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++)
72 blkcg_policy[i]->pd_free_fn(blkg->pd[i]);
74 if (blkg->blkcg != &blkcg_root)
75 blk_exit_rl(&blkg->rl);
80 * blkg_alloc - allocate a blkg
81 * @blkcg: block cgroup the new blkg is associated with
82 * @q: request_queue the new blkg is associated with
83 * @gfp_mask: allocation mask to use
85 * Allocate a new blkg assocating @blkcg and @q.
87 static struct blkcg_gq *blkg_alloc(struct blkcg *blkcg, struct request_queue *q,
90 struct blkcg_gq *blkg;
93 /* alloc and init base part */
94 blkg = kzalloc_node(sizeof(*blkg), gfp_mask, q->node);
99 INIT_LIST_HEAD(&blkg->q_node);
101 atomic_set(&blkg->refcnt, 1);
103 /* root blkg uses @q->root_rl, init rl only for !root blkgs */
104 if (blkcg != &blkcg_root) {
105 if (blk_init_rl(&blkg->rl, q, gfp_mask))
107 blkg->rl.blkg = blkg;
110 for (i = 0; i < BLKCG_MAX_POLS; i++) {
111 struct blkcg_policy *pol = blkcg_policy[i];
112 struct blkg_policy_data *pd;
114 if (!blkcg_policy_enabled(q, pol))
117 /* alloc per-policy data and attach it to blkg */
118 pd = pol->pd_alloc_fn(gfp_mask, q->node);
134 struct blkcg_gq *blkg_lookup_slowpath(struct blkcg *blkcg,
135 struct request_queue *q, bool update_hint)
137 struct blkcg_gq *blkg;
140 * Hint didn't match. Look up from the radix tree. Note that the
141 * hint can only be updated under queue_lock as otherwise @blkg
142 * could have already been removed from blkg_tree. The caller is
143 * responsible for grabbing queue_lock if @update_hint.
145 blkg = radix_tree_lookup(&blkcg->blkg_tree, q->id);
146 if (blkg && blkg->q == q) {
148 lockdep_assert_held(q->queue_lock);
149 rcu_assign_pointer(blkcg->blkg_hint, blkg);
156 EXPORT_SYMBOL_GPL(blkg_lookup_slowpath);
159 * If @new_blkg is %NULL, this function tries to allocate a new one as
160 * necessary using %GFP_NOWAIT. @new_blkg is always consumed on return.
162 static struct blkcg_gq *blkg_create(struct blkcg *blkcg,
163 struct request_queue *q,
164 struct blkcg_gq *new_blkg)
166 struct blkcg_gq *blkg;
167 struct bdi_writeback_congested *wb_congested;
170 WARN_ON_ONCE(!rcu_read_lock_held());
171 lockdep_assert_held(q->queue_lock);
173 /* blkg holds a reference to blkcg */
174 if (!css_tryget_online(&blkcg->css)) {
179 wb_congested = wb_congested_get_create(&q->backing_dev_info,
180 blkcg->css.id, GFP_NOWAIT);
188 new_blkg = blkg_alloc(blkcg, q, GFP_NOWAIT);
189 if (unlikely(!new_blkg)) {
191 goto err_put_congested;
195 blkg->wb_congested = wb_congested;
198 if (blkcg_parent(blkcg)) {
199 blkg->parent = __blkg_lookup(blkcg_parent(blkcg), q, false);
200 if (WARN_ON_ONCE(!blkg->parent)) {
202 goto err_put_congested;
204 blkg_get(blkg->parent);
207 /* invoke per-policy init */
208 for (i = 0; i < BLKCG_MAX_POLS; i++) {
209 struct blkcg_policy *pol = blkcg_policy[i];
211 if (blkg->pd[i] && pol->pd_init_fn)
212 pol->pd_init_fn(blkg->pd[i]);
216 spin_lock(&blkcg->lock);
217 ret = radix_tree_insert(&blkcg->blkg_tree, q->id, blkg);
219 hlist_add_head_rcu(&blkg->blkcg_node, &blkcg->blkg_list);
220 list_add(&blkg->q_node, &q->blkg_list);
222 for (i = 0; i < BLKCG_MAX_POLS; i++) {
223 struct blkcg_policy *pol = blkcg_policy[i];
225 if (blkg->pd[i] && pol->pd_online_fn)
226 pol->pd_online_fn(blkg->pd[i]);
230 spin_unlock(&blkcg->lock);
235 /* @blkg failed fully initialized, use the usual release path */
240 wb_congested_put(wb_congested);
242 css_put(&blkcg->css);
249 * blkg_lookup_create - lookup blkg, try to create one if not there
250 * @blkcg: blkcg of interest
251 * @q: request_queue of interest
253 * Lookup blkg for the @blkcg - @q pair. If it doesn't exist, try to
254 * create one. blkg creation is performed recursively from blkcg_root such
255 * that all non-root blkg's have access to the parent blkg. This function
256 * should be called under RCU read lock and @q->queue_lock.
