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/blk-cgroup.h>
30 #define MAX_KEY_LEN 100
33 * blkcg_pol_mutex protects blkcg_policy[] and policy [de]activation.
34 * blkcg_pol_register_mutex nests outside of it and synchronizes entire
35 * policy [un]register operations including cgroup file additions /
36 * removals. Putting cgroup file registration outside blkcg_pol_mutex
37 * allows grabbing it from cgroup callbacks.
39 static DEFINE_MUTEX(blkcg_pol_register_mutex);
40 static DEFINE_MUTEX(blkcg_pol_mutex);
42 struct blkcg blkcg_root;
43 EXPORT_SYMBOL_GPL(blkcg_root);
45 struct cgroup_subsys_state * const blkcg_root_css = &blkcg_root.css;
47 static struct blkcg_policy *blkcg_policy[BLKCG_MAX_POLS];
49 static LIST_HEAD(all_blkcgs); /* protected by blkcg_pol_mutex */
51 static bool blkcg_policy_enabled(struct request_queue *q,
52 const struct blkcg_policy *pol)
54 return pol && test_bit(pol->plid, q->blkcg_pols);
58 * blkg_free - free a blkg
61 * Free @blkg which may be partially allocated.
63 static void blkg_free(struct blkcg_gq *blkg)
70 for (i = 0; i < BLKCG_MAX_POLS; i++)
73 blk_exit_rl(&blkg->rl);
78 * blkg_alloc - allocate a blkg
79 * @blkcg: block cgroup the new blkg is associated with
80 * @q: request_queue the new blkg is associated with
81 * @gfp_mask: allocation mask to use
83 * Allocate a new blkg assocating @blkcg and @q.
85 static struct blkcg_gq *blkg_alloc(struct blkcg *blkcg, struct request_queue *q,
88 struct blkcg_gq *blkg;
91 /* alloc and init base part */
92 blkg = kzalloc_node(sizeof(*blkg), gfp_mask, q->node);
97 INIT_LIST_HEAD(&blkg->q_node);
99 atomic_set(&blkg->refcnt, 1);
101 /* root blkg uses @q->root_rl, init rl only for !root blkgs */
102 if (blkcg != &blkcg_root) {
103 if (blk_init_rl(&blkg->rl, q, gfp_mask))
105 blkg->rl.blkg = blkg;
108 for (i = 0; i < BLKCG_MAX_POLS; i++) {
109 struct blkcg_policy *pol = blkcg_policy[i];
110 struct blkg_policy_data *pd;
112 if (!blkcg_policy_enabled(q, pol))
115 /* alloc per-policy data and attach it to blkg */
116 pd = kzalloc_node(pol->pd_size, gfp_mask, q->node);
133 * __blkg_lookup - internal version of blkg_lookup()
134 * @blkcg: blkcg of interest
135 * @q: request_queue of interest
136 * @update_hint: whether to update lookup hint with the result or not
138 * This is internal version and shouldn't be used by policy
139 * implementations. Looks up blkgs for the @blkcg - @q pair regardless of
140 * @q's bypass state. If @update_hint is %true, the caller should be
141 * holding @q->queue_lock and lookup hint is updated on success.
143 struct blkcg_gq *__blkg_lookup(struct blkcg *blkcg, struct request_queue *q,
146 struct blkcg_gq *blkg;
148 blkg = rcu_dereference(blkcg->blkg_hint);
149 if (blkg && blkg->q == q)
153 * Hint didn't match. Look up from the radix tree. Note that the
154 * hint can only be updated under queue_lock as otherwise @blkg
155 * could have already been removed from blkg_tree. The caller is
156 * responsible for grabbing queue_lock if @update_hint.
158 blkg = radix_tree_lookup(&blkcg->blkg_tree, q->id);
159 if (blkg && blkg->q == q) {
161 lockdep_assert_held(q->queue_lock);
162 rcu_assign_pointer(blkcg->blkg_hint, blkg);
171 * blkg_lookup - lookup blkg for the specified blkcg - q pair
172 * @blkcg: blkcg of interest
173 * @q: request_queue of interest
175 * Lookup blkg for the @blkcg - @q pair. This function should be called
176 * under RCU read lock and is guaranteed to return %NULL if @q is bypassing
177 * - see blk_queue_bypass_start() for details.
179 struct blkcg_gq *blkg_lookup(struct blkcg *blkcg, struct request_queue *q)
181 WARN_ON_ONCE(!rcu_read_lock_held());
183 if (unlikely(blk_queue_bypass(q)))
185 return __blkg_lookup(blkcg, q, false);
187 EXPORT_SYMBOL_GPL(blkg_lookup);
190 * If @new_blkg is %NULL, this function tries to allocate a new one as
191 * necessary using %GFP_ATOMIC. @new_blkg is always consumed on return.
