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 #include <linux/ioprio.h>
14 #include <linux/kdev_t.h>
15 #include <linux/module.h>
16 #include <linux/err.h>
17 #include <linux/blkdev.h>
18 #include <linux/slab.h>
19 #include <linux/genhd.h>
20 #include <linux/delay.h>
21 #include <linux/atomic.h>
22 #include "blk-cgroup.h"
25 #define MAX_KEY_LEN 100
27 static DEFINE_MUTEX(blkcg_pol_mutex);
29 struct blkcg blkcg_root = { .cfq_weight = 2 * CFQ_WEIGHT_DEFAULT };
30 EXPORT_SYMBOL_GPL(blkcg_root);
32 static struct blkcg_policy *blkcg_policy[BLKCG_MAX_POLS];
34 static bool blkcg_policy_enabled(struct request_queue *q,
35 const struct blkcg_policy *pol)
37 return pol && test_bit(pol->plid, q->blkcg_pols);
41 * blkg_free - free a blkg
44 * Free @blkg which may be partially allocated.
46 static void blkg_free(struct blkcg_gq *blkg)
53 for (i = 0; i < BLKCG_MAX_POLS; i++) {
54 struct blkcg_policy *pol = blkcg_policy[i];
55 struct blkg_policy_data *pd = blkg->pd[i];
60 if (pol && pol->pd_exit_fn)
61 pol->pd_exit_fn(blkg);
66 blk_exit_rl(&blkg->rl);
71 * blkg_alloc - allocate a blkg
72 * @blkcg: block cgroup the new blkg is associated with
73 * @q: request_queue the new blkg is associated with
74 * @gfp_mask: allocation mask to use
76 * Allocate a new blkg assocating @blkcg and @q.
78 static struct blkcg_gq *blkg_alloc(struct blkcg *blkcg, struct request_queue *q,
81 struct blkcg_gq *blkg;
84 /* alloc and init base part */
85 blkg = kzalloc_node(sizeof(*blkg), gfp_mask, q->node);
90 INIT_LIST_HEAD(&blkg->q_node);
94 /* root blkg uses @q->root_rl, init rl only for !root blkgs */
95 if (blkcg != &blkcg_root) {
96 if (blk_init_rl(&blkg->rl, q, gfp_mask))
101 for (i = 0; i < BLKCG_MAX_POLS; i++) {
102 struct blkcg_policy *pol = blkcg_policy[i];
103 struct blkg_policy_data *pd;
105 if (!blkcg_policy_enabled(q, pol))
108 /* alloc per-policy data and attach it to blkg */
109 pd = kzalloc_node(pol->pd_size, gfp_mask, q->node);
116 /* invoke per-policy init */
118 pol->pd_init_fn(blkg);
128 static struct blkcg_gq *__blkg_lookup(struct blkcg *blkcg,
129 struct request_queue *q, bool update_hint)
131 struct blkcg_gq *blkg;
133 blkg = rcu_dereference(blkcg->blkg_hint);
134 if (blkg && blkg->q == q)
138 * Hint didn't match. Look up from the radix tree. Note that the
139 * hint can only be updated under queue_lock as otherwise @blkg
140 * could have already been removed from blkg_tree. The caller is
141 * responsible for grabbing queue_lock if @update_hint.
143 blkg = radix_tree_lookup(&blkcg->blkg_tree, q->id);
144 if (blkg && blkg->q == q) {
146 lockdep_assert_held(q->queue_lock);
147 rcu_assign_pointer(blkcg->blkg_hint, blkg);
156 * blkg_lookup - lookup blkg for the specified blkcg - q pair
157 * @blkcg: blkcg of interest
158 * @q: request_queue of interest
160 * Lookup blkg for the @blkcg - @q pair. This function should be called
161 * under RCU read lock and is guaranteed to return %NULL if @q is bypassing
162 * - see blk_queue_bypass_start() for details.