258 * Returns pointer to the looked up or created blkg on success, ERR_PTR()
259 * value on error. If @q is dead, returns ERR_PTR(-EINVAL). If @q is not
260 * dead and bypassing, returns ERR_PTR(-EBUSY).
262 struct blkcg_gq *blkg_lookup_create(struct blkcg *blkcg,
263 struct request_queue *q)
265 struct blkcg_gq *blkg;
267 WARN_ON_ONCE(!rcu_read_lock_held());
268 lockdep_assert_held(q->queue_lock);
271 * This could be the first entry point of blkcg implementation and
272 * we shouldn't allow anything to go through for a bypassing queue.
274 if (unlikely(blk_queue_bypass(q)))
275 return ERR_PTR(blk_queue_dying(q) ? -EINVAL : -EBUSY);
277 blkg = __blkg_lookup(blkcg, q, true);
282 * Create blkgs walking down from blkcg_root to @blkcg, so that all
283 * non-root blkgs have access to their parents.
286 struct blkcg *pos = blkcg;
287 struct blkcg *parent = blkcg_parent(blkcg);
289 while (parent && !__blkg_lookup(parent, q, false)) {
291 parent = blkcg_parent(parent);
294 blkg = blkg_create(pos, q, NULL);
295 if (pos == blkcg || IS_ERR(blkg))
300 static void blkg_destroy(struct blkcg_gq *blkg)
302 struct blkcg *blkcg = blkg->blkcg;
305 lockdep_assert_held(blkg->q->queue_lock);
306 lockdep_assert_held(&blkcg->lock);
308 /* Something wrong if we are trying to remove same group twice */
309 WARN_ON_ONCE(list_empty(&blkg->q_node));
310 WARN_ON_ONCE(hlist_unhashed(&blkg->blkcg_node));
312 for (i = 0; i < BLKCG_MAX_POLS; i++) {
313 struct blkcg_policy *pol = blkcg_policy[i];
315 if (blkg->pd[i] && pol->pd_offline_fn)
316 pol->pd_offline_fn(blkg->pd[i]);
318 blkg->online = false;
320 radix_tree_delete(&blkcg->blkg_tree, blkg->q->id);
321 list_del_init(&blkg->q_node);
322 hlist_del_init_rcu(&blkg->blkcg_node);
325 * Both setting lookup hint to and clearing it from @blkg are done
326 * under queue_lock. If it's not pointing to @blkg now, it never
327 * will. Hint assignment itself can race safely.
329 if (rcu_access_pointer(blkcg->blkg_hint) == blkg)
330 rcu_assign_pointer(blkcg->blkg_hint, NULL);
333 * Put the reference taken at the time of creation so that when all
334 * queues are gone, group can be destroyed.
340 * blkg_destroy_all - destroy all blkgs associated with a request_queue
341 * @q: request_queue of interest
343 * Destroy all blkgs associated with @q.
345 static void blkg_destroy_all(struct request_queue *q)
347 struct blkcg_gq *blkg, *n;
349 lockdep_assert_held(q->queue_lock);
351 list_for_each_entry_safe(blkg, n, &q->blkg_list, q_node) {
352 struct blkcg *blkcg = blkg->blkcg;
354 spin_lock(&blkcg->lock);
356 spin_unlock(&blkcg->lock);
361 * A group is RCU protected, but having an rcu lock does not mean that one
362 * can access all the fields of blkg and assume these are valid. For
363 * example, don't try to follow throtl_data and request queue links.
365 * Having a reference to blkg under an rcu allows accesses to only values
366 * local to groups like group stats and group rate limits.