193 static struct blkcg_gq *blkg_create(struct blkcg *blkcg,
194 struct request_queue *q,
195 struct blkcg_gq *new_blkg)
197 struct blkcg_gq *blkg;
198 struct bdi_writeback_congested *wb_congested;
201 WARN_ON_ONCE(!rcu_read_lock_held());
202 lockdep_assert_held(q->queue_lock);
204 /* blkg holds a reference to blkcg */
205 if (!css_tryget_online(&blkcg->css)) {
210 wb_congested = wb_congested_get_create(&q->backing_dev_info,
211 blkcg->css.id, GFP_ATOMIC);
219 new_blkg = blkg_alloc(blkcg, q, GFP_ATOMIC);
220 if (unlikely(!new_blkg)) {
222 goto err_put_congested;
226 blkg->wb_congested = wb_congested;
229 if (blkcg_parent(blkcg)) {
230 blkg->parent = __blkg_lookup(blkcg_parent(blkcg), q, false);
231 if (WARN_ON_ONCE(!blkg->parent)) {
233 goto err_put_congested;
235 blkg_get(blkg->parent);
238 /* invoke per-policy init */
239 for (i = 0; i < BLKCG_MAX_POLS; i++) {
240 struct blkcg_policy *pol = blkcg_policy[i];
242 if (blkg->pd[i] && pol->pd_init_fn)
243 pol->pd_init_fn(blkg);
247 spin_lock(&blkcg->lock);
248 ret = radix_tree_insert(&blkcg->blkg_tree, q->id, blkg);
250 hlist_add_head_rcu(&blkg->blkcg_node, &blkcg->blkg_list);
251 list_add(&blkg->q_node, &q->blkg_list);
253 for (i = 0; i < BLKCG_MAX_POLS; i++) {
254 struct blkcg_policy *pol = blkcg_policy[i];
256 if (blkg->pd[i] && pol->pd_online_fn)
257 pol->pd_online_fn(blkg);
261 spin_unlock(&blkcg->lock);
266 /* @blkg failed fully initialized, use the usual release path */
271 wb_congested_put(wb_congested);
273 css_put(&blkcg->css);
280 * blkg_lookup_create - lookup blkg, try to create one if not there
281 * @blkcg: blkcg of interest
282 * @q: request_queue of interest
284 * Lookup blkg for the @blkcg - @q pair. If it doesn't exist, try to
285 * create one. blkg creation is performed recursively from blkcg_root such
286 * that all non-root blkg's have access to the parent blkg. This function
287 * should be called under RCU read lock and @q->queue_lock.
289 * Returns pointer to the looked up or created blkg on success, ERR_PTR()
290 * value on error. If @q is dead, returns ERR_PTR(-EINVAL). If @q is not
291 * dead and bypassing, returns ERR_PTR(-EBUSY).
293 struct blkcg_gq *blkg_lookup_create(struct blkcg *blkcg,
294 struct request_queue *q)
296 struct blkcg_gq *blkg;
298 WARN_ON_ONCE(!rcu_read_lock_held());
299 lockdep_assert_held(q->queue_lock);
302 * This could be the first entry point of blkcg implementation and
303 * we shouldn't allow anything to go through for a bypassing queue.
305 if (unlikely(blk_queue_bypass(q)))
306 return ERR_PTR(blk_queue_dying(q) ? -EINVAL : -EBUSY);
308 blkg = __blkg_lookup(blkcg, q, true);
313 * Create blkgs walking down from blkcg_root to @blkcg, so that all
314 * non-root blkgs have access to their parents.
317 struct blkcg *pos = blkcg;
318 struct blkcg *parent = blkcg_parent(blkcg);
320 while (parent && !__blkg_lookup(parent, q, false)) {
322 parent = blkcg_parent(parent);
325 blkg = blkg_create(pos, q, NULL);
326 if (pos == blkcg || IS_ERR(blkg))
330 EXPORT_SYMBOL_GPL(blkg_lookup_create);
332 static void blkg_destroy(struct blkcg_gq *blkg)
334 struct blkcg *blkcg = blkg->blkcg;
337 lockdep_assert_held(blkg->q->queue_lock);
338 lockdep_assert_held(&blkcg->lock);
340 /* Something wrong if we are trying to remove same group twice */
341 WARN_ON_ONCE(list_empty(&blkg->q_node));
342 WARN_ON_ONCE(hlist_unhashed(&blkg->blkcg_node));
344 for (i = 0; i < BLKCG_MAX_POLS; i++) {
345 struct blkcg_policy *pol = blkcg_policy[i];
347 if (blkg->pd[i] && pol->pd_offline_fn)
348 pol->pd_offline_fn(blkg);
350 blkg->online = false;
352 radix_tree_delete(&blkcg->blkg_tree, blkg->q->id);
353 list_del_init(&blkg->q_node);
354 hlist_del_init_rcu(&blkg->blkcg_node);
357 * Both setting lookup hint to and clearing it from @blkg are done
358 * under queue_lock. If it's not pointing to @blkg now, it never
359 * will. Hint assignment itself can race safely.
