_pointer(blkcg->blkg_hint) == blkg)
346 rcu_assign_pointer(blkcg->blkg_hint, NULL);
349 * Put the reference taken at the time of creation so that when all
350 * queues are gone, group can be destroyed.
356 * blkg_destroy_all - destroy all blkgs associated with a request_queue
357 * @q: request_queue of interest
359 * Destroy all blkgs associated with @q.
361 static void blkg_destroy_all(struct request_queue *q)
363 struct blkcg_gq *blkg, *n;
365 lockdep_assert_held(q->queue_lock);
367 list_for_each_entry_safe(blkg, n, &q->blkg_list, q_node) {
368 struct blkcg *blkcg = blkg->blkcg;
370 spin_lock(&blkcg->lock);
372 spin_unlock(&blkcg->lock);
376 q->root_rl.blkg = NULL;
380 * A group is RCU protected, but having an rcu lock does not mean that one
381 * can access all the fields of blkg and assume these are valid. For
382 * example, don't try to follow throtl_data and request queue links.
384 * Having a reference to blkg under an rcu allows accesses to only values
385 * local to groups like group stats and group rate limits.
387 void __blkg_release_rcu(struct rcu_head *rcu_head)
389 struct blkcg_gq *blkg = container_of(rcu_head, struct blkcg_gq, rcu_head);
391 /* release the blkcg and parent blkg refs this blkg has been holding */
392 css_put(&blkg->blkcg->css);
394 blkg_put(blkg->parent);
396 wb_congested_put(blkg->wb_congested);
400 EXPORT_SYMBOL_GPL(__blkg_release_rcu);
403 * The next function used by blk_queue_for_each_rl(). It's a bit tricky
404 * because the root blkg uses @q->root_rl instead of its own rl.
406 struct request_list *__blk_queue_next_rl(struct request_list *rl,
407 struct request_queue *q)
409 struct list_head *ent;
410 struct blkcg_gq *blkg;
413 * Determine the current blkg list_head. The first entry is
414 * root_rl which is off @q->blkg_list and mapped to the head.
416 if (rl == &q->root_rl) {
418 /* There are no more block groups, hence no request lists */
422 blkg = container_of(rl, struct blkcg_gq, rl);
426 /* walk to the next list_head, skip root blkcg */
428 if (ent == &q->root_blkg->q_node)
430 if (ent == &q->blkg_list)
433 blkg = container_of(ent, struct blkcg_gq, q_node);
437 static int blkcg_reset_stats(struct cgroup_subsys_state *css,
438 struct cftype *cftype, u64 val)
440 struct blkcg *blkcg = css_to_blkcg(css);
441 struct blkcg_gq *blkg;
444 mutex_lock(&blkcg_pol_mutex);
445 spin_lock_irq(&blkcg->lock);
448 * Note that stat reset is racy - it doesn't synchronize against
449 * stat updates. This is a debug feature which shouldn't exist
450 * anyway. If you get hit by a race, retry.
452 hlist_for_each_entry(blkg, &blkcg->blkg_list, blkcg_node) {
453 blkg_rwstat_reset(&blkg->stat_bytes);
454 blkg_rwstat_reset(&blkg->stat_ios);
456 for (i = 0; i < BLKCG_MAX_POLS; i++) {
457 struct blkcg_policy *pol = blkcg_policy[i];
459 if (blkg->pd[i] && pol->pd_reset_stats_fn)
460 pol->pd_reset_stats_fn(blkg->pd[i]);
464 spin_unlock_irq(&blkcg->lock);
465 mutex_unlock(&blkcg_pol_mutex);
469 const char *blkg_dev_name(struct blkcg_gq *blkg)
471 /* some drivers (floppy) instantiate a queue w/o disk registered */
472 if (blkg->q->backing_dev_info.dev)
473 return dev_name(blkg->q->backing_dev_info.dev);
476 EXPORT_SYMBOL_GPL(blkg_dev_name);
479 * blkcg_print_blkgs - helper for printing per-blkg data
480 * @sf: seq_file to print to
481 * @blkcg: blkcg of interest
482 * @prfill: fill function to print out a blkg
483 * @pol: policy in question
484 * @data: data to be passed to @prfill
485 * @show_total: to print out sum of prfill return values or not
487 * This function invokes @prfill on each blkg of @blkcg if pd for the
488 * policy specified by @pol exists. @prfill is invoked with @sf, the
489 * policy data and @data and the matching queue lock held. If @show_total
490 * is %true, the sum of the return values from @prfill is printed with
491 * "Total" label at the end.
