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 struct blkcg *cgroup_to_blkcg(struct cgroup *cgroup)
36 return container_of(cgroup_subsys_state(cgroup, blkio_subsys_id),
39 EXPORT_SYMBOL_GPL(cgroup_to_blkcg);
41 static struct blkcg *task_blkcg(struct task_struct *tsk)
43 return container_of(task_subsys_state(tsk, blkio_subsys_id),
47 struct blkcg *bio_blkcg(struct bio *bio)
49 if (bio && bio->bi_css)
50 return container_of(bio->bi_css, struct blkcg, css);
51 return task_blkcg(current);
53 EXPORT_SYMBOL_GPL(bio_blkcg);
55 static bool blkcg_policy_enabled(struct request_queue *q,
56 const struct blkcg_policy *pol)
58 return pol && test_bit(pol->plid, q->blkcg_pols);
62 * blkg_free - free a blkg
65 * Free @blkg which may be partially allocated.
67 static void blkg_free(struct blkcg_gq *blkg)
74 for (i = 0; i < BLKCG_MAX_POLS; i++) {
75 struct blkcg_policy *pol = blkcg_policy[i];
76 struct blkg_policy_data *pd = blkg->pd[i];
81 if (pol && pol->pd_exit_fn)
82 pol->pd_exit_fn(blkg);
91 * blkg_alloc - allocate a blkg
92 * @blkcg: block cgroup the new blkg is associated with
93 * @q: request_queue the new blkg is associated with
95 * Allocate a new blkg assocating @blkcg and @q.
97 static struct blkcg_gq *blkg_alloc(struct blkcg *blkcg, struct request_queue *q)
99 struct blkcg_gq *blkg;
102 /* alloc and init base part */
103 blkg = kzalloc_node(sizeof(*blkg), GFP_ATOMIC, q->node);
108 INIT_LIST_HEAD(&blkg->q_node);
112 for (i = 0; i < BLKCG_MAX_POLS; i++) {
113 struct blkcg_policy *pol = blkcg_policy[i];
114 struct blkg_policy_data *pd;
116 if (!blkcg_policy_enabled(q, pol))
119 /* alloc per-policy data and attach it to blkg */
120 pd = kzalloc_node(pol->pd_size, GFP_ATOMIC, q->node);
130 /* invoke per-policy init */
131 for (i = 0; i < BLKCG_MAX_POLS; i++) {
132 struct blkcg_policy *pol = blkcg_policy[i];
134 if (blkcg_policy_enabled(blkg->q, pol))
135 pol->pd_init_fn(blkg);
141 static struct blkcg_gq *__blkg_lookup(struct blkcg *blkcg,
142 struct request_queue *q)
144 struct blkcg_gq *blkg;
146 blkg = rcu_dereference(blkcg->blkg_hint);
147 if (blkg && blkg->q == q)
151 * Hint didn't match. Look up from the radix tree. Note that we
152 * may not be holding queue_lock and thus are not sure whether
153 * @blkg from blkg_tree has already been removed or not, so we
154 * can't update hint to the lookup result. Leave it to the caller.
156 blkg = radix_tree_lookup(&blkcg->blkg_tree, q->id);
157 if (blkg && blkg->q == q)
164 * blkg_lookup - lookup blkg for the specified blkcg - q pair
165 * @blkcg: blkcg of interest
166 * @q: request_queue of interest
168 * Lookup blkg for the @blkcg - @q pair. This function should be called
169 * under RCU read lock and is guaranteed to return %NULL if @q is bypassing
170 * - see blk_queue_bypass_start() for details.