361 if (rcu_access_pointer(blkcg->blkg_hint) == blkg)
362 rcu_assign_pointer(blkcg->blkg_hint, NULL);
365 * Put the reference taken at the time of creation so that when all
366 * queues are gone, group can be destroyed.
372 * blkg_destroy_all - destroy all blkgs associated with a request_queue
373 * @q: request_queue of interest
375 * Destroy all blkgs associated with @q.
377 static void blkg_destroy_all(struct request_queue *q)
379 struct blkcg_gq *blkg, *n;
381 lockdep_assert_held(q->queue_lock);
383 list_for_each_entry_safe(blkg, n, &q->blkg_list, q_node) {
384 struct blkcg *blkcg = blkg->blkcg;
386 spin_lock(&blkcg->lock);
388 spin_unlock(&blkcg->lock);
392 q->root_rl.blkg = NULL;
396 * A group is RCU protected, but having an rcu lock does not mean that one
397 * can access all the fields of blkg and assume these are valid. For
398 * example, don't try to follow throtl_data and request queue links.
400 * Having a reference to blkg under an rcu allows accesses to only values
401 * local to groups like group stats and group rate limits.
403 void __blkg_release_rcu(struct rcu_head *rcu_head)
405 struct blkcg_gq *blkg = container_of(rcu_head, struct blkcg_gq, rcu_head);
408 /* tell policies that this one is being freed */
409 for (i = 0; i < BLKCG_MAX_POLS; i++) {
410 struct blkcg_policy *pol = blkcg_policy[i];
412 if (blkg->pd[i] && pol->pd_exit_fn)
413 pol->pd_exit_fn(blkg);
416 /* release the blkcg and parent blkg refs this blkg has been holding */
417 css_put(&blkg->blkcg->css);
419 blkg_put(blkg->parent);
421 wb_congested_put(blkg->wb_congested);
425 EXPORT_SYMBOL_GPL(__blkg_release_rcu);
428 * The next function used by blk_queue_for_each_rl(). It's a bit tricky
429 * because the root blkg uses @q->root_rl instead of its own rl.
431 struct request_list *__blk_queue_next_rl(struct request_list *rl,
432 struct request_queue *q)
434 struct list_head *ent;
435 struct blkcg_gq *blkg;
438 * Determine the current blkg list_head. The first entry is
439 * root_rl which is off @q->blkg_list and mapped to the head.
441 if (rl == &q->root_rl) {
443 /* There are no more block groups, hence no request lists */
447 blkg = container_of(rl, struct blkcg_gq, rl);
451 /* walk to the next list_head, skip root blkcg */
453 if (ent == &q->root_blkg->q_node)
455 if (ent == &q->blkg_list)
458 blkg = container_of(ent, struct blkcg_gq, q_node);
462 static int blkcg_reset_stats(struct cgroup_subsys_state *css,
463 struct cftype *cftype, u64 val)
465 struct blkcg *blkcg = css_to_blkcg(css);
466 struct blkcg_gq *blkg;
469 mutex_lock(&blkcg_pol_mutex);
470 spin_lock_irq(&blkcg->lock);
473 * Note that stat reset is racy - it doesn't synchronize against
474 * stat updates. This is a debug feature which shouldn't exist
475 * anyway. If you get hit by a race, retry.
477 hlist_for_each_entry(blkg, &blkcg->blkg_list, blkcg_node) {
478 for (i = 0; i < BLKCG_MAX_POLS; i++) {
479 struct blkcg_policy *pol = blkcg_policy[i];
481 if (blkcg_policy_enabled(blkg->q, pol) &&
482 pol->pd_reset_stats_fn)
483 pol->pd_reset_stats_fn(blkg);
487 spin_unlock_irq(&blkcg->lock);
488 mutex_unlock(&blkcg_pol_mutex);
492 static const char *blkg_dev_name(struct blkcg_gq *blkg)
494 /* some drivers (floppy) instantiate a queue w/o disk registered */
495 if (blkg->q->backing_dev_info.dev)
496 return dev_name(blkg->q->backing_dev_info.dev);
501 * blkcg_print_blkgs - helper for printing per-blkg data
502 * @sf: seq_file to print to
503 * @blkcg: blkcg of interest
504 * @prfill: fill function to print out a blkg
505 * @pol: policy in question
506 * @data: data to be passed to @prfill
507 * @show_total: to print out sum of prfill return values or not
509 * This function invokes @prfill on each blkg of @blkcg if pd for the
510 * policy specified by @pol exists. @prfill is invoked with @sf, the
511 * policy data and @data and the matching queue lock held. If @show_total
512 * is %true, the sum of the return values from @prfill is printed with
513 * "Total" label at the end.
