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 .cfq_leaf_weight = 2 * CFQ_WEIGHT_DEFAULT, };
31 EXPORT_SYMBOL_GPL(blkcg_root);
33 static struct blkcg_policy *blkcg_policy[BLKCG_MAX_POLS];
35 static bool blkcg_policy_enabled(struct request_queue *q,
36 const struct blkcg_policy *pol)
38 return pol && test_bit(pol->plid, q->blkcg_pols);
42 * blkg_free - free a blkg
45 * Free @blkg which may be partially allocated.
47 static void blkg_free(struct blkcg_gq *blkg)
54 for (i = 0; i < BLKCG_MAX_POLS; i++) {
55 struct blkcg_policy *pol = blkcg_policy[i];
56 struct blkg_policy_data *pd = blkg->pd[i];
61 if (pol && pol->pd_exit_fn)
62 pol->pd_exit_fn(blkg);
67 blk_exit_rl(&blkg->rl);
72 * blkg_alloc - allocate a blkg
73 * @blkcg: block cgroup the new blkg is associated with
74 * @q: request_queue the new blkg is associated with
75 * @gfp_mask: allocation mask to use
77 * Allocate a new blkg assocating @blkcg and @q.
79 static struct blkcg_gq *blkg_alloc(struct blkcg *blkcg, struct request_queue *q,
82 struct blkcg_gq *blkg;
85 /* alloc and init base part */
86 blkg = kzalloc_node(sizeof(*blkg), gfp_mask, q->node);
91 INIT_LIST_HEAD(&blkg->q_node);
95 /* root blkg uses @q->root_rl, init rl only for !root blkgs */
96 if (blkcg != &blkcg_root) {
97 if (blk_init_rl(&blkg->rl, q, gfp_mask))
102 for (i = 0; i < BLKCG_MAX_POLS; i++) {
103 struct blkcg_policy *pol = blkcg_policy[i];
104 struct blkg_policy_data *pd;
106 if (!blkcg_policy_enabled(q, pol))
109 /* alloc per-policy data and attach it to blkg */
110 pd = kzalloc_node(pol->pd_size, gfp_mask, q->node);
118 /* invoke per-policy init */
120 pol->pd_init_fn(blkg);
130 static struct blkcg_gq *__blkg_lookup(struct blkcg *blkcg,
131 struct request_queue *q, bool update_hint)
133 struct blkcg_gq *blkg;
135 blkg = rcu_dereference(blkcg->blkg_hint);
136 if (blkg && blkg->q == q)
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);
158 * blkg_lookup - lookup blkg for the specified blkcg - q pair
159 * @blkcg: blkcg of interest
160 * @q: request_queue of interest
162 * Lookup blkg for the @blkcg - @q pair. This function should be called
163 * under RCU read lock and is guaranteed to return %NULL if @q is bypassing
164 * - see blk_queue_bypass_start() for details.
166 struct blkcg_gq *blkg_lookup(struct blkcg *blkcg, struct request_queue *q)
168 WARN_ON_ONCE(!rcu_read_lock_held());
170 if (unlikely(blk_queue_bypass(q)))
172 return __blkg_lookup(blkcg, q, false);
174 EXPORT_SYMBOL_GPL(blkg_lookup);
177 * If @new_blkg is %NULL, this function tries to allocate a new one as
178 * necessary using %GFP_ATOMIC. @new_blkg is always consumed on return.