164 struct blkcg_gq *blkg_lookup(struct blkcg *blkcg, struct request_queue *q)
166 WARN_ON_ONCE(!rcu_read_lock_held());
168 if (unlikely(blk_queue_bypass(q)))
170 return __blkg_lookup(blkcg, q, false);
172 EXPORT_SYMBOL_GPL(blkg_lookup);
175 * If @new_blkg is %NULL, this function tries to allocate a new one as
176 * necessary using %GFP_ATOMIC. @new_blkg is always consumed on return.
178 static struct blkcg_gq *blkg_create(struct blkcg *blkcg,
179 struct request_queue *q,
180 struct blkcg_gq *new_blkg)
182 struct blkcg_gq *blkg;
185 WARN_ON_ONCE(!rcu_read_lock_held());
186 lockdep_assert_held(q->queue_lock);
188 /* blkg holds a reference to blkcg */
189 if (!css_tryget(&blkcg->css)) {
190 blkg = ERR_PTR(-EINVAL);
196 new_blkg = blkg_alloc(blkcg, q, GFP_ATOMIC);
197 if (unlikely(!new_blkg)) {
198 blkg = ERR_PTR(-ENOMEM);
205 spin_lock(&blkcg->lock);
206 ret = radix_tree_insert(&blkcg->blkg_tree, q->id, blkg);
208 hlist_add_head_rcu(&blkg->blkcg_node, &blkcg->blkg_list);
209 list_add(&blkg->q_node, &q->blkg_list);
211 spin_unlock(&blkcg->lock);
218 css_put(&blkcg->css);
225 * blkg_lookup_create - lookup blkg, try to create one if not there
226 * @blkcg: blkcg of interest
227 * @q: request_queue of interest
229 * Lookup blkg for the @blkcg - @q pair. If it doesn't exist, try to
230 * create one. This function should be called under RCU read lock and
233 * Returns pointer to the looked up or created blkg on success, ERR_PTR()
234 * value on error. If @q is dead, returns ERR_PTR(-EINVAL). If @q is not
235 * dead and bypassing, returns ERR_PTR(-EBUSY).
237 struct blkcg_gq *blkg_lookup_create(struct blkcg *blkcg,
238 struct request_queue *q)
240 struct blkcg_gq *blkg;
242 WARN_ON_ONCE(!rcu_read_lock_held());
243 lockdep_assert_held(q->queue_lock);
246 * This could be the first entry point of blkcg implementation and
247 * we shouldn't allow anything to go through for a bypassing queue.
249 if (unlikely(blk_queue_bypass(q)))
250 return ERR_PTR(blk_queue_dying(q) ? -EINVAL : -EBUSY);
252 blkg = __blkg_lookup(blkcg, q, true);
256 return blkg_create(blkcg, q, NULL);
258 EXPORT_SYMBOL_GPL(blkg_lookup_create);
260 static void blkg_destroy(struct blkcg_gq *blkg)
262 struct blkcg *blkcg = blkg->blkcg;
264 lockdep_assert_held(blkg->q->queue_lock);
265 lockdep_assert_held(&blkcg->lock);
267 /* Something wrong if we are trying to remove same group twice */
268 WARN_ON_ONCE(list_empty(&blkg->q_node));
269 WARN_ON_ONCE(hlist_unhashed(&blkg->blkcg_node));
271 radix_tree_delete(&blkcg->blkg_tree, blkg->q->id);
272 list_del_init(&blkg->q_node);
273 hlist_del_init_rcu(&blkg->blkcg_node);
276 * Both setting lookup hint to and clearing it from @blkg are done
277 * under queue_lock. If it's not pointing to @blkg now, it never
278 * will. Hint assignment itself can race safely.
280 if (rcu_dereference_raw(blkcg->blkg_hint) == blkg)
281 rcu_assign_pointer(blkcg->blkg_hint, NULL);
284 * Put the reference taken at the time of creation so that when all
285 * queues are gone, group can be destroyed.
291 * blkg_destroy_all - destroy all blkgs associated with a request_queue
292 * @q: request_queue of interest
294 * Destroy all blkgs associated with @q.
296 static void blkg_destroy_all(struct request_queue *q)
298 struct blkcg_gq *blkg, *n;
300 lockdep_assert_held(q->queue_lock);
302 list_for_each_entry_safe(blkg, n, &q->blkg_list, q_node) {
303 struct blkcg *blkcg = blkg->blkcg;
305 spin_lock(&blkcg->lock);
307 spin_unlock(&blkcg->lock);
311 * root blkg is destroyed. Just clear the pointer since
312 * root_rl does not take reference on root blkg.