493 * This is to be used to construct print functions for
494 * cftype->read_seq_string method.
496 void blkcg_print_blkgs(struct seq_file *sf, struct blkcg *blkcg,
497 u64 (*prfill)(struct seq_file *,
498 struct blkg_policy_data *, int),
499 const struct blkcg_policy *pol, int data,
502 struct blkcg_gq *blkg;
506 hlist_for_each_entry_rcu(blkg, &blkcg->blkg_list, blkcg_node) {
507 spin_lock_irq(blkg->q->queue_lock);
508 if (blkcg_policy_enabled(blkg->q, pol))
509 total += prfill(sf, blkg->pd[pol->plid], data);
510 spin_unlock_irq(blkg->q->queue_lock);
515 seq_printf(sf, "Total %llu\n", (unsigned long long)total);
517 EXPORT_SYMBOL_GPL(blkcg_print_blkgs);
520 * __blkg_prfill_u64 - prfill helper for a single u64 value
521 * @sf: seq_file to print to
522 * @pd: policy private data of interest
525 * Print @v to @sf for the device assocaited with @pd.
527 u64 __blkg_prfill_u64(struct seq_file *sf, struct blkg_policy_data *pd, u64 v)
529 const char *dname = blkg_dev_name(pd->blkg);
534 seq_printf(sf, "%s %llu\n", dname, (unsigned long long)v);
537 EXPORT_SYMBOL_GPL(__blkg_prfill_u64);
540 * __blkg_prfill_rwstat - prfill helper for a blkg_rwstat
541 * @sf: seq_file to print to
542 * @pd: policy private data of interest
543 * @rwstat: rwstat to print
545 * Print @rwstat to @sf for the device assocaited with @pd.
547 u64 __blkg_prfill_rwstat(struct seq_file *sf, struct blkg_policy_data *pd,
548 const struct blkg_rwstat *rwstat)
550 static const char *rwstr[] = {
551 [BLKG_RWSTAT_READ] = "Read",
552 [BLKG_RWSTAT_WRITE] = "Write",
553 [BLKG_RWSTAT_SYNC] = "Sync",
554 [BLKG_RWSTAT_ASYNC] = "Async",
556 const char *dname = blkg_dev_name(pd->blkg);
563 for (i = 0; i < BLKG_RWSTAT_NR; i++)
564 seq_printf(sf, "%s %s %llu\n", dname, rwstr[i],
565 (unsigned long long)atomic64_read(&rwstat->aux_cnt[i]));
567 v = atomic64_read(&rwstat->aux_cnt[BLKG_RWSTAT_READ]) +
568 atomic64_read(&rwstat->aux_cnt[BLKG_RWSTAT_WRITE]);
569 seq_printf(sf, "%s Total %llu\n", dname, (unsigned long long)v);
572 EXPORT_SYMBOL_GPL(__blkg_prfill_rwstat);
575 * blkg_prfill_stat - prfill callback for blkg_stat
576 * @sf: seq_file to print to
577 * @pd: policy private data of interest
578 * @off: offset to the blkg_stat in @pd
580 * prfill callback for printing a blkg_stat.
582 u64 blkg_prfill_stat(struct seq_file *sf, struct blkg_policy_data *pd, int off)
584 return __blkg_prfill_u64(sf, pd, blkg_stat_read((void *)pd + off));
586 EXPORT_SYMBOL_GPL(blkg_prfill_stat);
589 * blkg_prfill_rwstat - prfill callback for blkg_rwstat
590 * @sf: seq_file to print to
591 * @pd: policy private data of interest
592 * @off: offset to the blkg_rwstat in @pd
594 * prfill callback for printing a blkg_rwstat.