180 static struct blkcg_gq *blkg_create(struct blkcg *blkcg,
181 struct request_queue *q,
182 struct blkcg_gq *new_blkg)
184 struct blkcg_gq *blkg;
187 WARN_ON_ONCE(!rcu_read_lock_held());
188 lockdep_assert_held(q->queue_lock);
190 /* blkg holds a reference to blkcg */
191 if (!css_tryget(&blkcg->css)) {
198 new_blkg = blkg_alloc(blkcg, q, GFP_ATOMIC);
199 if (unlikely(!new_blkg)) {
206 /* link parent and insert */
207 if (blkcg_parent(blkcg)) {
208 blkg->parent = __blkg_lookup(blkcg_parent(blkcg), q, false);
209 if (WARN_ON_ONCE(!blkg->parent)) {
210 blkg = ERR_PTR(-EINVAL);
213 blkg_get(blkg->parent);
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 spin_unlock(&blkcg->lock);
227 /* @blkg failed fully initialized, use the usual release path */
232 css_put(&blkcg->css);
239 * blkg_lookup_create - lookup blkg, try to create one if not there
240 * @blkcg: blkcg of interest
241 * @q: request_queue of interest
243 * Lookup blkg for the @blkcg - @q pair. If it doesn't exist, try to
244 * create one. blkg creation is performed recursively from blkcg_root such
245 * that all non-root blkg's have access to the parent blkg. This function
246 * should be called under RCU read lock and @q->queue_lock.
248 * Returns pointer to the looked up or created blkg on success, ERR_PTR()
249 * value on error. If @q is dead, returns ERR_PTR(-EINVAL). If @q is not
250 * dead and bypassing, returns ERR_PTR(-EBUSY).
252 struct blkcg_gq *blkg_lookup_create(struct blkcg *blkcg,
253 struct request_queue *q)
255 struct blkcg_gq *blkg;
257 WARN_ON_ONCE(!rcu_read_lock_held());
258 lockdep_assert_held(q->queue_lock);
261 * This could be the first entry point of blkcg implementation and
262 * we shouldn't allow anything to go through for a bypassing queue.
264 if (unlikely(blk_queue_bypass(q)))
265 return ERR_PTR(blk_queue_dying(q) ? -EINVAL : -EBUSY);
267 blkg = __blkg_lookup(blkcg, q, true);
272 * Create blkgs walking down from blkcg_root to @blkcg, so that all
273 * non-root blkgs have access to their parents.
276 struct blkcg *pos = blkcg;
277 struct blkcg *parent = blkcg_parent(blkcg);
279 while (parent && !__blkg_lookup(parent, q, false)) {
281 parent = blkcg_parent(parent);
284 blkg = blkg_create(pos, q, NULL);
285 if (pos == blkcg || IS_ERR(blkg))
289 EXPORT_SYMBOL_GPL(blkg_lookup_create);
291 static void blkg_destroy(struct blkcg_gq *blkg)
293 struct blkcg *blkcg = blkg->blkcg;
295 lockdep_assert_held(blkg->q->queue_lock);
296 lockdep_assert_held(&blkcg->lock);
298 /* Something wrong if we are trying to remove same group twice */
299 WARN_ON_ONCE(list_empty(&blkg->q_node));
300 WARN_ON_ONCE(hlist_unhashed(&blkg->blkcg_node));
302 radix_tree_delete(&blkcg->blkg_tree, blkg->q->id);
303 list_del_init(&blkg->q_node);
304 hlist_del_init_rcu(&blkg->blkcg_node);
307 * Both setting lookup hint to and clearing it from @blkg are done
308 * under queue_lock. If it's not pointing to @blkg now, it never
309 * will. Hint assignment itself can race safely.
311 if (rcu_dereference_raw(blkcg->blkg_hint) == blkg)
312 rcu_assign_pointer(blkcg->blkg_hint, NULL);
315 * Put the reference taken at the time of creation so that when all
316 * queues are gone, group can be destroyed.
322 * blkg_destroy_all - destroy all blkgs associated with a request_queue
323 * @q: request_queue of interest
325 * Destroy all blkgs associated with @q.
327 static void blkg_destroy_all(struct request_queue *q)
329 struct blkcg_gq *blkg, *n;
331 lockdep_assert_held(q->queue_lock);
333 list_for_each_entry_safe(blkg, n, &q->blkg_list, q_node) {
334 struct blkcg *blkcg = blkg->blkcg;
336 spin_lock(&blkcg->lock);
338 spin_unlock(&blkcg->lock);
342 * root blkg is destroyed. Just clear the pointer since
343 * root_rl does not take reference on root blkg.