368 void __blkg_release_rcu(struct rcu_head *rcu_head)
370 struct blkcg_gq *blkg = container_of(rcu_head, struct blkcg_gq, rcu_head);
372 /* release the blkcg and parent blkg refs this blkg has been holding */
373 css_put(&blkg->blkcg->css);
375 blkg_put(blkg->parent);
377 wb_congested_put(blkg->wb_congested);
381 EXPORT_SYMBOL_GPL(__blkg_release_rcu);
384 * The next function used by blk_queue_for_each_rl(). It's a bit tricky
385 * because the root blkg uses @q->root_rl instead of its own rl.
387 struct request_list *__blk_queue_next_rl(struct request_list *rl,
388 struct request_queue *q)
390 struct list_head *ent;
391 struct blkcg_gq *blkg;
394 * Determine the current blkg list_head. The first entry is
395 * root_rl which is off @q->blkg_list and mapped to the head.
397 if (rl == &q->root_rl) {
399 /* There are no more block groups, hence no request lists */
403 blkg = container_of(rl, struct blkcg_gq, rl);
407 /* walk to the next list_head, skip root blkcg */
409 if (ent == &q->root_blkg->q_node)
411 if (ent == &q->blkg_list)
414 blkg = container_of(ent, struct blkcg_gq, q_node);
418 static int blkcg_reset_stats(struct cgroup_subsys_state *css,
419 struct cftype *cftype, u64 val)
421 struct blkcg *blkcg = css_to_blkcg(css);
422 struct blkcg_gq *blkg;
425 mutex_lock(&blkcg_pol_mutex);
426 spin_lock_irq(&blkcg->lock);
429 * Note that stat reset is racy - it doesn't synchronize against
430 * stat updates. This is a debug feature which shouldn't exist
431 * anyway. If you get hit by a race, retry.
433 hlist_for_each_entry(blkg, &blkcg->blkg_list, blkcg_node) {
434 for (i = 0; i < BLKCG_MAX_POLS; i++) {
435 struct blkcg_policy *pol = blkcg_policy[i];
437 if (blkg->pd[i] && pol->pd_reset_stats_fn)
438 pol->pd_reset_stats_fn(blkg->pd[i]);
442 spin_unlock_irq(&blkcg->lock);
443 mutex_unlock(&blkcg_pol_mutex);
447 static const char *blkg_dev_name(struct blkcg_gq *blkg)
449 /* some drivers (floppy) instantiate a queue w/o disk registered */
450 if (blkg->q->backing_dev_info.dev)
451 return dev_name(blkg->q->backing_dev_info.dev);
456 * blkcg_print_blkgs - helper for printing per-blkg data
457 * @sf: seq_file to print to
458 * @blkcg: blkcg of interest
459 * @prfill: fill function to print out a blkg
460 * @pol: policy in question
461 * @data: data to be passed to @prfill
462 * @show_total: to print out sum of prfill return values or not
464 * This function invokes @prfill on each blkg of @blkcg if pd for the
465 * policy specified by @pol exists. @prfill is invoked with @sf, the
466 * policy data and @data and the matching queue lock held. If @show_total
467 * is %true, the sum of the return values from @prfill is printed with
468 * "Total" label at the end.
470 * This is to be used to construct print functions for
471 * cftype->read_seq_string method.
473 void blkcg_print_blkgs(struct seq_file *sf, struct blkcg *blkcg,
474 u64 (*prfill)(struct seq_file *,
475 struct blkg_policy_data *, int),
476 const struct blkcg_policy *pol, int data,
479 struct blkcg_gq *blkg;
483 hlist_for_each_entry_rcu(blkg, &blkcg->blkg_list, blkcg_node) {
484 spin_lock_irq(blkg->q->queue_lock);
485 if (blkcg_policy_enabled(blkg->q, pol))
486 total += prfill(sf, blkg->pd[pol->plid], data);
487 spin_unlock_irq(blkg->q->queue_lock);
492 seq_printf(sf, "Total %llu\n", (unsigned long long)total);
494 EXPORT_SYMBOL_GPL(blkcg_print_blkgs);
497 * __blkg_prfill_u64 - prfill helper for a single u64 value
498 * @sf: seq_file to print to
499 * @pd: policy private data of interest
502 * Print @v to @sf for the device assocaited with @pd.