596 u64 blkg_prfill_rwstat(struct seq_file *sf, struct blkg_policy_data *pd,
599 struct blkg_rwstat rwstat = blkg_rwstat_read((void *)pd + off);
601 return __blkg_prfill_rwstat(sf, pd, &rwstat);
603 EXPORT_SYMBOL_GPL(blkg_prfill_rwstat);
605 static u64 blkg_prfill_rwstat_field(struct seq_file *sf,
606 struct blkg_policy_data *pd, int off)
608 struct blkg_rwstat rwstat = blkg_rwstat_read((void *)pd->blkg + off);
610 return __blkg_prfill_rwstat(sf, pd, &rwstat);
614 * blkg_print_stat_bytes - seq_show callback for blkg->stat_bytes
615 * @sf: seq_file to print to
618 * To be used as cftype->seq_show to print blkg->stat_bytes.
619 * cftype->private must be set to the blkcg_policy.
621 int blkg_print_stat_bytes(struct seq_file *sf, void *v)
623 blkcg_print_blkgs(sf, css_to_blkcg(seq_css(sf)),
624 blkg_prfill_rwstat_field, (void *)seq_cft(sf)->private,
625 offsetof(struct blkcg_gq, stat_bytes), true);
628 EXPORT_SYMBOL_GPL(blkg_print_stat_bytes);
631 * blkg_print_stat_bytes - seq_show callback for blkg->stat_ios
632 * @sf: seq_file to print to
635 * To be used as cftype->seq_show to print blkg->stat_ios. cftype->private
636 * must be set to the blkcg_policy.
638 int blkg_print_stat_ios(struct seq_file *sf, void *v)
640 blkcg_print_blkgs(sf, css_to_blkcg(seq_css(sf)),
641 blkg_prfill_rwstat_field, (void *)seq_cft(sf)->private,
642 offsetof(struct blkcg_gq, stat_ios), true);
645 EXPORT_SYMBOL_GPL(blkg_print_stat_ios);
647 static u64 blkg_prfill_rwstat_field_recursive(struct seq_file *sf,
648 struct blkg_policy_data *pd,
651 struct blkg_rwstat rwstat = blkg_rwstat_recursive_sum(pd->blkg,
653 return __blkg_prfill_rwstat(sf, pd, &rwstat);
657 * blkg_print_stat_bytes_recursive - recursive version of blkg_print_stat_bytes
658 * @sf: seq_file to print to
661 int blkg_print_stat_bytes_recursive(struct seq_file *sf, void *v)
663 blkcg_print_blkgs(sf, css_to_blkcg(seq_css(sf)),
664 blkg_prfill_rwstat_field_recursive,
665 (void *)seq_cft(sf)->private,
666 offsetof(struct blkcg_gq, stat_bytes), true);
669 EXPORT_SYMBOL_GPL(blkg_print_stat_bytes_recursive);
672 * blkg_print_stat_ios_recursive - recursive version of blkg_print_stat_ios
673 * @sf: seq_file to print to
676 int blkg_print_stat_ios_recursive(struct seq_file *sf, void *v)
678 blkcg_print_blkgs(sf, css_to_blkcg(seq_css(sf)),
679 blkg_prfill_rwstat_field_recursive,
680 (void *)seq_cft(sf)->private,
681 offsetof(struct blkcg_gq, stat_ios), true);
684 EXPORT_SYMBOL_GPL(blkg_print_stat_ios_recursive);
687 * blkg_stat_recursive_sum - collect hierarchical blkg_stat
688 * @blkg: blkg of interest
689 * @pol: blkcg_policy which contains the blkg_stat
690 * @off: offset to the blkg_stat in blkg_policy_data or @blkg
692 * Collect the blkg_stat specified by @blkg, @pol and @off and all its
693 * online descendants and their aux counts. The caller must be holding the
694 * queue lock for online tests.
696 * If @pol is NULL, blkg_stat is at @off bytes into @blkg; otherwise, it is
697 * at @off bytes into @blkg's blkg_policy_data of the policy.