515 * This is to be used to construct print functions for
516 * cftype->read_seq_string method.
518 void blkcg_print_blkgs(struct seq_file *sf, struct blkcg *blkcg,
519 u64 (*prfill)(struct seq_file *,
520 struct blkg_policy_data *, int),
521 const struct blkcg_policy *pol, int data,
524 struct blkcg_gq *blkg;
528 hlist_for_each_entry_rcu(blkg, &blkcg->blkg_list, blkcg_node) {
529 spin_lock_irq(blkg->q->queue_lock);
530 if (blkcg_policy_enabled(blkg->q, pol))
531 total += prfill(sf, blkg->pd[pol->plid], data);
532 spin_unlock_irq(blkg->q->queue_lock);
537 seq_printf(sf, "Total %llu\n", (unsigned long long)total);
539 EXPORT_SYMBOL_GPL(blkcg_print_blkgs);
542 * __blkg_prfill_u64 - prfill helper for a single u64 value
543 * @sf: seq_file to print to
544 * @pd: policy private data of interest
547 * Print @v to @sf for the device assocaited with @pd.
549 u64 __blkg_prfill_u64(struct seq_file *sf, struct blkg_policy_data *pd, u64 v)
551 const char *dname = blkg_dev_name(pd->blkg);
556 seq_printf(sf, "%s %llu\n", dname, (unsigned long long)v);
559 EXPORT_SYMBOL_GPL(__blkg_prfill_u64);
562 * __blkg_prfill_rwstat - prfill helper for a blkg_rwstat
563 * @sf: seq_file to print to
564 * @pd: policy private data of interest
565 * @rwstat: rwstat to print
567 * Print @rwstat to @sf for the device assocaited with @pd.
569 u64 __blkg_prfill_rwstat(struct seq_file *sf, struct blkg_policy_data *pd,
570 const struct blkg_rwstat *rwstat)
572 static const char *rwstr[] = {
573 [BLKG_RWSTAT_READ] = "Read",
574 [BLKG_RWSTAT_WRITE] = "Write",
575 [BLKG_RWSTAT_SYNC] = "Sync",
576 [BLKG_RWSTAT_ASYNC] = "Async",
578 const char *dname = blkg_dev_name(pd->blkg);
585 for (i = 0; i < BLKG_RWSTAT_NR; i++)
586 seq_printf(sf, "%s %s %llu\n", dname, rwstr[i],
587 (unsigned long long)rwstat->cnt[i]);
589 v = rwstat->cnt[BLKG_RWSTAT_READ] + rwstat->cnt[BLKG_RWSTAT_WRITE];
590 seq_printf(sf, "%s Total %llu\n", dname, (unsigned long long)v);
593 EXPORT_SYMBOL_GPL(__blkg_prfill_rwstat);
596 * blkg_prfill_stat - prfill callback for blkg_stat
597 * @sf: seq_file to print to
598 * @pd: policy private data of interest
599 * @off: offset to the blkg_stat in @pd
601 * prfill callback for printing a blkg_stat.
603 u64 blkg_prfill_stat(struct seq_file *sf, struct blkg_policy_data *pd, int off)
605 return __blkg_prfill_u64(sf, pd, blkg_stat_read((void *)pd + off));
607 EXPORT_SYMBOL_GPL(blkg_prfill_stat);
610 * blkg_prfill_rwstat - prfill callback for blkg_rwstat
611 * @sf: seq_file to print to
612 * @pd: policy private data of interest
613 * @off: offset to the blkg_rwstat in @pd
615 * prfill callback for printing a blkg_rwstat.
617 u64 blkg_prfill_rwstat(struct seq_file *sf, struct blkg_policy_data *pd,
620 struct blkg_rwstat rwstat = blkg_rwstat_read((void *)pd + off);
622 return __blkg_prfill_rwstat(sf, pd, &rwstat);
624 EXPORT_SYMBOL_GPL(blkg_prfill_rwstat);
627 * blkg_stat_recursive_sum - collect hierarchical blkg_stat
628 * @pd: policy private data of interest
629 * @off: offset to the blkg_stat in @pd
631 * Collect the blkg_stat specified by @off from @pd and all its online
632 * descendants and return the sum. The caller must be holding the queue
633 * lock for online tests.