504 u64 __blkg_prfill_u64(struct seq_file *sf, struct blkg_policy_data *pd, u64 v)
506 const char *dname = blkg_dev_name(pd->blkg);
511 seq_printf(sf, "%s %llu\n", dname, (unsigned long long)v);
514 EXPORT_SYMBOL_GPL(__blkg_prfill_u64);
517 * __blkg_prfill_rwstat - prfill helper for a blkg_rwstat
518 * @sf: seq_file to print to
519 * @pd: policy private data of interest
520 * @rwstat: rwstat to print
522 * Print @rwstat to @sf for the device assocaited with @pd.
524 u64 __blkg_prfill_rwstat(struct seq_file *sf, struct blkg_policy_data *pd,
525 const struct blkg_rwstat *rwstat)
527 static const char *rwstr[] = {
528 [BLKG_RWSTAT_READ] = "Read",
529 [BLKG_RWSTAT_WRITE] = "Write",
530 [BLKG_RWSTAT_SYNC] = "Sync",
531 [BLKG_RWSTAT_ASYNC] = "Async",
533 const char *dname = blkg_dev_name(pd->blkg);
540 for (i = 0; i < BLKG_RWSTAT_NR; i++)
541 seq_printf(sf, "%s %s %llu\n", dname, rwstr[i],
542 (unsigned long long)atomic64_read(&rwstat->aux_cnt[i]));
544 v = atomic64_read(&rwstat->aux_cnt[BLKG_RWSTAT_READ]) +
545 atomic64_read(&rwstat->aux_cnt[BLKG_RWSTAT_WRITE]);
546 seq_printf(sf, "%s Total %llu\n", dname, (unsigned long long)v);
549 EXPORT_SYMBOL_GPL(__blkg_prfill_rwstat);
552 * blkg_prfill_stat - prfill callback for blkg_stat
553 * @sf: seq_file to print to
554 * @pd: policy private data of interest
555 * @off: offset to the blkg_stat in @pd
557 * prfill callback for printing a blkg_stat.
559 u64 blkg_prfill_stat(struct seq_file *sf, struct blkg_policy_data *pd, int off)
561 return __blkg_prfill_u64(sf, pd, blkg_stat_read((void *)pd + off));
563 EXPORT_SYMBOL_GPL(blkg_prfill_stat);
566 * blkg_prfill_rwstat - prfill callback for blkg_rwstat
567 * @sf: seq_file to print to
568 * @pd: policy private data of interest
569 * @off: offset to the blkg_rwstat in @pd
571 * prfill callback for printing a blkg_rwstat.
573 u64 blkg_prfill_rwstat(struct seq_file *sf, struct blkg_policy_data *pd,
576 struct blkg_rwstat rwstat = blkg_rwstat_read((void *)pd + off);
578 return __blkg_prfill_rwstat(sf, pd, &rwstat);
580 EXPORT_SYMBOL_GPL(blkg_prfill_rwstat);
583 * blkg_stat_recursive_sum - collect hierarchical blkg_stat
584 * @pd: policy private data of interest
585 * @off: offset to the blkg_stat in @pd
587 * Collect the blkg_stat specified by @off from @pd and all its online
588 * descendants and their aux counts. The caller must be holding the queue
589 * lock for online tests.
591 u64 blkg_stat_recursive_sum(struct blkg_policy_data *pd, int off)
593 struct blkcg_policy *pol = blkcg_policy[pd->plid];
594 struct blkcg_gq *pos_blkg;
595 struct cgroup_subsys_state *pos_css;
598 lockdep_assert_held(pd->blkg->q->queue_lock);
601 blkg_for_each_descendant_pre(pos_blkg, pos_css, pd_to_blkg(pd)) {
602 struct blkg_policy_data *pos_pd = blkg_to_pd(pos_blkg, pol);
603 struct blkg_stat *stat = (void *)pos_pd + off;
605 if (pos_blkg->online)
606 sum += blkg_stat_read(stat) +
607 atomic64_read(&stat->aux_cnt);
613 EXPORT_SYMBOL_GPL(blkg_stat_recursive_sum);
616 * blkg_rwstat_recursive_sum - collect hierarchical blkg_rwstat
617 * @pd: policy private data of interest
618 * @off: offset to the blkg_stat in @pd
620 * Collect the blkg_rwstat specified by @off from @pd and all its online
621 * descendants and their aux counts. The caller must be holding the queue
622 * lock for online tests.