699 u64 blkg_stat_recursive_sum(struct blkcg_gq *blkg,
700 struct blkcg_policy *pol, int off)
702 struct blkcg_gq *pos_blkg;
703 struct cgroup_subsys_state *pos_css;
706 lockdep_assert_held(blkg->q->queue_lock);
709 blkg_for_each_descendant_pre(pos_blkg, pos_css, blkg) {
710 struct blkg_stat *stat;
712 if (!pos_blkg->online)
716 stat = (void *)blkg_to_pd(pos_blkg, pol) + off;
718 stat = (void *)blkg + off;
720 sum += blkg_stat_read(stat) + atomic64_read(&stat->aux_cnt);
726 EXPORT_SYMBOL_GPL(blkg_stat_recursive_sum);
729 * blkg_rwstat_recursive_sum - collect hierarchical blkg_rwstat
730 * @blkg: blkg of interest
731 * @pol: blkcg_policy which contains the blkg_rwstat
732 * @off: offset to the blkg_rwstat in blkg_policy_data or @blkg
734 * Collect the blkg_rwstat specified by @blkg, @pol and @off and all its
735 * online descendants and their aux counts. The caller must be holding the
736 * queue lock for online tests.
738 * If @pol is NULL, blkg_rwstat is at @off bytes into @blkg; otherwise, it
739 * is at @off bytes into @blkg's blkg_policy_data of the policy.
741 struct blkg_rwstat blkg_rwstat_recursive_sum(struct blkcg_gq *blkg,
742 struct blkcg_policy *pol, int off)
744 struct blkcg_gq *pos_blkg;
745 struct cgroup_subsys_state *pos_css;
746 struct blkg_rwstat sum = { };
749 lockdep_assert_held(blkg->q->queue_lock);
752 blkg_for_each_descendant_pre(pos_blkg, pos_css, blkg) {
753 struct blkg_rwstat *rwstat;
755 if (!pos_blkg->online)
759 rwstat = (void *)blkg_to_pd(pos_blkg, pol) + off;
761 rwstat = (void *)pos_blkg + off;
763 for (i = 0; i < BLKG_RWSTAT_NR; i++)
764 atomic64_add(atomic64_read(&rwstat->aux_cnt[i]) +
765 percpu_counter_sum_positive(&rwstat->cpu_cnt[i]),
772 EXPORT_SYMBOL_GPL(blkg_rwstat_recursive_sum);
775 * blkg_conf_prep - parse and prepare for per-blkg config update
776 * @blkcg: target block cgroup
777 * @pol: target policy
778 * @input: input string
779 * @ctx: blkg_conf_ctx to be filled
781 * Parse per-blkg config update from @input and initialize @ctx with the
782 * result. @ctx->blkg points to the blkg to be updated and @ctx->body the
783 * part of @input following MAJ:MIN. This function returns with RCU read
784 * lock and queue lock held and must be paired with blkg_conf_finish().
786 int blkg_conf_prep(struct blkcg *blkcg, const struct blkcg_policy *pol,
787 char *input, struct blkg_conf_ctx *ctx)
788 __acquires(rcu) __acquires(disk->queue->queue_lock)
790 struct gendisk *disk;
791 struct blkcg_gq *blkg;
792 struct module *owner;
793 unsigned int major, minor;
794 int key_len, part, ret;
797 if (sscanf(input, "%u:%u%n", &major, &minor, &key_len) != 2)
800 body = input + key_len;
803 body = skip_spaces(body);
805 disk = get_gendisk(MKDEV(major, minor), &part);
809 owner = disk->fops->owner;
816 spin_lock_irq(disk->queue->queue_lock);
818 if (blkcg_policy_enabled(disk->queue, pol))
819 blkg = blkg_lookup_create(blkcg, disk->queue);
821 blkg = ERR_PTR(-EOPNOTSUPP);
826 spin_unlock_irq(disk->queue->queue_lock);
827 owner = disk->fops->owner;
831 * If queue was bypassing, we should retry. Do so after a
832 * short msleep(). It isn't strictly necessary but queue
833 * can be bypassing for some time and it's always nice to
834 * avoid busy looping.
838 ret = restart_syscall();
848 EXPORT_SYMBOL_GPL(blkg_conf_prep);
851 * blkg_conf_finish - finish up per-blkg config update
852 * @ctx: blkg_conf_ctx intiailized by blkg_conf_prep()
854 * Finish up after per-blkg config update. This function must be paired
855 * with blkg_conf_prep().
