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)) {
196 new_blkg = blkg_alloc(blkcg, q, GFP_ATOMIC);
197 if (unlikely(!new_blkg)) {
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);
217 css_put(&blkcg->css);
224 * blkg_lookup_create - lookup blkg, try to create one if not there
225 * @blkcg: blkcg of interest
226 * @q: request_queue of interest
228 * Lookup blkg for the @blkcg - @q pair. If it doesn't exist, try to
229 * create one. This function should be called under RCU read lock and
232 * Returns pointer to the looked up or created blkg on success, ERR_PTR()
233 * value on error. If @q is dead, returns ERR_PTR(-EINVAL). If @q is not
234 * dead and bypassing, returns ERR_PTR(-EBUSY).
236 struct blkcg_gq *blkg_lookup_create(struct blkcg *blkcg,
237 struct request_queue *q)
239 struct blkcg_gq *blkg;
241 WARN_ON_ONCE(!rcu_read_lock_held());
242 lockdep_assert_held(q->queue_lock);
245 * This could be the first entry point of blkcg implementation and
246 * we shouldn't allow anything to go through for a bypassing queue.
248 if (unlikely(blk_queue_bypass(q)))
249 return ERR_PTR(blk_queue_dying(q) ? -EINVAL : -EBUSY);
251 blkg = __blkg_lookup(blkcg, q, true);
255 return blkg_create(blkcg, q, NULL);
257 EXPORT_SYMBOL_GPL(blkg_lookup_create);
259 static void blkg_destroy(struct blkcg_gq *blkg)
261 struct blkcg *blkcg = blkg->blkcg;
263 lockdep_assert_held(blkg->q->queue_lock);
264 lockdep_assert_held(&blkcg->lock);
266 /* Something wrong if we are trying to remove same group twice */
267 WARN_ON_ONCE(list_empty(&blkg->q_node));
268 WARN_ON_ONCE(hlist_unhashed(&blkg->blkcg_node));
270 radix_tree_delete(&blkcg->blkg_tree, blkg->q->id);
271 list_del_init(&blkg->q_node);
272 hlist_del_init_rcu(&blkg->blkcg_node);
275 * Both setting lookup hint to and clearing it from @blkg are done
276 * under queue_lock. If it's not pointing to @blkg now, it never
277 * will. Hint assignment itself can race safely.
279 if (rcu_dereference_raw(blkcg->blkg_hint) == blkg)
280 rcu_assign_pointer(blkcg->blkg_hint, NULL);
283 * Put the reference taken at the time of creation so that when all
284 * queues are gone, group can be destroyed.
290 * blkg_destroy_all - destroy all blkgs associated with a request_queue
291 * @q: request_queue of interest
293 * Destroy all blkgs associated with @q.
295 static void blkg_destroy_all(struct request_queue *q)
297 struct blkcg_gq *blkg, *n;
299 lockdep_assert_held(q->queue_lock);
301 list_for_each_entry_safe(blkg, n, &q->blkg_list, q_node) {
302 struct blkcg *blkcg = blkg->blkcg;
304 spin_lock(&blkcg->lock);
306 spin_unlock(&blkcg->lock);
310 * root blkg is destroyed. Just clear the pointer since
311 * root_rl does not take reference on root blkg.
314 q->root_rl.blkg = NULL;
317 static void blkg_rcu_free(struct rcu_head *rcu_head)
319 blkg_free(container_of(rcu_head, struct blkcg_gq, rcu_head));
322 void __blkg_release(struct blkcg_gq *blkg)
324 /* release the extra blkcg reference this blkg has been holding */
325 css_put(&blkg->blkcg->css);
328 * A group is freed in rcu manner. But having an rcu lock does not
329 * mean that one can access all the fields of blkg and assume these
330 * are valid. For example, don't try to follow throtl_data and
331 * request queue links.