635 u64 blkg_stat_recursive_sum(struct blkg_policy_data *pd, int off)
637 struct blkcg_policy *pol = blkcg_policy[pd->plid];
638 struct blkcg_gq *pos_blkg;
639 struct cgroup_subsys_state *pos_css;
642 lockdep_assert_held(pd->blkg->q->queue_lock);
645 blkg_for_each_descendant_pre(pos_blkg, pos_css, pd_to_blkg(pd)) {
646 struct blkg_policy_data *pos_pd = blkg_to_pd(pos_blkg, pol);
647 struct blkg_stat *stat = (void *)pos_pd + off;
649 if (pos_blkg->online)
650 sum += blkg_stat_read(stat);
656 EXPORT_SYMBOL_GPL(blkg_stat_recursive_sum);
659 * blkg_rwstat_recursive_sum - collect hierarchical blkg_rwstat
660 * @pd: policy private data of interest
661 * @off: offset to the blkg_stat in @pd
663 * Collect the blkg_rwstat specified by @off from @pd and all its online
664 * descendants and return the sum. The caller must be holding the queue
665 * lock for online tests.
667 struct blkg_rwstat blkg_rwstat_recursive_sum(struct blkg_policy_data *pd,
670 struct blkcg_policy *pol = blkcg_policy[pd->plid];
671 struct blkcg_gq *pos_blkg;
672 struct cgroup_subsys_state *pos_css;
673 struct blkg_rwstat sum = { };
676 lockdep_assert_held(pd->blkg->q->queue_lock);
679 blkg_for_each_descendant_pre(pos_blkg, pos_css, pd_to_blkg(pd)) {
680 struct blkg_policy_data *pos_pd = blkg_to_pd(pos_blkg, pol);
681 struct blkg_rwstat *rwstat = (void *)pos_pd + off;
682 struct blkg_rwstat tmp;
684 if (!pos_blkg->online)
687 tmp = blkg_rwstat_read(rwstat);
689 for (i = 0; i < BLKG_RWSTAT_NR; i++)
690 sum.cnt[i] += tmp.cnt[i];
696 EXPORT_SYMBOL_GPL(blkg_rwstat_recursive_sum);
699 * blkg_conf_prep - parse and prepare for per-blkg config update
700 * @blkcg: target block cgroup
701 * @pol: target policy
702 * @input: input string
703 * @ctx: blkg_conf_ctx to be filled
705 * Parse per-blkg config update from @input and initialize @ctx with the
706 * result. @ctx->blkg points to the blkg to be updated and @ctx->v the new
707 * value. This function returns with RCU read lock and queue lock held and
708 * must be paired with blkg_conf_finish().
710 int blkg_conf_prep(struct blkcg *blkcg, const struct blkcg_policy *pol,
711 const char *input, struct blkg_conf_ctx *ctx)
712 __acquires(rcu) __acquires(disk->queue->queue_lock)
714 struct gendisk *disk;
715 struct blkcg_gq *blkg;
716 unsigned int major, minor;
717 unsigned long long v;
720 if (sscanf(input, "%u:%u %llu", &major, &minor, &v) != 3)
723 disk = get_gendisk(MKDEV(major, minor), &part);
732 spin_lock_irq(disk->queue->queue_lock);
734 if (blkcg_policy_enabled(disk->queue, pol))
735 blkg = blkg_lookup_create(blkcg, disk->queue);
737 blkg = ERR_PTR(-EINVAL);
742 spin_unlock_irq(disk->queue->queue_lock);
745 * If queue was bypassing, we should retry. Do so after a
746 * short msleep(). It isn't strictly necessary but queue
747 * can be bypassing for some time and it's always nice to
748 * avoid busy looping.
752 ret = restart_syscall();
762 EXPORT_SYMBOL_GPL(blkg_conf_prep);
765 * blkg_conf_finish - finish up per-blkg config update
766 * @ctx: blkg_conf_ctx intiailized by blkg_conf_prep()
768 * Finish up after per-blkg config update. This function must be paired
769 * with blkg_conf_prep().
771 void blkg_conf_finish(struct blkg_conf_ctx *ctx)
772 __releases(ctx->disk->queue->queue_lock) __releases(rcu)
774 spin_unlock_irq(ctx->disk->queue->queue_lock);
778 EXPORT_SYMBOL_GPL(blkg_conf_finish);
780 struct cftype blkcg_files[] = {
782 .name = "reset_stats",
783 .write_u64 = blkcg_reset_stats,
789 * blkcg_css_offline - cgroup css_offline callback
790 * @css: css of interest
792 * This function is called when @css is about to go away and responsible
793 * for shooting down all blkgs associated with @css. blkgs should be
794 * removed while holding both q and blkcg locks. As blkcg lock is nested
795 * inside q lock, this function performs reverse double lock dancing.