857 void blkg_conf_finish(struct blkg_conf_ctx *ctx)
858 __releases(ctx->disk->queue->queue_lock) __releases(rcu)
860 struct module *owner;
862 spin_unlock_irq(ctx->disk->queue->queue_lock);
864 owner = ctx->disk->fops->owner;
868 EXPORT_SYMBOL_GPL(blkg_conf_finish);
870 static int blkcg_print_stat(struct seq_file *sf, void *v)
872 struct blkcg *blkcg = css_to_blkcg(seq_css(sf));
873 struct blkcg_gq *blkg;
877 hlist_for_each_entry_rcu(blkg, &blkcg->blkg_list, blkcg_node) {
879 struct blkg_rwstat rwstat;
880 u64 rbytes, wbytes, rios, wios;
882 dname = blkg_dev_name(blkg);
886 spin_lock_irq(blkg->q->queue_lock);
888 rwstat = blkg_rwstat_recursive_sum(blkg, NULL,
889 offsetof(struct blkcg_gq, stat_bytes));
890 rbytes = atomic64_read(&rwstat.aux_cnt[BLKG_RWSTAT_READ]);
891 wbytes = atomic64_read(&rwstat.aux_cnt[BLKG_RWSTAT_WRITE]);
893 rwstat = blkg_rwstat_recursive_sum(blkg, NULL,
894 offsetof(struct blkcg_gq, stat_ios));
895 rios = atomic64_read(&rwstat.aux_cnt[BLKG_RWSTAT_READ]);
896 wios = atomic64_read(&rwstat.aux_cnt[BLKG_RWSTAT_WRITE]);
898 spin_unlock_irq(blkg->q->queue_lock);
900 if (rbytes || wbytes || rios || wios)
901 seq_printf(sf, "%s rbytes=%llu wbytes=%llu rios=%llu wios=%llu\n",
902 dname, rbytes, wbytes, rios, wios);
909 static struct cftype blkcg_files[] = {
912 .flags = CFTYPE_NOT_ON_ROOT,
913 .seq_show = blkcg_print_stat,
918 static struct cftype blkcg_legacy_files[] = {
920 .name = "reset_stats",
921 .write_u64 = blkcg_reset_stats,
927 * blkcg_css_offline - cgroup css_offline callback
928 * @css: css of interest
930 * This function is called when @css is about to go away and responsible
931 * for shooting down all blkgs associated with @css. blkgs should be
932 * removed while holding both q and blkcg locks. As blkcg lock is nested
933 * inside q lock, this function performs reverse double lock dancing.
935 * This is the blkcg counterpart of ioc_release_fn().
937 static void blkcg_css_offline(struct cgroup_subsys_state *css)
939 struct blkcg *blkcg = css_to_blkcg(css);
941 spin_lock_irq(&blkcg->lock);
943 while (!hlist_empty(&blkcg->blkg_list)) {
944 struct blkcg_gq *blkg = hlist_entry(blkcg->blkg_list.first,
945 struct blkcg_gq, blkcg_node);
946 struct request_queue *q = blkg->q;
948 if (spin_trylock(q->queue_lock)) {
950 spin_unlock(q->queue_lock);
952 spin_unlock_irq(&blkcg->lock);
954 spin_lock_irq(&blkcg->lock);
958 spin_unlock_irq(&blkcg->lock);
960 wb_blkcg_offline(blkcg);
963 static void blkcg_css_free(struct cgroup_subsys_state *css)
965 struct blkcg *blkcg = css_to_blkcg(css);
968 mutex_lock(&blkcg_pol_mutex);
970 list_del(&blkcg->all_blkcgs_node);
972 for (i = 0; i < BLKCG_MAX_POLS; i++)
974 blkcg_policy[i]->cpd_free_fn(blkcg->cpd[i]);
976 mutex_unlock(&blkcg_pol_mutex);
981 static struct cgroup_subsys_state *
982 blkcg_css_alloc(struct cgroup_subsys_state *parent_css)
985 struct cgroup_subsys_state *ret;
988 mutex_lock(&blkcg_pol_mutex);
993 blkcg = kzalloc(sizeof(*blkcg), GFP_KERNEL);
995 ret = ERR_PTR(-ENOMEM);
1000 for (i = 0; i < BLKCG_MAX_POLS ; i++) {
1001 struct blkcg_policy *pol = blkcg_policy[i];
1002 struct blkcg_policy_data *cpd;
1005 * If the policy hasn't been attached yet, wait for it
1006 * to be attached before doing anything else. Otherwise,
1007 * check if the policy requires any specific per-cgroup
1008 * data: if it does, allocate and initialize it.