624 struct blkg_rwstat blkg_rwstat_recursive_sum(struct blkg_policy_data *pd,
627 struct blkcg_policy *pol = blkcg_policy[pd->plid];
628 struct blkcg_gq *pos_blkg;
629 struct cgroup_subsys_state *pos_css;
630 struct blkg_rwstat sum = { };
633 lockdep_assert_held(pd->blkg->q->queue_lock);
636 blkg_for_each_descendant_pre(pos_blkg, pos_css, pd_to_blkg(pd)) {
637 struct blkg_policy_data *pos_pd = blkg_to_pd(pos_blkg, pol);
638 struct blkg_rwstat *rwstat = (void *)pos_pd + off;
639 struct blkg_rwstat tmp;
641 if (!pos_blkg->online)
644 tmp = blkg_rwstat_read(rwstat);
646 for (i = 0; i < BLKG_RWSTAT_NR; i++)
647 atomic64_add(atomic64_read(&tmp.aux_cnt[i]) +
648 atomic64_read(&rwstat->aux_cnt[i]),
655 EXPORT_SYMBOL_GPL(blkg_rwstat_recursive_sum);
658 * blkg_conf_prep - parse and prepare for per-blkg config update
659 * @blkcg: target block cgroup
660 * @pol: target policy
661 * @input: input string
662 * @ctx: blkg_conf_ctx to be filled
664 * Parse per-blkg config update from @input and initialize @ctx with the
665 * result. @ctx->blkg points to the blkg to be updated and @ctx->v the new
666 * value. This function returns with RCU read lock and queue lock held and
667 * must be paired with blkg_conf_finish().
669 int blkg_conf_prep(struct blkcg *blkcg, const struct blkcg_policy *pol,
670 const char *input, struct blkg_conf_ctx *ctx)
671 __acquires(rcu) __acquires(disk->queue->queue_lock)
673 struct gendisk *disk;
674 struct blkcg_gq *blkg;
675 unsigned int major, minor;
676 unsigned long long v;
679 if (sscanf(input, "%u:%u %llu", &major, &minor, &v) != 3)
682 disk = get_gendisk(MKDEV(major, minor), &part);
691 spin_lock_irq(disk->queue->queue_lock);
693 if (blkcg_policy_enabled(disk->queue, pol))
694 blkg = blkg_lookup_create(blkcg, disk->queue);
696 blkg = ERR_PTR(-EINVAL);
701 spin_unlock_irq(disk->queue->queue_lock);
704 * If queue was bypassing, we should retry. Do so after a
705 * short msleep(). It isn't strictly necessary but queue
706 * can be bypassing for some time and it's always nice to
707 * avoid busy looping.
711 ret = restart_syscall();
721 EXPORT_SYMBOL_GPL(blkg_conf_prep);
724 * blkg_conf_finish - finish up per-blkg config update
725 * @ctx: blkg_conf_ctx intiailized by blkg_conf_prep()
727 * Finish up after per-blkg config update. This function must be paired
728 * with blkg_conf_prep().
730 void blkg_conf_finish(struct blkg_conf_ctx *ctx)
731 __releases(ctx->disk->queue->queue_lock) __releases(rcu)
733 spin_unlock_irq(ctx->disk->queue->queue_lock);
737 EXPORT_SYMBOL_GPL(blkg_conf_finish);
739 struct cftype blkcg_files[] = {
741 .name = "reset_stats",
742 .write_u64 = blkcg_reset_stats,
748 * blkcg_css_offline - cgroup css_offline callback
749 * @css: css of interest
751 * This function is called when @css is about to go away and responsible
752 * for shooting down all blkgs associated with @css. blkgs should be
753 * removed while holding both q and blkcg locks. As blkcg lock is nested
754 * inside q lock, this function performs reverse double lock dancing.
756 * This is the blkcg counterpart of ioc_release_fn().