315 q->root_rl.blkg = NULL;
318 static void blkg_rcu_free(struct rcu_head *rcu_head)
320 blkg_free(container_of(rcu_head, struct blkcg_gq, rcu_head));
323 void __blkg_release(struct blkcg_gq *blkg)
325 /* release the extra blkcg reference this blkg has been holding */
326 css_put(&blkg->blkcg->css);
329 * A group is freed in rcu manner. But having an rcu lock does not
330 * mean that one can access all the fields of blkg and assume these
331 * are valid. For example, don't try to follow throtl_data and
332 * request queue links.
334 * Having a reference to blkg under an rcu allows acess to only
335 * values local to groups like group stats and group rate limits
337 call_rcu(&blkg->rcu_head, blkg_rcu_free);
339 EXPORT_SYMBOL_GPL(__blkg_release);
342 * The next function used by blk_queue_for_each_rl(). It's a bit tricky
343 * because the root blkg uses @q->root_rl instead of its own rl.
345 struct request_list *__blk_queue_next_rl(struct request_list *rl,
346 struct request_queue *q)
348 struct list_head *ent;
349 struct blkcg_gq *blkg;
352 * Determine the current blkg list_head. The first entry is
353 * root_rl which is off @q->blkg_list and mapped to the head.
355 if (rl == &q->root_rl) {
357 /* There are no more block groups, hence no request lists */
361 blkg = container_of(rl, struct blkcg_gq, rl);
365 /* walk to the next list_head, skip root blkcg */
367 if (ent == &q->root_blkg->q_node)
369 if (ent == &q->blkg_list)
372 blkg = container_of(ent, struct blkcg_gq, q_node);
376 static int blkcg_reset_stats(struct cgroup *cgroup, struct cftype *cftype,
379 struct blkcg *blkcg = cgroup_to_blkcg(cgroup);
380 struct blkcg_gq *blkg;
381 struct hlist_node *n;
384 mutex_lock(&blkcg_pol_mutex);
385 spin_lock_irq(&blkcg->lock);
388 * Note that stat reset is racy - it doesn't synchronize against
389 * stat updates. This is a debug feature which shouldn't exist
390 * anyway. If you get hit by a race, retry.
392 hlist_for_each_entry(blkg, n, &blkcg->blkg_list, blkcg_node) {
393 for (i = 0; i < BLKCG_MAX_POLS; i++) {
394 struct blkcg_policy *pol = blkcg_policy[i];
396 if (blkcg_policy_enabled(blkg->q, pol) &&
397 pol->pd_reset_stats_fn)
398 pol->pd_reset_stats_fn(blkg);
402 spin_unlock_irq(&blkcg->lock);
403 mutex_unlock(&blkcg_pol_mutex);
407 static const char *blkg_dev_name(struct blkcg_gq *blkg)
409 /* some drivers (floppy) instantiate a queue w/o disk registered */
410 if (blkg->q->backing_dev_info.dev)
411 return dev_name(blkg->q->backing_dev_info.dev);
416 * blkcg_print_blkgs - helper for printing per-blkg data
417 * @sf: seq_file to print to
418 * @blkcg: blkcg of interest
419 * @prfill: fill function to print out a blkg
420 * @pol: policy in question
421 * @data: data to be passed to @prfill
422 * @show_total: to print out sum of prfill return values or not
424 * This function invokes @prfill on each blkg of @blkcg if pd for the
425 * policy specified by @pol exists. @prfill is invoked with @sf, the
426 * policy data and @data. If @show_total is %true, the sum of the return
427 * values from @prfill is printed with "Total" label at the end.
429 * This is to be used to construct print functions for
430 * cftype->read_seq_string method.