1010 if (!pol || !pol->cpd_alloc_fn)
1013 cpd = pol->cpd_alloc_fn(GFP_KERNEL);
1015 ret = ERR_PTR(-ENOMEM);
1018 blkcg->cpd[i] = cpd;
1021 if (pol->cpd_init_fn)
1022 pol->cpd_init_fn(cpd);
1025 spin_lock_init(&blkcg->lock);
1026 INIT_RADIX_TREE(&blkcg->blkg_tree, GFP_NOWAIT | __GFP_NOWARN);
1027 INIT_HLIST_HEAD(&blkcg->blkg_list);
1028 #ifdef CONFIG_CGROUP_WRITEBACK
1029 INIT_LIST_HEAD(&blkcg->cgwb_list);
1031 list_add_tail(&blkcg->all_blkcgs_node, &all_blkcgs);
1033 mutex_unlock(&blkcg_pol_mutex);
1037 for (i--; i >= 0; i--)
1039 blkcg_policy[i]->cpd_free_fn(blkcg->cpd[i]);
1042 mutex_unlock(&blkcg_pol_mutex);
1047 * blkcg_init_queue - initialize blkcg part of request queue
1048 * @q: request_queue to initialize
1050 * Called from blk_alloc_queue_node(). Responsible for initializing blkcg
1051 * part of new request_queue @q.
1054 * 0 on success, -errno on failure.
1056 int blkcg_init_queue(struct request_queue *q)
1058 struct blkcg_gq *new_blkg, *blkg;
1062 new_blkg = blkg_alloc(&blkcg_root, q, GFP_KERNEL);
1066 preloaded = !radix_tree_preload(GFP_KERNEL);
1069 * Make sure the root blkg exists and count the existing blkgs. As
1070 * @q is bypassing at this point, blkg_lookup_create() can't be
1071 * used. Open code insertion.
1074 spin_lock_irq(q->queue_lock);
1075 blkg = blkg_create(&blkcg_root, q, new_blkg);
1076 spin_unlock_irq(q->queue_lock);
1080 radix_tree_preload_end();
1083 blkg_free(new_blkg);
1084 return PTR_ERR(blkg);
1087 q->root_blkg = blkg;
1088 q->root_rl.blkg = blkg;
1090 ret = blk_throtl_init(q);
1092 spin_lock_irq(q->queue_lock);
1093 blkg_destroy_all(q);
1094 spin_unlock_irq(q->queue_lock);
1100 * blkcg_drain_queue - drain blkcg part of request_queue
1101 * @q: request_queue to drain
1103 * Called from blk_drain_queue(). Responsible for draining blkcg part.
1105 void blkcg_drain_queue(struct request_queue *q)
1107 lockdep_assert_held(q->queue_lock);
1110 * @q could be exiting and already have destroyed all blkgs as
1111 * indicated by NULL root_blkg. If so, don't confuse policies.
1116 blk_throtl_drain(q);
1120 * blkcg_exit_queue - exit and release blkcg part of request_queue
1121 * @q: request_queue being released
1123 * Called from blk_release_queue(). Responsible for exiting blkcg part.
1125 void blkcg_exit_queue(struct request_queue *q)
1127 spin_lock_irq(q->queue_lock);
1128 blkg_destroy_all(q);
1129 spin_unlock_irq(q->queue_lock);
1135 * We cannot support shared io contexts, as we have no mean to support
1136 * two tasks with the same ioc in two different groups without major rework
1137 * of the main cic data structures. For now we allow a task to change
1138 * its cgroup only if it's the only owner of its ioc.