758 static void blkcg_css_offline(struct cgroup_subsys_state *css)
760 struct blkcg *blkcg = css_to_blkcg(css);
762 spin_lock_irq(&blkcg->lock);
764 while (!hlist_empty(&blkcg->blkg_list)) {
765 struct blkcg_gq *blkg = hlist_entry(blkcg->blkg_list.first,
766 struct blkcg_gq, blkcg_node);
767 struct request_queue *q = blkg->q;
769 if (spin_trylock(q->queue_lock)) {
771 spin_unlock(q->queue_lock);
773 spin_unlock_irq(&blkcg->lock);
775 spin_lock_irq(&blkcg->lock);
779 spin_unlock_irq(&blkcg->lock);
781 wb_blkcg_offline(blkcg);
784 static void blkcg_css_free(struct cgroup_subsys_state *css)
786 struct blkcg *blkcg = css_to_blkcg(css);
789 mutex_lock(&blkcg_pol_mutex);
791 list_del(&blkcg->all_blkcgs_node);
793 for (i = 0; i < BLKCG_MAX_POLS; i++)
795 blkcg_policy[i]->cpd_free_fn(blkcg->cpd[i]);
797 mutex_unlock(&blkcg_pol_mutex);
802 static struct cgroup_subsys_state *
803 blkcg_css_alloc(struct cgroup_subsys_state *parent_css)
806 struct cgroup_subsys_state *ret;
809 mutex_lock(&blkcg_pol_mutex);
814 blkcg = kzalloc(sizeof(*blkcg), GFP_KERNEL);
816 ret = ERR_PTR(-ENOMEM);
821 for (i = 0; i < BLKCG_MAX_POLS ; i++) {
822 struct blkcg_policy *pol = blkcg_policy[i];
823 struct blkcg_policy_data *cpd;
826 * If the policy hasn't been attached yet, wait for it
827 * to be attached before doing anything else. Otherwise,
828 * check if the policy requires any specific per-cgroup
829 * data: if it does, allocate and initialize it.
831 if (!pol || !pol->cpd_alloc_fn)
834 cpd = pol->cpd_alloc_fn(GFP_KERNEL);
836 ret = ERR_PTR(-ENOMEM);
842 if (pol->cpd_init_fn)
843 pol->cpd_init_fn(cpd);
846 spin_lock_init(&blkcg->lock);
847 INIT_RADIX_TREE(&blkcg->blkg_tree, GFP_NOWAIT);
848 INIT_HLIST_HEAD(&blkcg->blkg_list);
849 #ifdef CONFIG_CGROUP_WRITEBACK
850 INIT_LIST_HEAD(&blkcg->cgwb_list);
852 list_add_tail(&blkcg->all_blkcgs_node, &all_blkcgs);
854 mutex_unlock(&blkcg_pol_mutex);
858 for (i--; i >= 0; i--)
860 blkcg_policy[i]->cpd_free_fn(blkcg->cpd[i]);
863 mutex_unlock(&blkcg_pol_mutex);
868 * blkcg_init_queue - initialize blkcg part of request queue
869 * @q: request_queue to initialize
871 * Called from blk_alloc_queue_node(). Responsible for initializing blkcg
872 * part of new request_queue @q.
875 * 0 on success, -errno on failure.
877 int blkcg_init_queue(struct request_queue *q)
879 struct blkcg_gq *new_blkg, *blkg;
883 new_blkg = blkg_alloc(&blkcg_root, q, GFP_KERNEL);
887 preloaded = !radix_tree_preload(GFP_KERNEL);
890 * Make sure the root blkg exists and count the existing blkgs. As
891 * @q is bypassing at this point, blkg_lookup_create() can't be
892 * used. Open code insertion.
895 spin_lock_irq(q->queue_lock);
896 blkg = blkg_create(&blkcg_root, q, new_blkg);
897 spin_unlock_irq(q->queue_lock);
901 radix_tree_preload_end();
905 return PTR_ERR(blkg);
909 q->root_rl.blkg = blkg;
911 ret = blk_throtl_init(q);
913 spin_lock_irq(q->queue_lock);
915 spin_unlock_irq(q->queue_lock);
921 * blkcg_drain_queue - drain blkcg part of request_queue
922 * @q: request_queue to drain
924 * Called from blk_drain_queue(). Responsible for draining blkcg part.
926 void blkcg_drain_queue(struct request_queue *q)
928 lockdep_assert_held(q->queue_lock);
931 * @q could be exiting and already have destroyed all blkgs as
932 * indicated by NULL root_blkg. If so, don't confuse policies.