333 * Having a reference to blkg under an rcu allows acess to only
334 * values local to groups like group stats and group rate limits
336 call_rcu(&blkg->rcu_head, blkg_rcu_free);
338 EXPORT_SYMBOL_GPL(__blkg_release);
341 * The next function used by blk_queue_for_each_rl(). It's a bit tricky
342 * because the root blkg uses @q->root_rl instead of its own rl.
344 struct request_list *__blk_queue_next_rl(struct request_list *rl,
345 struct request_queue *q)
347 struct list_head *ent;
348 struct blkcg_gq *blkg;
351 * Determine the current blkg list_head. The first entry is
352 * root_rl which is off @q->blkg_list and mapped to the head.
354 if (rl == &q->root_rl) {
356 /* There are no more block groups, hence no request lists */
360 blkg = container_of(rl, struct blkcg_gq, rl);
364 /* walk to the next list_head, skip root blkcg */
366 if (ent == &q->root_blkg->q_node)
368 if (ent == &q->blkg_list)
371 blkg = container_of(ent, struct blkcg_gq, q_node);
375 static int blkcg_reset_stats(struct cgroup *cgroup, struct cftype *cftype,
378 struct blkcg *blkcg = cgroup_to_blkcg(cgroup);
379 struct blkcg_gq *blkg;
380 struct hlist_node *n;
383 mutex_lock(&blkcg_pol_mutex);
384 spin_lock_irq(&blkcg->lock);
387 * Note that stat reset is racy - it doesn't synchronize against
388 * stat updates. This is a debug feature which shouldn't exist
389 * anyway. If you get hit by a race, retry.
391 hlist_for_each_entry(blkg, n, &blkcg->blkg_list, blkcg_node) {
392 for (i = 0; i < BLKCG_MAX_POLS; i++) {
393 struct blkcg_policy *pol = blkcg_policy[i];
395 if (blkcg_policy_enabled(blkg->q, pol) &&
396 pol->pd_reset_stats_fn)
397 pol->pd_reset_stats_fn(blkg);
401 spin_unlock_irq(&blkcg->lock);
402 mutex_unlock(&blkcg_pol_mutex);
406 static const char *blkg_dev_name(struct blkcg_gq *blkg)
408 /* some drivers (floppy) instantiate a queue w/o disk registered */
409 if (blkg->q->backing_dev_info.dev)
410 return dev_name(blkg->q->backing_dev_info.dev);
415 * blkcg_print_blkgs - helper for printing per-blkg data
416 * @sf: seq_file to print to
417 * @blkcg: blkcg of interest
418 * @prfill: fill function to print out a blkg
419 * @pol: policy in question
420 * @data: data to be passed to @prfill
421 * @show_total: to print out sum of prfill return values or not
423 * This function invokes @prfill on each blkg of @blkcg if pd for the
424 * policy specified by @pol exists. @prfill is invoked with @sf, the
425 * policy data and @data. If @show_total is %true, the sum of the return
426 * values from @prfill is printed with "Total" label at the end.
428 * This is to be used to construct print functions for
429 * cftype->read_seq_string method.
431 void blkcg_print_blkgs(struct seq_file *sf, struct blkcg *blkcg,
432 u64 (*prfill)(struct seq_file *,
433 struct blkg_policy_data *, int),
434 const struct blkcg_policy *pol, int data,
437 struct blkcg_gq *blkg;
438 struct hlist_node *n;
441 spin_lock_irq(&blkcg->lock);
442 hlist_for_each_entry(blkg, n, &blkcg->blkg_list, blkcg_node)
443 if (blkcg_policy_enabled(blkg->q, pol))
444 total += prfill(sf, blkg->pd[pol->plid], data);
445 spin_unlock_irq(&blkcg->lock);
448 seq_printf(sf, "Total %llu\n", (unsigned long long)total);
450 EXPORT_SYMBOL_GPL(blkcg_print_blkgs);
453 * __blkg_prfill_u64 - prfill helper for a single u64 value
454 * @sf: seq_file to print to
455 * @pd: policy private data of interest
458 * Print @v to @sf for the device assocaited with @pd.