1140 static int blkcg_can_attach(struct cgroup_taskset *tset)
1142 struct task_struct *task;
1143 struct cgroup_subsys_state *dst_css;
1144 struct io_context *ioc;
1147 /* task_lock() is needed to avoid races with exit_io_context() */
1148 cgroup_taskset_for_each(task, dst_css, tset) {
1150 ioc = task->io_context;
1151 if (ioc && atomic_read(&ioc->nr_tasks) > 1)
1160 static void blkcg_bind(struct cgroup_subsys_state *root_css)
1164 mutex_lock(&blkcg_pol_mutex);
1166 for (i = 0; i < BLKCG_MAX_POLS; i++) {
1167 struct blkcg_policy *pol = blkcg_policy[i];
1168 struct blkcg *blkcg;
1170 if (!pol || !pol->cpd_bind_fn)
1173 list_for_each_entry(blkcg, &all_blkcgs, all_blkcgs_node)
1174 if (blkcg->cpd[pol->plid])
1175 pol->cpd_bind_fn(blkcg->cpd[pol->plid]);
1177 mutex_unlock(&blkcg_pol_mutex);
1180 struct cgroup_subsys io_cgrp_subsys = {
1181 .css_alloc = blkcg_css_alloc,
1182 .css_offline = blkcg_css_offline,
1183 .css_free = blkcg_css_free,
1184 .can_attach = blkcg_can_attach,
1186 .dfl_cftypes = blkcg_files,
1187 .legacy_cftypes = blkcg_legacy_files,
1188 .legacy_name = "blkio",
1191 * This ensures that, if available, memcg is automatically enabled
1192 * together on the default hierarchy so that the owner cgroup can
1193 * be retrieved from writeback pages.
1195 .depends_on = 1 << memory_cgrp_id,
1198 EXPORT_SYMBOL_GPL(io_cgrp_subsys);
1201 * blkcg_activate_policy - activate a blkcg policy on a request_queue
1202 * @q: request_queue of interest
1203 * @pol: blkcg policy to activate
1205 * Activate @pol on @q. Requires %GFP_KERNEL context. @q goes through
1206 * bypass mode to populate its blkgs with policy_data for @pol.
1208 * Activation happens with @q bypassed, so nobody would be accessing blkgs
1209 * from IO path. Update of each blkg is protected by both queue and blkcg
1210 * locks so that holding either lock and testing blkcg_policy_enabled() is
1211 * always enough for dereferencing policy data.
1213 * The caller is responsible for synchronizing [de]activations and policy
1214 * [un]registerations. Returns 0 on success, -errno on failure.
1216 int blkcg_activate_policy(struct request_queue *q,
1217 const struct blkcg_policy *pol)
1219 struct blkg_policy_data *pd_prealloc = NULL;
1220 struct blkcg_gq *blkg;
1223 if (blkcg_policy_enabled(q, pol))
1226 blk_queue_bypass_start(q);
1229 pd_prealloc = pol->pd_alloc_fn(GFP_KERNEL, q->node);
1232 goto out_bypass_end;
1236 spin_lock_irq(q->queue_lock);
1238 list_for_each_entry(blkg, &q->blkg_list, q_node) {
1239 struct blkg_policy_data *pd;
1241 if (blkg->pd[pol->plid])
1244 pd = pol->pd_alloc_fn(GFP_NOWAIT | __GFP_NOWARN, q->node);
1246 swap(pd, pd_prealloc);
1248 spin_unlock_irq(q->queue_lock);
1252 blkg->pd[pol->plid] = pd;
1254 pd->plid = pol->plid;
1255 if (pol->pd_init_fn)
1256 pol->pd_init_fn(pd);
1259 __set_bit(pol->plid, q->blkcg_pols);
1262 spin_unlock_irq(q->queue_lock);
1264 blk_queue_bypass_end(q);
1266 pol->pd_free_fn(pd_prealloc);
1269 EXPORT_SYMBOL_GPL(blkcg_activate_policy);
1272 * blkcg_deactivate_policy - deactivate a blkcg policy on a request_queue
1273 * @q: request_queue of interest
1274 * @pol: blkcg policy to deactivate
1276 * Deactivate @pol on @q. Follows the same synchronization rules as
1277 * blkcg_activate_policy().