432 void blkcg_print_blkgs(struct seq_file *sf, struct blkcg *blkcg,
433 u64 (*prfill)(struct seq_file *,
434 struct blkg_policy_data *, int),
435 const struct blkcg_policy *pol, int data,
438 struct blkcg_gq *blkg;
439 struct hlist_node *n;
442 spin_lock_irq(&blkcg->lock);
443 hlist_for_each_entry(blkg, n, &blkcg->blkg_list, blkcg_node)
444 if (blkcg_policy_enabled(blkg->q, pol))
445 total += prfill(sf, blkg->pd[pol->plid], data);
446 spin_unlock_irq(&blkcg->lock);
449 seq_printf(sf, "Total %llu\n", (unsigned long long)total);
451 EXPORT_SYMBOL_GPL(blkcg_print_blkgs);
454 * __blkg_prfill_u64 - prfill helper for a single u64 value
455 * @sf: seq_file to print to
456 * @pd: policy private data of interest
459 * Print @v to @sf for the device assocaited with @pd.
461 u64 __blkg_prfill_u64(struct seq_file *sf, struct blkg_policy_data *pd, u64 v)
463 const char *dname = blkg_dev_name(pd->blkg);
468 seq_printf(sf, "%s %llu\n", dname, (unsigned long long)v);
471 EXPORT_SYMBOL_GPL(__blkg_prfill_u64);
474 * __blkg_prfill_rwstat - prfill helper for a blkg_rwstat
475 * @sf: seq_file to print to
476 * @pd: policy private data of interest
477 * @rwstat: rwstat to print
479 * Print @rwstat to @sf for the device assocaited with @pd.
481 u64 __blkg_prfill_rwstat(struct seq_file *sf, struct blkg_policy_data *pd,
482 const struct blkg_rwstat *rwstat)
484 static const char *rwstr[] = {
485 [BLKG_RWSTAT_READ] = "Read",
486 [BLKG_RWSTAT_WRITE] = "Write",
487 [BLKG_RWSTAT_SYNC] = "Sync",
488 [BLKG_RWSTAT_ASYNC] = "Async",
490 const char *dname = blkg_dev_name(pd->blkg);
497 for (i = 0; i < BLKG_RWSTAT_NR; i++)
498 seq_printf(sf, "%s %s %llu\n", dname, rwstr[i],
499 (unsigned long long)rwstat->cnt[i]);
501 v = rwstat->cnt[BLKG_RWSTAT_READ] + rwstat->cnt[BLKG_RWSTAT_WRITE];
502 seq_printf(sf, "%s Total %llu\n", dname, (unsigned long long)v);
507 * blkg_prfill_stat - prfill callback for blkg_stat
508 * @sf: seq_file to print to
509 * @pd: policy private data of interest
510 * @off: offset to the blkg_stat in @pd
512 * prfill callback for printing a blkg_stat.
514 u64 blkg_prfill_stat(struct seq_file *sf, struct blkg_policy_data *pd, int off)
516 return __blkg_prfill_u64(sf, pd, blkg_stat_read((void *)pd + off));
518 EXPORT_SYMBOL_GPL(blkg_prfill_stat);
521 * blkg_prfill_rwstat - prfill callback for blkg_rwstat
522 * @sf: seq_file to print to
523 * @pd: policy private data of interest
524 * @off: offset to the blkg_rwstat in @pd
526 * prfill callback for printing a blkg_rwstat.
528 u64 blkg_prfill_rwstat(struct seq_file *sf, struct blkg_policy_data *pd,
531 struct blkg_rwstat rwstat = blkg_rwstat_read((void *)pd + off);
533 return __blkg_prfill_rwstat(sf, pd, &rwstat);
535 EXPORT_SYMBOL_GPL(blkg_prfill_rwstat);
538 * blkg_conf_prep - parse and prepare for per-blkg config update
539 * @blkcg: target block cgroup
540 * @pol: target policy
541 * @input: input string
542 * @ctx: blkg_conf_ctx to be filled
544 * Parse per-blkg config update from @input and initialize @ctx with the
545 * result. @ctx->blkg points to the blkg to be updated and @ctx->v the new
546 * value. This function returns with RCU read lock and queue lock held and
547 * must be paired with blkg_conf_finish().