346 q->root_rl.blkg = NULL;
349 static void blkg_rcu_free(struct rcu_head *rcu_head)
351 blkg_free(container_of(rcu_head, struct blkcg_gq, rcu_head));
354 void __blkg_release(struct blkcg_gq *blkg)
356 /* release the blkcg and parent blkg refs this blkg has been holding */
357 css_put(&blkg->blkcg->css);
359 blkg_put(blkg->parent);
362 * A group is freed in rcu manner. But having an rcu lock does not
363 * mean that one can access all the fields of blkg and assume these
364 * are valid. For example, don't try to follow throtl_data and
365 * request queue links.
367 * Having a reference to blkg under an rcu allows acess to only
368 * values local to groups like group stats and group rate limits
370 call_rcu(&blkg->rcu_head, blkg_rcu_free);
372 EXPORT_SYMBOL_GPL(__blkg_release);
375 * The next function used by blk_queue_for_each_rl(). It's a bit tricky
376 * because the root blkg uses @q->root_rl instead of its own rl.
378 struct request_list *__blk_queue_next_rl(struct request_list *rl,
379 struct request_queue *q)
381 struct list_head *ent;
382 struct blkcg_gq *blkg;
385 * Determine the current blkg list_head. The first entry is
386 * root_rl which is off @q->blkg_list and mapped to the head.
388 if (rl == &q->root_rl) {
390 /* There are no more block groups, hence no request lists */
394 blkg = container_of(rl, struct blkcg_gq, rl);
398 /* walk to the next list_head, skip root blkcg */
400 if (ent == &q->root_blkg->q_node)
402 if (ent == &q->blkg_list)
405 blkg = container_of(ent, struct blkcg_gq, q_node);
409 static int blkcg_reset_stats(struct cgroup *cgroup, struct cftype *cftype,
412 struct blkcg *blkcg = cgroup_to_blkcg(cgroup);
413 struct blkcg_gq *blkg;
414 struct hlist_node *n;
417 mutex_lock(&blkcg_pol_mutex);
418 spin_lock_irq(&blkcg->lock);
421 * Note that stat reset is racy - it doesn't synchronize against
422 * stat updates. This is a debug feature which shouldn't exist
423 * anyway. If you get hit by a race, retry.
425 hlist_for_each_entry(blkg, n, &blkcg->blkg_list, blkcg_node) {
426 for (i = 0; i < BLKCG_MAX_POLS; i++) {
427 struct blkcg_policy *pol = blkcg_policy[i];
429 if (blkcg_policy_enabled(blkg->q, pol) &&
430 pol->pd_reset_stats_fn)
431 pol->pd_reset_stats_fn(blkg);
435 spin_unlock_irq(&blkcg->lock);
436 mutex_unlock(&blkcg_pol_mutex);
440 static const char *blkg_dev_name(struct blkcg_gq *blkg)
442 /* some drivers (floppy) instantiate a queue w/o disk registered */
443 if (blkg->q->backing_dev_info.dev)
444 return dev_name(blkg->q->backing_dev_info.dev);
449 * blkcg_print_blkgs - helper for printing per-blkg data
450 * @sf: seq_file to print to
451 * @blkcg: blkcg of interest
452 * @prfill: fill function to print out a blkg
453 * @pol: policy in question
454 * @data: data to be passed to @prfill
455 * @show_total: to print out sum of prfill return values or not
457 * This function invokes @prfill on each blkg of @blkcg if pd for the
458 * policy specified by @pol exists. @prfill is invoked with @sf, the
459 * policy data and @data. If @show_total is %true, the sum of the return
460 * values from @prfill is printed with "Total" label at the end.
462 * This is to be used to construct print functions for
463 * cftype->read_seq_string method.
465 void blkcg_print_blkgs(struct seq_file *sf, struct blkcg *blkcg,
466 u64 (*prfill)(struct seq_file *,
467 struct blkg_policy_data *, int),
468 const struct blkcg_policy *pol, int data,
471 struct blkcg_gq *blkg;
472 struct hlist_node *n;
475 spin_lock_irq(&blkcg->lock);
476 hlist_for_each_entry(blkg, n, &blkcg->blkg_list, blkcg_node)
477 if (blkcg_policy_enabled(blkg->q, pol))
478 total += prfill(sf, blkg->pd[pol->plid], data);
479 spin_unlock_irq(&blkcg->lock);
482 seq_printf(sf, "Total %llu\n", (unsigned long long)total);
484 EXPORT_SYMBOL_GPL(blkcg_print_blkgs);
487 * __blkg_prfill_u64 - prfill helper for a single u64 value
488 * @sf: seq_file to print to
489 * @pd: policy private data of interest
492 * Print @v to @sf for the device assocaited with @pd.