460 u64 __blkg_prfill_u64(struct seq_file *sf, struct blkg_policy_data *pd, u64 v)
462 const char *dname = blkg_dev_name(pd->blkg);
467 seq_printf(sf, "%s %llu\n", dname, (unsigned long long)v);
470 EXPORT_SYMBOL_GPL(__blkg_prfill_u64);
473 * __blkg_prfill_rwstat - prfill helper for a blkg_rwstat
474 * @sf: seq_file to print to
475 * @pd: policy private data of interest
476 * @rwstat: rwstat to print
478 * Print @rwstat to @sf for the device assocaited with @pd.
480 u64 __blkg_prfill_rwstat(struct seq_file *sf, struct blkg_policy_data *pd,
481 const struct blkg_rwstat *rwstat)
483 static const char *rwstr[] = {
484 [BLKG_RWSTAT_READ] = "Read",
485 [BLKG_RWSTAT_WRITE] = "Write",
486 [BLKG_RWSTAT_SYNC] = "Sync",
487 [BLKG_RWSTAT_ASYNC] = "Async",
489 const char *dname = blkg_dev_name(pd->blkg);
496 for (i = 0; i < BLKG_RWSTAT_NR; i++)
497 seq_printf(sf, "%s %s %llu\n", dname, rwstr[i],
498 (unsigned long long)rwstat->cnt[i]);
500 v = rwstat->cnt[BLKG_RWSTAT_READ] + rwstat->cnt[BLKG_RWSTAT_WRITE];
501 seq_printf(sf, "%s Total %llu\n", dname, (unsigned long long)v);
506 * blkg_prfill_stat - prfill callback for blkg_stat
507 * @sf: seq_file to print to
508 * @pd: policy private data of interest
509 * @off: offset to the blkg_stat in @pd
511 * prfill callback for printing a blkg_stat.
513 u64 blkg_prfill_stat(struct seq_file *sf, struct blkg_policy_data *pd, int off)
515 return __blkg_prfill_u64(sf, pd, blkg_stat_read((void *)pd + off));
517 EXPORT_SYMBOL_GPL(blkg_prfill_stat);
520 * blkg_prfill_rwstat - prfill callback for blkg_rwstat
521 * @sf: seq_file to print to
522 * @pd: policy private data of interest
523 * @off: offset to the blkg_rwstat in @pd
525 * prfill callback for printing a blkg_rwstat.
527 u64 blkg_prfill_rwstat(struct seq_file *sf, struct blkg_policy_data *pd,
530 struct blkg_rwstat rwstat = blkg_rwstat_read((void *)pd + off);
532 return __blkg_prfill_rwstat(sf, pd, &rwstat);
534 EXPORT_SYMBOL_GPL(blkg_prfill_rwstat);
537 * blkg_conf_prep - parse and prepare for per-blkg config update
538 * @blkcg: target block cgroup
539 * @pol: target policy
540 * @input: input string
541 * @ctx: blkg_conf_ctx to be filled
543 * Parse per-blkg config update from @input and initialize @ctx with the
544 * result. @ctx->blkg points to the blkg to be updated and @ctx->v the new
545 * value. This function returns with RCU read lock and queue lock held and
546 * must be paired with blkg_conf_finish().
548 int blkg_conf_prep(struct blkcg *blkcg, const struct blkcg_policy *pol,
549 const char *input, struct blkg_conf_ctx *ctx)
550 __acquires(rcu) __acquires(disk->queue->queue_lock)
552 struct gendisk *disk;
553 struct blkcg_gq *blkg;
554 unsigned int major, minor;
555 unsigned long long v;
558 if (sscanf(input, "%u:%u %llu", &major, &minor, &v) != 3)
561 disk = get_gendisk(MKDEV(major, minor), &part);
566 spin_lock_irq(disk->queue->queue_lock);
568 if (blkcg_policy_enabled(disk->queue, pol))
569 blkg = blkg_lookup_create(blkcg, disk->queue);
571 blkg = ERR_PTR(-EINVAL);
576 spin_unlock_irq(disk->queue->queue_lock);
579 * If queue was bypassing, we should retry. Do so after a
580 * short msleep(). It isn't strictly necessary but queue
581 * can be bypassing for some time and it's always nice to
582 * avoid busy looping.