172 struct blkcg_gq *blkg_lookup(struct blkcg *blkcg, struct request_queue *q)
174 WARN_ON_ONCE(!rcu_read_lock_held());
176 if (unlikely(blk_queue_bypass(q)))
178 return __blkg_lookup(blkcg, q);
180 EXPORT_SYMBOL_GPL(blkg_lookup);
182 static struct blkcg_gq *__blkg_lookup_create(struct blkcg *blkcg,
183 struct request_queue *q)
184 __releases(q->queue_lock) __acquires(q->queue_lock)
186 struct blkcg_gq *blkg;
189 WARN_ON_ONCE(!rcu_read_lock_held());
190 lockdep_assert_held(q->queue_lock);
192 /* lookup and update hint on success, see __blkg_lookup() for details */
193 blkg = __blkg_lookup(blkcg, q);
195 rcu_assign_pointer(blkcg->blkg_hint, blkg);
199 /* blkg holds a reference to blkcg */
200 if (!css_tryget(&blkcg->css))
201 return ERR_PTR(-EINVAL);
205 blkg = blkg_alloc(blkcg, q);
210 ret = radix_tree_preload(GFP_ATOMIC);
214 spin_lock(&blkcg->lock);
215 ret = radix_tree_insert(&blkcg->blkg_tree, q->id, blkg);
217 hlist_add_head_rcu(&blkg->blkcg_node, &blkcg->blkg_list);
218 list_add(&blkg->q_node, &q->blkg_list);
220 spin_unlock(&blkcg->lock);
222 radix_tree_preload_end();
229 css_put(&blkcg->css);
233 struct blkcg_gq *blkg_lookup_create(struct blkcg *blkcg,
234 struct request_queue *q)
237 * This could be the first entry point of blkcg implementation and
238 * we shouldn't allow anything to go through for a bypassing queue.
240 if (unlikely(blk_queue_bypass(q)))
241 return ERR_PTR(blk_queue_dead(q) ? -EINVAL : -EBUSY);
242 return __blkg_lookup_create(blkcg, q);
244 EXPORT_SYMBOL_GPL(blkg_lookup_create);
246 static void blkg_destroy(struct blkcg_gq *blkg)
248 struct request_queue *q = blkg->q;
249 struct blkcg *blkcg = blkg->blkcg;
251 lockdep_assert_held(q->queue_lock);
252 lockdep_assert_held(&blkcg->lock);
254 /* Something wrong if we are trying to remove same group twice */
255 WARN_ON_ONCE(list_empty(&blkg->q_node));
256 WARN_ON_ONCE(hlist_unhashed(&blkg->blkcg_node));
258 radix_tree_delete(&blkcg->blkg_tree, blkg->q->id);
259 list_del_init(&blkg->q_node);
260 hlist_del_init_rcu(&blkg->blkcg_node);
263 * Both setting lookup hint to and clearing it from @blkg are done
264 * under queue_lock. If it's not pointing to @blkg now, it never
265 * will. Hint assignment itself can race safely.
267 if (rcu_dereference_raw(blkcg->blkg_hint) == blkg)
268 rcu_assign_pointer(blkcg->blkg_hint, NULL);
271 * Put the reference taken at the time of creation so that when all
272 * queues are gone, group can be destroyed.
278 * blkg_destroy_all - destroy all blkgs associated with a request_queue
279 * @q: request_queue of interest
281 * Destroy all blkgs associated with @q.
283 static void blkg_destroy_all(struct request_queue *q)
285 struct blkcg_gq *blkg, *n;
287 lockdep_assert_held(q->queue_lock);
289 list_for_each_entry_safe(blkg, n, &q->blkg_list, q_node) {
290 struct blkcg *blkcg = blkg->blkcg;
292 spin_lock(&blkcg->lock);
294 spin_unlock(&blkcg->lock);
298 static void blkg_rcu_free(struct rcu_head *rcu_head)
300 blkg_free(container_of(rcu_head, struct blkcg_gq, rcu_head));
303 void __blkg_release(struct blkcg_gq *blkg)
305 /* release the extra blkcg reference this blkg has been holding */
306 css_put(&blkg->blkcg->css);
309 * A group is freed in rcu manner. But having an rcu lock does not
310 * mean that one can access all the fields of blkg and assume these
311 * are valid. For example, don't try to follow throtl_data and
312 * request queue links.
314 * Having a reference to blkg under an rcu allows acess to only
315 * values local to groups like group stats and group rate limits
317 call_rcu(&blkg->rcu_head, blkg_rcu_free);
319 EXPORT_SYMBOL_GPL(__blkg_release);
321 static int blkcg_reset_stats(struct cgroup *cgroup, struct cftype *cftype,
324 struct blkcg *blkcg = cgroup_to_blkcg(cgroup);
325 struct blkcg_gq *blkg;
326 struct hlist_node *n;
329 mutex_lock(&blkcg_pol_mutex);
330 spin_lock_irq(&blkcg->lock);
333 * Note that stat reset is racy - it doesn't synchronize against
334 * stat updates. This is a debug feature which shouldn't exist
335 * anyway. If you get hit by a race, retry.