797 * This is the blkcg counterpart of ioc_release_fn().
799 static void blkcg_css_offline(struct cgroup_subsys_state *css)
801 struct blkcg *blkcg = css_to_blkcg(css);
803 spin_lock_irq(&blkcg->lock);
805 while (!hlist_empty(&blkcg->blkg_list)) {
806 struct blkcg_gq *blkg = hlist_entry(blkcg->blkg_list.first,
807 struct blkcg_gq, blkcg_node);
808 struct request_queue *q = blkg->q;
810 if (spin_trylock(q->queue_lock)) {
812 spin_unlock(q->queue_lock);
814 spin_unlock_irq(&blkcg->lock);
816 spin_lock_irq(&blkcg->lock);
820 spin_unlock_irq(&blkcg->lock);
822 wb_blkcg_offline(blkcg);
825 static void blkcg_css_free(struct cgroup_subsys_state *css)
827 struct blkcg *blkcg = css_to_blkcg(css);
829 mutex_lock(&blkcg_pol_mutex);
830 list_del(&blkcg->all_blkcgs_node);
831 mutex_unlock(&blkcg_pol_mutex);
833 if (blkcg != &blkcg_root) {
836 for (i = 0; i < BLKCG_MAX_POLS; i++)
842 static struct cgroup_subsys_state *
843 blkcg_css_alloc(struct cgroup_subsys_state *parent_css)
846 struct cgroup_subsys_state *ret;
849 mutex_lock(&blkcg_pol_mutex);
856 blkcg = kzalloc(sizeof(*blkcg), GFP_KERNEL);
858 ret = ERR_PTR(-ENOMEM);
862 for (i = 0; i < BLKCG_MAX_POLS ; i++) {
863 struct blkcg_policy *pol = blkcg_policy[i];
864 struct blkcg_policy_data *cpd;
867 * If the policy hasn't been attached yet, wait for it
868 * to be attached before doing anything else. Otherwise,
869 * check if the policy requires any specific per-cgroup
870 * data: if it does, allocate and initialize it.
872 if (!pol || !pol->cpd_size)
875 BUG_ON(blkcg->pd[i]);
876 cpd = kzalloc(pol->cpd_size, GFP_KERNEL);
878 ret = ERR_PTR(-ENOMEM);
883 pol->cpd_init_fn(blkcg);
887 spin_lock_init(&blkcg->lock);
888 INIT_RADIX_TREE(&blkcg->blkg_tree, GFP_ATOMIC);
889 INIT_HLIST_HEAD(&blkcg->blkg_list);
890 #ifdef CONFIG_CGROUP_WRITEBACK
891 INIT_LIST_HEAD(&blkcg->cgwb_list);
893 list_add_tail(&blkcg->all_blkcgs_node, &all_blkcgs);
895 mutex_unlock(&blkcg_pol_mutex);
899 for (i--; i >= 0; i--)
903 mutex_unlock(&blkcg_pol_mutex);
908 * blkcg_init_queue - initialize blkcg part of request queue
909 * @q: request_queue to initialize
911 * Called from blk_alloc_queue_node(). Responsible for initializing blkcg
912 * part of new request_queue @q.
915 * 0 on success, -errno on failure.
917 int blkcg_init_queue(struct request_queue *q)
919 struct blkcg_gq *new_blkg, *blkg;
923 new_blkg = blkg_alloc(&blkcg_root, q, GFP_KERNEL);
927 preloaded = !radix_tree_preload(GFP_KERNEL);
930 * Make sure the root blkg exists and count the existing blkgs. As
931 * @q is bypassing at this point, blkg_lookup_create() can't be
932 * used. Open code insertion.
935 spin_lock_irq(q->queue_lock);
936 blkg = blkg_create(&blkcg_root, q, new_blkg);
937 spin_unlock_irq(q->queue_lock);
941 radix_tree_preload_end();
945 return PTR_ERR(blkg);
949 q->root_rl.blkg = blkg;
951 ret = blk_throtl_init(q);
953 spin_lock_irq(q->queue_lock);
955 spin_unlock_irq(q->queue_lock);
961 * blkcg_drain_queue - drain blkcg part of request_queue
962 * @q: request_queue to drain
964 * Called from blk_drain_queue(). Responsible for draining blkcg part.