586 ret = restart_syscall();
596 EXPORT_SYMBOL_GPL(blkg_conf_prep);
599 * blkg_conf_finish - finish up per-blkg config update
600 * @ctx: blkg_conf_ctx intiailized by blkg_conf_prep()
602 * Finish up after per-blkg config update. This function must be paired
603 * with blkg_conf_prep().
605 void blkg_conf_finish(struct blkg_conf_ctx *ctx)
606 __releases(ctx->disk->queue->queue_lock) __releases(rcu)
608 spin_unlock_irq(ctx->disk->queue->queue_lock);
612 EXPORT_SYMBOL_GPL(blkg_conf_finish);
614 struct cftype blkcg_files[] = {
616 .name = "reset_stats",
617 .write_u64 = blkcg_reset_stats,
623 * blkcg_css_offline - cgroup css_offline callback
624 * @cgroup: cgroup of interest
626 * This function is called when @cgroup is about to go away and responsible
627 * for shooting down all blkgs associated with @cgroup. blkgs should be
628 * removed while holding both q and blkcg locks. As blkcg lock is nested
629 * inside q lock, this function performs reverse double lock dancing.
631 * This is the blkcg counterpart of ioc_release_fn().
633 static void blkcg_css_offline(struct cgroup *cgroup)
635 struct blkcg *blkcg = cgroup_to_blkcg(cgroup);
637 spin_lock_irq(&blkcg->lock);
639 while (!hlist_empty(&blkcg->blkg_list)) {
640 struct blkcg_gq *blkg = hlist_entry(blkcg->blkg_list.first,
641 struct blkcg_gq, blkcg_node);
642 struct request_queue *q = blkg->q;
644 if (spin_trylock(q->queue_lock)) {
646 spin_unlock(q->queue_lock);
648 spin_unlock_irq(&blkcg->lock);
650 spin_lock_irq(&blkcg->lock);
654 spin_unlock_irq(&blkcg->lock);
657 static void blkcg_css_free(struct cgroup *cgroup)
659 struct blkcg *blkcg = cgroup_to_blkcg(cgroup);
661 if (blkcg != &blkcg_root)
665 static struct cgroup_subsys_state *blkcg_css_alloc(struct cgroup *cgroup)
667 static atomic64_t id_seq = ATOMIC64_INIT(0);
669 struct cgroup *parent = cgroup->parent;
676 blkcg = kzalloc(sizeof(*blkcg), GFP_KERNEL);
678 return ERR_PTR(-ENOMEM);
680 blkcg->cfq_weight = CFQ_WEIGHT_DEFAULT;
681 blkcg->id = atomic64_inc_return(&id_seq); /* root is 0, start from 1 */
683 spin_lock_init(&blkcg->lock);
684 INIT_RADIX_TREE(&blkcg->blkg_tree, GFP_ATOMIC);
685 INIT_HLIST_HEAD(&blkcg->blkg_list);
691 * blkcg_init_queue - initialize blkcg part of request queue
692 * @q: request_queue to initialize
694 * Called from blk_alloc_queue_node(). Responsible for initializing blkcg
695 * part of new request_queue @q.
698 * 0 on success, -errno on failure.
700 int blkcg_init_queue(struct request_queue *q)
704 return blk_throtl_init(q);
708 * blkcg_drain_queue - drain blkcg part of request_queue
709 * @q: request_queue to drain
711 * Called from blk_drain_queue(). Responsible for draining blkcg part.