549 int blkg_conf_prep(struct blkcg *blkcg, const struct blkcg_policy *pol,
550 const char *input, struct blkg_conf_ctx *ctx)
551 __acquires(rcu) __acquires(disk->queue->queue_lock)
553 struct gendisk *disk;
554 struct blkcg_gq *blkg;
555 unsigned int major, minor;
556 unsigned long long v;
559 if (sscanf(input, "%u:%u %llu", &major, &minor, &v) != 3)
562 disk = get_gendisk(MKDEV(major, minor), &part);
567 spin_lock_irq(disk->queue->queue_lock);
569 if (blkcg_policy_enabled(disk->queue, pol))
570 blkg = blkg_lookup_create(blkcg, disk->queue);
572 blkg = ERR_PTR(-EINVAL);
577 spin_unlock_irq(disk->queue->queue_lock);
580 * If queue was bypassing, we should retry. Do so after a
581 * short msleep(). It isn't strictly necessary but queue
582 * can be bypassing for some time and it's always nice to
583 * avoid busy looping.
587 ret = restart_syscall();
597 EXPORT_SYMBOL_GPL(blkg_conf_prep);
600 * blkg_conf_finish - finish up per-blkg config update
601 * @ctx: blkg_conf_ctx intiailized by blkg_conf_prep()
603 * Finish up after per-blkg config update. This function must be paired
604 * with blkg_conf_prep().
606 void blkg_conf_finish(struct blkg_conf_ctx *ctx)
607 __releases(ctx->disk->queue->queue_lock) __releases(rcu)
609 spin_unlock_irq(ctx->disk->queue->queue_lock);
613 EXPORT_SYMBOL_GPL(blkg_conf_finish);
615 struct cftype blkcg_files[] = {
617 .name = "reset_stats",
618 .write_u64 = blkcg_reset_stats,
624 * blkcg_css_offline - cgroup css_offline callback
625 * @cgroup: cgroup of interest
627 * This function is called when @cgroup is about to go away and responsible
628 * for shooting down all blkgs associated with @cgroup. blkgs should be
629 * removed while holding both q and blkcg locks. As blkcg lock is nested
630 * inside q lock, this function performs reverse double lock dancing.
632 * This is the blkcg counterpart of ioc_release_fn().
634 static void blkcg_css_offline(struct cgroup *cgroup)
636 struct blkcg *blkcg = cgroup_to_blkcg(cgroup);
638 spin_lock_irq(&blkcg->lock);
640 while (!hlist_empty(&blkcg->blkg_list)) {
641 struct blkcg_gq *blkg = hlist_entry(blkcg->blkg_list.first,
642 struct blkcg_gq, blkcg_node);
643 struct request_queue *q = blkg->q;
645 if (spin_trylock(q->queue_lock)) {
647 spin_unlock(q->queue_lock);
649 spin_unlock_irq(&blkcg->lock);
651 spin_lock_irq(&blkcg->lock);
655 spin_unlock_irq(&blkcg->lock);
658 static void blkcg_css_free(struct cgroup *cgroup)
660 struct blkcg *blkcg = cgroup_to_blkcg(cgroup);
662 if (blkcg != &blkcg_root)
666 static struct cgroup_subsys_state *blkcg_css_alloc(struct cgroup *cgroup)
668 static atomic64_t id_seq = ATOMIC64_INIT(0);
670 struct cgroup *parent = cgroup->parent;
677 blkcg = kzalloc(sizeof(*blkcg), GFP_KERNEL);
679 return ERR_PTR(-ENOMEM);
681 blkcg->cfq_weight = CFQ_WEIGHT_DEFAULT;
682 blkcg->id = atomic64_inc_return(&id_seq); /* root is 0, start from 1 */
684 spin_lock_init(&blkcg->lock);
685 INIT_RADIX_TREE(&blkcg->blkg_tree, GFP_ATOMIC);
686 INIT_HLIST_HEAD(&blkcg->blkg_list);
692 * blkcg_init_queue - initialize blkcg part of request queue
693 * @q: request_queue to initialize
695 * Called from blk_alloc_queue_node(). Responsible for initializing blkcg
696 * part of new request_queue @q.
699 * 0 on success, -errno on failure.