337 hlist_for_each_entry(blkg, n, &blkcg->blkg_list, blkcg_node) {
338 for (i = 0; i < BLKCG_MAX_POLS; i++) {
339 struct blkcg_policy *pol = blkcg_policy[i];
341 if (blkcg_policy_enabled(blkg->q, pol) &&
342 pol->pd_reset_stats_fn)
343 pol->pd_reset_stats_fn(blkg);
347 spin_unlock_irq(&blkcg->lock);
348 mutex_unlock(&blkcg_pol_mutex);
352 static const char *blkg_dev_name(struct blkcg_gq *blkg)
354 /* some drivers (floppy) instantiate a queue w/o disk registered */
355 if (blkg->q->backing_dev_info.dev)
356 return dev_name(blkg->q->backing_dev_info.dev);
361 * blkcg_print_blkgs - helper for printing per-blkg data
362 * @sf: seq_file to print to
363 * @blkcg: blkcg of interest
364 * @prfill: fill function to print out a blkg
365 * @pol: policy in question
366 * @data: data to be passed to @prfill
367 * @show_total: to print out sum of prfill return values or not
369 * This function invokes @prfill on each blkg of @blkcg if pd for the
370 * policy specified by @pol exists. @prfill is invoked with @sf, the
371 * policy data and @data. If @show_total is %true, the sum of the return
372 * values from @prfill is printed with "Total" label at the end.
374 * This is to be used to construct print functions for
375 * cftype->read_seq_string method.
377 void blkcg_print_blkgs(struct seq_file *sf, struct blkcg *blkcg,
378 u64 (*prfill)(struct seq_file *,
379 struct blkg_policy_data *, int),
380 const struct blkcg_policy *pol, int data,
383 struct blkcg_gq *blkg;
384 struct hlist_node *n;
387 spin_lock_irq(&blkcg->lock);
388 hlist_for_each_entry(blkg, n, &blkcg->blkg_list, blkcg_node)
389 if (blkcg_policy_enabled(blkg->q, pol))
390 total += prfill(sf, blkg->pd[pol->plid], data);
391 spin_unlock_irq(&blkcg->lock);
394 seq_printf(sf, "Total %llu\n", (unsigned long long)total);
396 EXPORT_SYMBOL_GPL(blkcg_print_blkgs);
399 * __blkg_prfill_u64 - prfill helper for a single u64 value
400 * @sf: seq_file to print to
401 * @pd: policy private data of interest
404 * Print @v to @sf for the device assocaited with @pd.
406 u64 __blkg_prfill_u64(struct seq_file *sf, struct blkg_policy_data *pd, u64 v)
408 const char *dname = blkg_dev_name(pd->blkg);
413 seq_printf(sf, "%s %llu\n", dname, (unsigned long long)v);
416 EXPORT_SYMBOL_GPL(__blkg_prfill_u64);
419 * __blkg_prfill_rwstat - prfill helper for a blkg_rwstat
420 * @sf: seq_file to print to
421 * @pd: policy private data of interest
422 * @rwstat: rwstat to print
424 * Print @rwstat to @sf for the device assocaited with @pd.
426 u64 __blkg_prfill_rwstat(struct seq_file *sf, struct blkg_policy_data *pd,
427 const struct blkg_rwstat *rwstat)
429 static const char *rwstr[] = {
430 [BLKG_RWSTAT_READ] = "Read",
431 [BLKG_RWSTAT_WRITE] = "Write",
432 [BLKG_RWSTAT_SYNC] = "Sync",
433 [BLKG_RWSTAT_ASYNC] = "Async",
435 const char *dname = blkg_dev_name(pd->blkg);
442 for (i = 0; i < BLKG_RWSTAT_NR; i++)
443 seq_printf(sf, "%s %s %llu\n", dname, rwstr[i],
444 (unsigned long long)rwstat->cnt[i]);
446 v = rwstat->cnt[BLKG_RWSTAT_READ] + rwstat->cnt[BLKG_RWSTAT_WRITE];
447 seq_printf(sf, "%s Total %llu\n", dname, (unsigned long long)v);
452 * blkg_prfill_stat - prfill callback for blkg_stat
453 * @sf: seq_file to print to
454 * @pd: policy private data of interest
455 * @off: offset to the blkg_stat in @pd
457 * prfill callback for printing a blkg_stat.