1279 void blkcg_deactivate_policy(struct request_queue *q,
1280 const struct blkcg_policy *pol)
1282 struct blkcg_gq *blkg;
1284 if (!blkcg_policy_enabled(q, pol))
1287 blk_queue_bypass_start(q);
1288 spin_lock_irq(q->queue_lock);
1290 __clear_bit(pol->plid, q->blkcg_pols);
1292 list_for_each_entry(blkg, &q->blkg_list, q_node) {
1293 /* grab blkcg lock too while removing @pd from @blkg */
1294 spin_lock(&blkg->blkcg->lock);
1296 if (blkg->pd[pol->plid]) {
1297 if (pol->pd_offline_fn)
1298 pol->pd_offline_fn(blkg->pd[pol->plid]);
1299 pol->pd_free_fn(blkg->pd[pol->plid]);
1300 blkg->pd[pol->plid] = NULL;
1303 spin_unlock(&blkg->blkcg->lock);
1306 spin_unlock_irq(q->queue_lock);
1307 blk_queue_bypass_end(q);
1309 EXPORT_SYMBOL_GPL(blkcg_deactivate_policy);
1312 * blkcg_policy_register - register a blkcg policy
1313 * @pol: blkcg policy to register
1315 * Register @pol with blkcg core. Might sleep and @pol may be modified on
1316 * successful registration. Returns 0 on success and -errno on failure.
1318 int blkcg_policy_register(struct blkcg_policy *pol)
1320 struct blkcg *blkcg;
1323 mutex_lock(&blkcg_pol_register_mutex);
1324 mutex_lock(&blkcg_pol_mutex);
1326 /* find an empty slot */
1328 for (i = 0; i < BLKCG_MAX_POLS; i++)
1329 if (!blkcg_policy[i])
1331 if (i >= BLKCG_MAX_POLS)
1336 blkcg_policy[pol->plid] = pol;
1338 /* allocate and install cpd's */
1339 if (pol->cpd_alloc_fn) {
1340 list_for_each_entry(blkcg, &all_blkcgs, all_blkcgs_node) {
1341 struct blkcg_policy_data *cpd;
1343 cpd = pol->cpd_alloc_fn(GFP_KERNEL);
1347 blkcg->cpd[pol->plid] = cpd;
1349 cpd->plid = pol->plid;
1350 pol->cpd_init_fn(cpd);
1354 mutex_unlock(&blkcg_pol_mutex);
1356 /* everything is in place, add intf files for the new policy */
1357 if (pol->dfl_cftypes)
1358 WARN_ON(cgroup_add_dfl_cftypes(&io_cgrp_subsys,
1360 if (pol->legacy_cftypes)
1361 WARN_ON(cgroup_add_legacy_cftypes(&io_cgrp_subsys,
1362 pol->legacy_cftypes));
1363 mutex_unlock(&blkcg_pol_register_mutex);
1367 if (pol->cpd_alloc_fn) {
1368 list_for_each_entry(blkcg, &all_blkcgs, all_blkcgs_node) {
1369 if (blkcg->cpd[pol->plid]) {
1370 pol->cpd_free_fn(blkcg->cpd[pol->plid]);
1371 blkcg->cpd[pol->plid] = NULL;
1375 blkcg_policy[pol->plid] = NULL;
1377 mutex_unlock(&blkcg_pol_mutex);
1378 mutex_unlock(&blkcg_pol_register_mutex);
1381 EXPORT_SYMBOL_GPL(blkcg_policy_register);
1384 * blkcg_policy_unregister - unregister a blkcg policy
1385 * @pol: blkcg policy to unregister
1387 * Undo blkcg_policy_register(@pol). Might sleep.
1389 void blkcg_policy_unregister(struct blkcg_policy *pol)
1391 struct blkcg *blkcg;
1393 mutex_lock(&blkcg_pol_register_mutex);
1395 if (WARN_ON(blkcg_policy[pol->plid] != pol))
1398 /* kill the intf files first */
1399 if (pol->dfl_cftypes)
1400 cgroup_rm_cftypes(pol->dfl_cftypes);
1401 if (pol->legacy_cftypes)
1402 cgroup_rm_cftypes(pol->legacy_cftypes);
1404 /* remove cpds and unregister */
1405 mutex_lock(&blkcg_pol_mutex);
1407 if (pol->cpd_alloc_fn) {
1408 list_for_each_entry(blkcg, &all_blkcgs, all_blkcgs_node) {
1409 if (blkcg->cpd[pol->plid]) {
1410 pol->cpd_free_fn(blkcg->cpd[pol->plid]);
1411 blkcg->cpd[pol->plid] = NULL;
1415 blkcg_policy[pol->plid] = NULL;
1417 mutex_unlock(&blkcg_pol_mutex);
1419 mutex_unlock(&blkcg_pol_register_mutex);
1421 EXPORT_SYMBOL_GPL(blkcg_policy_unregister);