701 int blkcg_init_queue(struct request_queue *q)
705 return blk_throtl_init(q);
709 * blkcg_drain_queue - drain blkcg part of request_queue
710 * @q: request_queue to drain
712 * Called from blk_drain_queue(). Responsible for draining blkcg part.
714 void blkcg_drain_queue(struct request_queue *q)
716 lockdep_assert_held(q->queue_lock);
722 * blkcg_exit_queue - exit and release blkcg part of request_queue
723 * @q: request_queue being released
725 * Called from blk_release_queue(). Responsible for exiting blkcg part.
727 void blkcg_exit_queue(struct request_queue *q)
729 spin_lock_irq(q->queue_lock);
731 spin_unlock_irq(q->queue_lock);
737 * We cannot support shared io contexts, as we have no mean to support
738 * two tasks with the same ioc in two different groups without major rework
739 * of the main cic data structures. For now we allow a task to change
740 * its cgroup only if it's the only owner of its ioc.
742 static int blkcg_can_attach(struct cgroup *cgrp, struct cgroup_taskset *tset)
744 struct task_struct *task;
745 struct io_context *ioc;
748 /* task_lock() is needed to avoid races with exit_io_context() */
749 cgroup_taskset_for_each(task, cgrp, tset) {
751 ioc = task->io_context;
752 if (ioc && atomic_read(&ioc->nr_tasks) > 1)
761 struct cgroup_subsys blkio_subsys = {
763 .css_alloc = blkcg_css_alloc,
764 .css_offline = blkcg_css_offline,
765 .css_free = blkcg_css_free,
766 .can_attach = blkcg_can_attach,
767 .subsys_id = blkio_subsys_id,
768 .base_cftypes = blkcg_files,
769 .module = THIS_MODULE,
772 * blkio subsystem is utterly broken in terms of hierarchy support.
773 * It treats all cgroups equally regardless of where they're
774 * located in the hierarchy - all cgroups are treated as if they're
775 * right below the root. Fix it and remove the following.
777 .broken_hierarchy = true,
779 EXPORT_SYMBOL_GPL(blkio_subsys);
782 * blkcg_activate_policy - activate a blkcg policy on a request_queue
783 * @q: request_queue of interest
784 * @pol: blkcg policy to activate
786 * Activate @pol on @q. Requires %GFP_KERNEL context. @q goes through
787 * bypass mode to populate its blkgs with policy_data for @pol.
789 * Activation happens with @q bypassed, so nobody would be accessing blkgs
790 * from IO path. Update of each blkg is protected by both queue and blkcg
791 * locks so that holding either lock and testing blkcg_policy_enabled() is
792 * always enough for dereferencing policy data.
794 * The caller is responsible for synchronizing [de]activations and policy
795 * [un]registerations. Returns 0 on success, -errno on failure.
797 int blkcg_activate_policy(struct request_queue *q,
798 const struct blkcg_policy *pol)
801 struct blkcg_gq *blkg, *new_blkg;
802 struct blkg_policy_data *pd, *n;
806 if (blkcg_policy_enabled(q, pol))
809 /* preallocations for root blkg */
810 new_blkg = blkg_alloc(&blkcg_root, q, GFP_KERNEL);
814 preloaded = !radix_tree_preload(GFP_KERNEL);
816 blk_queue_bypass_start(q);
819 * Make sure the root blkg exists and count the existing blkgs. As
820 * @q is bypassing at this point, blkg_lookup_create() can't be
821 * used. Open code it.
823 spin_lock_irq(q->queue_lock);
826 blkg = __blkg_lookup(&blkcg_root, q, false);
830 blkg = blkg_create(&blkcg_root, q, new_blkg);
834 radix_tree_preload_end();
841 q->root_rl.blkg = blkg;
843 list_for_each_entry(blkg, &q->blkg_list, q_node)
846 spin_unlock_irq(q->queue_lock);
848 /* allocate policy_data for all existing blkgs */
850 pd = kzalloc_node(pol->pd_size, GFP_KERNEL, q->node);
855 list_add_tail(&pd->alloc_node, &pds);
859 * Install the allocated pds. With @q bypassing, no new blkg
860 * should have been created while the queue lock was dropped.