494 u64 __blkg_prfill_u64(struct seq_file *sf, struct blkg_policy_data *pd, u64 v)
496 const char *dname = blkg_dev_name(pd->blkg);
501 seq_printf(sf, "%s %llu\n", dname, (unsigned long long)v);
504 EXPORT_SYMBOL_GPL(__blkg_prfill_u64);
507 * __blkg_prfill_rwstat - prfill helper for a blkg_rwstat
508 * @sf: seq_file to print to
509 * @pd: policy private data of interest
510 * @rwstat: rwstat to print
512 * Print @rwstat to @sf for the device assocaited with @pd.
514 u64 __blkg_prfill_rwstat(struct seq_file *sf, struct blkg_policy_data *pd,
515 const struct blkg_rwstat *rwstat)
517 static const char *rwstr[] = {
518 [BLKG_RWSTAT_READ] = "Read",
519 [BLKG_RWSTAT_WRITE] = "Write",
520 [BLKG_RWSTAT_SYNC] = "Sync",
521 [BLKG_RWSTAT_ASYNC] = "Async",
523 const char *dname = blkg_dev_name(pd->blkg);
530 for (i = 0; i < BLKG_RWSTAT_NR; i++)
531 seq_printf(sf, "%s %s %llu\n", dname, rwstr[i],
532 (unsigned long long)rwstat->cnt[i]);
534 v = rwstat->cnt[BLKG_RWSTAT_READ] + rwstat->cnt[BLKG_RWSTAT_WRITE];
535 seq_printf(sf, "%s Total %llu\n", dname, (unsigned long long)v);
540 * blkg_prfill_stat - prfill callback for blkg_stat
541 * @sf: seq_file to print to
542 * @pd: policy private data of interest
543 * @off: offset to the blkg_stat in @pd
545 * prfill callback for printing a blkg_stat.
547 u64 blkg_prfill_stat(struct seq_file *sf, struct blkg_policy_data *pd, int off)
549 return __blkg_prfill_u64(sf, pd, blkg_stat_read((void *)pd + off));
551 EXPORT_SYMBOL_GPL(blkg_prfill_stat);
554 * blkg_prfill_rwstat - prfill callback for blkg_rwstat
555 * @sf: seq_file to print to
556 * @pd: policy private data of interest
557 * @off: offset to the blkg_rwstat in @pd
559 * prfill callback for printing a blkg_rwstat.
561 u64 blkg_prfill_rwstat(struct seq_file *sf, struct blkg_policy_data *pd,
564 struct blkg_rwstat rwstat = blkg_rwstat_read((void *)pd + off);
566 return __blkg_prfill_rwstat(sf, pd, &rwstat);
568 EXPORT_SYMBOL_GPL(blkg_prfill_rwstat);
571 * blkg_conf_prep - parse and prepare for per-blkg config update
572 * @blkcg: target block cgroup
573 * @pol: target policy
574 * @input: input string
575 * @ctx: blkg_conf_ctx to be filled
577 * Parse per-blkg config update from @input and initialize @ctx with the
578 * result. @ctx->blkg points to the blkg to be updated and @ctx->v the new
579 * value. This function returns with RCU read lock and queue lock held and
580 * must be paired with blkg_conf_finish().