459 u64 blkg_prfill_stat(struct seq_file *sf, struct blkg_policy_data *pd, int off)
461 return __blkg_prfill_u64(sf, pd, blkg_stat_read((void *)pd + off));
463 EXPORT_SYMBOL_GPL(blkg_prfill_stat);
466 * blkg_prfill_rwstat - prfill callback for blkg_rwstat
467 * @sf: seq_file to print to
468 * @pd: policy private data of interest
469 * @off: offset to the blkg_rwstat in @pd
471 * prfill callback for printing a blkg_rwstat.
473 u64 blkg_prfill_rwstat(struct seq_file *sf, struct blkg_policy_data *pd,
476 struct blkg_rwstat rwstat = blkg_rwstat_read((void *)pd + off);
478 return __blkg_prfill_rwstat(sf, pd, &rwstat);
480 EXPORT_SYMBOL_GPL(blkg_prfill_rwstat);
483 * blkg_conf_prep - parse and prepare for per-blkg config update
484 * @blkcg: target block cgroup
485 * @pol: target policy
486 * @input: input string
487 * @ctx: blkg_conf_ctx to be filled
489 * Parse per-blkg config update from @input and initialize @ctx with the
490 * result. @ctx->blkg points to the blkg to be updated and @ctx->v the new
491 * value. This function returns with RCU read lock and queue lock held and
492 * must be paired with blkg_conf_finish().
494 int blkg_conf_prep(struct blkcg *blkcg, const struct blkcg_policy *pol,
495 const char *input, struct blkg_conf_ctx *ctx)
496 __acquires(rcu) __acquires(disk->queue->queue_lock)
498 struct gendisk *disk;
499 struct blkcg_gq *blkg;
500 unsigned int major, minor;
501 unsigned long long v;
504 if (sscanf(input, "%u:%u %llu", &major, &minor, &v) != 3)
507 disk = get_gendisk(MKDEV(major, minor), &part);
512 spin_lock_irq(disk->queue->queue_lock);
514 if (blkcg_policy_enabled(disk->queue, pol))
515 blkg = blkg_lookup_create(blkcg, disk->queue);
517 blkg = ERR_PTR(-EINVAL);
522 spin_unlock_irq(disk->queue->queue_lock);
525 * If queue was bypassing, we should retry. Do so after a
526 * short msleep(). It isn't strictly necessary but queue
527 * can be bypassing for some time and it's always nice to
528 * avoid busy looping.
532 ret = restart_syscall();
542 EXPORT_SYMBOL_GPL(blkg_conf_prep);
545 * blkg_conf_finish - finish up per-blkg config update
546 * @ctx: blkg_conf_ctx intiailized by blkg_conf_prep()
548 * Finish up after per-blkg config update. This function must be paired
549 * with blkg_conf_prep().
551 void blkg_conf_finish(struct blkg_conf_ctx *ctx)
552 __releases(ctx->disk->queue->queue_lock) __releases(rcu)
554 spin_unlock_irq(ctx->disk->queue->queue_lock);
558 EXPORT_SYMBOL_GPL(blkg_conf_finish);
560 struct cftype blkcg_files[] = {
562 .name = "reset_stats",
563 .write_u64 = blkcg_reset_stats,
569 * blkcg_pre_destroy - cgroup pre_destroy callback
570 * @cgroup: cgroup of interest
572 * This function is called when @cgroup is about to go away and responsible
573 * for shooting down all blkgs associated with @cgroup. blkgs should be
574 * removed while holding both q and blkcg locks. As blkcg lock is nested
575 * inside q lock, this function performs reverse double lock dancing.
577 * This is the blkcg counterpart of ioc_release_fn().