966 void blkcg_drain_queue(struct request_queue *q)
968 lockdep_assert_held(q->queue_lock);
971 * @q could be exiting and already have destroyed all blkgs as
972 * indicated by NULL root_blkg. If so, don't confuse policies.
981 * blkcg_exit_queue - exit and release blkcg part of request_queue
982 * @q: request_queue being released
984 * Called from blk_release_queue(). Responsible for exiting blkcg part.
986 void blkcg_exit_queue(struct request_queue *q)
988 spin_lock_irq(q->queue_lock);
990 spin_unlock_irq(q->queue_lock);
996 * We cannot support shared io contexts, as we have no mean to support
997 * two tasks with the same ioc in two different groups without major rework
998 * of the main cic data structures. For now we allow a task to change
999 * its cgroup only if it's the only owner of its ioc.
1001 static int blkcg_can_attach(struct cgroup_subsys_state *css,
1002 struct cgroup_taskset *tset)
1004 struct task_struct *task;
1005 struct io_context *ioc;
1008 /* task_lock() is needed to avoid races with exit_io_context() */
1009 cgroup_taskset_for_each(task, tset) {
1011 ioc = task->io_context;
1012 if (ioc && atomic_read(&ioc->nr_tasks) > 1)
1021 struct cgroup_subsys blkio_cgrp_subsys = {
1022 .css_alloc = blkcg_css_alloc,
1023 .css_offline = blkcg_css_offline,
1024 .css_free = blkcg_css_free,
1025 .can_attach = blkcg_can_attach,
1026 .legacy_cftypes = blkcg_files,
1029 * This ensures that, if available, memcg is automatically enabled
1030 * together on the default hierarchy so that the owner cgroup can
1031 * be retrieved from writeback pages.
1033 .depends_on = 1 << memory_cgrp_id,
1036 EXPORT_SYMBOL_GPL(blkio_cgrp_subsys);
1039 * blkcg_activate_policy - activate a blkcg policy on a request_queue
1040 * @q: request_queue of interest
1041 * @pol: blkcg policy to activate
1043 * Activate @pol on @q. Requires %GFP_KERNEL context. @q goes through
1044 * bypass mode to populate its blkgs with policy_data for @pol.
1046 * Activation happens with @q bypassed, so nobody would be accessing blkgs
1047 * from IO path. Update of each blkg is protected by both queue and blkcg
1048 * locks so that holding either lock and testing blkcg_policy_enabled() is
1049 * always enough for dereferencing policy data.
1051 * The caller is responsible for synchronizing [de]activations and policy
1052 * [un]registerations. Returns 0 on success, -errno on failure.
1054 int blkcg_activate_policy(struct request_queue *q,
1055 const struct blkcg_policy *pol)
1058 struct blkcg_gq *blkg;
1059 struct blkg_policy_data *pd, *nd;
1062 if (blkcg_policy_enabled(q, pol))
1065 /* count and allocate policy_data for all existing blkgs */
1066 blk_queue_bypass_start(q);
1067 spin_lock_irq(q->queue_lock);
1068 list_for_each_entry(blkg, &q->blkg_list, q_node)
1070 spin_unlock_irq(q->queue_lock);
1072 /* allocate per-blkg policy data for all existing blkgs */
1074 pd = kzalloc_node(pol->pd_size, GFP_KERNEL, q->node);
1079 list_add_tail(&pd->alloc_node, &pds);
1083 * Install the allocated pds and cpds. With @q bypassing, no new blkg
1084 * should have been created while the queue lock was dropped.
1086 spin_lock_irq(q->queue_lock);
1088 list_for_each_entry(blkg, &q->blkg_list, q_node) {
1089 if (WARN_ON(list_empty(&pds))) {
1090 /* umm... this shouldn't happen, just abort */
1094 pd = list_first_entry(&pds, struct blkg_policy_data, alloc_node);
1095 list_del_init(&pd->alloc_node);
1097 /* grab blkcg lock too while installing @pd on @blkg */
1098 spin_lock(&blkg->blkcg->lock);
1100 blkg->pd[pol->plid] = pd;
1102 pd->plid = pol->plid;
1103 pol->pd_init_fn(blkg);
1105 spin_unlock(&blkg->blkcg->lock);
1108 __set_bit(pol->plid, q->blkcg_pols);
1111 spin_unlock_irq(q->queue_lock);
1113 blk_queue_bypass_end(q);
1114 list_for_each_entry_safe(pd, nd, &pds, alloc_node)
1118 EXPORT_SYMBOL_GPL(blkcg_activate_policy);
1121 * blkcg_deactivate_policy - deactivate a blkcg policy on a request_queue
1122 * @q: request_queue of interest
1123 * @pol: blkcg policy to deactivate
1125 * Deactivate @pol on @q. Follows the same synchronization rules as
1126 * blkcg_activate_policy().