862 spin_lock_irq(q->queue_lock);
864 list_for_each_entry(blkg, &q->blkg_list, q_node) {
865 if (WARN_ON(list_empty(&pds))) {
866 /* umm... this shouldn't happen, just abort */
870 pd = list_first_entry(&pds, struct blkg_policy_data, alloc_node);
871 list_del_init(&pd->alloc_node);
873 /* grab blkcg lock too while installing @pd on @blkg */
874 spin_lock(&blkg->blkcg->lock);
876 blkg->pd[pol->plid] = pd;
878 pol->pd_init_fn(blkg);
880 spin_unlock(&blkg->blkcg->lock);
883 __set_bit(pol->plid, q->blkcg_pols);
886 spin_unlock_irq(q->queue_lock);
888 blk_queue_bypass_end(q);
889 list_for_each_entry_safe(pd, n, &pds, alloc_node)
893 EXPORT_SYMBOL_GPL(blkcg_activate_policy);
896 * blkcg_deactivate_policy - deactivate a blkcg policy on a request_queue
897 * @q: request_queue of interest
898 * @pol: blkcg policy to deactivate
900 * Deactivate @pol on @q. Follows the same synchronization rules as
901 * blkcg_activate_policy().
903 void blkcg_deactivate_policy(struct request_queue *q,
904 const struct blkcg_policy *pol)
906 struct blkcg_gq *blkg;
908 if (!blkcg_policy_enabled(q, pol))
911 blk_queue_bypass_start(q);
912 spin_lock_irq(q->queue_lock);
914 __clear_bit(pol->plid, q->blkcg_pols);
916 /* if no policy is left, no need for blkgs - shoot them down */
917 if (bitmap_empty(q->blkcg_pols, BLKCG_MAX_POLS))
920 list_for_each_entry(blkg, &q->blkg_list, q_node) {
921 /* grab blkcg lock too while removing @pd from @blkg */
922 spin_lock(&blkg->blkcg->lock);
925 pol->pd_exit_fn(blkg);
927 kfree(blkg->pd[pol->plid]);
928 blkg->pd[pol->plid] = NULL;
930 spin_unlock(&blkg->blkcg->lock);
933 spin_unlock_irq(q->queue_lock);
934 blk_queue_bypass_end(q);
936 EXPORT_SYMBOL_GPL(blkcg_deactivate_policy);
939 * blkcg_policy_register - register a blkcg policy
940 * @pol: blkcg policy to register
942 * Register @pol with blkcg core. Might sleep and @pol may be modified on
943 * successful registration. Returns 0 on success and -errno on failure.
945 int blkcg_policy_register(struct blkcg_policy *pol)
949 if (WARN_ON(pol->pd_size < sizeof(struct blkg_policy_data)))
952 mutex_lock(&blkcg_pol_mutex);
954 /* find an empty slot */
956 for (i = 0; i < BLKCG_MAX_POLS; i++)
957 if (!blkcg_policy[i])
959 if (i >= BLKCG_MAX_POLS)
962 /* register and update blkgs */
964 blkcg_policy[i] = pol;
966 /* everything is in place, add intf files for the new policy */
968 WARN_ON(cgroup_add_cftypes(&blkio_subsys, pol->cftypes));
971 mutex_unlock(&blkcg_pol_mutex);
974 EXPORT_SYMBOL_GPL(blkcg_policy_register);
977 * blkcg_policy_unregister - unregister a blkcg policy
978 * @pol: blkcg policy to unregister
980 * Undo blkcg_policy_register(@pol). Might sleep.
982 void blkcg_policy_unregister(struct blkcg_policy *pol)
984 mutex_lock(&blkcg_pol_mutex);
986 if (WARN_ON(blkcg_policy[pol->plid] != pol))
989 /* kill the intf files first */
991 cgroup_rm_cftypes(&blkio_subsys, pol->cftypes);
993 /* unregister and update blkgs */
994 blkcg_policy[pol->plid] = NULL;
996 mutex_unlock(&blkcg_pol_mutex);
998 EXPORT_SYMBOL_GPL(blkcg_policy_unregister);