713 void blkcg_drain_queue(struct request_queue *q)
715 lockdep_assert_held(q->queue_lock);
721 * blkcg_exit_queue - exit and release blkcg part of request_queue
722 * @q: request_queue being released
724 * Called from blk_release_queue(). Responsible for exiting blkcg part.
726 void blkcg_exit_queue(struct request_queue *q)
728 spin_lock_irq(q->queue_lock);
730 spin_unlock_irq(q->queue_lock);
736 * We cannot support shared io contexts, as we have no mean to support
737 * two tasks with the same ioc in two different groups without major rework
738 * of the main cic data structures. For now we allow a task to change
739 * its cgroup only if it's the only owner of its ioc.
741 static int blkcg_can_attach(struct cgroup *cgrp, struct cgroup_taskset *tset)
743 struct task_struct *task;
744 struct io_context *ioc;
747 /* task_lock() is needed to avoid races with exit_io_context() */
748 cgroup_taskset_for_each(task, cgrp, tset) {
750 ioc = task->io_context;
751 if (ioc && atomic_read(&ioc->nr_tasks) > 1)
760 struct cgroup_subsys blkio_subsys = {
762 .css_alloc = blkcg_css_alloc,
763 .css_offline = blkcg_css_offline,
764 .css_free = blkcg_css_free,
765 .can_attach = blkcg_can_attach,
766 .subsys_id = blkio_subsys_id,
767 .base_cftypes = blkcg_files,
768 .module = THIS_MODULE,
771 * blkio subsystem is utterly broken in terms of hierarchy support.
772 * It treats all cgroups equally regardless of where they're
773 * located in the hierarchy - all cgroups are treated as if they're
774 * right below the root. Fix it and remove the following.
776 .broken_hierarchy = true,
778 EXPORT_SYMBOL_GPL(blkio_subsys);
781 * blkcg_activate_policy - activate a blkcg policy on a request_queue
782 * @q: request_queue of interest
783 * @pol: blkcg policy to activate
785 * Activate @pol on @q. Requires %GFP_KERNEL context. @q goes through
786 * bypass mode to populate its blkgs with policy_data for @pol.
788 * Activation happens with @q bypassed, so nobody would be accessing blkgs
789 * from IO path. Update of each blkg is protected by both queue and blkcg
790 * locks so that holding either lock and testing blkcg_policy_enabled() is
791 * always enough for dereferencing policy data.
793 * The caller is responsible for synchronizing [de]activations and policy
794 * [un]registerations. Returns 0 on success, -errno on failure.
796 int blkcg_activate_policy(struct request_queue *q,
797 const struct blkcg_policy *pol)
800 struct blkcg_gq *blkg, *new_blkg;
801 struct blkg_policy_data *pd, *n;
805 if (blkcg_policy_enabled(q, pol))
808 /* preallocations for root blkg */
809 new_blkg = blkg_alloc(&blkcg_root, q, GFP_KERNEL);
813 preloaded = !radix_tree_preload(GFP_KERNEL);
815 blk_queue_bypass_start(q);
818 * Make sure the root blkg exists and count the existing blkgs. As
819 * @q is bypassing at this point, blkg_lookup_create() can't be
820 * used. Open code it.
822 spin_lock_irq(q->queue_lock);
825 blkg = __blkg_lookup(&blkcg_root, q, false);
829 blkg = blkg_create(&blkcg_root, q, new_blkg);
833 radix_tree_preload_end();
840 q->root_rl.blkg = blkg;
842 list_for_each_entry(blkg, &q->blkg_list, q_node)
845 spin_unlock_irq(q->queue_lock);
847 /* allocate policy_data for all existing blkgs */
849 pd = kzalloc_node(pol->pd_size, GFP_KERNEL, q->node);
854 list_add_tail(&pd->alloc_node, &pds);
858 * Install the allocated pds. With @q bypassing, no new blkg
859 * should have been created while the queue lock was dropped.