1128 void blkcg_deactivate_policy(struct request_queue *q,
1129 const struct blkcg_policy *pol)
1131 struct blkcg_gq *blkg;
1133 if (!blkcg_policy_enabled(q, pol))
1136 blk_queue_bypass_start(q);
1137 spin_lock_irq(q->queue_lock);
1139 __clear_bit(pol->plid, q->blkcg_pols);
1141 list_for_each_entry(blkg, &q->blkg_list, q_node) {
1142 /* grab blkcg lock too while removing @pd from @blkg */
1143 spin_lock(&blkg->blkcg->lock);
1145 if (pol->pd_offline_fn)
1146 pol->pd_offline_fn(blkg);
1147 if (pol->pd_exit_fn)
1148 pol->pd_exit_fn(blkg);
1150 kfree(blkg->pd[pol->plid]);
1151 blkg->pd[pol->plid] = NULL;
1153 spin_unlock(&blkg->blkcg->lock);
1156 spin_unlock_irq(q->queue_lock);
1157 blk_queue_bypass_end(q);
1159 EXPORT_SYMBOL_GPL(blkcg_deactivate_policy);
1162 * blkcg_policy_register - register a blkcg policy
1163 * @pol: blkcg policy to register
1165 * Register @pol with blkcg core. Might sleep and @pol may be modified on
1166 * successful registration. Returns 0 on success and -errno on failure.
1168 int blkcg_policy_register(struct blkcg_policy *pol)
1170 struct blkcg *blkcg;
1173 if (WARN_ON(pol->pd_size < sizeof(struct blkg_policy_data)))
1176 mutex_lock(&blkcg_pol_register_mutex);
1177 mutex_lock(&blkcg_pol_mutex);
1179 /* find an empty slot */
1181 for (i = 0; i < BLKCG_MAX_POLS; i++)
1182 if (!blkcg_policy[i])
1184 if (i >= BLKCG_MAX_POLS)
1189 blkcg_policy[pol->plid] = pol;
1191 /* allocate and install cpd's */
1192 if (pol->cpd_size) {
1193 list_for_each_entry(blkcg, &all_blkcgs, all_blkcgs_node) {
1194 struct blkcg_policy_data *cpd;
1196 cpd = kzalloc(pol->cpd_size, GFP_KERNEL);
1198 mutex_unlock(&blkcg_pol_mutex);
1202 blkcg->pd[pol->plid] = cpd;
1203 cpd->plid = pol->plid;
1204 pol->cpd_init_fn(blkcg);
1208 mutex_unlock(&blkcg_pol_mutex);
1210 /* everything is in place, add intf files for the new policy */
1212 WARN_ON(cgroup_add_legacy_cftypes(&blkio_cgrp_subsys,
1214 mutex_unlock(&blkcg_pol_register_mutex);
1218 if (pol->cpd_size) {
1219 list_for_each_entry(blkcg, &all_blkcgs, all_blkcgs_node) {
1220 kfree(blkcg->pd[pol->plid]);
1221 blkcg->pd[pol->plid] = NULL;
1224 blkcg_policy[pol->plid] = NULL;
1226 mutex_unlock(&blkcg_pol_mutex);
1227 mutex_unlock(&blkcg_pol_register_mutex);
1230 EXPORT_SYMBOL_GPL(blkcg_policy_register);
1233 * blkcg_policy_unregister - unregister a blkcg policy
1234 * @pol: blkcg policy to unregister
1236 * Undo blkcg_policy_register(@pol). Might sleep.
1238 void blkcg_policy_unregister(struct blkcg_policy *pol)
1240 struct blkcg *blkcg;
1242 mutex_lock(&blkcg_pol_register_mutex);
1244 if (WARN_ON(blkcg_policy[pol->plid] != pol))
1247 /* kill the intf files first */
1249 cgroup_rm_cftypes(pol->cftypes);
1251 /* remove cpds and unregister */
1252 mutex_lock(&blkcg_pol_mutex);
1254 if (pol->cpd_size) {
1255 list_for_each_entry(blkcg, &all_blkcgs, all_blkcgs_node) {
1256 kfree(blkcg->pd[pol->plid]);
1257 blkcg->pd[pol->plid] = NULL;
1260 blkcg_policy[pol->plid] = NULL;
1262 mutex_unlock(&blkcg_pol_mutex);
1264 mutex_unlock(&blkcg_pol_register_mutex);
1266 EXPORT_SYMBOL_GPL(blkcg_policy_unregister);