582 int blkg_conf_prep(struct blkcg *blkcg, const struct blkcg_policy *pol,
583 const char *input, struct blkg_conf_ctx *ctx)
584 __acquires(rcu) __acquires(disk->queue->queue_lock)
586 struct gendisk *disk;
587 struct blkcg_gq *blkg;
588 unsigned int major, minor;
589 unsigned long long v;
592 if (sscanf(input, "%u:%u %llu", &major, &minor, &v) != 3)
595 disk = get_gendisk(MKDEV(major, minor), &part);
600 spin_lock_irq(disk->queue->queue_lock);
602 if (blkcg_policy_enabled(disk->queue, pol))
603 blkg = blkg_lookup_create(blkcg, disk->queue);
605 blkg = ERR_PTR(-EINVAL);
610 spin_unlock_irq(disk->queue->queue_lock);
613 * If queue was bypassing, we should retry. Do so after a
614 * short msleep(). It isn't strictly necessary but queue
615 * can be bypassing for some time and it's always nice to
616 * avoid busy looping.
620 ret = restart_syscall();
630 EXPORT_SYMBOL_GPL(blkg_conf_prep);
633 * blkg_conf_finish - finish up per-blkg config update
634 * @ctx: blkg_conf_ctx intiailized by blkg_conf_prep()
636 * Finish up after per-blkg config update. This function must be paired
637 * with blkg_conf_prep().
639 void blkg_conf_finish(struct blkg_conf_ctx *ctx)
640 __releases(ctx->disk->queue->queue_lock) __releases(rcu)
642 spin_unlock_irq(ctx->disk->queue->queue_lock);
646 EXPORT_SYMBOL_GPL(blkg_conf_finish);
648 struct cftype blkcg_files[] = {
650 .name = "reset_stats",
651 .write_u64 = blkcg_reset_stats,
657 * blkcg_css_offline - cgroup css_offline callback
658 * @cgroup: cgroup of interest
660 * This function is called when @cgroup is about to go away and responsible
661 * for shooting down all blkgs associated with @cgroup. blkgs should be
662 * removed while holding both q and blkcg locks. As blkcg lock is nested
663 * inside q lock, this function performs reverse double lock dancing.
665 * This is the blkcg counterpart of ioc_release_fn().
667 static void blkcg_css_offline(struct cgroup *cgroup)
669 struct blkcg *blkcg = cgroup_to_blkcg(cgroup);
671 spin_lock_irq(&blkcg->lock);
673 while (!hlist_empty(&blkcg->blkg_list)) {
674 struct blkcg_gq *blkg = hlist_entry(blkcg->blkg_list.first,
675 struct blkcg_gq, blkcg_node);
676 struct request_queue *q = blkg->q;
678 if (spin_trylock(q->queue_lock)) {
680 spin_unlock(q->queue_lock);
682 spin_unlock_irq(&blkcg->lock);
684 spin_lock_irq(&blkcg->lock);
688 spin_unlock_irq(&blkcg->lock);
691 static void blkcg_css_free(struct cgroup *cgroup)
693 struct blkcg *blkcg = cgroup_to_blkcg(cgroup);
695 if (blkcg != &blkcg_root)
699 static struct cgroup_subsys_state *blkcg_css_alloc(struct cgroup *cgroup)
701 static atomic64_t id_seq = ATOMIC64_INIT(0);
703 struct cgroup *parent = cgroup->parent;
710 blkcg = kzalloc(sizeof(*blkcg), GFP_KERNEL);
712 return ERR_PTR(-ENOMEM);
714 blkcg->cfq_weight = CFQ_WEIGHT_DEFAULT;
715 blkcg->cfq_leaf_weight = CFQ_WEIGHT_DEFAULT;
716 blkcg->id = atomic64_inc_return(&id_seq); /* root is 0, start from 1 */
718 spin_lock_init(&blkcg->lock);
719 INIT_RADIX_TREE(&blkcg->blkg_tree, GFP_ATOMIC);
720 INIT_HLIST_HEAD(&blkcg->blkg_list);
726 * blkcg_init_queue - initialize blkcg part of request queue
727 * @q: request_queue to initialize
729 * Called from blk_alloc_queue_node(). Responsible for initializing blkcg
730 * part of new request_queue @q.
733 * 0 on success, -errno on failure.
735 int blkcg_init_queue(struct request_queue *q)
739 return blk_throtl_init(q);
743 * blkcg_drain_queue - drain blkcg part of request_queue
744 * @q: request_queue to drain
746 * Called from blk_drain_queue(). Responsible for draining blkcg part.