861 spin_lock_irq(q->queue_lock);
863 list_for_each_entry(blkg, &q->blkg_list, q_node) {
864 if (WARN_ON(list_empty(&pds))) {
865 /* umm... this shouldn't happen, just abort */
869 pd = list_first_entry(&pds, struct blkg_policy_data, alloc_node);
870 list_del_init(&pd->alloc_node);
872 /* grab blkcg lock too while installing @pd on @blkg */
873 spin_lock(&blkg->blkcg->lock);
875 blkg->pd[pol->plid] = pd;
877 pol->pd_init_fn(blkg);
879 spin_unlock(&blkg->blkcg->lock);
882 __set_bit(pol->plid, q->blkcg_pols);
885 spin_unlock_irq(q->queue_lock);
887 blk_queue_bypass_end(q);
888 list_for_each_entry_safe(pd, n, &pds, alloc_node)
892 EXPORT_SYMBOL_GPL(blkcg_activate_policy);
895 * blkcg_deactivate_policy - deactivate a blkcg policy on a request_queue
896 * @q: request_queue of interest
897 * @pol: blkcg policy to deactivate
899 * Deactivate @pol on @q. Follows the same synchronization rules as
900 * blkcg_activate_policy().
902 void blkcg_deactivate_policy(struct request_queue *q,
903 const struct blkcg_policy *pol)
905 struct blkcg_gq *blkg;
907 if (!blkcg_policy_enabled(q, pol))
910 blk_queue_bypass_start(q);
911 spin_lock_irq(q->queue_lock);
913 __clear_bit(pol->plid, q->blkcg_pols);
915 /* if no policy is left, no need for blkgs - shoot them down */
916 if (bitmap_empty(q->blkcg_pols, BLKCG_MAX_POLS))
919 list_for_each_entry(blkg, &q->blkg_list, q_node) {
920 /* grab blkcg lock too while removing @pd from @blkg */
921 spin_lock(&blkg->blkcg->lock);
924 pol->pd_exit_fn(blkg);
926 kfree(blkg->pd[pol->plid]);
927 blkg->pd[pol->plid] = NULL;
929 spin_unlock(&blkg->blkcg->lock);
932 spin_unlock_irq(q->queue_lock);
933 blk_queue_bypass_end(q);
935 EXPORT_SYMBOL_GPL(blkcg_deactivate_policy);
938 * blkcg_policy_register - register a blkcg policy
939 * @pol: blkcg policy to register
941 * Register @pol with blkcg core. Might sleep and @pol may be modified on
942 * successful registration. Returns 0 on success and -errno on failure.
944 int blkcg_policy_register(struct blkcg_policy *pol)
948 if (WARN_ON(pol->pd_size < sizeof(struct blkg_policy_data)))
951 mutex_lock(&blkcg_pol_mutex);
953 /* find an empty slot */
955 for (i = 0; i < BLKCG_MAX_POLS; i++)
956 if (!blkcg_policy[i])
958 if (i >= BLKCG_MAX_POLS)
961 /* register and update blkgs */
963 blkcg_policy[i] = pol;
965 /* everything is in place, add intf files for the new policy */
967 WARN_ON(cgroup_add_cftypes(&blkio_subsys, pol->cftypes));
970 mutex_unlock(&blkcg_pol_mutex);
973 EXPORT_SYMBOL_GPL(blkcg_policy_register);
976 * blkcg_policy_unregister - unregister a blkcg policy
977 * @pol: blkcg policy to unregister
979 * Undo blkcg_policy_register(@pol). Might sleep.
981 void blkcg_policy_unregister(struct blkcg_policy *pol)
983 mutex_lock(&blkcg_pol_mutex);
985 if (WARN_ON(blkcg_policy[pol->plid] != pol))
988 /* kill the intf files first */
990 cgroup_rm_cftypes(&blkio_subsys, pol->cftypes);
992 /* unregister and update blkgs */
993 blkcg_policy[pol->plid] = NULL;
995 mutex_unlock(&blkcg_pol_mutex);
997 EXPORT_SYMBOL_GPL(blkcg_policy_unregister);