941 * blkcg_exit_queue - exit and release blkcg part of request_queue
942 * @q: request_queue being released
944 * Called from blk_release_queue(). Responsible for exiting blkcg part.
946 void blkcg_exit_queue(struct request_queue *q)
948 spin_lock_irq(q->queue_lock);
950 spin_unlock_irq(q->queue_lock);
956 * We cannot support shared io contexts, as we have no mean to support
957 * two tasks with the same ioc in two different groups without major rework
958 * of the main cic data structures. For now we allow a task to change
959 * its cgroup only if it's the only owner of its ioc.
961 static int blkcg_can_attach(struct cgroup_subsys_state *css,
962 struct cgroup_taskset *tset)
964 struct task_struct *task;
965 struct io_context *ioc;
968 /* task_lock() is needed to avoid races with exit_io_context() */
969 cgroup_taskset_for_each(task, tset) {
971 ioc = task->io_context;
972 if (ioc && atomic_read(&ioc->nr_tasks) > 1)
981 struct cgroup_subsys blkio_cgrp_subsys = {
982 .css_alloc = blkcg_css_alloc,
983 .css_offline = blkcg_css_offline,
984 .css_free = blkcg_css_free,
985 .can_attach = blkcg_can_attach,
986 .legacy_cftypes = blkcg_files,
989 * This ensures that, if available, memcg is automatically enabled
990 * together on the default hierarchy so that the owner cgroup can
991 * be retrieved from writeback pages.
993 .depends_on = 1 << memory_cgrp_id,
996 EXPORT_SYMBOL_GPL(blkio_cgrp_subsys);
999 * blkcg_activate_policy - activate a blkcg policy on a request_queue
1000 * @q: request_queue of interest
1001 * @pol: blkcg policy to activate
1003 * Activate @pol on @q. Requires %GFP_KERNEL context. @q goes through
1004 * bypass mode to populate its blkgs with policy_data for @pol.
1006 * Activation happens with @q bypassed, so nobody would be accessing blkgs
1007 * from IO path. Update of each blkg is protected by both queue and blkcg
1008 * locks so that holding either lock and testing blkcg_policy_enabled() is
1009 * always enough for dereferencing policy data.
1011 * The caller is responsible for synchronizing [de]activations and policy
1012 * [un]registerations. Returns 0 on success, -errno on failure.
1014 int blkcg_activate_policy(struct request_queue *q,
1015 const struct blkcg_policy *pol)
1017 struct blkg_policy_data *pd_prealloc = NULL;
1018 struct blkcg_gq *blkg;
1021 if (blkcg_policy_enabled(q, pol))
1024 blk_queue_bypass_start(q);
1027 pd_prealloc = pol->pd_alloc_fn(GFP_KERNEL, q->node);
1030 goto out_bypass_end;
1034 spin_lock_irq(q->queue_lock);
1036 list_for_each_entry(blkg, &q->blkg_list, q_node) {
1037 struct blkg_policy_data *pd;
1039 if (blkg->pd[pol->plid])
1042 pd = pol->pd_alloc_fn(GFP_NOWAIT, q->node);
1044 swap(pd, pd_prealloc);
1046 spin_unlock_irq(q->queue_lock);
1050 blkg->pd[pol->plid] = pd;
1052 pd->plid = pol->plid;
1053 if (pol->pd_init_fn)
1054 pol->pd_init_fn(pd);
1057 __set_bit(pol->plid, q->blkcg_pols);
1060 spin_unlock_irq(q->queue_lock);
1062 blk_queue_bypass_end(q);
1064 pol->pd_free_fn(pd_prealloc);
1067 EXPORT_SYMBOL_GPL(blkcg_activate_policy);
1070 * blkcg_deactivate_policy - deactivate a blkcg policy on a request_queue
1071 * @q: request_queue of interest
1072 * @pol: blkcg policy to deactivate
1074 * Deactivate @pol on @q. Follows the same synchronization rules as
1075 * blkcg_activate_policy().