579 static int blkcg_pre_destroy(struct cgroup *cgroup)
581 struct blkcg *blkcg = cgroup_to_blkcg(cgroup);
583 spin_lock_irq(&blkcg->lock);
585 while (!hlist_empty(&blkcg->blkg_list)) {
586 struct blkcg_gq *blkg = hlist_entry(blkcg->blkg_list.first,
587 struct blkcg_gq, blkcg_node);
588 struct request_queue *q = blkg->q;
590 if (spin_trylock(q->queue_lock)) {
592 spin_unlock(q->queue_lock);
594 spin_unlock_irq(&blkcg->lock);
596 spin_lock_irq(&blkcg->lock);
600 spin_unlock_irq(&blkcg->lock);
604 static void blkcg_destroy(struct cgroup *cgroup)
606 struct blkcg *blkcg = cgroup_to_blkcg(cgroup);
608 if (blkcg != &blkcg_root)
612 static struct cgroup_subsys_state *blkcg_create(struct cgroup *cgroup)
614 static atomic64_t id_seq = ATOMIC64_INIT(0);
616 struct cgroup *parent = cgroup->parent;
623 blkcg = kzalloc(sizeof(*blkcg), GFP_KERNEL);
625 return ERR_PTR(-ENOMEM);
627 blkcg->cfq_weight = CFQ_WEIGHT_DEFAULT;
628 blkcg->id = atomic64_inc_return(&id_seq); /* root is 0, start from 1 */
630 spin_lock_init(&blkcg->lock);
631 INIT_RADIX_TREE(&blkcg->blkg_tree, GFP_ATOMIC);
632 INIT_HLIST_HEAD(&blkcg->blkg_list);
638 * blkcg_init_queue - initialize blkcg part of request queue
639 * @q: request_queue to initialize
641 * Called from blk_alloc_queue_node(). Responsible for initializing blkcg
642 * part of new request_queue @q.
645 * 0 on success, -errno on failure.
647 int blkcg_init_queue(struct request_queue *q)
651 return blk_throtl_init(q);
655 * blkcg_drain_queue - drain blkcg part of request_queue
656 * @q: request_queue to drain
658 * Called from blk_drain_queue(). Responsible for draining blkcg part.
660 void blkcg_drain_queue(struct request_queue *q)
662 lockdep_assert_held(q->queue_lock);
668 * blkcg_exit_queue - exit and release blkcg part of request_queue
669 * @q: request_queue being released
671 * Called from blk_release_queue(). Responsible for exiting blkcg part.
673 void blkcg_exit_queue(struct request_queue *q)
675 spin_lock_irq(q->queue_lock);
677 spin_unlock_irq(q->queue_lock);
683 * We cannot support shared io contexts, as we have no mean to support
684 * two tasks with the same ioc in two different groups without major rework
685 * of the main cic data structures. For now we allow a task to change
686 * its cgroup only if it's the only owner of its ioc.
688 static int blkcg_can_attach(struct cgroup *cgrp, struct cgroup_taskset *tset)
690 struct task_struct *task;
691 struct io_context *ioc;
694 /* task_lock() is needed to avoid races with exit_io_context() */
695 cgroup_taskset_for_each(task, cgrp, tset) {
697 ioc = task->io_context;
698 if (ioc && atomic_read(&ioc->nr_tasks) > 1)
707 struct cgroup_subsys blkio_subsys = {
709 .create = blkcg_create,
710 .can_attach = blkcg_can_attach,
711 .pre_destroy = blkcg_pre_destroy,
712 .destroy = blkcg_destroy,
713 .subsys_id = blkio_subsys_id,
714 .base_cftypes = blkcg_files,
715 .module = THIS_MODULE,
717 EXPORT_SYMBOL_GPL(blkio_subsys);
720 * blkcg_activate_policy - activate a blkcg policy on a request_queue
721 * @q: request_queue of interest
722 * @pol: blkcg policy to activate
724 * Activate @pol on @q. Requires %GFP_KERNEL context. @q goes through
725 * bypass mode to populate its blkgs with policy_data for @pol.
727 * Activation happens with @q bypassed, so nobody would be accessing blkgs
728 * from IO path. Update of each blkg is protected by both queue and blkcg
729 * locks so that holding either lock and testing blkcg_policy_enabled() is
730 * always enough for dereferencing policy data.
732 * The caller is responsible for synchronizing [de]activations and policy
733 * [un]registerations. Returns 0 on success, -errno on failure.
735 int blkcg_activate_policy(struct request_queue *q,
736 const struct blkcg_policy *pol)
739 struct blkcg_gq *blkg;
740 struct blkg_policy_data *pd, *n;
743 if (blkcg_policy_enabled(q, pol))
746 blk_queue_bypass_start(q);
748 /* make sure the root blkg exists and count the existing blkgs */
749 spin_lock_irq(q->queue_lock);
752 blkg = __blkg_lookup_create(&blkcg_root, q);
761 list_for_each_entry(blkg, &q->blkg_list, q_node)
764 spin_unlock_irq(q->queue_lock);
766 /* allocate policy_data for all existing blkgs */
768 pd = kzalloc_node(pol->pd_size, GFP_KERNEL, q->node);
773 list_add_tail(&pd->alloc_node, &pds);
777 * Install the allocated pds. With @q bypassing, no new blkg
778 * should have been created while the queue lock was dropped.