748 void blkcg_drain_queue(struct request_queue *q)
750 lockdep_assert_held(q->queue_lock);
756 * blkcg_exit_queue - exit and release blkcg part of request_queue
757 * @q: request_queue being released
759 * Called from blk_release_queue(). Responsible for exiting blkcg part.
761 void blkcg_exit_queue(struct request_queue *q)
763 spin_lock_irq(q->queue_lock);
765 spin_unlock_irq(q->queue_lock);
771 * We cannot support shared io contexts, as we have no mean to support
772 * two tasks with the same ioc in two different groups without major rework
773 * of the main cic data structures. For now we allow a task to change
774 * its cgroup only if it's the only owner of its ioc.
776 static int blkcg_can_attach(struct cgroup *cgrp, struct cgroup_taskset *tset)
778 struct task_struct *task;
779 struct io_context *ioc;
782 /* task_lock() is needed to avoid races with exit_io_context() */
783 cgroup_taskset_for_each(task, cgrp, tset) {
785 ioc = task->io_context;
786 if (ioc && atomic_read(&ioc->nr_tasks) > 1)
795 struct cgroup_subsys blkio_subsys = {
797 .css_alloc = blkcg_css_alloc,
798 .css_offline = blkcg_css_offline,
799 .css_free = blkcg_css_free,
800 .can_attach = blkcg_can_attach,
801 .subsys_id = blkio_subsys_id,
802 .base_cftypes = blkcg_files,
803 .module = THIS_MODULE,
806 * blkio subsystem is utterly broken in terms of hierarchy support.
807 * It treats all cgroups equally regardless of where they're
808 * located in the hierarchy - all cgroups are treated as if they're
809 * right below the root. Fix it and remove the following.
811 .broken_hierarchy = true,
813 EXPORT_SYMBOL_GPL(blkio_subsys);
816 * blkcg_activate_policy - activate a blkcg policy on a request_queue
817 * @q: request_queue of interest
818 * @pol: blkcg policy to activate
820 * Activate @pol on @q. Requires %GFP_KERNEL context. @q goes through
821 * bypass mode to populate its blkgs with policy_data for @pol.
823 * Activation happens with @q bypassed, so nobody would be accessing blkgs
824 * from IO path. Update of each blkg is protected by both queue and blkcg
825 * locks so that holding either lock and testing blkcg_policy_enabled() is
826 * always enough for dereferencing policy data.
828 * The caller is responsible for synchronizing [de]activations and policy
829 * [un]registerations. Returns 0 on success, -errno on failure.
831 int blkcg_activate_policy(struct request_queue *q,
832 const struct blkcg_policy *pol)
835 struct blkcg_gq *blkg, *new_blkg;
836 struct blkg_policy_data *pd, *n;
840 if (blkcg_policy_enabled(q, pol))
843 /* preallocations for root blkg */
844 new_blkg = blkg_alloc(&blkcg_root, q, GFP_KERNEL);
848 preloaded = !radix_tree_preload(GFP_KERNEL);
850 blk_queue_bypass_start(q);
853 * Make sure the root blkg exists and count the existing blkgs. As
854 * @q is bypassing at this point, blkg_lookup_create() can't be
855 * used. Open code it.
857 spin_lock_irq(q->queue_lock);
860 blkg = __blkg_lookup(&blkcg_root, q, false);
864 blkg = blkg_create(&blkcg_root, q, new_blkg);
868 radix_tree_preload_end();
875 q->root_rl.blkg = blkg;
877 list_for_each_entry(blkg, &q->blkg_list, q_node)
880 spin_unlock_irq(q->queue_lock);
882 /* allocate policy_data for all existing blkgs */
884 pd = kzalloc_node(pol->pd_size, GFP_KERNEL, q->node);
889 list_add_tail(&pd->alloc_node, &pds);
893 * Install the allocated pds. With @q bypassing, no new blkg
894 * should have been created while the queue lock was dropped.