1077 void blkcg_deactivate_policy(struct request_queue *q,
1078 const struct blkcg_policy *pol)
1080 struct blkcg_gq *blkg;
1082 if (!blkcg_policy_enabled(q, pol))
1085 blk_queue_bypass_start(q);
1086 spin_lock_irq(q->queue_lock);
1088 __clear_bit(pol->plid, q->blkcg_pols);
1090 list_for_each_entry(blkg, &q->blkg_list, q_node) {
1091 /* grab blkcg lock too while removing @pd from @blkg */
1092 spin_lock(&blkg->blkcg->lock);
1094 if (blkg->pd[pol->plid]) {
1095 if (pol->pd_offline_fn)
1096 pol->pd_offline_fn(blkg->pd[pol->plid]);
1097 pol->pd_free_fn(blkg->pd[pol->plid]);
1098 blkg->pd[pol->plid] = NULL;
1101 spin_unlock(&blkg->blkcg->lock);
1104 spin_unlock_irq(q->queue_lock);
1105 blk_queue_bypass_end(q);
1107 EXPORT_SYMBOL_GPL(blkcg_deactivate_policy);
1110 * blkcg_policy_register - register a blkcg policy
1111 * @pol: blkcg policy to register
1113 * Register @pol with blkcg core. Might sleep and @pol may be modified on
1114 * successful registration. Returns 0 on success and -errno on failure.
1116 int blkcg_policy_register(struct blkcg_policy *pol)
1118 struct blkcg *blkcg;
1121 mutex_lock(&blkcg_pol_register_mutex);
1122 mutex_lock(&blkcg_pol_mutex);
1124 /* find an empty slot */
1126 for (i = 0; i < BLKCG_MAX_POLS; i++)
1127 if (!blkcg_policy[i])
1129 if (i >= BLKCG_MAX_POLS)
1134 blkcg_policy[pol->plid] = pol;
1136 /* allocate and install cpd's */
1137 if (pol->cpd_alloc_fn) {
1138 list_for_each_entry(blkcg, &all_blkcgs, all_blkcgs_node) {
1139 struct blkcg_policy_data *cpd;
1141 cpd = pol->cpd_alloc_fn(GFP_KERNEL);
1143 mutex_unlock(&blkcg_pol_mutex);
1147 blkcg->cpd[pol->plid] = cpd;
1149 cpd->plid = pol->plid;
1150 pol->cpd_init_fn(cpd);
1154 mutex_unlock(&blkcg_pol_mutex);
1156 /* everything is in place, add intf files for the new policy */
1158 WARN_ON(cgroup_add_legacy_cftypes(&blkio_cgrp_subsys,
1160 mutex_unlock(&blkcg_pol_register_mutex);
1164 if (pol->cpd_alloc_fn) {
1165 list_for_each_entry(blkcg, &all_blkcgs, all_blkcgs_node) {
1166 if (blkcg->cpd[pol->plid]) {
1167 pol->cpd_free_fn(blkcg->cpd[pol->plid]);
1168 blkcg->cpd[pol->plid] = NULL;
1172 blkcg_policy[pol->plid] = NULL;
1174 mutex_unlock(&blkcg_pol_mutex);
1175 mutex_unlock(&blkcg_pol_register_mutex);
1178 EXPORT_SYMBOL_GPL(blkcg_policy_register);
1181 * blkcg_policy_unregister - unregister a blkcg policy
1182 * @pol: blkcg policy to unregister
1184 * Undo blkcg_policy_register(@pol). Might sleep.
1186 void blkcg_policy_unregister(struct blkcg_policy *pol)
1188 struct blkcg *blkcg;
1190 mutex_lock(&blkcg_pol_register_mutex);
1192 if (WARN_ON(blkcg_policy[pol->plid] != pol))
1195 /* kill the intf files first */
1197 cgroup_rm_cftypes(pol->cftypes);
1199 /* remove cpds and unregister */
1200 mutex_lock(&blkcg_pol_mutex);
1202 if (pol->cpd_alloc_fn) {
1203 list_for_each_entry(blkcg, &all_blkcgs, all_blkcgs_node) {
1204 if (blkcg->cpd[pol->plid]) {
1205 pol->cpd_free_fn(blkcg->cpd[pol->plid]);
1206 blkcg->cpd[pol->plid] = NULL;
1210 blkcg_policy[pol->plid] = NULL;
1212 mutex_unlock(&blkcg_pol_mutex);
1214 mutex_unlock(&blkcg_pol_register_mutex);
1216 EXPORT_SYMBOL_GPL(blkcg_policy_unregister);