780 spin_lock_irq(q->queue_lock);
782 list_for_each_entry(blkg, &q->blkg_list, q_node) {
783 if (WARN_ON(list_empty(&pds))) {
784 /* umm... this shouldn't happen, just abort */
788 pd = list_first_entry(&pds, struct blkg_policy_data, alloc_node);
789 list_del_init(&pd->alloc_node);
791 /* grab blkcg lock too while installing @pd on @blkg */
792 spin_lock(&blkg->blkcg->lock);
794 blkg->pd[pol->plid] = pd;
796 pol->pd_init_fn(blkg);
798 spin_unlock(&blkg->blkcg->lock);
801 __set_bit(pol->plid, q->blkcg_pols);
804 spin_unlock_irq(q->queue_lock);
806 blk_queue_bypass_end(q);
807 list_for_each_entry_safe(pd, n, &pds, alloc_node)
811 EXPORT_SYMBOL_GPL(blkcg_activate_policy);
814 * blkcg_deactivate_policy - deactivate a blkcg policy on a request_queue
815 * @q: request_queue of interest
816 * @pol: blkcg policy to deactivate
818 * Deactivate @pol on @q. Follows the same synchronization rules as
819 * blkcg_activate_policy().
821 void blkcg_deactivate_policy(struct request_queue *q,
822 const struct blkcg_policy *pol)
824 struct blkcg_gq *blkg;
826 if (!blkcg_policy_enabled(q, pol))
829 blk_queue_bypass_start(q);
830 spin_lock_irq(q->queue_lock);
832 __clear_bit(pol->plid, q->blkcg_pols);
834 /* if no policy is left, no need for blkgs - shoot them down */
835 if (bitmap_empty(q->blkcg_pols, BLKCG_MAX_POLS))
838 list_for_each_entry(blkg, &q->blkg_list, q_node) {
839 /* grab blkcg lock too while removing @pd from @blkg */
840 spin_lock(&blkg->blkcg->lock);
843 pol->pd_exit_fn(blkg);
845 kfree(blkg->pd[pol->plid]);
846 blkg->pd[pol->plid] = NULL;
848 spin_unlock(&blkg->blkcg->lock);
851 spin_unlock_irq(q->queue_lock);
852 blk_queue_bypass_end(q);
854 EXPORT_SYMBOL_GPL(blkcg_deactivate_policy);
857 * blkcg_policy_register - register a blkcg policy
858 * @pol: blkcg policy to register
860 * Register @pol with blkcg core. Might sleep and @pol may be modified on
861 * successful registration. Returns 0 on success and -errno on failure.
863 int blkcg_policy_register(struct blkcg_policy *pol)
867 if (WARN_ON(pol->pd_size < sizeof(struct blkg_policy_data)))
870 mutex_lock(&blkcg_pol_mutex);
872 /* find an empty slot */
874 for (i = 0; i < BLKCG_MAX_POLS; i++)
875 if (!blkcg_policy[i])
877 if (i >= BLKCG_MAX_POLS)
880 /* register and update blkgs */
882 blkcg_policy[i] = pol;
884 /* everything is in place, add intf files for the new policy */
886 WARN_ON(cgroup_add_cftypes(&blkio_subsys, pol->cftypes));
889 mutex_unlock(&blkcg_pol_mutex);
892 EXPORT_SYMBOL_GPL(blkcg_policy_register);
895 * blkcg_policy_unregister - unregister a blkcg policy
896 * @pol: blkcg policy to unregister
898 * Undo blkcg_policy_register(@pol). Might sleep.
900 void blkcg_policy_unregister(struct blkcg_policy *pol)
902 mutex_lock(&blkcg_pol_mutex);
904 if (WARN_ON(blkcg_policy[pol->plid] != pol))
907 /* kill the intf files first */
909 cgroup_rm_cftypes(&blkio_subsys, pol->cftypes);
911 /* unregister and update blkgs */
912 blkcg_policy[pol->plid] = NULL;
914 mutex_unlock(&blkcg_pol_mutex);
916 EXPORT_SYMBOL_GPL(blkcg_policy_unregister);