896 spin_lock_irq(q->queue_lock);
898 list_for_each_entry(blkg, &q->blkg_list, q_node) {
899 if (WARN_ON(list_empty(&pds))) {
900 /* umm... this shouldn't happen, just abort */
904 pd = list_first_entry(&pds, struct blkg_policy_data, alloc_node);
905 list_del_init(&pd->alloc_node);
907 /* grab blkcg lock too while installing @pd on @blkg */
908 spin_lock(&blkg->blkcg->lock);
910 blkg->pd[pol->plid] = pd;
912 pd->plid = pol->plid;
913 pol->pd_init_fn(blkg);
915 spin_unlock(&blkg->blkcg->lock);
918 __set_bit(pol->plid, q->blkcg_pols);
921 spin_unlock_irq(q->queue_lock);
923 blk_queue_bypass_end(q);
924 list_for_each_entry_safe(pd, n, &pds, alloc_node)
928 EXPORT_SYMBOL_GPL(blkcg_activate_policy);
931 * blkcg_deactivate_policy - deactivate a blkcg policy on a request_queue
932 * @q: request_queue of interest
933 * @pol: blkcg policy to deactivate
935 * Deactivate @pol on @q. Follows the same synchronization rules as
936 * blkcg_activate_policy().
938 void blkcg_deactivate_policy(struct request_queue *q,
939 const struct blkcg_policy *pol)
941 struct blkcg_gq *blkg;
943 if (!blkcg_policy_enabled(q, pol))
946 blk_queue_bypass_start(q);
947 spin_lock_irq(q->queue_lock);
949 __clear_bit(pol->plid, q->blkcg_pols);
951 /* if no policy is left, no need for blkgs - shoot them down */
952 if (bitmap_empty(q->blkcg_pols, BLKCG_MAX_POLS))
955 list_for_each_entry(blkg, &q->blkg_list, q_node) {
956 /* grab blkcg lock too while removing @pd from @blkg */
957 spin_lock(&blkg->blkcg->lock);
960 pol->pd_exit_fn(blkg);
962 kfree(blkg->pd[pol->plid]);
963 blkg->pd[pol->plid] = NULL;
965 spin_unlock(&blkg->blkcg->lock);
968 spin_unlock_irq(q->queue_lock);
969 blk_queue_bypass_end(q);
971 EXPORT_SYMBOL_GPL(blkcg_deactivate_policy);
974 * blkcg_policy_register - register a blkcg policy
975 * @pol: blkcg policy to register
977 * Register @pol with blkcg core. Might sleep and @pol may be modified on
978 * successful registration. Returns 0 on success and -errno on failure.
980 int blkcg_policy_register(struct blkcg_policy *pol)
984 if (WARN_ON(pol->pd_size < sizeof(struct blkg_policy_data)))
987 mutex_lock(&blkcg_pol_mutex);
989 /* find an empty slot */
991 for (i = 0; i < BLKCG_MAX_POLS; i++)
992 if (!blkcg_policy[i])
994 if (i >= BLKCG_MAX_POLS)
997 /* register and update blkgs */
999 blkcg_policy[i] = pol;
1001 /* everything is in place, add intf files for the new policy */
1003 WARN_ON(cgroup_add_cftypes(&blkio_subsys, pol->cftypes));
1006 mutex_unlock(&blkcg_pol_mutex);
1009 EXPORT_SYMBOL_GPL(blkcg_policy_register);
1012 * blkcg_policy_unregister - unregister a blkcg policy
1013 * @pol: blkcg policy to unregister
1015 * Undo blkcg_policy_register(@pol). Might sleep.
1017 void blkcg_policy_unregister(struct blkcg_policy *pol)
1019 mutex_lock(&blkcg_pol_mutex);
1021 if (WARN_ON(blkcg_policy[pol->plid] != pol))
1024 /* kill the intf files first */
1026 cgroup_rm_cftypes(&blkio_subsys, pol->cftypes);
1028 /* unregister and update blkgs */
1029 blkcg_policy[pol->plid] = NULL;
1031 mutex_unlock(&blkcg_pol_mutex);
1033 EXPORT_SYMBOL_GPL(blkcg_policy_unregister);